From: Michel Dänzer <daenzer@vmware.com>
Date: Wed, 10 Mar 2010 17:33:17 +0000 (+0100)
Subject: Merge branch '7.8'
X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=commitdiff_plain;h=a56da1005d30da60701e33b75d5f4f37096df060;hp=766d90f1a201d8fd19e49df14d712efbb82223b4

Merge branch '7.8'
---

diff --git a/Makefile b/Makefile
index c4ee9e583cf..4a813120468 100644
--- a/Makefile
+++ b/Makefile
@@ -180,7 +180,7 @@ ultrix-gcc:
 
 # Rules for making release tarballs
 
-VERSION=7.8-devel
+VERSION=7.9-devel
 DIRECTORY = Mesa-$(VERSION)
 LIB_NAME = MesaLib-$(VERSION)
 DEMO_NAME = MesaDemos-$(VERSION)
@@ -215,7 +215,6 @@ MAIN_FILES = \
 	$(DIRECTORY)/include/GL/glx_mangle.h				\
 	$(DIRECTORY)/include/GL/glfbdev.h				\
 	$(DIRECTORY)/include/GL/mesa_wgl.h				\
-	$(DIRECTORY)/include/GL/mglmesa.h				\
 	$(DIRECTORY)/include/GL/osmesa.h				\
 	$(DIRECTORY)/include/GL/vms_x_fix.h				\
 	$(DIRECTORY)/include/GL/wglext.h				\
diff --git a/SConstruct b/SConstruct
index e1c4a1898ce..6ed44ddd067 100644
--- a/SConstruct
+++ b/SConstruct
@@ -110,9 +110,11 @@ Export([
 #######################################################################
 # Environment setup
 
-# Always build trace driver
+# Always build trace and identity drivers
 if 'trace' not in env['drivers']:
     env['drivers'].append('trace')
+if 'identity' not in env['drivers']:
+    env['drivers'].append('identity')
 
 # Includes
 env.Append(CPPPATH = [
@@ -120,6 +122,7 @@ env.Append(CPPPATH = [
 	'#/src/gallium/include',
 	'#/src/gallium/auxiliary',
 	'#/src/gallium/drivers',
+	'#/src/gallium/winsys',
 ])
 
 if env['msvc']:
diff --git a/configs/autoconf.in b/configs/autoconf.in
index 30637877f3b..bf34f3bffad 100644
--- a/configs/autoconf.in
+++ b/configs/autoconf.in
@@ -73,6 +73,7 @@ EGL_DRIVERS_DIRS = @EGL_DRIVERS_DIRS@
 GALLIUM_DIRS = @GALLIUM_DIRS@
 GALLIUM_DRIVERS_DIRS = @GALLIUM_DRIVERS_DIRS@
 GALLIUM_WINSYS_DIRS = @GALLIUM_WINSYS_DIRS@
+GALLIUM_TARGET_DIRS = @GALLIUM_TARGET_DIRS@
 GALLIUM_WINSYS_DRM_DIRS = @GALLIUM_WINSYS_DRM_DIRS@
 GALLIUM_STATE_TRACKERS_DIRS = @GALLIUM_STATE_TRACKERS_DIRS@
 GALLIUM_AUXILIARIES = $(TOP)/src/gallium/auxiliary/libgallium.a
@@ -83,7 +84,6 @@ PROGRAM_DIRS = @PROGRAM_DIRS@
 
 # Driver specific build vars
 DRI_DIRS = @DRI_DIRS@ 
-WINDOW_SYSTEM = @WINDOW_SYSTEM@
 EGL_DISPLAYS = @EGL_DISPLAYS@
 
 # Dependencies
diff --git a/configs/default b/configs/default
index ad6d93c92ff..a6a92efe8fb 100644
--- a/configs/default
+++ b/configs/default
@@ -9,7 +9,7 @@ CONFIG_NAME = default
 
 # Version info
 MESA_MAJOR=7
-MESA_MINOR=8
+MESA_MINOR=9
 MESA_TINY=0
 MESA_VERSION = $(MESA_MAJOR).$(MESA_MINOR).$(MESA_TINY)
 
@@ -86,7 +86,7 @@ MOTIF_CFLAGS = -I/usr/include/Motif1.2
 
 # Directories to build
 LIB_DIR = lib
-SRC_DIRS = glsl mesa gallium egl gallium/winsys glu glut/glx glew glw
+SRC_DIRS = glsl mesa gallium egl gallium/winsys gallium/targets glu glut/glx glew glw
 GLU_DIRS = sgi
 DRIVER_DIRS = x11 osmesa
 # Which subdirs under $(TOP)/progs/ to enter:
@@ -100,7 +100,8 @@ GALLIUM_DIRS = auxiliary drivers state_trackers
 GALLIUM_AUXILIARIES = $(TOP)/src/gallium/auxiliary/libgallium.a
 GALLIUM_DRIVERS_DIRS = softpipe failover svga i915 i965 r300 trace identity
 GALLIUM_DRIVERS = $(foreach DIR,$(GALLIUM_DRIVERS_DIRS),$(TOP)/src/gallium/drivers/$(DIR)/lib$(DIR).a)
-GALLIUM_WINSYS_DIRS = drm xlib
+GALLIUM_WINSYS_DIRS = drm null xlib
+GALLIUM_TARGET_DIRS = libgl-xlib
 GALLIUM_WINSYS_DRM_DIRS = swrast
 GALLIUM_STATE_TRACKERS_DIRS = glx vega
 
diff --git a/configs/freebsd-dri b/configs/freebsd-dri
index 0b91cfc37e9..2c697e1c6e4 100644
--- a/configs/freebsd-dri
+++ b/configs/freebsd-dri
@@ -44,7 +44,6 @@ GLW_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -L/usr/local/lib -lGL -lXt -lX11
 SRC_DIRS = glx gallium mesa glu glut/glx glew glw
 DRIVER_DIRS = dri
 PROGRAM_DIRS = 
-WINDOW_SYSTEM=dri
 
 DRM_SOURCE_PATH=$(TOP)/../drm
 
diff --git a/configs/linux-cell b/configs/linux-cell
index e89a08cd934..306b9e01f47 100644
--- a/configs/linux-cell
+++ b/configs/linux-cell
@@ -37,7 +37,7 @@ CXXFLAGS = $(COMMON_C_CPP_FLAGS)
 
 
 # Omitting glw here:
-SRC_DIRS = glsl mesa gallium gallium/winsys glu glut/glx glew
+SRC_DIRS = glsl mesa gallium gallium/winsys gallium/targets glu glut/glx glew
 
 # Build no traditional Mesa drivers:
 DRIVER_DIRS =
diff --git a/configs/linux-dri b/configs/linux-dri
index e8e8ccfcf94..04bcfd859d1 100644
--- a/configs/linux-dri
+++ b/configs/linux-dri
@@ -58,8 +58,8 @@ PROGRAM_DIRS := egl $(PROGRAM_DIRS)
 EGL_DRIVERS_DIRS = glx
 
 DRIVER_DIRS = dri
-WINDOW_SYSTEM = dri
 GALLIUM_WINSYS_DIRS = drm
+GALLIUM_TARGET_DIRS =
 GALLIUM_WINSYS_DRM_DIRS = vmware intel i965
 GALLIUM_STATE_TRACKERS_DIRS = egl
 
diff --git a/configs/linux-dri-xcb b/configs/linux-dri-xcb
index c5ed89c4b34..1ed980aa36e 100644
--- a/configs/linux-dri-xcb
+++ b/configs/linux-dri-xcb
@@ -51,7 +51,5 @@ SRC_DIRS = glx gallium mesa glu glut/glx glew glw
 PROGRAM_DIRS = xdemos
 
 DRIVER_DIRS = dri
-WINDOW_SYSTEM=dri
-
 DRI_DIRS = i810 i915 mach64 mga r128 r200 r300 radeon \
 	savage sis tdfx unichrome
diff --git a/configs/linux-egl b/configs/linux-egl
index 2c2834b81d7..28165ed5b5a 100644
--- a/configs/linux-egl
+++ b/configs/linux-egl
@@ -47,11 +47,11 @@ GL_LIB_DEPS   = $(EXTRA_LIB_PATH) -lX11 -lXext -lXxf86vm -lXdamage -lXfixes \
 
 
 # Directories
-SRC_DIRS = gallium mesa gallium/winsys glu egl
+SRC_DIRS = gallium mesa gallium/winsys gallium/targets glu egl
 PROGRAM_DIRS = egl
 
 DRIVER_DIRS = dri
-WINDOW_SYSTEM = dri
 GALLIUM_WINSYS_DIRS = egl_drm
+GALLIUM_TARGET_DIRS =
 
 DRI_DIRS = intel 
diff --git a/configs/linux-i965 b/configs/linux-i965
index e66abc347bb..7656a2adc5e 100644
--- a/configs/linux-i965
+++ b/configs/linux-i965
@@ -6,3 +6,4 @@ CONFIG_NAME = linux-i965
 
 GALLIUM_DRIVER_DIRS = i965
 GALLIUM_WINSYS_DIRS = drm/i965/xlib
+GALLIUM_TARGET_DIRS =
diff --git a/configs/linux-indirect b/configs/linux-indirect
index 0914fba19bc..1c7dd857db6 100644
--- a/configs/linux-indirect
+++ b/configs/linux-indirect
@@ -50,4 +50,3 @@ GL_LIB_DEPS   = $(EXTRA_LIB_PATH) -lX11 -lXext -lXxf86vm -lm -lpthread -ldl
 SRC_DIRS = glx glu glut/glx glew glw
 DRIVER_DIRS =
 PROGRAM_DIRS = 
-WINDOW_SYSTEM=dri
diff --git a/configs/linux-opengl-es b/configs/linux-opengl-es
index 259c26a931b..76054aad14e 100644
--- a/configs/linux-opengl-es
+++ b/configs/linux-opengl-es
@@ -6,7 +6,7 @@ CONFIG_NAME = linux-opengl-es
 
 # Directories to build
 LIB_DIR = lib
-SRC_DIRS = egl glsl mesa/es gallium gallium/winsys
+SRC_DIRS = egl glsl mesa/es gallium gallium/winsys gallium/targets
 PROGRAM_DIRS = es1/screen es1/xegl es2/xegl
 
 # egl st needs this
diff --git a/configure.ac b/configure.ac
index 8ca274932cc..70e158f5d7a 100644
--- a/configure.ac
+++ b/configure.ac
@@ -457,7 +457,6 @@ CORE_DIRS="glsl mesa"
 
 SRC_DIRS="glew"
 GLU_DIRS="sgi"
-WINDOW_SYSTEM=""
 GALLIUM_DIRS="auxiliary drivers state_trackers"
 GALLIUM_WINSYS_DIRS=""
 GALLIUM_WINSYS_DRM_DIRS=""
@@ -472,7 +471,6 @@ xlib)
 dri)
     SRC_DIRS="$SRC_DIRS glx"
     DRIVER_DIRS="dri"
-    WINDOW_SYSTEM="dri"
     GALLIUM_WINSYS_DIRS="$GALLIUM_WINSYS_DIRS drm"
     ;;
 osmesa)
@@ -482,7 +480,6 @@ esac
 AC_SUBST([SRC_DIRS])
 AC_SUBST([GLU_DIRS])
 AC_SUBST([DRIVER_DIRS])
-AC_SUBST([WINDOW_SYSTEM])
 AC_SUBST([GALLIUM_DIRS])
 AC_SUBST([GALLIUM_WINSYS_DIRS])
 AC_SUBST([GALLIUM_WINSYS_DRM_DIRS])
diff --git a/docs/install.html b/docs/install.html
index 5aea92e0b51..3962ea5c91e 100644
--- a/docs/install.html
+++ b/docs/install.html
@@ -361,7 +361,7 @@ To build Mesa with SCons for Windows on Linux using the MinGW crosscompiler tool
 This will create:
 </p>
 <ul>
-<li>build/windows-x86-debug/gallium/winsys/gdi/opengl32.dll &mdash; Mesa + Gallium + softpipe, binary compatible with Windows's opengl32.dll 
+<li>build/windows-x86-debug/gallium/targets/libgl-gdi/opengl32.dll &mdash; Mesa + Gallium + softpipe, binary compatible with Windows's opengl32.dll 
 <li>build/windows-x86-debug/glut/glx/glut32.dll
 <li>progs/build/windows-x86-debug/wgl/wglinfo.exe
 <li>progs/build/windows-x86-debug/trivial/tri.exe
diff --git a/docs/relnotes-7.9.html b/docs/relnotes-7.9.html
new file mode 100644
index 00000000000..f7d5016dc69
--- /dev/null
+++ b/docs/relnotes-7.9.html
@@ -0,0 +1,50 @@
+<HTML>
+
+<TITLE>Mesa Release Notes</TITLE>
+
+<head><link rel="stylesheet" type="text/css" href="mesa.css"></head>
+
+<BODY>
+
+<body bgcolor="#eeeeee">
+
+<H1>Mesa 7.9 Release Notes / date TBD</H1>
+
+<p>
+Mesa 7.9 is a new development release.
+People who are concerned with stability and reliability should stick
+with a previous release or wait for Mesa 7.9.1.
+</p>
+<p>
+Mesa 7.9 implements the OpenGL 2.1 API, but the version reported by
+glGetString(GL_VERSION) depends on the particular driver being used.
+Some drivers don't support all the features required in OpenGL 2.1.
+</p>
+<p>
+See the <a href="install.html">Compiling/Installing page</a> for prerequisites
+for DRI hardware acceleration.
+</p>
+
+
+<h2>MD5 checksums</h2>
+<pre>
+tbd
+</pre>
+
+
+<h2>New features</h2>
+<ul>
+</ul>
+
+
+<h2>Bug fixes</h2>
+<ul>
+</ul>
+
+
+<h2>Changes</h2>
+<ul>
+</ul>
+
+</body>
+</html>
diff --git a/include/GL/glut_h.dja b/include/GL/glut_h.dja
deleted file mode 100644
index e76dcb985e3..00000000000
--- a/include/GL/glut_h.dja
+++ /dev/null
@@ -1,340 +0,0 @@
-
-/*
- * Mesa 3-D graphics library
- * Version:  3.1
- * Copyright (C) 1995-1998  Brian Paul
- *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Library 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
- * Library General Public License for more details.
- *
- * You should have received a copy of the GNU Library General Public
- * License along with this library; if not, write to the Free
- * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
- */
-
-
-/*
- * This header file is based on the REAL glut.h by Mark J. Kilgard.
- *
- * The DJGPP/ALLEGRO (DJA) GLUT implementation was written by
- * Bernhard Tschirren (bernie-t@geocities.com) for the sole purpose
- * of compiling all the sample programs (which use GLUT). Therefore,
- * is NOT AT ALL a complete version of GLUT!
- */
-
-
-#ifndef __AGLUT_H__
-#define __AGLUT_H__
-
-#include <GL/gl.h>
-#include <GL/glu.h>
-
-#define GLUTCALLBACK
-#define APIENTRY
-#define GLUTAPI             extern
-
-#define GLUT_RGB			0
-#define GLUT_RGBA			GLUT_RGB
-#define GLUT_INDEX			1
-#define GLUT_SINGLE			0
-#define GLUT_DOUBLE			2
-#define GLUT_ACCUM			4
-#define GLUT_ALPHA			8
-#define GLUT_DEPTH			16
-#define GLUT_STENCIL        32
-
-/* Mouse buttons. */
-#define GLUT_LEFT_BUTTON    0
-#define GLUT_MIDDLE_BUTTON  1
-#define GLUT_RIGHT_BUTTON   2
-
-/* Mouse button  state. */
-#define GLUT_DOWN			0
-#define GLUT_UP				1
-
-/* function keys */
-#define GLUT_KEY_F1			1
-#define GLUT_KEY_F2			2
-#define GLUT_KEY_F3			3
-#define GLUT_KEY_F4			4
-#define GLUT_KEY_F5			5
-#define GLUT_KEY_F6			6
-#define GLUT_KEY_F7			7
-#define GLUT_KEY_F8			8
-#define GLUT_KEY_F9			9
-#define GLUT_KEY_F10        10
-#define GLUT_KEY_F11        11
-#define GLUT_KEY_F12        12
-
-/* directional keys */
-#define GLUT_KEY_LEFT       100
-#define GLUT_KEY_UP			101
-#define GLUT_KEY_RIGHT      102
-#define GLUT_KEY_DOWN       103
-#define GLUT_KEY_PAGE_UP    104
-#define GLUT_KEY_PAGE_DOWN  105
-#define GLUT_KEY_HOME       106
-#define GLUT_KEY_END        107
-#define GLUT_KEY_INSERT     108
-
-/* Entry/exit  state. */
-#define GLUT_LEFT			0
-#define GLUT_ENTERED        1
-
-/* Visibility  state. */
-#define GLUT_NOT_VISIBLE    0
-#define GLUT_VISIBLE        1
-
-/* Color index component selection values. */
-#define GLUT_RED			0
-#define GLUT_GREEN			1
-#define GLUT_BLUE			2
-
-/* Layers for use. */
-#define GLUT_NORMAL			0
-#define GLUT_OVERLAY        1
-
-/* Stroke font constants (use these in GLUT program). */
-#define GLUT_STROKE_ROMAN           ((void*)0)
-#define GLUT_STROKE_MONO_ROMAN		((void*)1)
-
-/* Bitmap font constants (use these in GLUT program). */
-#define GLUT_BITMAP_9_BY_15         ((void*)2)
-#define GLUT_BITMAP_8_BY_13         ((void*)3)
-#define GLUT_BITMAP_TIMES_ROMAN_10	((void*)4)
-#define GLUT_BITMAP_TIMES_ROMAN_24	((void*)5)
-#define GLUT_BITMAP_HELVETICA_10	((void*)6)
-#define GLUT_BITMAP_HELVETICA_12	((void*)7)
-#define GLUT_BITMAP_HELVETICA_18    ((void*)8)
-
-/* glutGet parameters. */
-#define GLUT_WINDOW_X                   100
-#define GLUT_WINDOW_Y                   101
-#define GLUT_WINDOW_WIDTH               102
-#define GLUT_WINDOW_HEIGHT              103
-#define GLUT_WINDOW_BUFFER_SIZE         104
-#define GLUT_WINDOW_STENCIL_SIZE        105
-#define GLUT_WINDOW_DEPTH_SIZE          106
-#define GLUT_WINDOW_RED_SIZE            107
-#define GLUT_WINDOW_GREEN_SIZE          108
-#define GLUT_WINDOW_BLUE_SIZE           109
-#define GLUT_WINDOW_ALPHA_SIZE          110
-#define GLUT_WINDOW_ACCUM_RED_SIZE      111
-#define GLUT_WINDOW_ACCUM_GREEN_SIZE	112
-#define GLUT_WINDOW_ACCUM_BLUE_SIZE     113
-#define GLUT_WINDOW_ACCUM_ALPHA_SIZE	114
-#define GLUT_WINDOW_DOUBLEBUFFER        115
-#define GLUT_WINDOW_RGBA                116
-#define GLUT_WINDOW_PARENT              117
-#define GLUT_WINDOW_NUM_CHILDREN        118
-#define GLUT_WINDOW_COLORMAP_SIZE       119
-#define GLUT_WINDOW_NUM_SAMPLES         120
-#define GLUT_WINDOW_STEREO              121
-#define GLUT_WINDOW_CURSOR              122
-#define GLUT_SCREEN_WIDTH               200
-#define GLUT_SCREEN_HEIGHT              201
-#define GLUT_SCREEN_WIDTH_MM            202
-#define GLUT_SCREEN_HEIGHT_MM           203
-#define GLUT_MENU_NUM_ITEMS             300
-#define GLUT_DISPLAY_MODE_POSSIBLE      400
-#define GLUT_INIT_WINDOW_X              500
-#define GLUT_INIT_WINDOW_Y              501
-#define GLUT_INIT_WINDOW_WIDTH          502
-#define GLUT_INIT_WINDOW_HEIGHT         503
-#define GLUT_INIT_DISPLAY_MODE          504
-#define GLUT_ELAPSED_TIME               700
-#define GLUT_WINDOW_FORMAT_ID           123
-
-/* glutDeviceGet parameters. */
-#define GLUT_HAS_KEYBOARD               600
-#define GLUT_HAS_MOUSE                  601
-#define GLUT_HAS_SPACEBALL              602
-#define GLUT_HAS_DIAL_AND_BUTTON_BOX	603
-#define GLUT_HAS_TABLET                 604
-#define GLUT_NUM_MOUSE_BUTTONS          605
-#define GLUT_NUM_SPACEBALL_BUTTONS      606
-#define GLUT_NUM_BUTTON_BOX_BUTTONS     607
-#define GLUT_NUM_DIALS                  608
-#define GLUT_NUM_TABLET_BUTTONS         609
-#define GLUT_DEVICE_IGNORE_KEY_REPEAT   610
-#define GLUT_DEVICE_KEY_REPEAT          611
-#define GLUT_HAS_JOYSTICK               612
-#define GLUT_OWNS_JOYSTICK              613
-#define GLUT_JOYSTICK_BUTTONS           614
-#define GLUT_JOYSTICK_AXES              615
-#define GLUT_JOYSTICK_POLL_RATE         616
-
-/* glutLayerGet parameters. */
-#define GLUT_OVERLAY_POSSIBLE           800
-#define GLUT_LAYER_IN_USE               801
-#define GLUT_HAS_OVERLAY                802
-#define GLUT_TRANSPARENT_INDEX          803
-#define GLUT_NORMAL_DAMAGED             804
-#define GLUT_OVERLAY_DAMAGED            805
-
-/* glutVideoResizeGet parameters. */
-#define GLUT_VIDEO_RESIZE_POSSIBLE      900
-#define GLUT_VIDEO_RESIZE_IN_USE        901
-#define GLUT_VIDEO_RESIZE_X_DELTA       902
-#define GLUT_VIDEO_RESIZE_Y_DELTA       903
-#define GLUT_VIDEO_RESIZE_WIDTH_DELTA	904
-#define GLUT_VIDEO_RESIZE_HEIGHT_DELTA	905
-#define GLUT_VIDEO_RESIZE_X             906
-#define GLUT_VIDEO_RESIZE_Y             907
-#define GLUT_VIDEO_RESIZE_WIDTH         908
-#define GLUT_VIDEO_RESIZE_HEIGHT        909
-
-/* glutUseLayer parameters. */
-#define GLUT_NORMAL                     0
-#define GLUT_OVERLAY                    1
-
-/* glutGetModifiers return mask. */
-#define GLUT_ACTIVE_SHIFT               1
-#define GLUT_ACTIVE_CTRL                2
-#define GLUT_ACTIVE_ALT                 4
-
-/* glutSetCursor parameters. */
-/* Basic arrows. */
-#define GLUT_CURSOR_RIGHT_ARROW         0
-#define GLUT_CURSOR_LEFT_ARROW          1
-/* Symbolic cursor shapes. */
-#define GLUT_CURSOR_INFO                2
-#define GLUT_CURSOR_DESTROY             3
-#define GLUT_CURSOR_HELP                4
-#define GLUT_CURSOR_CYCLE               5
-#define GLUT_CURSOR_SPRAY               6
-#define GLUT_CURSOR_WAIT                7
-#define GLUT_CURSOR_TEXT                8
-#define GLUT_CURSOR_CROSSHAIR           9
-/* Directional cursors. */
-#define GLUT_CURSOR_UP_DOWN             10
-#define GLUT_CURSOR_LEFT_RIGHT          11
-/* Sizing cursors. */
-#define GLUT_CURSOR_TOP_SIDE            12
-#define GLUT_CURSOR_BOTTOM_SIDE         13
-#define GLUT_CURSOR_LEFT_SIDE           14
-#define GLUT_CURSOR_RIGHT_SIDE          15
-#define GLUT_CURSOR_TOP_LEFT_CORNER     16
-#define GLUT_CURSOR_TOP_RIGHT_CORNER	17
-#define GLUT_CURSOR_BOTTOM_RIGHT_CORNER	18
-#define GLUT_CURSOR_BOTTOM_LEFT_CORNER	19
-/* Inherit from parent window. */
-#define GLUT_CURSOR_INHERIT             100
-/* Blank cursor. */
-#define GLUT_CURSOR_NONE                101
-/* Fullscreen crosshair (if available). */
-#define GLUT_CURSOR_FULL_CROSSHAIR      102
-
-/* GLUT initialization sub-API. */
-GLUTAPI void APIENTRY glutInit(int *argcp, char **argv);
-GLUTAPI void APIENTRY glutInitDisplayMode(unsigned int mode);
-GLUTAPI void APIENTRY glutInitWindowPosition(int x, int y);
-GLUTAPI void APIENTRY glutInitWindowSize(int width, int height);
-GLUTAPI void APIENTRY glutMainLoop(void);
-
-/* GLUT window sub-API. */
-GLUTAPI int APIENTRY glutCreateWindow(const char *title);
-GLUTAPI int APIENTRY glutCreateSubWindow(int win, int x, int y, int width, int height);
-GLUTAPI void APIENTRY glutDestroyWindow(int win);
-GLUTAPI void APIENTRY glutPostRedisplay(void);
-GLUTAPI void APIENTRY glutSwapBuffers(void);
-GLUTAPI int APIENTRY glutGetWindow(void);
-GLUTAPI void APIENTRY glutSetWindow(int win);
-GLUTAPI void APIENTRY glutSetWindowTitle(const char *title);
-GLUTAPI void APIENTRY glutSetIconTitle(const char *title);
-GLUTAPI void APIENTRY glutPositionWindow(int x, int y);
-GLUTAPI void APIENTRY glutReshapeWindow(int width, int height);
-GLUTAPI void APIENTRY glutPopWindow(void);
-GLUTAPI void APIENTRY glutPushWindow(void);
-GLUTAPI void APIENTRY glutIconifyWindow(void);
-GLUTAPI void APIENTRY glutShowWindow(void);
-GLUTAPI void APIENTRY glutHideWindow(void);
-
-/* GLUT overlay sub-API. */
-GLUTAPI void APIENTRY glutEstablishOverlay(void);
-GLUTAPI void APIENTRY glutRemoveOverlay(void);
-GLUTAPI void APIENTRY glutUseLayer(GLenum layer);
-GLUTAPI void APIENTRY glutPostOverlayRedisplay(void);
-GLUTAPI void APIENTRY glutShowOverlay(void);
-GLUTAPI void APIENTRY glutHideOverlay(void);
-
-/* GLUT menu sub-API. */
-GLUTAPI int APIENTRY glutCreateMenu(void (GLUTCALLBACK *)(int));
-GLUTAPI void APIENTRY glutDestroyMenu(int menu);
-GLUTAPI int APIENTRY glutGetMenu(void);
-GLUTAPI void APIENTRY glutSetMenu(int menu);
-GLUTAPI void APIENTRY glutAddMenuEntry(const char *label, int value);
-GLUTAPI void APIENTRY glutAddSubMenu(const char *label, int submenu);
-GLUTAPI void APIENTRY glutChangeToMenuEntry(int item, const char *label, int value);
-GLUTAPI void APIENTRY glutChangeToSubMenu(int item, const char *label, int submenu);
-GLUTAPI void APIENTRY glutRemoveMenuItem(int item);
-GLUTAPI void APIENTRY glutAttachMenu(int button);
-GLUTAPI void APIENTRY glutDetachMenu(int button);
-
-/* GLUT window callback sub-API. */
-GLUTAPI void APIENTRY glutDisplayFunc(void (GLUTCALLBACK * func)(void));
-GLUTAPI void APIENTRY glutReshapeFunc(void (GLUTCALLBACK * func)(int width, int height));
-GLUTAPI void APIENTRY glutKeyboardFunc(void (GLUTCALLBACK * func)(unsigned char key, int x, int y));
-GLUTAPI void APIENTRY glutMouseFunc(void (GLUTCALLBACK * func)(int button, int state, int x, int y));
-GLUTAPI void APIENTRY glutMotionFunc(void (GLUTCALLBACK * func)(int x, int y));
-GLUTAPI void APIENTRY glutPassiveMotionFunc(void (GLUTCALLBACK * func)(int x, int y));
-GLUTAPI void APIENTRY glutEntryFunc(void (GLUTCALLBACK * func)(int state));
-GLUTAPI void APIENTRY glutVisibilityFunc(void (GLUTCALLBACK * func)(int state));
-GLUTAPI void APIENTRY glutIdleFunc(void (GLUTCALLBACK * func)(void));
-GLUTAPI void APIENTRY glutTimerFunc(unsigned int millis, void (GLUTCALLBACK * func)(int value), int value);
-GLUTAPI void APIENTRY glutMenuStateFunc(void (GLUTCALLBACK * func)(int state));
-GLUTAPI void APIENTRY glutSpecialFunc(void (GLUTCALLBACK * func)(int key, int x, int y));
-GLUTAPI void APIENTRY glutSpaceballMotionFunc(void (GLUTCALLBACK * func)(int x, int y, int z));
-GLUTAPI void APIENTRY glutSpaceballRotateFunc(void (GLUTCALLBACK * func)(int x, int y, int z));
-GLUTAPI void APIENTRY glutSpaceballButtonFunc(void (GLUTCALLBACK * func)(int button, int state));
-GLUTAPI void APIENTRY glutButtonBoxFunc(void (GLUTCALLBACK * func)(int button, int state));
-GLUTAPI void APIENTRY glutDialsFunc(void (GLUTCALLBACK * func)(int dial, int value));
-GLUTAPI void APIENTRY glutTabletMotionFunc(void (GLUTCALLBACK * func)(int x, int y));
-GLUTAPI void APIENTRY glutTabletButtonFunc(void (GLUTCALLBACK * func)(int button, int state, int x, int y));
-GLUTAPI void APIENTRY glutMenuStatusFunc(void (GLUTCALLBACK * func)(int status, int x, int y));
-GLUTAPI void APIENTRY glutOverlayDisplayFunc(void (GLUTCALLBACK * func)(void));
-GLUTAPI void APIENTRY glutWindowStatusFunc(void (GLUTCALLBACK * func)(int state));
-
-/* GLUT color index sub-API. */
-GLUTAPI void APIENTRY glutSetColor(int, GLfloat red, GLfloat green, GLfloat blue);
-GLUTAPI GLfloat APIENTRY glutGetColor(int ndx, int component);
-GLUTAPI void APIENTRY glutCopyColormap(int win);
-
-/* GLUT state retrieval sub-API. */
-GLUTAPI int APIENTRY glutGet(GLenum type);
-GLUTAPI int APIENTRY glutDeviceGet(GLenum type);
-
-/* GLUT font sub-API */
-GLUTAPI void APIENTRY glutBitmapCharacter(void *font, int character);
-GLUTAPI int APIENTRY glutBitmapWidth(void *font, int character);
-GLUTAPI void APIENTRY glutStrokeCharacter(void *font, int character);
-GLUTAPI int APIENTRY glutStrokeWidth(void *font, int character);
-
-/* GLUT pre-built models sub-API */
-GLUTAPI void APIENTRY glutWireSphere(GLdouble radius, GLint slices, GLint stacks);
-GLUTAPI void APIENTRY glutSolidSphere(GLdouble radius, GLint slices, GLint stacks);
-GLUTAPI void APIENTRY glutWireCone(GLdouble base, GLdouble height, GLint slices, GLint stacks);
-GLUTAPI void APIENTRY glutSolidCone(GLdouble base, GLdouble height, GLint slices, GLint stacks);
-GLUTAPI void APIENTRY glutWireCube(GLdouble size);
-GLUTAPI void APIENTRY glutSolidCube(GLdouble size);
-GLUTAPI void APIENTRY glutWireTorus(GLdouble innerRadius, GLdouble outerRadius, GLint sides, GLint rings);
-GLUTAPI void APIENTRY glutSolidTorus(GLdouble innerRadius, GLdouble outerRadius, GLint sides, GLint rings);
-GLUTAPI void APIENTRY glutWireDodecahedron(void);
-GLUTAPI void APIENTRY glutSolidDodecahedron(void);
-GLUTAPI void APIENTRY glutWireTeapot(GLdouble size);
-GLUTAPI void APIENTRY glutSolidTeapot(GLdouble size);
-GLUTAPI void APIENTRY glutWireOctahedron(void);
-GLUTAPI void APIENTRY glutSolidOctahedron(void);
-GLUTAPI void APIENTRY glutWireTetrahedron(void);
-GLUTAPI void APIENTRY glutSolidTetrahedron(void);
-GLUTAPI void APIENTRY glutWireIcosahedron(void);
-GLUTAPI void APIENTRY glutSolidIcosahedron(void);
-
-#endif /* __AGLUT_H__ */
diff --git a/include/GL/mglmesa.h b/include/GL/mglmesa.h
deleted file mode 100644
index 0f9f7895714..00000000000
--- a/include/GL/mglmesa.h
+++ /dev/null
@@ -1,79 +0,0 @@
-/****************************************************************************
-*
-*                      Mesa bindings for SciTech MGL
-*
-*                   Copyright (C) 1996 SciTech Software.
-*                           All rights reserved.
-*
-* Filename:     mglmesa.h
-* Version:      Revision: 1.1.1.1
-*
-* Language:     ANSI C
-* Environment:  Any
-*
-* Description:  Header file for the Mesa/OpenGL interface bindings for the
-*               SciTech MGL graphics library. Uses the MGL internal
-*               device context structures to get direct access to the
-*               high performance MGL rasterization functions for maximum
-*               performance. Utilizes the VESA VBE/AF Accelerator Functions
-*               via the MGL's accelerated device driver functions, as well
-*               as basic DirectDraw accelerated functions provided by the
-*               MGL.
-*
-* This library is free software; you can redistribute it and/or
-* modify it under the terms of the GNU Library 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
-* Library General Public License for more details.
-*
-* You should have received a copy of the GNU Library General Public
-* License along with this library; if not, write to the Free
-* Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
-*
-*
-****************************************************************************/
-
-#ifndef __MGLMESA_H
-#define __MGLMESA_H
-
-#include "mgraph.h"
-
-/*------------------------- Function Prototypes ---------------------------*/
-
-#ifdef  __cplusplus
-extern "C" {            /* Use "C" linkage when in C++ mode */
-#endif
-
-#ifndef __WINDOWS__
-#define GLAPIENTRY
-#endif
-
-#ifdef  __WINDOWS__
-bool    GLAPIENTRY MGLMesaInitDLL(MGLCallbacks *cb,char *version);
-#endif
-void    GLAPIENTRY MGLMesaChooseVisual(MGLDC *dc,MGLVisual *visual);
-bool    GLAPIENTRY MGLMesaSetVisual(MGLDC *dc,MGLVisual *visual);
-bool    GLAPIENTRY MGLMesaCreateContext(MGLDC *dc,bool forceMemDC);
-void    GLAPIENTRY MGLMesaDestroyContext(MGLDC *dc);
-void    GLAPIENTRY MGLMesaMakeCurrent(MGLDC *dc);
-void    GLAPIENTRY MGLMesaSwapBuffers(MGLDC *dc,bool waitVRT);
-
-/* Palette manipulation support. The reason we provide palette manipulation
- * routines is so that when rendering in double buffered modes with a
- * software backbuffer, the palette for the backbuffer is kept consistent
- * with the hardware front buffer.
- */
-
-void    GLAPIENTRY MGLMesaSetPaletteEntry(MGLDC *dc,int entry,uchar red,uchar green,uchar blue);
-void    GLAPIENTRY MGLMesaSetPalette(MGLDC *dc,palette_t *pal,int numColors,int startIndex);
-void    GLAPIENTRY MGLMesaRealizePalette(MGLDC *dc,int numColors,int startIndex,int waitVRT);
-
-#ifdef  __cplusplus
-}                       /* End of "C" linkage for C++   */
-#endif  /* __cplusplus */
-
-#endif  /* __MGLMESA_H */
diff --git a/progs/gallium/python/retrace/interpreter.py b/progs/gallium/python/retrace/interpreter.py
index b30469dfaee..1a961812553 100755
--- a/progs/gallium/python/retrace/interpreter.py
+++ b/progs/gallium/python/retrace/interpreter.py
@@ -551,7 +551,6 @@ class Context(Object):
                 data = vbuf.buffer.read()
                 values = unpack_from(format, data, offset)
                 sys.stdout.write('\t\t{' + ', '.join(map(str, values)) + '},\n')
-                assert len(values) == velem.nr_components
             sys.stdout.write('\t},\n')
         sys.stdout.flush()
 
diff --git a/progs/gallium/unit/u_format_test.c b/progs/gallium/unit/u_format_test.c
index 2dfba5538bf..5274311e035 100644
--- a/progs/gallium/unit/u_format_test.c
+++ b/progs/gallium/unit/u_format_test.c
@@ -30,546 +30,10 @@
 #include <stdio.h>
 
 #include "util/u_format.h"
+#include "util/u_format_tests.h"
 #include "util/u_format_pack.h"
 
 
-#define MAX_PACKED_BYTES 16
-
-
-/**
- * A (packed, unpacked) color pair.
- */
-struct util_format_test_case
-{
-   enum pipe_format format;
-
-   /**
-    * Mask of the bits that actually meaningful data. Used to mask out the
-    * "X" channels.
-    */
-   uint8_t mask[MAX_PACKED_BYTES];
-
-   uint8_t packed[MAX_PACKED_BYTES];
-
-   /**
-    * RGBA.
-    */
-   double unpacked[4];
-};
-
-
-/*
- * Helper macros to create the packed bytes for longer words.
- */
-
-#define PACKED_1x8(x)          {x, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_2x8(x, y)       {x, y, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_3x8(x, y, z)    {x, y, z, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_4x8(x, y, z, w) {x, y, z, w, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
-
-#define PACKED_1x16(x)          {(x) & 0xff, (x) >> 8,          0,        0,          0,        0,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_2x16(x, y)       {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8,          0,        0,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_3x16(x, y, z)    {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8, (z) & 0xff, (z) >> 8,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
-#define PACKED_4x16(x, y, z, w) {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8, (z) & 0xff, (z) >> 8, (w) & 0xff, (w) >> 8, 0, 0, 0, 0, 0, 0, 0, 0}
-
-#define PACKED_1x32(x)          {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24,          0,                 0,                  0,         0,          0,                 0,                  0,         0,          0,                 0,                  0,         0}
-#define PACKED_2x32(x, y)       {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24,          0,                 0,                  0,         0,          0,                 0,                  0,         0}
-#define PACKED_3x32(x, y, z)    {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24, (z) & 0xff, ((z) >> 8) & 0xff, ((z) >> 16) & 0xff, (z) >> 24,          0,                 0,                  0,         0}
-#define PACKED_4x32(x, y, z, w) {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24, (z) & 0xff, ((z) >> 8) & 0xff, ((z) >> 16) & 0xff, (z) >> 24, (w) & 0xff, ((w) >> 8) & 0xff, ((w) >> 16) & 0xff, (w) >> 24}
-
-
-/**
- * Test cases.
- *
- * These were manually entered. We could generate these
- *
- * To keep this to a we cover only the corner cases, which should produce
- * good enough coverage since that pixel format transformations are afine for
- * non SRGB formats.
- */
-static const struct util_format_test_case
-test_cases[] =
-{
-
-   /*
-    * 32-bit rendertarget formats
-    */
-
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x0000ff00), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00ff0000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x0000ff00), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xff000000), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x0000ff00), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xff000000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000003ff), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000ffc00), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x3ff00000), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xc0000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   /*
-    * 16-bit rendertarget formats
-    */
-
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x03e0), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x7c00), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x8000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x000f), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x00f0), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0f00), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xf000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x07e0), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xf800), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   /*
-    * Luminance/intensity/alpha formats
-    */
-
-   {PIPE_FORMAT_L8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_L8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_A8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {0.0, 0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_I8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_I8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x00ff), {1.0, 1.0, 1.0, 0.0}},
-   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xff00), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_L16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_L16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   /*
-    * TODO: SRGB formats
-    */
-
-   /*
-    * Mixed-signed formats
-    */
-
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00000000), { 0.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x0000007f), { 1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00000081), {-1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00007f00), { 0.0,  1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00008100), { 0.0, -1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00ff0000), { 0.0,  0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0xff000000), { 0.0,  0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), { 0.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x000f), { 1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0011), {-1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x01e0), { 0.0,  1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0220), { 0.0, -1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0xfc00), { 0.0,  0.0, 1.0, 1.0}},
-
-   /*
-    * TODO: Depth-stencil formats
-    */
-
-   /*
-    * TODO: YUV formats
-    */
-
-   /*
-    * TODO: Compressed formats
-    */
-
-   /*
-    * Standard 8-bit integer formats
-    */
-
-   {PIPE_FORMAT_R8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0x00), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0xff), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0xff), {1.0, 1.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0x00, 0x00), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0xff, 0x00), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0xff), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0xff, 0xff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0x00, 0x00, 0x00), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0xff, 0x00, 0x00), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0xff, 0x00), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0xff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0xff, 0xff, 0xff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R8_USCALED, PACKED_1x8(0xff), PACKED_1x8(0x00), {  0.0,   0.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8_USCALED, PACKED_1x8(0xff), PACKED_1x8(0xff), {255.0,   0.0,   0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {  0.0,   0.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0x00), {255.0,   0.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0xff), {  0.0, 255.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0xff), {255.0, 255.0,   0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {  0.0,   0.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0x00, 0x00), {255.0,   0.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0xff, 0x00), {  0.0, 255.0,   0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0xff), {  0.0,   0.0, 255.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0xff, 0xff), {255.0, 255.0, 255.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {  0.0,   0.0,   0.0,   0.0}},
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0x00, 0x00, 0x00), {255.0,   0.0,   0.0,   0.0}},
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0xff, 0x00, 0x00), {  0.0, 255.0,   0.0,   0.0}},
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0xff, 0x00), {  0.0,   0.0, 255.0,   0.0}},
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0xff), {  0.0,   0.0,   0.0, 255.0}},
-   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0xff, 0xff, 0xff), {255.0, 255.0, 255.0, 255.0}},
-
-   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), { 0.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x7f), { 1.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x81), {-1.0,  0.0,  0.0,  1.0}},
-
-   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), { 0.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x7f, 0x00), { 1.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x81, 0x00), {-1.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x7f), { 0.0,  1.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x81), { 0.0, -1.0,  0.0,  1.0}},
-
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), { 0.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x7f, 0x00, 0x00), { 1.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x81, 0x00, 0x00), {-1.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x7f, 0x00), { 0.0,  1.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x81, 0x00), { 0.0, -1.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x7f), { 0.0,  0.0,  1.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x81), { 0.0,  0.0, -1.0,  1.0}},
-
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), { 0.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x7f, 0x00, 0x00, 0x00), { 1.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x81, 0x00, 0x00, 0x00), {-1.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x7f, 0x00, 0x00), { 0.0,  1.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x81, 0x00, 0x00), { 0.0, -1.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x7f, 0x00), { 0.0,  0.0,  1.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x81, 0x00), { 0.0,  0.0, -1.0,  0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x7f), { 0.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x81), { 0.0,  0.0,  0.0, -1.0}},
-
-   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x00), {   0.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x7f), { 127.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x80), {-128.0,    0.0,    0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {   0.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x7f, 0x00), { 127.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x80, 0x00), {-128.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x7f), {   0.0,  127.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x80), {   0.0, -128.0,    0.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {   0.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x7f, 0x00, 0x00), { 127.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x80, 0x00, 0x00), {-128.0,    0.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x7f, 0x00), {   0.0,  127.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x80, 0x00), {   0.0, -128.0,    0.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x7f), {   0.0,    0.0,  127.0, 1.0}},
-   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x80), {   0.0,    0.0, -128.0, 1.0}},
-
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {   0.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x7f, 0x00, 0x00, 0x00), { 127.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x80, 0x00, 0x00, 0x00), {-128.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x7f, 0x00, 0x00), {   0.0,  127.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x80, 0x00, 0x00), {   0.0, -128.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x7f, 0x00), {   0.0,    0.0,  127.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x80, 0x00), {   0.0,    0.0, -128.0,    0.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x7f), {   0.0,    0.0,    0.0,  127.0}},
-   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x80), {   0.0,    0.0,    0.0, -128.0}},
-
-   /*
-    * Standard 16-bit integer formats
-    */
-
-   {PIPE_FORMAT_R16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0x0000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0xffff), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0xffff), {1.0, 1.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0x0000, 0x0000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0xffff, 0x0000), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0xffff), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0xffff, 0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0x0000, 0x0000, 0x0000), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0xffff, 0x0000, 0x0000), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0xffff, 0x0000), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0xffff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R16_USCALED, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {    0.0,     0.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16_USCALED, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {65535.0,     0.0,     0.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {    0.0,     0.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0x0000), {65535.0,     0.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0xffff), {    0.0, 65535.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0xffff), {65535.0, 65535.0,     0.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {    0.0,     0.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0x0000, 0x0000), {65535.0,     0.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0xffff, 0x0000), {    0.0, 65535.0,     0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0xffff), {    0.0,     0.0, 65535.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0xffff, 0xffff), {65535.0, 65535.0, 65535.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {    0.0,     0.0,     0.0,     0.0}},
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0x0000, 0x0000, 0x0000), {65535.0,     0.0,     0.0,     0.0}},
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0xffff, 0x0000, 0x0000), {    0.0, 65535.0,     0.0,     0.0}},
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0xffff, 0x0000), {    0.0,     0.0, 65535.0,     0.0}},
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0xffff), {    0.0,     0.0,     0.0, 65535.0}},
-   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), {65535.0, 65535.0, 65535.0, 65535.0}},
-
-   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x7fff), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x8001), {  -1.0,    0.0,    0.0,    1.0}},
-
-   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x7fff, 0x0000), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x8001, 0x0000), {  -1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x7fff), {   0.0,    1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x8001), {   0.0,   -1.0,    0.0,    1.0}},
-
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x7fff, 0x0000, 0x0000), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x8001, 0x0000, 0x0000), {  -1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x7fff, 0x0000), {   0.0,    1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x8001, 0x0000), {   0.0,   -1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x7fff), {   0.0,    0.0,    1.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x8001), {   0.0,    0.0,   -1.0,    1.0}},
-
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {   0.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x7fff, 0x0000, 0x0000, 0x0000), {   1.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x8001, 0x0000, 0x0000, 0x0000), {  -1.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x7fff, 0x0000, 0x0000), {   0.0,    1.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x8001, 0x0000, 0x0000), {   0.0,   -1.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x7fff, 0x0000), {   0.0,    0.0,    1.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x8001, 0x0000), {   0.0,    0.0,   -1.0,    0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x7fff), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x8001), {   0.0,    0.0,    0.0,   -1.0}},
-
-   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {     0.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x7fff), { 32767.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x8000), {-32768.0,      0.0,      0.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {     0.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x7fff, 0x0000), { 32767.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x8000, 0x0000), {-32768.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x7fff), {     0.0,  32767.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x8000), {     0.0, -32768.0,      0.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {     0.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x7fff, 0x0000, 0x0000), { 32767.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x8000, 0x0000, 0x0000), {-32768.0,      0.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x7fff, 0x0000), {     0.0,  32767.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x8000, 0x0000), {     0.0, -32768.0,      0.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x7fff), {     0.0,      0.0,  32767.0,   1.0}},
-   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x8000), {     0.0,      0.0, -32768.0,   1.0}},
-
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {     0.0,      0.0,      0.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x7fff, 0x0000, 0x0000, 0x0000), { 32767.0,      0.0,      0.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x8000, 0x0000, 0x0000, 0x0000), {-32768.0,      0.0,      0.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x7fff, 0x0000, 0x0000), {     0.0,  32767.0,      0.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x8000, 0x0000, 0x0000), {     0.0, -32768.0,      0.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x7fff, 0x0000), {     0.0,      0.0,  32767.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x8000, 0x0000), {     0.0,      0.0, -32768.0,      0.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x7fff), {     0.0,      0.0,      0.0,  32767.0}},
-   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x8000), {     0.0,      0.0,      0.0, -32768.0}},
-
-   /*
-    * Standard 32-bit integer formats
-    *
-    * NOTE: We can't accurately represent integers larger than +/-0x1000000
-    * with single precision floats, so that's as far as we test.
-    */
-
-   {PIPE_FORMAT_R32_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xffffffff, 0x00000000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xffffffff), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xffffffff, 0xffffffff), {1.0, 1.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xffffffff, 0x00000000, 0x00000000), {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xffffffff, 0x00000000), {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xffffffff), {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xffffffff, 0x00000000, 0x00000000, 0x00000000), {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xffffffff, 0x00000000, 0x00000000), {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xffffffff, 0x00000000), {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xffffffff), {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_R32_USCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {       0.0,        0.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32_USCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x01000000), {16777216.0,        0.0,        0.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {       0.0,        0.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x00000000), {16777216.0,        0.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x01000000), {       0.0, 16777216.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x01000000), {16777216.0, 16777216.0,        0.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {       0.0,        0.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x00000000, 0x00000000), {16777216.0,        0.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x01000000, 0x00000000), {       0.0, 16777216.0,        0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x01000000), {       0.0,        0.0, 16777216.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x01000000, 0x01000000), {16777216.0, 16777216.0, 16777216.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {       0.0,        0.0,        0.0,        0.0}},
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x00000000, 0x00000000, 0x00000000), {16777216.0,        0.0,        0.0,        0.0}},
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x01000000, 0x00000000, 0x00000000), {       0.0, 16777216.0,        0.0,        0.0}},
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x01000000, 0x00000000), {       0.0,        0.0, 16777216.0,        0.0}},
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x01000000), {       0.0,        0.0,        0.0, 16777216.0}},
-   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x01000000, 0x01000000, 0x01000000), {16777216.0, 16777216.0, 16777216.0, 16777216.0}},
-
-   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x7fffffff), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x80000001), {  -1.0,    0.0,    0.0,    1.0}},
-
-   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x7fffffff, 0x00000000), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x80000001, 0x00000000), {  -1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x7fffffff), {   0.0,    1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x80000001), {   0.0,   -1.0,    0.0,    1.0}},
-
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x7fffffff, 0x00000000, 0x00000000), {   1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x80000001, 0x00000000, 0x00000000), {  -1.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x7fffffff, 0x00000000), {   0.0,    1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x80000001, 0x00000000), {   0.0,   -1.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x7fffffff), {   0.0,    0.0,    1.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x80000001), {   0.0,    0.0,   -1.0,    1.0}},
-
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x7fffffff, 0x00000000, 0x00000000, 0x00000000), {   1.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x80000001, 0x00000000, 0x00000000, 0x00000000), {  -1.0,    0.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x7fffffff, 0x00000000, 0x00000000), {   0.0,    1.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x80000001, 0x00000000, 0x00000000), {   0.0,   -1.0,    0.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x7fffffff, 0x00000000), {   0.0,    0.0,    1.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x80000001, 0x00000000), {   0.0,    0.0,   -1.0,    0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x7fffffff), {   0.0,    0.0,    0.0,    1.0}},
-   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x80000001), {   0.0,    0.0,    0.0,   -1.0}},
-
-   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {        0.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x01000000), { 16777216.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {-16777216.0,         0.0,         0.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {        0.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x00000000), { 16777216.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xff000000, 0x00000000), {-16777216.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x01000000), {        0.0,  16777216.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xff000000), {        0.0, -16777216.0,         0.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {        0.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x00000000, 0x00000000), { 16777216.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xff000000, 0x00000000, 0x00000000), {-16777216.0,         0.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x01000000, 0x00000000), {        0.0,  16777216.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xff000000, 0x00000000), {        0.0, -16777216.0,         0.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x01000000), {        0.0,         0.0,  16777216.0,   1.0}},
-   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xff000000), {        0.0,         0.0, -16777216.0,   1.0}},
-
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {        0.0,         0.0,         0.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x00000000, 0x00000000, 0x00000000), { 16777216.0,         0.0,         0.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xff000000, 0x00000000, 0x00000000, 0x00000000), {-16777216.0,         0.0,         0.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x01000000, 0x00000000, 0x00000000), {        0.0,  16777216.0,         0.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xff000000, 0x00000000, 0x00000000), {        0.0, -16777216.0,         0.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x01000000, 0x00000000), {        0.0,         0.0,  16777216.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xff000000, 0x00000000), {        0.0,         0.0, -16777216.0,         0.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x01000000), {        0.0,         0.0,         0.0,  16777216.0}},
-   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xff000000), {        0.0,         0.0,         0.0, -16777216.0}},
-
-   /*
-    * Standard 32-bit float formats
-    */
-
-   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {  0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0x3f800000), {  1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0xbf800000), { -1.0, 0.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), { 0.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x3f800000, 0x00000000), { 1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xbf800000, 0x00000000), {-1.0,  0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x3f800000), { 0.0,  1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xbf800000), { 0.0, -1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x3f800000, 0x3f800000), { 1.0,  1.0, 0.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), { 0.0,  0.0,  0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x3f800000, 0x00000000, 0x00000000), { 1.0,  0.0,  0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xbf800000, 0x00000000, 0x00000000), {-1.0,  0.0,  0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x3f800000, 0x00000000), { 0.0,  1.0,  0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xbf800000, 0x00000000), { 0.0, -1.0,  0.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x3f800000), { 0.0,  0.0,  1.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xbf800000), { 0.0,  0.0, -1.0, 1.0}},
-   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x3f800000, 0x3f800000, 0x3f800000), { 1.0,  1.0,  1.0, 1.0}},
-
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), { 0.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x3f800000, 0x00000000, 0x00000000, 0x00000000), { 1.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xbf800000, 0x00000000, 0x00000000, 0x00000000), {-1.0,  0.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x3f800000, 0x00000000, 0x00000000), { 0.0,  1.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xbf800000, 0x00000000, 0x00000000), { 0.0, -1.0,  0.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x3f800000, 0x00000000), { 0.0,  0.0,  1.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xbf800000, 0x00000000), { 0.0,  0.0, -1.0,  0.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x3f800000), { 0.0,  0.0,  0.0,  1.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xbf800000), { 0.0,  0.0,  0.0, -1.0}},
-   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000), { 1.0,  1.0,  1.0,  1.0}},
-};
-
-
 static boolean
 test_format_unpack_4f(const struct util_format_test_case *test)
 {
@@ -596,7 +60,7 @@ test_format_unpack_4f(const struct util_format_test_case *test)
 static boolean
 test_format_pack_4f(const struct util_format_test_case *test)
 {
-   uint8_t packed[MAX_PACKED_BYTES];
+   uint8_t packed[UTIL_FORMAT_MAX_PACKED_BYTES];
    unsigned i;
    boolean success;
 
@@ -605,7 +69,7 @@ test_format_pack_4f(const struct util_format_test_case *test)
    util_format_pack_4f(test->format, packed, test->unpacked[0], test->unpacked[1], test->unpacked[2], test->unpacked[3]);
 
    success = TRUE;
-   for (i = 0; i < MAX_PACKED_BYTES; ++i)
+   for (i = 0; i < UTIL_FORMAT_MAX_PACKED_BYTES; ++i)
       if ((test->packed[i] & test->mask[i]) != (packed[i] & test->mask[i]))
          success = FALSE;
 
@@ -627,13 +91,27 @@ test_format_pack_4f(const struct util_format_test_case *test)
 }
 
 
-static void
+static boolean
 convert_4f_to_4ub(uint8_t *dst, const double *src)
 {
    unsigned i;
+   boolean accurate = TRUE;
 
-   for (i = 0; i < 4; ++i)
-      dst[i] = CLAMP(src[i], 0.0, 1.0) * 255.0;
+   for (i = 0; i < 4; ++i) {
+      if (src[i] < 0.0) {
+         accurate = FALSE;
+         dst[i] = 0;
+      }
+      else if (src[i] > 1.0) {
+         accurate = FALSE;
+         dst[i] = 255;
+      }
+      else {
+         dst[i] = src[i] * 255.0;
+      }
+   }
+
+   return accurate;
 }
 
 
@@ -667,18 +145,23 @@ static boolean
 test_format_pack_4ub(const struct util_format_test_case *test)
 {
    uint8_t unpacked[4];
-   uint8_t packed[MAX_PACKED_BYTES];
+   uint8_t packed[UTIL_FORMAT_MAX_PACKED_BYTES];
    unsigned i;
    boolean success;
 
-   convert_4f_to_4ub(unpacked, test->unpacked);
+   if (!convert_4f_to_4ub(unpacked, test->unpacked)) {
+      /*
+       * Skip test cases which cannot be represented by four unorm bytes.
+       */
+      return TRUE;
+   }
 
    memset(packed, 0, sizeof packed);
 
    util_format_pack_4ub(test->format, packed, unpacked[0], unpacked[1], unpacked[2], unpacked[3]);
 
    success = TRUE;
-   for (i = 0; i < MAX_PACKED_BYTES; ++i)
+   for (i = 0; i < UTIL_FORMAT_MAX_PACKED_BYTES; ++i)
       if ((test->packed[i] & test->mask[i]) != (packed[i] & test->mask[i]))
          success = FALSE;
 
@@ -711,15 +194,16 @@ test_one(test_func_t func, const char *suffix)
    unsigned i;
    bool success = TRUE;
 
-   for (i = 0; i < sizeof(test_cases)/sizeof(test_cases[0]); ++i) {
-      if (test_cases[i].format != last_format) {
+   for (i = 0; i < util_format_nr_test_cases; ++i) {
+      const struct util_format_test_case *test = &util_format_test_cases[i];
+      if (test->format != last_format) {
          const struct util_format_description *format_desc;
-         format_desc = util_format_description(test_cases[i].format);
-         printf("Testing %s.%s ...\n", format_desc->name, suffix);
-         last_format = test_cases[i].format;
+         format_desc = util_format_description(test->format);
+         printf("Testing util_format_%s_%s ...\n", format_desc->short_name, suffix);
+         last_format = test->format;
       }
 
-      if (!func(&test_cases[i]))
+      if (!func(&util_format_test_cases[i]))
         success = FALSE;
    }
 
diff --git a/progs/tests/fbotest1.c b/progs/tests/fbotest1.c
index 0cd7f95c355..a95fdff74c3 100644
--- a/progs/tests/fbotest1.c
+++ b/progs/tests/fbotest1.c
@@ -36,8 +36,8 @@ Display( void )
 
    /* draw to user framebuffer */
    glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, MyFB);
-   glDrawBuffer(GL_COLOR_ATTACHMENT1_EXT);
-   glReadBuffer(GL_COLOR_ATTACHMENT1_EXT);
+   glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT);
+   glReadBuffer(GL_COLOR_ATTACHMENT0_EXT);
 
    status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT);
    if (status != GL_FRAMEBUFFER_COMPLETE_EXT) {
@@ -161,7 +161,7 @@ Init( void )
    assert(i == MyFB);
 
    CheckError(__LINE__);
-   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT1_EXT,
+   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT,
                                 GL_RENDERBUFFER_EXT, MyRB);
 
    glRenderbufferStorageEXT(GL_RENDERBUFFER_EXT, GL_RGB, Width, Height);
diff --git a/progs/tests/fbotest2.c b/progs/tests/fbotest2.c
index f9c506193f6..872b46279e6 100644
--- a/progs/tests/fbotest2.c
+++ b/progs/tests/fbotest2.c
@@ -40,8 +40,8 @@ Display( void )
 
    /* draw to user framebuffer */
    glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, MyFB);
-   glDrawBuffer(GL_COLOR_ATTACHMENT1_EXT);
-   glReadBuffer(GL_COLOR_ATTACHMENT1_EXT);
+   glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT);
+   glReadBuffer(GL_COLOR_ATTACHMENT0_EXT);
 
    status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT);
    if (status != GL_FRAMEBUFFER_COMPLETE_EXT) {
@@ -163,7 +163,7 @@ Init( void )
    glGenRenderbuffersEXT(1, &ColorRb);
    glBindRenderbufferEXT(GL_RENDERBUFFER_EXT, ColorRb);
    assert(glIsRenderbufferEXT(ColorRb));
-   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT1_EXT,
+   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT,
                                 GL_RENDERBUFFER_EXT, ColorRb);
    glRenderbufferStorageEXT(GL_RENDERBUFFER_EXT, GL_RGB, Width, Height);
 
diff --git a/progs/tests/fbotest3.c b/progs/tests/fbotest3.c
index 8e288b38b83..c176f82d2ba 100644
--- a/progs/tests/fbotest3.c
+++ b/progs/tests/fbotest3.c
@@ -50,8 +50,8 @@ Display( void )
 
    /* draw to user framebuffer */
    glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, MyFB);
-   glDrawBuffer(GL_COLOR_ATTACHMENT1_EXT);
-   glReadBuffer(GL_COLOR_ATTACHMENT1_EXT);
+   glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT);
+   glReadBuffer(GL_COLOR_ATTACHMENT0_EXT);
 
    status = glCheckFramebufferStatusEXT(GL_FRAMEBUFFER_EXT);
    if (status != GL_FRAMEBUFFER_COMPLETE_EXT) {
@@ -189,7 +189,7 @@ Init( void )
    glGenRenderbuffersEXT(1, &ColorRb);
    glBindRenderbufferEXT(GL_RENDERBUFFER_EXT, ColorRb);
    assert(glIsRenderbufferEXT(ColorRb));
-   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT1_EXT,
+   glFramebufferRenderbufferEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT,
                                 GL_RENDERBUFFER_EXT, ColorRb);
    glRenderbufferStorageEXT(GL_RENDERBUFFER_EXT, GL_RGB, Width, Height);
 
diff --git a/progs/xdemos/omlsync.c b/progs/xdemos/omlsync.c
index a2baf4ad720..061d6c68613 100644
--- a/progs/xdemos/omlsync.c
+++ b/progs/xdemos/omlsync.c
@@ -249,11 +249,13 @@ int main(int argc, char *argv[])
 		glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
 		glRectf(0, 0, width, height);
 
-		glXSwapBuffersMscOML(disp, winGL, 0, divisor, remainder);
-
-		if (wait_interval) {
+		if (!wait_interval)
+			glXSwapBuffersMscOML(disp, winGL, 0, divisor,
+					     remainder);
+		else {
 			glXWaitForMscOML(disp, winGL, msc + wait_interval,
-					 0, 0, &ust, &msc, &sbc);
+					 divisor, remainder, &ust, &msc, &sbc);
+			glXSwapBuffersMscOML(disp, winGL, 0, 0, 0);
 		}
 	}
 
diff --git a/scons/llvm.py b/scons/llvm.py
index 37c503ec986..c8e50c5916d 100644
--- a/scons/llvm.py
+++ b/scons/llvm.py
@@ -29,7 +29,9 @@ Tool-specific initialization for LLVM
 
 import os
 import os.path
+import re
 import sys
+import distutils.version
 
 import SCons.Errors
 import SCons.Util
@@ -60,71 +62,108 @@ def generate(env):
 
     if env['platform'] == 'windows':
         # XXX: There is no llvm-config on Windows, so assume a standard layout
-        if llvm_dir is not None:
-            env.Prepend(CPPPATH = [os.path.join(llvm_dir, 'include')])
-            env.AppendUnique(CPPDEFINES = [
-                '__STDC_LIMIT_MACROS', 
-                '__STDC_CONSTANT_MACROS',
-                'HAVE_STDINT_H',
+        if llvm_dir is None:
+            return
+
+        # Try to determine the LLVM version from llvm/Config/config.h
+        llvm_config = os.path.join(llvm_dir, 'include/llvm/Config/config.h')
+        if not os.path.exists(llvm_config):
+            print 'scons: could not find %s' % llvm_config
+            return
+        llvm_version_re = re.compile(r'^#define PACKAGE_VERSION "([^"]*)"')
+        llvm_version = None
+        for line in open(llvm_config, 'rt'):
+            mo = llvm_version_re.match(line)
+            if mo:
+                llvm_version = mo.group(1)
+                llvm_version = distutils.version.LooseVersion(llvm_version)
+                break
+        if llvm_version is None:
+            print 'scons: could not determine the LLVM version from %s' % llvm_config
+            return
+
+        if llvm_version >= distutils.version.LooseVersion('2.7'):
+            print 'scons: Ignoring unsupported LLVM version %s' % llvm_version
+            print 'scons: See http://www.llvm.org/bugs/show_bug.cgi?id=6429'
+            return
+
+        env.Prepend(CPPPATH = [os.path.join(llvm_dir, 'include')])
+        env.AppendUnique(CPPDEFINES = [
+            '__STDC_LIMIT_MACROS', 
+            '__STDC_CONSTANT_MACROS',
+            'HAVE_STDINT_H',
+        ])
+        env.Prepend(LIBPATH = [os.path.join(llvm_dir, 'lib')])
+        if llvm_version >= distutils.version.LooseVersion('2.7'):
+            # 2.7
+            env.Prepend(LIBS = [
+                'LLVMLinker', 'LLVMipo', 'LLVMInterpreter',
+                'LLVMInstrumentation', 'LLVMJIT', 'LLVMExecutionEngine',
+                'LLVMBitWriter', 'LLVMX86Disassembler', 'LLVMX86AsmParser',
+                'LLVMMCParser', 'LLVMX86AsmPrinter', 'LLVMX86CodeGen',
+                'LLVMSelectionDAG', 'LLVMX86Info', 'LLVMAsmPrinter',
+                'LLVMCodeGen', 'LLVMScalarOpts', 'LLVMInstCombine',
+                'LLVMTransformUtils', 'LLVMipa', 'LLVMAsmParser',
+                'LLVMArchive', 'LLVMBitReader', 'LLVMAnalysis', 'LLVMTarget',
+                'LLVMMC', 'LLVMCore', 'LLVMSupport', 'LLVMSystem',
             ])
-            env.Prepend(LIBPATH = [os.path.join(llvm_dir, 'lib')])
+        else:
+            # 2.6
             env.Prepend(LIBS = [
-                'LLVMX86AsmParser',
-                'LLVMX86AsmPrinter',
-                'LLVMX86CodeGen',
-                'LLVMX86Info',
-                'LLVMLinker',
-                'LLVMipo',
-                'LLVMInterpreter',
-                'LLVMInstrumentation',
-                'LLVMJIT',
-                'LLVMExecutionEngine',
-                'LLVMDebugger',
-                'LLVMBitWriter',
-                'LLVMAsmParser',
-                'LLVMArchive',
-                'LLVMBitReader',
-                'LLVMSelectionDAG',
-                'LLVMAsmPrinter',
-                'LLVMCodeGen',
-                'LLVMScalarOpts',
-                'LLVMTransformUtils',
-                'LLVMipa',
-                'LLVMAnalysis',
-                'LLVMTarget',
-                'LLVMMC',
-                'LLVMCore',
-                'LLVMSupport',
+                'LLVMX86AsmParser', 'LLVMX86AsmPrinter', 'LLVMX86CodeGen',
+                'LLVMX86Info', 'LLVMLinker', 'LLVMipo', 'LLVMInterpreter',
+                'LLVMInstrumentation', 'LLVMJIT', 'LLVMExecutionEngine',
+                'LLVMDebugger', 'LLVMBitWriter', 'LLVMAsmParser',
+                'LLVMArchive', 'LLVMBitReader', 'LLVMSelectionDAG',
+                'LLVMAsmPrinter', 'LLVMCodeGen', 'LLVMScalarOpts',
+                'LLVMTransformUtils', 'LLVMipa', 'LLVMAnalysis',
+                'LLVMTarget', 'LLVMMC', 'LLVMCore', 'LLVMSupport',
                 'LLVMSystem',
-                'imagehlp',
-                'psapi',
             ])
-            if env['msvc']:
-                # Some of the LLVM C headers use the inline keyword without
-                # defining it.
-                env.Append(CPPDEFINES = [('inline', '__inline')])
-                if env['debug']:
-                    # LLVM libraries are static, build with /MT, and they
-                    # automatically link agains LIBCMT. When we're doing a
-                    # debug build we'll be linking against LIBCMTD, so disable
-                    # that.
-                    env.Append(LINKFLAGS = ['/nodefaultlib:LIBCMT'])
-            env['LLVM_VERSION'] = '2.6'
-        return
+        env.Append(LIBS = [
+            'imagehlp',
+            'psapi',
+        ])
+        if env['msvc']:
+            # Some of the LLVM C headers use the inline keyword without
+            # defining it.
+            env.Append(CPPDEFINES = [('inline', '__inline')])
+            if env['debug']:
+                # LLVM libraries are static, build with /MT, and they
+                # automatically link agains LIBCMT. When we're doing a
+                # debug build we'll be linking against LIBCMTD, so disable
+                # that.
+                env.Append(LINKFLAGS = ['/nodefaultlib:LIBCMT'])
     elif env.Detect('llvm-config'):
-        version = env.backtick('llvm-config --version').rstrip()
+        llvm_version = env.backtick('llvm-config --version').rstrip()
+        llvm_version = distutils.version.LooseVersion(llvm_version)
+
+        if llvm_version >= distutils.version.LooseVersion('2.7'):
+            print 'scons: Ignoring unsupported LLVM version %s' % llvm_version
+            print 'scons: See http://www.llvm.org/bugs/show_bug.cgi?id=6429'
+            return
 
         try:
             env.ParseConfig('llvm-config --cppflags')
             env.ParseConfig('llvm-config --libs jit interpreter nativecodegen bitwriter')
             env.ParseConfig('llvm-config --ldflags')
         except OSError:
-            print 'llvm-config version %s failed' % version
+            print 'llvm-config version %s failed' % llvm_version
         else:
-            if env['platform'] == 'windows':
-                env.Append(LIBS = ['imagehlp', 'psapi'])
             env['LINK'] = env['CXX']
-            env['LLVM_VERSION'] = version
+    else:
+        return
+
+    assert llvm_version is not None
+
+    print 'scons: Found LLVM version %s' % llvm_version
+    env['LLVM_VERSION'] = llvm_version
+
+    # Define HAVE_LLVM macro with the major/minor version number (e.g., 0x0206 for 2.6)
+    llvm_version_major = int(llvm_version.version[0])
+    llvm_version_minor = int(llvm_version.version[1])
+    llvm_version_hex = '0x%02x%02x' % (llvm_version_major, llvm_version_minor)
+    env.Prepend(CPPDEFINES = [('HAVE_LLVM', llvm_version_hex)])
 
 def exists(env):
     return True
diff --git a/src/SConscript b/src/SConscript
index cd4896ada40..cf6db730da2 100644
--- a/src/SConscript
+++ b/src/SConscript
@@ -1,13 +1,12 @@
 Import('*')
 
 SConscript('glsl/SConscript')
-SConscript('gallium/SConscript')
 
 if 'mesa' in env['statetrackers']:
-	SConscript('mesa/SConscript')
+    SConscript('mesa/SConscript')
 
-SConscript('gallium/winsys/SConscript')
+SConscript('gallium/SConscript')
 
 if platform != 'embedded':
-	SConscript('glut/glx/SConscript')
-	SConscript('glew/SConscript')
+    SConscript('glut/glx/SConscript')
+    SConscript('glew/SConscript')
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index d56c5c84617..c833d83e65b 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -7,6 +7,9 @@ SConscript('auxiliary/SConscript')
 for driver in env['drivers']:
 	SConscript(os.path.join('drivers', driver, 'SConscript'))
 
+# Needed by some state trackers
+SConscript('winsys/null/SConscript')
+
 SConscript('state_trackers/python/SConscript')
 if platform != 'embedded':
 	SConscript('state_trackers/glx/xlib/SConscript')
@@ -15,3 +18,7 @@ if platform != 'embedded':
 
 if platform == 'windows':
 	SConscript('state_trackers/wgl/SConscript')
+
+SConscript('winsys/SConscript')
+
+SConscript('targets/SConscript')
diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index 1d0930e024b..0ac18426d97 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -107,6 +107,7 @@ C_SOURCES = \
 	util/u_draw_quad.c \
 	util/u_format_access.c \
 	util/u_format_table.c \
+	util/u_format_tests.c \
 	util/u_gen_mipmap.c \
 	util/u_handle_table.c \
 	util/u_hash_table.c \
@@ -130,7 +131,8 @@ C_SOURCES = \
 	vl/vl_mpeg12_mc_renderer.c \
 	vl/vl_compositor.c \
 	vl/vl_csc.c \
-	vl/vl_shader_build.c
+	vl/vl_shader_build.c \
+	target-helpers/wrap_screen.c
 
 GALLIVM_SOURCES = \
         gallivm/lp_bld_alpha.c \
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index f365c4bbdd6..2be16776fb8 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -7,6 +7,8 @@ env.Append(CPPPATH = [
     'util',
 ])
 
+env.Tool('udis86')
+
 env.CodeGenerate(
     target = 'indices/u_indices_gen.c', 
     script = 'indices/u_indices_gen.py', 
@@ -149,6 +151,7 @@ source = [
     'util/u_draw_quad.c',
     'util/u_format_access.c',
     'util/u_format_table.c',
+    'util/u_format_tests.c',
     'util/u_gen_mipmap.c',
     'util/u_handle_table.c',
     'util/u_hash.c',
@@ -172,6 +175,7 @@ source = [
     'vl/vl_compositor.c',
     'vl/vl_csc.c',
     'vl/vl_shader_build.c',
+    'target-helpers/wrap_screen.c',
 ]
 
 if drawllvm:
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c
index a6a07e72c2f..900c64df4b9 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.c
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.c
@@ -43,6 +43,7 @@ struct cso_cache {
    struct cso_hash *vs_hash;
    struct cso_hash *rasterizer_hash;
    struct cso_hash *sampler_hash;
+   struct cso_hash *velements_hash;
    int    max_size;
 
    cso_sanitize_callback sanitize_cb;
@@ -108,6 +109,9 @@ static struct cso_hash *_cso_hash_for_type(struct cso_cache *sc, enum cso_cache_
    case CSO_VERTEX_SHADER:
       hash = sc->vs_hash;
       break;
+   case CSO_VELEMENTS:
+      hash = sc->velements_hash;
+      break;
    }
 
    return hash;
@@ -161,6 +165,13 @@ static void delete_vs_state(void *state, void *data)
    FREE(state);
 }
 
+static void delete_velements(void *state, void *data)
+{
+   struct cso_velements *cso = (struct cso_velements *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
 
 static INLINE void delete_cso(void *state, enum cso_cache_type type)
 {
@@ -183,6 +194,9 @@ static INLINE void delete_cso(void *state, enum cso_cache_type type)
    case CSO_VERTEX_SHADER:
       delete_vs_state(state, 0);
       break;
+   case CSO_VELEMENTS:
+      delete_velements(state, 0);
+      break;
    default:
       assert(0);
       FREE(state);
@@ -294,6 +308,7 @@ struct cso_cache *cso_cache_create(void)
    sc->rasterizer_hash    = cso_hash_create();
    sc->fs_hash            = cso_hash_create();
    sc->vs_hash            = cso_hash_create();
+   sc->velements_hash     = cso_hash_create();
    sc->sanitize_cb        = sanitize_cb;
    sc->sanitize_data      = 0;
 
@@ -325,6 +340,9 @@ void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type,
    case CSO_VERTEX_SHADER:
       hash = sc->vs_hash;
       break;
+   case CSO_VELEMENTS:
+      hash = sc->velements_hash;
+      break;
    }
 
    iter = cso_hash_first_node(hash);
@@ -351,6 +369,7 @@ void cso_cache_delete(struct cso_cache *sc)
    cso_for_each_state(sc, CSO_VERTEX_SHADER, delete_vs_state, 0);
    cso_for_each_state(sc, CSO_RASTERIZER, delete_rasterizer_state, 0);
    cso_for_each_state(sc, CSO_SAMPLER, delete_sampler_state, 0);
+   cso_for_each_state(sc, CSO_VELEMENTS, delete_velements, 0);
 
    cso_hash_delete(sc->blend_hash);
    cso_hash_delete(sc->sampler_hash);
@@ -358,6 +377,7 @@ void cso_cache_delete(struct cso_cache *sc)
    cso_hash_delete(sc->rasterizer_hash);
    cso_hash_delete(sc->fs_hash);
    cso_hash_delete(sc->vs_hash);
+   cso_hash_delete(sc->velements_hash);
    FREE(sc);
 }
 
@@ -372,6 +392,7 @@ void cso_set_maximum_cache_size(struct cso_cache *sc, int number)
    sanitize_hash(sc, sc->vs_hash, CSO_VERTEX_SHADER, sc->max_size);
    sanitize_hash(sc, sc->rasterizer_hash, CSO_RASTERIZER, sc->max_size);
    sanitize_hash(sc, sc->sampler_hash, CSO_SAMPLER, sc->max_size);
+   sanitize_hash(sc, sc->velements_hash, CSO_VELEMENTS, sc->max_size);
 }
 
 int cso_maximum_cache_size(const struct cso_cache *sc)
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.h b/src/gallium/auxiliary/cso_cache/cso_cache.h
index eea60b940bb..fb09b83c623 100644
--- a/src/gallium/auxiliary/cso_cache/cso_cache.h
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.h
@@ -53,6 +53,7 @@
   * - rasterizer (old setup)
   * - sampler
   * - vertex shader
+  * - vertex elements
   *
   * Things that are not constant state objects include:
   * - blend_color
@@ -90,7 +91,8 @@ enum cso_cache_type {
    CSO_DEPTH_STENCIL_ALPHA,
    CSO_RASTERIZER,
    CSO_FRAGMENT_SHADER,
-   CSO_VERTEX_SHADER
+   CSO_VERTEX_SHADER,
+   CSO_VELEMENTS
 };
 
 typedef void (*cso_state_callback)(void *ctx, void *obj);
@@ -144,6 +146,18 @@ struct cso_sampler {
    struct pipe_context *context;
 };
 
+struct cso_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velems[PIPE_MAX_ATTRIBS];
+};
+
+struct cso_velements {
+   struct cso_velems_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
 unsigned cso_construct_key(void *item, int item_size);
 
 struct cso_cache *cso_cache_create(void);
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index a7335c340ca..6500891a10c 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -89,6 +89,7 @@ struct cso_context {
    void *rasterizer, *rasterizer_saved;
    void *fragment_shader, *fragment_shader_saved, *geometry_shader;
    void *vertex_shader, *vertex_shader_saved, *geometry_shader_saved;
+   void *velements, *velements_saved;
 
    struct pipe_clip_state clip;
    struct pipe_clip_state clip_saved;
@@ -174,6 +175,20 @@ static boolean delete_vs_state(struct cso_context *ctx, void *state)
    return FALSE;
 }
 
+static boolean delete_vertex_elements(struct cso_context *ctx,
+                                      void *state)
+{
+   struct cso_velements *cso = (struct cso_velements *)state;
+
+   if (ctx->velements == cso->data)
+      return FALSE;
+
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
 
 static INLINE boolean delete_cso(struct cso_context *ctx,
                                  void *state, enum cso_cache_type type)
@@ -197,6 +212,9 @@ static INLINE boolean delete_cso(struct cso_context *ctx,
    case CSO_VERTEX_SHADER:
       return delete_vs_state(ctx, state);
       break;
+   case CSO_VELEMENTS:
+      return delete_vertex_elements(ctx, state);
+      break;
    default:
       assert(0);
       FREE(state);
@@ -271,6 +289,7 @@ void cso_release_all( struct cso_context *ctx )
       ctx->pipe->bind_depth_stencil_alpha_state( ctx->pipe, NULL );
       ctx->pipe->bind_fs_state( ctx->pipe, NULL );
       ctx->pipe->bind_vs_state( ctx->pipe, NULL );
+      ctx->pipe->bind_vertex_elements_state( ctx->pipe, NULL );
    }
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
@@ -1130,7 +1149,6 @@ void cso_restore_geometry_shader(struct cso_context *ctx)
    ctx->geometry_shader_saved = NULL;
 }
 
-
 /* clip state */
 
 static INLINE void
@@ -1180,3 +1198,66 @@ cso_restore_clip(struct cso_context *ctx)
       ctx->pipe->set_clip_state(ctx->pipe, &ctx->clip_saved);
    }
 }
+
+enum pipe_error cso_set_vertex_elements(struct cso_context *ctx,
+                                        unsigned count,
+                                        const struct pipe_vertex_element *states)
+{
+   unsigned key_size, hash_key;
+   struct cso_hash_iter iter;
+   void *handle;
+   struct cso_velems_state velems_state;
+
+   /* need to include the count into the stored state data too.
+      Otherwise first few count pipe_vertex_elements could be identical even if count
+      is different, and there's no guarantee the hash would be different in that
+      case neither */
+   key_size = sizeof(struct pipe_vertex_element) * count + sizeof(unsigned);
+   velems_state.count = count;
+   memcpy(velems_state.velems, states, sizeof(struct pipe_vertex_element) * count);
+   hash_key = cso_construct_key((void*)&velems_state, key_size);
+   iter = cso_find_state_template(ctx->cache, hash_key, CSO_VELEMENTS, (void*)&velems_state, key_size);
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_velements *cso = MALLOC(sizeof(struct cso_velements));
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(&cso->state, &velems_state, key_size);
+      cso->data = ctx->pipe->create_vertex_elements_state(ctx->pipe, count, &cso->state.velems[0]);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_vertex_elements_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_VELEMENTS, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_velements *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->velements != handle) {
+      ctx->velements = handle;
+      ctx->pipe->bind_vertex_elements_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_vertex_elements(struct cso_context *ctx)
+{
+   assert(!ctx->velements_saved);
+   ctx->velements_saved = ctx->velements;
+}
+
+void cso_restore_vertex_elements(struct cso_context *ctx)
+{
+   if (ctx->velements != ctx->velements_saved) {
+      ctx->velements = ctx->velements_saved;
+      ctx->pipe->bind_vertex_elements_state(ctx->pipe, ctx->velements_saved);
+   }
+   ctx->velements_saved = NULL;
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index 251a9a644f8..9c16abd28dd 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -122,6 +122,12 @@ void
 cso_restore_vertex_sampler_textures(struct cso_context *cso);
 
 
+enum pipe_error cso_set_vertex_elements(struct cso_context *ctx,
+                                        unsigned count,
+                                        const struct pipe_vertex_element *states);
+void cso_save_vertex_elements(struct cso_context *ctx);
+void cso_restore_vertex_elements(struct cso_context *ctx);
+
 
 /* These aren't really sensible -- most of the time the api provides
  * object semantics for shaders anyway, and the cases where it doesn't
@@ -157,7 +163,6 @@ void cso_save_geometry_shader(struct cso_context *cso);
 void cso_restore_geometry_shader(struct cso_context *cso);
 
 
-
 enum pipe_error cso_set_framebuffer(struct cso_context *cso,
                                     const struct pipe_framebuffer_state *fb);
 void cso_save_framebuffer(struct cso_context *cso);
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 6d90a6c42fd..a8cdc57ad96 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -307,9 +307,8 @@ draw_arrays_instanced(struct draw_context *draw,
       tgsi_dump(draw->vs.vertex_shader->state.tokens, 0);
       debug_printf("Elements:\n");
       for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
-         debug_printf("  format=%s comps=%u\n",
-                      util_format_name(draw->pt.vertex_element[i].src_format),
-                      draw->pt.vertex_element[i].nr_components);
+         debug_printf("  format=%s\n",
+                      util_format_name(draw->pt.vertex_element[i].src_format));
       }
       debug_printf("Buffers:\n");
       for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_decompose.h b/src/gallium/auxiliary/draw/draw_pt_decompose.h
index 4ca5b520204..3c44f7c11ee 100644
--- a/src/gallium/auxiliary/draw/draw_pt_decompose.h
+++ b/src/gallium/auxiliary/draw/draw_pt_decompose.h
@@ -105,40 +105,20 @@ static void FUNC( ARGS,
 
 
    case PIPE_PRIM_QUADS:
-      if (flatfirst) {
-         for (i = 0; i+3 < count; i += 4) {
-            QUAD( (i + 1),
-                  (i + 2),
-                  (i + 3),
-                  (i + 0) );
-         }
-      }
-      else {
-         for (i = 0; i+3 < count; i += 4) {
-            QUAD( (i + 0),
-                  (i + 1),
-                  (i + 2),
-                  (i + 3));
-         }
+      for (i = 0; i+3 < count; i += 4) {
+         QUAD( (i + 0),
+               (i + 1),
+               (i + 2),
+               (i + 3));
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (flatfirst) {
-         for (i = 0; i+3 < count; i += 2) {
-            QUAD( (i + 1),
-                  (i + 3),
-                  (i + 2),
-                  (i + 0) );
-         }
-      }
-      else {
-         for (i = 0; i+3 < count; i += 2) {
-            QUAD( (i + 2),
-                  (i + 0),
-                  (i + 1),
-                  (i + 3));
-         }
+      for (i = 0; i+3 < count; i += 2) {
+         QUAD( (i + 2),
+               (i + 0),
+               (i + 1),
+               (i + 3));
       }
       break;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
index 62822a3d562..7cba8547f15 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
@@ -118,39 +118,21 @@ static void FUNC( struct draw_pt_front_end *frontend,
 
    case PIPE_PRIM_QUADS:
       for (i = 0; i+3 < count; i += 4) {
-         if (flatfirst) {
-            QUAD( vcache,
-                  get_elt(elts, i + 0),
-                  get_elt(elts, i + 1),
-                  get_elt(elts, i + 2),
-                  get_elt(elts, i + 3) );
-         }
-         else {
-            QUAD( vcache,
-                  get_elt(elts, i + 0),
-                  get_elt(elts, i + 1),
-                  get_elt(elts, i + 2),
-                  get_elt(elts, i + 3) );
-         }
+         QUAD( vcache,
+               get_elt(elts, i + 0),
+               get_elt(elts, i + 1),
+               get_elt(elts, i + 2),
+               get_elt(elts, i + 3) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
       for (i = 0; i+3 < count; i += 2) {
-         if (flatfirst) {
-            QUAD( vcache,
-                  get_elt(elts, i + 0),
-                  get_elt(elts, i + 1),
-                  get_elt(elts, i + 3),
-                  get_elt(elts, i + 2) );
-         }
-         else {
-            QUAD( vcache,
-                  get_elt(elts, i + 2),
-                  get_elt(elts, i + 0),
-                  get_elt(elts, i + 1),
-                  get_elt(elts, i + 3) );
-         }
+         QUAD( vcache,
+               get_elt(elts, i + 2),
+               get_elt(elts, i + 0),
+               get_elt(elts, i + 1),
+               get_elt(elts, i + 3) );
       }
       break;
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
index 32f9e5201c5..e2c67883972 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
@@ -644,13 +644,26 @@ lp_build_abs(struct lp_build_context *bld,
 
    if(type.floating) {
       /* Mask out the sign bit */
-      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
-      unsigned long long absMask = ~(1ULL << (type.width - 1));
-      LLVMValueRef mask = lp_build_int_const_scalar(type, ((unsigned long long) absMask));
-      a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      a = LLVMBuildAnd(bld->builder, a, mask, "");
-      a = LLVMBuildBitCast(bld->builder, a, vec_type, "");
-      return a;
+      if (type.length == 1) {
+         LLVMTypeRef int_type = LLVMIntType(type.width);
+         LLVMTypeRef float_type = LLVMFloatType();
+         unsigned long long absMask = ~(1ULL << (type.width - 1));
+         LLVMValueRef mask = LLVMConstInt(int_type, absMask, 0);
+         a = LLVMBuildBitCast(bld->builder, a, int_type, "");
+         a = LLVMBuildAnd(bld->builder, a, mask, "");
+         a = LLVMBuildBitCast(bld->builder, a, float_type, "");
+         return a;
+      }
+      else {
+         /* vector of floats */
+         LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+         unsigned long long absMask = ~(1ULL << (type.width - 1));
+         LLVMValueRef mask = lp_build_int_const_scalar(type, ((unsigned long long) absMask));
+         a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+         a = LLVMBuildAnd(bld->builder, a, mask, "");
+         a = LLVMBuildBitCast(bld->builder, a, vec_type, "");
+         return a;
+      }
    }
 
    if(type.width*type.length == 128 && util_cpu_caps.has_ssse3) {
@@ -753,7 +766,7 @@ lp_build_set_sign(struct lp_build_context *bld,
 
 
 /**
- * Convert vector of int to vector of float.
+ * Convert vector of (or scalar) int to vector of (or scalar) float.
  */
 LLVMValueRef
 lp_build_int_to_float(struct lp_build_context *bld,
@@ -764,7 +777,11 @@ lp_build_int_to_float(struct lp_build_context *bld,
    assert(type.floating);
    /*assert(lp_check_value(type, a));*/
 
-   {
+   if (type.length == 1) {
+      LLVMTypeRef float_type = LLVMFloatType();
+      return LLVMBuildSIToFP(bld->builder, a, float_type, "");
+   }
+   else {
       LLVMTypeRef vec_type = lp_build_vec_type(type);
       /*LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);*/
       LLVMValueRef res;
@@ -921,12 +938,18 @@ lp_build_itrunc(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
 
    assert(type.floating);
-   assert(lp_check_value(type, a));
 
-   return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+   if (type.length == 1) {
+      LLVMTypeRef int_type = LLVMIntType(type.width);
+      return LLVMBuildFPTrunc(bld->builder, a, int_type, "");
+   }
+   else {
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      assert(lp_check_value(type, a));
+      return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+   }
 }
 
 
@@ -939,6 +962,15 @@ lp_build_iround(struct lp_build_context *bld,
    LLVMValueRef res;
 
    assert(type.floating);
+
+   if (type.length == 1) {
+      /* scalar float to int */
+      LLVMTypeRef int_type = LLVMIntType(type.width);
+      /* XXX we want rounding here! */
+      res = LLVMBuildFPToSI(bld->builder, a, int_type, "");
+      return res;
+   }
+
    assert(lp_check_value(type, a));
 
    if(util_cpu_caps.has_sse4_1) {
@@ -1207,6 +1239,7 @@ lp_build_polynomial(struct lp_build_context *bld,
                     unsigned num_coeffs)
 {
    const struct lp_type type = bld->type;
+   LLVMTypeRef float_type = LLVMFloatType();
    LLVMValueRef res = NULL;
    unsigned i;
 
@@ -1216,7 +1249,13 @@ lp_build_polynomial(struct lp_build_context *bld,
                    __FUNCTION__);
 
    for (i = num_coeffs; i--; ) {
-      LLVMValueRef coeff = lp_build_const_scalar(type, coeffs[i]);
+      LLVMValueRef coeff;
+
+      if (type.length == 1)
+         coeff = LLVMConstReal(float_type, coeffs[i]);
+      else
+         coeff = lp_build_const_scalar(type, coeffs[i]);
+
       if(res)
          res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res));
       else
@@ -1410,11 +1449,87 @@ lp_build_log2_approx(struct lp_build_context *bld,
 }
 
 
+/** scalar version of above function */
+static void
+lp_build_float_log2_approx(struct lp_build_context *bld,
+                           LLVMValueRef x,
+                           LLVMValueRef *p_exp,
+                           LLVMValueRef *p_floor_log2,
+                           LLVMValueRef *p_log2)
+{
+   const struct lp_type type = bld->type;
+   LLVMTypeRef float_type = LLVMFloatType();
+   LLVMTypeRef int_type = LLVMIntType(type.width);
+
+   LLVMValueRef expmask = LLVMConstInt(int_type, 0x7f800000, 0);
+   LLVMValueRef mantmask = LLVMConstInt(int_type, 0x007fffff, 0);
+   LLVMValueRef one = LLVMConstBitCast(bld->one, int_type);
+
+   LLVMValueRef i = NULL;
+   LLVMValueRef exp = NULL;
+   LLVMValueRef mant = NULL;
+   LLVMValueRef logexp = NULL;
+   LLVMValueRef logmant = NULL;
+   LLVMValueRef res = NULL;
+
+   if(p_exp || p_floor_log2 || p_log2) {
+      /* TODO: optimize the constant case */
+      if(LLVMIsConstant(x))
+         debug_printf("%s: inefficient/imprecise constant arithmetic\n",
+                      __FUNCTION__);
+
+      assert(type.floating && type.width == 32);
+
+      i = LLVMBuildBitCast(bld->builder, x, int_type, "");
+
+      /* exp = (float) exponent(x) */
+      exp = LLVMBuildAnd(bld->builder, i, expmask, "");
+   }
+
+   if(p_floor_log2 || p_log2) {
+      LLVMValueRef c23 = LLVMConstInt(int_type, 23, 0);
+      LLVMValueRef c127 = LLVMConstInt(int_type, 127, 0);
+      logexp = LLVMBuildLShr(bld->builder, exp, c23, "");
+      logexp = LLVMBuildSub(bld->builder, logexp, c127, "");
+      logexp = LLVMBuildSIToFP(bld->builder, logexp, float_type, "");
+   }
+
+   if(p_log2) {
+      /* mant = (float) mantissa(x) */
+      mant = LLVMBuildAnd(bld->builder, i, mantmask, "");
+      mant = LLVMBuildOr(bld->builder, mant, one, "");
+      mant = LLVMBuildBitCast(bld->builder, mant, float_type, "");
+
+      logmant = lp_build_polynomial(bld, mant, lp_build_log2_polynomial,
+                                    Elements(lp_build_log2_polynomial));
+
+      /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/
+      logmant = LLVMBuildMul(bld->builder, logmant, LLVMBuildSub(bld->builder, mant, bld->one, ""), "");
+
+      res = LLVMBuildAdd(bld->builder, logmant, logexp, "");
+   }
+
+   if(p_exp)
+      *p_exp = exp;
+
+   if(p_floor_log2)
+      *p_floor_log2 = logexp;
+
+   if(p_log2)
+      *p_log2 = res;
+}
+
+
 LLVMValueRef
 lp_build_log2(struct lp_build_context *bld,
               LLVMValueRef x)
 {
    LLVMValueRef res;
-   lp_build_log2_approx(bld, x, NULL, NULL, &res);
+   if (bld->type.length == 1) {
+      lp_build_float_log2_approx(bld, x, NULL, NULL, &res);
+   }
+   else {
+      lp_build_log2_approx(bld, x, NULL, NULL, &res);
+   }
    return res;
 }
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.c b/src/gallium/auxiliary/gallivm/lp_bld_const.c
index c8eaa8c3940..53447757e8e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_const.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_const.c
@@ -264,10 +264,16 @@ lp_build_one(struct lp_type type)
    for(i = 1; i < type.length; ++i)
       elems[i] = elems[0];
 
-   return LLVMConstVector(elems, type.length);
+   if (type.length == 1)
+      return elems[0];
+   else
+      return LLVMConstVector(elems, type.length);
 }
                
 
+/**
+ * Build constant-valued vector from a scalar value.
+ */
 LLVMValueRef
 lp_build_const_scalar(struct lp_type type,
                       double val)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
index 2726747eaea..7c585fda788 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
@@ -198,7 +198,7 @@ lp_build_compare(LLVMBuilderRef builder,
 
          return res;
       }
-   }
+   } /* if (type.width * type.length == 128) */
 #endif
 
    if(type.floating) {
@@ -238,20 +238,25 @@ lp_build_compare(LLVMBuilderRef builder,
       cond = LLVMBuildFCmp(builder, op, a, b, "");
       res = LLVMBuildSelect(builder, cond, ones, zeros, "");
 #else
-      debug_printf("%s: warning: using slow element-wise vector comparison\n",
-                   __FUNCTION__);
       res = LLVMGetUndef(int_vec_type);
-      for(i = 0; i < type.length; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-         cond = LLVMBuildFCmp(builder, op,
-                              LLVMBuildExtractElement(builder, a, index, ""),
-                              LLVMBuildExtractElement(builder, b, index, ""),
-                              "");
-         cond = LLVMBuildSelect(builder, cond,
-                                LLVMConstExtractElement(ones, index),
-                                LLVMConstExtractElement(zeros, index),
-                                "");
-         res = LLVMBuildInsertElement(builder, res, cond, index, "");
+      if (type.length == 1) {
+         res = LLVMBuildFCmp(builder, op, a, b, "");
+      }
+      else {
+         debug_printf("%s: warning: using slow element-wise float"
+                      " vector comparison\n", __FUNCTION__);
+         for (i = 0; i < type.length; ++i) {
+            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            cond = LLVMBuildFCmp(builder, op,
+                                 LLVMBuildExtractElement(builder, a, index, ""),
+                                 LLVMBuildExtractElement(builder, b, index, ""),
+                                 "");
+            cond = LLVMBuildSelect(builder, cond,
+                                   LLVMConstExtractElement(ones, index),
+                                   LLVMConstExtractElement(zeros, index),
+                                   "");
+            res = LLVMBuildInsertElement(builder, res, cond, index, "");
+         }
       }
 #endif
    }
@@ -286,20 +291,26 @@ lp_build_compare(LLVMBuilderRef builder,
       cond = LLVMBuildICmp(builder, op, a, b, "");
       res = LLVMBuildSelect(builder, cond, ones, zeros, "");
 #else
-      debug_printf("%s: warning: using slow element-wise int vector comparison\n",
-                   __FUNCTION__);
       res = LLVMGetUndef(int_vec_type);
-      for(i = 0; i < type.length; ++i) {
-         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-         cond = LLVMBuildICmp(builder, op,
-                              LLVMBuildExtractElement(builder, a, index, ""),
-                              LLVMBuildExtractElement(builder, b, index, ""),
-                              "");
-         cond = LLVMBuildSelect(builder, cond,
-                                LLVMConstExtractElement(ones, index),
-                                LLVMConstExtractElement(zeros, index),
-                                "");
-         res = LLVMBuildInsertElement(builder, res, cond, index, "");
+      if (type.length == 1) {
+         res = LLVMBuildICmp(builder, op, a, b, "");
+      }
+      else {
+         debug_printf("%s: warning: using slow element-wise int"
+                      " vector comparison\n", __FUNCTION__);
+
+         for(i = 0; i < type.length; ++i) {
+            LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+            cond = LLVMBuildICmp(builder, op,
+                                 LLVMBuildExtractElement(builder, a, index, ""),
+                                 LLVMBuildExtractElement(builder, b, index, ""),
+                                 "");
+            cond = LLVMBuildSelect(builder, cond,
+                                   LLVMConstExtractElement(ones, index),
+                                   LLVMConstExtractElement(zeros, index),
+                                   "");
+            res = LLVMBuildInsertElement(builder, res, cond, index, "");
+         }
       }
 #endif
    }
@@ -339,26 +350,31 @@ lp_build_select(struct lp_build_context *bld,
    if(a == b)
       return a;
 
-   if(type.floating) {
-      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
-      a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, int_vec_type, "");
+   if (type.length == 1) {
+      res = LLVMBuildSelect(bld->builder, mask, a, b, "");
    }
+   else {
+      if(type.floating) {
+         LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+         a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+         b = LLVMBuildBitCast(bld->builder, b, int_vec_type, "");
+      }
 
-   a = LLVMBuildAnd(bld->builder, a, mask, "");
+      a = LLVMBuildAnd(bld->builder, a, mask, "");
 
-   /* This often gets translated to PANDN, but sometimes the NOT is
-    * pre-computed and stored in another constant. The best strategy depends
-    * on available registers, so it is not a big deal -- hopefully LLVM does
-    * the right decision attending the rest of the program.
-    */
-   b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), "");
+      /* This often gets translated to PANDN, but sometimes the NOT is
+       * pre-computed and stored in another constant. The best strategy depends
+       * on available registers, so it is not a big deal -- hopefully LLVM does
+       * the right decision attending the rest of the program.
+       */
+      b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), "");
 
-   res = LLVMBuildOr(bld->builder, a, b, "");
+      res = LLVMBuildOr(bld->builder, a, b, "");
 
-   if(type.floating) {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+      if(type.floating) {
+         LLVMTypeRef vec_type = lp_build_vec_type(type);
+         res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+      }
    }
 
    return res;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.c b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
index bc360ad77ad..4c61d107494 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_pack.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
@@ -256,7 +256,9 @@ lp_build_pack2(LLVMBuilderRef builder,
                LLVMValueRef lo,
                LLVMValueRef hi)
 {
+#if !(HAVE_LLVM >= 0x0207)
    LLVMTypeRef src_vec_type = lp_build_vec_type(src_type);
+#endif
    LLVMTypeRef dst_vec_type = lp_build_vec_type(dst_type);
    LLVMValueRef shuffle;
    LLVMValueRef res;
@@ -272,11 +274,14 @@ lp_build_pack2(LLVMBuilderRef builder,
       switch(src_type.width) {
       case 32:
          if(dst_type.sign) {
+#if HAVE_LLVM >= 0x0207
+            res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packssdw.128", dst_vec_type, lo, hi);
+#else
             res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packssdw.128", src_vec_type, lo, hi);
+#endif
          }
          else {
             if (util_cpu_caps.has_sse4_1) {
-               /* PACKUSDW is the only instrinsic with a consistent signature */
                return lp_build_intrinsic_binary(builder, "llvm.x86.sse41.packusdw", dst_vec_type, lo, hi);
             }
             else {
@@ -288,9 +293,17 @@ lp_build_pack2(LLVMBuilderRef builder,
 
       case 16:
          if(dst_type.sign)
+#if HAVE_LLVM >= 0x0207
+            res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packsswb.128", dst_vec_type, lo, hi);
+#else
             res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packsswb.128", src_vec_type, lo, hi);
+#endif
          else
+#if HAVE_LLVM >= 0x0207
+            res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packuswb.128", dst_vec_type, lo, hi);
+#else
             res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packuswb.128", src_vec_type, lo, hi);
+#endif
          break;
 
       default:
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
index 6a026e468e1..311c9f1b9e4 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
@@ -62,6 +62,18 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
    if(!sampler)
       return;
 
+   /*
+    * We don't copy sampler state over unless it is actually enabled, to avoid
+    * spurious recompiles, as the sampler static state is part of the shader
+    * key.
+    *
+    * Ideally the state tracker or cso_cache module would make all state
+    * canonical, but until that happens it's better to be safe than sorry here.
+    *
+    * XXX: Actually there's much more than can be done here, especially
+    * regarding 1D/2D/3D/CUBE textures, wrap modes, etc.
+    */
+
    state->format            = texture->format;
    state->target            = texture->target;
    state->pot_width         = util_is_pot(texture->width0);
@@ -74,8 +86,12 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
    state->min_img_filter    = sampler->min_img_filter;
    state->min_mip_filter    = sampler->min_mip_filter;
    state->mag_img_filter    = sampler->mag_img_filter;
+
    state->compare_mode      = sampler->compare_mode;
-   state->compare_func      = sampler->compare_func;
+   if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
+      state->compare_func   = sampler->compare_func;
+   }
+
    state->normalized_coords = sampler->normalized_coords;
    state->lod_bias          = sampler->lod_bias;
    state->min_lod           = sampler->min_lod;
@@ -146,8 +162,7 @@ lp_build_sample_offset(struct lp_build_context *bld,
                        const struct util_format_description *format_desc,
                        LLVMValueRef x,
                        LLVMValueRef y,
-                       LLVMValueRef y_stride,
-                       LLVMValueRef data_ptr)
+                       LLVMValueRef y_stride)
 {
    LLVMValueRef x_stride;
    LLVMValueRef offset;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
index 5ba0925bb69..68db91d6fd6 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
@@ -148,8 +148,7 @@ lp_build_sample_offset(struct lp_build_context *bld,
                        const struct util_format_description *format_desc,
                        LLVMValueRef x,
                        LLVMValueRef y,
-                       LLVMValueRef y_stride,
-                       LLVMValueRef data_ptr);
+                       LLVMValueRef y_stride);
 
 
 void
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 9058f76c1df..a965d394f44 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -65,6 +65,14 @@ struct lp_build_sample_context
 
    const struct util_format_description *format_desc;
 
+   /** regular scalar float type */
+   struct lp_type float_type;
+   struct lp_build_context float_bld;
+
+   /** regular scalar float type */
+   struct lp_type int_type;
+   struct lp_build_context int_bld;
+
    /** Incoming coordinates type and build context */
    struct lp_type coord_type;
    struct lp_build_context coord_bld;
@@ -108,6 +116,27 @@ wrap_mode_uses_border_color(unsigned mode)
 }
 
 
+static LLVMValueRef
+lp_build_get_mipmap_level(struct lp_build_sample_context *bld,
+                          LLVMValueRef data_array, LLVMValueRef level)
+{
+   LLVMValueRef indexes[2], data_ptr;
+   indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indexes[1] = level;
+   data_ptr = LLVMBuildGEP(bld->builder, data_array, indexes, 2, "");
+   data_ptr = LLVMBuildLoad(bld->builder, data_ptr, "");
+   return data_ptr;
+}
+
+
+static LLVMValueRef
+lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
+                                LLVMValueRef data_array, int level)
+{
+   LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0);
+   return lp_build_get_mipmap_level(bld, data_array, lvl);
+}
+
 
 /**
  * Gen code to fetch a texel from a texture at int coords (x, y).
@@ -171,8 +200,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
    /* convert x,y coords to linear offset from start of texture, in bytes */
    offset = lp_build_sample_offset(&bld->uint_coord_bld,
                                    bld->format_desc,
-                                   x, y, y_stride,
-                                   data_ptr);
+                                   x, y, y_stride);
 
    assert(bld->format_desc->block.width == 1);
    assert(bld->format_desc->block.height == 1);
@@ -210,19 +238,22 @@ lp_build_sample_packed(struct lp_build_sample_context *bld,
                        LLVMValueRef x,
                        LLVMValueRef y,
                        LLVMValueRef y_stride,
-                       LLVMValueRef data_ptr)
+                       LLVMValueRef data_array)
 {
    LLVMValueRef offset;
+   LLVMValueRef data_ptr;
 
    offset = lp_build_sample_offset(&bld->uint_coord_bld,
                                    bld->format_desc,
-                                   x, y, y_stride,
-                                   data_ptr);
+                                   x, y, y_stride);
 
    assert(bld->format_desc->block.width == 1);
    assert(bld->format_desc->block.height == 1);
    assert(bld->format_desc->block.bits <= bld->texel_type.width);
 
+   /* get pointer to mipmap level 0 data */
+   data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0);
+
    return lp_build_gather(bld->builder,
                           bld->texel_type.length,
                           bld->format_desc->block.bits,
@@ -711,7 +742,210 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
 
 
 /**
- * Sample 2D texture with nearest filtering.
+ * Codegen equivalent for u_minify().
+ * Return max(1, base_size >> level);
+ */
+static LLVMValueRef
+lp_build_minify(struct lp_build_sample_context *bld,
+                LLVMValueRef base_size,
+                LLVMValueRef level)
+{
+   LLVMValueRef size = LLVMBuildAShr(bld->builder, base_size, level, "minify");
+   size = lp_build_max(&bld->int_coord_bld, size, bld->int_coord_bld.one);
+   return size;
+}
+
+
+static int
+texture_dims(enum pipe_texture_target tex)
+{
+   switch (tex) {
+   case PIPE_TEXTURE_1D:
+      return 1;
+   case PIPE_TEXTURE_2D:
+   case PIPE_TEXTURE_CUBE:
+      return 2;
+   case PIPE_TEXTURE_3D:
+      return 3;
+   default:
+      assert(0 && "bad texture target in texture_dims()");
+      return 2;
+   }
+}
+
+
+/**
+ * Generate code to compute texture level of detail (lambda).
+ * \param s  vector of texcoord s values
+ * \param t  vector of texcoord t values
+ * \param r  vector of texcoord r values
+ * \param width  scalar int texture width
+ * \param height  scalar int texture height
+ * \param depth  scalar int texture depth
+ */
+static LLVMValueRef
+lp_build_lod_selector(struct lp_build_sample_context *bld,
+                      LLVMValueRef s,
+                      LLVMValueRef t,
+                      LLVMValueRef r,
+                      LLVMValueRef width,
+                      LLVMValueRef height,
+                      LLVMValueRef depth)
+
+{
+   const int dims = texture_dims(bld->static_state->target);
+   struct lp_build_context *coord_bld = &bld->coord_bld;
+   struct lp_build_context *float_bld = &bld->float_bld;
+   LLVMValueRef lod_bias = LLVMConstReal(LLVMFloatType(), bld->static_state->lod_bias);
+   LLVMValueRef min_lod = LLVMConstReal(LLVMFloatType(), bld->static_state->min_lod);
+   LLVMValueRef max_lod = LLVMConstReal(LLVMFloatType(), bld->static_state->max_lod);
+
+   LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   LLVMValueRef index1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
+   LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
+
+   LLVMValueRef s0, s1, s2;
+   LLVMValueRef t0, t1, t2;
+   LLVMValueRef r0, r1, r2;
+   LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
+   LLVMValueRef rho, lod;
+
+   /*
+    * dsdx = abs(s[1] - s[0]);
+    * dsdy = abs(s[2] - s[0]);
+    * dtdx = abs(t[1] - t[0]);
+    * dtdy = abs(t[2] - t[0]);
+    * drdx = abs(r[1] - r[0]);
+    * drdy = abs(r[2] - r[0]);
+    * XXX we're assuming a four-element quad in 2x2 layout here.
+    */
+   s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0");
+   s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1");
+   s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2");
+   dsdx = LLVMBuildSub(bld->builder, s1, s0, "");
+   dsdx = lp_build_abs(float_bld, dsdx);
+   dsdy = LLVMBuildSub(bld->builder, s2, s0, "");
+   dsdy = lp_build_abs(float_bld, dsdy);
+   if (dims > 1) {
+      t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0");
+      t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1");
+      t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2");
+      dtdx = LLVMBuildSub(bld->builder, t1, t0, "");
+      dtdx = lp_build_abs(float_bld, dtdx);
+      dtdy = LLVMBuildSub(bld->builder, t2, t0, "");
+      dtdy = lp_build_abs(float_bld, dtdy);
+      if (dims > 2) {
+         r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0");
+         r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1");
+         r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2");
+         drdx = LLVMBuildSub(bld->builder, r1, r0, "");
+         drdx = lp_build_abs(float_bld, drdx);
+         drdy = LLVMBuildSub(bld->builder, r2, r0, "");
+         drdy = lp_build_abs(float_bld, drdy);
+      }
+   }
+
+   /* Compute rho = max of all partial derivatives scaled by texture size.
+    * XXX this could be vectorized somewhat
+    */
+   rho = LLVMBuildMul(bld->builder,
+                      lp_build_max(float_bld, dsdx, dsdy),
+                      lp_build_int_to_float(float_bld, width), "");
+   if (dims > 1) {
+      LLVMValueRef max;
+      max = LLVMBuildMul(bld->builder,
+                         lp_build_max(float_bld, dtdx, dtdy),
+                         lp_build_int_to_float(float_bld, height), "");
+      rho = lp_build_max(float_bld, rho, max);
+      if (dims > 2) {
+         max = LLVMBuildMul(bld->builder,
+                            lp_build_max(float_bld, drdx, drdy),
+                            lp_build_int_to_float(float_bld, depth), "");
+         rho = lp_build_max(float_bld, rho, max);
+      }
+   }
+
+   /* compute lod = log2(rho) */
+   lod = lp_build_log2(float_bld, rho);
+
+   /* add lod bias */
+   lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "LOD bias");
+
+   /* clamp lod */
+   lod = lp_build_clamp(float_bld, lod, min_lod, max_lod);
+
+   return lod;
+}
+
+
+/**
+ * For PIPE_TEX_MIPFILTER_NEAREST, convert float LOD to integer
+ * mipmap level index.
+ * Note: this is all scalar code.
+ * \param lod  scalar float texture level of detail
+ * \param level_out  returns integer 
+ */
+static void
+lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod,
+                           LLVMValueRef *level_out)
+{
+   struct lp_build_context *float_bld = &bld->float_bld;
+   struct lp_build_context *int_bld = &bld->int_bld;
+   LLVMValueRef last_level, level;
+
+   LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
+
+   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
+                                               bld->builder, unit);
+
+   /* convert float lod to integer */
+   level = lp_build_iround(float_bld, lod);
+
+   /* clamp level to legal range of levels */
+   *level_out = lp_build_clamp(int_bld, level, zero, last_level);
+}
+
+
+/**
+ * For PIPE_TEX_MIPFILTER_LINEAR, convert float LOD to integer to
+ * two (adjacent) mipmap level indexes.  Later, we'll sample from those
+ * two mipmap levels and interpolate between them.
+ */
+static void
+lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod,
+                           LLVMValueRef *level0_out,
+                           LLVMValueRef *level1_out,
+                           LLVMValueRef *weight_out)
+{
+   struct lp_build_context *coord_bld = &bld->coord_bld;
+   struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMValueRef last_level, level;
+
+   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
+                                               bld->builder, unit);
+
+   /* convert float lod to integer */
+   level = lp_build_ifloor(coord_bld, lod);
+
+   /* compute level 0 and clamp to legal range of levels */
+   *level0_out = lp_build_clamp(int_coord_bld, level,
+                                int_coord_bld->zero,
+                                last_level);
+   /* compute level 1 and clamp to legal range of levels */
+   *level1_out = lp_build_add(int_coord_bld, *level0_out, int_coord_bld->one);
+   *level1_out = lp_build_min(int_coord_bld, *level1_out, int_coord_bld->zero);
+
+   *weight_out = lp_build_fract(coord_bld, lod);
+}
+
+
+
+/**
+ * Sample 2D texture with nearest filtering, no mipmapping.
  */
 static void
 lp_build_sample_2d_nearest_soa(struct lp_build_sample_context *bld,
@@ -720,10 +954,11 @@ lp_build_sample_2d_nearest_soa(struct lp_build_sample_context *bld,
                                LLVMValueRef width,
                                LLVMValueRef height,
                                LLVMValueRef stride,
-                               LLVMValueRef data_ptr,
+                               LLVMValueRef data_array,
                                LLVMValueRef *texel)
 {
    LLVMValueRef x, y;
+   LLVMValueRef data_ptr;
 
    x = lp_build_sample_wrap_nearest(bld, s, width,
                                     bld->static_state->pot_width,
@@ -735,10 +970,66 @@ lp_build_sample_2d_nearest_soa(struct lp_build_sample_context *bld,
    lp_build_name(x, "tex.x.wrapped");
    lp_build_name(y, "tex.y.wrapped");
 
+   /* get pointer to mipmap level 0 data */
+   data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0);
+
    lp_build_sample_texel_soa(bld, width, height, x, y, stride, data_ptr, texel);
 }
 
 
+/**
+ * Sample 2D texture with nearest filtering, nearest mipmap.
+ */
+static void
+lp_build_sample_2d_nearest_mip_nearest_soa(struct lp_build_sample_context *bld,
+                                           unsigned unit,
+                                           LLVMValueRef s,
+                                           LLVMValueRef t,
+                                           LLVMValueRef width,
+                                           LLVMValueRef height,
+                                           LLVMValueRef width_vec,
+                                           LLVMValueRef height_vec,
+                                           LLVMValueRef stride,
+                                           LLVMValueRef data_array,
+                                           LLVMValueRef *texel)
+{
+   LLVMValueRef x, y;
+   LLVMValueRef lod, ilevel, ilevel_vec;
+   LLVMValueRef data_ptr;
+
+   /* compute float LOD */
+   lod = lp_build_lod_selector(bld, s, t, NULL, width, height, NULL);
+
+   /* convert LOD to int */
+   lp_build_nearest_mip_level(bld, unit, lod, &ilevel);
+
+   ilevel_vec = lp_build_broadcast_scalar(&bld->int_coord_bld, ilevel);
+
+   /* compute width_vec, height at mipmap level 'ilevel' */
+   width_vec = lp_build_minify(bld, width_vec, ilevel_vec);
+   height_vec = lp_build_minify(bld, height_vec, ilevel_vec);
+   stride = lp_build_minify(bld, stride, ilevel_vec);
+
+   x = lp_build_sample_wrap_nearest(bld, s, width_vec,
+                                    bld->static_state->pot_width,
+                                    bld->static_state->wrap_s);
+   y = lp_build_sample_wrap_nearest(bld, t, height_vec,
+                                    bld->static_state->pot_height,
+                                    bld->static_state->wrap_t);
+
+   lp_build_name(x, "tex.x.wrapped");
+   lp_build_name(y, "tex.y.wrapped");
+
+   /* get pointer to mipmap level [ilevel] data */
+   if (0)
+      data_ptr = lp_build_get_mipmap_level(bld, data_array, ilevel);
+   else
+      data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0);
+
+   lp_build_sample_texel_soa(bld, width_vec, height_vec, x, y, stride, data_ptr, texel);
+}
+
+
 /**
  * Sample 2D texture with bilinear filtering.
  */
@@ -749,7 +1040,7 @@ lp_build_sample_2d_linear_soa(struct lp_build_sample_context *bld,
                               LLVMValueRef width,
                               LLVMValueRef height,
                               LLVMValueRef stride,
-                              LLVMValueRef data_ptr,
+                              LLVMValueRef data_array,
                               LLVMValueRef *texel)
 {
    LLVMValueRef s_fpart;
@@ -757,6 +1048,7 @@ lp_build_sample_2d_linear_soa(struct lp_build_sample_context *bld,
    LLVMValueRef x0, x1;
    LLVMValueRef y0, y1;
    LLVMValueRef neighbors[2][2][4];
+   LLVMValueRef data_ptr;
    unsigned chan;
 
    lp_build_sample_wrap_linear(bld, s, width, bld->static_state->pot_width,
@@ -764,6 +1056,9 @@ lp_build_sample_2d_linear_soa(struct lp_build_sample_context *bld,
    lp_build_sample_wrap_linear(bld, t, height, bld->static_state->pot_height,
                                bld->static_state->wrap_t, &y0, &y1, &t_fpart);
 
+   /* get pointer to mipmap level 0 data */
+   data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0);
+
    lp_build_sample_texel_soa(bld, width, height, x0, y0, stride, data_ptr, neighbors[0][0]);
    lp_build_sample_texel_soa(bld, width, height, x1, y0, stride, data_ptr, neighbors[0][1]);
    lp_build_sample_texel_soa(bld, width, height, x0, y1, stride, data_ptr, neighbors[1][0]);
@@ -818,7 +1113,7 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
                               LLVMValueRef width,
                               LLVMValueRef height,
                               LLVMValueRef stride,
-                              LLVMValueRef data_ptr,
+                              LLVMValueRef data_array,
                               LLVMValueRef *texel)
 {
    LLVMBuilderRef builder = bld->builder;
@@ -835,7 +1130,7 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
    LLVMValueRef packed, packed_lo, packed_hi;
    LLVMValueRef unswizzled[4];
 
-   lp_build_context_init(&i32, builder, lp_type_int(32));
+   lp_build_context_init(&i32, builder, lp_type_int_vec(32));
    lp_build_context_init(&h16, builder, lp_type_ufixed(16));
    lp_build_context_init(&u8n, builder, lp_type_unorm(8));
 
@@ -958,10 +1253,10 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
     * The higher 8 bits of the resulting elements will be zero.
     */
 
-   neighbors[0][0] = lp_build_sample_packed(bld, x0, y0, stride, data_ptr);
-   neighbors[0][1] = lp_build_sample_packed(bld, x1, y0, stride, data_ptr);
-   neighbors[1][0] = lp_build_sample_packed(bld, x0, y1, stride, data_ptr);
-   neighbors[1][1] = lp_build_sample_packed(bld, x1, y1, stride, data_ptr);
+   neighbors[0][0] = lp_build_sample_packed(bld, x0, y0, stride, data_array);
+   neighbors[0][1] = lp_build_sample_packed(bld, x1, y0, stride, data_array);
+   neighbors[1][0] = lp_build_sample_packed(bld, x0, y1, stride, data_array);
+   neighbors[1][1] = lp_build_sample_packed(bld, x1, y1, stride, data_array);
 
    neighbors[0][0] = LLVMBuildBitCast(builder, neighbors[0][0], u8n_vec_type, "");
    neighbors[0][1] = LLVMBuildBitCast(builder, neighbors[0][1], u8n_vec_type, "");
@@ -1044,194 +1339,11 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
 }
 
 
-static int
-texture_dims(enum pipe_texture_target tex)
-{
-   switch (tex) {
-   case PIPE_TEXTURE_1D:
-      return 1;
-   case PIPE_TEXTURE_2D:
-   case PIPE_TEXTURE_CUBE:
-      return 2;
-   case PIPE_TEXTURE_3D:
-      return 3;
-   default:
-      assert(0 && "bad texture target in texture_dims()");
-      return 2;
-   }
-}
-
-
-/**
- * Generate code to compute texture level of detail (lambda).
- * \param s  vector of texcoord s values
- * \param t  vector of texcoord t values
- * \param r  vector of texcoord r values
- * \param width  scalar int texture width
- * \param height  scalar int texture height
- * \param depth  scalar int texture depth
- */
-static LLVMValueRef
-lp_build_lod_selector(struct lp_build_sample_context *bld,
-                      LLVMValueRef s,
-                      LLVMValueRef t,
-                      LLVMValueRef r,
-                      LLVMValueRef width,
-                      LLVMValueRef height,
-                      LLVMValueRef depth)
-
-{
-   const int dims = texture_dims(bld->static_state->target);
-   struct lp_build_context *coord_bld = &bld->coord_bld;
-
-   LLVMValueRef lod_bias = lp_build_const_scalar(bld->coord_bld.type,
-                                                 bld->static_state->lod_bias);
-   LLVMValueRef min_lod = lp_build_const_scalar(bld->coord_bld.type,
-                                                bld->static_state->min_lod);
-   LLVMValueRef max_lod = lp_build_const_scalar(bld->coord_bld.type,
-                                                bld->static_state->max_lod);
-
-   LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   LLVMValueRef index1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
-   LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-
-   LLVMValueRef s0, s1, s2;
-   LLVMValueRef t0, t1, t2;
-   LLVMValueRef r0, r1, r2;
-   LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
-   LLVMValueRef rho, lod;
-
-   /*
-    * dsdx = abs(s[1] - s[0]);
-    * dsdy = abs(s[2] - s[0]);
-    * dtdx = abs(t[1] - t[0]);
-    * dtdy = abs(t[2] - t[0]);
-    * drdx = abs(r[1] - r[0]);
-    * drdy = abs(r[2] - r[0]);
-    * XXX we're assuming a four-element quad in 2x2 layout here.
-    */
-   s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0");
-   s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1");
-   s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2");
-   dsdx = lp_build_abs(coord_bld, lp_build_sub(coord_bld, s1, s0));
-   dsdy = lp_build_abs(coord_bld, lp_build_sub(coord_bld, s2, s0));
-   if (dims > 1) {
-      t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0");
-      t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1");
-      t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2");
-      dtdx = lp_build_abs(coord_bld, lp_build_sub(coord_bld, t1, t0));
-      dtdy = lp_build_abs(coord_bld, lp_build_sub(coord_bld, t2, t0));
-      if (dims > 2) {
-         r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0");
-         r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1");
-         r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2");
-         drdx = lp_build_abs(coord_bld, lp_build_sub(coord_bld, r1, r0));
-         drdy = lp_build_abs(coord_bld, lp_build_sub(coord_bld, r2, r0));
-      }
-   }
-
-   /* Compute rho = max of all partial derivatives scaled by texture size.
-    * XXX this can be vectorized somewhat
-    */
-   rho = lp_build_mul(coord_bld,
-                       lp_build_max(coord_bld, dsdx, dsdy),
-                       lp_build_int_to_float(coord_bld, width));
-   if (dims > 1) {
-      LLVMValueRef max;
-      max = lp_build_mul(coord_bld,
-                         lp_build_max(coord_bld, dtdx, dtdy),
-                         lp_build_int_to_float(coord_bld, height));
-      rho = lp_build_max(coord_bld, rho, max);
-      if (dims > 2) {
-         max = lp_build_mul(coord_bld,
-                            lp_build_max(coord_bld, drdx, drdy),
-                            lp_build_int_to_float(coord_bld, depth));
-         rho = lp_build_max(coord_bld, rho, max);
-      }
-   }
-
-   /* compute lod = log2(rho) */
-   lod = lp_build_log2(coord_bld, rho);
-
-   /* add lod bias */
-   lod = lp_build_add(coord_bld, lod, lod_bias);
-
-   /* clamp lod */
-   lod = lp_build_clamp(coord_bld, lod, min_lod, max_lod);
-
-   return lod;
-}
-
-
-/**
- * For PIPE_TEX_MIPFILTER_NEAREST, convert float LOD to integer
- * mipmap level index.
- * \param lod  scalar float texture level of detail
- * \param level_out  returns integer 
- */
-static void
-lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
-                           unsigned unit,
-                           LLVMValueRef lod,
-                           LLVMValueRef *level_out)
-{
-   struct lp_build_context *coord_bld = &bld->coord_bld;
-   struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
-   LLVMValueRef last_level, level;
-
-   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
-                                               bld->builder, unit);
-
-   /* convert float lod to integer */
-   level = lp_build_iround(coord_bld, lod);
-
-   /* clamp level to legal range of levels */
-   *level_out = lp_build_clamp(int_coord_bld, level,
-                               int_coord_bld->zero,
-                               last_level);
-}
-
-
-/**
- * For PIPE_TEX_MIPFILTER_LINEAR, convert float LOD to integer to
- * two (adjacent) mipmap level indexes.  Later, we'll sample from those
- * two mipmap levels and interpolate between them.
- */
-static void
-lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
-                           unsigned unit,
-                           LLVMValueRef lod,
-                           LLVMValueRef *level0_out,
-                           LLVMValueRef *level1_out,
-                           LLVMValueRef *weight_out)
-{
-   struct lp_build_context *coord_bld = &bld->coord_bld;
-   struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
-   LLVMValueRef last_level, level;
-
-   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
-                                               bld->builder, unit);
-
-   /* convert float lod to integer */
-   level = lp_build_ifloor(coord_bld, lod);
-
-   /* compute level 0 and clamp to legal range of levels */
-   *level0_out = lp_build_clamp(int_coord_bld, level,
-                                int_coord_bld->zero,
-                                last_level);
-   /* compute level 1 and clamp to legal range of levels */
-   *level1_out = lp_build_add(int_coord_bld, *level0_out, int_coord_bld->one);
-   *level1_out = lp_build_min(int_coord_bld, *level1_out, int_coord_bld->zero);
-
-   *weight_out = lp_build_fract(coord_bld, lod);
-}
-
-
-
 /**
  * Build texture sampling code.
  * 'texel' will return a vector of four LLVMValueRefs corresponding to
  * R, G, B, A.
+ * \param type  vector float type to use for coords, etc.
  */
 void
 lp_build_sample_soa(LLVMBuilderRef builder,
@@ -1245,17 +1357,19 @@ lp_build_sample_soa(LLVMBuilderRef builder,
                     LLVMValueRef *texel)
 {
    struct lp_build_sample_context bld;
-   LLVMValueRef width;
-   LLVMValueRef height;
-   LLVMValueRef stride;
-   LLVMValueRef data_ptr;
+   LLVMValueRef width, width_vec;
+   LLVMValueRef height, height_vec;
+   LLVMValueRef stride, stride_vec;
+   LLVMValueRef data_array;
    LLVMValueRef s;
    LLVMValueRef t;
    LLVMValueRef r;
+   boolean done = FALSE;
 
    (void) lp_build_lod_selector;   /* temporary to silence warning */
    (void) lp_build_nearest_mip_level;
    (void) lp_build_linear_mip_levels;
+   (void) lp_build_minify;
 
    /* Setup our build context */
    memset(&bld, 0, sizeof bld);
@@ -1263,10 +1377,16 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    bld.static_state = static_state;
    bld.dynamic_state = dynamic_state;
    bld.format_desc = util_format_description(static_state->format);
+
+   bld.float_type = lp_type_float(32);
+   bld.int_type = lp_type_int(32);
    bld.coord_type = type;
    bld.uint_coord_type = lp_uint_type(type);
    bld.int_coord_type = lp_int_type(type);
    bld.texel_type = type;
+
+   lp_build_context_init(&bld.float_bld, builder, bld.float_type);
+   lp_build_context_init(&bld.int_bld, builder, bld.int_type);
    lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
    lp_build_context_init(&bld.uint_coord_bld, builder, bld.uint_coord_type);
    lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
@@ -1276,36 +1396,63 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    width = dynamic_state->width(dynamic_state, builder, unit);
    height = dynamic_state->height(dynamic_state, builder, unit);
    stride = dynamic_state->stride(dynamic_state, builder, unit);
-   data_ptr = dynamic_state->data_ptr(dynamic_state, builder, unit);
+   data_array = dynamic_state->data_ptr(dynamic_state, builder, unit);
+   /* Note that data_array is an array[level] of pointers to texture images */
 
    s = coords[0];
    t = coords[1];
    r = coords[2];
 
-   width = lp_build_broadcast_scalar(&bld.uint_coord_bld, width);
-   height = lp_build_broadcast_scalar(&bld.uint_coord_bld, height);
-   stride = lp_build_broadcast_scalar(&bld.uint_coord_bld, stride);
+   width_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, width);
+   height_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, height);
+   stride_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, stride);
 
    if(static_state->target == PIPE_TEXTURE_1D)
       t = bld.coord_bld.zero;
 
-   switch (static_state->min_img_filter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      lp_build_sample_2d_nearest_soa(&bld, s, t, width, height,
-                                     stride, data_ptr, texel);
+   switch (static_state->min_mip_filter) {
+   case PIPE_TEX_MIPFILTER_NONE:
       break;
-   case PIPE_TEX_FILTER_LINEAR:
-      if(lp_format_is_rgba8(bld.format_desc) &&
-         is_simple_wrap_mode(static_state->wrap_s) &&
-         is_simple_wrap_mode(static_state->wrap_t))
-         lp_build_sample_2d_linear_aos(&bld, s, t, width, height,
-                                       stride, data_ptr, texel);
-      else
-         lp_build_sample_2d_linear_soa(&bld, s, t, width, height,
-                                       stride, data_ptr, texel);
+   case PIPE_TEX_MIPFILTER_NEAREST:
+
+      switch (static_state->min_img_filter) {
+      case PIPE_TEX_FILTER_NEAREST:
+         lp_build_sample_2d_nearest_mip_nearest_soa(&bld, unit,
+                                                    s, t,
+                                                    width, height,
+                                                    width_vec, height_vec,
+                                                    stride_vec,
+                                                    data_array, texel);
+         done = TRUE;
+         break;
+      }
+
+      break;
+   case PIPE_TEX_MIPFILTER_LINEAR:
       break;
    default:
-      assert(0);
+      assert(0 && "invalid mip filter");
+   }
+
+   if (!done) {
+      switch (static_state->min_img_filter) {
+      case PIPE_TEX_FILTER_NEAREST:
+         lp_build_sample_2d_nearest_soa(&bld, s, t, width_vec, height_vec,
+                                        stride_vec, data_array, texel);
+         break;
+      case PIPE_TEX_FILTER_LINEAR:
+         if(lp_format_is_rgba8(bld.format_desc) &&
+            is_simple_wrap_mode(static_state->wrap_s) &&
+            is_simple_wrap_mode(static_state->wrap_t))
+            lp_build_sample_2d_linear_aos(&bld, s, t, width_vec, height_vec,
+                                          stride_vec, data_array, texel);
+         else
+            lp_build_sample_2d_linear_soa(&bld, s, t, width_vec, height_vec,
+                                          stride_vec, data_array, texel);
+         break;
+      default:
+         assert(0);
+      }
    }
 
    /* FIXME: respect static_state->min_mip_filter */;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 5f2c2a54ee9..fbb664d43a0 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -152,8 +152,7 @@ static void lp_exec_mask_init(struct lp_exec_mask *mask, struct lp_build_context
 static void lp_exec_mask_update(struct lp_exec_mask *mask)
 {
    mask->exec_mask = mask->cond_mask;
-   if (mask->cond_stack_size > 0)
-      mask->has_mask = TRUE;
+   mask->has_mask = (mask->cond_stack_size > 0);
 }
 
 static void lp_exec_mask_cond_push(struct lp_exec_mask *mask,
@@ -384,6 +383,11 @@ emit_store(
       assert(0);
       break;
 
+   case TGSI_FILE_PREDICATE:
+      /* FIXME */
+      assert(0);
+      break;
+
    default:
       assert( 0 );
    }
@@ -581,6 +585,17 @@ emit_instruction(
    if (indirect_temp_reference(inst))
       return FALSE;
 
+   /*
+    * Stores and write masks are handled in a general fashion after the long
+    * instruction opcode switch statement.
+    *
+    * Although not stricitly necessary, we avoid generating instructions for
+    * channels which won't be stored, in cases where's that easy. For some
+    * complex instructions, like texture sampling, it is more convenient to
+    * assume a full writemask and then let LLVM optimization passes eliminate
+    * redundant code.
+    */
+
    assert(info->num_dst <= 1);
    if(info->num_dst) {
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
@@ -1126,7 +1141,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TEX:
-      /* XXX what about dst0 writemask? */
       emit_tex( bld, inst, FALSE, FALSE, dst0 );
       break;
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.h b/src/gallium/auxiliary/gallivm/lp_bld_type.h
index 16946cc28a2..4daa904e633 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_type.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_type.h
@@ -103,7 +103,7 @@ struct lp_type {
    unsigned width:14;
 
    /**
-    * Vector length.
+    * Vector length.  If length==1, this is a scalar (float/int) type.
     *
     * width*length should be a power of two greater or equal to eight.
     *
@@ -139,11 +139,28 @@ struct lp_build_context
 };
 
 
+/** Create scalar float type */
 static INLINE struct lp_type
 lp_type_float(unsigned width)
 {
    struct lp_type res_type;
 
+   memset(&res_type, 0, sizeof res_type);
+   res_type.floating = TRUE;
+   res_type.sign = TRUE;
+   res_type.width = width;
+   res_type.length = 1;
+
+   return res_type;
+}
+
+
+/** Create vector of float type */
+static INLINE struct lp_type
+lp_type_float_vec(unsigned width)
+{
+   struct lp_type res_type;
+
    memset(&res_type, 0, sizeof res_type);
    res_type.floating = TRUE;
    res_type.sign = TRUE;
@@ -154,11 +171,27 @@ lp_type_float(unsigned width)
 }
 
 
+/** Create scalar int type */
 static INLINE struct lp_type
 lp_type_int(unsigned width)
 {
    struct lp_type res_type;
 
+   memset(&res_type, 0, sizeof res_type);
+   res_type.sign = TRUE;
+   res_type.width = width;
+   res_type.length = 1;
+
+   return res_type;
+}
+
+
+/** Create vector int type */
+static INLINE struct lp_type
+lp_type_int_vec(unsigned width)
+{
+   struct lp_type res_type;
+
    memset(&res_type, 0, sizeof res_type);
    res_type.sign = TRUE;
    res_type.width = width;
@@ -168,11 +201,26 @@ lp_type_int(unsigned width)
 }
 
 
+/** Create scalar uint type */
 static INLINE struct lp_type
 lp_type_uint(unsigned width)
 {
    struct lp_type res_type;
 
+   memset(&res_type, 0, sizeof res_type);
+   res_type.width = width;
+   res_type.length = 1;
+
+   return res_type;
+}
+
+
+/** Create vector uint type */
+static INLINE struct lp_type
+lp_type_uint_vec(unsigned width)
+{
+   struct lp_type res_type;
+
    memset(&res_type, 0, sizeof res_type);
    res_type.width = width;
    res_type.length = LP_NATIVE_VECTOR_WIDTH / width;
diff --git a/src/gallium/auxiliary/target-helpers/wrap_screen.c b/src/gallium/auxiliary/target-helpers/wrap_screen.c
new file mode 100644
index 00000000000..5fe30139386
--- /dev/null
+++ b/src/gallium/auxiliary/target-helpers/wrap_screen.c
@@ -0,0 +1,65 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * 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 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell
+ */
+
+#include "target-helpers/wrap_screen.h"
+#include "trace/tr_public.h"
+#include "identity/id_public.h"
+#include "util/u_debug.h"
+
+
+/* Centralized code to inject common wrapping layers:
+ */
+struct pipe_screen *
+gallium_wrap_screen( struct pipe_screen *screen )
+{
+   /* Screen wrapping functions are required not to fail.  If it is
+    * impossible to wrap a screen, the unwrapped screen should be
+    * returned instead.  Any failure condition should be returned in
+    * an OUT argument.
+    *
+    * Otherwise it is really messy trying to clean up in this code.
+    */
+   if (debug_get_bool_option("GALLIUM_WRAP", FALSE)) {
+      screen = identity_screen_create(screen);
+   }
+
+   if (debug_get_bool_option("GALLIUM_TRACE", FALSE)) {
+      screen = trace_screen_create( screen );
+   }
+
+   return screen;
+}
+
+
+
+
diff --git a/src/gallium/auxiliary/target-helpers/wrap_screen.h b/src/gallium/auxiliary/target-helpers/wrap_screen.h
new file mode 100644
index 00000000000..7e76beb7c5a
--- /dev/null
+++ b/src/gallium/auxiliary/target-helpers/wrap_screen.h
@@ -0,0 +1,16 @@
+#ifndef WRAP_SCREEN_HELPER_H
+#define WRAP_SCREEN_HELPER_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_screen;
+
+/* Centralized code to inject common wrapping layers.  Other layers
+ * can be introduced by specific targets, but these are the generally
+ * helpful ones we probably want everywhere.
+ */
+struct pipe_screen *
+gallium_wrap_screen( struct pipe_screen *screen );
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
index 0b263a9db5c..4d0737ccd3d 100644
--- a/src/gallium/auxiliary/util/u_blit.c
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -63,6 +63,7 @@ struct blit_state
    struct pipe_sampler_state sampler;
    struct pipe_viewport_state viewport;
    struct pipe_clip_state clip;
+   struct pipe_vertex_element velem[2];
 
    void *vs;
    void *fs[TGSI_WRITEMASK_XYZW + 1];
@@ -114,6 +115,15 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
    ctx->sampler.mag_img_filter = 0; /* set later */
    ctx->sampler.normalized_coords = 1;
 
+   /* vertex elements state */
+   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      ctx->velem[i].src_offset = i * 4 * sizeof(float);
+      ctx->velem[i].instance_divisor = 0;
+      ctx->velem[i].vertex_buffer_index = 0;
+      ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    /* vertex shader - still required to provide the linkage between
     * fragment shader input semantics and vertex_element/buffers.
     */
@@ -410,12 +420,14 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    cso_save_fragment_shader(ctx->cso);
    cso_save_vertex_shader(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* set misc state we care about */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    /* sampler */
    ctx->sampler.min_img_filter = filter;
@@ -480,6 +492,7 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    cso_restore_fragment_shader(ctx->cso);
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 
    pipe_texture_reference(&tex, NULL);
 }
@@ -564,12 +577,14 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_save_fragment_shader(ctx->cso);
    cso_save_vertex_shader(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* set misc state we care about */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    /* sampler */
    ctx->sampler.min_img_filter = filter;
@@ -628,4 +643,5 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_restore_fragment_shader(ctx->cso);
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 }
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index 0ba09d33bfc..33d09085f0b 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -88,6 +88,8 @@ struct blitter_context_priv
    void *dsa_write_depth_keep_stencil;
    void *dsa_keep_depth_stencil;
 
+   void *velem_state;
+
    /* Sampler state for clamping to a miplevel. */
    void *sampler_state[PIPE_MAX_TEXTURE_LEVELS];
 
@@ -108,6 +110,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    struct pipe_depth_stencil_alpha_state dsa = { { 0 } };
    struct pipe_rasterizer_state rs_state = { 0 };
    struct pipe_sampler_state *sampler_state;
+   struct pipe_vertex_element velem[2];
    unsigned i;
 
    ctx = CALLOC_STRUCT(blitter_context_priv);
@@ -122,6 +125,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    ctx->blitter.saved_rs_state = INVALID_PTR;
    ctx->blitter.saved_fs = INVALID_PTR;
    ctx->blitter.saved_vs = INVALID_PTR;
+   ctx->blitter.saved_velem_state = INVALID_PTR;
    ctx->blitter.saved_fb_state.nr_cbufs = ~0;
    ctx->blitter.saved_num_textures = ~0;
    ctx->blitter.saved_num_sampler_states = ~0;
@@ -170,6 +174,16 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    rs_state.flatshade = 1;
    ctx->rs_state = pipe->create_rasterizer_state(pipe, &rs_state);
 
+   /* vertex elements state */
+   memset(&velem[0], 0, sizeof(velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      velem[i].src_offset = i * 4 * sizeof(float);
+      velem[i].instance_divisor = 0;
+      velem[i].vertex_buffer_index = 0;
+      velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+   ctx->velem_state = pipe->create_vertex_elements_state(pipe, 2, &velem[0]);
+
    /* fragment shaders are created on-demand */
 
    /* vertex shaders */
@@ -219,6 +233,7 @@ void util_blitter_destroy(struct blitter_context *blitter)
    pipe->delete_rasterizer_state(pipe, ctx->rs_state);
    pipe->delete_vs_state(pipe, ctx->vs_col);
    pipe->delete_vs_state(pipe, ctx->vs_tex);
+   pipe->delete_vertex_elements_state(pipe, ctx->velem_state);
 
    for (i = 0; i < PIPE_MAX_TEXTURE_TYPES; i++) {
       if (ctx->fs_texfetch_col[i])
@@ -246,7 +261,8 @@ static void blitter_check_saved_CSOs(struct blitter_context_priv *ctx)
           ctx->blitter.saved_dsa_state != INVALID_PTR &&
           ctx->blitter.saved_rs_state != INVALID_PTR &&
           ctx->blitter.saved_fs != INVALID_PTR &&
-          ctx->blitter.saved_vs != INVALID_PTR);
+          ctx->blitter.saved_vs != INVALID_PTR &&
+          ctx->blitter.saved_velem_state != INVALID_PTR);
 }
 
 static void blitter_restore_CSOs(struct blitter_context_priv *ctx)
@@ -259,12 +275,14 @@ static void blitter_restore_CSOs(struct blitter_context_priv *ctx)
    pipe->bind_rasterizer_state(pipe, ctx->blitter.saved_rs_state);
    pipe->bind_fs_state(pipe, ctx->blitter.saved_fs);
    pipe->bind_vs_state(pipe, ctx->blitter.saved_vs);
+   pipe->bind_vertex_elements_state(pipe, ctx->blitter.saved_velem_state);
 
    ctx->blitter.saved_blend_state = INVALID_PTR;
    ctx->blitter.saved_dsa_state = INVALID_PTR;
    ctx->blitter.saved_rs_state = INVALID_PTR;
    ctx->blitter.saved_fs = INVALID_PTR;
    ctx->blitter.saved_vs = INVALID_PTR;
+   ctx->blitter.saved_velem_state = INVALID_PTR;
 
    pipe->set_stencil_ref(pipe, &ctx->blitter.saved_stencil_ref);
 
@@ -569,6 +587,7 @@ void util_blitter_clear(struct blitter_context *blitter,
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
 
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, num_cbufs));
    pipe->bind_vs_state(pipe, ctx->vs_col);
 
@@ -634,6 +653,7 @@ static void util_blitter_do_copy(struct blitter_context *blitter,
    pipe->bind_vs_state(pipe, ctx->vs_tex);
    pipe->bind_fragment_sampler_states(pipe, 1,
       blitter_get_sampler_state(ctx, src->level));
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
    pipe->set_fragment_sampler_textures(pipe, 1, &src->texture);
    pipe->set_framebuffer_state(pipe, &fb_state);
 
@@ -807,6 +827,7 @@ void util_blitter_fill(struct blitter_context *blitter,
    pipe->bind_rasterizer_state(pipe, ctx->rs_state);
    pipe->bind_fs_state(pipe, blitter_get_fs_col(ctx, 1));
    pipe->bind_vs_state(pipe, ctx->vs_col);
+   pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
 
    /* set a framebuffer state */
    fb_state.width = dst->width;
diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
index 92008fce992..ecafdabafae 100644
--- a/src/gallium/auxiliary/util/u_blitter.h
+++ b/src/gallium/auxiliary/util/u_blitter.h
@@ -43,6 +43,7 @@ struct blitter_context
    /* Private members, really. */
    void *saved_blend_state;   /**< blend state */
    void *saved_dsa_state;     /**< depth stencil alpha state */
+   void *saved_velem_state;   /**< vertex elements state */
    void *saved_rs_state;      /**< rasterizer state */
    void *saved_fs, *saved_vs; /**< fragment shader, vertex shader */
 
@@ -172,6 +173,13 @@ void util_blitter_save_depth_stencil_alpha(struct blitter_context *blitter,
    blitter->saved_dsa_state = state;
 }
 
+static INLINE
+void util_blitter_save_vertex_elements(struct blitter_context *blitter,
+                                       void *state)
+{
+   blitter->saved_velem_state = state;
+}
+
 static INLINE
 void util_blitter_save_stencil_ref(struct blitter_context *blitter,
                                    const struct pipe_stencil_ref *state)
diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c
index 14506e84519..8c194102bfc 100644
--- a/src/gallium/auxiliary/util/u_draw_quad.c
+++ b/src/gallium/auxiliary/util/u_draw_quad.c
@@ -45,8 +45,6 @@ util_draw_vertex_buffer(struct pipe_context *pipe,
                         uint num_attribs)
 {
    struct pipe_vertex_buffer vbuffer;
-   struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
-   uint i;
 
    assert(num_attribs <= PIPE_MAX_ATTRIBS);
 
@@ -58,15 +56,7 @@ util_draw_vertex_buffer(struct pipe_context *pipe,
    vbuffer.max_index = num_verts - 1;
    pipe->set_vertex_buffers(pipe, 1, &vbuffer);
 
-   /* tell pipe about the vertex attributes */
-   for (i = 0; i < num_attribs; i++) {
-      velements[i].src_offset = i * 4 * sizeof(float);
-      velements[i].instance_divisor = 0;
-      velements[i].vertex_buffer_index = 0;
-      velements[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
-      velements[i].nr_components = 4;
-   }
-   pipe->set_vertex_elements(pipe, num_attribs, velements);
+   /* note: vertex elements already set by caller */
 
    /* draw */
    pipe->draw_arrays(pipe, prim_type, 0, num_verts);
diff --git a/src/gallium/auxiliary/util/u_dump_state.c b/src/gallium/auxiliary/util/u_dump_state.c
index ae7afd7311e..52cf3ef4ce0 100644
--- a/src/gallium/auxiliary/util/u_dump_state.c
+++ b/src/gallium/auxiliary/util/u_dump_state.c
@@ -700,7 +700,6 @@ util_dump_vertex_element(struct os_stream *stream, const struct pipe_vertex_elem
    util_dump_member(stream, uint, state, src_offset);
 
    util_dump_member(stream, uint, state, vertex_buffer_index);
-   util_dump_member(stream, uint, state, nr_components);
 
    util_dump_member(stream, format, state, src_format);
 
diff --git a/src/gallium/auxiliary/util/u_format.h b/src/gallium/auxiliary/util/u_format.h
index e8fa0022b5b..c08fdcafcc8 100644
--- a/src/gallium/auxiliary/util/u_format.h
+++ b/src/gallium/auxiliary/util/u_format.h
@@ -120,8 +120,14 @@ struct util_format_channel_description
 struct util_format_description
 {
    enum pipe_format format;
+
    const char *name;
 
+   /**
+    * Short name, striped of the prefix, lower case.
+    */
+   const char *short_name;
+
    /**
     * Pixel block dimensions.
     */
@@ -139,6 +145,15 @@ struct util_format_description
     */
    unsigned is_array:1;
 
+   /**
+    * Whether the pixel format can be described as a bitfield structure.
+    *
+    * In particular:
+    * - pixel depth must be 8, 16, or 32 bits;
+    * - all channels must be unsigned, signed, or void
+    */
+   unsigned is_bitmask:1;
+
    /**
     * Whether channels have mixed types (ignoring UTIL_FORMAT_TYPE_VOID).
     */
@@ -400,6 +415,16 @@ util_format_has_alpha(enum pipe_format format)
    }
 }
 
+/**
+ * Return the number of components stored.
+ * Formats with block size != 1x1 will always have 1 component (the block).
+ */
+static INLINE unsigned
+util_format_get_nr_components(enum pipe_format format)
+{
+   const struct util_format_description *desc = util_format_description(format);
+   return desc->nr_channels;
+}
 
 /*
  * Format access functions.
diff --git a/src/gallium/auxiliary/util/u_format_pack.py b/src/gallium/auxiliary/util/u_format_pack.py
index 3f33f7cc021..409d024c637 100644
--- a/src/gallium/auxiliary/util/u_format_pack.py
+++ b/src/gallium/auxiliary/util/u_format_pack.py
@@ -252,9 +252,6 @@ def conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=True
     if src_channel.type == FLOAT and dst_channel.type == FLOAT:
         return '(%s)%s' % (dst_native_type, value)
     
-    if not src_channel.norm and not dst_channel.norm:
-        return '(%s)%s' % (dst_native_type, value)
-
     if clamp:
         value = clamp_expr(src_channel, dst_channel, dst_native_type, value)
 
@@ -280,15 +277,15 @@ def conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=True
             value = '(%s * %s)' % (value, scale)
         return '(%s)%s' % (dst_native_type, value)
 
-    if not src_channel.norm and not dst_channel.norm:
-        # neither is normalized -- just cast
-        return '(%s)%s' % (dst_native_type, value)
-
     if src_channel.type in (SIGNED, UNSIGNED) and dst_channel.type in (SIGNED, UNSIGNED):
+        if not src_channel.norm and not dst_channel.norm:
+            # neither is normalized -- just cast
+            return '(%s)%s' % (dst_native_type, value)
+
         src_one = get_one(src_channel)
         dst_one = get_one(dst_channel)
 
-        if src_one > dst_one and src_channel.norm:
+        if src_one > dst_one and src_channel.norm and dst_channel.norm:
             # We can just bitshift
             src_shift = get_one_shift(src_channel)
             dst_shift = get_one_shift(dst_channel)
@@ -296,7 +293,7 @@ def conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=True
         else:
             # We need to rescale using an intermediate type big enough to hold the multiplication of both
             tmp_native_type = intermediate_native_type(src_channel.size + dst_channel.size, src_channel.sign and dst_channel.sign)
-            value = '(%s)%s' % (tmp_native_type, value)
+            value = '((%s)%s)' % (tmp_native_type, value)
             value = '(%s * 0x%x / 0x%x)' % (value, dst_one, src_one)
         value = '(%s)%s' % (dst_native_type, value)
         return value
@@ -307,6 +304,8 @@ def conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=True
 def generate_format_unpack(format, dst_channel, dst_native_type, dst_suffix):
     '''Generate the function to unpack pixels from a particular format'''
 
+    assert format.layout == PLAIN
+
     name = format.short_name()
 
     src_native_type = native_type(format)
@@ -314,32 +313,99 @@ def generate_format_unpack(format, dst_channel, dst_native_type, dst_suffix):
     print 'static INLINE void'
     print 'util_format_%s_unpack_%s(%s *dst, const void *src)' % (name, dst_suffix, dst_native_type)
     print '{'
-    print '   union util_format_%s pixel;' % format.short_name()
-    print '   memcpy(&pixel, src, sizeof pixel);'
-    bswap_format(format)
+    
+    if format.is_bitmask():
+        depth = format.block_size()
+        print '   uint%u_t value = *(uint%u_t *)src;' % (depth, depth) 
 
-    assert format.layout == PLAIN
+        # Declare the intermediate variables
+        for i in range(format.nr_channels()):
+            src_channel = format.channels[i]
+            if src_channel.type == UNSIGNED:
+                print '   uint%u_t %s;' % (depth, src_channel.name)
+            elif src_channel.type == SIGNED:
+                print '   int%u_t %s;' % (depth, src_channel.name)
 
-    for i in range(4):
-        swizzle = format.swizzles[i]
-        if swizzle < 4:
-            src_channel = format.channels[swizzle]
-            value = 'pixel.chan.%s' % src_channel.name 
-            value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
-        elif swizzle == SWIZZLE_0:
-            value = '0'
-        elif swizzle == SWIZZLE_1:
-            value = get_one(dst_channel)
-        elif swizzle == SWIZZLE_NONE:
-            value = '0'
-        else:
-            assert False
-        if format.colorspace == ZS:
-            if i == 3:
+        print '#ifdef PIPE_ARCH_BIG_ENDIAN'
+        print '   value = util_bswap%u(value);' % depth
+        print '#endif'
+
+        # Compute the intermediate unshifted values 
+        shift = 0
+        for i in range(format.nr_channels()):
+            src_channel = format.channels[i]
+            value = 'value'
+            if src_channel.type == UNSIGNED:
+                if shift:
+                    value = '%s >> %u' % (value, shift)
+                if shift + src_channel.size < depth:
+                    value = '(%s) & 0x%x' % (value, (1 << src_channel.size) - 1)
+            elif src_channel.type == SIGNED:
+                if shift + src_channel.size < depth:
+                    # Align the sign bit
+                    lshift = depth - (shift + src_channel.size)
+                    value = '%s << %u' % (value, lshift)
+                # Cast to signed
+                value = '(int%u_t)(%s) ' % (depth, value)
+                if src_channel.size < depth:
+                    # Align the LSB bit
+                    rshift = depth - src_channel.size
+                    value = '(%s) >> %u' % (value, rshift)
+            else:
+                value = None
+                
+            if value is not None:
+                print '   %s = %s;' % (src_channel.name, value)
+                
+            shift += src_channel.size
+
+        # Convert, swizzle, and store final values
+        for i in range(4):
+            swizzle = format.swizzles[i]
+            if swizzle < 4:
+                src_channel = format.channels[swizzle]
+                value = src_channel.name 
+                value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+            elif swizzle == SWIZZLE_0:
+                value = '0'
+            elif swizzle == SWIZZLE_1:
+                value = get_one(dst_channel)
+            elif swizzle == SWIZZLE_NONE:
+                value = '0'
+            else:
+                assert False
+            if format.colorspace == ZS:
+                if i == 3:
+                    value = get_one(dst_channel)
+                elif i >= 1:
+                    value = 'dst[0]'
+            print '   dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i])
+        
+    else:
+        print '   union util_format_%s pixel;' % format.short_name()
+        print '   memcpy(&pixel, src, sizeof pixel);'
+        bswap_format(format)
+    
+        for i in range(4):
+            swizzle = format.swizzles[i]
+            if swizzle < 4:
+                src_channel = format.channels[swizzle]
+                value = 'pixel.chan.%s' % src_channel.name 
+                value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+            elif swizzle == SWIZZLE_0:
+                value = '0'
+            elif swizzle == SWIZZLE_1:
                 value = get_one(dst_channel)
-            elif i >= 1:
-                value = 'dst[0]'
-        print '   dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i])
+            elif swizzle == SWIZZLE_NONE:
+                value = '0'
+            else:
+                assert False
+            if format.colorspace == ZS:
+                if i == 3:
+                    value = get_one(dst_channel)
+                elif i >= 1:
+                    value = 'dst[0]'
+            print '   dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i])
 
     print '}'
     print
@@ -352,31 +418,70 @@ def generate_format_pack(format, src_channel, src_native_type, src_suffix):
 
     dst_native_type = native_type(format)
 
-    print 'static INLINE void'
-    print 'util_format_%s_pack_%s(void *dst, %s r, %s g, %s b, %s a)' % (name, src_suffix, src_native_type, src_native_type, src_native_type, src_native_type)
-    print '{'
-    print '   union util_format_%s pixel;' % format.short_name()
-
     assert format.layout == PLAIN
 
     inv_swizzle = format.inv_swizzles()
+    
+    print 'static INLINE void'
+    print 'util_format_%s_pack_%s(void *dst, %s r, %s g, %s b, %s a)' % (name, src_suffix, src_native_type, src_native_type, src_native_type, src_native_type)
+    print '{'
+    
+    if format.is_bitmask():
+        depth = format.block_size()
+        print '   uint%u_t value = 0;' % depth 
+
+        shift = 0
+        for i in range(4):
+            dst_channel = format.channels[i]
+            if inv_swizzle[i] is not None:
+                value = 'rgba'[inv_swizzle[i]]
+                value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+                if format.colorspace == ZS:
+                    if i == 3:
+                        value = get_one(dst_channel)
+                    elif i >= 1:
+                        value = '0'
+                if dst_channel.type in (UNSIGNED, SIGNED):
+                    if shift + dst_channel.size < depth:
+                        value = '(%s) & 0x%x' % (value, (1 << dst_channel.size) - 1)
+                    if shift:
+                        value = '(%s) << %u' % (value, shift)
+                    if dst_channel.type == SIGNED:
+                        # Cast to unsigned
+                        value = '(uint%u_t)(%s) ' % (depth, value)
+                else:
+                    value = None
+                if value is not None:
+                    print '   value |= %s;' % (value)
+                
+            shift += dst_channel.size
 
-    for i in range(4):
-        dst_channel = format.channels[i]
-        width = dst_channel.size
-        if inv_swizzle[i] is None:
-            continue
-        value = 'rgba'[inv_swizzle[i]]
-        value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
-        if format.colorspace == ZS:
-            if i == 3:
-                value = get_one(dst_channel)
-            elif i >= 1:
-                value = '0'
-        print '   pixel.chan.%s = %s;' % (dst_channel.name, value)
+        print '#ifdef PIPE_ARCH_BIG_ENDIAN'
+        print '   value = util_bswap%u(value);' % depth
+        print '#endif'
+        
+        print '   *(uint%u_t *)dst = value;' % depth 
 
-    bswap_format(format)
-    print '   memcpy(dst, &pixel, sizeof pixel);'
+    else:
+        print '   union util_format_%s pixel;' % format.short_name()
+    
+        for i in range(4):
+            dst_channel = format.channels[i]
+            width = dst_channel.size
+            if inv_swizzle[i] is None:
+                continue
+            value = 'rgba'[inv_swizzle[i]]
+            value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+            if format.colorspace == ZS:
+                if i == 3:
+                    value = get_one(dst_channel)
+                elif i >= 1:
+                    value = '0'
+            print '   pixel.chan.%s = %s;' % (dst_channel.name, value)
+    
+        bswap_format(format)
+        print '   memcpy(dst, &pixel, sizeof pixel);'
+        
     print '}'
     print
     
diff --git a/src/gallium/auxiliary/util/u_format_parse.py b/src/gallium/auxiliary/util/u_format_parse.py
index 250926418ec..f74dc5e88a4 100755
--- a/src/gallium/auxiliary/util/u_format_parse.py
+++ b/src/gallium/auxiliary/util/u_format_parse.py
@@ -78,7 +78,7 @@ class Channel:
         if self.type == UNSIGNED:
             return (1 << self.size) - 1
         if self.type == SIGNED:
-            return self.size - 1
+            return (1 << (self.size - 1)) - 1
         assert False
     
     def min(self):
@@ -166,17 +166,11 @@ class Format:
         return True
 
     def is_bitmask(self):
-        if self.block_size() > 32:
-            return False
-        if not self.is_pot():
+        if self.block_size() not in (8, 16, 32):
             return False
         for channel in self.channels:
-            if not is_pot(channel.size):
-                return True
             if channel.type not in (VOID, UNSIGNED, SIGNED):
                 return False
-            if channel.size >= 32:
-                return False
         return True
 
     def inv_swizzles(self):
diff --git a/src/gallium/auxiliary/util/u_format_table.py b/src/gallium/auxiliary/util/u_format_table.py
index 4e29d15f3bb..fb68852a530 100755
--- a/src/gallium/auxiliary/util/u_format_table.py
+++ b/src/gallium/auxiliary/util/u_format_table.py
@@ -90,11 +90,13 @@ def write_format_table(formats):
     print 'util_format_none_description = {'
     print "   PIPE_FORMAT_NONE,"
     print "   \"PIPE_FORMAT_NONE\","
+    print "   \"none\","
     print "   {0, 0, 0},"
     print "   0,"
     print "   0,"
     print "   0,"
     print "   0,"
+    print "   0,"
     print "   {{0, 0, 0}, {0, 0, 0}, {0, 0, 0}, {0, 0, 0}},"
     print "   {0, 0, 0, 0},"
     print "   0"
@@ -105,10 +107,12 @@ def write_format_table(formats):
         print 'util_format_%s_description = {' % (format.short_name(),)
         print "   %s," % (format.name,)
         print "   \"%s\"," % (format.name,)
+        print "   \"%s\"," % (format.short_name(),)
         print "   {%u, %u, %u},\t/* block */" % (format.block_width, format.block_height, format.block_size())
         print "   %s," % (layout_map(format.layout),)
         print "   %u,\t/* nr_channels */" % (format.nr_channels(),)
         print "   %s,\t/* is_array */" % (bool_map(format.is_array()),)
+        print "   %s,\t/* is_bitmask */" % (bool_map(format.is_bitmask()),)
         print "   %s,\t/* is_mixed */" % (bool_map(format.is_mixed()),)
         print "   {"
         for i in range(4):
diff --git a/src/gallium/auxiliary/util/u_format_tests.c b/src/gallium/auxiliary/util/u_format_tests.c
new file mode 100644
index 00000000000..182a4740448
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_tests.c
@@ -0,0 +1,544 @@
+/**************************************************************************
+ *
+ * Copyright 2009-2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE 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 "u_memory.h"
+#include "u_format_tests.h"
+
+
+/*
+ * Helper macros to create the packed bytes for longer words.
+ */
+
+#define PACKED_1x8(x)          {x, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_2x8(x, y)       {x, y, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_3x8(x, y, z)    {x, y, z, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_4x8(x, y, z, w) {x, y, z, w, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
+
+#define PACKED_1x16(x)          {(x) & 0xff, (x) >> 8,          0,        0,          0,        0,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_2x16(x, y)       {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8,          0,        0,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_3x16(x, y, z)    {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8, (z) & 0xff, (z) >> 8,          0,        0, 0, 0, 0, 0, 0, 0, 0, 0}
+#define PACKED_4x16(x, y, z, w) {(x) & 0xff, (x) >> 8, (y) & 0xff, (y) >> 8, (z) & 0xff, (z) >> 8, (w) & 0xff, (w) >> 8, 0, 0, 0, 0, 0, 0, 0, 0}
+
+#define PACKED_1x32(x)          {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24,          0,                 0,                  0,         0,          0,                 0,                  0,         0,          0,                 0,                  0,         0}
+#define PACKED_2x32(x, y)       {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24,          0,                 0,                  0,         0,          0,                 0,                  0,         0}
+#define PACKED_3x32(x, y, z)    {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24, (z) & 0xff, ((z) >> 8) & 0xff, ((z) >> 16) & 0xff, (z) >> 24,          0,                 0,                  0,         0}
+#define PACKED_4x32(x, y, z, w) {(x) & 0xff, ((x) >> 8) & 0xff, ((x) >> 16) & 0xff, (x) >> 24, (y) & 0xff, ((y) >> 8) & 0xff, ((y) >> 16) & 0xff, (y) >> 24, (z) & 0xff, ((z) >> 8) & 0xff, ((z) >> 16) & 0xff, (z) >> 24, (w) & 0xff, ((w) >> 8) & 0xff, ((w) >> 16) & 0xff, (w) >> 24}
+
+
+/**
+ * Test cases.
+ *
+ * These were manually entered. We could generate these
+ *
+ * To keep this to a we cover only the corner cases, which should produce
+ * good enough coverage since that pixel format transformations are afine for
+ * non SRGB formats.
+ */
+const struct util_format_test_case
+util_format_test_cases[] =
+{
+
+   /*
+    * 32-bit rendertarget formats
+    */
+
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x0000ff00), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0x00ff0000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8X8_UNORM, PACKED_1x32(0x00ffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x0000ff00), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xff000000), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_X8R8G8B8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x0000ff00), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8B8G8R8_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x000000ff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x0000ff00), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0x00ff0000), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xff000000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_X8B8G8R8_UNORM, PACKED_1x32(0xffffff00), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000003ff), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x000ffc00), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x3ff00000), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xc0000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R10G10B10A2_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   /*
+    * 16-bit rendertarget formats
+    */
+
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x03e0), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x7c00), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x8000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x000f), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x00f0), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0f00), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xf000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B4G4R4A4_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x001f), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x07e0), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xf800), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   /*
+    * Luminance/intensity/alpha formats
+    */
+
+   {PIPE_FORMAT_L8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_L8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_A8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {0.0, 0.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_I8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_I8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x00ff), {1.0, 1.0, 1.0, 0.0}},
+   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xff00), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_L8A8_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_L16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_L16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   /*
+    * TODO: SRGB formats
+    */
+
+   /*
+    * Mixed-signed formats
+    */
+
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x00, 0x00, 0x00, 0x00), { 0.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x7f, 0x00, 0x00, 0x00), { 1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x81, 0x00, 0x00, 0x00), {-1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x00, 0x7f, 0x00, 0x00), { 0.0,  1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x00, 0x81, 0x00, 0x00), { 0.0, -1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8SG8SB8UX8U_NORM, PACKED_4x8(0xff, 0xff, 0xff, 0x00), PACKED_4x8(0x00, 0x00, 0xff, 0x00), { 0.0,  0.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), { 0.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x000f), { 1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0011), {-1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x01e0), { 0.0,  1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0x0220), { 0.0, -1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R5SG5SB6U_NORM, PACKED_1x16(0xffff), PACKED_1x16(0xfc00), { 0.0,  0.0, 1.0, 1.0}},
+
+   /*
+    * TODO: Depth-stencil formats
+    */
+
+   /*
+    * TODO: YUV formats
+    */
+
+   /*
+    * TODO: Compressed formats
+    */
+
+   /*
+    * Standard 8-bit integer formats
+    */
+
+   {PIPE_FORMAT_R8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8_UNORM, PACKED_1x8(0xff), PACKED_1x8(0xff), {1.0, 0.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0x00), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0xff), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_UNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0xff), {1.0, 1.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0x00, 0x00), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0xff, 0x00), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0xff), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_UNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0xff, 0xff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0x00, 0x00, 0x00), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0xff, 0x00, 0x00), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0xff, 0x00), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0xff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8A8_UNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0xff, 0xff, 0xff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R8_USCALED, PACKED_1x8(0xff), PACKED_1x8(0x00), {  0.0,   0.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8_USCALED, PACKED_1x8(0xff), PACKED_1x8(0xff), {255.0,   0.0,   0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {  0.0,   0.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0x00), {255.0,   0.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0xff), {  0.0, 255.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_USCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0xff, 0xff), {255.0, 255.0,   0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {  0.0,   0.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0x00, 0x00), {255.0,   0.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0xff, 0x00), {  0.0, 255.0,   0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0xff), {  0.0,   0.0, 255.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_USCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0xff, 0xff, 0xff), {255.0, 255.0, 255.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {  0.0,   0.0,   0.0,   0.0}},
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0x00, 0x00, 0x00), {255.0,   0.0,   0.0,   0.0}},
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0xff, 0x00, 0x00), {  0.0, 255.0,   0.0,   0.0}},
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0xff, 0x00), {  0.0,   0.0, 255.0,   0.0}},
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0xff), {  0.0,   0.0,   0.0, 255.0}},
+   {PIPE_FORMAT_R8G8B8A8_USCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0xff, 0xff, 0xff, 0xff), {255.0, 255.0, 255.0, 255.0}},
+
+   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x00), { 0.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x7f), { 1.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8_SNORM, PACKED_1x8(0xff), PACKED_1x8(0x81), {-1.0,  0.0,  0.0,  1.0}},
+
+   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), { 0.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x7f, 0x00), { 1.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x81, 0x00), {-1.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x7f), { 0.0,  1.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8_SNORM, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x81), { 0.0, -1.0,  0.0,  1.0}},
+
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), { 0.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x7f, 0x00, 0x00), { 1.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x81, 0x00, 0x00), {-1.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x7f, 0x00), { 0.0,  1.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x81, 0x00), { 0.0, -1.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x7f), { 0.0,  0.0,  1.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8_SNORM, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x81), { 0.0,  0.0, -1.0,  1.0}},
+
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), { 0.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x7f, 0x00, 0x00, 0x00), { 1.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x81, 0x00, 0x00, 0x00), {-1.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x7f, 0x00, 0x00), { 0.0,  1.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x81, 0x00, 0x00), { 0.0, -1.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x7f, 0x00), { 0.0,  0.0,  1.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x81, 0x00), { 0.0,  0.0, -1.0,  0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x7f), { 0.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R8G8B8A8_SNORM, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x81), { 0.0,  0.0,  0.0, -1.0}},
+
+   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x00), {   0.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x7f), { 127.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8_SSCALED, PACKED_1x8(0xff), PACKED_1x8(0x80), {-128.0,    0.0,    0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x00), {   0.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x7f, 0x00), { 127.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x80, 0x00), {-128.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x7f), {   0.0,  127.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8_SSCALED, PACKED_2x8(0xff, 0xff), PACKED_2x8(0x00, 0x80), {   0.0, -128.0,    0.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x00), {   0.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x7f, 0x00, 0x00), { 127.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x80, 0x00, 0x00), {-128.0,    0.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x7f, 0x00), {   0.0,  127.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x80, 0x00), {   0.0, -128.0,    0.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x7f), {   0.0,    0.0,  127.0, 1.0}},
+   {PIPE_FORMAT_R8G8B8_SSCALED, PACKED_3x8(0xff, 0xff, 0xff), PACKED_3x8(0x00, 0x00, 0x80), {   0.0,    0.0, -128.0, 1.0}},
+
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x00), {   0.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x7f, 0x00, 0x00, 0x00), { 127.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x80, 0x00, 0x00, 0x00), {-128.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x7f, 0x00, 0x00), {   0.0,  127.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x80, 0x00, 0x00), {   0.0, -128.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x7f, 0x00), {   0.0,    0.0,  127.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x80, 0x00), {   0.0,    0.0, -128.0,    0.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x7f), {   0.0,    0.0,    0.0,  127.0}},
+   {PIPE_FORMAT_R8G8B8A8_SSCALED, PACKED_4x8(0xff, 0xff, 0xff, 0xff), PACKED_4x8(0x00, 0x00, 0x00, 0x80), {   0.0,    0.0,    0.0, -128.0}},
+
+   /*
+    * Standard 16-bit integer formats
+    */
+
+   {PIPE_FORMAT_R16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16_UNORM, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {1.0, 0.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0x0000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0xffff), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16_UNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0xffff), {1.0, 1.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0x0000, 0x0000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0xffff, 0x0000), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0xffff), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_R16G16B16_UNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0xffff, 0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0x0000, 0x0000, 0x0000), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0xffff, 0x0000, 0x0000), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0xffff, 0x0000), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0xffff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R16G16B16A16_UNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R16_USCALED, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {    0.0,     0.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16_USCALED, PACKED_1x16(0xffff), PACKED_1x16(0xffff), {65535.0,     0.0,     0.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {    0.0,     0.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0x0000), {65535.0,     0.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0xffff), {    0.0, 65535.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_USCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0xffff, 0xffff), {65535.0, 65535.0,     0.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {    0.0,     0.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0x0000, 0x0000), {65535.0,     0.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0xffff, 0x0000), {    0.0, 65535.0,     0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0xffff), {    0.0,     0.0, 65535.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_USCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0xffff, 0xffff, 0xffff), {65535.0, 65535.0, 65535.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {    0.0,     0.0,     0.0,     0.0}},
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0x0000, 0x0000, 0x0000), {65535.0,     0.0,     0.0,     0.0}},
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0xffff, 0x0000, 0x0000), {    0.0, 65535.0,     0.0,     0.0}},
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0xffff, 0x0000), {    0.0,     0.0, 65535.0,     0.0}},
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0xffff), {    0.0,     0.0,     0.0, 65535.0}},
+   {PIPE_FORMAT_R16G16B16A16_USCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), {65535.0, 65535.0, 65535.0, 65535.0}},
+
+   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x7fff), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16_SNORM, PACKED_1x16(0xffff), PACKED_1x16(0x8001), {  -1.0,    0.0,    0.0,    1.0}},
+
+   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x7fff, 0x0000), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x8001, 0x0000), {  -1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x7fff), {   0.0,    1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16_SNORM, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x8001), {   0.0,   -1.0,    0.0,    1.0}},
+
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x7fff, 0x0000, 0x0000), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x8001, 0x0000, 0x0000), {  -1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x7fff, 0x0000), {   0.0,    1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x8001, 0x0000), {   0.0,   -1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x7fff), {   0.0,    0.0,    1.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16_SNORM, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x8001), {   0.0,    0.0,   -1.0,    1.0}},
+
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {   0.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x7fff, 0x0000, 0x0000, 0x0000), {   1.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x8001, 0x0000, 0x0000, 0x0000), {  -1.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x7fff, 0x0000, 0x0000), {   0.0,    1.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x8001, 0x0000, 0x0000), {   0.0,   -1.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x7fff, 0x0000), {   0.0,    0.0,    1.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x8001, 0x0000), {   0.0,    0.0,   -1.0,    0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x7fff), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R16G16B16A16_SNORM, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x8001), {   0.0,    0.0,    0.0,   -1.0}},
+
+   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x0000), {     0.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x7fff), { 32767.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16_SSCALED, PACKED_1x16(0xffff), PACKED_1x16(0x8000), {-32768.0,      0.0,      0.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x0000), {     0.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x7fff, 0x0000), { 32767.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x8000, 0x0000), {-32768.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x7fff), {     0.0,  32767.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16_SSCALED, PACKED_2x16(0xffff, 0xffff), PACKED_2x16(0x0000, 0x8000), {     0.0, -32768.0,      0.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x0000), {     0.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x7fff, 0x0000, 0x0000), { 32767.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x8000, 0x0000, 0x0000), {-32768.0,      0.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x7fff, 0x0000), {     0.0,  32767.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x8000, 0x0000), {     0.0, -32768.0,      0.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x7fff), {     0.0,      0.0,  32767.0,   1.0}},
+   {PIPE_FORMAT_R16G16B16_SSCALED, PACKED_3x16(0xffff, 0xffff, 0xffff), PACKED_3x16(0x0000, 0x0000, 0x8000), {     0.0,      0.0, -32768.0,   1.0}},
+
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x0000), {     0.0,      0.0,      0.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x7fff, 0x0000, 0x0000, 0x0000), { 32767.0,      0.0,      0.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x8000, 0x0000, 0x0000, 0x0000), {-32768.0,      0.0,      0.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x7fff, 0x0000, 0x0000), {     0.0,  32767.0,      0.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x8000, 0x0000, 0x0000), {     0.0, -32768.0,      0.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x7fff, 0x0000), {     0.0,      0.0,  32767.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x8000, 0x0000), {     0.0,      0.0, -32768.0,      0.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x7fff), {     0.0,      0.0,      0.0,  32767.0}},
+   {PIPE_FORMAT_R16G16B16A16_SSCALED, PACKED_4x16(0xffff, 0xffff, 0xffff, 0xffff), PACKED_4x16(0x0000, 0x0000, 0x0000, 0x8000), {     0.0,      0.0,      0.0, -32768.0}},
+
+   /*
+    * Standard 32-bit integer formats
+    *
+    * NOTE: We can't accurately represent integers larger than +/-0x1000000
+    * with single precision floats, so that's as far as we test.
+    */
+
+   {PIPE_FORMAT_R32_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32_UNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0xffffffff), {1.0, 0.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xffffffff, 0x00000000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xffffffff), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_UNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xffffffff, 0xffffffff), {1.0, 1.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xffffffff, 0x00000000, 0x00000000), {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xffffffff, 0x00000000), {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xffffffff), {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_UNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xffffffff, 0x00000000, 0x00000000, 0x00000000), {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xffffffff, 0x00000000, 0x00000000), {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xffffffff, 0x00000000), {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xffffffff), {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32A32_UNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_R32_USCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {       0.0,        0.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32_USCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x01000000), {16777216.0,        0.0,        0.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {       0.0,        0.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x00000000), {16777216.0,        0.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x01000000), {       0.0, 16777216.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_USCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x01000000), {16777216.0, 16777216.0,        0.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {       0.0,        0.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x00000000, 0x00000000), {16777216.0,        0.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x01000000, 0x00000000), {       0.0, 16777216.0,        0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x01000000), {       0.0,        0.0, 16777216.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_USCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x01000000, 0x01000000), {16777216.0, 16777216.0, 16777216.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {       0.0,        0.0,        0.0,        0.0}},
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x00000000, 0x00000000, 0x00000000), {16777216.0,        0.0,        0.0,        0.0}},
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x01000000, 0x00000000, 0x00000000), {       0.0, 16777216.0,        0.0,        0.0}},
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x01000000, 0x00000000), {       0.0,        0.0, 16777216.0,        0.0}},
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x01000000), {       0.0,        0.0,        0.0, 16777216.0}},
+   {PIPE_FORMAT_R32G32B32A32_USCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x01000000, 0x01000000, 0x01000000), {16777216.0, 16777216.0, 16777216.0, 16777216.0}},
+
+   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x7fffffff), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32_SNORM, PACKED_1x32(0xffffffff), PACKED_1x32(0x80000001), {  -1.0,    0.0,    0.0,    1.0}},
+
+   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x7fffffff, 0x00000000), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x80000001, 0x00000000), {  -1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x7fffffff), {   0.0,    1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32_SNORM, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x80000001), {   0.0,   -1.0,    0.0,    1.0}},
+
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x7fffffff, 0x00000000, 0x00000000), {   1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x80000001, 0x00000000, 0x00000000), {  -1.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x7fffffff, 0x00000000), {   0.0,    1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x80000001, 0x00000000), {   0.0,   -1.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x7fffffff), {   0.0,    0.0,    1.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32_SNORM, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x80000001), {   0.0,    0.0,   -1.0,    1.0}},
+
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {   0.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x7fffffff, 0x00000000, 0x00000000, 0x00000000), {   1.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x80000001, 0x00000000, 0x00000000, 0x00000000), {  -1.0,    0.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x7fffffff, 0x00000000, 0x00000000), {   0.0,    1.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x80000001, 0x00000000, 0x00000000), {   0.0,   -1.0,    0.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x7fffffff, 0x00000000), {   0.0,    0.0,    1.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x80000001, 0x00000000), {   0.0,    0.0,   -1.0,    0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x7fffffff), {   0.0,    0.0,    0.0,    1.0}},
+   {PIPE_FORMAT_R32G32B32A32_SNORM, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x80000001), {   0.0,    0.0,    0.0,   -1.0}},
+
+   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {        0.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0x01000000), { 16777216.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32_SSCALED, PACKED_1x32(0xffffffff), PACKED_1x32(0xff000000), {-16777216.0,         0.0,         0.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), {        0.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x01000000, 0x00000000), { 16777216.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xff000000, 0x00000000), {-16777216.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x01000000), {        0.0,  16777216.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32_SSCALED, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xff000000), {        0.0, -16777216.0,         0.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), {        0.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x01000000, 0x00000000, 0x00000000), { 16777216.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xff000000, 0x00000000, 0x00000000), {-16777216.0,         0.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x01000000, 0x00000000), {        0.0,  16777216.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xff000000, 0x00000000), {        0.0, -16777216.0,         0.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x01000000), {        0.0,         0.0,  16777216.0,   1.0}},
+   {PIPE_FORMAT_R32G32B32_SSCALED, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xff000000), {        0.0,         0.0, -16777216.0,   1.0}},
+
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), {        0.0,         0.0,         0.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x01000000, 0x00000000, 0x00000000, 0x00000000), { 16777216.0,         0.0,         0.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xff000000, 0x00000000, 0x00000000, 0x00000000), {-16777216.0,         0.0,         0.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x01000000, 0x00000000, 0x00000000), {        0.0,  16777216.0,         0.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xff000000, 0x00000000, 0x00000000), {        0.0, -16777216.0,         0.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x01000000, 0x00000000), {        0.0,         0.0,  16777216.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xff000000, 0x00000000), {        0.0,         0.0, -16777216.0,         0.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x01000000), {        0.0,         0.0,         0.0,  16777216.0}},
+   {PIPE_FORMAT_R32G32B32A32_SSCALED, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xff000000), {        0.0,         0.0,         0.0, -16777216.0}},
+
+   /*
+    * Standard 32-bit float formats
+    */
+
+   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0x00000000), {  0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0x3f800000), {  1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32_FLOAT, PACKED_1x32(0xffffffff), PACKED_1x32(0xbf800000), { -1.0, 0.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x00000000), { 0.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x3f800000, 0x00000000), { 1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0xbf800000, 0x00000000), {-1.0,  0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0x3f800000), { 0.0,  1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x00000000, 0xbf800000), { 0.0, -1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_R32G32_FLOAT, PACKED_2x32(0xffffffff, 0xffffffff), PACKED_2x32(0x3f800000, 0x3f800000), { 1.0,  1.0, 0.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x00000000), { 0.0,  0.0,  0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x3f800000, 0x00000000, 0x00000000), { 1.0,  0.0,  0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0xbf800000, 0x00000000, 0x00000000), {-1.0,  0.0,  0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x3f800000, 0x00000000), { 0.0,  1.0,  0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0xbf800000, 0x00000000), { 0.0, -1.0,  0.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0x3f800000), { 0.0,  0.0,  1.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x00000000, 0x00000000, 0xbf800000), { 0.0,  0.0, -1.0, 1.0}},
+   {PIPE_FORMAT_R32G32B32_FLOAT, PACKED_3x32(0xffffffff, 0xffffffff, 0xffffffff), PACKED_3x32(0x3f800000, 0x3f800000, 0x3f800000), { 1.0,  1.0,  1.0, 1.0}},
+
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x00000000), { 0.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x3f800000, 0x00000000, 0x00000000, 0x00000000), { 1.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0xbf800000, 0x00000000, 0x00000000, 0x00000000), {-1.0,  0.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x3f800000, 0x00000000, 0x00000000), { 0.0,  1.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0xbf800000, 0x00000000, 0x00000000), { 0.0, -1.0,  0.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x3f800000, 0x00000000), { 0.0,  0.0,  1.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0xbf800000, 0x00000000), { 0.0,  0.0, -1.0,  0.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0x3f800000), { 0.0,  0.0,  0.0,  1.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x00000000, 0x00000000, 0x00000000, 0xbf800000), { 0.0,  0.0,  0.0, -1.0}},
+   {PIPE_FORMAT_R32G32B32A32_FLOAT, PACKED_4x32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff), PACKED_4x32(0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000), { 1.0,  1.0,  1.0,  1.0}},
+};
+
+
+const unsigned util_format_nr_test_cases = Elements(util_format_test_cases);
diff --git a/src/gallium/auxiliary/util/u_format_tests.h b/src/gallium/auxiliary/util/u_format_tests.h
new file mode 100644
index 00000000000..2d4d9d5fa9d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_tests.h
@@ -0,0 +1,69 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ **************************************************************************/
+
+
+#ifndef U_FORMAT_TESTS_H_
+#define U_FORMAT_TESTS_H_
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+
+
+#define UTIL_FORMAT_MAX_PACKED_BYTES 16
+
+
+/**
+ * A (packed, unpacked) color pair.
+ */
+struct util_format_test_case
+{
+   enum pipe_format format;
+
+   /**
+    * Mask of the bits that actually meaningful data. Used to mask out the
+    * "X" channels.
+    */
+   uint8_t mask[UTIL_FORMAT_MAX_PACKED_BYTES];
+
+   uint8_t packed[UTIL_FORMAT_MAX_PACKED_BYTES];
+
+   /**
+    * RGBA.
+    */
+   double unpacked[4];
+};
+
+
+extern const struct util_format_test_case
+util_format_test_cases[];
+
+
+extern const unsigned util_format_nr_test_cases;
+
+
+#endif /* U_FORMAT_TESTS_H_ */
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
index fc027e48e4e..d421bee8efe 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -62,6 +62,7 @@ struct gen_mipmap_state
    struct pipe_rasterizer_state rasterizer;
    struct pipe_sampler_state sampler;
    struct pipe_clip_state clip;
+   struct pipe_vertex_element velem[2];
 
    void *vs;
    void *fs2d, *fsCube;
@@ -1307,6 +1308,15 @@ util_create_gen_mipmap(struct pipe_context *pipe,
    ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
    ctx->sampler.normalized_coords = 1;
 
+   /* vertex elements state */
+   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
+   for (i = 0; i < 2; i++) {
+      ctx->velem[i].src_offset = i * 4 * sizeof(float);
+      ctx->velem[i].instance_divisor = 0;
+      ctx->velem[i].vertex_buffer_index = 0;
+      ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    /* vertex shader - still needed to specify mapping from fragment
     * shader input semantics to vertex elements 
     */
@@ -1501,12 +1511,14 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    cso_save_vertex_shader(ctx->cso);
    cso_save_viewport(ctx->cso);
    cso_save_clip(ctx->cso);
+   cso_save_vertex_elements(ctx->cso);
 
    /* bind our state */
    cso_set_blend(ctx->cso, &ctx->blend);
    cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_clip(ctx->cso, &ctx->clip);
+   cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
 
    cso_set_fragment_shader_handle(ctx->cso, fs);
    cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
@@ -1593,4 +1605,5 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    cso_restore_vertex_shader(ctx->cso);
    cso_restore_viewport(ctx->cso);
    cso_restore_clip(ctx->cso);
+   cso_restore_vertex_elements(ctx->cso);
 }
diff --git a/src/gallium/auxiliary/util/u_simple_screen.c b/src/gallium/auxiliary/util/u_simple_screen.c
index 53f3c16dbcc..9203cb6580c 100644
--- a/src/gallium/auxiliary/util/u_simple_screen.c
+++ b/src/gallium/auxiliary/util/u_simple_screen.c
@@ -59,22 +59,7 @@ pass_user_buffer_create(struct pipe_screen *screen,
    return buffer;
 }
 
-static struct pipe_buffer *
-pass_surface_buffer_create(struct pipe_screen *screen,
-                           unsigned width, unsigned height,
-                           enum pipe_format format,
-                           unsigned usage,
-                           unsigned tex_usage,
-                           unsigned *stride)
-{
-   struct pipe_buffer *buffer =
-      screen->winsys->surface_buffer_create(screen->winsys, width, height,
-                                            format, usage, tex_usage, stride);
 
-   buffer->screen = screen;
-
-   return buffer;
-}
 
 static void *
 pass_buffer_map(struct pipe_screen *screen,
@@ -135,7 +120,6 @@ u_simple_screen_init(struct pipe_screen *screen)
 {
    screen->buffer_create = pass_buffer_create;
    screen->user_buffer_create = pass_user_buffer_create;
-   screen->surface_buffer_create = pass_surface_buffer_create;
 
    screen->buffer_map = pass_buffer_map;
    screen->buffer_unmap = pass_buffer_unmap;
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
index ba23435f698..6d461cb8800 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.c
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -230,6 +230,7 @@ static bool
 init_pipe_state(struct vl_compositor *c)
 {
    struct pipe_sampler_state sampler;
+   struct pipe_vertex_element vertex_elems[2];
 
    assert(c);
 
@@ -251,15 +252,27 @@ init_pipe_state(struct vl_compositor *c)
    /*sampler.border_color[i] = ;*/
    /*sampler.max_anisotropy = ;*/
    c->sampler = c->pipe->create_sampler_state(c->pipe, &sampler);
-	
+
+   vertex_elems[0].src_offset = 0;
+   vertex_elems[0].instance_divisor = 0;
+   vertex_elems[0].vertex_buffer_index = 0;
+   vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+   vertex_elems[1].src_offset = 0;
+   vertex_elems[1].instance_divisor = 0;
+   vertex_elems[1].vertex_buffer_index = 1;
+   vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+   c->vertex_elems = c->pipe->create_vertex_elements_state(c->pipe, 2, vertex_elems);
+
+
    return true;
 }
 
 static void cleanup_pipe_state(struct vl_compositor *c)
 {
    assert(c);
-	
+
    c->pipe->delete_sampler_state(c->pipe, c->sampler);
+   c->pipe->delete_vertex_elements_state(c->pipe, c->vertex_elems);
 }
 
 static bool
@@ -314,12 +327,6 @@ init_buffers(struct vl_compositor *c)
 
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[0].buffer);
 
-   c->vertex_elems[0].src_offset = 0;
-   c->vertex_elems[0].instance_divisor = 0;
-   c->vertex_elems[0].vertex_buffer_index = 0;
-   c->vertex_elems[0].nr_components = 2;
-   c->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    /*
     * Create our texcoord buffer and texcoord buffer element
     * Texcoord buffer contains the TCs for mapping the rendered surface to the 4 vertices
@@ -344,12 +351,6 @@ init_buffers(struct vl_compositor *c)
 
    pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[1].buffer);
 
-   c->vertex_elems[1].src_offset = 0;
-   c->vertex_elems[1].instance_divisor = 0;
-   c->vertex_elems[1].vertex_buffer_index = 1;
-   c->vertex_elems[1].nr_components = 2;
-   c->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    /*
     * Create our vertex shader's constant buffer
     * Const buffer contains scaling and translation vectors
@@ -483,7 +484,7 @@ void vl_compositor_render(struct vl_compositor          *compositor,
    compositor->pipe->bind_vs_state(compositor->pipe, compositor->vertex_shader);
    compositor->pipe->bind_fs_state(compositor->pipe, compositor->fragment_shader);
    compositor->pipe->set_vertex_buffers(compositor->pipe, 2, compositor->vertex_bufs);
-   compositor->pipe->set_vertex_elements(compositor->pipe, 2, compositor->vertex_elems);
+   compositor->pipe->bind_vertex_elements_state(compositor->pipe, compositor->vertex_elems);
    compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, compositor->vs_const_buf);
    compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, compositor->fs_const_buf);
 
diff --git a/src/gallium/auxiliary/vl/vl_compositor.h b/src/gallium/auxiliary/vl/vl_compositor.h
index 6a9a3fd7af1..51755554da1 100644
--- a/src/gallium/auxiliary/vl/vl_compositor.h
+++ b/src/gallium/auxiliary/vl/vl_compositor.h
@@ -43,10 +43,10 @@ struct vl_compositor
    void *sampler;
    void *vertex_shader;
    void *fragment_shader;
+   void *vertex_elems;
    struct pipe_viewport_state viewport;
    struct pipe_scissor_state scissor;
    struct pipe_vertex_buffer vertex_bufs[2];
-   struct pipe_vertex_element vertex_elems[2];
    struct pipe_buffer *vs_const_buf, *fs_const_buf;
 };
 
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
index f323de0ea55..0763b5bb0e4 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -708,6 +708,7 @@ static bool
 init_pipe_state(struct vl_mpeg12_mc_renderer *r)
 {
    struct pipe_sampler_state sampler;
+   struct pipe_vertex_element vertex_elems[8];
    unsigned filters[5];
    unsigned i;
 
@@ -771,6 +772,59 @@ init_pipe_state(struct vl_mpeg12_mc_renderer *r)
       r->samplers.all[i] = r->pipe->create_sampler_state(r->pipe, &sampler);
    }
 
+   /* Position element */
+   vertex_elems[0].src_offset = 0;
+   vertex_elems[0].instance_divisor = 0;
+   vertex_elems[0].vertex_buffer_index = 0;
+   vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Luma, texcoord element */
+   vertex_elems[1].src_offset = sizeof(struct vertex2f);
+   vertex_elems[1].instance_divisor = 0;
+   vertex_elems[1].vertex_buffer_index = 0;
+   vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cr texcoord element */
+   vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
+   vertex_elems[2].instance_divisor = 0;
+   vertex_elems[2].vertex_buffer_index = 0;
+   vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cb texcoord element */
+   vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
+   vertex_elems[3].instance_divisor = 0;
+   vertex_elems[3].vertex_buffer_index = 0;
+   vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface top field texcoord element */
+   vertex_elems[4].src_offset = 0;
+   vertex_elems[4].instance_divisor = 0;
+   vertex_elems[4].vertex_buffer_index = 1;
+   vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface bottom field texcoord element */
+   vertex_elems[5].src_offset = sizeof(struct vertex2f);
+   vertex_elems[5].instance_divisor = 0;
+   vertex_elems[5].vertex_buffer_index = 1;
+   vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface top field texcoord element */
+   vertex_elems[6].src_offset = 0;
+   vertex_elems[6].instance_divisor = 0;
+   vertex_elems[6].vertex_buffer_index = 2;
+   vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface bottom field texcoord element */
+   vertex_elems[7].src_offset = sizeof(struct vertex2f);
+   vertex_elems[7].instance_divisor = 0;
+   vertex_elems[7].vertex_buffer_index = 2;
+   vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* need versions with 4,6 and 8 vertex elems */
+   r->vertex_elems[0] = r->pipe->create_vertex_elements_state(r->pipe, 4, vertex_elems);
+   r->vertex_elems[1] = r->pipe->create_vertex_elements_state(r->pipe, 6, vertex_elems);
+   r->vertex_elems[2] = r->pipe->create_vertex_elements_state(r->pipe, 8, vertex_elems);
+
    return true;
 }
 
@@ -783,6 +837,8 @@ cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
 
    for (i = 0; i < 5; ++i)
       r->pipe->delete_sampler_state(r->pipe, r->samplers.all[i]);
+   for (i = 0; i < 3; i++)
+      r->pipe->delete_vertex_elements_state(r->pipe, r->vertex_elems[i]);
 }
 
 static bool
@@ -888,62 +944,6 @@ init_buffers(struct vl_mpeg12_mc_renderer *r)
       );
    }
 
-   /* Position element */
-   r->vertex_elems[0].src_offset = 0;
-   r->vertex_elems[0].instance_divisor = 0;
-   r->vertex_elems[0].vertex_buffer_index = 0;
-   r->vertex_elems[0].nr_components = 2;
-   r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Luma, texcoord element */
-   r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[1].instance_divisor = 0;
-   r->vertex_elems[1].vertex_buffer_index = 0;
-   r->vertex_elems[1].nr_components = 2;
-   r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Chroma Cr texcoord element */
-   r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
-   r->vertex_elems[2].instance_divisor = 0;
-   r->vertex_elems[2].vertex_buffer_index = 0;
-   r->vertex_elems[2].nr_components = 2;
-   r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Chroma Cb texcoord element */
-   r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
-   r->vertex_elems[3].instance_divisor = 0;
-   r->vertex_elems[3].vertex_buffer_index = 0;
-   r->vertex_elems[3].nr_components = 2;
-   r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* First ref surface top field texcoord element */
-   r->vertex_elems[4].src_offset = 0;
-   r->vertex_elems[4].instance_divisor = 0;
-   r->vertex_elems[4].vertex_buffer_index = 1;
-   r->vertex_elems[4].nr_components = 2;
-   r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* First ref surface bottom field texcoord element */
-   r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[5].instance_divisor = 0;
-   r->vertex_elems[5].vertex_buffer_index = 1;
-   r->vertex_elems[5].nr_components = 2;
-   r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Second ref surface top field texcoord element */
-   r->vertex_elems[6].src_offset = 0;
-   r->vertex_elems[6].instance_divisor = 0;
-   r->vertex_elems[6].vertex_buffer_index = 2;
-   r->vertex_elems[6].nr_components = 2;
-   r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-   /* Second ref surface bottom field texcoord element */
-   r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
-   r->vertex_elems[7].instance_divisor = 0;
-   r->vertex_elems[7].vertex_buffer_index = 2;
-   r->vertex_elems[7].nr_components = 2;
-   r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
    r->vs_const_buf = pipe_buffer_create
    (
       r->pipe->screen,
@@ -1307,7 +1307,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[0]);
       r->pipe->set_fragment_sampler_textures(r->pipe, 3, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 3, r->samplers.all);
       r->pipe->bind_vs_state(r->pipe, r->i_vs);
@@ -1320,7 +1320,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->past;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1334,7 +1334,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->past;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1348,7 +1348,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1362,7 +1362,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[1]);
       r->textures.individual.ref[0] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
       r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
@@ -1376,7 +1376,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0) {
       r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[2]);
       r->textures.individual.ref[0] = r->past;
       r->textures.individual.ref[1] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
@@ -1391,7 +1391,7 @@ flush(struct vl_mpeg12_mc_renderer *r)
 
    if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ ) {
       r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
-      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->pipe->bind_vertex_elements_state(r->pipe, r->vertex_elems[2]);
       r->textures.individual.ref[0] = r->past;
       r->textures.individual.ref[1] = r->future;
       r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
index f00b8c7b8b1..a11a3e7307b 100644
--- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -66,8 +66,8 @@ struct vl_mpeg12_mc_renderer
    struct pipe_buffer *vs_const_buf;
    struct pipe_buffer *fs_const_buf;
    struct pipe_framebuffer_state fb_state;
-   struct pipe_vertex_element vertex_elems[8];
-	
+   void *vertex_elems[3];
+
    union
    {
       void *all[5];
diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
index 9080addba44..4608e97adbb 100644
--- a/src/gallium/docs/source/context.rst
+++ b/src/gallium/docs/source/context.rst
@@ -24,6 +24,7 @@ CSO objects handled by the context object:
 * :ref:`Depth, Stencil, & Alpha`: ``*_depth_stencil_alpha_state``
 * :ref:`Shader`: These have two sets of methods. ``*_fs_state`` is for
   fragment shaders, and ``*_vs_state`` is for vertex shaders.
+* :ref:`Vertex Elements`: ``*_vertex_elements_state``
 
 
 Resource Binding State
@@ -60,7 +61,6 @@ objects. They all follow simple, one-method binding calls, e.g.
   not have the scissor test enabled, then the scissor bounds never need to
   be set since they will not be used.
 * ``set_viewport_state``
-* ``set_vertex_elements``
 
 
 Clearing
diff --git a/src/gallium/docs/source/cso/velems.rst b/src/gallium/docs/source/cso/velems.rst
new file mode 100644
index 00000000000..8e758fae103
--- /dev/null
+++ b/src/gallium/docs/source/cso/velems.rst
@@ -0,0 +1,24 @@
+.. _vertex,elements
+
+Vertex Elements
+===============
+
+This state controls format etc. of the input attributes contained
+in the pipe_vertex_buffer(s). There's one pipe_vertex_element array member
+for each input attribute.
+
+Members
+-------
+
+src_offset
+    The byte offset of the attribute in the buffer given by
+    vertex_buffer_index for the first vertex.
+instance_divisor
+    The instance data rate divisor, used for instancing.
+    0 means this is per-vertex data, n means per-instance data used for
+    n consecutive instances (n > 0).
+vertex_buffer_index
+    The vertex buffer this attribute lives in. Several attributes may
+    live in the same vertex buffer.
+src_format
+    The format of the attribute data. One of the PIPE_FORMAT tokens.
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index c92f8e5cba2..8769b826b5f 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -21,6 +21,7 @@ SPU_CODE_MODULE = ../spu/g3d_spu.a
 
 SOURCES = \
 	cell_batch.c \
+	cell_buffer.c \
 	cell_clear.c \
 	cell_context.c \
 	cell_draw_arrays.c \
diff --git a/src/gallium/drivers/cell/ppu/cell_buffer.c b/src/gallium/drivers/cell/ppu/cell_buffer.c
new file mode 100644
index 00000000000..f56a28d62e2
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_buffer.c
@@ -0,0 +1,118 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE 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 "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "cell_screen.h"
+#include "cell_buffer.h"
+
+
+static void *
+cell_buffer_map(struct pipe_screen *screen,
+                    struct pipe_buffer *buf,
+                    unsigned flags)
+{
+   struct cell_buffer *cell_buf = cell_buffer(buf);
+   return cell_buf->data;
+}
+
+
+static void
+cell_buffer_unmap(struct pipe_screen *screen,
+                      struct pipe_buffer *buf)
+{
+}
+
+
+static void
+cell_buffer_destroy(struct pipe_buffer *buf)
+{
+   struct cell_buffer *sbuf = cell_buffer(buf);
+
+   if (!sbuf->userBuffer)
+      align_free(sbuf->data);
+      
+   FREE(sbuf);
+}
+
+
+static struct pipe_buffer *
+cell_buffer_create(struct pipe_screen *screen,
+                       unsigned alignment,
+                       unsigned usage,
+                       unsigned size)
+{
+   struct cell_buffer *buffer = CALLOC_STRUCT(cell_buffer);
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.alignment = MAX2(alignment, 16);
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   buffer->data = align_malloc(size, alignment);
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+cell_user_buffer_create(struct pipe_screen *screen,
+                            void *ptr,
+                            unsigned bytes)
+{
+   struct cell_buffer *buffer;
+
+   buffer = CALLOC_STRUCT(cell_buffer);
+   if(!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+void
+cell_init_screen_buffer_funcs(struct pipe_screen *screen)
+{
+   screen->buffer_create = cell_buffer_create;
+   screen->user_buffer_create = cell_user_buffer_create;
+   screen->buffer_map = cell_buffer_map;
+   screen->buffer_unmap = cell_buffer_unmap;
+   screen->buffer_destroy = cell_buffer_destroy;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_buffer.h b/src/gallium/drivers/cell/ppu/cell_buffer.h
new file mode 100644
index 00000000000..ef0a8a70e59
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_buffer.h
@@ -0,0 +1,55 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE 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 SP_BUFFER_H
+#define SP_BUFFER_H
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+
+
+struct cell_buffer
+{
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
+};
+
+
+/** Cast wrapper */
+static INLINE struct cell_buffer *
+cell_buffer( struct pipe_buffer *buf )
+{
+   return (struct cell_buffer *)buf;
+}
+
+
+void
+cell_init_screen_buffer_funcs(struct pipe_screen *screen);
+
+
+#endif /* SP_BUFFER_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
index 5bff9869fd0..afcea616d59 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.c
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -36,7 +36,6 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_format.h"
 #include "util/u_memory.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_screen.h"
 
 #include "draw/draw_context.h"
@@ -137,7 +136,7 @@ cell_create_context(struct pipe_screen *screen,
    memset(cell, 0, sizeof(*cell));
 
    cell->winsys = NULL;		/* XXX: fixme - get this from screen? */
-   cell->pipe.winsys = screen->winsys;
+   cell->pipe.winsys = NULL;
    cell->pipe.screen = screen;
    cell->pipe.priv = priv;
    cell->pipe.destroy = cell_destroy_context;
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index a77cc5b9067..28f80b82cd5 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -34,7 +34,7 @@
 #include "pipe/p_defines.h"
 #include "draw/draw_vertex.h"
 #include "draw/draw_vbuf.h"
-#include "cell_winsys.h"
+/*#include "cell_winsys.h"*/
 #include "cell/common.h"
 #include "rtasm/rtasm_ppc_spe.h"
 #include "tgsi/tgsi_scan.h"
@@ -93,6 +93,11 @@ struct cell_buffer_list
    struct cell_buffer_node *head;
 };
 
+struct cell_velems_state
+{
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+}
 
 /**
  * Per-context state, subclass of pipe_context.
@@ -110,6 +115,7 @@ struct cell_context
    const struct pipe_rasterizer_state *rasterizer;
    const struct cell_vertex_shader_state *vs;
    const struct cell_fragment_shader_state *fs;
+   const struct cell_velems_state *velems;
 
    struct spe_function logic_op;
 
@@ -125,8 +131,6 @@ struct cell_context
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
    uint num_vertex_buffers;
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-   uint num_vertex_elements;
 
    ubyte *cbuf_map[PIPE_MAX_COLOR_BUFS];
    ubyte *zsbuf_map;
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index bffd0fac6fe..15d4e8338f4 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -33,48 +33,18 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_inlines.h"
 
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
 #include "cell_state.h"
 #include "cell_flush.h"
+#include "cell_buffer.h"
 
 #include "draw/draw_context.h"
 
 
 
-static void
-cell_map_constant_buffers(struct cell_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-   for (i = 0; i < 2; i++) {
-      if (sp->constants[i] && sp->constants[i]->size) {
-         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i],
-                                                   PIPE_BUFFER_USAGE_CPU_READ);
-         cell_flush_buffer_range(sp, sp->mapped_constants[i], 
-                                 sp->constants[i]->size);
-      }
-   }
-
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX, 0,
-                                   sp->mapped_constants[PIPE_SHADER_VERTEX],
-                                   sp->constants[PIPE_SHADER_VERTEX]->size);
-}
-
-static void
-cell_unmap_constant_buffers(struct cell_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-   for (i = 0; i < 2; i++) {
-      if (sp->constants[i] && sp->constants[i]->size)
-         ws->buffer_unmap(ws, sp->constants[i]);
-      sp->mapped_constants[i] = NULL;
-   }
-}
 
 
 
@@ -93,33 +63,27 @@ cell_draw_range_elements(struct pipe_context *pipe,
                          unsigned max_index,
                          unsigned mode, unsigned start, unsigned count)
 {
-   struct cell_context *sp = cell_context(pipe);
-   struct draw_context *draw = sp->draw;
+   struct cell_context *cell = cell_context(pipe);
+   struct draw_context *draw = cell->draw;
    unsigned i;
 
-   if (sp->dirty)
-      cell_update_derived( sp );
+   if (cell->dirty)
+      cell_update_derived( cell );
 
 #if 0
-   cell_map_surfaces(sp);
+   cell_map_surfaces(cell);
 #endif
-   cell_map_constant_buffers(sp);
 
    /*
     * Map vertex buffers
     */
-   for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf = pipe_buffer_map(pipe->screen,
-                                           sp->vertex_buffer[i].buffer,
-                                           PIPE_BUFFER_USAGE_CPU_READ);
-      cell_flush_buffer_range(sp, buf, sp->vertex_buffer[i].buffer->size);
+   for (i = 0; i < cell->num_vertex_buffers; i++) {
+      void *buf = cell_buffer(cell->vertex_buffer[i].buffer)->data;
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes = pipe_buffer_map(pipe->screen,
-                                                      indexBuffer,
-                                                      PIPE_BUFFER_USAGE_CPU_READ);
+      void *mapped_indexes = cell_buffer(indexBuffer)->data;
       draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
    }
    else {
@@ -134,17 +98,19 @@ cell_draw_range_elements(struct pipe_context *pipe,
    /*
     * unmap vertex/index buffers - will cause draw module to flush
     */
-   for (i = 0; i < sp->num_vertex_buffers; i++) {
+   for (i = 0; i < cell->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
-   /* Note: leave drawing surfaces mapped */
-   cell_unmap_constant_buffers(sp);
+   /*
+    * TODO: Flush only when a user vertex/index buffer is present
+    * (or even better, modify draw module to do this
+    * internally when this condition is seen?)
+    */
+   draw_flush(draw);
 }
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_fence.c b/src/gallium/drivers/cell/ppu/cell_fence.c
index e10071529a8..035ef41b898 100644
--- a/src/gallium/drivers/cell/ppu/cell_fence.c
+++ b/src/gallium/drivers/cell/ppu/cell_fence.c
@@ -92,7 +92,6 @@ cell_add_buffer_to_list(struct cell_context *cell,
                         struct cell_buffer_list *list,
                         struct pipe_buffer *buffer)
 {
-   struct pipe_screen *ps = cell->pipe.screen;
    struct cell_buffer_node *node = CALLOC_STRUCT(cell_buffer_node);
    /* create new list node which references the buffer, insert at head */
    if (node) {
@@ -157,8 +156,13 @@ cell_add_fenced_textures(struct cell_context *cell)
          printf("Adding texture %p buffer %p to list\n",
                 ct, ct->tiled_buffer[level]);
 #endif
-         if (ct->buffer)
+#if 00
+         /* XXX this needs to be fixed/restored!
+          * Maybe keep pointers to textures, not buffers.
+          */
+         if (ct->base.buffer)
             cell_add_buffer_to_list(cell, list, ct->buffer);
+#endif
       }
    }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 3d8b4409c75..dce10ae7f86 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -292,17 +292,23 @@ cell_set_sampler_textures(struct pipe_context *pipe,
 static void
 cell_map_surfaces(struct cell_context *cell)
 {
+#if 0
    struct pipe_screen *screen = cell->pipe.screen;
+#endif
    uint i;
 
    for (i = 0; i < 1; i++) {
       struct pipe_surface *ps = cell->framebuffer.cbufs[i];
       if (ps) {
          struct cell_texture *ct = cell_texture(ps->texture);
+#if 0
          cell->cbuf_map[i] = screen->buffer_map(screen,
                                                 ct->buffer,
                                                 (PIPE_BUFFER_USAGE_GPU_READ |
                                                  PIPE_BUFFER_USAGE_GPU_WRITE));
+#else
+         cell->cbuf_map[i] = ct->data;
+#endif
       }
    }
 
@@ -310,10 +316,14 @@ cell_map_surfaces(struct cell_context *cell)
       struct pipe_surface *ps = cell->framebuffer.zsbuf;
       if (ps) {
          struct cell_texture *ct = cell_texture(ps->texture);
+#if 0
          cell->zsbuf_map = screen->buffer_map(screen,
                                               ct->buffer,
                                               (PIPE_BUFFER_USAGE_GPU_READ |
                                                PIPE_BUFFER_USAGE_GPU_WRITE));
+#else
+         cell->zsbuf_map = ct->data;
+#endif
       }
    }
 }
@@ -325,17 +335,17 @@ cell_map_surfaces(struct cell_context *cell)
 static void
 cell_unmap_surfaces(struct cell_context *cell)
 {
-   struct pipe_screen *screen = cell->pipe.screen;
+   /*struct pipe_screen *screen = cell->pipe.screen;*/
    uint i;
 
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       struct pipe_surface *ps = cell->framebuffer.cbufs[i];
       if (ps && cell->cbuf_map[i]) {
-         struct cell_texture *ct = cell_texture(ps->texture);
+         /*struct cell_texture *ct = cell_texture(ps->texture);*/
          assert(ps->texture);
-         assert(ct->buffer);
+         /*assert(ct->buffer);*/
 
-         screen->buffer_unmap(screen, ct->buffer);
+         /*screen->buffer_unmap(screen, ct->buffer);*/
          cell->cbuf_map[i] = NULL;
       }
    }
@@ -343,8 +353,8 @@ cell_unmap_surfaces(struct cell_context *cell)
    {
       struct pipe_surface *ps = cell->framebuffer.zsbuf;
       if (ps && cell->zsbuf_map) {
-         struct cell_texture *ct = cell_texture(ps->texture);
-         screen->buffer_unmap(screen, ct->buffer);
+         /*struct cell_texture *ct = cell_texture(ps->texture);*/
+         /*screen->buffer_unmap(screen, ct->buffer);*/
          cell->zsbuf_map = NULL;
       }
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_public.h b/src/gallium/drivers/cell/ppu/cell_public.h
new file mode 100644
index 00000000000..7e2e093565d
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_public.h
@@ -0,0 +1,10 @@
+#ifndef CELL_PUBLIC_H
+#define CELL_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+cell_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index a43f8638dcd..31fd963d19a 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -28,7 +28,6 @@
 
 #include "util/u_memory.h"
 #include "util/u_simple_screen.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
@@ -36,7 +35,10 @@
 #include "cell_context.h"
 #include "cell_screen.h"
 #include "cell_texture.h"
-#include "cell_winsys.h"
+#include "cell_buffer.h"
+#include "cell_public.h"
+
+#include "state_tracker/sw_winsys.h"
 
 
 static const char *
@@ -134,19 +136,28 @@ cell_is_format_supported( struct pipe_screen *screen,
                           unsigned tex_usage, 
                           unsigned geom_flags )
 {
-   /* cell supports most formats, XXX for now anyway */
+   struct sw_winsys *winsys = cell_screen(screen)->winsys;
+
    if (format == PIPE_FORMAT_DXT5_RGBA ||
        format == PIPE_FORMAT_A8B8G8R8_SRGB)
       return FALSE;
-   else
-      return TRUE;
+
+   if (tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
+      if (!winsys->is_displaytarget_format_supported(winsys, format))
+         return FALSE;
+   }
+
+   /* This is often a lie.  Pull in logic from llvmpipe to fix.
+    */
+   return TRUE;
 }
 
 
 static void
 cell_destroy_screen( struct pipe_screen *screen )
 {
-   struct pipe_winsys *winsys = screen->winsys;
+   struct cell_screen *sp_screen = cell_screen(screen);
+   struct sw_winsys *winsys = sp_screen->winsys;
 
    if(winsys->destroy)
       winsys->destroy(winsys);
@@ -155,32 +166,34 @@ cell_destroy_screen( struct pipe_screen *screen )
 }
 
 
+
 /**
  * Create a new pipe_screen object
  * Note: we're not presently subclassing pipe_screen (no cell_screen) but
  * that would be the place to put SPU thread/context info...
  */
 struct pipe_screen *
-cell_create_screen(struct pipe_winsys *winsys)
+cell_create_screen(struct sw_winsys *winsys)
 {
-   struct pipe_screen *screen = CALLOC_STRUCT(pipe_screen);
+   struct cell_screen *screen = CALLOC_STRUCT(cell_screen);
 
    if (!screen)
       return NULL;
 
    screen->winsys = winsys;
+   screen->base.winsys = NULL;
 
-   screen->destroy = cell_destroy_screen;
+   screen->base.destroy = cell_destroy_screen;
 
-   screen->get_name = cell_get_name;
-   screen->get_vendor = cell_get_vendor;
-   screen->get_param = cell_get_param;
-   screen->get_paramf = cell_get_paramf;
-   screen->is_format_supported = cell_is_format_supported;
-   screen->context_create = cell_create_context;
+   screen->base.get_name = cell_get_name;
+   screen->base.get_vendor = cell_get_vendor;
+   screen->base.get_param = cell_get_param;
+   screen->base.get_paramf = cell_get_paramf;
+   screen->base.is_format_supported = cell_is_format_supported;
+   screen->base.context_create = cell_create_context;
 
-   cell_init_screen_texture_funcs(screen);
-   u_simple_screen_init(screen);
+   cell_init_screen_texture_funcs(&screen->base);
+   cell_init_screen_buffer_funcs(&screen->base);
 
-   return screen;
+   return &screen->base;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.h b/src/gallium/drivers/cell/ppu/cell_screen.h
index c7e15889d66..baff9d3b7d4 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.h
+++ b/src/gallium/drivers/cell/ppu/cell_screen.h
@@ -30,12 +30,26 @@
 #define CELL_SCREEN_H
 
 
-struct pipe_screen;
-struct pipe_winsys;
+#include "pipe/p_screen.h"
 
+struct sw_winsys;
 
-extern struct pipe_screen *
-cell_create_screen(struct pipe_winsys *winsys);
+struct cell_screen {
+   struct pipe_screen base;
+
+   struct sw_winsys *winsys;
+
+   /* Increments whenever textures are modified.  Contexts can track
+    * this.
+    */
+   unsigned timestamp;          
+};
+
+static INLINE struct cell_screen *
+cell_screen( struct pipe_screen *pipe )
+{
+   return (struct cell_screen *)pipe;
+}
 
 
 #endif /* CELL_SCREEN_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index a59c7828ac3..424e2628a95 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -245,16 +245,13 @@ cell_emit_state(struct cell_context *cell)
       uint i, j;
       float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
       uint32_t *ibuf = (uint32_t *) buf;
-      const float *constants = pipe_buffer_map(cell->pipe.screen,
-                                               cell->constants[shader],
-                                               PIPE_BUFFER_USAGE_CPU_READ);
+      const float *constants = cell->mapped_constants[shader];
       ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS;
       ibuf[4] = num_const;
       j = 8;
       for (i = 0; i < num_const; i++) {
          buf[j++] = constants[i];
       }
-      pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader]);
    }
 
    if (cell->dirty & (CELL_NEW_FRAMEBUFFER |
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index 9b2f86fdfba..9e29ddc2d45 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -28,13 +28,13 @@
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
-#include "util/u_simple_screen.h"
 #include "draw/draw_context.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "cell_context.h"
 #include "cell_state.h"
 #include "cell_gen_fp.h"
+#include "cell_buffer.h"
 
 
 /** cast wrapper */
@@ -183,17 +183,29 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 static void
 cell_set_constant_buffer(struct pipe_context *pipe,
                          uint shader, uint index,
-                         struct pipe_buffer *buf)
+                         struct pipe_buffer *constants)
 {
    struct cell_context *cell = cell_context(pipe);
+   unsigned size = constants ? constants->size : 0;
+   const void *data = constants ? cell_buffer(constants)->data : NULL;
 
    assert(shader < PIPE_SHADER_TYPES);
    assert(index == 0);
 
+   if (cell->constants[shader] == constants)
+      return;
+
    draw_flush(cell->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&cell->constants[shader], buf);
+   pipe_buffer_reference(&cell->constants[shader], constants);
+
+   if(shader == PIPE_SHADER_VERTEX) {
+      draw_set_mapped_constant_buffer(cell->draw, PIPE_SHADER_VERTEX, 0,
+                                      data, size);
+   }
+
+   cell->mapped_constants[shader] = data;
 
    if (shader == PIPE_SHADER_VERTEX)
       cell->dirty |= CELL_NEW_VS_CONSTANTS;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_vertex.c b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
index fbe55c84721..9510ea9ac2b 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_vertex.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
@@ -32,24 +32,44 @@
 #include "cell_context.h"
 #include "cell_state.h"
 
+#include "util/u_memory.h"
 #include "draw/draw_context.h"
 
 
-static void
-cell_set_vertex_elements(struct pipe_context *pipe,
-                         unsigned count,
-                         const struct pipe_vertex_element *elements)
+void *
+cell_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
 {
-   struct cell_context *cell = cell_context(pipe);
-
+   struct cell_velems_state *velems;
    assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct cell_velems_state *) MALLOC(sizeof(struct cell_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
+void
+cell_bind_vertex_elements_state(struct pipe_context *pipe,
+                                void *velems)
+{
+   struct cell_context *cell = cell_context(pipe);
+   struct cell_velems_state *cell_velems = (struct cell_velems_state *) velems;
 
-   memcpy(cell->vertex_element, elements, count * sizeof(elements[0]));
-   cell->num_vertex_elements = count;
+   cell->velems = cell_velems;
 
    cell->dirty |= CELL_NEW_VERTEX;
 
-   draw_set_vertex_elements(cell->draw, count, elements);
+   if (cell_velems)
+      draw_set_vertex_elements(cell->draw, cell_velems->count, cell_velems->velem);
+}
+
+void
+cell_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
 }
 
 
@@ -75,5 +95,7 @@ void
 cell_init_vertex_functions(struct cell_context *cell)
 {
    cell->pipe.set_vertex_buffers = cell_set_vertex_buffers;
-   cell->pipe.set_vertex_elements = cell_set_vertex_elements;
+   cell->pipe.create_vertex_elements_state = cell_create_vertex_elements_state;
+   cell->pipe.bind_vertex_elements_state = cell_bind_vertex_elements_state;
+   cell->pipe.delete_vertex_elements_state = cell_delete_vertex_elements_state;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index fad290dfa0e..c8a1acd86a9 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -34,20 +34,23 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "util/u_inlines.h"
-#include "util/u_simple_screen.h"
 
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
 #include "cell_context.h"
+#include "cell_screen.h"
 #include "cell_state.h"
 #include "cell_texture.h"
 
+#include "state_tracker/sw_winsys.h"
 
 
-static void
-cell_texture_layout(struct cell_texture *ct)
+
+static boolean
+cell_texture_layout(struct pipe_screen *screen, 
+                    struct cell_texture *ct)
 {
    struct pipe_texture *pt = &ct->base;
    unsigned level;
@@ -83,9 +86,34 @@ cell_texture_layout(struct cell_texture *ct)
       height = u_minify(height, 1);
       depth = u_minify(depth, 1);
    }
+
+   ct->data = align_malloc(ct->buffer_size, 16);
+ 
+   return ct->data != NULL;
 }
 
 
+/**
+ * Texture layout for simple color buffers.
+ */
+static boolean
+cell_displaytarget_layout(struct pipe_screen *screen,
+                          struct cell_texture * ct)
+{
+   struct sw_winsys *winsys = cell_screen(screen)->winsys;
+
+   /* Round up the surface size to a multiple of the tile size?
+    */
+   ct->dt = winsys->displaytarget_create(winsys,
+                                          ct->base.format,
+                                          ct->base.width0, 
+                                          ct->base.height0,
+                                          16,
+                                          &ct->dt_stride );
+
+   return ct->dt != NULL;
+}
+
 static struct pipe_texture *
 cell_texture_create(struct pipe_screen *screen,
                     const struct pipe_texture *templat)
@@ -98,31 +126,46 @@ cell_texture_create(struct pipe_screen *screen,
    pipe_reference_init(&ct->base.reference, 1);
    ct->base.screen = screen;
 
-   cell_texture_layout(ct);
+   /* Create both a displaytarget (linear) and regular texture
+    * (twiddled).  Convert twiddled->linear at flush_frontbuffer time.
+    */
+   if (ct->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+                             PIPE_TEXTURE_USAGE_SCANOUT |
+                             PIPE_TEXTURE_USAGE_SHARED)) {
+      if (!cell_displaytarget_layout(screen, ct))
+         goto fail;
+   }
 
-   ct->buffer = screen->buffer_create(screen, 32, PIPE_BUFFER_USAGE_PIXEL,
-                                   ct->buffer_size);
+   if (!cell_texture_layout(screen, ct))
+      goto fail;
 
-   if (!ct->buffer) {
-      FREE(ct);
-      return NULL;
+   return &ct->base;
+
+fail:
+   if (ct->dt) {
+      struct sw_winsys *winsys = cell_screen(screen)->winsys;
+      winsys->displaytarget_destroy(winsys, ct->dt);
    }
 
-   return &ct->base;
+   FREE(ct);
+
+   return NULL;
 }
 
 
 static void
 cell_texture_destroy(struct pipe_texture *pt)
 {
+   struct cell_screen *screen = cell_screen(pt->screen);
+   struct sw_winsys *winsys = screen->winsys;
    struct cell_texture *ct = cell_texture(pt);
 
-   if (ct->mapped) {
-      pipe_buffer_unmap(ct->buffer->screen, ct->buffer);
-      ct->mapped = NULL;
+   if (ct->dt) {
+      /* display target */
+      winsys->displaytarget_destroy(winsys, ct->dt);
    }
 
-   pipe_buffer_reference(&ct->buffer, NULL);
+   align_free(ct->data);
 
    FREE(ct);
 }
@@ -389,10 +432,8 @@ cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
 
    assert(transfer->texture);
 
-   if (!ct->mapped) {
-      /* map now */
-      ct->mapped = pipe_buffer_map(screen, ct->buffer,
-                                   pipe_transfer_buffer_flags(transfer));
+   if (ct->mapped == NULL) {
+      ct->mapped = ct->data;
    }
 
    /*
@@ -442,9 +483,8 @@ cell_transfer_unmap(struct pipe_screen *screen,
    const uint stride = ct->stride[level];
 
    if (!ct->mapped) {
-      /* map now */
-      ct->mapped = pipe_buffer_map(screen, ct->buffer,
-                                   PIPE_BUFFER_USAGE_CPU_READ);
+      assert(0);
+      return;
    }
 
    if (transfer->usage & PIPE_TRANSFER_WRITE) {
@@ -467,6 +507,50 @@ cell_transfer_unmap(struct pipe_screen *screen,
 }
 
 
+
+/* This used to be overriden by the co-state tracker, but really needs
+ * to be active with sw_winsys.
+ *
+ * Contrasting with llvmpipe and softpipe, this is the only place
+ * where we use the ct->dt display target in any real sense.
+ *
+ * Basically just untwiddle our local data into the linear
+ * displaytarget.
+ */
+static void
+cell_flush_frontbuffer(struct pipe_screen *_screen,
+                       struct pipe_surface *surface,
+                       void *context_private)
+{
+   struct cell_screen *screen = cell_screen(_screen);
+   struct sw_winsys *winsys = screen->winsys;
+   struct cell_texture *ct = cell_texture(surface->texture);
+
+   if (!ct->dt)
+      return;
+
+   /* Need to untwiddle from our internal representation here:
+    */
+   {
+      unsigned *map = winsys->displaytarget_map(winsys, ct->dt,
+                                                (PIPE_BUFFER_USAGE_CPU_READ |
+                                                 PIPE_BUFFER_USAGE_CPU_WRITE));
+      unsigned *src = (unsigned *)(ct->data + ct->level_offset[surface->level]);
+
+      untwiddle_image_uint(surface->width,
+                           surface->height,
+                           TILE_SIZE,
+                           map,
+                           ct->dt_stride,
+                           src);
+
+      winsys->displaytarget_unmap(winsys, ct->dt);
+   }
+
+   winsys->displaytarget_display(winsys, ct->dt, context_private);
+}
+
+
 void
 cell_init_screen_texture_funcs(struct pipe_screen *screen)
 {
@@ -481,4 +565,6 @@ cell_init_screen_texture_funcs(struct pipe_screen *screen)
 
    screen->transfer_map = cell_transfer_map;
    screen->transfer_unmap = cell_transfer_unmap;
+
+   screen->flush_frontbuffer = cell_flush_frontbuffer;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
index 3ffc0bfdb51..2be05793122 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -28,6 +28,7 @@
 #ifndef CELL_TEXTURE_H
 #define CELL_TEXTURE_H
 
+#include "cell/common.h"
 
 struct cell_context;
 struct pipe_texture;
@@ -43,8 +44,20 @@ struct cell_texture
    unsigned long level_offset[CELL_MAX_TEXTURE_LEVELS];
    unsigned long stride[CELL_MAX_TEXTURE_LEVELS];
 
-   /** The tiled texture data is held in this buffer */
-   struct pipe_buffer *buffer;
+   /**
+    * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
+    * usage.
+    */
+   struct sw_displaytarget *dt;
+   unsigned dt_stride;
+
+   /**
+    * Malloc'ed data for regular textures, or a mapping to dt above.
+    */
+   void *data;
+
+   /* Size of the linear buffer??
+    */
    unsigned long buffer_size;
 
    /** The buffer above, mapped.  This is the memory from which the
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
index cf8cd411598..3d389d6ea36 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -31,7 +31,6 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_math.h"
 
 #include "cell_context.h"
diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.h b/src/gallium/drivers/cell/ppu/cell_winsys.h
deleted file mode 100644
index e227e065ff3..00000000000
--- a/src/gallium/drivers/cell/ppu/cell_winsys.h
+++ /dev/null
@@ -1,47 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-
-#ifndef CELL_WINSYS_H
-#define CELL_WINSYS_H
-
-#include "pipe/p_compiler.h"
-
-
-/**
- * Very simple winsys at this time.
- * Will probably eventually add SPU control info.
- */
-struct cell_winsys
-{
-   uint dummy;
-};
-
-
-
-
-#endif
diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c
index 2ccc5d3e605..659e40cbf03 100644
--- a/src/gallium/drivers/failover/fo_context.c
+++ b/src/gallium/drivers/failover/fo_context.c
@@ -27,7 +27,6 @@
 
 
 #include "pipe/p_defines.h"
-#include "util/u_simple_screen.h"
 #include "util/u_memory.h"
 #include "pipe/p_context.h"
 
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
index bb1a168ea7a..4a754465bbe 100644
--- a/src/gallium/drivers/failover/fo_context.h
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -78,6 +78,7 @@ struct failover_context {
    const struct fo_state     *rasterizer;
    const struct fo_state     *fragment_shader;
    const struct fo_state     *vertex_shader;
+   const struct fo_state     *vertex_elements;
 
    struct pipe_blend_color blend_color;
    struct pipe_stencil_ref stencil_ref;
@@ -89,10 +90,8 @@ struct failover_context {
    struct pipe_texture *vertex_textures[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_elements[PIPE_MAX_ATTRIBS];
 
    uint num_vertex_buffers;
-   uint num_vertex_elements;
 
    void *sw_sampler_state[PIPE_MAX_SAMPLERS];
    void *hw_sampler_state[PIPE_MAX_SAMPLERS];
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
index 970606a3f50..0247fb803b2 100644
--- a/src/gallium/drivers/failover/fo_state.c
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -255,9 +255,52 @@ failover_delete_vs_state(struct pipe_context *pipe,
    free(state);
 }
 
+
+
+static void *
+failover_create_vertex_elements_state( struct pipe_context *pipe,
+                                       unsigned count,
+                                       const struct pipe_vertex_element *velems )
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_vertex_elements_state(failover->sw, count, velems);
+   state->hw_state = failover->hw->create_vertex_elements_state(failover->hw, count, velems);
+
+   return state;
+}
+
+static void
+failover_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems )
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state*)velems;
+
+   failover->vertex_elements = state;
+   failover->dirty |= FO_NEW_VERTEX_ELEMENT;
+   failover->sw->bind_vertex_elements_state( failover->sw, velems );
+   failover->hw->bind_vertex_elements_state( failover->hw, velems );
+}
+
+static void
+failover_delete_vertex_elements_state( struct pipe_context *pipe,
+                                       void *velems )
+{
+   struct fo_state *state = (struct fo_state*)velems;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_vertex_elements_state(failover->sw, state->sw_state);
+   failover->hw->delete_vertex_elements_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
 static void 
 failover_set_polygon_stipple( struct pipe_context *pipe,
-			      const struct pipe_poly_stipple *stipple )
+                              const struct pipe_poly_stipple *stipple )
 {
    struct failover_context *failover = failover_context(pipe);
 
@@ -490,22 +533,6 @@ failover_set_vertex_buffers(struct pipe_context *pipe,
 }
 
 
-static void
-failover_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *vertex_elements)
-{
-   struct failover_context *failover = failover_context(pipe);
-
-   memcpy(failover->vertex_elements, vertex_elements,
-          count * sizeof(vertex_elements[0]));
-
-   failover->dirty |= FO_NEW_VERTEX_ELEMENT;
-   failover->num_vertex_elements = count;
-   failover->sw->set_vertex_elements( failover->sw, count, vertex_elements );
-   failover->hw->set_vertex_elements( failover->hw, count, vertex_elements );
-}
-
 void
 failover_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
@@ -543,6 +570,9 @@ failover_init_state_functions( struct failover_context *failover )
    failover->pipe.create_vs_state = failover_create_vs_state;
    failover->pipe.bind_vs_state   = failover_bind_vs_state;
    failover->pipe.delete_vs_state = failover_delete_vs_state;
+   failover->pipe.create_vertex_elements_state = failover_create_vertex_elements_state;
+   failover->pipe.bind_vertex_elements_state = failover_bind_vertex_elements_state;
+   failover->pipe.delete_vertex_elements_state = failover_delete_vertex_elements_state;
 
    failover->pipe.set_blend_color = failover_set_blend_color;
    failover->pipe.set_stencil_ref = failover_set_stencil_ref;
@@ -554,6 +584,5 @@ failover_init_state_functions( struct failover_context *failover )
    failover->pipe.set_vertex_sampler_textures = failover_set_vertex_sampler_textures;
    failover->pipe.set_viewport_state = failover_set_viewport_state;
    failover->pipe.set_vertex_buffers = failover_set_vertex_buffers;
-   failover->pipe.set_vertex_elements = failover_set_vertex_elements;
    failover->pipe.set_constant_buffer = failover_set_constant_buffer;
 }
diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c
index 5c000808425..09ca1944971 100644
--- a/src/gallium/drivers/failover/fo_state_emit.c
+++ b/src/gallium/drivers/failover/fo_state_emit.c
@@ -81,6 +81,10 @@ failover_state_emit( struct failover_context *failover )
       failover->sw->bind_vs_state( failover->sw,
                                    failover->vertex_shader->sw_state );
 
+   if (failover->dirty & FO_NEW_VERTEX_ELEMENT)
+      failover->sw->bind_vertex_elements_state( failover->sw,
+                                                failover->vertex_elements->sw_state );
+
    if (failover->dirty & FO_NEW_STIPPLE)
       failover->sw->set_polygon_stipple( failover->sw, &failover->poly_stipple );
 
@@ -116,11 +120,5 @@ failover_state_emit( struct failover_context *failover )
                                         failover->vertex_buffers );
    }
 
-   if (failover->dirty & FO_NEW_VERTEX_ELEMENT) {
-      failover->sw->set_vertex_elements( failover->sw,
-                                         failover->num_vertex_elements,
-                                         failover->vertex_elements );
-   }
-
    failover->dirty = 0;
 }
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index da769e7b290..49945376837 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -148,7 +148,7 @@ struct i915_state
 
    /** Describes the current hardware vertex layout */
    struct vertex_info vertex_info;
-   
+
    unsigned id;			/* track lost context events */
 };
 
@@ -187,6 +187,14 @@ struct i915_sampler_state {
    unsigned maxlod;
 };
 
+struct i915_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
+#define I915_MAX_TEXTURE_2D_LEVELS 11  /* max 1024x1024 */
+#define I915_MAX_TEXTURE_3D_LEVELS  8  /* max 128x128x128 */
+
 struct i915_texture {
    struct pipe_texture base;
 
@@ -199,7 +207,7 @@ struct i915_texture {
    unsigned sw_tiled; /**< tiled with software flags */
    unsigned hw_tiled; /**< tiled with hardware fences */
 
-   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned nr_images[I915_MAX_TEXTURE_2D_LEVELS];
 
    /* Explicitly store the offset of each image for each cube face or
     * depth value.  Pretty much have to accept that hardware formats
@@ -207,7 +215,7 @@ struct i915_texture {
     * compute the offsets of depth/cube images within a mipmap level,
     * so have to store them as a lookup table:
     */
-   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];   /**< array [depth] of offsets */
+   unsigned *image_offset[I915_MAX_TEXTURE_2D_LEVELS];   /**< array [depth] of offsets */
 
    /* The data is held here:
     */
@@ -247,7 +255,6 @@ struct i915_context
 
    unsigned num_samplers;
    unsigned num_textures;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    struct intel_batchbuffer *batch;
diff --git a/src/gallium/drivers/i915/i915_debug_fp.c b/src/gallium/drivers/i915/i915_debug_fp.c
index 066e7392d18..f41c51f2991 100644
--- a/src/gallium/drivers/i915/i915_debug_fp.c
+++ b/src/gallium/drivers/i915/i915_debug_fp.c
@@ -28,7 +28,6 @@
 
 #include "i915_reg.h"
 #include "i915_debug.h"
-#include "util/u_simple_screen.h"
 #include "util/u_debug.h"
 
 
diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
index 72bd2635506..e5bf4a20bd0 100644
--- a/src/gallium/drivers/i915/i915_screen.c
+++ b/src/gallium/drivers/i915/i915_screen.c
@@ -116,11 +116,11 @@ i915_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 11; /* max 1024x1024 */
+      return I915_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return I915_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 11; /* max 1024x1024 */
+      return I915_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
       return 1;
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 62169918e2b..377d8425a5c 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -742,21 +742,45 @@ static void i915_set_vertex_buffers(struct pipe_context *pipe,
    draw_set_vertex_buffers(i915->draw, count, buffers);
 }
 
-static void i915_set_vertex_elements(struct pipe_context *pipe,
-                                     unsigned count,
-                                     const struct pipe_vertex_element *elements)
+static void *
+i915_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
+{
+   struct i915_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct i915_velems_state *) MALLOC(sizeof(struct i915_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
+static void
+i915_bind_vertex_elements_state(struct pipe_context *pipe,
+                                void *velems)
 {
    struct i915_context *i915 = i915_context(pipe);
+   struct i915_velems_state *i915_velems = (struct i915_velems_state *) velems;
+
    /* Because we change state before the draw_set_vertex_buffers call
     * we need a flush here, just to be sure.
     */
    draw_flush(i915->draw);
 
-   i915->num_vertex_elements = count;
    /* pass-through to draw module */
-   draw_set_vertex_elements(i915->draw, count, elements);
+   if (i915_velems) {
+      draw_set_vertex_elements(i915->draw,
+            i915_velems->count, i915_velems->velem);
+   }
 }
 
+static void
+i915_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 i915_init_state_functions( struct i915_context *i915 )
@@ -782,6 +806,9 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.create_vs_state = i915_create_vs_state;
    i915->base.bind_vs_state = i915_bind_vs_state;
    i915->base.delete_vs_state = i915_delete_vs_state;
+   i915->base.create_vertex_elements_state = i915_create_vertex_elements_state;
+   i915->base.bind_vertex_elements_state = i915_bind_vertex_elements_state;
+   i915->base.delete_vertex_elements_state = i915_delete_vertex_elements_state;
 
    i915->base.set_blend_color = i915_set_blend_color;
    i915->base.set_stencil_ref = i915_set_stencil_ref;
@@ -794,5 +821,4 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.set_fragment_sampler_textures = i915_set_sampler_textures;
    i915->base.set_viewport_state = i915_set_viewport_state;
    i915->base.set_vertex_buffers = i915_set_vertex_buffers;
-   i915->base.set_vertex_elements = i915_set_vertex_elements;
 }
diff --git a/src/gallium/drivers/i915/i915_texture.c b/src/gallium/drivers/i915/i915_texture.c
index 7ba222c78b7..3ce52cdcdbd 100644
--- a/src/gallium/drivers/i915/i915_texture.c
+++ b/src/gallium/drivers/i915/i915_texture.c
@@ -96,7 +96,7 @@ i915_miptree_set_level_info(struct i915_texture *tex,
                              unsigned nr_images,
                              unsigned w, unsigned h, unsigned d)
 {
-   assert(level < PIPE_MAX_TEXTURE_LEVELS);
+   assert(level < Elements(tex->nr_images));
 
    tex->nr_images[level] = nr_images;
 
@@ -219,12 +219,12 @@ i915_miptree_layout_2d(struct i915_texture *tex)
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->width0);
 
    /* used for scanouts that need special layouts */
-   if (pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+   if (pt->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT)
       if (i915_scanout_layout(tex))
          return;
 
-   /* for shared buffers we use something very like scanout */
-   if (pt->tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+   /* shared buffers needs to be compatible with X servers */
+   if (pt->tex_usage & PIPE_TEXTURE_USAGE_SHARED)
       if (i915_display_target_layout(tex))
          return;
 
@@ -369,12 +369,12 @@ i945_miptree_layout_2d(struct i915_texture *tex)
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->height0);
 
    /* used for scanouts that need special layouts */
-   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_SCANOUT)
       if (i915_scanout_layout(tex))
          return;
 
-   /* for shared buffers we use some very like scanout */
-   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+   /* shared buffers needs to be compatible with X servers */
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_SHARED)
       if (i915_display_target_layout(tex))
          return;
 
@@ -642,7 +642,7 @@ i915_texture_create(struct pipe_screen *screen,
 
 
    /* for scanouts and cursors, cursors arn't scanouts */
-   if (templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY && templat->width0 != 64)
+   if (templat->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT && templat->width0 != 64)
       buf_usage = INTEL_NEW_SCANOUT;
    else
       buf_usage = INTEL_NEW_TEXTURE;
@@ -673,19 +673,24 @@ fail:
 }
 
 static struct pipe_texture *
-i915_texture_blanket(struct pipe_screen * screen,
-                     const struct pipe_texture *base,
-                     const unsigned *stride,
-                     struct pipe_buffer *buffer)
+i915_texture_from_handle(struct pipe_screen * screen,
+                         const struct pipe_texture *templat,
+                         struct winsys_handle *whandle)
 {
-#if 0
+   struct i915_screen *is = i915_screen(screen);
    struct i915_texture *tex;
+   struct intel_winsys *iws = is->iws;
+   struct intel_buffer *buffer;
+   unsigned stride;
+
    assert(screen);
 
+   buffer = iws->buffer_from_handle(iws, whandle, &stride);
+
    /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
+   if (templat->target != PIPE_TEXTURE_2D ||
+       templat->last_level != 0 ||
+       templat->depth0 != 1) {
       return NULL;
    }
 
@@ -693,23 +698,33 @@ i915_texture_blanket(struct pipe_screen * screen,
    if (!tex)
       return NULL;
 
-   tex->base = *base;
+   tex->base = *templat;
    pipe_reference_init(&tex->base.reference, 1);
    tex->base.screen = screen;
 
-   tex->stride = stride[0];
+   tex->stride = stride;
 
-   i915_miptree_set_level_info(tex, 0, 1, base->width0, base->height0, 1);
+   i915_miptree_set_level_info(tex, 0, 1, templat->width0, templat->height0, 1);
    i915_miptree_set_image_offset(tex, 0, 0, 0, 0);
 
-   pipe_buffer_reference(&tex->buffer, buffer);
+   tex->buffer = buffer;
 
    return &tex->base;
-#else
-   return NULL;
-#endif
 }
 
+static boolean
+i915_texture_get_handle(struct pipe_screen * screen,
+                        struct pipe_texture *texture,
+                        struct winsys_handle *whandle)
+{
+   struct i915_screen *is = i915_screen(screen);
+   struct i915_texture *tex = (struct i915_texture *)texture;
+   struct intel_winsys *iws = is->iws;
+
+   return iws->buffer_get_handle(iws, tex->buffer, whandle, tex->stride);
+}
+
+
 static void
 i915_texture_destroy(struct pipe_texture *pt)
 {
@@ -723,7 +738,7 @@ i915_texture_destroy(struct pipe_texture *pt)
 
    iws->buffer_destroy(iws, tex->buffer);
 
-   for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++)
+   for (i = 0; i < Elements(tex->image_offset); i++)
       if (tex->image_offset[i])
          FREE(tex->image_offset[i]);
 
@@ -869,7 +884,8 @@ void
 i915_init_screen_texture_functions(struct i915_screen *is)
 {
    is->base.texture_create = i915_texture_create;
-   is->base.texture_blanket = i915_texture_blanket;
+   is->base.texture_from_handle = i915_texture_from_handle;
+   is->base.texture_get_handle = i915_texture_get_handle;
    is->base.texture_destroy = i915_texture_destroy;
    is->base.get_tex_surface = i915_get_tex_surface;
    is->base.tex_surface_destroy = i915_tex_surface_destroy;
@@ -878,53 +894,3 @@ i915_init_screen_texture_functions(struct i915_screen *is)
    is->base.transfer_unmap = i915_transfer_unmap;
    is->base.tex_transfer_destroy = i915_tex_transfer_destroy;
 }
-
-struct pipe_texture *
-i915_texture_blanket_intel(struct pipe_screen *screen,
-                           struct pipe_texture *base,
-                           unsigned stride,
-                           struct intel_buffer *buffer)
-{
-   struct i915_texture *tex;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   tex = CALLOC_STRUCT(i915_texture);
-   if (!tex)
-      return NULL;
-
-   tex->base = *base;
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
-
-   tex->stride = stride;
-
-   i915_miptree_set_level_info(tex, 0, 1, base->width0, base->height0, 1);
-   i915_miptree_set_image_offset(tex, 0, 0, 0, 0);
-
-   tex->buffer = buffer;
-
-   return &tex->base;
-}
-
-boolean
-i915_get_texture_buffer_intel(struct pipe_texture *texture,
-                              struct intel_buffer **buffer,
-                              unsigned *stride)
-{
-   struct i915_texture *tex = (struct i915_texture *)texture;
-
-   if (!texture)
-      return FALSE;
-
-   *stride = tex->stride;
-   *buffer = tex->buffer;
-
-   return TRUE;
-}
diff --git a/src/gallium/drivers/i915/intel_winsys.h b/src/gallium/drivers/i915/intel_winsys.h
index b3a802b0e29..00fd0c1efea 100644
--- a/src/gallium/drivers/i915/intel_winsys.h
+++ b/src/gallium/drivers/i915/intel_winsys.h
@@ -33,6 +33,7 @@ struct intel_buffer;
 struct intel_batchbuffer;
 struct pipe_texture;
 struct pipe_fence_handle;
+struct winsys_handle;
 
 enum intel_buffer_usage
 {
@@ -128,6 +129,25 @@ struct intel_winsys {
                                          unsigned size, unsigned alignment,
                                          enum intel_buffer_type type);
 
+   /**
+    * Creates a buffer from a handle.
+    * Used to implement pipe_screen::texture_from_handle.
+    * Also provides the stride information needed for the
+    * texture via the stride argument.
+    */
+   struct intel_buffer *(*buffer_from_handle)(struct intel_winsys *iws,
+                                              struct winsys_handle *whandle,
+                                              unsigned *stride);
+
+   /**
+    * Used to implement pipe_screen::texture_get_handle.
+    * The winsys might need the stride information.
+    */
+   boolean (*buffer_get_handle)(struct intel_winsys *iws,
+                                struct intel_buffer *buffer,
+                                struct winsys_handle *whandle,
+                                unsigned stride);
+
    /**
     * Fence a buffer with a fence reg.
     * Not to be confused with pipe_fence_handle.
@@ -204,23 +224,4 @@ struct intel_winsys {
 struct pipe_screen *i915_create_screen(struct intel_winsys *iws, unsigned pci_id);
 
 
-/**
- * Get the intel_winsys buffer backing the texture.
- *
- * TODO UGLY
- */
-boolean i915_get_texture_buffer_intel(struct pipe_texture *texture,
-                                      struct intel_buffer **buffer,
-                                      unsigned *stride);
-
-/**
- * Wrap a intel_winsys buffer with a texture blanket.
- *
- * TODO UGLY
- */
-struct pipe_texture * i915_texture_blanket_intel(struct pipe_screen *screen,
-                                                 struct pipe_texture *tmplt,
-                                                 unsigned pitch,
-                                                 struct intel_buffer *buffer);
-
 #endif
diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h
index 12cfa7b049c..f5b1a06576b 100644
--- a/src/gallium/drivers/i965/brw_context.h
+++ b/src/gallium/drivers/i965/brw_context.h
@@ -351,7 +351,7 @@ struct brw_vs_prog_data {
 
 /* Size == 0 if output either not written, or always [0,0,0,1]
  */
-struct brw_vs_ouput_sizes {
+struct brw_vs_output_sizes {
    GLubyte output_size[PIPE_MAX_SHADER_OUTPUTS];
 };
 
@@ -546,14 +546,13 @@ struct brw_context
       const struct brw_blend_state *blend;
       const struct brw_rasterizer_state *rast;
       const struct brw_depth_stencil_state *zstencil;
+      const struct brw_vertex_element_packet *velems;
 
       const struct brw_sampler *sampler[PIPE_MAX_SAMPLERS];
       unsigned num_samplers;
 
       struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
       struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-      struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-      unsigned num_vertex_elements;
       unsigned num_textures;
       unsigned num_vertex_buffers;
 
diff --git a/src/gallium/drivers/i965/brw_draw_upload.c b/src/gallium/drivers/i965/brw_draw_upload.c
index 9f136eec71c..0820ba20a08 100644
--- a/src/gallium/drivers/i965/brw_draw_upload.c
+++ b/src/gallium/drivers/i965/brw_draw_upload.c
@@ -42,141 +42,6 @@
 
 
 
-static unsigned brw_translate_surface_format( unsigned id )
-{
-   switch (id) {
-   case PIPE_FORMAT_R64_FLOAT:
-      return BRW_SURFACEFORMAT_R64_FLOAT;
-   case PIPE_FORMAT_R64G64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64_FLOAT;
-   case PIPE_FORMAT_R64G64B64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64B64_FLOAT;
-   case PIPE_FORMAT_R64G64B64A64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT;
-
-   case PIPE_FORMAT_R32_FLOAT:
-      return BRW_SURFACEFORMAT_R32_FLOAT;
-   case PIPE_FORMAT_R32G32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32_FLOAT;
-   case PIPE_FORMAT_R32G32B32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32B32_FLOAT;
-   case PIPE_FORMAT_R32G32B32A32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
-
-   case PIPE_FORMAT_R32_UNORM:
-      return BRW_SURFACEFORMAT_R32_UNORM;
-   case PIPE_FORMAT_R32G32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32_UNORM;
-   case PIPE_FORMAT_R32G32B32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32B32_UNORM;
-   case PIPE_FORMAT_R32G32B32A32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32B32A32_UNORM;
-
-   case PIPE_FORMAT_R32_USCALED:
-      return BRW_SURFACEFORMAT_R32_USCALED;
-   case PIPE_FORMAT_R32G32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32_USCALED;
-   case PIPE_FORMAT_R32G32B32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32B32_USCALED;
-   case PIPE_FORMAT_R32G32B32A32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32B32A32_USCALED;
-
-   case PIPE_FORMAT_R32_SNORM:
-      return BRW_SURFACEFORMAT_R32_SNORM;
-   case PIPE_FORMAT_R32G32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32_SNORM;
-   case PIPE_FORMAT_R32G32B32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32B32_SNORM;
-   case PIPE_FORMAT_R32G32B32A32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32B32A32_SNORM;
-
-   case PIPE_FORMAT_R32_SSCALED:
-      return BRW_SURFACEFORMAT_R32_SSCALED;
-   case PIPE_FORMAT_R32G32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32_SSCALED;
-   case PIPE_FORMAT_R32G32B32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32B32_SSCALED;
-   case PIPE_FORMAT_R32G32B32A32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED;
-
-   case PIPE_FORMAT_R16_UNORM:
-      return BRW_SURFACEFORMAT_R16_UNORM;
-   case PIPE_FORMAT_R16G16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16_UNORM;
-   case PIPE_FORMAT_R16G16B16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16B16_UNORM;
-   case PIPE_FORMAT_R16G16B16A16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16B16A16_UNORM;
-
-   case PIPE_FORMAT_R16_USCALED:
-      return BRW_SURFACEFORMAT_R16_USCALED;
-   case PIPE_FORMAT_R16G16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16_USCALED;
-   case PIPE_FORMAT_R16G16B16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16B16_USCALED;
-   case PIPE_FORMAT_R16G16B16A16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16B16A16_USCALED;
-
-   case PIPE_FORMAT_R16_SNORM:
-      return BRW_SURFACEFORMAT_R16_SNORM;
-   case PIPE_FORMAT_R16G16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16_SNORM;
-   case PIPE_FORMAT_R16G16B16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16B16_SNORM;
-   case PIPE_FORMAT_R16G16B16A16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16B16A16_SNORM;
-
-   case PIPE_FORMAT_R16_SSCALED:
-      return BRW_SURFACEFORMAT_R16_SSCALED;
-   case PIPE_FORMAT_R16G16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16_SSCALED;
-   case PIPE_FORMAT_R16G16B16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16B16_SSCALED;
-   case PIPE_FORMAT_R16G16B16A16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED;
-
-   case PIPE_FORMAT_R8_UNORM:
-      return BRW_SURFACEFORMAT_R8_UNORM;
-   case PIPE_FORMAT_R8G8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8_UNORM;
-   case PIPE_FORMAT_R8G8B8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
-   case PIPE_FORMAT_R8G8B8A8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
-
-   case PIPE_FORMAT_R8_USCALED:
-      return BRW_SURFACEFORMAT_R8_USCALED;
-   case PIPE_FORMAT_R8G8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8_USCALED;
-   case PIPE_FORMAT_R8G8B8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8B8_USCALED;
-   case PIPE_FORMAT_R8G8B8A8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8B8A8_USCALED;
-
-   case PIPE_FORMAT_R8_SNORM:
-      return BRW_SURFACEFORMAT_R8_SNORM;
-   case PIPE_FORMAT_R8G8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8_SNORM;
-   case PIPE_FORMAT_R8G8B8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8B8_SNORM;
-   case PIPE_FORMAT_R8G8B8A8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8B8A8_SNORM;
-
-   case PIPE_FORMAT_R8_SSCALED:
-      return BRW_SURFACEFORMAT_R8_SSCALED;
-   case PIPE_FORMAT_R8G8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8_SSCALED;
-   case PIPE_FORMAT_R8G8B8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8B8_SSCALED;
-   case PIPE_FORMAT_R8G8B8A8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED;
-
-   default:
-      assert(0);
-      return 0;
-   }
-}
-
 static unsigned get_index_type(int type)
 {
    switch (type) {
@@ -315,75 +180,16 @@ static int brw_emit_vertex_buffers( struct brw_context *brw )
 
 
 
-
 static int brw_emit_vertex_elements(struct brw_context *brw)
 {
-   GLuint nr = brw->curr.num_vertex_elements;
-   GLuint i;
+   const struct brw_vertex_element_packet *brw_velems = brw->curr.velems;
+   unsigned size = brw_velems->header.length + 2;
 
+   /* why is this here */
    brw_emit_query_begin(brw);
 
-   /* If the VS doesn't read any inputs (calculating vertex position from
-    * a state variable for some reason, for example), emit a single pad
-    * VERTEX_ELEMENT struct and bail.
-    *
-    * The stale VB state stays in place, but they don't do anything unless
-    * a VE loads from them.
-    */
-   if (nr == 0) {
-      BEGIN_BATCH(3, IGNORE_CLIPRECTS);
-      OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
-      OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
-		BRW_VE0_VALID |
-		(BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
-		(0 << BRW_VE0_SRC_OFFSET_SHIFT));
-      OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
-      ADVANCE_BATCH();
-      return 0;
-   }
-
-   /* Now emit vertex element (VEP) state packets.
-    *
-    */
-   BEGIN_BATCH(1 + nr * 2, IGNORE_CLIPRECTS);
-   OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + nr * 2) - 2));
-   for (i = 0; i < nr; i++) {
-      const struct pipe_vertex_element *input = &brw->curr.vertex_element[i];
-      uint32_t format = brw_translate_surface_format( input->src_format );
-      uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
-
-      switch (input->nr_components) {
-      case 0: comp0 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
-      case 1: comp1 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
-      case 2: comp2 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
-      case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
-	 break;
-      }
-
-      OUT_BATCH((input->vertex_buffer_index << BRW_VE0_INDEX_SHIFT) |
-		BRW_VE0_VALID |
-		(format << BRW_VE0_FORMAT_SHIFT) |
-		(input->src_offset << BRW_VE0_SRC_OFFSET_SHIFT));
+   brw_batchbuffer_data(brw->batch, brw_velems, size * 4, IGNORE_CLIPRECTS);
 
-      if (BRW_IS_IGDNG(brw))
-          OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
-                    (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
-                    (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
-                    (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
-      else
-          OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
-                    (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
-                    (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
-                    (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
-                    ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
-   }
-   ADVANCE_BATCH();
    return 0;
 }
 
@@ -396,10 +202,11 @@ static int brw_emit_vertices( struct brw_context *brw )
    if (ret)
       return ret;
 
+   /* XXX should separate this? */
    ret = brw_emit_vertex_elements( brw );
    if (ret)
       return ret;
-   
+
    return 0;
 }
 
@@ -407,7 +214,8 @@ static int brw_emit_vertices( struct brw_context *brw )
 const struct brw_tracked_state brw_vertices = {
    .dirty = {
       .mesa = (PIPE_NEW_INDEX_RANGE |
-               PIPE_NEW_VERTEX_BUFFER),
+               PIPE_NEW_VERTEX_BUFFER |
+               PIPE_NEW_VERTEX_ELEMENT),
       .brw = BRW_NEW_BATCH,
       .cache = 0,
    },
diff --git a/src/gallium/drivers/i965/brw_pipe_vertex.c b/src/gallium/drivers/i965/brw_pipe_vertex.c
index e3c48e31493..d6a840857ec 100644
--- a/src/gallium/drivers/i965/brw_pipe_vertex.c
+++ b/src/gallium/drivers/i965/brw_pipe_vertex.c
@@ -1,22 +1,251 @@
 #include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_structs.h"
 
+#include "util/u_memory.h"
+#include "util/u_format.h"
 
-static void brw_set_vertex_elements( struct pipe_context *pipe,
-				     unsigned count,
-				     const struct pipe_vertex_element *elements )
+
+static unsigned brw_translate_surface_format( unsigned id )
+{
+   switch (id) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return BRW_SURFACEFORMAT_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return BRW_SURFACEFORMAT_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return BRW_SURFACEFORMAT_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return BRW_SURFACEFORMAT_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return BRW_SURFACEFORMAT_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return BRW_SURFACEFORMAT_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return BRW_SURFACEFORMAT_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return BRW_SURFACEFORMAT_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return BRW_SURFACEFORMAT_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return BRW_SURFACEFORMAT_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return BRW_SURFACEFORMAT_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return BRW_SURFACEFORMAT_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return BRW_SURFACEFORMAT_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return BRW_SURFACEFORMAT_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED;
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static void brw_translate_vertex_elements(struct brw_context *brw,
+                                          struct brw_vertex_element_packet *brw_velems,
+                                          const struct pipe_vertex_element *attribs,
+                                          unsigned count)
+{
+   unsigned i;
+
+   /* If the VS doesn't read any inputs (calculating vertex position from
+    * a state variable for some reason, for example), emit a single pad
+    * VERTEX_ELEMENT struct and bail.
+    *
+    * The stale VB state stays in place, but they don't do anything unless
+    * a VE loads from them.
+    */
+   brw_velems->header.opcode = CMD_VERTEX_ELEMENT;
+
+   if (count == 0) {
+      brw_velems->header.length = 1;
+      brw_velems->ve[0].ve0.src_offset = 0;
+      brw_velems->ve[0].ve0.src_format = BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
+      brw_velems->ve[0].ve0.valid = 1;
+      brw_velems->ve[0].ve0.vertex_buffer_index = 0;
+      brw_velems->ve[0].ve1.dst_offset = 0;
+      brw_velems->ve[0].ve1.vfcomponent0 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent1 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent2 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent3 = BRW_VE1_COMPONENT_STORE_1_FLT;
+      return;
+   }
+
+
+   /* Now emit vertex element (VEP) state packets.
+    *
+    */
+   brw_velems->header.length = (1 + count * 2) - 2;
+   for (i = 0; i < count; i++) {
+      const struct pipe_vertex_element *input = &attribs[i];
+      unsigned nr_components = util_format_get_nr_components(input->src_format);
+
+      uint32_t format = brw_translate_surface_format( input->src_format );
+      uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
+
+      switch (nr_components) {
+      case 0: comp0 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 1: comp1 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 2: comp2 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
+         break;
+      }
+
+      brw_velems->ve[i].ve0.src_offset = input->src_offset;
+      brw_velems->ve[i].ve0.src_format = format;
+      brw_velems->ve[i].ve0.valid = 1;
+      brw_velems->ve[i].ve0.vertex_buffer_index = input->vertex_buffer_index;
+      brw_velems->ve[i].ve1.vfcomponent0 = comp0;
+      brw_velems->ve[i].ve1.vfcomponent1 = comp1;
+      brw_velems->ve[i].ve1.vfcomponent2 = comp2;
+      brw_velems->ve[i].ve1.vfcomponent3 = comp3;
+
+      if (BRW_IS_IGDNG(brw))
+         brw_velems->ve[i].ve1.dst_offset = 0;
+      else
+         brw_velems->ve[i].ve1.dst_offset = i * 4;
+   }
+}
+
+static void* brw_create_vertex_elements_state( struct pipe_context *pipe,
+                                               unsigned count,
+                                               const struct pipe_vertex_element *attribs )
 {
+   /* note: for the brw_swtnl.c code (if ever we need draw fallback) we'd also need
+      to store the original data */
    struct brw_context *brw = brw_context(pipe);
+   struct brw_vertex_element_packet *velems;
+   assert(count <= BRW_VEP_MAX);
+   velems = (struct brw_vertex_element_packet *) MALLOC(sizeof(struct brw_vertex_element_packet));
+   if (velems) {
+      brw_translate_vertex_elements(brw, velems, attribs, count);
+   }
+   return velems;
+}
 
-   memcpy(brw->curr.vertex_element, elements, count * sizeof(elements[0]));
-   brw->curr.num_vertex_elements = count;
+static void brw_bind_vertex_elements_state(struct pipe_context *pipe,
+                                           void *velems)
+{
+   struct brw_context *brw = brw_context(pipe);
+   struct brw_vertex_element_packet *brw_velems = (struct brw_vertex_element_packet *) velems;
+
+   brw->curr.velems = brw_velems;
 
    brw->state.dirty.mesa |= PIPE_NEW_VERTEX_ELEMENT;
 }
 
+static void brw_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
+
 
 static void brw_set_vertex_buffers(struct pipe_context *pipe,
-				   unsigned count,
-				   const struct pipe_vertex_buffer *buffers)
+                                   unsigned count,
+                                   const struct pipe_vertex_buffer *buffers)
 {
    struct brw_context *brw = brw_context(pipe);
    unsigned i;
@@ -49,7 +278,9 @@ void
 brw_pipe_vertex_init( struct brw_context *brw )
 {
    brw->base.set_vertex_buffers = brw_set_vertex_buffers;
-   brw->base.set_vertex_elements = brw_set_vertex_elements;
+   brw->base.create_vertex_elements_state = brw_create_vertex_elements_state;
+   brw->base.bind_vertex_elements_state = brw_bind_vertex_elements_state;
+   brw->base.delete_vertex_elements_state = brw_delete_vertex_elements_state;
 }
 
 
diff --git a/src/gallium/drivers/i965/brw_screen.c b/src/gallium/drivers/i965/brw_screen.c
index 66f3aad8b21..cef83ffea80 100644
--- a/src/gallium/drivers/i965/brw_screen.c
+++ b/src/gallium/drivers/i965/brw_screen.c
@@ -174,11 +174,11 @@ brw_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 11; /* max 1024x1024 */
+      return BRW_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return BRW_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 11; /* max 1024x1024 */
+      return BRW_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
       return 1;
diff --git a/src/gallium/drivers/i965/brw_screen.h b/src/gallium/drivers/i965/brw_screen.h
index 7226d9228b7..e0f3cd2a9fe 100644
--- a/src/gallium/drivers/i965/brw_screen.h
+++ b/src/gallium/drivers/i965/brw_screen.h
@@ -100,6 +100,9 @@ struct brw_surface
 };
 
 
+#define BRW_MAX_TEXTURE_2D_LEVELS 11  /* max 1024x1024 */
+#define BRW_MAX_TEXTURE_3D_LEVELS  8  /* max 128x128x128 */
+
 
 struct brw_texture
 {
@@ -107,9 +110,9 @@ struct brw_texture
    struct brw_winsys_buffer *bo;
    struct brw_surface_state ss;
 
-   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned *image_offset[BRW_MAX_TEXTURE_2D_LEVELS];
+   unsigned nr_images[BRW_MAX_TEXTURE_2D_LEVELS];
+   unsigned level_offset[BRW_MAX_TEXTURE_2D_LEVELS];
 
    boolean compressed;
    unsigned brw_target;
diff --git a/src/gallium/drivers/i965/brw_screen_texture.c b/src/gallium/drivers/i965/brw_screen_texture.c
index e38fdf1869b..cc79bfc7715 100644
--- a/src/gallium/drivers/i965/brw_screen_texture.c
+++ b/src/gallium/drivers/i965/brw_screen_texture.c
@@ -231,8 +231,8 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen,
       goto fail;
 
    
-   if (templ->tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                           PIPE_TEXTURE_USAGE_PRIMARY)) {
+   if (templ->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
+                           PIPE_TEXTURE_USAGE_SHARED)) {
       buffer_type = BRW_BUFFER_TYPE_SCANOUT;
    }
    else {
@@ -303,14 +303,121 @@ fail:
    return NULL;
 }
 
-static struct pipe_texture *brw_texture_blanket(struct pipe_screen *screen,
-						const struct pipe_texture *templ,
-						const unsigned *stride,
-						struct pipe_buffer *buffer)
+static struct pipe_texture * 
+brw_texture_from_handle(struct pipe_screen *screen,
+                        const struct pipe_texture *templ,
+                        struct winsys_handle *whandle)
 {
+   struct brw_screen *bscreen = brw_screen(screen);
+   struct brw_texture *tex;
+   struct brw_winsys_buffer *buffer;
+   unsigned tiling;
+   unsigned pitch;
+
+   if (templ->target != PIPE_TEXTURE_2D ||
+       templ->last_level != 0 ||
+       templ->depth0 != 1)
+      return NULL;
+
+   if (util_format_is_compressed(templ->format))
+      return NULL;
+
+   tex = CALLOC_STRUCT(brw_texture);
+   if (!tex)
+      return NULL;
+
+   if (bscreen->sws->bo_from_handle(bscreen->sws, whandle, &pitch, &tiling, &buffer) != PIPE_OK)
+      goto fail;
+
+   memcpy(&tex->base, templ, sizeof *templ);
+   pipe_reference_init(&tex->base.reference, 1);
+   tex->base.screen = screen;
+
+   /* XXX: cpp vs. blocksize
+    */
+   tex->cpp = util_format_get_blocksize(tex->base.format);
+   tex->tiling = tiling;
+
+   make_empty_list(&tex->views[0]);
+   make_empty_list(&tex->views[1]);
+
+   if (!brw_texture_layout(bscreen, tex))
+      goto fail;
+
+   /* XXX Maybe some more checks? */
+   if ((pitch / tex->cpp) < tex->pitch)
+      goto fail;
+
+   tex->pitch = pitch / tex->cpp;
+
+   tex->bo = buffer;
+
+   /* fix this warning */
+#if 0
+   if (tex->size > buffer->size)
+      goto fail;
+#endif
+
+   tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
+   tex->ss.ss0.surface_type = translate_tex_target(tex->base.target);
+   tex->ss.ss0.surface_format = translate_tex_format(tex->base.format);
+   assert(tex->ss.ss0.surface_format != BRW_SURFACEFORMAT_INVALID);
+
+   /* This is ok for all textures with channel width 8bit or less:
+    */
+/*    tex->ss.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
+
+
+   /* XXX: what happens when tex->bo->offset changes???
+    */
+   tex->ss.ss1.base_addr = 0; /* reloc */
+   tex->ss.ss2.mip_count = tex->base.last_level;
+   tex->ss.ss2.width = tex->base.width0 - 1;
+   tex->ss.ss2.height = tex->base.height0 - 1;
+
+   switch (tex->tiling) {
+   case BRW_TILING_NONE:
+      tex->ss.ss3.tiled_surface = 0;
+      tex->ss.ss3.tile_walk = 0;
+      break;
+   case BRW_TILING_X:
+      tex->ss.ss3.tiled_surface = 1;
+      tex->ss.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+      break;
+   case BRW_TILING_Y:
+      tex->ss.ss3.tiled_surface = 1;
+      tex->ss.ss3.tile_walk = BRW_TILEWALK_YMAJOR;
+      break;
+   }
+
+   tex->ss.ss3.pitch = (tex->pitch * tex->cpp) - 1;
+   tex->ss.ss3.depth = tex->base.depth0 - 1;
+
+   tex->ss.ss4.min_lod = 0;
+
+   return &tex->base;
+
+fail:
+   FREE(tex);
    return NULL;
 }
 
+static boolean
+brw_texture_get_handle(struct pipe_screen *screen,
+                       struct pipe_texture *texture,
+                       struct winsys_handle *whandle)
+{
+   struct brw_screen *bscreen = brw_screen(screen);
+   struct brw_texture *tex = brw_texture(texture);
+   unsigned stride;
+
+   stride = tex->pitch * tex->cpp;
+
+   return bscreen->sws->bo_get_handle(tex->bo, whandle, stride);
+}
+
+
+
 static void brw_texture_destroy(struct pipe_texture *pt)
 {
    struct brw_texture *tex = brw_texture(pt);
@@ -451,127 +558,13 @@ brw_tex_transfer_destroy(struct pipe_transfer *trans)
 }
 
 
-/*
- * Functions exported to the winsys
- */
-
-boolean brw_texture_get_winsys_buffer(struct pipe_texture *texture,
-                                      struct brw_winsys_buffer **buffer,
-                                      unsigned *stride)
-{
-   struct brw_texture *tex = brw_texture(texture);
-
-   *buffer = tex->bo;
-   if (stride)
-      *stride = tex->pitch * tex->cpp;
-
-   return TRUE;
-}
-
-struct pipe_texture * 
-brw_texture_blanket_winsys_buffer(struct pipe_screen *screen,
-                                  const struct pipe_texture *templ,
-                                  unsigned pitch,
-				  unsigned tiling,
-                                  struct brw_winsys_buffer *buffer)
-{
-   struct brw_screen *bscreen = brw_screen(screen);
-   struct brw_texture *tex;
-   GLuint format;
-
-   if (templ->target != PIPE_TEXTURE_2D ||
-       templ->last_level != 0 ||
-       templ->depth0 != 1)
-      return NULL;
-
-   if (util_format_is_compressed(templ->format))
-      return NULL;
-
-   tex = CALLOC_STRUCT(brw_texture);
-   if (!tex)
-      return NULL;
-
-   memcpy(&tex->base, templ, sizeof *templ);
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
-
-   /* XXX: cpp vs. blocksize
-    */
-   tex->cpp = util_format_get_blocksize(tex->base.format);
-   tex->tiling = tiling;
-
-   make_empty_list(&tex->views[0]);
-   make_empty_list(&tex->views[1]);
-
-   if (!brw_texture_layout(bscreen, tex))
-      goto fail;
-
-   /* XXX Maybe some more checks? */
-   if ((pitch / tex->cpp) < tex->pitch)
-      goto fail;
-
-   tex->pitch = pitch / tex->cpp;
-
-   tex->bo = buffer;
-
-   /* fix this warning */
-#if 0
-   if (tex->size > buffer->size)
-      goto fail;
-#endif
-
-   tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
-   tex->ss.ss0.surface_type = translate_tex_target(tex->base.target);
-
-   format = translate_tex_format(tex->base.format);
-   assert(format != BRW_SURFACEFORMAT_INVALID);
-   tex->ss.ss0.surface_format = format;
-
-   /* This is ok for all textures with channel width 8bit or less:
-    */
-/*    tex->ss.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
-
-
-   /* XXX: what happens when tex->bo->offset changes???
-    */
-   tex->ss.ss1.base_addr = 0; /* reloc */
-   tex->ss.ss2.mip_count = tex->base.last_level;
-   tex->ss.ss2.width = tex->base.width0 - 1;
-   tex->ss.ss2.height = tex->base.height0 - 1;
-
-   switch (tex->tiling) {
-   case BRW_TILING_NONE:
-      tex->ss.ss3.tiled_surface = 0;
-      tex->ss.ss3.tile_walk = 0;
-      break;
-   case BRW_TILING_X:
-      tex->ss.ss3.tiled_surface = 1;
-      tex->ss.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
-      break;
-   case BRW_TILING_Y:
-      tex->ss.ss3.tiled_surface = 1;
-      tex->ss.ss3.tile_walk = BRW_TILEWALK_YMAJOR;
-      break;
-   }
-
-   tex->ss.ss3.pitch = (tex->pitch * tex->cpp) - 1;
-   tex->ss.ss3.depth = tex->base.depth0 - 1;
-
-   tex->ss.ss4.min_lod = 0;
-
-   return &tex->base;
-
-fail:
-   FREE(tex);
-   return NULL;
-}
-
 void brw_screen_tex_init( struct brw_screen *brw_screen )
 {
    brw_screen->base.is_format_supported = brw_is_format_supported;
    brw_screen->base.texture_create = brw_texture_create;
+   brw_screen->base.texture_from_handle = brw_texture_from_handle;
+   brw_screen->base.texture_get_handle = brw_texture_get_handle;
    brw_screen->base.texture_destroy = brw_texture_destroy;
-   brw_screen->base.texture_blanket = brw_texture_blanket;
    brw_screen->base.get_tex_transfer = brw_get_tex_transfer;
    brw_screen->base.transfer_map = brw_transfer_map;
    brw_screen->base.transfer_unmap = brw_transfer_unmap;
diff --git a/src/gallium/drivers/i965/brw_structs.h b/src/gallium/drivers/i965/brw_structs.h
index bf10bc04de7..e97ddeb5e1c 100644
--- a/src/gallium/drivers/i965/brw_structs.h
+++ b/src/gallium/drivers/i965/brw_structs.h
@@ -28,7 +28,7 @@
   * Authors:
   *   Keith Whitwell <keith@tungstengraphics.com>
   */
-        
+
 
 #ifndef BRW_STRUCTS_H
 #define BRW_STRUCTS_H
@@ -1149,7 +1149,7 @@ struct brw_vertex_element_state
       GLuint valid:1; 
       GLuint vertex_buffer_index:5; 
    } ve0;
-   
+
    struct
    {
       GLuint dst_offset:8; 
diff --git a/src/gallium/drivers/i965/brw_winsys.h b/src/gallium/drivers/i965/brw_winsys.h
index c82d00f4a47..f30c7f18132 100644
--- a/src/gallium/drivers/i965/brw_winsys.h
+++ b/src/gallium/drivers/i965/brw_winsys.h
@@ -162,6 +162,16 @@ struct brw_winsys_screen {
                                unsigned alignment,
                                struct brw_winsys_buffer **bo_out);
 
+   enum pipe_error (*bo_from_handle)(struct brw_winsys_screen *sws,
+                                     struct winsys_handle *whandle,
+                                     unsigned *stride,
+                                     unsigned *tiling,
+                                     struct brw_winsys_buffer **bo_out);
+
+   enum pipe_error (*bo_get_handle)(struct brw_winsys_buffer *buffer,
+                                    struct winsys_handle *whandle,
+                                    unsigned stride);
+
    /* Destroy a buffer when our refcount goes to zero:
     */
    void (*bo_destroy)(struct brw_winsys_buffer *buffer);
@@ -257,28 +267,6 @@ bo_reference(struct brw_winsys_buffer **ptr, struct brw_winsys_buffer *buf)
 struct pipe_screen *brw_create_screen(struct brw_winsys_screen *iws, unsigned pci_id);
 
 
-/**
- * Get the brw_winsys buffer backing the texture.
- *
- * TODO UGLY
- */
-struct pipe_texture;
-boolean brw_texture_get_winsys_buffer(struct pipe_texture *texture,
-                                      struct brw_winsys_buffer **buffer,
-                                      unsigned *stride);
-
-/**
- * Wrap a brw_winsys buffer with a texture blanket.
- *
- * TODO UGLY
- */
-struct pipe_texture * 
-brw_texture_blanket_winsys_buffer(struct pipe_screen *screen,
-                                  const struct pipe_texture *template,
-                                  unsigned pitch,
-				  unsigned tiling,
-                                  struct brw_winsys_buffer *buffer);
-
 
 /*************************************************************************
  * Cooperative dumping between winsys and driver.  TODO: make this
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index 8248b2a4132..baf0ae44016 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -377,6 +377,42 @@ identity_delete_vs_state(struct pipe_context *_pipe,
                          vs);
 }
 
+
+static void *
+identity_create_vertex_elements_state(struct pipe_context *_pipe,
+                                      unsigned num_elements,
+                                      const struct pipe_vertex_element *vertex_elements)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   return pipe->create_vertex_elements_state(pipe,
+                                             num_elements,
+                                             vertex_elements);
+}
+
+static void
+identity_bind_vertex_elements_state(struct pipe_context *_pipe,
+                                    void *velems)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   pipe->bind_vertex_elements_state(pipe,
+                                    velems);
+}
+
+static void
+identity_delete_vertex_elements_state(struct pipe_context *_pipe,
+                                      void *velems)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   pipe->delete_vertex_elements_state(pipe,
+                                      velems);
+}
+
 static void
 identity_set_blend_color(struct pipe_context *_pipe,
                          const struct pipe_blend_color *blend_color)
@@ -563,20 +599,6 @@ identity_set_vertex_buffers(struct pipe_context *_pipe,
                             num_buffers,
                             buffers);
 }
-
-static void
-identity_set_vertex_elements(struct pipe_context *_pipe,
-                             unsigned num_elements,
-                             const struct pipe_vertex_element *vertex_elements)
-{
-   struct identity_context *id_pipe = identity_context(_pipe);
-   struct pipe_context *pipe = id_pipe->pipe;
-
-   pipe->set_vertex_elements(pipe,
-                             num_elements,
-                             vertex_elements);
-}
-
 static void
 identity_surface_copy(struct pipe_context *_pipe,
                       struct pipe_surface *_dst,
@@ -733,6 +755,9 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.create_vs_state = identity_create_vs_state;
    id_pipe->base.bind_vs_state = identity_bind_vs_state;
    id_pipe->base.delete_vs_state = identity_delete_vs_state;
+   id_pipe->base.create_vertex_elements_state = identity_create_vertex_elements_state;
+   id_pipe->base.bind_vertex_elements_state = identity_bind_vertex_elements_state;
+   id_pipe->base.delete_vertex_elements_state = identity_delete_vertex_elements_state;
    id_pipe->base.set_blend_color = identity_set_blend_color;
    id_pipe->base.set_stencil_ref = identity_set_stencil_ref;
    id_pipe->base.set_clip_state = identity_set_clip_state;
@@ -744,7 +769,6 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.set_fragment_sampler_textures = identity_set_fragment_sampler_textures;
    id_pipe->base.set_vertex_sampler_textures = identity_set_vertex_sampler_textures;
    id_pipe->base.set_vertex_buffers = identity_set_vertex_buffers;
-   id_pipe->base.set_vertex_elements = identity_set_vertex_elements;
    id_pipe->base.surface_copy = identity_surface_copy;
    id_pipe->base.surface_fill = identity_surface_fill;
    id_pipe->base.clear = identity_clear;
diff --git a/src/gallium/drivers/identity/id_drm.c b/src/gallium/drivers/identity/id_drm.c
index f258c38cd71..936ccc444a8 100644
--- a/src/gallium/drivers/identity/id_drm.c
+++ b/src/gallium/drivers/identity/id_drm.c
@@ -63,62 +63,6 @@ identity_drm_create_screen(struct drm_api *_api, int fd,
    return identity_screen_create(screen);
 }
 
-
-static struct pipe_texture *
-identity_drm_texture_from_shared_handle(struct drm_api *_api,
-                                        struct pipe_screen *_screen,
-                                        struct pipe_texture *templ,
-                                        const char *name,
-                                        unsigned stride,
-                                        unsigned handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct drm_api *api = id_api->api;
-   struct pipe_texture *result;
-
-   result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle);
-
-   result = identity_texture_create(identity_screen(_screen), result);
-
-   return result;
-}
-
-static boolean
-identity_drm_shared_handle_from_texture(struct drm_api *_api,
-                                        struct pipe_screen *_screen,
-                                        struct pipe_texture *_texture,
-                                        unsigned *stride,
-                                        unsigned *handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_texture *id_texture = identity_texture(_texture);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_texture *texture = id_texture->texture;
-   struct drm_api *api = id_api->api;
-
-   return api->shared_handle_from_texture(api, screen, texture, stride, handle);
-}
-
-static boolean
-identity_drm_local_handle_from_texture(struct drm_api *_api,
-                                       struct pipe_screen *_screen,
-                                       struct pipe_texture *_texture,
-                                       unsigned *stride,
-                                       unsigned *handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_texture *id_texture = identity_texture(_texture);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_texture *texture = id_texture->texture;
-   struct drm_api *api = id_api->api;
-
-   return api->local_handle_from_texture(api, screen, texture, stride, handle);
-}
-
 static void
 identity_drm_destroy(struct drm_api *_api)
 {
@@ -145,9 +89,6 @@ identity_drm_create(struct drm_api *api)
    id_api->base.name = api->name;
    id_api->base.driver_name = api->driver_name;
    id_api->base.create_screen = identity_drm_create_screen;
-   id_api->base.texture_from_shared_handle = identity_drm_texture_from_shared_handle;
-   id_api->base.shared_handle_from_texture = identity_drm_shared_handle_from_texture;
-   id_api->base.local_handle_from_texture = identity_drm_local_handle_from_texture;
    id_api->base.destroy = identity_drm_destroy;
    id_api->api = api;
 
diff --git a/src/gallium/drivers/identity/id_screen.c b/src/gallium/drivers/identity/id_screen.c
index b85492114a3..b9d0f003d74 100644
--- a/src/gallium/drivers/identity/id_screen.c
+++ b/src/gallium/drivers/identity/id_screen.c
@@ -135,27 +135,40 @@ identity_screen_texture_create(struct pipe_screen *_screen,
 }
 
 static struct pipe_texture *
-identity_screen_texture_blanket(struct pipe_screen *_screen,
-                                const struct pipe_texture *templat,
-                                const unsigned *stride,
-                                struct pipe_buffer *_buffer)
+identity_screen_texture_from_handle(struct pipe_screen *_screen,
+                                    const struct pipe_texture *templ,
+                                    struct winsys_handle *handle)
 {
    struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_buffer *id_buffer = identity_buffer(_buffer);
    struct pipe_screen *screen = id_screen->screen;
-   struct pipe_buffer *buffer = id_buffer->buffer;
    struct pipe_texture *result;
 
-   result = screen->texture_blanket(screen,
-                                    templat,
-                                    stride,
-                                    buffer);
+   /* TODO trace call */
 
-   if (result)
-      return identity_texture_create(id_screen, result);
-   return NULL;
+   result = screen->texture_from_handle(screen, templ, handle);
+
+   result = identity_texture_create(identity_screen(_screen), result);
+
+   return result;
+}
+
+static boolean
+identity_screen_texture_get_handle(struct pipe_screen *_screen,
+                                   struct pipe_texture *_texture,
+                                   struct winsys_handle *handle)
+{
+   struct identity_screen *id_screen = identity_screen(_screen);
+   struct identity_texture *id_texture = identity_texture(_texture);
+   struct pipe_screen *screen = id_screen->screen;
+   struct pipe_texture *texture = id_texture->texture;
+
+   /* TODO trace call */
+
+   return screen->texture_get_handle(screen, texture, handle);
 }
 
+
+
 static void
 identity_screen_texture_destroy(struct pipe_texture *_texture)
 {
@@ -298,31 +311,6 @@ identity_screen_user_buffer_create(struct pipe_screen *_screen,
    return NULL;
 }
 
-static struct pipe_buffer *
-identity_screen_surface_buffer_create(struct pipe_screen *_screen,
-                                      unsigned width,
-                                      unsigned height,
-                                      enum pipe_format format,
-                                      unsigned usage,
-                                      unsigned tex_usage,
-                                      unsigned *stride)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_buffer *result;
-
-   result = screen->surface_buffer_create(screen,
-                                          width,
-                                          height,
-                                          format,
-                                          usage,
-                                          tex_usage,
-                                          stride);
-
-   if (result)
-      return identity_buffer_create(id_screen, result);
-   return NULL;
-}
 
 static void *
 identity_screen_buffer_map(struct pipe_screen *_screen,
@@ -495,7 +483,8 @@ identity_screen_create(struct pipe_screen *screen)
    id_screen->base.is_format_supported = identity_screen_is_format_supported;
    id_screen->base.context_create = identity_screen_context_create;
    id_screen->base.texture_create = identity_screen_texture_create;
-   id_screen->base.texture_blanket = identity_screen_texture_blanket;
+   id_screen->base.texture_from_handle = identity_screen_texture_from_handle;
+   id_screen->base.texture_get_handle = identity_screen_texture_get_handle;
    id_screen->base.texture_destroy = identity_screen_texture_destroy;
    id_screen->base.get_tex_surface = identity_screen_get_tex_surface;
    id_screen->base.tex_surface_destroy = identity_screen_tex_surface_destroy;
@@ -505,7 +494,6 @@ identity_screen_create(struct pipe_screen *screen)
    id_screen->base.transfer_unmap = identity_screen_transfer_unmap;
    id_screen->base.buffer_create = identity_screen_buffer_create;
    id_screen->base.user_buffer_create = identity_screen_user_buffer_create;
-   id_screen->base.surface_buffer_create = identity_screen_surface_buffer_create;
    if (screen->buffer_map)
       id_screen->base.buffer_map = identity_screen_buffer_map;
    if (screen->buffer_map_range)
diff --git a/src/gallium/drivers/llvmpipe/lp_buffer.c b/src/gallium/drivers/llvmpipe/lp_buffer.c
index 9eda9720818..dab20cb6397 100644
--- a/src/gallium/drivers/llvmpipe/lp_buffer.c
+++ b/src/gallium/drivers/llvmpipe/lp_buffer.c
@@ -30,10 +30,10 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 
-#include "lp_winsys.h"
 #include "lp_screen.h"
 #include "lp_buffer.h"
 
+#include "state_tracker/sw_winsys.h"
 
 static void *
 llvmpipe_buffer_map(struct pipe_screen *screen,
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index e31ae6a3fc1..d94efec16a4 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -145,6 +145,10 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.bind_vs_state   = llvmpipe_bind_vs_state;
    llvmpipe->pipe.delete_vs_state = llvmpipe_delete_vs_state;
 
+   llvmpipe->pipe.create_vertex_elements_state = llvmpipe_create_vertex_elements_state;
+   llvmpipe->pipe.bind_vertex_elements_state = llvmpipe_bind_vertex_elements_state;
+   llvmpipe->pipe.delete_vertex_elements_state = llvmpipe_delete_vertex_elements_state;
+
    llvmpipe->pipe.set_blend_color = llvmpipe_set_blend_color;
    llvmpipe->pipe.set_stencil_ref = llvmpipe_set_stencil_ref;
    llvmpipe->pipe.set_clip_state = llvmpipe_set_clip_state;
@@ -157,7 +161,6 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.set_viewport_state = llvmpipe_set_viewport_state;
 
    llvmpipe->pipe.set_vertex_buffers = llvmpipe_set_vertex_buffers;
-   llvmpipe->pipe.set_vertex_elements = llvmpipe_set_vertex_elements;
 
    llvmpipe->pipe.draw_arrays = llvmpipe_draw_arrays;
    llvmpipe->pipe.draw_elements = llvmpipe_draw_elements;
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 955c7eb8e0e..217ec59b688 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -46,6 +46,7 @@ struct lp_fragment_shader;
 struct lp_vertex_shader;
 struct lp_blend_state;
 struct setup_context;
+struct lp_velems_state;
 
 struct llvmpipe_context {
    struct pipe_context pipe;  /**< base class */
@@ -58,6 +59,7 @@ struct llvmpipe_context {
    const struct pipe_rasterizer_state *rasterizer;
    struct lp_fragment_shader *fs;
    const struct lp_vertex_shader *vs;
+   const struct lp_velems_state *velems;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -71,13 +73,11 @@ struct llvmpipe_context {
    struct pipe_texture *vertex_textures[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
 
    unsigned num_samplers;
    unsigned num_textures;
    unsigned num_vertex_samplers;
    unsigned num_vertex_textures;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    unsigned dirty; /**< Mask of LP_NEW_x flags */
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index bacff500d63..08c8f93794d 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -58,7 +58,9 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
       elem_types[LP_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
       elem_types[LP_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
       elem_types[LP_JIT_TEXTURE_STRIDE] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_DATA]   = LLVMPointerType(LLVMInt8Type(), 0);
+      elem_types[LP_JIT_TEXTURE_DATA] =
+         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
+                       LP_MAX_TEXTURE_2D_LEVELS);
 
       texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 0ebb2826fa2..5cc7a12c03b 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -39,6 +39,7 @@
 #include "gallivm/lp_bld_struct.h"
 
 #include "pipe/p_state.h"
+#include "lp_texture.h"
 
 
 struct llvmpipe_screen;
@@ -51,7 +52,7 @@ struct lp_jit_texture
    uint32_t depth;
    uint32_t last_level;
    uint32_t stride;
-   const void *data;
+   const void *data[LP_MAX_TEXTURE_2D_LEVELS];
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_public.h b/src/gallium/drivers/llvmpipe/lp_public.h
new file mode 100644
index 00000000000..ec6b660b48e
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_public.h
@@ -0,0 +1,10 @@
+#ifndef LP_PUBLIC_H
+#define LP_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+llvmpipe_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index f84ede675b3..5093f58bb19 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -34,11 +34,13 @@
 #include "lp_texture.h"
 #include "lp_buffer.h"
 #include "lp_fence.h"
-#include "lp_winsys.h"
 #include "lp_jit.h"
 #include "lp_screen.h"
 #include "lp_context.h"
 #include "lp_debug.h"
+#include "lp_public.h"
+
+#include "state_tracker/sw_winsys.h"
 
 #ifdef DEBUG
 int LP_DEBUG = 0;
@@ -107,11 +109,11 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 9;  /* max 256x256x256 */
+      return LP_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
@@ -167,7 +169,7 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
                               unsigned geom_flags )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    const struct util_format_description *format_desc;
 
    format_desc = util_format_description(format);
@@ -238,18 +240,6 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
 }
 
 
-static struct pipe_buffer *
-llvmpipe_surface_buffer_create(struct pipe_screen *screen,
-                               unsigned width, unsigned height,
-                               enum pipe_format format,
-                               unsigned tex_usage,
-                               unsigned usage,
-                               unsigned *stride)
-{
-   /* This function should never be used */
-   assert(0);
-   return NULL;
-}
 
 
 static void
@@ -258,7 +248,7 @@ llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
                            void *context_private)
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    struct llvmpipe_texture *texture = llvmpipe_texture(surface->texture);
 
    assert(texture->dt);
@@ -271,7 +261,7 @@ static void
 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    lp_jit_screen_cleanup(screen);
 
@@ -288,7 +278,7 @@ llvmpipe_destroy_screen( struct pipe_screen *_screen )
  * Note: we're not presently subclassing pipe_screen (no llvmpipe_screen).
  */
 struct pipe_screen *
-llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
+llvmpipe_create_screen(struct sw_winsys *winsys)
 {
    struct llvmpipe_screen *screen = CALLOC_STRUCT(llvmpipe_screen);
 
@@ -309,7 +299,6 @@ llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
    screen->base.get_paramf = llvmpipe_get_paramf;
    screen->base.is_format_supported = llvmpipe_is_format_supported;
 
-   screen->base.surface_buffer_create = llvmpipe_surface_buffer_create;
    screen->base.context_create = llvmpipe_create_context;
    screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 4a1b4d6f3e2..f4e62cbf08e 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -43,14 +43,14 @@
 #include "pipe/p_defines.h"
 
 
-struct llvmpipe_winsys;
+struct sw_winsys;
 
 
 struct llvmpipe_screen
 {
    struct pipe_screen base;
 
-   struct llvmpipe_winsys *winsys;
+   struct sw_winsys *winsys;
 
    LLVMModuleRef module;
    LLVMExecutionEngineRef engine;
@@ -76,4 +76,5 @@ llvmpipe_screen( struct pipe_screen *pipe )
 }
 
 
+
 #endif /* LP_SCREEN_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index b0713c3b71d..fbe14924cb1 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -46,7 +46,7 @@
 #include "lp_rast.h"
 #include "lp_setup_context.h"
 #include "lp_screen.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
@@ -472,19 +472,25 @@ lp_setup_set_sampler_textures( struct setup_context *setup,
          jit_tex->depth = tex->depth0;
          jit_tex->last_level = tex->last_level;
          jit_tex->stride = lp_tex->stride[0];
-         if(!lp_tex->dt) {
-            jit_tex->data = lp_tex->data;
+         if (!lp_tex->dt) {
+            /* regular texture - setup array of mipmap level pointers */
+            int j;
+            for (j = 0; j < LP_MAX_TEXTURE_2D_LEVELS; j++) {
+               jit_tex->data[j] =
+                  (ubyte *) lp_tex->data + lp_tex->level_offset[j];
+            }
          }
          else {
+            /* display target texture/surface */
             /*
              * XXX: Where should this be unmapped?
              */
 
             struct llvmpipe_screen *screen = llvmpipe_screen(tex->screen);
-            struct llvmpipe_winsys *winsys = screen->winsys;
-            jit_tex->data = winsys->displaytarget_map(winsys, lp_tex->dt,
+            struct sw_winsys *winsys = screen->winsys;
+            jit_tex->data[0] = winsys->displaytarget_map(winsys, lp_tex->dt,
                                                       PIPE_BUFFER_USAGE_CPU_READ);
-            assert(jit_tex->data);
+            assert(jit_tex->data[0]);
          }
 
          /* the scene references this texture */
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 24291da91e4..671e74465c0 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -231,57 +231,29 @@ lp_setup_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
 
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride));
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
@@ -415,57 +387,28 @@ lp_setup_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 9beba32271f..6dbdc195bfc 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -119,6 +119,10 @@ struct lp_vertex_shader {
    struct draw_vertex_shader *draw_data;
 };
 
+struct lp_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
 
 
 void *
@@ -176,8 +180,14 @@ void *llvmpipe_create_vs_state(struct pipe_context *,
 void llvmpipe_bind_vs_state(struct pipe_context *, void *);
 void llvmpipe_delete_vs_state(struct pipe_context *, void *);
 
+void *llvmpipe_create_vertex_elements_state(struct pipe_context *,
+                                            unsigned count,
+                                            const struct pipe_vertex_element *);
+void llvmpipe_bind_vertex_elements_state(struct pipe_context *, void *);
+void llvmpipe_delete_vertex_elements_state(struct pipe_context *, void *);
+
 void llvmpipe_set_polygon_stipple( struct pipe_context *,
-				  const struct pipe_poly_stipple * );
+                                   const struct pipe_poly_stipple * );
 
 void llvmpipe_set_scissor_state( struct pipe_context *,
                                  const struct pipe_scissor_state * );
@@ -194,10 +204,6 @@ llvmpipe_set_vertex_sampler_textures(struct pipe_context *,
 void llvmpipe_set_viewport_state( struct pipe_context *,
                                   const struct pipe_viewport_state * );
 
-void llvmpipe_set_vertex_elements(struct pipe_context *,
-                                  unsigned count,
-                                  const struct pipe_vertex_element *);
-
 void llvmpipe_set_vertex_buffers(struct pipe_context *,
                                  unsigned count,
                                  const struct pipe_vertex_buffer *);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
index 57ac25ea0cb..f6427aa908e 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_vertex.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
@@ -35,24 +35,41 @@
 #include "draw/draw_context.h"
 
 
+void *
+llvmpipe_create_vertex_elements_state(struct pipe_context *pipe,
+                                      unsigned count,
+                                      const struct pipe_vertex_element *attribs)
+{
+   struct lp_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct lp_velems_state *) MALLOC(sizeof(struct lp_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
 void
-llvmpipe_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *attribs)
+llvmpipe_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   struct lp_velems_state *lp_velems = (struct lp_velems_state *) velems;
 
-   assert(count <= PIPE_MAX_ATTRIBS);
-
-   memcpy(llvmpipe->vertex_element, attribs,
-          count * sizeof(struct pipe_vertex_element));
-   llvmpipe->num_vertex_elements = count;
+   llvmpipe->velems = lp_velems;
 
    llvmpipe->dirty |= LP_NEW_VERTEX;
 
-   draw_set_vertex_elements(llvmpipe->draw, count, attribs);
+   if (velems)
+      draw_set_vertex_elements(llvmpipe->draw, lp_velems->count, lp_velems->velem);
 }
 
+void
+llvmpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 llvmpipe_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
index 632462460a3..5a3cf37d6d8 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
@@ -51,10 +51,11 @@
 
 
 /**
- * This provides the bridge between the sampler state store in lp_jit_context
- * and lp_jit_texture and the sampler code generator. It provides the
- * texture layout information required by the texture sampler code generator
- * in terms of the state stored in lp_jit_context and lp_jit_texture in runtime.
+ * This provides the bridge between the sampler state store in
+ * lp_jit_context and lp_jit_texture and the sampler code
+ * generator. It provides the texture layout information required by
+ * the texture sampler code generator in terms of the state stored in
+ * lp_jit_context and lp_jit_texture in runtime.
  */
 struct llvmpipe_sampler_dynamic_state
 {
@@ -79,6 +80,9 @@ struct lp_llvm_sampler_soa
 
 /**
  * Fetch the specified member of the lp_jit_texture structure.
+ * \param emit_load  if TRUE, emit the LLVM load instruction to actually
+ *                   fetch the field's value.  Otherwise, just emit the
+ *                   GEP code to address the field.
  *
  * @sa http://llvm.org/docs/GetElementPtr.html
  */
@@ -87,9 +91,11 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
                        LLVMBuilderRef builder,
                        unsigned unit,
                        unsigned member_index,
-                       const char *member_name)
+                       const char *member_name,
+                       boolean emit_load)
 {
-   struct llvmpipe_sampler_dynamic_state *state = (struct llvmpipe_sampler_dynamic_state *)base;
+   struct llvmpipe_sampler_dynamic_state *state =
+      (struct llvmpipe_sampler_dynamic_state *)base;
    LLVMValueRef indices[4];
    LLVMValueRef ptr;
    LLVMValueRef res;
@@ -107,7 +113,10 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
-   res = LLVMBuildLoad(builder, ptr, "");
+   if (emit_load)
+      res = LLVMBuildLoad(builder, ptr, "");
+   else
+      res = ptr;
 
    lp_build_name(res, "context.texture%u.%s", unit, member_name);
 
@@ -116,28 +125,30 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
 
 /**
- * Helper macro to instantiate the functions that generate the code to fetch
- * the members of lp_jit_texture to fulfill the sampler code generator requests.
+ * Helper macro to instantiate the functions that generate the code to
+ * fetch the members of lp_jit_texture to fulfill the sampler code
+ * generator requests.
  *
- * This complexity is the price we have to pay to keep the texture sampler code
- * generator a reusable module without dependencies to llvmpipe internals.
+ * This complexity is the price we have to pay to keep the texture
+ * sampler code generator a reusable module without dependencies to
+ * llvmpipe internals.
  */
-#define LP_LLVM_TEXTURE_MEMBER(_name, _index) \
+#define LP_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    lp_llvm_texture_##_name( struct lp_sampler_dynamic_state *base, \
                             LLVMBuilderRef builder, \
                             unsigned unit) \
    { \
-      return lp_llvm_texture_member(base, builder, unit, _index, #_name ); \
+      return lp_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
    }
 
 
-LP_LLVM_TEXTURE_MEMBER(width,      LP_JIT_TEXTURE_WIDTH)
-LP_LLVM_TEXTURE_MEMBER(height,     LP_JIT_TEXTURE_HEIGHT)
-LP_LLVM_TEXTURE_MEMBER(depth,      LP_JIT_TEXTURE_DEPTH)
-LP_LLVM_TEXTURE_MEMBER(last_level, LP_JIT_TEXTURE_LAST_LEVEL)
-LP_LLVM_TEXTURE_MEMBER(stride,     LP_JIT_TEXTURE_STRIDE)
-LP_LLVM_TEXTURE_MEMBER(data_ptr,   LP_JIT_TEXTURE_DATA)
+LP_LLVM_TEXTURE_MEMBER(width,      LP_JIT_TEXTURE_WIDTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(height,     LP_JIT_TEXTURE_HEIGHT, TRUE)
+LP_LLVM_TEXTURE_MEMBER(depth,      LP_JIT_TEXTURE_DEPTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(last_level, LP_JIT_TEXTURE_LAST_LEVEL, TRUE)
+LP_LLVM_TEXTURE_MEMBER(stride,     LP_JIT_TEXTURE_STRIDE, TRUE)
+LP_LLVM_TEXTURE_MEMBER(data_ptr,   LP_JIT_TEXTURE_DATA, FALSE)
 
 
 static void
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index 7f456355428..74b7b4ec5e1 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -42,7 +42,7 @@
 #include "lp_screen.h"
 #include "lp_texture.h"
 #include "lp_tile_size.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 
 /**
@@ -93,7 +93,7 @@ static boolean
 llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
                               struct llvmpipe_texture *lpt)
 {
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    /* Round up the surface size to a multiple of the tile size to
     * avoid tile clipping.
@@ -125,7 +125,8 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
    lpt->base.screen = &screen->base;
 
    if (lpt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                              PIPE_TEXTURE_USAGE_PRIMARY)) {
+                              PIPE_TEXTURE_USAGE_SCANOUT |
+                              PIPE_TEXTURE_USAGE_SHARED)) {
       if (!llvmpipe_displaytarget_layout(screen, lpt))
          goto fail;
    }
@@ -142,43 +143,6 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
 }
 
 
-static struct pipe_texture *
-llvmpipe_texture_blanket(struct pipe_screen * screen,
-                         const struct pipe_texture *base,
-                         const unsigned *stride,
-                         struct pipe_buffer *buffer)
-{
-   /* FIXME */
-#if 0
-   struct llvmpipe_texture *lpt;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   lpt = CALLOC_STRUCT(llvmpipe_texture);
-   if (!lpt)
-      return NULL;
-
-   lpt->base = *base;
-   pipe_reference_init(&lpt->base.reference, 1);
-   lpt->base.screen = screen;
-   lpt->stride[0] = stride[0];
-
-   pipe_buffer_reference(&lpt->buffer, buffer);
-
-   return &lpt->base;
-#else
-   debug_printf("llvmpipe_texture_blanket() not implemented!");
-   return NULL;
-#endif
-}
-
-
 static void
 llvmpipe_texture_destroy(struct pipe_texture *pt)
 {
@@ -187,7 +151,7 @@ llvmpipe_texture_destroy(struct pipe_texture *pt)
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+      struct sw_winsys *winsys = screen->winsys;
       winsys->displaytarget_destroy(winsys, lpt->dt);
    }
    else {
@@ -357,7 +321,7 @@ llvmpipe_transfer_map( struct pipe_screen *_screen,
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+      struct sw_winsys *winsys = screen->winsys;
 
       map = winsys->displaytarget_map(winsys, lpt->dt,
                                       pipe_transfer_buffer_flags(transfer));
@@ -398,7 +362,7 @@ llvmpipe_transfer_unmap(struct pipe_screen *screen,
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = lp_screen->winsys;
+      struct sw_winsys *winsys = lp_screen->winsys;
       winsys->displaytarget_unmap(winsys, lpt->dt);
    }
 }
@@ -408,7 +372,6 @@ void
 llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = llvmpipe_texture_create;
-   screen->texture_blanket = llvmpipe_texture_blanket;
    screen->texture_destroy = llvmpipe_texture_destroy;
 
    screen->get_tex_surface = llvmpipe_get_tex_surface;
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.h b/src/gallium/drivers/llvmpipe/lp_texture.h
index 87c905bc027..b23f929b167 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.h
+++ b/src/gallium/drivers/llvmpipe/lp_texture.h
@@ -32,24 +32,29 @@
 #include "pipe/p_state.h"
 
 
+#define LP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K for now */
+#define LP_MAX_TEXTURE_3D_LEVELS 10  /* 512 x 512 x 512 for now */
+
+
 struct pipe_context;
 struct pipe_screen;
 struct llvmpipe_context;
-struct llvmpipe_displaytarget;
+
+struct sw_displaytarget;
 
 
 struct llvmpipe_texture
 {
    struct pipe_texture base;
 
-   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned stride[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long level_offset[LP_MAX_TEXTURE_2D_LEVELS];
+   unsigned stride[LP_MAX_TEXTURE_2D_LEVELS];
 
    /**
     * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
     * usage.
     */
-   struct llvmpipe_displaytarget *dt;
+   struct sw_displaytarget *dt;
 
    /**
     * Malloc'ed data for regular textures, or a mapping to dt above.
diff --git a/src/gallium/drivers/llvmpipe/lp_winsys.h b/src/gallium/drivers/llvmpipe/lp_winsys.h
deleted file mode 100644
index ce11fa93041..00000000000
--- a/src/gallium/drivers/llvmpipe/lp_winsys.h
+++ /dev/null
@@ -1,125 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007-2009 VMware, Inc.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * @file
- * llvmpipe public interface.
- */
-
-
-#ifndef LP_WINSYS_H
-#define LP_WINSYS_H
-
-
-#include "pipe/p_compiler.h" /* for boolean */
-#include "pipe/p_format.h"
-
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-
-struct pipe_screen;
-struct pipe_context;
-
-
-/**
- * Opaque pointer.
- */
-struct llvmpipe_displaytarget;
-
-
-/**
- * This is the interface that llvmpipe expects any window system
- * hosting it to implement.
- * 
- * llvmpipe is for the most part a self sufficient driver. The only thing it
- * does not know is how to display a surface.
- */
-struct llvmpipe_winsys
-{
-   void 
-   (*destroy)( struct llvmpipe_winsys *ws );
-
-   boolean
-   (*is_displaytarget_format_supported)( struct llvmpipe_winsys *ws,
-                                         enum pipe_format format );
-   
-   /**
-    * Allocate storage for a render target.
-    * 
-    * Often surfaces which are meant to be blitted to the front screen (i.e.,
-    * display targets) must be allocated with special characteristics, memory 
-    * pools, or obtained directly from the windowing system.
-    *  
-    * This callback is invoked by the pipe_screen when creating a texture marked
-    * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag to get the underlying 
-    * storage.
-    */
-   struct llvmpipe_displaytarget *
-   (*displaytarget_create)( struct llvmpipe_winsys *ws,
-                            enum pipe_format format,
-                            unsigned width, unsigned height,
-                            unsigned alignment,
-                            unsigned *stride );
-
-   void *
-   (*displaytarget_map)( struct llvmpipe_winsys *ws, 
-                         struct llvmpipe_displaytarget *dt,
-                         unsigned flags );
-
-   void
-   (*displaytarget_unmap)( struct llvmpipe_winsys *ws,
-                           struct llvmpipe_displaytarget *dt );
-
-   /**
-    * @sa pipe_screen:flush_frontbuffer.
-    *
-    * This call will likely become asynchronous eventually.
-    */
-   void
-   (*displaytarget_display)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt,
-                             void *context_private );
-
-   void 
-   (*displaytarget_destroy)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt );
-};
-
-
-
-struct pipe_screen *
-llvmpipe_create_screen( struct llvmpipe_winsys * );
-
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* LP_WINSYS_H */
diff --git a/src/gallium/drivers/nouveau/nouveau_screen.c b/src/gallium/drivers/nouveau/nouveau_screen.c
index 3c2f771b51e..b1ad686022a 100644
--- a/src/gallium/drivers/nouveau/nouveau_screen.c
+++ b/src/gallium/drivers/nouveau/nouveau_screen.c
@@ -4,6 +4,7 @@
 
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
+#include "util/u_format.h"
 
 #include <stdio.h>
 #include <errno.h>
@@ -12,6 +13,9 @@
 #include "nouveau_winsys.h"
 #include "nouveau_screen.h"
 
+/* XXX this should go away */
+#include "state_tracker/drm_api.h"
+
 static const char *
 nouveau_screen_get_name(struct pipe_screen *pscreen)
 {
@@ -120,7 +124,7 @@ nouveau_screen_map_flags(unsigned pipe)
 	if (pipe & PIPE_BUFFER_USAGE_DONTBLOCK)
 		flags |= NOUVEAU_BO_NOWAIT;
 	else
-	if (pipe & 0 /*PIPE_BUFFER_USAGE_UNSYNCHRONIZED*/)
+	if (pipe & PIPE_BUFFER_USAGE_UNSYNCHRONIZED)
 		flags |= NOUVEAU_BO_NOSYNC;
 
 	return flags;
@@ -231,6 +235,72 @@ nouveau_screen_fence_finish(struct pipe_screen *screen,
 	return 0;
 }
 
+
+/*
+ * Both texture_{from|get}_handle use drm api defines directly which they
+ * shouldn't do. The problem is that from|get are pipe functions and as
+ * such they should be defined in the pipe level. If nouveau had a propper
+ * winsys interface we would have added from|get to that interface using
+ * the winsys_handle struct as done with other drivers. However this code
+ * calls directly into the libdrm_nouveau.so functions (nouveau_bo_*). So
+ * we need to translate the handle into something they understand.
+ */
+static struct pipe_texture *
+nouveau_screen_texture_from_handle(struct pipe_screen *pscreen,
+				   const struct pipe_texture *templ,
+				   struct winsys_handle *whandle)
+{
+	struct nouveau_device *dev = nouveau_screen(pscreen)->device;
+	struct pipe_texture *pt;
+	struct pipe_buffer *pb;
+	int ret;
+
+	pb = CALLOC(1, sizeof(struct pipe_buffer) + sizeof(struct nouveau_bo*));
+	if (!pb)
+		return NULL;
+
+	ret = nouveau_bo_handle_ref(dev, whandle->handle, (struct nouveau_bo**)(pb+1));
+	if (ret) {
+		debug_printf("%s: ref name 0x%08x failed with %d\n",
+			     __func__, whandle->handle, ret);
+		FREE(pb);
+		return NULL;
+	}
+
+	pipe_reference_init(&pb->reference, 1);
+	pb->screen = pscreen;
+	pb->alignment = 0;
+	pb->usage = PIPE_BUFFER_USAGE_GPU_READ_WRITE |
+		    PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+	pb->size = nouveau_bo(pb)->size;
+	pt = nouveau_screen(pscreen)->texture_blanket(pscreen, templ,
+                                                      &whandle->stride, pb);
+	pipe_buffer_reference(&pb, NULL);
+	return pt;
+}
+
+static boolean
+nouveau_screen_texture_get_handle(struct pipe_screen *pscreen,
+				  struct pipe_texture *pt,
+				  struct winsys_handle *whandle)
+{
+	struct nouveau_miptree *mt = nouveau_miptree(pt);
+
+	if (!mt || !mt->bo)
+		return false;
+
+	whandle->stride = util_format_get_stride(mt->base.format, mt->base.width0);
+
+	if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) { 
+		return nouveau_bo_handle_get(mt->bo, &whandle->handle) == 0;
+	} else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+		whandle->handle = mt->bo->handle;
+		return TRUE;
+	} else {
+		return FALSE;
+	}
+}
+
 int
 nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
 {
@@ -258,6 +328,9 @@ nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
 	pscreen->fence_signalled = nouveau_screen_fence_signalled;
 	pscreen->fence_finish = nouveau_screen_fence_finish;
 
+	pscreen->texture_from_handle = nouveau_screen_texture_from_handle;
+	pscreen->texture_get_handle = nouveau_screen_texture_get_handle;
+
 	return 0;
 }
 
diff --git a/src/gallium/drivers/nouveau/nouveau_screen.h b/src/gallium/drivers/nouveau/nouveau_screen.h
index a7927d88dfc..f4a7a2bc234 100644
--- a/src/gallium/drivers/nouveau/nouveau_screen.h
+++ b/src/gallium/drivers/nouveau/nouveau_screen.h
@@ -6,6 +6,18 @@ struct nouveau_screen {
 	struct nouveau_device *device;
 	struct nouveau_channel *channel;
 
+        /**
+         * Create a new texture object, using the given template info, but on top of
+         * existing memory.
+         * 
+         * It is assumed that the buffer data is layed out according to the expected
+         * by the hardware. NULL will be returned if any inconsistency is found.  
+         */
+        struct pipe_texture * (*texture_blanket)(struct pipe_screen *,
+                                                 const struct pipe_texture *templat,
+                                                 const unsigned *stride,
+                                                 struct pipe_buffer *buffer);
+
 	int (*pre_pipebuffer_map_callback) (struct pipe_screen *pscreen,
 		struct pipe_buffer *pb, unsigned usage);
 };
diff --git a/src/gallium/drivers/nouveau/nouveau_util.h b/src/gallium/drivers/nouveau/nouveau_util.h
index a10114beab9..7f16e31c3f0 100644
--- a/src/gallium/drivers/nouveau/nouveau_util.h
+++ b/src/gallium/drivers/nouveau/nouveau_util.h
@@ -88,4 +88,104 @@ static INLINE unsigned log2i(unsigned i)
 	return r;
 }
 
+struct u_split_prim {
+   void *priv;
+   void (*emit)(void *priv, unsigned start, unsigned count);
+   void (*edge)(void *priv, boolean enabled);
+
+   unsigned mode;
+   unsigned start;
+   unsigned p_start;
+   unsigned p_end;
+
+   int repeat_first:1;
+   int close_first:1;
+   int edgeflag_off:1;
+};
+
+static inline void
+u_split_prim_init(struct u_split_prim *s,
+                  unsigned mode, unsigned start, unsigned count)
+{
+   if (mode == PIPE_PRIM_LINE_LOOP) {
+      s->mode = PIPE_PRIM_LINE_STRIP;
+      s->close_first = 1;
+   } else {
+      s->mode = mode;
+      s->close_first = 0;
+   }
+   s->start = start;
+   s->p_start = start;
+   s->p_end = start + count;
+   s->edgeflag_off = 0;
+   s->repeat_first = 0;
+}
+
+static INLINE boolean
+u_split_prim_next(struct u_split_prim *s, unsigned max_verts)
+{
+   int repeat = 0;
+
+   if (s->repeat_first) {
+      s->emit(s->priv, s->start, 1);
+      max_verts--;
+      if (s->edgeflag_off) {
+         s->edge(s->priv, TRUE);
+         s->edgeflag_off = FALSE;
+      }
+   }
+
+   if (s->p_start + s->close_first + max_verts >= s->p_end) {
+      s->emit(s->priv, s->p_start, s->p_end - s->p_start);
+      if (s->close_first)
+         s->emit(s->priv, s->start, 1);
+      return TRUE;
+   }
+
+   switch (s->mode) {
+   case PIPE_PRIM_LINES:
+      max_verts &= ~1;
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+      repeat = 1;
+      break;
+   case PIPE_PRIM_POLYGON:
+      max_verts--;
+      s->emit(s->priv, s->p_start, max_verts);
+      s->edge(s->priv, FALSE);
+      s->emit(s->priv, s->p_start + max_verts, 1);
+      s->p_start += max_verts;
+      s->repeat_first = TRUE;
+      s->edgeflag_off = TRUE;
+      return FALSE;
+   case PIPE_PRIM_TRIANGLES:
+      max_verts = max_verts - (max_verts % 3);
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      /* to ensure winding stays correct, always split
+       * on an even number of generated triangles
+       */
+      max_verts = max_verts & ~1;
+      repeat = 2;
+      break;
+   case PIPE_PRIM_TRIANGLE_FAN:
+      s->repeat_first = TRUE;
+      repeat = 1;
+      break;
+   case PIPE_PRIM_QUADS:
+      max_verts &= ~3;
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      max_verts &= ~1;
+      repeat = 2;
+      break;
+   default:
+      break;
+   }
+
+   s->emit (s->priv, s->p_start, max_verts);
+   s->p_start += (max_verts - repeat);
+   return FALSE;
+}
+
 #endif
diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h
index ea259aadf35..1786460aec1 100644
--- a/src/gallium/drivers/nv30/nv30_context.h
+++ b/src/gallium/drivers/nv30/nv30_context.h
@@ -107,6 +107,11 @@ struct nv30_state {
 	struct nouveau_stateobj *hw[NV30_STATE_MAX];
 };
 
+struct nv30_vtxelt_state {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+};
+
 struct nv30_context {
 	struct pipe_context pipe;
 
@@ -142,8 +147,7 @@ struct nv30_context {
 	unsigned dirty_samplers;
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
+	struct nv30_vtxelt_state *vtxelt;
 };
 
 static INLINE struct nv30_context *
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index 697b1b92ff6..bfa27b632f7 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -18,7 +18,7 @@ nv30_miptree_layout(struct nv30_miptree *nv30mt)
 		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
 		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
 		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-		                           PIPE_TEXTURE_USAGE_PRIMARY);
+		                           PIPE_TEXTURE_USAGE_SCANOUT);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -78,7 +78,7 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	    pt->height0 & (pt->height0 - 1))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
-	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
+	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
 	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
@@ -232,8 +232,9 @@ void
 nv30_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv30_miptree_create;
-	pscreen->texture_blanket = nv30_miptree_blanket;
 	pscreen->texture_destroy = nv30_miptree_destroy;
 	pscreen->get_tex_surface = nv30_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv30_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv30_miptree_blanket;
 }
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index d911c807074..24b15a63ac4 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -669,15 +669,34 @@ nv30_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	/*nv30->draw_dirty |= NV30_NEW_ARRAYS;*/
 }
 
+static void *
+nv30_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv30_vtxelt_state *cso = CALLOC_STRUCT(nv30_vtxelt_state);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+/*	nv30_vtxelt_construct(cso);*/
+
+	return (void *)cso;
+}
+
 static void
-nv30_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv30_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv30_context *nv30 = nv30_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv30->vtxelt, ve, sizeof(*ve) * count);
-	nv30->vtxelt_nr = count;
+static void
+nv30_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv30_context *nv30 = nv30_context(pipe);
 
+	nv30->vtxelt = hwcso;
 	nv30->dirty |= NV30_NEW_ARRAYS;
 	/*nv30->draw_dirty |= NV30_NEW_ARRAYS;*/
 }
@@ -722,7 +741,10 @@ nv30_init_state_functions(struct nv30_context *nv30)
 	nv30->pipe.set_scissor_state = nv30_set_scissor_state;
 	nv30->pipe.set_viewport_state = nv30_set_viewport_state;
 
+	nv30->pipe.create_vertex_elements_state = nv30_vtxelts_state_create;
+	nv30->pipe.delete_vertex_elements_state = nv30_vtxelts_state_delete;
+	nv30->pipe.bind_vertex_elements_state = nv30_vtxelts_state_bind;
+
 	nv30->pipe.set_vertex_buffers = nv30_set_vertex_buffers;
-	nv30->pipe.set_vertex_elements = nv30_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h
index e42e872de75..66c26360cb2 100644
--- a/src/gallium/drivers/nv30/nv30_state.h
+++ b/src/gallium/drivers/nv30/nv30_state.h
@@ -70,6 +70,8 @@ struct nv30_fragment_program {
 	struct nouveau_stateobj *so;
 };
 
+#define NV30_MAX_TEXTURE_LEVELS  16
+
 struct nv30_miptree {
 	struct pipe_texture base;
 	struct nouveau_bo *bo;
@@ -80,7 +82,7 @@ struct nv30_miptree {
 	struct {
 		uint pitch;
 		uint *image_offset;
-	} level[PIPE_MAX_TEXTURE_LEVELS];
+	} level[NV30_MAX_TEXTURE_LEVELS];
 };
 
 #endif
diff --git a/src/gallium/drivers/nv30/nv30_vbo.c b/src/gallium/drivers/nv30/nv30_vbo.c
index e48823a9138..f3856bb5a5e 100644
--- a/src/gallium/drivers/nv30/nv30_vbo.c
+++ b/src/gallium/drivers/nv30/nv30_vbo.c
@@ -492,16 +492,16 @@ nv30_vbo_validate(struct nv30_context *nv30)
 	int hw;
 
 	vtxbuf = so_new(3, 17, 18);
-	so_method(vtxbuf, rankine, NV34TCL_VTXBUF_ADDRESS(0), nv30->vtxelt_nr);
+	so_method(vtxbuf, rankine, NV34TCL_VTXBUF_ADDRESS(0), nv30->vtxelt->num_elements);
 	vtxfmt = so_new(1, 16, 0);
-	so_method(vtxfmt, rankine, NV34TCL_VTXFMT(0), nv30->vtxelt_nr);
+	so_method(vtxfmt, rankine, NV34TCL_VTXFMT(0), nv30->vtxelt->num_elements);
 
-	for (hw = 0; hw < nv30->vtxelt_nr; hw++) {
+	for (hw = 0; hw < nv30->vtxelt->num_elements; hw++) {
 		struct pipe_vertex_element *ve;
 		struct pipe_vertex_buffer *vb;
 		unsigned type, ncomp;
 
-		ve = &nv30->vtxelt[hw];
+		ve = &nv30->vtxelt->pipe[hw];
 		vb = &nv30->vtxbuf[ve->vertex_buffer_index];
 
 		if (!vb->stride) {
diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h
index 97fb6a2ef94..2550ec654b3 100644
--- a/src/gallium/drivers/nv40/nv40_context.h
+++ b/src/gallium/drivers/nv40/nv40_context.h
@@ -107,6 +107,12 @@ struct nv40_state {
 	struct nouveau_stateobj *hw[NV40_STATE_MAX];
 };
 
+
+struct nv40_vtxelt_state {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+};
+
 struct nv40_context {
 	struct pipe_context pipe;
 
@@ -157,8 +163,7 @@ struct nv40_context {
 	unsigned dirty_samplers;
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
+	struct nv40_vtxelt_state *vtxelt;
 };
 
 static INLINE struct nv40_context *
diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c
index 85d7e1f1972..62e97bcea48 100644
--- a/src/gallium/drivers/nv40/nv40_miptree.c
+++ b/src/gallium/drivers/nv40/nv40_miptree.c
@@ -20,7 +20,7 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
 		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
 		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-		                           PIPE_TEXTURE_USAGE_PRIMARY);
+		                           PIPE_TEXTURE_USAGE_SCANOUT);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -80,7 +80,7 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	    pt->height0 & (pt->height0 - 1))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
-	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
+	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
 	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
@@ -227,9 +227,10 @@ void
 nv40_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv40_miptree_create;
-	pscreen->texture_blanket = nv40_miptree_blanket;
 	pscreen->texture_destroy = nv40_miptree_destroy;
 	pscreen->get_tex_surface = nv40_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv40_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv40_miptree_blanket;
 }
 
diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c
index 4f28675e64c..ee471e6ce4c 100644
--- a/src/gallium/drivers/nv40/nv40_state.c
+++ b/src/gallium/drivers/nv40/nv40_state.c
@@ -684,15 +684,34 @@ nv40_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	nv40->draw_dirty |= NV40_NEW_ARRAYS;
 }
 
+static void *
+nv40_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv40_vtxelt_state *cso = CALLOC_STRUCT(nv40_vtxelt_state);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+/*	nv40_vtxelt_construct(cso);*/
+
+	return (void *)cso;
+}
+
 static void
-nv40_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv40_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv40_context *nv40 = nv40_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv40->vtxelt, ve, sizeof(*ve) * count);
-	nv40->vtxelt_nr = count;
+static void
+nv40_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv40_context *nv40 = nv40_context(pipe);
 
+	nv40->vtxelt = hwcso;
 	nv40->dirty |= NV40_NEW_ARRAYS;
 	nv40->draw_dirty |= NV40_NEW_ARRAYS;
 }
@@ -737,7 +756,10 @@ nv40_init_state_functions(struct nv40_context *nv40)
 	nv40->pipe.set_scissor_state = nv40_set_scissor_state;
 	nv40->pipe.set_viewport_state = nv40_set_viewport_state;
 
+	nv40->pipe.create_vertex_elements_state = nv40_vtxelts_state_create;
+	nv40->pipe.delete_vertex_elements_state = nv40_vtxelts_state_delete;
+	nv40->pipe.bind_vertex_elements_state = nv40_vtxelts_state_bind;
+
 	nv40->pipe.set_vertex_buffers = nv40_set_vertex_buffers;
-	nv40->pipe.set_vertex_elements = nv40_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h
index 192074e7471..e2e69420eae 100644
--- a/src/gallium/drivers/nv40/nv40_state.h
+++ b/src/gallium/drivers/nv40/nv40_state.h
@@ -73,6 +73,8 @@ struct nv40_fragment_program {
 	struct nouveau_stateobj *so;
 };
 
+#define NV40_MAX_TEXTURE_LEVELS  16
+
 struct nv40_miptree {
 	struct pipe_texture base;
 	struct nouveau_bo *bo;
@@ -83,7 +85,7 @@ struct nv40_miptree {
 	struct {
 		uint pitch;
 		uint *image_offset;
-	} level[PIPE_MAX_TEXTURE_LEVELS];
+	} level[NV40_MAX_TEXTURE_LEVELS];
 };
 
 #endif
diff --git a/src/gallium/drivers/nv40/nv40_state_emit.c b/src/gallium/drivers/nv40/nv40_state_emit.c
index 8990f303ce4..297d71f4fac 100644
--- a/src/gallium/drivers/nv40/nv40_state_emit.c
+++ b/src/gallium/drivers/nv40/nv40_state_emit.c
@@ -174,7 +174,7 @@ nv40_state_validate_swtnl(struct nv40_context *nv40)
 
 	if (nv40->draw_dirty & NV40_NEW_ARRAYS) {
 		draw_set_vertex_buffers(draw, nv40->vtxbuf_nr, nv40->vtxbuf);
-		draw_set_vertex_elements(draw, nv40->vtxelt_nr, nv40->vtxelt);	
+		draw_set_vertex_elements(draw, nv40->vtxelt->num_elements, nv40->vtxelt->pipe);	
 	}
 
 	nv40_state_do_validate(nv40, swtnl_states);
diff --git a/src/gallium/drivers/nv40/nv40_vbo.c b/src/gallium/drivers/nv40/nv40_vbo.c
index 7812460d2ed..fabdf4bf23b 100644
--- a/src/gallium/drivers/nv40/nv40_vbo.c
+++ b/src/gallium/drivers/nv40/nv40_vbo.c
@@ -493,16 +493,16 @@ nv40_vbo_validate(struct nv40_context *nv40)
 	int hw;
 
 	vtxbuf = so_new(3, 17, 18);
-	so_method(vtxbuf, curie, NV40TCL_VTXBUF_ADDRESS(0), nv40->vtxelt_nr);
+	so_method(vtxbuf, curie, NV40TCL_VTXBUF_ADDRESS(0), nv40->vtxelt->num_elements);
 	vtxfmt = so_new(1, 16, 0);
-	so_method(vtxfmt, curie, NV40TCL_VTXFMT(0), nv40->vtxelt_nr);
+	so_method(vtxfmt, curie, NV40TCL_VTXFMT(0), nv40->vtxelt->num_elements);
 
-	for (hw = 0; hw < nv40->vtxelt_nr; hw++) {
+	for (hw = 0; hw < nv40->vtxelt->num_elements; hw++) {
 		struct pipe_vertex_element *ve;
 		struct pipe_vertex_buffer *vb;
 		unsigned type, ncomp;
 
-		ve = &nv40->vtxelt[hw];
+		ve = &nv40->vtxelt->pipe[hw];
 		vb = &nv40->vtxbuf[ve->vertex_buffer_index];
 
 		if (!vb->stride) {
diff --git a/src/gallium/drivers/nv50/Makefile b/src/gallium/drivers/nv50/Makefile
index 612aea28a34..5d622e1c13c 100644
--- a/src/gallium/drivers/nv50/Makefile
+++ b/src/gallium/drivers/nv50/Makefile
@@ -16,6 +16,7 @@ C_SOURCES = \
 	nv50_surface.c \
 	nv50_tex.c \
 	nv50_transfer.c \
-	nv50_vbo.c
+	nv50_vbo.c \
+	nv50_push.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/nv50/nv50_clear.c b/src/gallium/drivers/nv50/nv50_clear.c
index e0b2d2880b0..8afc95c9fc6 100644
--- a/src/gallium/drivers/nv50/nv50_clear.c
+++ b/src/gallium/drivers/nv50/nv50_clear.c
@@ -36,7 +36,7 @@ nv50_clear(struct pipe_context *pipe, unsigned buffers,
 	struct pipe_framebuffer_state *fb = &nv50->framebuffer;
 	unsigned mode = 0, i;
 
-	if (!nv50_state_validate(nv50))
+	if (!nv50_state_validate(nv50, 64))
 		return;
 
 	if (buffers & PIPE_CLEAR_COLOR && fb->nr_cbufs) {
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 7be12fcdef4..0eb42f323ff 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -46,43 +46,13 @@ static void
 nv50_destroy(struct pipe_context *pipe)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	int i;
 
-        if (nv50->state.fb)
-		so_ref(NULL, &nv50->state.fb);
-	if (nv50->state.blend)
-		so_ref(NULL, &nv50->state.blend);
-	if (nv50->state.blend_colour)
-		so_ref(NULL, &nv50->state.blend_colour);
-	if (nv50->state.zsa)
-		so_ref(NULL, &nv50->state.zsa);
-	if (nv50->state.rast)
-		so_ref(NULL, &nv50->state.rast);
-	if (nv50->state.stipple)
-		so_ref(NULL, &nv50->state.stipple);
-	if (nv50->state.scissor)
-		so_ref(NULL, &nv50->state.scissor);
-	if (nv50->state.viewport)
-		so_ref(NULL, &nv50->state.viewport);
-	if (nv50->state.tsc_upload)
-		so_ref(NULL, &nv50->state.tsc_upload);
-	if (nv50->state.tic_upload)
-		so_ref(NULL, &nv50->state.tic_upload);
-	if (nv50->state.vertprog)
-		so_ref(NULL, &nv50->state.vertprog);
-	if (nv50->state.fragprog)
-		so_ref(NULL, &nv50->state.fragprog);
-	if (nv50->state.geomprog)
-		so_ref(NULL, &nv50->state.geomprog);
-	if (nv50->state.fp_linkage)
-		so_ref(NULL, &nv50->state.fp_linkage);
-	if (nv50->state.gp_linkage)
-		so_ref(NULL, &nv50->state.gp_linkage);
-	if (nv50->state.vtxfmt)
-		so_ref(NULL, &nv50->state.vtxfmt);
-	if (nv50->state.vtxbuf)
-		so_ref(NULL, &nv50->state.vtxbuf);
-	if (nv50->state.vtxattr)
-		so_ref(NULL, &nv50->state.vtxattr);
+	for (i = 0; i < 64; i++) {
+		if (!nv50->state.hw[i])
+			continue;
+		so_ref(NULL, &nv50->state.hw[i]);
+	}
 
 	draw_destroy(nv50->draw);
 
@@ -123,7 +93,6 @@ nv50_create(struct pipe_screen *pscreen, void *priv)
 	nv50->pipe.is_buffer_referenced = nouveau_is_buffer_referenced;
 
 	screen->base.channel->user_private = nv50;
-	screen->base.channel->flush_notify = nv50_state_flush_notify;
 
 	nv50_init_surface_functions(nv50);
 	nv50_init_state_functions(nv50);
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index c540594b949..8793c2aac5d 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -72,6 +72,12 @@ struct nv50_sampler_stateobj {
 	unsigned tsc[8];
 };
 
+struct nv50_vtxelt_stateobj {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+	uint32_t hw[16];
+};
+
 static INLINE unsigned
 get_tile_height(uint32_t tile_mode)
 {
@@ -90,10 +96,12 @@ struct nv50_miptree_level {
 	unsigned tile_mode;
 };
 
+#define NV50_MAX_TEXTURE_LEVELS 16
+
 struct nv50_miptree {
 	struct nouveau_miptree base;
 
-	struct nv50_miptree_level level[PIPE_MAX_TEXTURE_LEVELS];
+	struct nv50_miptree_level level[NV50_MAX_TEXTURE_LEVELS];
 	int image_nr;
 	int total_size;
 };
@@ -115,30 +123,12 @@ nv50_surface(struct pipe_surface *pt)
 }
 
 struct nv50_state {
-	unsigned dirty;
+	struct nouveau_stateobj *hw[64];
+	uint64_t hw_dirty;
 
-	struct nouveau_stateobj *fb;
-	struct nouveau_stateobj *blend;
-	struct nouveau_stateobj *blend_colour;
-	struct nouveau_stateobj *zsa;
-	struct nouveau_stateobj *stencil_ref;
-	struct nouveau_stateobj *rast;
-	struct nouveau_stateobj *stipple;
-	struct nouveau_stateobj *scissor;
-	unsigned scissor_enabled;
-	struct nouveau_stateobj *viewport;
-	struct nouveau_stateobj *tsc_upload;
-	struct nouveau_stateobj *tic_upload;
 	unsigned miptree_nr[PIPE_SHADER_TYPES];
-	struct nouveau_stateobj *vertprog;
-	struct nouveau_stateobj *fragprog;
-	struct nouveau_stateobj *geomprog;
-	struct nouveau_stateobj *fp_linkage;
-	struct nouveau_stateobj *gp_linkage;
-	struct nouveau_stateobj *vtxfmt;
 	struct nouveau_stateobj *vtxbuf;
 	struct nouveau_stateobj *vtxattr;
-	struct nouveau_stateobj *instbuf;
 	unsigned vtxelt_nr;
 };
 
@@ -167,14 +157,13 @@ struct nv50_context {
 	struct pipe_buffer *constbuf[PIPE_SHADER_TYPES];
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
+	struct nv50_vtxelt_stateobj *vtxelt;
 	struct nv50_sampler_stateobj *sampler[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 	unsigned sampler_nr[PIPE_SHADER_TYPES];
 	struct nv50_miptree *miptree[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 	unsigned miptree_nr[PIPE_SHADER_TYPES];
 
-	uint16_t vbo_fifo;
+	unsigned vbo_fifo;
 };
 
 static INLINE struct nv50_context *
@@ -216,24 +205,36 @@ extern void nv50_draw_elements_instanced(struct pipe_context *pipe,
 					 unsigned count,
 					 unsigned startInstance,
 					 unsigned instanceCount);
-extern void nv50_vbo_validate(struct nv50_context *nv50);
+extern void nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso);
+extern struct nouveau_stateobj *nv50_vbo_validate(struct nv50_context *nv50);
+
+/* nv50_push.c */
+extern void
+nv50_push_elements_instanced(struct pipe_context *, struct pipe_buffer *,
+			     unsigned idxsize, unsigned mode, unsigned start,
+			     unsigned count, unsigned i_start,
+			     unsigned i_count);
 
 /* nv50_clear.c */
 extern void nv50_clear(struct pipe_context *pipe, unsigned buffers,
 		       const float *rgba, double depth, unsigned stencil);
 
 /* nv50_program.c */
-extern void nv50_vertprog_validate(struct nv50_context *nv50);
-extern void nv50_fragprog_validate(struct nv50_context *nv50);
-extern void nv50_geomprog_validate(struct nv50_context *nv50);
-extern void nv50_fp_linkage_validate(struct nv50_context *nv50);
-extern void nv50_gp_linkage_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_vertprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_fragprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_geomprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_fp_linkage_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_gp_linkage_validate(struct nv50_context *nv50);
 extern void nv50_program_destroy(struct nv50_context *nv50,
 				 struct nv50_program *p);
 
 /* nv50_state_validate.c */
-extern boolean nv50_state_validate(struct nv50_context *nv50);
-extern void nv50_state_flush_notify(struct nouveau_channel *chan);
+extern boolean nv50_state_validate(struct nv50_context *nv50, unsigned dwords);
 
 extern void nv50_so_init_sifc(struct nv50_context *nv50,
 			      struct nouveau_stateobj *so,
@@ -241,7 +242,8 @@ extern void nv50_so_init_sifc(struct nv50_context *nv50,
 			      unsigned offset, unsigned size);
 
 /* nv50_tex.c */
-extern void nv50_tex_validate(struct nv50_context *);
+extern void nv50_tex_relocs(struct nv50_context *);
+extern struct nouveau_stateobj *nv50_tex_validate(struct nv50_context *);
 
 /* nv50_transfer.c */
 extern void
@@ -255,4 +257,35 @@ nv50_upload_sifc(struct nv50_context *nv50,
 struct pipe_context *
 nv50_create(struct pipe_screen *pscreen, void *priv);
 
+static INLINE unsigned
+nv50_prim(unsigned mode)
+{
+	switch (mode) {
+	case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
+	case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
+	case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
+	case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
+	case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
+	case PIPE_PRIM_TRIANGLE_STRIP:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
+	case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
+	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
+	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
+	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
+	case PIPE_PRIM_LINES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
+	case PIPE_PRIM_LINE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
+	case PIPE_PRIM_TRIANGLES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
+	case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
+	default:
+		break;
+	}
+
+	NOUVEAU_ERR("invalid primitive type %d\n", mode);
+	return NV50TCL_VERTEX_BEGIN_POINTS;
+}
+
 #endif
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 3f9d869d7a4..e091cae6024 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -104,7 +104,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 		tile_flags = 0x7400;
 		break;
 	default:
-		if ((pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) &&
+		if ((pt->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT) &&
 		    util_format_get_blocksizebits(pt->format) == 32)
 			tile_flags = 0x7a00;
 		else
@@ -255,9 +255,10 @@ void
 nv50_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv50_miptree_create;
-	pscreen->texture_blanket = nv50_miptree_blanket;
 	pscreen->texture_destroy = nv50_miptree_destroy;
 	pscreen->get_tex_surface = nv50_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv50_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv50_miptree_blanket;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 2372cbbef69..c857816b31a 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -4270,7 +4270,7 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 	FREE(up);
 }
 
-void
+struct nouveau_stateobj *
 nv50_vertprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4286,6 +4286,9 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_VERTPROG))
+		return NULL;
+
 	so = so_new(5, 7, 2);
 	so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4301,11 +4304,10 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.high_temp);
 	so_method(so, tesla, NV50TCL_VP_START_ID, 1);
 	so_data  (so, 0); /* program start offset */
-	so_ref(so, &nv50->state.vertprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
-void
+struct nouveau_stateobj *
 nv50_fragprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4321,6 +4323,9 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_FRAGPROG))
+		return NULL;
+
 	so = so_new(6, 7, 2);
 	so_method(so, tesla, NV50TCL_FP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4337,11 +4342,10 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.regs[3]);
 	so_method(so, tesla, NV50TCL_FP_START_ID, 1);
 	so_data  (so, 0); /* program start offset */
-	so_ref(so, &nv50->state.fragprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
-void
+struct nouveau_stateobj *
 nv50_geomprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4357,6 +4361,9 @@ nv50_geomprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_GEOMPROG))
+		return NULL;
+
 	so = so_new(6, 7, 2);
 	so_method(so, tesla, NV50TCL_GP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4373,8 +4380,7 @@ nv50_geomprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.vert_count);
 	so_method(so, tesla, NV50TCL_GP_START_ID, 1);
 	so_data  (so, 0);
-	so_ref(so, &nv50->state.geomprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static uint32_t
@@ -4454,7 +4460,7 @@ nv50_vec4_map(uint32_t *map32, int mid, uint8_t zval, uint32_t lin[4],
 	return mid;
 }
 
-void
+struct nouveau_stateobj *
 nv50_fp_linkage_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4580,8 +4586,7 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
 	so_method(so, tesla, NV50TCL_GP_ENABLE, 1);
 	so_data  (so, (vp->type == PIPE_SHADER_GEOMETRY) ? 1 : 0);
 
-	so_ref(so, &nv50->state.fp_linkage);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static int
@@ -4615,7 +4620,7 @@ construct_vp_gp_mapping(uint32_t *map32, int m,
 	return m;
 }
 
-void
+struct nouveau_stateobj *
 nv50_gp_linkage_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4625,10 +4630,8 @@ nv50_gp_linkage_validate(struct nv50_context *nv50)
 	uint32_t map[16];
 	int m = 0;
 
-	if (!gp) {
-		so_ref(NULL, &nv50->state.gp_linkage);
-		return;
-	}
+	if (!gp)
+		return NULL;
 	memset(map, 0, sizeof(map));
 
 	m = construct_vp_gp_mapping(map, m, vp, gp);
@@ -4646,8 +4649,7 @@ nv50_gp_linkage_validate(struct nv50_context *nv50)
 	so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), m);
 	so_datap (so, map, m);
 
-	so_ref(so, &nv50->state.gp_linkage);
-	so_ref(NULL, &so);
+	return so;
 }
 
 void
diff --git a/src/gallium/drivers/nv50/nv50_push.c b/src/gallium/drivers/nv50/nv50_push.c
new file mode 100644
index 00000000000..96a1f32d304
--- /dev/null
+++ b/src/gallium/drivers/nv50/nv50_push.c
@@ -0,0 +1,326 @@
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+
+#include "nouveau/nouveau_util.h"
+#include "nv50_context.h"
+
+struct push_context {
+   struct nv50_context *nv50;
+
+   unsigned vtx_size;
+
+   void *idxbuf;
+   unsigned idxsize;
+
+   float edgeflag;
+   int edgeflag_attr;
+
+   struct {
+      void *map;
+      unsigned stride;
+      unsigned divisor;
+      unsigned step;
+      void (*push)(struct nouveau_channel *, void *);
+   } attr[16];
+   unsigned attr_nr;
+};
+
+static void
+emit_b32_1(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b32_2(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+}
+
+static void
+emit_b32_3(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+   OUT_RING(chan, v[2]);
+}
+
+static void
+emit_b32_4(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+   OUT_RING(chan, v[2]);
+   OUT_RING(chan, v[3]);
+}
+
+static void
+emit_b16_1(struct nouveau_channel *chan, void *data)
+{
+   uint16_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b16_3(struct nouveau_channel *chan, void *data)
+{
+   uint16_t *v = data;
+
+   OUT_RING(chan, (v[1] << 16) | v[0]);
+   OUT_RING(chan, v[2]);
+}
+
+static void
+emit_b08_1(struct nouveau_channel *chan, void *data)
+{
+   uint8_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b08_3(struct nouveau_channel *chan, void *data)
+{
+   uint8_t *v = data;
+
+   OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
+}
+
+static INLINE void
+emit_vertex(struct push_context *ctx, unsigned n)
+{
+   struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+   int i;
+
+   if (ctx->edgeflag_attr < 16) {
+      float *edgeflag = ctx->attr[ctx->edgeflag_attr].map +
+                        ctx->attr[ctx->edgeflag_attr].stride * n;
+
+      if (*edgeflag != ctx->edgeflag) {
+         BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+         OUT_RING  (chan, *edgeflag ? 1 : 0);
+         ctx->edgeflag = *edgeflag;
+      }
+   }
+
+   BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, ctx->vtx_size);
+   for (i = 0; i < ctx->attr_nr; i++)
+      ctx->attr[i].push(chan, ctx->attr[i].map + ctx->attr[i].stride * n);
+}
+
+static void
+emit_edgeflag(void *priv, boolean enabled)
+{
+   struct push_context *ctx = priv;
+   struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+
+   BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+   OUT_RING  (chan, enabled ? 1 : 0);
+}
+
+static void
+emit_elt08(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint8_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_elt16(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint16_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_elt32(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint32_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_verts(void *priv, unsigned start, unsigned count)
+{
+   while (count--)
+      emit_vertex(priv, start++);
+}
+
+void
+nv50_push_elements_instanced(struct pipe_context *pipe,
+                             struct pipe_buffer *idxbuf, unsigned idxsize,
+                             unsigned mode, unsigned start, unsigned count,
+                             unsigned i_start, unsigned i_count)
+{
+   struct nv50_context *nv50 = nv50_context(pipe);
+   struct nouveau_grobj *tesla = nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+   struct push_context ctx;
+   const unsigned p_overhead = 4 + /* begin/end */
+                               4; /* potential edgeflag enable/disable */
+   const unsigned v_overhead = 1 + /* VERTEX_DATA packet header */
+                               2; /* potential edgeflag modification */
+   struct u_split_prim s;
+   unsigned vtx_size;
+   boolean nzi = FALSE;
+   int i;
+
+   ctx.nv50 = nv50;
+   ctx.attr_nr = 0;
+   ctx.idxbuf = NULL;
+   ctx.vtx_size = 0;
+   ctx.edgeflag = 0.5f;
+   ctx.edgeflag_attr = nv50->vertprog->cfg.edgeflag_in;
+
+   /* map vertex buffers, determine vertex size */
+   for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+      struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
+      struct pipe_vertex_buffer *vb = &nv50->vtxbuf[ve->vertex_buffer_index];
+      struct nouveau_bo *bo = nouveau_bo(vb->buffer);
+      unsigned size, nr_components, n;
+
+      if (!(nv50->vbo_fifo & (1 << i)))
+         continue;
+      n = ctx.attr_nr++;
+
+      if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
+         assert(bo->map);
+         return;
+      }
+      ctx.attr[n].map = bo->map + vb->buffer_offset + ve->src_offset;
+      nouveau_bo_unmap(bo);
+
+      ctx.attr[n].stride = vb->stride;
+      ctx.attr[n].divisor = ve->instance_divisor;
+      if (ctx.attr[n].divisor) {
+         ctx.attr[n].step = i_start % ve->instance_divisor;
+         ctx.attr[n].map += i_start * vb->stride;
+      }
+
+      size = util_format_get_component_bits(ve->src_format,
+                                            UTIL_FORMAT_COLORSPACE_RGB, 0);
+      nr_components = util_format_get_nr_components(ve->src_format);
+      switch (size) {
+      case 8:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b08_1; break;
+         case 2: ctx.attr[n].push = emit_b16_1; break;
+         case 3: ctx.attr[n].push = emit_b08_3; break;
+         case 4: ctx.attr[n].push = emit_b32_1; break;
+         }
+         ctx.vtx_size++;
+         break;
+      case 16:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b16_1; break;
+         case 2: ctx.attr[n].push = emit_b32_1; break;
+         case 3: ctx.attr[n].push = emit_b16_3; break;
+         case 4: ctx.attr[n].push = emit_b32_2; break;
+         }
+         ctx.vtx_size += (nr_components + 1) >> 1;
+         break;
+      case 32:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b32_1; break;
+         case 2: ctx.attr[n].push = emit_b32_2; break;
+         case 3: ctx.attr[n].push = emit_b32_3; break;
+         case 4: ctx.attr[n].push = emit_b32_4; break;
+         }
+         ctx.vtx_size += nr_components;
+         break;
+      default:
+         assert(0);
+         return;
+      }
+   }
+   vtx_size = ctx.vtx_size + v_overhead;
+
+   /* map index buffer, if present */
+   if (idxbuf) {
+      struct nouveau_bo *bo = nouveau_bo(idxbuf);
+
+      if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
+         assert(bo->map);
+         return;
+      }
+      ctx.idxbuf = bo->map;
+      ctx.idxsize = idxsize;
+      nouveau_bo_unmap(bo);
+   }
+
+   s.priv = &ctx;
+   s.edge = emit_edgeflag;
+   if (idxbuf) {
+      if (idxsize == 1)
+         s.emit = emit_elt08;
+      else
+      if (idxsize == 2)
+         s.emit = emit_elt16;
+      else
+         s.emit = emit_elt32;
+   } else
+      s.emit = emit_verts;
+
+   /* per-instance loop */
+   BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+   OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+   OUT_RING  (chan, i_start);
+   while (i_count--) {
+      unsigned max_verts;
+      boolean done;
+
+      for (i = 0; i < ctx.attr_nr; i++) {
+         if (!ctx.attr[i].divisor ||
+              ctx.attr[i].divisor != ++ctx.attr[i].step)
+            continue;
+         ctx.attr[i].step = 0;
+         ctx.attr[i].map += ctx.attr[i].stride;
+      }
+
+      u_split_prim_init(&s, mode, start, count);
+      do {
+         if (AVAIL_RING(chan) < p_overhead + (6 * vtx_size)) {
+            FIRE_RING(chan);
+            if (!nv50_state_validate(nv50, p_overhead + (6 * vtx_size))) {
+               assert(0);
+               return;
+            }
+         }
+
+         max_verts  = AVAIL_RING(chan);
+         max_verts -= p_overhead;
+         max_verts /= vtx_size;
+
+         BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+         OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1 << 28) : 0));
+         done = u_split_prim_next(&s, max_verts);
+         BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+         OUT_RING  (chan, 0);
+      } while (!done);
+
+      nzi = TRUE;
+   }
+}
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index eed6031eafa..7e2e8aa336e 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -95,6 +95,8 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen,
 static int
 nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 {
+	struct nv50_screen *screen = nv50_screen(pscreen);
+
 	switch (param) {
 	case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
 		return 32;
@@ -132,9 +134,9 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 		return 1;
 	case NOUVEAU_CAP_HW_VTXBUF:
-		return 1;
+		return screen->force_push ? 0 : 1;
 	case NOUVEAU_CAP_HW_IDXBUF:
-		return 1;
+		return screen->force_push ? 0 : 1;
 	case PIPE_CAP_INDEP_BLEND_ENABLE:
 		return 1;
 	case PIPE_CAP_INDEP_BLEND_FUNC:
@@ -202,28 +204,6 @@ nv50_screen_destroy(struct pipe_screen *pscreen)
 	FREE(screen);
 }
 
-static int
-nv50_pre_pipebuffer_map(struct pipe_screen *pscreen, struct pipe_buffer *pb,
-	unsigned usage)
-{
-	struct nv50_screen *screen = nv50_screen(pscreen);
-	struct nv50_context *ctx = screen->cur_ctx;
-
-	if (!(pb->usage & PIPE_BUFFER_USAGE_VERTEX))
-		return 0;
-
-	/* Our vtxbuf got mapped, it can no longer be considered part of current
-	 * state, remove it to avoid emitting reloc markers.
-	 */
-	if (ctx && ctx->state.vtxbuf && so_bo_is_reloc(ctx->state.vtxbuf,
-			nouveau_bo(pb))) {
-		so_ref(NULL, &ctx->state.vtxbuf);
-		ctx->dirty |= NV50_NEW_ARRAYS;
-	}
-
-	return 0;
-}
-
 struct pipe_screen *
 nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 {
@@ -252,7 +232,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	pscreen->get_paramf = nv50_screen_get_paramf;
 	pscreen->is_format_supported = nv50_screen_is_format_supported;
 	pscreen->context_create = nv50_create;
-	screen->base.pre_pipebuffer_map_callback = nv50_pre_pipebuffer_map;
 
 	nv50_screen_init_miptree_functions(pscreen);
 	nv50_transfer_init_screen_functions(pscreen);
@@ -508,10 +487,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_method(so, screen->tesla, NV50TCL_LINKED_TSC, 1);
 	so_data  (so, 1);
 
-	/* activate first scissor rectangle */
-	so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
-	so_data  (so, 1);
-
 	so_method(so, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
 	so_data  (so, 1); /* default edgeflag to TRUE */
 
@@ -520,6 +495,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref (NULL, &so);
 	nouveau_pushbuf_flush(chan, 0);
 
+	screen->force_push = debug_get_bool_option("NV50_ALWAYS_PUSH", FALSE);
 	return pscreen;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index 2687b721277..d1bc80cb9ed 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -28,6 +28,8 @@ struct nv50_screen {
 	struct nouveau_bo *tsc;
 
 	struct nouveau_stateobj *static_init;
+
+	boolean force_push;
 };
 
 static INLINE struct nv50_screen *
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 7d304907b65..b0e5552eff4 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -302,7 +302,7 @@ static void *
 nv50_rasterizer_state_create(struct pipe_context *pipe,
 			     const struct pipe_rasterizer_state *cso)
 {
-	struct nouveau_stateobj *so = so_new(15, 21, 0);
+	struct nouveau_stateobj *so = so_new(16, 22, 0);
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_rasterizer_stateobj *rso =
 		CALLOC_STRUCT(nv50_rasterizer_stateobj);
@@ -314,6 +314,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
 	 * 	- point_sprite / sprite_coord_mode
 	 */
 
+	so_method(so, tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
+	so_data  (so, cso->scissor);
+
 	so_method(so, tesla, NV50TCL_SHADE_MODEL, 1);
 	so_data  (so, cso->flatshade ? NV50TCL_SHADE_MODEL_FLAT :
 				       NV50TCL_SHADE_MODEL_SMOOTH);
@@ -720,15 +723,34 @@ nv50_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	nv50->dirty |= NV50_NEW_ARRAYS;
 }
 
+static void *
+nv50_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv50_vtxelt_stateobj *cso = CALLOC_STRUCT(nv50_vtxelt_stateobj);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+	nv50_vtxelt_construct(cso);
+
+	return (void *)cso;
+}
+
 static void
-nv50_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv50_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv50->vtxelt, ve, sizeof(*ve) * count);
-	nv50->vtxelt_nr = count;
+static void
+nv50_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
 
+	nv50->vtxelt = hwcso;
 	nv50->dirty |= NV50_NEW_ARRAYS;
 }
 
@@ -778,7 +800,10 @@ nv50_init_state_functions(struct nv50_context *nv50)
 	nv50->pipe.set_scissor_state = nv50_set_scissor_state;
 	nv50->pipe.set_viewport_state = nv50_set_viewport_state;
 
+	nv50->pipe.create_vertex_elements_state = nv50_vtxelts_state_create;
+	nv50->pipe.delete_vertex_elements_state = nv50_vtxelts_state_delete;
+	nv50->pipe.bind_vertex_elements_state = nv50_vtxelts_state_bind;
+
 	nv50->pipe.set_vertex_buffers = nv50_set_vertex_buffers;
-	nv50->pipe.set_vertex_elements = nv50_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index c974cc92dcc..2c8e7ca7982 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -25,8 +25,8 @@
 #include "nv50_context.h"
 #include "nouveau/nouveau_stateobj.h"
 
-static void
-nv50_state_validate_fb(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_fb(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nouveau_stateobj *so = so_new(32, 79, 18);
@@ -167,12 +167,7 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 	so_data  (so, w << 16);
 	so_data  (so, h << 16);
 
-	/* we set scissors to framebuffer size when they're 'turned off' */
-	nv50->dirty |= NV50_NEW_SCISSOR;
-	so_ref(NULL, &nv50->state.scissor);
-
-	so_ref(so, &nv50->state.fb);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static void
@@ -199,263 +194,256 @@ nv50_validate_samplers(struct nv50_context *nv50, struct nouveau_stateobj *so,
 	}
 }
 
-static void
-nv50_state_emit(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_blend(struct nv50_context *nv50)
 {
-	struct nv50_screen *screen = nv50->screen;
-	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->blend->so, &so);
+	return so;
+}
 
-	/* XXX: this is racy for multiple contexts active on separate
-	 * threads.
-	 */
-	if (screen->cur_ctx != nv50) {
-		if (nv50->state.fb)
-			nv50->state.dirty |= NV50_NEW_FRAMEBUFFER;
-		if (nv50->state.blend)
-			nv50->state.dirty |= NV50_NEW_BLEND;
-		if (nv50->state.zsa)
-			nv50->state.dirty |= NV50_NEW_ZSA;
-		if (nv50->state.vertprog)
-			nv50->state.dirty |= NV50_NEW_VERTPROG;
-		if (nv50->state.fragprog)
-			nv50->state.dirty |= NV50_NEW_FRAGPROG;
-		if (nv50->state.geomprog)
-			nv50->state.dirty |= NV50_NEW_GEOMPROG;
-		if (nv50->state.rast)
-			nv50->state.dirty |= NV50_NEW_RASTERIZER;
-		if (nv50->state.blend_colour)
-			nv50->state.dirty |= NV50_NEW_BLEND_COLOUR;
-		if (nv50->state.stencil_ref)
-			nv50->state.dirty |= NV50_NEW_STENCIL_REF;
-		if (nv50->state.stipple)
-			nv50->state.dirty |= NV50_NEW_STIPPLE;
-		if (nv50->state.scissor)
-			nv50->state.dirty |= NV50_NEW_SCISSOR;
-		if (nv50->state.viewport)
-			nv50->state.dirty |= NV50_NEW_VIEWPORT;
-		if (nv50->state.tsc_upload)
-			nv50->state.dirty |= NV50_NEW_SAMPLER;
-		if (nv50->state.tic_upload)
-			nv50->state.dirty |= NV50_NEW_TEXTURE;
-		if (nv50->state.vtxfmt && nv50->state.vtxbuf)
-			nv50->state.dirty |= NV50_NEW_ARRAYS;
-		screen->cur_ctx = nv50;
-	}
+static struct nouveau_stateobj *
+validate_zsa(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->zsa->so, &so);
+	return so;
+}
 
-	if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER)
-		so_emit(chan, nv50->state.fb);
-	if (nv50->state.dirty & NV50_NEW_BLEND)
-		so_emit(chan, nv50->state.blend);
-	if (nv50->state.dirty & NV50_NEW_ZSA)
-		so_emit(chan, nv50->state.zsa);
-	if (nv50->state.dirty & NV50_NEW_VERTPROG)
-		so_emit(chan, nv50->state.vertprog);
-	if (nv50->state.dirty & NV50_NEW_FRAGPROG)
-		so_emit(chan, nv50->state.fragprog);
-	if (nv50->state.dirty & NV50_NEW_GEOMPROG && nv50->state.geomprog)
-		so_emit(chan, nv50->state.geomprog);
-	if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-				 NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
-		so_emit(chan, nv50->state.fp_linkage);
-	if ((nv50->state.dirty & (NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG))
-	    && nv50->state.gp_linkage)
-		so_emit(chan, nv50->state.gp_linkage);
-	if (nv50->state.dirty & NV50_NEW_RASTERIZER)
-		so_emit(chan, nv50->state.rast);
-	if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
-		so_emit(chan, nv50->state.blend_colour);
-	if (nv50->state.dirty & NV50_NEW_STENCIL_REF)
-		so_emit(chan, nv50->state.stencil_ref);
-	if (nv50->state.dirty & NV50_NEW_STIPPLE)
-		so_emit(chan, nv50->state.stipple);
-	if (nv50->state.dirty & NV50_NEW_SCISSOR)
-		so_emit(chan, nv50->state.scissor);
-	if (nv50->state.dirty & NV50_NEW_VIEWPORT)
-		so_emit(chan, nv50->state.viewport);
-	if (nv50->state.dirty & NV50_NEW_SAMPLER)
-		so_emit(chan, nv50->state.tsc_upload);
-	if (nv50->state.dirty & NV50_NEW_TEXTURE)
-		so_emit(chan, nv50->state.tic_upload);
-	if (nv50->state.dirty & NV50_NEW_ARRAYS) {
-		so_emit(chan, nv50->state.vtxfmt);
-		so_emit(chan, nv50->state.vtxbuf);
-		if (nv50->state.vtxattr)
-			so_emit(chan, nv50->state.vtxattr);
-	}
-	nv50->state.dirty = 0;
+static struct nouveau_stateobj *
+validate_rast(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->rasterizer->so, &so);
+	return so;
 }
 
-void
-nv50_state_flush_notify(struct nouveau_channel *chan)
+static struct nouveau_stateobj *
+validate_blend_colour(struct nv50_context *nv50)
 {
-	struct nv50_context *nv50 = chan->user_private;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(1, 4, 0);
+
+	so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
+	so_data  (so, fui(nv50->blend_colour.color[0]));
+	so_data  (so, fui(nv50->blend_colour.color[1]));
+	so_data  (so, fui(nv50->blend_colour.color[2]));
+	so_data  (so, fui(nv50->blend_colour.color[3]));
+	return so;
+}
 
-	if (nv50->state.tic_upload && !(nv50->dirty & NV50_NEW_TEXTURE))
-		so_emit(chan, nv50->state.tic_upload);
+static struct nouveau_stateobj *
+validate_stencil_ref(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so = so_new(2, 2, 0);
 
-	so_emit_reloc_markers(chan, nv50->state.fb);
-	so_emit_reloc_markers(chan, nv50->state.vertprog);
-	so_emit_reloc_markers(chan, nv50->state.fragprog);
-	so_emit_reloc_markers(chan, nv50->state.vtxbuf);
-	so_emit_reloc_markers(chan, nv50->screen->static_init);
+	so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1);
+	so_data  (so, nv50->stencil_ref.ref_value[0]);
+	so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1);
+	so_data  (so, nv50->stencil_ref.ref_value[1]);
+	return so;
+}
 
-	if (nv50->state.instbuf)
-		so_emit_reloc_markers(chan, nv50->state.instbuf);
+static struct nouveau_stateobj *
+validate_stipple(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(1, 32, 0);
+	int i;
+
+	so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
+	for (i = 0; i < 32; i++)
+		so_data(so, util_bswap32(nv50->stipple.stipple[i]));
+	return so;
 }
 
-boolean
-nv50_state_validate(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_scissor(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
+        struct pipe_scissor_state *s = &nv50->scissor;
 	struct nouveau_stateobj *so;
-	unsigned i;
 
-	if (nv50->dirty & NV50_NEW_FRAMEBUFFER)
-		nv50_state_validate_fb(nv50);
+	so = so_new(1, 2, 0);
+	so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
+	so_data  (so, (s->maxx << 16) | s->minx);
+	so_data  (so, (s->maxy << 16) | s->miny);
+	return so;
+}
+
+static struct nouveau_stateobj *
+validate_viewport(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(5, 9, 0);
+
+	so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
+	so_data  (so, fui(nv50->viewport.translate[0]));
+	so_data  (so, fui(nv50->viewport.translate[1]));
+	so_data  (so, fui(nv50->viewport.translate[2]));
+	so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3);
+	so_data  (so, fui(nv50->viewport.scale[0]));
+	so_data  (so, fui(nv50->viewport.scale[1]));
+	so_data  (so, fui(nv50->viewport.scale[2]));
+
+	so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
+	so_data  (so, 1);
+	/* 0x0000 = remove whole primitive only (xyz)
+	 * 0x1018 = remove whole primitive only (xy), clamp z
+	 * 0x1080 = clip primitive (xyz)
+	 * 0x1098 = clip primitive (xy), clamp z
+	 */
+	so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
+	so_data  (so, 0x1080);
+	/* no idea what 0f90 does */
+	so_method(so, tesla, 0x0f90, 1);
+	so_data  (so, 0);
+
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_BLEND)
-		so_ref(nv50->blend->so, &nv50->state.blend);
+static struct nouveau_stateobj *
+validate_sampler(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so;
+	unsigned nr = 0, i;
 
-	if (nv50->dirty & NV50_NEW_ZSA)
-		so_ref(nv50->zsa->so, &nv50->state.zsa);
+	for (i = 0; i < PIPE_SHADER_TYPES; ++i)
+		nr += nv50->sampler_nr[i];
 
-	if (nv50->dirty & (NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB))
-		nv50_vertprog_validate(nv50);
+	so = so_new(1 + 5 * PIPE_SHADER_TYPES,
+		    1 + 19 * PIPE_SHADER_TYPES + nr * 8,
+		    PIPE_SHADER_TYPES * 2);
 
-	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB))
-		nv50_fragprog_validate(nv50);
+	nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX);
+	nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT);
 
-	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB))
-		nv50_geomprog_validate(nv50);
+	so_method(so, tesla, 0x1334, 1); /* flush TSC */
+	so_data  (so, 0);
 
-	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-			   NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
-		nv50_fp_linkage_validate(nv50);
+	return so;
+}
 
-	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_VERTPROG))
-		nv50_gp_linkage_validate(nv50);
+static struct nouveau_stateobj *
+validate_vtxbuf(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->state.vtxbuf, &so);
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_RASTERIZER)
-		so_ref(nv50->rasterizer->so, &nv50->state.rast);
+static struct nouveau_stateobj *
+validate_vtxattr(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->state.vtxattr, &so);
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_BLEND_COLOUR) {
-		so = so_new(1, 4, 0);
-		so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
-		so_data  (so, fui(nv50->blend_colour.color[0]));
-		so_data  (so, fui(nv50->blend_colour.color[1]));
-		so_data  (so, fui(nv50->blend_colour.color[2]));
-		so_data  (so, fui(nv50->blend_colour.color[3]));
-		so_ref(so, &nv50->state.blend_colour);
-		so_ref(NULL, &so);
-	}
+struct state_validate {
+	struct nouveau_stateobj *(*func)(struct nv50_context *nv50);
+	unsigned states;
+} validate_list[] = {
+	{ validate_fb             , NV50_NEW_FRAMEBUFFER                      },
+	{ validate_blend          , NV50_NEW_BLEND                            },
+	{ validate_zsa            , NV50_NEW_ZSA                              },
+	{ nv50_vertprog_validate  , NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB  },
+	{ nv50_fragprog_validate  , NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB  },
+	{ nv50_geomprog_validate  , NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB  },
+	{ nv50_fp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG |
+				    NV50_NEW_FRAGPROG | NV50_NEW_RASTERIZER   },
+	{ nv50_gp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG     },
+	{ validate_rast           , NV50_NEW_RASTERIZER                       },
+	{ validate_blend_colour   , NV50_NEW_BLEND_COLOUR                     },
+	{ validate_stencil_ref    , NV50_NEW_STENCIL_REF                      },
+	{ validate_stipple        , NV50_NEW_STIPPLE                          },
+	{ validate_scissor        , NV50_NEW_SCISSOR                          },
+	{ validate_viewport       , NV50_NEW_VIEWPORT                         },
+	{ validate_sampler        , NV50_NEW_SAMPLER                          },
+	{ nv50_tex_validate       , NV50_NEW_TEXTURE | NV50_NEW_SAMPLER       },
+	{ nv50_vbo_validate       , NV50_NEW_ARRAYS                           },
+	{ validate_vtxbuf         , NV50_NEW_ARRAYS                           },
+	{ validate_vtxattr        , NV50_NEW_ARRAYS                           },
+	{}
+};
+#define validate_list_len (sizeof(validate_list) / sizeof(validate_list[0]))
 
-	if (nv50->dirty & NV50_NEW_STENCIL_REF) {
-		so = so_new(2, 2, 0);
-		so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1);
-		so_data  (so, nv50->stencil_ref.ref_value[0]);
-		so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1);
-		so_data  (so, nv50->stencil_ref.ref_value[1]);
-		so_ref(so, &nv50->state.stencil_ref);
-		so_ref(NULL, &so);
-	}
+boolean
+nv50_state_validate(struct nv50_context *nv50, unsigned wait_dwords)
+{
+	struct nouveau_channel *chan = nv50->screen->base.channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	unsigned nr_relocs = 128, nr_dwords = wait_dwords + 128 + 4;
+	int ret, i;
 
-	if (nv50->dirty & NV50_NEW_STIPPLE) {
-		so = so_new(1, 32, 0);
-		so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
-		for (i = 0; i < 32; i++)
-			so_data(so, util_bswap32(nv50->stipple.stipple[i]));
-		so_ref(so, &nv50->state.stipple);
-		so_ref(NULL, &so);
-	}
+	for (i = 0; i < validate_list_len; i++) {
+		struct state_validate *validate = &validate_list[i];
+		struct nouveau_stateobj *so;
 
-	if (nv50->dirty & (NV50_NEW_SCISSOR | NV50_NEW_RASTERIZER)) {
-		struct pipe_rasterizer_state *rast = &nv50->rasterizer->pipe;
-		struct pipe_scissor_state *s = &nv50->scissor;
+		if (!(nv50->dirty & validate->states))
+			continue;
 
-		if (nv50->state.scissor &&
-		    (rast->scissor == 0 && nv50->state.scissor_enabled == 0))
-			goto scissor_uptodate;
-		nv50->state.scissor_enabled = rast->scissor;
+		so = validate->func(nv50);
+		if (!so)
+			continue;
 
-		so = so_new(1, 2, 0);
-		so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
-		if (nv50->state.scissor_enabled) {
-			so_data(so, (s->maxx << 16) | s->minx);
-			so_data(so, (s->maxy << 16) | s->miny);
-		} else {
-			so_data(so, (nv50->framebuffer.width << 16));
-			so_data(so, (nv50->framebuffer.height << 16));
-		}
-		so_ref(so, &nv50->state.scissor);
-		so_ref(NULL, &so);
-		nv50->state.dirty |= NV50_NEW_SCISSOR;
-	}
-scissor_uptodate:
-
-	if (nv50->dirty & (NV50_NEW_VIEWPORT | NV50_NEW_RASTERIZER)) {
-		if (nv50->state.viewport &&
-		    !(nv50->dirty & NV50_NEW_VIEWPORT))
-			goto viewport_uptodate;
-
-		so = so_new(5, 9, 0);
-		so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
-		so_data  (so, fui(nv50->viewport.translate[0]));
-		so_data  (so, fui(nv50->viewport.translate[1]));
-		so_data  (so, fui(nv50->viewport.translate[2]));
-		so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3);
-		so_data  (so, fui(nv50->viewport.scale[0]));
-		so_data  (so, fui(nv50->viewport.scale[1]));
-		so_data  (so, fui(nv50->viewport.scale[2]));
-
-		so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
-		so_data  (so, 1);
-		/* 0x0000 = remove whole primitive only (xyz)
-		 * 0x1018 = remove whole primitive only (xy), clamp z
-		 * 0x1080 = clip primitive (xyz)
-		 * 0x1098 = clip primitive (xy), clamp z
-		 */
-		so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
-		so_data  (so, 0x1080);
-		/* no idea what 0f90 does */
-		so_method(so, tesla, 0x0f90, 1);
-		so_data  (so, 0);
+		nr_dwords += (so->total + so->cur);
+		nr_relocs += so->cur_reloc;
 
-		so_ref(so, &nv50->state.viewport);
+		so_ref(so, &nv50->state.hw[i]);
 		so_ref(NULL, &so);
-		nv50->state.dirty |= NV50_NEW_VIEWPORT;
+		nv50->state.hw_dirty |= (1 << i);
 	}
-viewport_uptodate:
-
-	if (nv50->dirty & NV50_NEW_SAMPLER) {
-		unsigned nr = 0;
-
-		for (i = 0; i < PIPE_SHADER_TYPES; ++i)
-			nr += nv50->sampler_nr[i];
+	nv50->dirty = 0;
 
-		so = so_new(1 + 5 * PIPE_SHADER_TYPES,
-			    1 + 19 * PIPE_SHADER_TYPES + nr * 8,
-			    PIPE_SHADER_TYPES * 2);
+	if (nv50->screen->cur_ctx != nv50) {
+		for (i = 0; i < validate_list_len; i++) {
+			if (!nv50->state.hw[i] ||
+			    (nv50->state.hw_dirty & (1 << i)))
+				continue;
 
-		nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX);
-		nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT);
+			nr_dwords += (nv50->state.hw[i]->total +
+				      nv50->state.hw[i]->cur);
+			nr_relocs += nv50->state.hw[i]->cur_reloc;
+			nv50->state.hw_dirty |= (1 << i);
+		}
 
-		so_method(so, tesla, 0x1334, 1); /* flush TSC */
-		so_data  (so, 0);
+		nv50->screen->cur_ctx = nv50;
+	}
 
-		so_ref(so, &nv50->state.tsc_upload);
-		so_ref(NULL, &so);
+	ret = MARK_RING(chan, nr_dwords, nr_relocs);
+	if (ret) {
+		debug_printf("MARK_RING(%d, %d) failed: %d\n",
+			     nr_dwords, nr_relocs, ret);
+		return FALSE;
 	}
 
-	if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER))
-		nv50_tex_validate(nv50);
+	while (nv50->state.hw_dirty) {
+		i = ffs(nv50->state.hw_dirty) - 1;
+		nv50->state.hw_dirty &= ~(1 << i);
 
-	if (nv50->dirty & NV50_NEW_ARRAYS)
-		nv50_vbo_validate(nv50);
+		so_emit(chan, nv50->state.hw[i]);
+	}
 
-	nv50->state.dirty |= nv50->dirty;
-	nv50->dirty = 0;
-	nv50_state_emit(nv50);
+	/* Yes, really, we need to do this.  If a buffer that is referenced
+	 * on the hardware isn't part of changed state above, without doing
+	 * this the kernel is given no clue that the buffer is being used
+	 * still.  This can cause all sorts of fun issues.
+	 */
+	nv50_tex_relocs(nv50);
+	so_emit_reloc_markers(chan, nv50->state.hw[0]); /* fb */
+	so_emit_reloc_markers(chan, nv50->state.hw[3]); /* vp */
+	so_emit_reloc_markers(chan, nv50->state.hw[4]); /* fp */
+	so_emit_reloc_markers(chan, nv50->state.hw[17]); /* vb */
+	so_emit_reloc_markers(chan, nv50->screen->static_init);
 
+	/* No idea.. */
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
 	return TRUE;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index de0560e20cd..4c48b12cd87 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -24,6 +24,7 @@
 #include "nv50_texture.h"
 
 #include "nouveau/nouveau_stateobj.h"
+#include "nouveau/nouveau_reloc.h"
 
 #include "util/u_format.h"
 
@@ -195,6 +196,35 @@ nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so,
 }
 
 void
+nv50_tex_relocs(struct nv50_context *nv50)
+{
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
+	int p, unit;
+
+	p = PIPE_SHADER_FRAGMENT;
+	for (unit = 0; unit < nv50->miptree_nr[p]; unit++) {
+		if (!nv50->miptree[p][unit])
+			continue;
+		nouveau_reloc_emit(chan, nv50->screen->tic,
+				   ((p * 32) + unit) * 32, NULL,
+				   nv50->miptree[p][unit]->base.bo, 0, 0,
+				   NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+				   NOUVEAU_BO_RD, 0, 0);
+	}
+
+	p = PIPE_SHADER_VERTEX;
+	for (unit = 0; unit < nv50->miptree_nr[p]; unit++) {
+		if (!nv50->miptree[p][unit])
+			continue;
+		nouveau_reloc_emit(chan, nv50->screen->tic,
+				   ((p * 32) + unit) * 32, NULL,
+				   nv50->miptree[p][unit]->base.bo, 0, 0,
+				   NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+				   NOUVEAU_BO_RD, 0, 0);
+	}
+}
+
+struct nouveau_stateobj *
 nv50_tex_validate(struct nv50_context *nv50)
 {
 	struct nouveau_stateobj *so;
@@ -217,12 +247,11 @@ nv50_tex_validate(struct nv50_context *nv50)
 		so_ref(NULL, &so);
 
 		NOUVEAU_ERR("failed tex validate\n");
-		return;
+		return NULL;
 	}
 
 	so_method(so, tesla, 0x1330, 1); /* flush TIC */
 	so_data  (so, 0);
 
-	so_ref(so, &nv50->state.tic_upload);
-	so_ref(NULL, &so);
+	return so;
 }
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index 1c8ee0b9adf..6b9c1ee231e 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -25,53 +25,9 @@
 #include "util/u_inlines.h"
 #include "util/u_format.h"
 
+#include "nouveau/nouveau_util.h"
 #include "nv50_context.h"
 
-static boolean
-nv50_push_elements_u08(struct nv50_context *, uint8_t *, unsigned);
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *, uint16_t *, unsigned);
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
-
-static boolean
-nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
-
-#define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
-
-static INLINE unsigned
-nv50_prim(unsigned mode)
-{
-	switch (mode) {
-	case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
-	case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
-	case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
-	case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
-	case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
-	case PIPE_PRIM_TRIANGLE_STRIP:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
-	case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
-	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
-	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
-	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
-	case PIPE_PRIM_LINES_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
-	case PIPE_PRIM_LINE_STRIP_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
-	case PIPE_PRIM_TRIANGLES_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
-	case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
-	default:
-		break;
-	}
-
-	NOUVEAU_ERR("invalid primitive type %d\n", mode);
-	return NV50TCL_VERTEX_BEGIN_POINTS;
-}
-
 static INLINE uint32_t
 nv50_vbo_type_to_hw(enum pipe_format format)
 {
@@ -139,15 +95,16 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 	uint32_t hw_type, hw_size;
 	enum pipe_format pf = ve->src_format;
 	const struct util_format_description *desc;
-	unsigned size;
+	unsigned size, nr_components;
 
 	desc = util_format_description(pf);
 	assert(desc);
 
 	size = util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0);
+	nr_components = util_format_get_nr_components(pf);
 
 	hw_type = nv50_vbo_type_to_hw(pf);
-	hw_size = nv50_vbo_size_to_hw(size, ve->nr_components);
+	hw_size = nv50_vbo_size_to_hw(size, nr_components);
 
 	if (!hw_type || !hw_size) {
 		NOUVEAU_ERR("unsupported vbo format: %s\n", util_format_name(pf));
@@ -161,250 +118,58 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 	return (hw_type | hw_size);
 }
 
-/* For instanced drawing from user buffers, hitting the FIFO repeatedly
- * with the same vertex data is probably worse than uploading all data.
- */
-static boolean
-nv50_upload_vtxbuf(struct nv50_context *nv50, unsigned i)
-{
-	struct nv50_screen *nscreen = nv50->screen;
-	struct pipe_screen *pscreen = &nscreen->base.base;
-	struct pipe_buffer *buf = nscreen->strm_vbuf[i];
-	struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-	uint8_t *src;
-	unsigned size = align(vb->buffer->size, 4096);
-
-	if (buf && buf->size < size)
-		pipe_buffer_reference(&nscreen->strm_vbuf[i], NULL);
-
-	if (!nscreen->strm_vbuf[i]) {
-		nscreen->strm_vbuf[i] = pipe_buffer_create(
-			pscreen, 0, PIPE_BUFFER_USAGE_VERTEX, size);
-		buf = nscreen->strm_vbuf[i];
-	}
-
-	src = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ);
-	if (!src)
-		return FALSE;
-	src += vb->buffer_offset;
-
-	size = (vb->max_index + 1) * vb->stride + 16; /* + 16 is for stride 0 */
-	if (vb->buffer_offset + size > vb->buffer->size)
-		size = vb->buffer->size - vb->buffer_offset;
-
-	pipe_buffer_write(pscreen, buf, vb->buffer_offset, size, src);
-	pipe_buffer_unmap(pscreen, vb->buffer);
-
-	vb->buffer = buf; /* don't pipe_reference, this is a private copy */
-	return TRUE;
-}
-
-static void
-nv50_upload_user_vbufs(struct nv50_context *nv50)
-{
-	unsigned i;
-
-	if (nv50->vbo_fifo)
-		nv50->dirty |= NV50_NEW_ARRAYS;
-	if (!(nv50->dirty & NV50_NEW_ARRAYS))
-		return;
-
-	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
-		if (nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX)
-			continue;
-		nv50_upload_vtxbuf(nv50, i);
-	}
-}
-
-static void
-nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
-{
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	float v[4];
-
-	util_format_read_4f(nv50->vtxelt[i].src_format,
-			    v, 0, data, 0, 0, 0, 1, 1);
-
-	switch (nv50->vtxelt[i].nr_components) {
-	case 4:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_4F_X(i), 4);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		OUT_RINGf (chan, v[2]);
-		OUT_RINGf (chan, v[3]);
-		break;
-	case 3:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_3F_X(i), 3);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		OUT_RINGf (chan, v[2]);
-		break;
-	case 2:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_2F_X(i), 2);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		break;
-	case 1:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_1F(i), 1);
-		OUT_RINGf (chan, v[0]);
-		break;
-	default:
-		assert(0);
-		break;
-	}
-}
-
-static unsigned
-init_per_instance_arrays_immd(struct nv50_context *nv50,
-			      unsigned startInstance,
-			      unsigned pos[16], unsigned step[16])
-{
-	struct nouveau_bo *bo;
-	unsigned i, b, count = 0;
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		++count;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-
-		pos[i] = nv50->vtxelt[i].src_offset +
-			nv50->vtxbuf[b].buffer_offset +
-			startInstance * nv50->vtxbuf[b].stride;
-		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-		if (!bo->map)
-			nouveau_bo_map(bo, NOUVEAU_BO_RD);
-
-		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
-	}
-
-	return count;
-}
-
-static unsigned
-init_per_instance_arrays(struct nv50_context *nv50,
-			 unsigned startInstance,
-			 unsigned pos[16], unsigned step[16])
-{
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
+struct instance {
 	struct nouveau_bo *bo;
-	struct nouveau_stateobj *so;
-	unsigned i, b, count = 0;
-	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
-	if (nv50->vbo_fifo)
-		return init_per_instance_arrays_immd(nv50, startInstance,
-						     pos, step);
-
-	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		++count;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-
-		pos[i] = nv50->vtxelt[i].src_offset +
-			nv50->vtxbuf[b].buffer_offset +
-			startInstance * nv50->vtxbuf[b].stride;
-
-		if (!startInstance) {
-			step[i] = 0;
-			continue;
-		}
-		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
-		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
-	}
-
-	if (count && startInstance) {
-		so_ref (so, &nv50->state.instbuf); /* for flush notify */
-		so_emit(chan, nv50->state.instbuf);
-	}
-	so_ref (NULL, &so);
-
-	return count;
-}
+	unsigned delta;
+	unsigned stride;
+	unsigned step;
+	unsigned divisor;
+};
 
 static void
-step_per_instance_arrays_immd(struct nv50_context *nv50,
-			      unsigned pos[16], unsigned step[16])
+instance_init(struct nv50_context *nv50, struct instance *a, unsigned first)
 {
-	struct nouveau_bo *bo;
-	unsigned i, b;
+	int i;
 
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		if (++step[i] != nv50->vtxelt[i].instance_divisor)
-			continue;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
+		struct pipe_vertex_buffer *vb;
 
-		step[i] = 0;
-		pos[i] += nv50->vtxbuf[b].stride;
+		a[i].divisor = ve->instance_divisor;
+		if (a[i].divisor) {
+			vb = &nv50->vtxbuf[ve->vertex_buffer_index];
 
-		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+			a[i].bo = nouveau_bo(vb->buffer);
+			a[i].stride = vb->stride;
+			a[i].step = first % a[i].divisor;
+			a[i].delta = vb->buffer_offset + ve->src_offset +
+				     (first * a[i].stride);
+		}
 	}
 }
 
 static void
-step_per_instance_arrays(struct nv50_context *nv50,
-			 unsigned pos[16], unsigned step[16])
+instance_step(struct nv50_context *nv50, struct instance *a)
 {
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	struct nouveau_bo *bo;
-	struct nouveau_stateobj *so;
-	unsigned i, b;
-	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
-	if (nv50->vbo_fifo) {
-		step_per_instance_arrays_immd(nv50, pos, step);
-		return;
-	}
-
-	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
+	int i;
 
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		if (!a[i].divisor)
 			continue;
-		b = nv50->vtxelt[i].vertex_buffer_index;
 
-		if (++step[i] == nv50->vtxelt[i].instance_divisor) {
-			step[i] = 0;
-			pos[i] += nv50->vtxbuf[b].stride;
+		BEGIN_RING(chan, tesla,
+			   NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
+		OUT_RELOCh(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		OUT_RELOCl(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		if (++a[i].step == a[i].divisor) {
+			a[i].step = 0;
+			a[i].delta += a[i].stride;
 		}
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
-		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 	}
-
-	so_ref (so, &nv50->state.instbuf); /* for flush notify */
-	so_ref (NULL, &so);
-
-	so_emit(chan, nv50->state.instbuf);
-}
-
-static INLINE void
-nv50_unmap_vbufs(struct nv50_context *nv50)
-{
-        unsigned i;
-
-        for (i = 0; i < nv50->vtxbuf_nr; ++i)
-                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
-                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
 }
 
 void
@@ -415,198 +180,207 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	unsigned i, nz_divisors;
-	unsigned step[16], pos[16];
-
-	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50_upload_user_vbufs(nv50);
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	nv50_state_validate(nv50);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 10 + 16*3))
+		return;
 
-	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, NULL, 0, mode, start,
+					     count, startInstance,
+					     instanceCount);
+		return;
+	}
 
 	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
 	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
 	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 7 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
 
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (nv50->vbo_fifo)
-		nv50_push_arrays(nv50, start, count);
-	else {
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, prim);
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
 		OUT_RING  (chan, start);
 		OUT_RING  (chan, count);
-	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	for (i = 1; i < instanceCount; i++) {
-		if (nz_divisors) /* any non-zero array divisors ? */
-			step_per_instance_arrays(nv50, pos, step);
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
-
-		if (nv50->vbo_fifo)
-			nv50_push_arrays(nv50, start, count);
-		else {
-			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-			OUT_RING  (chan, start);
-			OUT_RING  (chan, count);
-		}
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
-	}
-	nv50_unmap_vbufs(nv50);
 
-	so_ref(NULL, &nv50->state.instbuf);
+		prim |= (1 << 28);
+	}
 }
 
 void
 nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 		 unsigned count)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	boolean ret;
-
-	nv50_state_validate(nv50);
-
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (nv50->vbo_fifo)
-		ret = nv50_push_arrays(nv50, start, count);
-	else {
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-		OUT_RING  (chan, start);
-		OUT_RING  (chan, count);
-		ret = TRUE;
-	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	nv50_unmap_vbufs(nv50);
-
-        /* XXX: not sure what to do if ret != TRUE: flush and retry?
-         */
-        assert(ret);
+	nv50_draw_arrays_instanced(pipe, mode, start, count, 0, 1);
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
-			      unsigned start, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-
-	map += start;
+struct inline_ctx {
+	struct nv50_context *nv50;
+	void *map;
+};
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u08(nv50, map, count);
+static void
+inline_elt08(void *priv, unsigned start, unsigned count)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint8_t *map = (uint8_t *)ctx->map + start;
 
 	if (count & 1) {
 		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
 		map++;
-		count--;
+		count &= ~1;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt16(void *priv, unsigned start, unsigned count)
 {
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-
-	map += start;
-
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u16(nv50, map, count);
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint16_t *map = (uint16_t *)ctx->map + start;
 
 	if (count & 1) {
 		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
+		count &= ~1;
 		map++;
-		count--;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt32(void *priv, unsigned start, unsigned count)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, count);
+	OUT_RINGp    (chan, (uint32_t *)ctx->map + start, count);
+}
+
+static void
+inline_edgeflag(void *priv, boolean enabled)
 {
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+
+	BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+	OUT_RING  (chan, enabled ? 1 : 0);
+}
+
+static void
+nv50_draw_elements_inline(struct pipe_context *pipe,
+			  struct pipe_buffer *indexBuffer, unsigned indexSize,
+			  unsigned mode, unsigned start, unsigned count,
+			  unsigned startInstance, unsigned instanceCount)
+{
+	struct pipe_screen *pscreen = pipe->screen;
+	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct instance a[16];
+	struct inline_ctx ctx;
+	struct u_split_prim s;
+	boolean nzi = FALSE;
+	unsigned overhead;
+
+	overhead = 16*3; /* potential instance adjustments */
+	overhead += 4; /* Begin()/End() */
+	overhead += 4; /* potential edgeflag disable/reenable */
+	overhead += 3; /* potentially 3 VTX_ELT_U16/U32 packet headers */
+
+	s.priv = &ctx;
+	if (indexSize == 1)
+		s.emit = inline_elt08;
+	else
+	if (indexSize == 2)
+		s.emit = inline_elt16;
+	else
+		s.emit = inline_elt32;
+	s.edge = inline_edgeflag;
+
+	ctx.nv50 = nv50;
+	ctx.map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+	assert(ctx.map);
+	if (!ctx.map)
+		return;
 
-	map += start;
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, overhead + 6 + 3))
+		return;
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u32(nv50, map, count);
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		unsigned max_verts;
+		boolean done;
+
+		u_split_prim_init(&s, mode, start, count);
+		do {
+			if (AVAIL_RING(chan) < (overhead + 6)) {
+				FIRE_RING(chan);
+				if (!nv50_state_validate(nv50, (overhead + 6))) {
+					assert(0);
+					return;
+				}
+			}
 
-	while (count) {
-		unsigned nr = count > 2047 ? 2047 : count;
+			max_verts = AVAIL_RING(chan) - overhead;
+			if (max_verts > 2047)
+				max_verts = 2047;
+			if (indexSize != 4)
+				max_verts <<= 1;
+			instance_step(nv50, a);
 
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
-		OUT_RINGp (chan, map, nr);
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+			OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1<<28) : 0));
+			done = u_split_prim_next(&s, max_verts);
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+			OUT_RING  (chan, 0);
+		} while (!done);
 
-		count -= nr;
-		map += nr;
+		nzi = TRUE;
 	}
-	return TRUE;
-}
 
-static INLINE void
-nv50_draw_elements_inline(struct nv50_context *nv50,
-			  void *map, unsigned indexSize,
-			  unsigned start, unsigned count)
-{
-	switch (indexSize) {
-	case 1:
-		nv50_draw_elements_inline_u08(nv50, map, start, count);
-		break;
-	case 2:
-		nv50_draw_elements_inline_u16(nv50, map, start, count);
-		break;
-	case 4:
-		nv50_draw_elements_inline_u32(nv50, map, start, count);
-		break;
-	}
+	pipe_buffer_unmap(pscreen, indexBuffer);
 }
 
 void
@@ -617,49 +391,68 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 			     unsigned startInstance, unsigned instanceCount)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	struct pipe_screen *pscreen = pipe->screen;
-	void *map;
-	unsigned i, nz_divisors;
-	unsigned step[16], pos[16];
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
-
-	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50_upload_user_vbufs(nv50);
-
-	nv50_state_validate(nv50);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 13 + 16*3))
+		return;
 
-	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, indexBuffer, indexSize,
+					     mode, start, count, startInstance,
+					     instanceCount);
+		return;
+	} else
+	if (!(indexBuffer->usage & PIPE_BUFFER_USAGE_INDEX) || indexSize == 1) {
+		nv50_draw_elements_inline(pipe, indexBuffer, indexSize,
+					  mode, start, count, startInstance,
+					  instanceCount);
+		return;
+	}
 
 	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
 	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
 	OUT_RING  (chan, startInstance);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	for (i = 1; i < instanceCount; ++i) {
-		if (nz_divisors) /* any non-zero array divisors ? */
-			step_per_instance_arrays(nv50, pos, step);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 10 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
 
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
-
-		nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
+		OUT_RING  (chan, prim);
+		if (indexSize == 4) {
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
+			OUT_RING  (chan, count);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       start << 2, count << 2);
+		} else
+		if (indexSize == 2) {
+			unsigned vb_start = (start & ~1);
+			unsigned vb_end = (start + count + 1) & ~1;
+			unsigned dwords = (vb_end - vb_start) >> 1;
+
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, ((start & 1) << 31) | count);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
+			OUT_RING  (chan, dwords);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       vb_start << 1, dwords << 2);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, 0);
+		}
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
-	}
-	nv50_unmap_vbufs(nv50);
 
-	so_ref(NULL, &nv50->state.instbuf);
+		prim |= (1 << 28);
+	}
 }
 
 void
@@ -667,51 +460,8 @@ nv50_draw_elements(struct pipe_context *pipe,
 		   struct pipe_buffer *indexBuffer, unsigned indexSize,
 		   unsigned mode, unsigned start, unsigned count)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct pipe_screen *pscreen = pipe->screen;
-	void *map;
-	
-	nv50_state_validate(nv50);
-
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (!nv50->vbo_fifo && indexSize == 4) {
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
-		OUT_RING  (chan, count);
-		nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
-				       start << 2, count << 2);
-	} else
-	if (!nv50->vbo_fifo && indexSize == 2) {
-		unsigned vb_start = (start & ~1);
-		unsigned vb_end = (start + count + 1) & ~1;
-		unsigned dwords = (vb_end - vb_start) >> 1;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
-		OUT_RING  (chan, ((start & 1) << 31) | count);
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
-		OUT_RING  (chan, dwords);
-		nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
-				       vb_start << 1, dwords << 2);
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
-		OUT_RING  (chan, 0);
-	} else {
-		map = pipe_buffer_map(pscreen, indexBuffer,
-				      PIPE_BUFFER_USAGE_CPU_READ);
-		nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-		nv50_unmap_vbufs(nv50);
-		pipe_buffer_unmap(pscreen, indexBuffer);
-	}
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
+	nv50_draw_elements_instanced(pipe, indexBuffer, indexSize,
+				     mode, start, count, 0, 1);
 }
 
 static INLINE boolean
@@ -726,6 +476,7 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 	struct nouveau_bo *bo = nouveau_bo(vb->buffer);
 	float v[4];
 	int ret;
+	unsigned nr_components = util_format_get_nr_components(ve->src_format);
 
 	ret = nouveau_bo_map(bo, NOUVEAU_BO_RD);
 	if (ret)
@@ -736,9 +487,10 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 			    0, 0, 1, 1);
 	so = *pso;
 	if (!so)
-		*pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
+		*pso = so = so_new(nv50->vtxelt->num_elements,
+				   nv50->vtxelt->num_elements * 4, 0);
 
-	switch (ve->nr_components) {
+	switch (nr_components) {
 	case 4:
 		so_method(so, tesla, NV50TCL_VTX_ATTR_4F_X(attrib), 4);
 		so_data  (so, fui(v[0]));
@@ -775,6 +527,18 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 }
 
 void
+nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso)
+{
+	unsigned i;
+
+	for (i = 0; i < cso->num_elements; ++i) {
+		struct pipe_vertex_element *ve = &cso->pipe[i];
+
+		cso->hw[i] = nv50_vbo_vtxelt_to_hw(ve);
+	}
+}
+
+struct nouveau_stateobj *
 nv50_vbo_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -783,30 +547,31 @@ nv50_vbo_validate(struct nv50_context *nv50)
 
 	/* don't validate if Gallium took away our buffers */
 	if (nv50->vtxbuf_nr == 0)
-		return;
-	nv50->vbo_fifo = 0;
+		return NULL;
+
+	if (nv50->screen->force_push ||
+	    nv50->vertprog->cfg.edgeflag_in < 16)
+		nv50->vbo_fifo = 0xffff;
 
-	for (i = 0; i < nv50->vtxbuf_nr; ++i)
+	for (i = 0; i < nv50->vtxbuf_nr; i++) {
 		if (nv50->vtxbuf[i].stride &&
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
+	}
 
-	if (NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
-
-	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
+	n_ve = MAX2(nv50->vtxelt->num_elements, nv50->state.vtxelt_nr);
 
 	vtxattr = NULL;
-	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt_nr * 4);
+	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt->num_elements * 4);
 	vtxfmt = so_new(1, n_ve, 0);
 	so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
 
-	for (i = 0; i < nv50->vtxelt_nr; i++) {
-		struct pipe_vertex_element *ve = &nv50->vtxelt[i];
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
 		struct pipe_vertex_buffer *vb =
 			&nv50->vtxbuf[ve->vertex_buffer_index];
 		struct nouveau_bo *bo = nouveau_bo(vb->buffer);
-		uint32_t hw = nv50_vbo_vtxelt_to_hw(ve);
+		uint32_t hw = nv50->vtxelt->hw[i];
 
 		if (!vb->stride &&
 		    nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) {
@@ -821,13 +586,13 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		}
 
 		if (nv50->vbo_fifo) {
-			so_data  (vtxfmt, hw |
-				  (ve->instance_divisor ? (1 << 4) : i));
+			so_data  (vtxfmt, hw | (ve->instance_divisor ? (1 << 4) : i));
 			so_method(vtxbuf, tesla,
 				  NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 			so_data  (vtxbuf, 0);
 			continue;
 		}
+
 		so_data(vtxfmt, hw | i);
 
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3);
@@ -855,355 +620,13 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 		so_data  (vtxbuf, 0);
 	}
-	nv50->state.vtxelt_nr = nv50->vtxelt_nr;
+	nv50->state.vtxelt_nr = nv50->vtxelt->num_elements;
 
-	so_ref (vtxfmt, &nv50->state.vtxfmt);
 	so_ref (vtxbuf, &nv50->state.vtxbuf);
 	so_ref (vtxattr, &nv50->state.vtxattr);
 	so_ref (NULL, &vtxbuf);
-	so_ref (NULL, &vtxfmt);
 	so_ref (NULL, &vtxattr);
+	return vtxfmt;
 }
 
-typedef void (*pfn_push)(struct nouveau_channel *, void *);
-
-struct nv50_vbo_emitctx
-{
-	pfn_push push[16];
-	uint8_t *map[16];
-	unsigned stride[16];
-	unsigned nr_ve;
-	unsigned vtx_dwords;
-	unsigned vtx_max;
-
-	float edgeflag;
-	unsigned ve_edgeflag;
-};
-
-static INLINE void
-emit_vtx_next(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i) {
-		emit->push[i](chan, emit->map[i]);
-		emit->map[i] += emit->stride[i];
-	}
-}
-
-static INLINE void
-emit_vtx(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit,
-	 uint32_t vi)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i)
-		emit->push[i](chan, emit->map[i] + emit->stride[i] * vi);
-}
-
-static INLINE boolean
-nv50_map_vbufs(struct nv50_context *nv50)
-{
-	int i;
-
-	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
-		struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-		unsigned size = vb->stride * (vb->max_index + 1) + 16;
-
-		if (nouveau_bo(vb->buffer)->map)
-			continue;
-
-		size = vb->stride * (vb->max_index + 1) + 16;
-		size = MIN2(size, vb->buffer->size);
-		if (!size)
-			size = vb->buffer->size;
-
-		if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
-					 0, size, NOUVEAU_BO_RD))
-			break;
-	}
-
-	if (i == nv50->vtxbuf_nr)
-		return TRUE;
-	for (; i >= 0; --i)
-		nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-	return FALSE;
-}
-
-static void
-emit_b32_1(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b32_2(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-}
-
-static void
-emit_b32_3(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b32_4(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-	OUT_RING(chan, v[3]);
-}
-
-static void
-emit_b16_1(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b16_3(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, (v[1] << 16) | v[0]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b08_1(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b08_3(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
-}
-
-static boolean
-emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
-	     unsigned start)
-{
-	unsigned i;
-
-	if (nv50_map_vbufs(nv50) == FALSE)
-		return FALSE;
-
-	emit->ve_edgeflag = nv50->vertprog->cfg.edgeflag_in;
-
-	emit->edgeflag = 0.5f;
-	emit->nr_ve = 0;
-	emit->vtx_dwords = 0;
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		struct pipe_vertex_element *ve;
-		struct pipe_vertex_buffer *vb;
-		unsigned n, size;
-		const struct util_format_description *desc;
-
-		ve = &nv50->vtxelt[i];
-		vb = &nv50->vtxbuf[ve->vertex_buffer_index];
-		if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
-			continue;
-		n = emit->nr_ve++;
-
-		emit->stride[n] = vb->stride;
-		emit->map[n] = (uint8_t *)nouveau_bo(vb->buffer)->map +
-			vb->buffer_offset +
-			(start * vb->stride + ve->src_offset);
-
-		desc = util_format_description(ve->src_format);
-		assert(desc);
-
-		size = util_format_get_component_bits(
-			ve->src_format, UTIL_FORMAT_COLORSPACE_RGB, 0);
-
-		assert(ve->nr_components > 0 && ve->nr_components <= 4);
-
-		/* It shouldn't be necessary to push the implicit 1s
-		 * for case 3 and size 8 cases 1, 2, 3.
-		 */
-		switch (size) {
-		default:
-			NOUVEAU_ERR("unsupported vtxelt size: %u\n", size);
-			return FALSE;
-		case 32:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b32_1; break;
-			case 2: emit->push[n] = emit_b32_2; break;
-			case 3: emit->push[n] = emit_b32_3; break;
-			case 4: emit->push[n] = emit_b32_4; break;
-			}
-			emit->vtx_dwords += ve->nr_components;
-			break;
-		case 16:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b16_1; break;
-			case 2: emit->push[n] = emit_b32_1; break;
-			case 3: emit->push[n] = emit_b16_3; break;
-			case 4: emit->push[n] = emit_b32_2; break;
-			}
-			emit->vtx_dwords += (ve->nr_components + 1) >> 1;
-			break;
-		case 8:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b08_1; break;
-			case 2: emit->push[n] = emit_b16_1; break;
-			case 3: emit->push[n] = emit_b08_3; break;
-			case 4: emit->push[n] = emit_b32_1; break;
-			}
-			emit->vtx_dwords += 1;
-			break;
-		}
-	}
-
-	emit->vtx_max = 512 / emit->vtx_dwords;
-	if (emit->ve_edgeflag < 16)
-		emit->vtx_max = 1;
-
-	return TRUE;
-}
-
-static INLINE void
-set_edgeflag(struct nouveau_channel *chan,
-	     struct nouveau_grobj *tesla,
-	     struct nv50_vbo_emitctx *emit, uint32_t index)
-{
-	unsigned i = emit->ve_edgeflag;
-
-	if (i < 16) {
-		float f = *((float *)(emit->map[i] + index * emit->stride[i]));
-
-		if (emit->edgeflag != f) {
-			emit->edgeflag = f;
-
-			BEGIN_RING(chan, tesla, 0x15e4, 1);
-			OUT_RING  (chan, f ? 1 : 0);
-		}
-	}
-}
-
-static boolean
-nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
 
-	if (emit_prepare(nv50, &emit, start) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx_next(chan, &emit);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index 1f69daec819..61b54af4ddf 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -19,7 +19,8 @@ C_SOURCES = \
 	r300_state_invariant.c \
 	r300_vs.c \
 	r300_texture.c \
-	r300_tgsi_to_rc.c
+	r300_tgsi_to_rc.c \
+	r300_transfer.c
 
 LIBRARY_INCLUDES = \
 	-I$(TOP)/src/mesa/drivers/dri/r300/compiler \
diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript
index 183aa17f9b3..27b2e309932 100644
--- a/src/gallium/drivers/r300/SConscript
+++ b/src/gallium/drivers/r300/SConscript
@@ -30,6 +30,7 @@ r300 = env.ConvenienceLibrary(
         'r300_vs.c',
         'r300_texture.c',
         'r300_tgsi_to_rc.c',
+        'r300_transfer.c',
     ] + r300compiler) + r300compiler
 
 Export('r300')
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index 513cc0f5d44..b7ad6b20206 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -36,6 +36,7 @@ static void r300_blitter_save_states(struct r300_context* r300)
     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
     util_blitter_save_viewport(r300->blitter, &r300->viewport);
     util_blitter_save_clip(r300->blitter, &r300->clip);
+    util_blitter_save_vertex_elements(r300->blitter, r300->velems);
 }
 
 /* Clear currently bound buffers. */
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index 86b98a4ba52..923e1e541ff 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -60,7 +60,6 @@ static void r300_destroy_context(struct pipe_context* context)
     FREE(r300->rs_block_state.state);
     FREE(r300->scissor_state.state);
     FREE(r300->textures_state.state);
-    FREE(r300->vertex_stream_state.state);
     FREE(r300->vap_output_state.state);
     FREE(r300->viewport_state.state);
     FREE(r300->ztop_state.state);
@@ -72,11 +71,8 @@ r300_is_texture_referenced(struct pipe_context *pipe,
                            struct pipe_texture *texture,
                            unsigned face, unsigned level)
 {
-    struct pipe_buffer* buf = 0;
-
-    r300_get_texture_buffer(pipe->screen, texture, &buf, NULL);
-
-    return pipe->is_buffer_referenced(pipe, buf);
+    return pipe->is_buffer_referenced(pipe,
+                                      ((struct r300_texture *)texture)->buffer);
 }
 
 static unsigned int
@@ -86,7 +82,14 @@ r300_is_buffer_referenced(struct pipe_context *pipe,
     /* This only checks to see whether actual hardware buffers are
      * referenced. Since we use managed BOs and transfers, it's actually not
      * possible for pipe_buffers to ever reference the actual hardware, so
-     * buffers are never referenced. */
+     * buffers are never referenced. 
+     */
+
+    /* XXX: that doesn't make sense given that
+     * r300_is_texture_referenced is implemented on top of this
+     * function and hardware can certainly refer to textures
+     * directly...
+     */
     return 0;
 }
 
@@ -143,7 +146,6 @@ static void r300_setup_atoms(struct r300_context* r300)
     r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
     r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
     r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
-    r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
     r300->vap_output_state.state = CALLOC_STRUCT(r300_vap_output_state);
     r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
     r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
@@ -159,6 +161,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     if (!r300)
         return NULL;
 
+    r300screen->ctx = (struct pipe_context*)r300;
+
     r300->winsys = radeon_winsys;
 
     r300->context.winsys = (struct pipe_winsys*)radeon_winsys;
@@ -212,7 +216,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->invariant_state.dirty = TRUE;
 
     r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300);
-    r300->dirty_state = R300_NEW_KITCHEN_SINK;
     r300->dirty_hw++;
 
     r300->blitter = util_blitter_create(&r300->context);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 0d1518a05bc..985e3391126 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -124,13 +124,15 @@ struct r300_texture_format_state {
     uint32_t format2; /* R300_TX_FORMAT2: 0x4500 */
 };
 
+#define R300_MAX_TEXTURE_LEVELS 13
+
 struct r300_texture_fb_state {
     /* Colorbuffer. */
-    uint32_t colorpitch[PIPE_MAX_TEXTURE_LEVELS]; /* R300_RB3D_COLORPITCH[0-3]*/
+    uint32_t colorpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_COLORPITCH[0-3]*/
     uint32_t us_out_fmt; /* R300_US_OUT_FMT[0-3] */
 
     /* Zbuffer. */
-    uint32_t depthpitch[PIPE_MAX_TEXTURE_LEVELS]; /* R300_RB3D_DEPTHPITCH */
+    uint32_t depthpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_DEPTHPITCH */
     uint32_t zb_format; /* R300_ZB_FORMAT */
 };
 
@@ -236,16 +238,16 @@ struct r300_texture {
     struct pipe_texture tex;
 
     /* Offsets into the buffer. */
-    unsigned offset[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned offset[R300_MAX_TEXTURE_LEVELS];
 
     /* A pitch for each mip-level */
-    unsigned pitch[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned pitch[R300_MAX_TEXTURE_LEVELS];
 
     /* Size of one zslice or face based on the texture target */
-    unsigned layer_size[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned layer_size[R300_MAX_TEXTURE_LEVELS];
 
     /* Whether the mipmap level is macrotiled. */
-    enum r300_buffer_tiling mip_macrotile[PIPE_MAX_TEXTURE_LEVELS];
+    enum r300_buffer_tiling mip_macrotile[R300_MAX_TEXTURE_LEVELS];
 
     /**
      * If non-zero, override the natural texture layout with
@@ -276,6 +278,23 @@ struct r300_texture {
     enum r300_buffer_tiling microtile, macrotile;
 };
 
+struct r300_vertex_info {
+    /* Parent class */
+    struct vertex_info vinfo;
+
+    /* R300_VAP_PROG_STREAK_CNTL_[0-7] */
+    uint32_t vap_prog_stream_cntl[8];
+    /* R300_VAP_PROG_STREAK_CNTL_EXT_[0-7] */
+    uint32_t vap_prog_stream_cntl_ext[8];
+};
+
+struct r300_vertex_element_state {
+    unsigned count;
+    struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+
+    struct r300_vertex_stream_state vertex_stream;
+};
+
 extern struct pipe_viewport_state r300_viewport_identity;
 
 struct r300_context {
@@ -348,8 +367,7 @@ struct r300_context {
     int vertex_buffer_count;
     int vertex_buffer_max_index;
     /* Vertex elements for Gallium. */
-    struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-    int vertex_element_count;
+    struct r300_vertex_element_state *velems;
 
     /* Vertex info for Draw. */
     struct vertex_info vertex_info;
@@ -368,6 +386,8 @@ struct r300_context {
     boolean polygon_offset_enabled;
     /* Z buffer bit depth. */
     uint32_t zbuffer_bpp;
+    /* Whether scissor is enabled. */
+    boolean scissor_enabled;
 };
 
 /* Convenience cast wrapper. */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index addb28bded3..55e9217fd32 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -675,7 +675,7 @@ void r300_emit_scissor_state(struct r300_context* r300,
     maxx = fb->width;
     maxy = fb->height;
 
-    if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) {
+    if (r300->scissor_enabled) {
         minx = MAX2(minx, scissor->minx);
         miny = MAX2(miny, scissor->miny);
         maxx = MIN2(maxx, scissor->maxx);
@@ -757,9 +757,9 @@ void r300_emit_textures_state(struct r300_context *r300,
 void r300_emit_aos(struct r300_context* r300, unsigned offset)
 {
     struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_vertex_element *velem = r300->velems->velem;
     int i;
-    unsigned size1, size2, aos_count = r300->vertex_element_count;
+    unsigned size1, size2, aos_count = r300->velems->count;
     unsigned packet_size = (aos_count * 3 + 1) / 2;
     CS_LOCALS(r300);
 
@@ -794,6 +794,30 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
     END_CS;
 }
 
+void r300_emit_vertex_buffer(struct r300_context* r300)
+{
+    CS_LOCALS(r300);
+
+    DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, "
+            "vertex size %d\n", r300->vbo,
+            r300->vertex_info.size);
+    /* Set the pointer to our vertex buffer. The emitted values are this:
+     * PACKET3 [3D_LOAD_VBPNTR]
+     * COUNT   [1]
+     * FORMAT  [size | stride << 8]
+     * OFFSET  [offset into BO]
+     * VBPNTR  [relocated BO]
+     */
+    BEGIN_CS(7);
+    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
+    OUT_CS(1);
+    OUT_CS(r300->vertex_info.size |
+            (r300->vertex_info.size << 8));
+    OUT_CS(r300->vbo_offset);
+    OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_CS;
+}
+
 void r300_emit_vertex_stream_state(struct r300_context* r300,
                                    unsigned size, void* state)
 {
@@ -868,7 +892,7 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
     CS_LOCALS(r300);
 
     if (!r300screen->caps->has_tcl) {
-        debug_printf("r300: Implementation error: emit_vertex_shader called,"
+        debug_printf("r300: Implementation error: emit_vs_state called,"
                 " but has_tcl is FALSE!\n");
         return;
     }
@@ -907,7 +931,7 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300,
     CS_LOCALS(r300);
 
     if (!r300screen->caps->has_tcl) {
-        debug_printf("r300: Implementation error: emit_vertex_shader called,"
+        debug_printf("r300: Implementation error: emit_vs_constant_buffer called,"
         " but has_tcl is FALSE!\n");
         return;
     }
@@ -980,7 +1004,7 @@ void r300_emit_buffer_validate(struct r300_context *r300,
         (struct r300_textures_state*)r300->textures_state.state;
     struct r300_texture* tex;
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_vertex_element *velem = r300->velems->velem;
     struct pipe_buffer *pbuf;
     unsigned i;
     boolean invalid = FALSE;
@@ -1038,7 +1062,7 @@ validate:
     }
     /* ...vertex buffers for HWTCL path... */
     if (do_validate_vertex_buffers) {
-        for (i = 0; i < r300->vertex_element_count; i++) {
+        for (i = 0; i < r300->velems->count; i++) {
             pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
 
             if (!r300->winsys->add_buffer(r300->winsys, pbuf,
@@ -1135,8 +1159,10 @@ void r300_emit_dirty_state(struct r300_context* r300)
     assert(r300->dirty_state == 0);
     */
 
-    /* Finally, emit the VBO. */
-    /* r300_emit_vertex_buffer(r300); */
+    /* Emit the VBO for SWTCL. */
+    if (!r300screen->caps->has_tcl) {
+        r300_emit_vertex_buffer(r300);
+    }
 
     r300->dirty_hw++;
 }
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index e37d3092703..70de152713d 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -61,6 +61,12 @@ static void r300_flush(struct pipe_context* pipe,
                 atom->dirty = TRUE;
             }
         }
+
+        /* Unmark HWTCL state for SWTCL. */
+        if (!r300_screen(pipe->screen)->caps->has_tcl) {
+            r300->vs_state.dirty = FALSE;
+            r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS;
+        }
     }
 
     /* reset flushed query */
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index 3c2625269b8..9e71e61c303 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -207,6 +207,7 @@ static void r300_translate_fragment_shader(
         DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n",
             compiler.Base.ErrorMsg);
         assert(0);
+        abort();
     }
 
     /* And, finally... */
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 6c891029a56..9c001ae186d 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -143,7 +143,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 {
     struct pipe_vertex_element* velem;
     struct pipe_vertex_buffer* vbuf;
-    unsigned vertex_element_count = r300->vertex_element_count;
+    unsigned vertex_element_count = r300->velems->count;
     unsigned i, v, vbi, dw, elem_offset, dwords;
 
     /* Size of the vertex, in dwords. */
@@ -166,7 +166,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        velem = &r300->vertex_element[i];
+        velem = &r300->velems->velem[i];
         offset[i] = velem->src_offset / 4;
         size[i] = util_format_get_blocksize(velem->src_format) / 4;
         vertex_size += size[i];
@@ -183,18 +183,19 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
         }
     }
 
-    dwords = 10 + count * vertex_size;
+    dwords = 9 + count * vertex_size;
 
     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
-    r300_emit_buffer_validate(r300, FALSE, 0);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
     BEGIN_CS(dwords);
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
             r300_translate_primitive(mode));
@@ -202,7 +203,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
     /* Emit vertices. */
     for (v = 0; v < count; v++) {
         for (i = 0; i < vertex_element_count; i++) {
-            velem = &r300->vertex_element[i];
+            velem = &r300->velems->velem[i];
             vbi = velem->vertex_buffer_index;
             elem_offset = offset[i] + stride[vbi] * (v + start);
 
@@ -215,7 +216,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Unmap buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        vbi = r300->vertex_element[i].vertex_buffer_index;
+        vbi = r300->velems->velem[i].vertex_buffer_index;
 
         if (map[vbi]) {
             vbuf = &r300->vertex_buffer[vbi];
@@ -238,15 +239,16 @@ static void r300_emit_draw_arrays(struct r300_context *r300,
 
     if (alt_num_verts) {
         assert(count < (1 << 24));
-        BEGIN_CS(10);
+        BEGIN_CS(9);
         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
     } else {
-        BEGIN_CS(8);
+        BEGIN_CS(7);
     }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
            r300_translate_primitive(mode) |
@@ -278,18 +280,19 @@ static void r300_emit_draw_elements(struct r300_context *r300,
     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
         count, minIndex, maxIndex);
 
-    maxIndex = MIN2(maxIndex, ((1 << 24) - 1));
+    maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
 
     if (alt_num_verts) {
-        BEGIN_CS(16);
+        BEGIN_CS(15);
         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
     } else {
-        BEGIN_CS(14);
+        BEGIN_CS(13);
     }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(maxIndex);
+    OUT_CS(minIndex);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
     if (indexSize == 4) {
         count_dwords = count;
@@ -450,7 +453,7 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
         /* Make sure there are at least 128 spare dwords in the command buffer.
          * (most of it being consumed by emit_aos) */
         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
-        r300_emit_buffer_validate(r300, TRUE, 0);
+        r300_emit_buffer_validate(r300, TRUE, NULL);
         r300_emit_dirty_state(r300);
 
         if (alt_num_verts || count <= 65535) {
@@ -468,7 +471,7 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                 /* Again, we emit both AOS and draw_arrays so there should be
                  * at least 128 spare dwords. */
                 if (count && r300_reserve_cs_space(r300, 128)) {
-                    r300_emit_buffer_validate(r300, TRUE, 0);
+                    r300_emit_buffer_validate(r300, TRUE, NULL);
                     r300_emit_dirty_state(r300);
                 }
             } while (count);
@@ -688,6 +691,7 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
     CS_LOCALS(r300);
 
     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
@@ -711,6 +715,7 @@ static void r300_render_draw(struct vbuf_render* render,
     CS_LOCALS(r300);
 
     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
     BEGIN_CS(dwords);
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index d397a8eb2b4..5880eecd5fe 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -20,14 +21,13 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
-#include "util/u_inlines.h"
 #include "util/u_format.h"
 #include "util/u_memory.h"
 #include "util/u_simple_screen.h"
 
 #include "r300_context.h"
-#include "r300_screen.h"
 #include "r300_texture.h"
+#include "r300_transfer.h"
 
 #include "radeon_winsys.h"
 #include "r300_winsys.h"
@@ -231,14 +231,16 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
     /* Check colorbuffer format support. */
     if ((usage & (PIPE_TEXTURE_USAGE_RENDER_TARGET |
                   PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                  PIPE_TEXTURE_USAGE_PRIMARY)) &&
+                  PIPE_TEXTURE_USAGE_SCANOUT |
+                  PIPE_TEXTURE_USAGE_SHARED)) &&
         /* 2101010 cannot be rendered to on non-r5xx. */
         (is_r500 || !is_color2101010) &&
         r300_is_colorbuffer_format_supported(format)) {
         retval |= usage &
             (PIPE_TEXTURE_USAGE_RENDER_TARGET |
              PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-             PIPE_TEXTURE_USAGE_PRIMARY);
+             PIPE_TEXTURE_USAGE_SCANOUT |
+             PIPE_TEXTURE_USAGE_SHARED);
     }
 
     /* Check depth-stencil format support. */
@@ -250,70 +252,6 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
     return retval == usage;
 }
 
-static struct pipe_transfer*
-r300_get_tex_transfer(struct pipe_screen *screen,
-                      struct pipe_texture *texture,
-                      unsigned face, unsigned level, unsigned zslice,
-                      enum pipe_transfer_usage usage, unsigned x, unsigned y,
-                      unsigned w, unsigned h)
-{
-    struct r300_texture *tex = (struct r300_texture *)texture;
-    struct r300_transfer *trans;
-    struct r300_screen *rscreen = r300_screen(screen);
-    unsigned offset;
-
-    offset = r300_texture_get_offset(tex, level, zslice, face);  /* in bytes */
-
-    trans = CALLOC_STRUCT(r300_transfer);
-    if (trans) {
-        pipe_texture_reference(&trans->transfer.texture, texture);
-        trans->transfer.x = x;
-        trans->transfer.y = y;
-        trans->transfer.width = w;
-        trans->transfer.height = h;
-        trans->transfer.stride = r300_texture_get_stride(rscreen, tex, level);
-        trans->transfer.usage = usage;
-        trans->transfer.zslice = zslice;
-        trans->transfer.face = face;
-
-        trans->offset = offset;
-    }
-    return &trans->transfer;
-}
-
-static void
-r300_tex_transfer_destroy(struct pipe_transfer *trans)
-{
-   pipe_texture_reference(&trans->texture, NULL);
-   FREE(trans);
-}
-
-static void* r300_transfer_map(struct pipe_screen* screen,
-                              struct pipe_transfer* transfer)
-{
-    struct r300_texture* tex = (struct r300_texture*)transfer->texture;
-    char* map;
-    enum pipe_format format = tex->tex.format;
-
-    map = pipe_buffer_map(screen, tex->buffer,
-                          pipe_transfer_buffer_flags(transfer));
-
-    if (!map) {
-        return NULL;
-    }
-
-    return map + r300_transfer(transfer)->offset +
-        transfer->y / util_format_get_blockheight(format) * transfer->stride +
-        transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
-}
-
-static void r300_transfer_unmap(struct pipe_screen* screen,
-                                struct pipe_transfer* transfer)
-{
-    struct r300_texture* tex = (struct r300_texture*)transfer->texture;
-    pipe_buffer_unmap(screen, tex->buffer);
-}
-
 static void r300_destroy_screen(struct pipe_screen* pscreen)
 {
     struct r300_screen* r300screen = r300_screen(pscreen);
@@ -350,13 +288,11 @@ struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys)
     r300screen->screen.get_paramf = r300_get_paramf;
     r300screen->screen.is_format_supported = r300_is_format_supported;
     r300screen->screen.context_create = r300_create_context;
-    r300screen->screen.get_tex_transfer = r300_get_tex_transfer;
-    r300screen->screen.tex_transfer_destroy = r300_tex_transfer_destroy;
-    r300screen->screen.transfer_map = r300_transfer_map;
-    r300screen->screen.transfer_unmap = r300_transfer_unmap;
 
     r300_init_screen_texture_functions(&r300screen->screen);
+    r300_init_screen_transfer_functions(&r300screen->screen);
     u_simple_screen_init(&r300screen->screen);
 
     return &r300screen->screen;
 }
+
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 502fbfa5a24..484bde6a6be 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -27,6 +28,8 @@
 
 #include "r300_chipset.h"
 
+#define R300_TEXTURE_USAGE_TRANSFER PIPE_TEXTURE_USAGE_CUSTOM
+
 struct radeon_winsys;
 
 struct r300_screen {
@@ -35,6 +38,10 @@ struct r300_screen {
 
     struct radeon_winsys* radeon_winsys;
 
+    /* XXX This hack will be removed once texture transfers become part of
+     * pipe_context. */
+    struct pipe_context* ctx;
+
     /* Chipset capabilities */
     struct r300_capabilities* caps;
 
@@ -42,25 +49,14 @@ struct r300_screen {
     unsigned debug;
 };
 
-struct r300_transfer {
-    /* Parent class */
-    struct pipe_transfer transfer;
-
-    /* Offset from start of buffer. */
-    unsigned offset;
-};
 
 /* Convenience cast wrapper. */
 static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) {
     return (struct r300_screen*)screen;
 }
 
-/* Convenience cast wrapper. */
-static INLINE struct r300_transfer*
-r300_transfer(struct pipe_transfer* transfer)
-{
-    return (struct r300_transfer*)transfer;
-}
+/* Creates a new r300 screen. */
+struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
 
 /* Debug functionality. */
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 1f6f99d3e52..bd4c2766cb1 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -571,6 +571,7 @@ static void
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_screen* r300screen = r300_screen(pipe->screen);
+    struct pipe_framebuffer_state *old_state = r300->fb_state.state;
     unsigned max_width, max_height;
     uint32_t zbuffer_bpp = 0;
 
@@ -595,23 +596,30 @@ static void
         return;
     }
 
-
     if (r300->draw) {
         draw_flush(r300->draw);
     }
 
-    memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
+    r300->fb_state.dirty = TRUE;
 
-    r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
-                          (state->zsbuf ? 10 : 0) + 8;
+    /* If nr_cbufs is changed from zero to non-zero or vice versa... */
+    if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
+        r300->blend_state.dirty = TRUE;
+    }
+    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
+    if (!!old_state->zsbuf != !!state->zsbuf) {
+        r300->dsa_state.dirty = TRUE;
+    }
+    if (!r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
+    }
 
     r300_fb_update_tiling_flags(r300, r300->fb_state.state, state);
 
-    /* XXX wait what */
-    r300->blend_state.dirty = TRUE;
-    r300->dsa_state.dirty = TRUE;
-    r300->fb_state.dirty = TRUE;
-    r300->scissor_state.dirty = TRUE;
+    memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
+
+    r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
+                          (state->zsbuf ? 10 : 0) + 8;
 
     /* Polygon offset depends on the zbuffer bit depth. */
     if (state->zsbuf && r300->polygon_offset_enabled) {
@@ -716,8 +724,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     rs->vap_control_status = R300_VC_32BIT_SWAP;
 #endif
 
-    /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
-     * Else, enable HW TCL and force Draw's TCL off. */
+    /* If no TCL engine is present, turn off the HW TCL. */
     if (!r300screen->caps->has_tcl) {
         rs->vap_control_status |= R300_VAP_TCL_BYPASS;
     }
@@ -807,6 +814,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_rs_state* rs = (struct r300_rs_state*)state;
+    boolean scissor_was_enabled = r300->scissor_enabled;
 
     if (r300->draw) {
         draw_flush(r300->draw);
@@ -815,20 +823,17 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 
     if (rs) {
         r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
+        r300->scissor_enabled = rs->rs.scissor;
     } else {
         r300->polygon_offset_enabled = FALSE;
+        r300->scissor_enabled = FALSE;
     }
 
     UPDATE_STATE(state, r300->rs_state);
     r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0);
 
-    /* XXX Why is this still needed, dammit!? */
-    r300->scissor_state.dirty = TRUE;
-    r300->viewport_state.dirty = TRUE;
-
-    /* XXX Clean these up when we move to atom emits */
-    if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
-        r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+    if (scissor_was_enabled != r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
     }
 }
 
@@ -973,7 +978,9 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
     memcpy(r300->scissor_state.state, state,
         sizeof(struct pipe_scissor_state));
 
-    r300->scissor_state.dirty = TRUE;
+    if (r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
+    }
 }
 
 static void r300_set_viewport_state(struct pipe_context* pipe,
@@ -1039,19 +1046,17 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
     if (r300->draw) {
         draw_flush(r300->draw);
         draw_set_vertex_buffers(r300->draw, count, buffers);
-    } else {
-        r300->vertex_stream_state.dirty = TRUE;
     }
 }
 
 static boolean r300_validate_aos(struct r300_context *r300)
 {
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_vertex_element *velem = r300->velems->velem;
     int i;
 
     /* Check if formats and strides are aligned to the size of DWORD. */
-    for (i = 0; i < r300->vertex_element_count; i++) {
+    for (i = 0; i < r300->velems->count; i++) {
         if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
             util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
             return FALSE;
@@ -1060,20 +1065,209 @@ static boolean r300_validate_aos(struct r300_context *r300)
     return TRUE;
 }
 
-static void r300_set_vertex_elements(struct pipe_context* pipe,
-                                    unsigned count,
-                                    const struct pipe_vertex_element* elements)
+static void r300_draw_emit_attrib(struct r300_context* r300,
+                                  enum attrib_emit emit,
+                                  enum interp_mode interp,
+                                  int index)
 {
-    struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct tgsi_shader_info* info = &vs->info;
+    int output;
+
+    output = draw_find_shader_output(r300->draw,
+                                     info->output_semantic_name[index],
+                                     info->output_semantic_index[index]);
+    draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output);
+}
+
+static void r300_draw_emit_all_attribs(struct r300_context* r300)
+{
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct r300_shader_semantics* vs_outputs = &vs->outputs;
+    int i, gen_count;
+
+    /* Position. */
+    if (vs_outputs->pos != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                              vs_outputs->pos);
+    } else {
+        assert(0);
+    }
+
+    /* Point size. */
+    if (vs_outputs->psize != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
+                              vs_outputs->psize);
+    }
+
+    /* Colors. */
+    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+        if (vs_outputs->color[i] != ATTR_UNUSED) {
+            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
+                                  vs_outputs->color[i]);
+        }
+    }
+
+    /* XXX Back-face colors. */
 
-    memcpy(r300->vertex_element,
-           elements,
-           sizeof(struct pipe_vertex_element) * count);
-    r300->vertex_element_count = count;
+    /* Texture coordinates. */
+    gen_count = 0;
+    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+        if (vs_outputs->generic[i] != ATTR_UNUSED) {
+            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                                  vs_outputs->generic[i]);
+            gen_count++;
+        }
+    }
+
+    /* Fog coordinates. */
+    if (vs_outputs->fog != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                              vs_outputs->fog);
+        gen_count++;
+    }
+
+    /* XXX magic */
+    assert(gen_count <= 8);
+}
+
+/* Update the PSC tables. */
+static void r300_vertex_psc(struct r300_vertex_element_state *velems)
+{
+    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
+    uint16_t type, swizzle;
+    enum pipe_format format;
+    unsigned i;
+
+    assert(velems->count <= 16);
+
+    /* Vertex shaders have no semantics on their inputs,
+     * so PSC should just route stuff based on the vertex elements,
+     * and not on attrib information. */
+    for (i = 0; i < velems->count; i++) {
+        format = velems->velem[i].src_format;
+
+        type = r300_translate_vertex_data_type(format) |
+            (i << R300_DST_VEC_LOC_SHIFT);
+        swizzle = r300_translate_vertex_data_swizzle(format);
+
+        if (i & 1) {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+        } else {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
+        }
+    }
+
+    /* Set the last vector in the PSC. */
+    if (i) {
+        i -= 1;
+    }
+    vstream->vap_prog_stream_cntl[i >> 1] |=
+        (R300_LAST_VEC << (i & 1 ? 16 : 0));
+
+    vstream->count = (i >> 1) + 1;
+}
+
+/* Update the PSC tables for SW TCL, using Draw. */
+static void r300_swtcl_vertex_psc(struct r300_context *r300,
+                                  struct r300_vertex_element_state *velems)
+{
+    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct vertex_info* vinfo = &r300->vertex_info;
+    uint16_t type, swizzle;
+    enum pipe_format format;
+    unsigned i, attrib_count;
+    int* vs_output_tab = vs->stream_loc_notcl;
+
+    /* For each Draw attribute, route it to the fragment shader according
+     * to the vs_output_tab. */
+    attrib_count = vinfo->num_attribs;
+    DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
+    for (i = 0; i < attrib_count; i++) {
+        DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
+               " vs_output_tab %d\n", vinfo->attrib[i].src_index,
+               vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
+               vs_output_tab[i]);
+    }
+
+    for (i = 0; i < attrib_count; i++) {
+        /* Make sure we have a proper destination for our attribute. */
+        assert(vs_output_tab[i] != -1);
+
+        format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
+
+        /* Obtain the type of data in this attribute. */
+        type = r300_translate_vertex_data_type(format) |
+            vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
+
+        /* Obtain the swizzle for this attribute. Note that the default
+         * swizzle in the hardware is not XYZW! */
+        swizzle = r300_translate_vertex_data_swizzle(format);
+
+        /* Add the attribute to the PSC table. */
+        if (i & 1) {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+        } else {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
+        }
+    }
+
+    /* Set the last vector in the PSC. */
+    if (i) {
+        i -= 1;
+    }
+    vstream->vap_prog_stream_cntl[i >> 1] |=
+        (R300_LAST_VEC << (i & 1 ? 16 : 0));
+
+    vstream->count = (i >> 1) + 1;
+}
+
+static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
+                                               unsigned count,
+                                               const struct pipe_vertex_element* attribs)
+{
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_screen* r300screen = r300_screen(pipe->screen);
+    struct r300_vertex_element_state *velems;
+
+    assert(count <= PIPE_MAX_ATTRIBS);
+    velems = CALLOC_STRUCT(r300_vertex_element_state);
+    if (velems != NULL) {
+        velems->count = count;
+        memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count);
+
+        if (r300screen->caps->has_tcl) {
+            r300_vertex_psc(velems);
+        } else {
+            memset(&r300->vertex_info, 0, sizeof(struct vertex_info));
+            r300_draw_emit_all_attribs(r300);
+            draw_compute_vertex_size(&r300->vertex_info);
+            r300_swtcl_vertex_psc(r300, velems);
+        }
+    }
+    return velems;
+}
+
+static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
+                                            void *state)
+{
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_vertex_element_state *velems = state;
+
+    if (velems == NULL) {
+        return;
+    }
+
+    r300->velems = velems;
 
     if (r300->draw) {
         draw_flush(r300->draw);
-        draw_set_vertex_elements(r300->draw, count, elements);
+        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
     }
 
     if (!r300_validate_aos(r300)) {
@@ -1081,6 +1275,14 @@ static void r300_set_vertex_elements(struct pipe_context* pipe,
         assert(0);
         abort();
     }
+
+    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
+    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
+}
+
+static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
+{
+   FREE(state);
 }
 
 static void* r300_create_vs_state(struct pipe_context* pipe,
@@ -1088,68 +1290,71 @@ static void* r300_create_vs_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
-        /* Copy state directly into shader. */
-        vs->state = *shader;
-        vs->state.tokens = tgsi_dup_tokens(shader->tokens);
-
-        tgsi_scan_shader(shader->tokens, &vs->info);
+    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
+    r300_vertex_shader_common_init(vs, shader);
 
-        return (void*)vs;
+    if (r300_screen(pipe->screen)->caps->has_tcl) {
+        r300_translate_vertex_shader(r300, vs);
     } else {
-        return draw_create_vertex_shader(r300->draw, shader);
+        vs->draw_vs = draw_create_vertex_shader(r300->draw, shader);
     }
+
+    return vs;
 }
 
 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
 
-    if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
+    if (vs == NULL) {
+        r300->vs_state.state = NULL;
+        return;
+    }
+    if (vs == r300->vs_state.state) {
+        return;
+    }
+    r300->vs_state.state = vs;
 
-        if (vs == NULL) {
-            r300->vs_state.state = NULL;
-            return;
-        } else if (!vs->translated) {
-            r300_translate_vertex_shader(r300, vs);
-        }
+    // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
+    if (r300->fs) {
+        r300_vertex_shader_setup_wpos(r300);
+    }
+    memcpy(r300->vap_output_state.state, &vs->vap_out,
+           sizeof(struct r300_vap_output_state));
+    r300->vap_output_state.dirty = TRUE;
 
-        UPDATE_STATE(shader, r300->vs_state);
+    /* The majority of the RS block bits is dependent on the vertex shader. */
+    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+
+    if (r300_screen(pipe->screen)->caps->has_tcl) {
+        r300->vs_state.dirty = TRUE;
         r300->vs_state.size = vs->code.length + 9;
 
-        r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
-        r300->vap_output_state.dirty = TRUE;
-        r300->vertex_stream_state.dirty = TRUE; /* XXX needed for TCL bypass */
         r300->pvs_flush.dirty = TRUE;
 
-        if (r300->fs) {
-            r300_vertex_shader_setup_wpos(r300);
-        }
-
         r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
     } else {
         draw_flush(r300->draw);
         draw_bind_vertex_shader(r300->draw,
-                (struct draw_vertex_shader*)shader);
+                (struct draw_vertex_shader*)vs->draw_vs);
     }
 }
 
 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
 
     if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
-
         rc_constants_destroy(&vs->code.constants);
-        FREE((void*)vs->state.tokens);
-        FREE(shader);
     } else {
         draw_delete_vertex_shader(r300->draw,
-                (struct draw_vertex_shader*)shader);
+                (struct draw_vertex_shader*)vs->draw_vs);
     }
+
+    FREE((void*)vs->state.tokens);
+    FREE(shader);
 }
 
 static void r300_set_constant_buffer(struct pipe_context *pipe,
@@ -1201,8 +1406,10 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
     pipe_buffer_unmap(pipe->screen, buf);
 
     if (shader == PIPE_SHADER_VERTEX) {
-        r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
-        r300->pvs_flush.dirty = TRUE;
+        if (r300screen->caps->has_tcl) {
+            r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
+            r300->pvs_flush.dirty = TRUE;
+        }
     }
     else if (shader == PIPE_SHADER_FRAGMENT)
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
@@ -1250,7 +1457,10 @@ void r300_init_state_functions(struct r300_context* r300)
     r300->context.set_viewport_state = r300_set_viewport_state;
 
     r300->context.set_vertex_buffers = r300_set_vertex_buffers;
-    r300->context.set_vertex_elements = r300_set_vertex_elements;
+
+    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
+    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
+    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
 
     r300->context.create_vs_state = r300_create_vs_state;
     r300->context.bind_vs_state = r300_bind_vs_state;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 9c8e907fdf7..6b9f61acd7b 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -37,187 +37,6 @@
 /* r300_state_derived: Various bits of state which are dependent upon
  * currently bound CSO data. */
 
-static void r300_draw_emit_attrib(struct r300_context* r300,
-                                  enum attrib_emit emit,
-                                  enum interp_mode interp,
-                                  int index)
-{
-    struct r300_vertex_shader* vs = r300->vs_state.state;
-    struct tgsi_shader_info* info = &vs->info;
-    int output;
-
-    output = draw_find_shader_output(r300->draw,
-                                     info->output_semantic_name[index],
-                                     info->output_semantic_index[index]);
-    draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output);
-}
-
-static void r300_draw_emit_all_attribs(struct r300_context* r300)
-{
-    struct r300_vertex_shader* vs = r300->vs_state.state;
-    struct r300_shader_semantics* vs_outputs = &vs->outputs;
-    int i, gen_count;
-
-    /* Position. */
-    if (vs_outputs->pos != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->pos);
-    } else {
-        assert(0);
-    }
-
-    /* Point size. */
-    if (vs_outputs->psize != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
-                              vs_outputs->psize);
-    }
-
-    /* Colors. */
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED) {
-            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
-                                  vs_outputs->color[i]);
-        }
-    }
-
-    /* XXX Back-face colors. */
-
-    /* Texture coordinates. */
-    gen_count = 0;
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        if (vs_outputs->generic[i] != ATTR_UNUSED) {
-            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                                  vs_outputs->generic[i]);
-            gen_count++;
-        }
-    }
-
-    /* Fog coordinates. */
-    if (vs_outputs->fog != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->fog);
-        gen_count++;
-    }
-
-    /* XXX magic */
-    assert(gen_count <= 8);
-}
-
-/* Update the PSC tables. */
-/* XXX move this function into r300_state.c after TCL-bypass gets removed
- * XXX because this one is dependent only on vertex elements. */
-static void r300_vertex_psc(struct r300_context* r300)
-{
-    struct r300_vertex_shader* vs = r300->vs_state.state;
-    struct r300_vertex_stream_state *vformat =
-        (struct r300_vertex_stream_state*)r300->vertex_stream_state.state;
-    uint16_t type, swizzle;
-    enum pipe_format format;
-    unsigned i;
-    int identity[16] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
-    int* stream_tab;
-
-    memset(vformat, 0, sizeof(struct r300_vertex_stream_state));
-
-    stream_tab = identity;
-
-    /* Vertex shaders have no semantics on their inputs,
-     * so PSC should just route stuff based on the vertex elements,
-     * and not on attrib information. */
-    DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements"
-            " in psc\n",
-            vs->info.num_inputs,
-            r300->vertex_element_count);
-
-    for (i = 0; i < r300->vertex_element_count; i++) {
-        format = r300->vertex_element[i].src_format;
-
-        type = r300_translate_vertex_data_type(format) |
-            (stream_tab[i] << R300_DST_VEC_LOC_SHIFT);
-        swizzle = r300_translate_vertex_data_swizzle(format);
-
-        if (i & 1) {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
-        } else {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
-        }
-    }
-
-    assert(i <= 15);
-
-    /* Set the last vector in the PSC. */
-    if (i) {
-        i -= 1;
-    }
-    vformat->vap_prog_stream_cntl[i >> 1] |=
-        (R300_LAST_VEC << (i & 1 ? 16 : 0));
-
-    vformat->count = (i >> 1) + 1;
-    r300->vertex_stream_state.size = (1 + vformat->count) * 2;
-}
-
-/* Update the PSC tables for SW TCL, using Draw. */
-static void r300_swtcl_vertex_psc(struct r300_context* r300)
-{
-    struct r300_vertex_shader* vs = r300->vs_state.state;
-    struct r300_vertex_stream_state *vformat =
-        (struct r300_vertex_stream_state*)r300->vertex_stream_state.state;
-    struct vertex_info* vinfo = &r300->vertex_info;
-    uint16_t type, swizzle;
-    enum pipe_format format;
-    unsigned i, attrib_count;
-    int* vs_output_tab = vs->stream_loc_notcl;
-
-    memset(vformat, 0, sizeof(struct r300_vertex_stream_state));
-
-    /* For each Draw attribute, route it to the fragment shader according
-     * to the vs_output_tab. */
-    attrib_count = vinfo->num_attribs;
-    DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
-    for (i = 0; i < attrib_count; i++) {
-        DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
-               " vs_output_tab %d\n", vinfo->attrib[i].src_index,
-               vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
-               vs_output_tab[i]);
-    }
-
-    for (i = 0; i < attrib_count; i++) {
-        /* Make sure we have a proper destination for our attribute. */
-        assert(vs_output_tab[i] != -1);
-
-        format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
-
-        /* Obtain the type of data in this attribute. */
-        type = r300_translate_vertex_data_type(format) |
-            vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
-
-        /* Obtain the swizzle for this attribute. Note that the default
-         * swizzle in the hardware is not XYZW! */
-        swizzle = r300_translate_vertex_data_swizzle(format);
-
-        /* Add the attribute to the PSC table. */
-        if (i & 1) {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
-        } else {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
-        }
-    }
-
-    /* Set the last vector in the PSC. */
-    if (i) {
-        i -= 1;
-    }
-    vformat->vap_prog_stream_cntl[i >> 1] |=
-        (R300_LAST_VEC << (i & 1 ? 16 : 0));
-
-    vformat->count = (i >> 1) + 1;
-    r300->vertex_stream_state.size = (1 + vformat->count) * 2;
-}
-
 static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
                         boolean swizzle_0001)
 {
@@ -432,23 +251,8 @@ static void r300_update_rs_block(struct r300_context* r300,
 static void r300_update_derived_shader_state(struct r300_context* r300)
 {
     struct r300_vertex_shader* vs = r300->vs_state.state;
-    struct r300_screen* r300screen = r300_screen(r300->context.screen);
-    struct r300_vap_output_state *vap_out =
-        (struct r300_vap_output_state*)r300->vap_output_state.state;
-
-    /* XXX Mmm, delicious hax */
-    memset(&r300->vertex_info, 0, sizeof(struct vertex_info));
-    memcpy(vap_out, vs->hwfmt, sizeof(uint)*4);
 
     r300_update_rs_block(r300, &vs->outputs, &r300->fs->inputs);
-
-    if (r300screen->caps->has_tcl) {
-        r300_vertex_psc(r300);
-    } else {
-        r300_draw_emit_all_attribs(r300);
-        draw_compute_vertex_size(&r300->vertex_info);
-        r300_swtcl_vertex_psc(r300);
-    }
 }
 
 static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa)
@@ -581,9 +385,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
 
 void r300_update_derived_state(struct r300_context* r300)
 {
-    if (r300->rs_block_state.dirty ||
-        r300->vertex_stream_state.dirty || /* XXX put updating this state out of this file */
-        r300->rs_state.dirty) {  /* XXX and remove this one (tcl_bypass dependency) */
+    if (r300->rs_block_state.dirty) {
         r300_update_derived_shader_state(r300);
     }
 
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index 2f3a56e1fbc..a32924ed0a3 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -348,39 +348,12 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count)
     return 0;
 }
 
-/* Utility function to count the number of components in RGBAZS formats.
- * XXX should go to util or p_format.h */
-static INLINE unsigned pf_component_count(enum pipe_format format) {
-    unsigned count = 0;
-
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 1)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 2)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 3)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 0)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) {
-        count++;
-    }
-
-    return count;
-}
-
 /* Translate pipe_formats into PSC vertex types. */
 static INLINE uint16_t
 r300_translate_vertex_data_type(enum pipe_format format) {
     uint32_t result = 0;
     const struct util_format_description *desc;
-    unsigned components = pf_component_count(format);
+    unsigned components = util_format_get_nr_components(format);
 
     desc = util_format_description(format);
 
@@ -453,7 +426,6 @@ r300_translate_vertex_data_type(enum pipe_format format) {
 static INLINE uint16_t
 r300_translate_vertex_data_swizzle(enum pipe_format format) {
     const struct util_format_description *desc = util_format_description(format);
-    unsigned swizzle[4], i;
 
     assert(format);
 
@@ -463,25 +435,10 @@ r300_translate_vertex_data_swizzle(enum pipe_format format) {
         return 0;
     }
 
-    /* Swizzles for 8bits formats are in the reversed order, not sure why. */
-    if (desc->channel[0].size == 8) {
-        for (i = 0; i < 4; i++) {
-            if (desc->swizzle[i] <= 3) {
-                swizzle[i] = 3 - desc->swizzle[i];
-            } else {
-                swizzle[i] = desc->swizzle[i];
-            }
-        }
-    } else {
-        for (i = 0; i < 4; i++) {
-            swizzle[i] = desc->swizzle[i];
-        }
-    }
-
-    return ((swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
-            (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
-            (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
-            (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
+    return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
+            (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
+            (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
+            (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
             (0xf << R300_WRITE_ENA_SHIFT));
 }
 
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index c0144f64b4a..04124afd4dc 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -617,18 +617,23 @@ static unsigned r300_texture_get_tile_size(struct r300_texture* tex,
 /* Return true if macrotiling should be enabled on the miplevel. */
 static boolean r300_texture_macro_switch(struct r300_texture *tex,
                                          unsigned level,
-                                         boolean rv350_mode)
+                                         boolean rv350_mode,
+                                         int dim)
 {
-    unsigned tile_width, width;
+    unsigned tile, texdim;
 
-    tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH, TRUE);
-    width = u_minify(tex->tex.width0, level);
+    tile = r300_texture_get_tile_size(tex, dim, TRUE);
+    if (dim == TILE_WIDTH) {
+        texdim = u_minify(tex->tex.width0, level);
+    } else {
+        texdim = u_minify(tex->tex.height0, level);
+    }
 
     /* See TX_FILTER1_n.MACRO_SWITCH. */
     if (rv350_mode) {
-        return width >= tile_width;
+        return texdim >= tile;
     } else {
-        return width > tile_width;
+        return texdim > tile;
     }
 }
 
@@ -692,9 +697,10 @@ static void r300_setup_miptree(struct r300_screen* screen,
 
     for (i = 0; i <= base->last_level; i++) {
         /* Let's see if this miplevel can be macrotiled. */
-        tex->mip_macrotile[i] = (tex->macrotile == R300_BUFFER_TILED &&
-                                 r300_texture_macro_switch(tex, i, rv350_mode)) ?
-                                 R300_BUFFER_TILED : R300_BUFFER_LINEAR;
+        tex->mip_macrotile[i] =
+            (tex->macrotile == R300_BUFFER_TILED &&
+             r300_texture_macro_switch(tex, i, rv350_mode, TILE_WIDTH)) ?
+             R300_BUFFER_TILED : R300_BUFFER_LINEAR;
 
         stride = r300_texture_get_stride(screen, tex, i);
         nblocksy = r300_texture_get_nblocksy(tex, i);
@@ -724,10 +730,46 @@ static void r300_setup_flags(struct r300_texture* tex)
                    !util_is_power_of_two(tex->tex.height0);
 }
 
+static void r300_setup_tiling(struct pipe_screen *screen,
+                              struct r300_texture *tex)
+{
+    enum pipe_format format = tex->tex.format;
+    boolean rv350_mode = r300_screen(screen)->caps->family >= CHIP_FAMILY_RV350;
+
+    if (util_format_is_compressed(format)) {
+        return;
+    }
+
+    if (tex->tex.width0 == 1 ||
+        tex->tex.height0 == 1) {
+        return;
+    }
+
+    /* Set microtiling. */
+    switch (util_format_get_blocksize(format)) {
+        case 1:
+        case 4:
+            tex->microtile = R300_BUFFER_TILED;
+            break;
+
+        /* XXX Square-tiling doesn't work with kernel older than 2.6.34,
+         * XXX need to check the DRM version */
+        /*case 2:
+        case 8:
+            tex->microtile = R300_BUFFER_SQUARETILED;
+            break;*/
+    }
+
+    /* Set macrotiling. */
+    if (r300_texture_macro_switch(tex, 0, rv350_mode, TILE_WIDTH) &&
+        r300_texture_macro_switch(tex, 0, rv350_mode, TILE_HEIGHT)) {
+        tex->macrotile = R300_BUFFER_TILED;
+    }
+}
+
 /* Create a new texture. */
-static struct pipe_texture*
-    r300_texture_create(struct pipe_screen* screen,
-                        const struct pipe_texture* template)
+static struct pipe_texture* r300_texture_create(struct pipe_screen* screen,
+                                         const struct pipe_texture* template)
 {
     struct r300_texture* tex = CALLOC_STRUCT(r300_texture);
     struct r300_screen* rscreen = r300_screen(screen);
@@ -742,6 +784,9 @@ static struct pipe_texture*
     tex->tex.screen = screen;
 
     r300_setup_flags(tex);
+    if (!(template->tex_usage & R300_TEXTURE_USAGE_TRANSFER)) {
+        r300_setup_tiling(screen, tex);
+    }
     r300_setup_miptree(rscreen, tex);
     r300_setup_texture_state(rscreen, tex);
 
@@ -806,14 +851,17 @@ static void r300_tex_surface_destroy(struct pipe_surface* s)
     FREE(s);
 }
 
+
 static struct pipe_texture*
-    r300_texture_blanket(struct pipe_screen* screen,
-                         const struct pipe_texture* base,
-                         const unsigned* stride,
-                         struct pipe_buffer* buffer)
+    r300_texture_from_handle(struct pipe_screen* screen,
+                             const struct pipe_texture* base,
+                             struct winsys_handle *whandle)
 {
-    struct r300_texture* tex;
+    struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys;
     struct r300_screen* rscreen = r300_screen(screen);
+    struct pipe_buffer *buffer;
+    struct r300_texture* tex;
+    unsigned stride;
 
     /* Support only 2D textures without mipmaps */
     if (base->target != PIPE_TEXTURE_2D ||
@@ -822,6 +870,11 @@ static struct pipe_texture*
         return NULL;
     }
 
+    buffer = winsys->buffer_from_handle(winsys, screen, whandle, &stride);
+    if (!buffer) {
+        return NULL;
+    }
+
     tex = CALLOC_STRUCT(r300_texture);
     if (!tex) {
         return NULL;
@@ -831,17 +884,38 @@ static struct pipe_texture*
     pipe_reference_init(&tex->tex.reference, 1);
     tex->tex.screen = screen;
 
-    tex->stride_override = *stride;
-    tex->pitch[0] = *stride / util_format_get_blocksize(base->format);
+    tex->stride_override = stride;
+    tex->pitch[0] = stride / util_format_get_blocksize(base->format);
 
     r300_setup_flags(tex);
     r300_setup_texture_state(rscreen, tex);
 
-    pipe_buffer_reference(&tex->buffer, buffer);
+    /* one ref already taken */
+    tex->buffer = buffer;
 
     return (struct pipe_texture*)tex;
 }
 
+static boolean
+    r300_texture_get_handle(struct pipe_screen* screen,
+                            struct pipe_texture *texture,
+                            struct winsys_handle *whandle)
+{
+    struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys;
+    struct r300_texture* tex = (struct r300_texture*)texture;
+    unsigned stride;
+
+    if (!tex) {
+        return FALSE;
+    }
+
+    stride = r300_texture_get_stride(r300_screen(screen), tex, 0);
+
+    winsys->buffer_get_handle(winsys, tex->buffer, stride, whandle);
+
+    return TRUE;
+}
+
 static struct pipe_video_surface *
 r300_video_surface_create(struct pipe_screen *screen,
                           enum pipe_video_chroma_format chroma_format,
@@ -893,30 +967,13 @@ static void r300_video_surface_destroy(struct pipe_video_surface *vsfc)
 void r300_init_screen_texture_functions(struct pipe_screen* screen)
 {
     screen->texture_create = r300_texture_create;
+    screen->texture_from_handle = r300_texture_from_handle;
+    screen->texture_get_handle = r300_texture_get_handle;
     screen->texture_destroy = r300_texture_destroy;
     screen->get_tex_surface = r300_get_tex_surface;
     screen->tex_surface_destroy = r300_tex_surface_destroy;
-    screen->texture_blanket = r300_texture_blanket;
 
     screen->video_surface_create = r300_video_surface_create;
     screen->video_surface_destroy= r300_video_surface_destroy;
 }
 
-boolean r300_get_texture_buffer(struct pipe_screen* screen,
-                                struct pipe_texture* texture,
-                                struct pipe_buffer** buffer,
-                                unsigned* stride)
-{
-    struct r300_texture* tex = (struct r300_texture*)texture;
-    if (!tex) {
-        return FALSE;
-    }
-
-    pipe_buffer_reference(buffer, tex->buffer);
-
-    if (stride) {
-        *stride = r300_texture_get_stride(r300_screen(screen), tex, 0);
-    }
-
-    return TRUE;
-}
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 46a5fb6188b..138b62784e6 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -60,13 +60,11 @@ r300_video_surface(struct pipe_video_surface *pvs)
     return (struct r300_video_surface *)pvs;
 }
 
-#ifndef R300_WINSYS_H
-
+/* Used internally for texture_is_referenced()
+ */
 boolean r300_get_texture_buffer(struct pipe_screen* screen,
                                 struct pipe_texture* texture,
                                 struct pipe_buffer** buffer,
                                 unsigned* stride);
 
-#endif /* R300_WINSYS_H */
-
 #endif /* R300_TEXTURE_H */
diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c
new file mode 100644
index 00000000000..ec89681a3c0
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_transfer.c
@@ -0,0 +1,270 @@
+/*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
+ *
+ * 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
+ * THE AUTHOR(S) 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. */
+
+#include "r300_context.h"
+#include "r300_transfer.h"
+#include "r300_texture.h"
+#include "r300_screen.h"
+
+#include "util/u_memory.h"
+#include "util/u_format.h"
+
+struct r300_transfer {
+    /* Parent class */
+    struct pipe_transfer transfer;
+
+    /* Pipe context. */
+    struct pipe_context *ctx;
+
+    /* Parameters of get_tex_transfer. */
+    unsigned x, y, level, zslice, face;
+
+    /* Offset from start of buffer. */
+    unsigned offset;
+
+    /* Detiled texture. */
+    struct r300_texture *detiled_texture;
+
+    /* Transfer and format flags. */
+    unsigned buffer_usage, render_target_usage;
+};
+
+/* Convenience cast wrapper. */
+static INLINE struct r300_transfer*
+r300_transfer(struct pipe_transfer* transfer)
+{
+    return (struct r300_transfer*)transfer;
+}
+
+/* Copy from a tiled texture to a detiled one. */
+static void r300_copy_from_tiled_texture(struct pipe_context *ctx,
+                                         struct r300_transfer *r300transfer)
+{
+    struct pipe_screen *screen = ctx->screen;
+    struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
+    struct pipe_texture *tex = transfer->texture;
+    struct pipe_surface *src, *dst;
+
+    src = screen->get_tex_surface(screen, tex, r300transfer->face,
+                                  r300transfer->level, r300transfer->zslice,
+                                  PIPE_BUFFER_USAGE_GPU_READ |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    dst = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex,
+                                  0, 0, 0,
+                                  PIPE_BUFFER_USAGE_GPU_WRITE |
+                                  PIPE_BUFFER_USAGE_PIXEL |
+                                  r300transfer->buffer_usage);
+
+    ctx->surface_copy(ctx, dst, 0, 0, src, r300transfer->x, r300transfer->y,
+                      transfer->width, transfer->height);
+
+    pipe_surface_reference(&src, NULL);
+    pipe_surface_reference(&dst, NULL);
+}
+
+/* Copy a detiled texture to a tiled one. */
+static void r300_copy_into_tiled_texture(struct pipe_context *ctx,
+                                         struct r300_transfer *r300transfer)
+{
+    struct pipe_screen *screen = ctx->screen;
+    struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
+    struct pipe_texture *tex = transfer->texture;
+    struct pipe_surface *src, *dst;
+
+    src = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex,
+                                  0, 0, 0,
+                                  PIPE_BUFFER_USAGE_GPU_READ |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    dst = screen->get_tex_surface(screen, tex, r300transfer->face,
+                                  r300transfer->level, r300transfer->zslice,
+                                  PIPE_BUFFER_USAGE_GPU_WRITE |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    /* XXX this flush prevents the following DRM error from occuring:
+     * [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation !
+     * Reproducible with perf/copytex. */
+    ctx->flush(ctx, 0, NULL);
+
+    ctx->surface_copy(ctx, dst, r300transfer->x, r300transfer->y, src, 0, 0,
+                      transfer->width, transfer->height);
+
+    /* XXX this flush fixes a few piglit tests (e.g. glean/pixelFormats). */
+    ctx->flush(ctx, 0, NULL);
+
+    pipe_surface_reference(&src, NULL);
+    pipe_surface_reference(&dst, NULL);
+}
+
+static struct pipe_transfer*
+r300_get_tex_transfer(struct pipe_screen *screen,
+                      struct pipe_texture *texture,
+                      unsigned face, unsigned level, unsigned zslice,
+                      enum pipe_transfer_usage usage, unsigned x, unsigned y,
+                      unsigned w, unsigned h)
+{
+    struct r300_texture *tex = (struct r300_texture *)texture;
+    struct r300_transfer *trans;
+    struct r300_screen *r300screen = r300_screen(screen);
+    struct pipe_texture template;
+
+    trans = CALLOC_STRUCT(r300_transfer);
+    if (trans) {
+        /* Initialize the transfer object. */
+        pipe_texture_reference(&trans->transfer.texture, texture);
+        trans->transfer.usage = usage;
+        trans->transfer.width = w;
+        trans->transfer.height = h;
+        trans->ctx = r300screen->ctx;
+        trans->x = x;
+        trans->y = y;
+        trans->level = level;
+        trans->zslice = zslice;
+        trans->face = face;
+
+        /* If the texture is tiled, we must create a temporary detiled texture
+         * for this transfer. */
+        if (tex->microtile || tex->macrotile) {
+            trans->buffer_usage = pipe_transfer_buffer_flags(&trans->transfer);
+            trans->render_target_usage =
+                util_format_is_depth_or_stencil(texture->format) ?
+                PIPE_TEXTURE_USAGE_DEPTH_STENCIL :
+                PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+            template.target = PIPE_TEXTURE_2D;
+            template.format = texture->format;
+            template.width0 = w;
+            template.height0 = h;
+            template.depth0 = 0;
+            template.last_level = 0;
+            template.nr_samples = 0;
+            template.tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+                                 R300_TEXTURE_USAGE_TRANSFER;
+
+            /* For texture reading, the temporary (detiled) texture is used as
+             * a render target when blitting from a tiled texture. */
+            if (usage & PIPE_TRANSFER_READ) {
+                template.tex_usage |= trans->render_target_usage;
+            }
+            /* For texture writing, the temporary texture is used as a sampler
+             * when blitting into a tiled texture. */
+            if (usage & PIPE_TRANSFER_WRITE) {
+                template.tex_usage |= PIPE_TEXTURE_USAGE_SAMPLER;
+            }
+
+            /* Create the temporary texture. */
+            trans->detiled_texture =
+                (struct r300_texture*)screen->texture_create(screen, &template);
+            assert(!trans->detiled_texture->microtile &&
+                   !trans->detiled_texture->macrotile);
+
+            /* Set the stride.
+             * Parameters x, y, level, zslice, and face remain zero. */
+            trans->transfer.stride =
+                r300_texture_get_stride(r300screen, trans->detiled_texture, 0);
+
+            if (usage & PIPE_TRANSFER_READ) {
+                /* We cannot map a tiled texture directly because the data is
+                 * in a different order, therefore we do detiling using a blit. */
+                r300_copy_from_tiled_texture(r300screen->ctx, trans);
+            }
+        } else {
+            trans->transfer.x = x;
+            trans->transfer.y = y;
+            trans->transfer.stride =
+                r300_texture_get_stride(r300screen, tex, level);
+            trans->transfer.level = level;
+            trans->transfer.zslice = zslice;
+            trans->transfer.face = face;
+            trans->offset = r300_texture_get_offset(tex, level, zslice, face);
+        }
+    }
+    return &trans->transfer;
+}
+
+static void r300_tex_transfer_destroy(struct pipe_transfer *trans)
+{
+    struct r300_transfer *r300transfer = r300_transfer(trans);
+
+    if (r300transfer->detiled_texture) {
+        if (trans->usage & PIPE_TRANSFER_WRITE) {
+            r300_copy_into_tiled_texture(r300transfer->ctx, r300transfer);
+        }
+
+        pipe_texture_reference(
+            (struct pipe_texture**)&r300transfer->detiled_texture, NULL);
+    }
+    pipe_texture_reference(&trans->texture, NULL);
+    FREE(trans);
+}
+
+static void* r300_transfer_map(struct pipe_screen *screen,
+                               struct pipe_transfer *transfer)
+{
+    struct r300_transfer *r300transfer = r300_transfer(transfer);
+    struct r300_texture *tex = (struct r300_texture*)transfer->texture;
+    char *map;
+    enum pipe_format format = tex->tex.format;
+
+    if (r300transfer->detiled_texture) {
+        /* The detiled texture is of the same size as the region being mapped
+         * (no offset needed). */
+        return pipe_buffer_map(screen,
+                               r300transfer->detiled_texture->buffer,
+                               pipe_transfer_buffer_flags(transfer));
+    } else {
+        /* Tiling is disabled. */
+        map = pipe_buffer_map(screen, tex->buffer,
+                              pipe_transfer_buffer_flags(transfer));
+
+        if (!map) {
+            return NULL;
+        }
+
+        return map + r300_transfer(transfer)->offset +
+            transfer->y / util_format_get_blockheight(format) * transfer->stride +
+            transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+    }
+}
+
+static void r300_transfer_unmap(struct pipe_screen *screen,
+                                struct pipe_transfer *transfer)
+{
+    struct r300_transfer *r300transfer = r300_transfer(transfer);
+    struct r300_texture *tex = (struct r300_texture*)transfer->texture;
+
+    if (r300transfer->detiled_texture) {
+        pipe_buffer_unmap(screen, r300transfer->detiled_texture->buffer);
+    } else {
+        pipe_buffer_unmap(screen, tex->buffer);
+    }
+}
+
+void r300_init_screen_transfer_functions(struct pipe_screen *screen)
+{
+    screen->get_tex_transfer = r300_get_tex_transfer;
+    screen->tex_transfer_destroy = r300_tex_transfer_destroy;
+    screen->transfer_map = r300_transfer_map;
+    screen->transfer_unmap = r300_transfer_unmap;
+}
diff --git a/src/gallium/drivers/r300/r300_transfer.h b/src/gallium/drivers/r300/r300_transfer.h
new file mode 100644
index 00000000000..60d1d3dc85c
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_transfer.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
+ *
+ * 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
+ * THE AUTHOR(S) 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. */
+
+#ifndef R300_TRANSFER
+#define R300_TRANSFER
+
+#include "pipe/p_screen.h"
+
+void r300_init_screen_transfer_functions(struct pipe_screen *screen);
+
+#endif
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 60a04bbfeda..379939ac750 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -89,95 +89,41 @@ static void r300_shader_read_vs_outputs(
                 assert(0);
         }
     }
+
+    /* WPOS is a straight copy of POSITION and it's always emitted. */
+    vs_outputs->wpos = i;
 }
 
-static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs)
+/* This function sets up:
+ * - VAP mapping, which maps VS registers to output semantics and
+ *   at the same time it indicates which attributes are enabled and should
+ *   be rasterized.
+ * - Stream mapping to VS outputs if TCL is not present. */
+static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs)
 {
     struct r300_shader_semantics* vs_outputs = &vs->outputs;
-    uint32_t* hwfmt = vs->hwfmt;
-    int i, gen_count;
+    struct r300_vap_output_state *vap_out = &vs->vap_out;
+    int *stream_loc = vs->stream_loc_notcl;
+    int i, gen_count, tabi = 0;
     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
                               vs_outputs->bcolor[1] != ATTR_UNUSED;
 
-    /* Do the actual vertex_info setup.
-     *
-     * vertex_info has four uints of hardware-specific data in it.
-     * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
-     * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
-     * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
-     * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
-
-    hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
+    vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */
 
     /* Position. */
     if (vs_outputs->pos != ATTR_UNUSED) {
-        hwfmt[1] |= R300_INPUT_CNTL_POS;
-        hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+        vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
+        vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+
+        stream_loc[tabi++] = 0;
     } else {
         assert(0);
     }
 
     /* Point size. */
     if (vs_outputs->psize != ATTR_UNUSED) {
-        hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
-    }
-
-    /* Colors. */
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
-            vs_outputs->color[1] != ATTR_UNUSED) {
-            hwfmt[1] |= R300_INPUT_CNTL_COLOR;
-            hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
-        }
-    }
-
-    /* Back-face colors. */
-    if (any_bcolor_used) {
-        for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-            hwfmt[1] |= R300_INPUT_CNTL_COLOR;
-            hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
-        }
-    }
+        vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 
-    /* Texture coordinates. */
-    gen_count = 0;
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        if (vs_outputs->generic[i] != ATTR_UNUSED) {
-            hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
-            hwfmt[3] |= (4 << (3 * gen_count));
-            gen_count++;
-        }
-    }
-
-    /* Fog coordinates. */
-    if (vs_outputs->fog != ATTR_UNUSED) {
-        hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
-        hwfmt[3] |= (4 << (3 * gen_count));
-        gen_count++;
-    }
-
-    /* XXX magic */
-    assert(gen_count <= 8);
-
-    /* WPOS. */
-    vs->wpos_tex_output = gen_count;
-}
-
-/* Sets up stream mapping to equivalent VS outputs if TCL is bypassed
- * or isn't present. */
-static void r300_stream_locations_notcl(
-    struct r300_shader_semantics* vs_outputs,
-    int* stream_loc)
-{
-    int i, tabi = 0, gen_count;
-    boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
-                              vs_outputs->bcolor[1] != ATTR_UNUSED;
-
-    /* Position. */
-    stream_loc[tabi++] = 0;
-
-    /* Point size. */
-    if (vs_outputs->psize != ATTR_UNUSED) {
         stream_loc[tabi++] = 1;
     }
 
@@ -185,6 +131,9 @@ static void r300_stream_locations_notcl(
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
         if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
             vs_outputs->color[1] != ATTR_UNUSED) {
+            vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+            vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
+
             stream_loc[tabi++] = 2 + i;
         }
     }
@@ -192,6 +141,9 @@ static void r300_stream_locations_notcl(
     /* Back-face colors. */
     if (any_bcolor_used) {
         for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+            vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+            vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
+
             stream_loc[tabi++] = 4 + i;
         }
     }
@@ -200,6 +152,9 @@ static void r300_stream_locations_notcl(
     gen_count = 0;
     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
         if (vs_outputs->generic[i] != ATTR_UNUSED) {
+            vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
+            vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
+
             assert(tabi < 16);
             stream_loc[tabi++] = 6 + gen_count;
             gen_count++;
@@ -208,17 +163,22 @@ static void r300_stream_locations_notcl(
 
     /* Fog coordinates. */
     if (vs_outputs->fog != ATTR_UNUSED) {
+        vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
+        vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
+
         assert(tabi < 16);
         stream_loc[tabi++] = 6 + gen_count;
         gen_count++;
     }
 
+    /* XXX magic */
+    assert(gen_count <= 8);
+
     /* WPOS. */
-    if (vs_outputs->wpos != ATTR_UNUSED) {
-        assert(tabi < 16);
-        stream_loc[tabi++] = 6 + gen_count;
-        gen_count++;
-    }
+    vs->wpos_tex_output = gen_count;
+
+    assert(tabi < 16);
+    stream_loc[tabi++] = 6 + gen_count;
 
     for (; tabi < 16;) {
         stream_loc[tabi++] = -1;
@@ -294,26 +254,16 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
     }
 }
 
-static void r300_insert_wpos(struct r300_vertex_program_compiler* c,
-                             struct r300_shader_semantics* outputs)
+void r300_vertex_shader_common_init(struct r300_vertex_shader *vs,
+                                    const struct pipe_shader_state *shader)
 {
-    int i, lastOutput = 0;
-
-    /* Find the max output index. */
-    lastOutput = MAX2(lastOutput, outputs->psize);
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        lastOutput = MAX2(lastOutput, outputs->color[i]);
-        lastOutput = MAX2(lastOutput, outputs->bcolor[i]);
-    }
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        lastOutput = MAX2(lastOutput, outputs->generic[i]);
-    }
-    lastOutput = MAX2(lastOutput, outputs->fog);
+    /* Copy state directly into shader. */
+    vs->state = *shader;
+    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
+    tgsi_scan_shader(shader->tokens, &vs->info);
 
-    /* Set WPOS after the last output. */
-    lastOutput++;
-    rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */
-    outputs->wpos = lastOutput;
+    r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
+    r300_init_vs_output_mapping(vs);
 }
 
 void r300_translate_vertex_shader(struct r300_context* r300,
@@ -322,9 +272,6 @@ void r300_translate_vertex_shader(struct r300_context* r300,
     struct r300_vertex_program_compiler compiler;
     struct tgsi_to_rc ttr;
 
-    /* Initialize. */
-    r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
-
     /* Setup the compiler */
     rc_init(&compiler.Base);
 
@@ -348,10 +295,7 @@ void r300_translate_vertex_shader(struct r300_context* r300,
     compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
 
     /* Insert the WPOS output. */
-    r300_insert_wpos(&compiler, &vs->outputs);
-
-    r300_shader_vap_output_fmt(vs);
-    r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl);
+    rc_copy_output(&compiler.Base, 0, vs->outputs.wpos);
 
     /* Invoke the compiler */
     r3xx_compile_vertex_program(&compiler);
@@ -363,30 +307,29 @@ void r300_translate_vertex_shader(struct r300_context* r300,
 
     /* And, finally... */
     rc_destroy(&compiler.Base);
-    vs->translated = TRUE;
 }
 
 boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
 {
     struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct r300_vap_output_state *vap_out = &vs->vap_out;
     int tex_output = vs->wpos_tex_output;
     uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
-    uint32_t* hwfmt = vs->hwfmt;
 
     if (r300->fs->inputs.wpos != ATTR_UNUSED) {
         /* Enable WPOS in VAP. */
-        if (!(hwfmt[1] & tex_fmt)) {
-            hwfmt[1] |= tex_fmt;
-            hwfmt[3] |= (4 << (3 * tex_output));
+        if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) {
+            vap_out->vap_vsm_vtx_assm |= tex_fmt;
+            vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output));
 
             assert(tex_output < 8);
             return TRUE;
         }
     } else {
         /* Disable WPOS in VAP. */
-        if (hwfmt[1] & tex_fmt) {
-            hwfmt[1] &= ~tex_fmt;
-            hwfmt[3] &= ~(4 << (3 * tex_output));
+        if (vap_out->vap_vsm_vtx_assm & tex_fmt) {
+            vap_out->vap_vsm_vtx_assm &= ~tex_fmt;
+            vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output));
             return TRUE;
         }
     }
diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h
index 18cfeee3cd4..f6f0b86b683 100644
--- a/src/gallium/drivers/r300/r300_vs.h
+++ b/src/gallium/drivers/r300/r300_vs.h
@@ -28,6 +28,7 @@
 #include "tgsi/tgsi_scan.h"
 #include "radeon_code.h"
 
+#include "r300_context.h"
 #include "r300_shader_semantics.h"
 
 struct r300_context;
@@ -38,7 +39,7 @@ struct r300_vertex_shader {
 
     struct tgsi_shader_info info;
     struct r300_shader_semantics outputs;
-    uint hwfmt[4];
+    struct r300_vap_output_state vap_out;
 
     /* Stream locations for SWTCL or if TCL is bypassed. */
     int stream_loc_notcl[16];
@@ -46,13 +47,17 @@ struct r300_vertex_shader {
     /* Output stream location for WPOS. */
     int wpos_tex_output;
 
-    /* Has this shader been translated yet? */
-    boolean translated;
-
+    /* HWTCL-specific.  */
     /* Machine code (if translated) */
     struct r300_vertex_program_code code;
+
+    /* SWTCL-specific. */
+    void *draw_vs;
 };
 
+void r300_vertex_shader_common_init(struct r300_vertex_shader *vs,
+                                    const struct pipe_shader_state *shader);
+
 void r300_translate_vertex_shader(struct r300_context* r300,
                                   struct r300_vertex_shader* vs);
 
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 40fb8a95ca5..ddf2b790039 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -40,11 +40,6 @@ struct radeon_winsys;
 struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
 
 
-boolean r300_get_texture_buffer(struct pipe_screen* screen,
-                                struct pipe_texture* texture,
-                                struct pipe_buffer** buffer,
-                                unsigned* stride);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index e4ac49fa85f..1c6e4ae0768 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -6,6 +6,7 @@ LIBNAME = softpipe
 C_SOURCES = \
 	sp_fs_exec.c \
 	sp_fs_sse.c \
+	sp_buffer.c \
 	sp_clear.c \
 	sp_flush.c \
 	sp_query.c \
@@ -32,7 +33,6 @@ C_SOURCES = \
 	sp_tex_tile_cache.c \
 	sp_tile_cache.c \
 	sp_surface.c \
-	sp_video_context.c \
-	sp_winsys.c
+	sp_video_context.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index 3042e556c64..27ab00b036e 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -7,6 +7,7 @@ softpipe = env.ConvenienceLibrary(
 	source = [
 		'sp_fs_exec.c',
 		'sp_fs_sse.c',
+		'sp_buffer.c',
 		'sp_clear.c',
 		'sp_context.c',
 		'sp_draw_arrays.c',
@@ -33,8 +34,7 @@ softpipe = env.ConvenienceLibrary(
 		'sp_tex_tile_cache.c',
 		'sp_texture.c',
 		'sp_tile_cache.c',
-		'sp_video_context.c',
-		'sp_winsys.c'
+		'sp_video_context.c'
 	])
 
 Export('softpipe')
diff --git a/src/gallium/drivers/softpipe/sp_buffer.c b/src/gallium/drivers/softpipe/sp_buffer.c
new file mode 100644
index 00000000000..8f390250869
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_buffer.c
@@ -0,0 +1,118 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE 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 "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "sp_screen.h"
+#include "sp_buffer.h"
+
+
+static void *
+softpipe_buffer_map(struct pipe_screen *screen,
+                    struct pipe_buffer *buf,
+                    unsigned flags)
+{
+   struct softpipe_buffer *softpipe_buf = softpipe_buffer(buf);
+   return softpipe_buf->data;
+}
+
+
+static void
+softpipe_buffer_unmap(struct pipe_screen *screen,
+                      struct pipe_buffer *buf)
+{
+}
+
+
+static void
+softpipe_buffer_destroy(struct pipe_buffer *buf)
+{
+   struct softpipe_buffer *sbuf = softpipe_buffer(buf);
+
+   if (!sbuf->userBuffer)
+      align_free(sbuf->data);
+      
+   FREE(sbuf);
+}
+
+
+static struct pipe_buffer *
+softpipe_buffer_create(struct pipe_screen *screen,
+                       unsigned alignment,
+                       unsigned usage,
+                       unsigned size)
+{
+   struct softpipe_buffer *buffer = CALLOC_STRUCT(softpipe_buffer);
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.alignment = MAX2(alignment, 16);
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   buffer->data = align_malloc(size, alignment);
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+softpipe_user_buffer_create(struct pipe_screen *screen,
+                            void *ptr,
+                            unsigned bytes)
+{
+   struct softpipe_buffer *buffer;
+
+   buffer = CALLOC_STRUCT(softpipe_buffer);
+   if(!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+void
+softpipe_init_screen_buffer_funcs(struct pipe_screen *screen)
+{
+   screen->buffer_create = softpipe_buffer_create;
+   screen->user_buffer_create = softpipe_user_buffer_create;
+   screen->buffer_map = softpipe_buffer_map;
+   screen->buffer_unmap = softpipe_buffer_unmap;
+   screen->buffer_destroy = softpipe_buffer_destroy;
+}
diff --git a/src/gallium/drivers/softpipe/sp_buffer.h b/src/gallium/drivers/softpipe/sp_buffer.h
new file mode 100644
index 00000000000..9d8e56a176b
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_buffer.h
@@ -0,0 +1,55 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE 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 SP_BUFFER_H
+#define SP_BUFFER_H
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+
+
+struct softpipe_buffer
+{
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
+};
+
+
+/** Cast wrapper */
+static INLINE struct softpipe_buffer *
+softpipe_buffer( struct pipe_buffer *buf )
+{
+   return (struct softpipe_buffer *)buf;
+}
+
+
+void
+softpipe_init_screen_buffer_funcs(struct pipe_screen *screen);
+
+
+#endif /* SP_BUFFER_H */
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index ddc35bcd629..30494719f7b 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -210,7 +210,7 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->dump_fs = debug_get_bool_option( "GALLIUM_DUMP_FS", FALSE );
    softpipe->dump_gs = debug_get_bool_option( "SOFTPIPE_DUMP_GS", FALSE );
 
-   softpipe->pipe.winsys = screen->winsys;
+   softpipe->pipe.winsys = NULL;
    softpipe->pipe.screen = screen;
    softpipe->pipe.destroy = softpipe_destroy;
    softpipe->pipe.priv = priv;
@@ -245,6 +245,10 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->pipe.bind_gs_state   = softpipe_bind_gs_state;
    softpipe->pipe.delete_gs_state = softpipe_delete_gs_state;
 
+   softpipe->pipe.create_vertex_elements_state = softpipe_create_vertex_elements_state;
+   softpipe->pipe.bind_vertex_elements_state = softpipe_bind_vertex_elements_state;
+   softpipe->pipe.delete_vertex_elements_state = softpipe_delete_vertex_elements_state;
+
    softpipe->pipe.set_blend_color = softpipe_set_blend_color;
    softpipe->pipe.set_stencil_ref = softpipe_set_stencil_ref;
    softpipe->pipe.set_clip_state = softpipe_set_clip_state;
@@ -257,7 +261,6 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->pipe.set_viewport_state = softpipe_set_viewport_state;
 
    softpipe->pipe.set_vertex_buffers = softpipe_set_vertex_buffers;
-   softpipe->pipe.set_vertex_elements = softpipe_set_vertex_elements;
 
    softpipe->pipe.draw_arrays = softpipe_draw_arrays;
    softpipe->pipe.draw_elements = softpipe_draw_elements;
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 95def72c541..9a8158e6a22 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -45,6 +45,7 @@ struct softpipe_tile_cache;
 struct softpipe_tex_tile_cache;
 struct sp_fragment_shader;
 struct sp_vertex_shader;
+struct sp_velems_state;
 
 
 struct softpipe_context {
@@ -59,6 +60,7 @@ struct softpipe_context {
    struct sp_fragment_shader *fs;
    struct sp_vertex_shader *vs;
    struct sp_geometry_shader *gs;
+   struct sp_velems_state *velems;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -72,13 +74,11 @@ struct softpipe_context {
    struct pipe_texture *vertex_textures[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
 
    unsigned num_samplers;
    unsigned num_textures;
    unsigned num_vertex_samplers;
    unsigned num_vertex_textures;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    unsigned dirty; /**< Mask of SP_NEW_x flags */
@@ -93,7 +93,7 @@ struct softpipe_context {
    ubyte *mapped_vbuffer[PIPE_MAX_ATTRIBS];
 
    /** Mapped constant buffers */
-   void *mapped_constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
+   const void *mapped_constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
 
    /** Vertex format */
    struct vertex_info vertex_info;
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index b2acc36bf7a..7b77eb239fd 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -33,90 +33,19 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_inlines.h"
 #include "util/u_prim.h"
 
 #include "sp_context.h"
 #include "sp_query.h"
 #include "sp_state.h"
+#include "sp_buffer.h"
 
 #include "draw/draw_context.h"
 
 
 
-static void
-softpipe_map_constant_buffers(struct softpipe_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-
-   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      uint j;
-
-      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
-         if (sp->constants[i][j] && sp->constants[i][j]->size) {
-            sp->mapped_constants[i][j] = ws->buffer_map(ws,
-                                                        sp->constants[i][j],
-                                                        PIPE_BUFFER_USAGE_CPU_READ);
-         }
-      }
-   }
-
-   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
-      if (sp->constants[PIPE_SHADER_VERTEX][i]) {
-         draw_set_mapped_constant_buffer(sp->draw,
-                                         PIPE_SHADER_VERTEX,
-                                         i,
-                                         sp->mapped_constants[PIPE_SHADER_VERTEX][i],
-                                         sp->constants[PIPE_SHADER_VERTEX][i]->size);
-      }
-      if (sp->constants[PIPE_SHADER_GEOMETRY][i]) {
-         draw_set_mapped_constant_buffer(sp->draw,
-                                         PIPE_SHADER_GEOMETRY,
-                                         i,
-                                         sp->mapped_constants[PIPE_SHADER_GEOMETRY][i],
-                                         sp->constants[PIPE_SHADER_GEOMETRY][i]->size);
-      }
-   }
-}
-
-
-static void
-softpipe_unmap_constant_buffers(struct softpipe_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
 
-   /* really need to flush all prims since the vert/frag shaders const buffers
-    * are going away now.
-    */
-   draw_flush(sp->draw);
-
-   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
-      draw_set_mapped_constant_buffer(sp->draw,
-                                      PIPE_SHADER_VERTEX,
-                                      i,
-                                      NULL,
-                                      0);
-      draw_set_mapped_constant_buffer(sp->draw,
-                                      PIPE_SHADER_GEOMETRY,
-                                      i,
-                                      NULL,
-                                      0);
-   }
-
-   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      uint j;
-
-      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
-         if (sp->constants[i][j] && sp->constants[i][j]->size) {
-            ws->buffer_unmap(ws, sp->constants[i][j]);
-         }
-         sp->mapped_constants[i][j] = NULL;
-      }
-   }
-}
 
 
 /**
@@ -261,25 +190,16 @@ softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
    }
 
    softpipe_map_transfers(sp);
-   softpipe_map_constant_buffers(sp);
 
    /* Map vertex buffers */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf;
-
-      buf = pipe_buffer_map(pipe->screen,
-                            sp->vertex_buffer[i].buffer,
-                            PIPE_BUFFER_USAGE_CPU_READ);
+      void *buf = softpipe_buffer(sp->vertex_buffer[i].buffer)->data;
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
 
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes;
-
-      mapped_indexes = pipe_buffer_map(pipe->screen,
-                                       indexBuffer,
-                                       PIPE_BUFFER_USAGE_CPU_READ);
+      void *mapped_indexes = softpipe_buffer(indexBuffer)->data;
       draw_set_mapped_element_buffer_range(draw,
                                            indexSize,
                                            minIndex,
@@ -300,15 +220,18 @@ softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
    /* unmap vertex/index buffers - will cause draw module to flush */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
-   /* Note: leave drawing surfaces mapped */
-   softpipe_unmap_constant_buffers(sp);
+   /*
+    * TODO: Flush only when a user vertex/index buffer is present
+    * (or even better, modify draw module to do this
+    * internally when this condition is seen?)
+    */
+   draw_flush(draw);
 
+   /* Note: leave drawing surfaces mapped */
    sp->dirty_render_cache = TRUE;
 }
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index 98c08eaffaf..6749243ab41 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -264,57 +264,29 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-2], stride),
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 4) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
 
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-         }
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride));
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-2], stride),
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 2) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-1], stride),
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-         }
       }
       break;
 
@@ -448,56 +420,28 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-2, stride),
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 4) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-1, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-         }
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-2, stride),
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 2) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-1, stride),
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-         }
       }
       break;
 
diff --git a/src/gallium/drivers/softpipe/sp_public.h b/src/gallium/drivers/softpipe/sp_public.h
new file mode 100644
index 00000000000..62d0903d87a
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_public.h
@@ -0,0 +1,10 @@
+#ifndef SP_PUBLIC_H
+#define SP_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+softpipe_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 6ec63fe698a..e0a2ef604e9 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -27,15 +27,16 @@
 
 
 #include "util/u_memory.h"
-#include "util/u_simple_screen.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
+#include "state_tracker/sw_winsys.h"
+
 #include "sp_texture.h"
-#include "sp_winsys.h"
 #include "sp_screen.h"
 #include "sp_context.h"
+#include "sp_buffer.h"
+#include "sp_public.h"
 
 
 static const char *
@@ -83,11 +84,11 @@ softpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 13; /* max 4Kx4K */
+      return SP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 9;  /* max 256x256x256 */
+      return SP_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 13; /* max 4Kx4K */
+      return SP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
@@ -145,6 +146,8 @@ softpipe_is_format_supported( struct pipe_screen *screen,
                               unsigned tex_usage, 
                               unsigned geom_flags )
 {
+   struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
    assert(target == PIPE_TEXTURE_1D ||
           target == PIPE_TEXTURE_2D ||
           target == PIPE_TEXTURE_3D ||
@@ -166,15 +169,25 @@ softpipe_is_format_supported( struct pipe_screen *screen,
    case PIPE_FORMAT_NONE:
       return FALSE;
    default:
-      return TRUE;
+      break;
    }
+
+   if(tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
+      if(!winsys->is_displaytarget_format_supported(winsys, format))
+         return FALSE;
+   }
+
+   /* XXX: this is often a lie.  Pull in logic from llvmpipe to fix.
+    */
+   return TRUE;
 }
 
 
 static void
 softpipe_destroy_screen( struct pipe_screen *screen )
 {
-   struct pipe_winsys *winsys = screen->winsys;
+   struct softpipe_screen *sp_screen = softpipe_screen(screen);
+   struct sw_winsys *winsys = sp_screen->winsys;
 
    if(winsys->destroy)
       winsys->destroy(winsys);
@@ -183,21 +196,37 @@ softpipe_destroy_screen( struct pipe_screen *screen )
 }
 
 
+/* This is often overriden by the co-state tracker.
+ */
+static void
+softpipe_flush_frontbuffer(struct pipe_screen *_screen,
+                           struct pipe_surface *surface,
+                           void *context_private)
+{
+   struct softpipe_screen *screen = softpipe_screen(_screen);
+   struct sw_winsys *winsys = screen->winsys;
+   struct softpipe_texture *texture = softpipe_texture(surface->texture);
+
+   assert(texture->dt);
+   if (texture->dt)
+      winsys->displaytarget_display(winsys, texture->dt, context_private);
+}
 
 /**
  * Create a new pipe_screen object
  * Note: we're not presently subclassing pipe_screen (no softpipe_screen).
  */
 struct pipe_screen *
-softpipe_create_screen(struct pipe_winsys *winsys)
+softpipe_create_screen(struct sw_winsys *winsys)
 {
    struct softpipe_screen *screen = CALLOC_STRUCT(softpipe_screen);
 
    if (!screen)
       return NULL;
 
-   screen->base.winsys = winsys;
+   screen->winsys = winsys;
 
+   screen->base.winsys = NULL;
    screen->base.destroy = softpipe_destroy_screen;
 
    screen->base.get_name = softpipe_get_name;
@@ -206,9 +235,10 @@ softpipe_create_screen(struct pipe_winsys *winsys)
    screen->base.get_paramf = softpipe_get_paramf;
    screen->base.is_format_supported = softpipe_is_format_supported;
    screen->base.context_create = softpipe_create_context;
+   screen->base.flush_frontbuffer = softpipe_flush_frontbuffer;
 
    softpipe_init_screen_texture_funcs(&screen->base);
-   u_simple_screen_init(&screen->base);
+   softpipe_init_screen_buffer_funcs(&screen->base);
 
    return &screen->base;
 }
diff --git a/src/gallium/drivers/softpipe/sp_screen.h b/src/gallium/drivers/softpipe/sp_screen.h
index 3d4bfd3e840..f741454c9e5 100644
--- a/src/gallium/drivers/softpipe/sp_screen.h
+++ b/src/gallium/drivers/softpipe/sp_screen.h
@@ -35,10 +35,13 @@
 #include "pipe/p_defines.h"
 
 
+struct sw_winsys;
 
 struct softpipe_screen {
    struct pipe_screen base;
 
+   struct sw_winsys *winsys;
+
    /* Increments whenever textures are modified.  Contexts can track
     * this.
     */
@@ -55,4 +58,5 @@ softpipe_screen( struct pipe_screen *pipe )
 }
 
 
+
 #endif /* SP_SCREEN_H */
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 4370bbeaee2..6b01c0f4d72 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -100,6 +100,11 @@ struct sp_geometry_shader {
    struct draw_geometry_shader *draw_data;
 };
 
+struct sp_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
 
 void *
 softpipe_create_blend_state(struct pipe_context *,
@@ -160,8 +165,14 @@ void *softpipe_create_gs_state(struct pipe_context *,
 void softpipe_bind_gs_state(struct pipe_context *, void *);
 void softpipe_delete_gs_state(struct pipe_context *, void *);
 
+void *softpipe_create_vertex_elements_state(struct pipe_context *,
+                                            unsigned count,
+                                            const struct pipe_vertex_element *);
+void softpipe_bind_vertex_elements_state(struct pipe_context *, void *);
+void softpipe_delete_vertex_elements_state(struct pipe_context *, void *);
+
 void softpipe_set_polygon_stipple( struct pipe_context *,
-				  const struct pipe_poly_stipple * );
+                                   const struct pipe_poly_stipple * );
 
 void softpipe_set_scissor_state( struct pipe_context *,
                                  const struct pipe_scissor_state * );
@@ -178,10 +189,6 @@ softpipe_set_vertex_sampler_textures(struct pipe_context *,
 void softpipe_set_viewport_state( struct pipe_context *,
                                   const struct pipe_viewport_state * );
 
-void softpipe_set_vertex_elements(struct pipe_context *,
-                                  unsigned count,
-                                  const struct pipe_vertex_element *);
-
 void softpipe_set_vertex_buffers(struct pipe_context *,
                                  unsigned count,
                                  const struct pipe_vertex_buffer *);
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index c88e2137510..2b089c28316 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -28,6 +28,7 @@
 #include "sp_context.h"
 #include "sp_state.h"
 #include "sp_fs.h"
+#include "sp_buffer.h"
 
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
@@ -163,26 +164,33 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs)
    FREE( state );
 }
 
-
-
 void
 softpipe_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             struct pipe_buffer *buf)
+                             struct pipe_buffer *constants)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   unsigned size = constants ? constants->size : 0;
+   const void *data = constants ? softpipe_buffer(constants)->data : NULL;
 
    assert(shader < PIPE_SHADER_TYPES);
-   assert(index < PIPE_MAX_CONSTANT_BUFFERS);
+   assert(index == 0);
 
    draw_flush(softpipe->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&softpipe->constants[shader][index], buf);
+   pipe_buffer_reference(&softpipe->constants[shader][index], constants);
 
+   if(shader == PIPE_SHADER_VERTEX) {
+      draw_set_mapped_constant_buffer(softpipe->draw, PIPE_SHADER_VERTEX, index,
+                                      data, size);
+   }
+
+   softpipe->mapped_constants[shader][index] = data;
    softpipe->dirty |= SP_NEW_CONSTANTS;
 }
 
+
 void *
 softpipe_create_gs_state(struct pipe_context *pipe,
                          const struct pipe_shader_state *templ)
diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c
index b491d92ed15..462f4d2655e 100644
--- a/src/gallium/drivers/softpipe/sp_state_vertex.c
+++ b/src/gallium/drivers/softpipe/sp_state_vertex.c
@@ -32,27 +32,45 @@
 #include "sp_context.h"
 #include "sp_state.h"
 
+#include "util/u_memory.h"
 #include "draw/draw_context.h"
 
 
+void *
+softpipe_create_vertex_elements_state(struct pipe_context *pipe,
+                                      unsigned count,
+                                      const struct pipe_vertex_element *attribs)
+{
+   struct sp_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct sp_velems_state *) MALLOC(sizeof(struct sp_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
 void
-softpipe_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *attribs)
+softpipe_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   struct sp_velems_state *sp_velems = (struct sp_velems_state *) velems;
 
-   assert(count <= PIPE_MAX_ATTRIBS);
-
-   memcpy(softpipe->vertex_element, attribs,
-          count * sizeof(struct pipe_vertex_element));
-   softpipe->num_vertex_elements = count;
+   softpipe->velems = sp_velems;
 
    softpipe->dirty |= SP_NEW_VERTEX;
 
-   draw_set_vertex_elements(softpipe->draw, count, attribs);
+   if (sp_velems)
+      draw_set_vertex_elements(softpipe->draw, sp_velems->count, sp_velems->velem);
 }
 
+void
+softpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 softpipe_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 32d261b5ffc..11d184effb9 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -36,11 +36,13 @@
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_simple_screen.h"
 
 #include "sp_context.h"
 #include "sp_texture.h"
 #include "sp_screen.h"
-#include "sp_winsys.h"
+
+#include "state_tracker/sw_winsys.h"
 
 
 /**
@@ -72,11 +74,9 @@ softpipe_texture_layout(struct pipe_screen *screen,
       depth = u_minify(depth, 1);
    }
 
-   spt->buffer = screen->buffer_create(screen, 32,
-                                       PIPE_BUFFER_USAGE_PIXEL,
-                                       buffer_size);
+   spt->data = align_malloc(buffer_size, 16);
 
-   return spt->buffer != NULL;
+   return spt->data != NULL;
 }
 
 
@@ -87,19 +87,18 @@ static boolean
 softpipe_displaytarget_layout(struct pipe_screen *screen,
                               struct softpipe_texture * spt)
 {
-   unsigned usage = (PIPE_BUFFER_USAGE_CPU_READ_WRITE |
-                     PIPE_BUFFER_USAGE_GPU_READ_WRITE);
-   unsigned tex_usage = spt->base.tex_usage;
-
-   spt->buffer = screen->surface_buffer_create( screen, 
-                                                spt->base.width0, 
-                                                spt->base.height0,
-                                                spt->base.format,
-                                                usage,
-                                                tex_usage,
-                                                &spt->stride[0]);
-
-   return spt->buffer != NULL;
+   struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
+   /* Round up the surface size to a multiple of the tile size?
+    */
+   spt->dt = winsys->displaytarget_create(winsys,
+                                          spt->base.format,
+                                          spt->base.width0, 
+                                          spt->base.height0,
+                                          16,
+                                          &spt->stride[0] );
+
+   return spt->dt != NULL;
 }
 
 
@@ -123,7 +122,8 @@ softpipe_texture_create(struct pipe_screen *screen,
                util_is_power_of_two(template->depth0));
 
    if (spt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                              PIPE_TEXTURE_USAGE_PRIMARY)) {
+                              PIPE_TEXTURE_USAGE_SCANOUT |
+                              PIPE_TEXTURE_USAGE_SHARED)) {
       if (!softpipe_displaytarget_layout(screen, spt))
          goto fail;
    }
@@ -140,46 +140,24 @@ softpipe_texture_create(struct pipe_screen *screen,
 }
 
 
-/**
- * Create a new pipe_texture which wraps an existing buffer.
- */
-static struct pipe_texture *
-softpipe_texture_blanket(struct pipe_screen * screen,
-                         const struct pipe_texture *base,
-                         const unsigned *stride,
-                         struct pipe_buffer *buffer)
-{
-   struct softpipe_texture *spt;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   spt = CALLOC_STRUCT(softpipe_texture);
-   if (!spt)
-      return NULL;
-
-   spt->base = *base;
-   pipe_reference_init(&spt->base.reference, 1);
-   spt->base.screen = screen;
-   spt->stride[0] = stride[0];
-
-   pipe_buffer_reference(&spt->buffer, buffer);
-
-   return &spt->base;
-}
 
 
 static void
 softpipe_texture_destroy(struct pipe_texture *pt)
 {
+   struct softpipe_screen *screen = softpipe_screen(pt->screen);
    struct softpipe_texture *spt = softpipe_texture(pt);
 
-   pipe_buffer_reference(&spt->buffer, NULL);
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = screen->winsys;
+      winsys->displaytarget_destroy(winsys, spt->dt);
+   }
+   else {
+      /* regular texture */
+      align_free(spt->data);
+   }
+
    FREE(spt);
 }
 
@@ -359,9 +337,20 @@ softpipe_transfer_map( struct pipe_screen *screen,
    spt = softpipe_texture(transfer->texture);
    format = transfer->texture->format;
 
-   map = pipe_buffer_map(screen, spt->buffer, pipe_transfer_buffer_flags(transfer));
-   if (map == NULL)
-      return NULL;
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
+      map = winsys->displaytarget_map(winsys, spt->dt,
+                                      pipe_transfer_buffer_flags(transfer));
+      if (map == NULL)
+         return NULL;
+   }
+   else {
+      map = spt->data;
+      if (map == NULL)
+         return NULL;
+   }
 
    /* May want to different things here depending on read/write nature
     * of the map:
@@ -393,7 +382,11 @@ softpipe_transfer_unmap(struct pipe_screen *screen,
    assert(transfer->texture);
    spt = softpipe_texture(transfer->texture);
 
-   pipe_buffer_unmap( screen, spt->buffer );
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+      winsys->displaytarget_unmap(winsys, spt->dt);
+   }
 
    if (transfer->usage & PIPE_TRANSFER_WRITE) {
       /* Mark the texture as dirty to expire the tile caches. */
@@ -458,7 +451,6 @@ void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = softpipe_texture_create;
-   screen->texture_blanket = softpipe_texture_blanket;
    screen->texture_destroy = softpipe_texture_destroy;
 
    screen->get_tex_surface = softpipe_get_tex_surface;
@@ -474,24 +466,4 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 }
 
 
-/**
- * Return pipe_buffer handle and stride for given texture object.
- * XXX used for???
- */
-boolean
-softpipe_get_texture_buffer( struct pipe_texture *texture,
-                             struct pipe_buffer **buf,
-                             unsigned *stride )
-{
-   struct softpipe_texture *tex = (struct softpipe_texture *) texture;
 
-   if (!tex)
-      return FALSE;
-
-   pipe_buffer_reference(buf, tex->buffer);
-
-   if (stride)
-      *stride = tex->stride[0];
-
-   return TRUE;
-}
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 2ef64e1e7c3..1c8636d1d56 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -33,6 +33,10 @@
 #include "pipe/p_video_state.h"
 
 
+#define SP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K */
+#define SP_MAX_TEXTURE_3D_LEVELS 9   /* 512 x 512 x 512 */
+
+
 struct pipe_context;
 struct pipe_screen;
 struct softpipe_context;
@@ -42,12 +46,19 @@ struct softpipe_texture
 {
    struct pipe_texture base;
 
-   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned stride[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long level_offset[SP_MAX_TEXTURE_2D_LEVELS];
+   unsigned stride[SP_MAX_TEXTURE_2D_LEVELS];
 
-   /* The data is held here:
+   /**
+    * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
+    * usage.
+    */
+   struct sw_displaytarget *dt;
+
+   /**
+    * Malloc'ed data for regular textures, or a mapping to dt above.
     */
-   struct pipe_buffer *buffer;
+   void *data;
 
    /* True if texture images are power-of-two in all dimensions:
     */
diff --git a/src/gallium/drivers/softpipe/sp_winsys.c b/src/gallium/drivers/softpipe/sp_winsys.c
deleted file mode 100644
index 0a6245ed2cd..00000000000
--- a/src/gallium/drivers/softpipe/sp_winsys.c
+++ /dev/null
@@ -1,245 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/**
- * @file
- * Malloc softpipe winsys. Uses malloc for all memory allocations.
- * 
- * @author Keith Whitwell
- * @author Brian Paul
- * @author Jose Fonseca
- */
-
-
-#include "util/u_simple_screen.h"/* port to just p_screen */
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "util/u_inlines.h"
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "sp_winsys.h"
-
-
-struct st_softpipe_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-};
-
-
-/** Cast wrapper */
-static INLINE struct st_softpipe_buffer *
-st_softpipe_buffer( struct pipe_buffer *buf )
-{
-   return (struct st_softpipe_buffer *)buf;
-}
-
-
-static void *
-st_softpipe_buffer_map(struct pipe_winsys *winsys, 
-                       struct pipe_buffer *buf,
-                       unsigned flags)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = st_softpipe_buf->data;
-   return st_softpipe_buf->mapped;
-}
-
-
-static void
-st_softpipe_buffer_unmap(struct pipe_winsys *winsys, 
-                         struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = NULL;
-}
-
-
-static void
-st_softpipe_buffer_destroy(struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *oldBuf = st_softpipe_buffer(buf);
-
-   if (oldBuf->data) {
-      if (!oldBuf->userBuffer)
-         align_free(oldBuf->data);
-
-      oldBuf->data = NULL;
-   }
-
-   FREE(oldBuf);
-}
-
-
-static void
-st_softpipe_flush_frontbuffer(struct pipe_winsys *winsys,
-                              struct pipe_surface *surf,
-                              void *context_private)
-{
-}
-
-
-
-static const char *
-st_softpipe_get_name(struct pipe_winsys *winsys)
-{
-   return "softpipe";
-}
-
-
-static struct pipe_buffer *
-st_softpipe_buffer_create(struct pipe_winsys *winsys, 
-                          unsigned alignment, 
-                          unsigned usage,
-                          unsigned size)
-{
-   struct st_softpipe_buffer *buffer = CALLOC_STRUCT(st_softpipe_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-   buffer->data = align_malloc(size, alignment);
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-st_softpipe_user_buffer_create(struct pipe_winsys *winsys, 
-                               void *ptr, 
-                               unsigned bytes)
-{
-   struct st_softpipe_buffer *buffer;
-   
-   buffer = CALLOC_STRUCT(st_softpipe_buffer);
-   if(!buffer)
-      return NULL;
-   
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-
-   return &buffer->base;
-}
-
-
-static struct pipe_buffer *
-st_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
-                                  unsigned width, unsigned height,
-                                  enum pipe_format format,
-                                  unsigned usage,
-                                  unsigned tex_usage,
-                                  unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-st_softpipe_fence_reference(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle **ptr,
-                            struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-st_softpipe_fence_signalled(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle *fence,
-                            unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-st_softpipe_fence_finish(struct pipe_winsys *winsys, 
-                         struct pipe_fence_handle *fence,
-                         unsigned flag)
-{
-   return 0;
-}
-
-
-static void
-st_softpipe_destroy(struct pipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-struct pipe_screen *
-softpipe_create_screen_malloc(void)
-{
-   static struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(pipe_winsys);
-   if(!winsys)
-      return NULL;
-
-   winsys->destroy = st_softpipe_destroy;
-   
-   winsys->buffer_create = st_softpipe_buffer_create;
-   winsys->user_buffer_create = st_softpipe_user_buffer_create;
-   winsys->buffer_map = st_softpipe_buffer_map;
-   winsys->buffer_unmap = st_softpipe_buffer_unmap;
-   winsys->buffer_destroy = st_softpipe_buffer_destroy;
-
-   winsys->surface_buffer_create = st_softpipe_surface_buffer_create;
-
-   winsys->fence_reference = st_softpipe_fence_reference;
-   winsys->fence_signalled = st_softpipe_fence_signalled;
-   winsys->fence_finish = st_softpipe_fence_finish;
-
-   winsys->flush_frontbuffer = st_softpipe_flush_frontbuffer;
-   winsys->get_name = st_softpipe_get_name;
-
-   screen = softpipe_create_screen(winsys);
-   if(!screen)
-      st_softpipe_destroy(winsys);
-
-   return screen;
-}
diff --git a/src/gallium/drivers/softpipe/sp_winsys.h b/src/gallium/drivers/softpipe/sp_winsys.h
deleted file mode 100644
index 6e3920c49b2..00000000000
--- a/src/gallium/drivers/softpipe/sp_winsys.h
+++ /dev/null
@@ -1,73 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/* This is the interface that softpipe requires any window system
- * hosting it to implement.  This is the only include file in softpipe
- * which is public.
- */
-
-
-#ifndef SP_WINSYS_H
-#define SP_WINSYS_H
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#include "pipe/p_defines.h"
-
-struct pipe_screen;
-struct pipe_winsys;
-struct pipe_context;
-struct pipe_texture;
-struct pipe_buffer;
-
-
-
-/**
- * Create a softpipe screen that uses the
- * given winsys for allocating buffers.
- */
-struct pipe_screen *softpipe_create_screen( struct pipe_winsys * );
-
-/**
- * Create a softpipe screen that uses
- * regular malloc to create all its buffers.
- */
-struct pipe_screen *softpipe_create_screen_malloc(void);
-
-boolean
-softpipe_get_texture_buffer( struct pipe_texture *texture,
-                             struct pipe_buffer **buf,
-                             unsigned *stride );
-
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* SP_WINSYS_H */
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index 03302e2a6ec..791d30edc0e 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -169,6 +169,11 @@ struct svga_sampler_state {
    unsigned view_max_lod;
 };
 
+struct svga_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
 /* Use to calculate differences between state emitted to hardware and
  * current driver-calculated state.  
  */
@@ -178,13 +183,13 @@ struct svga_state
    const struct svga_depth_stencil_state *depth;
    const struct svga_rasterizer_state *rast;
    const struct svga_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   const struct svga_velems_state *velems;
 
    struct pipe_texture *texture[PIPE_MAX_SAMPLERS]; /* or texture ID's? */
    struct svga_fragment_shader *fs;
    struct svga_vertex_shader *vs;
 
    struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
    struct pipe_buffer *cb[PIPE_SHADER_TYPES];
 
    struct pipe_framebuffer_state framebuffer;
@@ -204,7 +209,6 @@ struct svga_state
 
    unsigned num_samplers;
    unsigned num_textures;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
    unsigned reduced_prim;
 
diff --git a/src/gallium/drivers/svga/svga_pipe_sampler.c b/src/gallium/drivers/svga/svga_pipe_sampler.c
index acba2b8f9da..1a8ef296cac 100644
--- a/src/gallium/drivers/svga/svga_pipe_sampler.c
+++ b/src/gallium/drivers/svga/svga_pipe_sampler.c
@@ -155,7 +155,7 @@ static void svga_bind_sampler_states(struct pipe_context *pipe,
    /* Check for no-op */
    if (num == svga->curr.num_samplers &&
        !memcmp(svga->curr.sampler, sampler, num * sizeof(void *))) {
-      debug_printf("sampler noop\n");
+      if (0) debug_printf("sampler noop\n");
       return;
    }
 
diff --git a/src/gallium/drivers/svga/svga_pipe_vertex.c b/src/gallium/drivers/svga/svga_pipe_vertex.c
index 836b8441da2..1715a47fc62 100644
--- a/src/gallium/drivers/svga/svga_pipe_vertex.c
+++ b/src/gallium/drivers/svga/svga_pipe_vertex.c
@@ -26,6 +26,7 @@
 #include "util/u_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
+#include "util/u_memory.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "svga_screen.h"
@@ -64,20 +65,37 @@ static void svga_set_vertex_buffers(struct pipe_context *pipe,
    svga->dirty |= SVGA_NEW_VBUFFER;
 }
 
-static void svga_set_vertex_elements(struct pipe_context *pipe,
-                                     unsigned count,
-                                     const struct pipe_vertex_element *elements)
+
+static void *
+svga_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
 {
-   struct svga_context *svga = svga_context(pipe);
-   unsigned i;
+   struct svga_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct svga_velems_state *) MALLOC(sizeof(struct svga_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
 
-   for (i = 0; i < count; i++)
-      svga->curr.ve[i] = elements[i];
+static void svga_bind_vertex_elements_state(struct pipe_context *pipe,
+                                            void *velems)
+{
+   struct svga_context *svga = svga_context(pipe);
+   struct svga_velems_state *svga_velems = (struct svga_velems_state *) velems;
 
-   svga->curr.num_vertex_elements = count;
+   svga->curr.velems = svga_velems;
    svga->dirty |= SVGA_NEW_VELEMENT;
 }
 
+static void svga_delete_vertex_elements_state(struct pipe_context *pipe,
+                                              void *velems)
+{
+   FREE(velems);
+}
 
 void svga_cleanup_vertex_state( struct svga_context *svga )
 {
@@ -91,7 +109,9 @@ void svga_cleanup_vertex_state( struct svga_context *svga )
 void svga_init_vertex_functions( struct svga_context *svga )
 {
    svga->pipe.set_vertex_buffers = svga_set_vertex_buffers;
-   svga->pipe.set_vertex_elements = svga_set_vertex_elements;
+   svga->pipe.create_vertex_elements_state = svga_create_vertex_elements_state;
+   svga->pipe.bind_vertex_elements_state = svga_bind_vertex_elements_state;
+   svga->pipe.delete_vertex_elements_state = svga_delete_vertex_elements_state;
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_screen_texture.c b/src/gallium/drivers/svga/svga_screen_texture.c
index e9792e063e0..5b581debfc7 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.c
+++ b/src/gallium/drivers/svga/svga_screen_texture.c
@@ -315,7 +315,11 @@ svga_texture_create(struct pipe_screen *screen,
       tex->key.cachable = 0;
    }
 
-   if(templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) {
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_SHARED) {
+      tex->key.cachable = 0;
+   }
+
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT) {
       tex->key.flags |= SVGA3D_SURFACE_HINT_SCANOUT;
       tex->key.cachable = 0;
    }
@@ -355,80 +359,18 @@ error1:
 }
 
 
-static struct pipe_texture *
-svga_texture_blanket(struct pipe_screen * screen,
-                     const struct pipe_texture *base,
-                     const unsigned *stride,
-                     struct pipe_buffer *buffer)
-{
-   struct svga_texture *tex;
-   struct svga_buffer *sbuf = svga_buffer(buffer);
-   struct svga_winsys_screen *sws = svga_winsys_screen(screen);
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   /**
-    * We currently can't do texture blanket on
-    * SVGA3D_BUFFER. Need to blit to a temporary surface?
-    */
-
-   assert(sbuf->handle);
-   if (!sbuf->handle)
-      return NULL;
-
-   if (svga_translate_format(base->format) != sbuf->key.format) {
-      unsigned f1 = svga_translate_format(base->format);
-      unsigned f2 = sbuf->key.format;
-
-      /* It's okay for XRGB and ARGB or depth with/out stencil to get mixed up */
-      if ( !( (f1 == SVGA3D_X8R8G8B8 && f2 == SVGA3D_A8R8G8B8) ||
-              (f1 == SVGA3D_A8R8G8B8 && f2 == SVGA3D_X8R8G8B8) ||
-              (f1 == SVGA3D_Z_D24X8 && f2 == SVGA3D_Z_D24S8) ) ) {
-         debug_printf("%s wrong format %u != %u\n", __FUNCTION__, f1, f2);
-         return NULL;
-      }
-   }
-
-   tex = CALLOC_STRUCT(svga_texture);
-   if (!tex)
-      return NULL;
-
-   tex->base = *base;
-   
-
-   if (sbuf->key.format == 1)
-      tex->base.format = PIPE_FORMAT_B8G8R8X8_UNORM;
-   else if (sbuf->key.format == 2)
-      tex->base.format = PIPE_FORMAT_B8G8R8A8_UNORM;
-
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
 
-   SVGA_DBG(DEBUG_DMA, "blanket sid %p\n", sbuf->handle);
 
-   /* We don't own this storage, so don't try to cache it.
-    */
-   assert(sbuf->key.cachable == 0);
-   tex->key.cachable = 0;
-   sws->surface_reference(sws, &tex->handle, sbuf->handle);
 
-   return &tex->base;
-}
-
-
-struct pipe_texture *
-svga_screen_texture_wrap_surface(struct pipe_screen *screen,
-				 struct pipe_texture *base,
-				 enum SVGA3dSurfaceFormat format,
-				 struct svga_winsys_surface *srf)
+static struct pipe_texture *
+svga_screen_texture_from_handle(struct pipe_screen *screen,
+                                const struct pipe_texture *base,
+                                struct winsys_handle *whandle)
 {
+   struct svga_winsys_screen *sws = svga_winsys_screen(screen);
+   struct svga_winsys_surface *srf;
    struct svga_texture *tex;
+   enum SVGA3dSurfaceFormat format = 0;
    assert(screen);
 
    /* Only supports one type */
@@ -438,6 +380,8 @@ svga_screen_texture_wrap_surface(struct pipe_screen *screen,
       return NULL;
    }
 
+   srf = sws->surface_from_handle(sws, whandle, &format);
+
    if (!srf)
       return NULL;
 
@@ -478,6 +422,22 @@ svga_screen_texture_wrap_surface(struct pipe_screen *screen,
 }
 
 
+static boolean 
+svga_screen_texture_get_handle(struct pipe_screen *screen,
+                               struct pipe_texture *texture,
+                               struct winsys_handle *whandle)
+{
+   struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
+   unsigned stride;
+
+   assert(svga_texture(texture)->key.cachable == 0);
+   svga_texture(texture)->key.cachable = 0;
+   stride = util_format_get_nblocksx(texture->format, texture->width0) *
+            util_format_get_blocksize(texture->format);
+   return sws->surface_get_handle(sws, svga_texture(texture)->handle, stride, whandle);
+}
+
+
 static void
 svga_texture_destroy(struct pipe_texture *pt)
 {
@@ -955,10 +915,11 @@ void
 svga_screen_init_texture_functions(struct pipe_screen *screen)
 {
    screen->texture_create = svga_texture_create;
+   screen->texture_from_handle = svga_screen_texture_from_handle;
+   screen->texture_get_handle = svga_screen_texture_get_handle;
    screen->texture_destroy = svga_texture_destroy;
    screen->get_tex_surface = svga_get_tex_surface;
    screen->tex_surface_destroy = svga_tex_surface_destroy;
-   screen->texture_blanket = svga_texture_blanket;
    screen->get_tex_transfer = svga_get_tex_transfer;
    screen->transfer_map = svga_transfer_map;
    screen->transfer_unmap = svga_transfer_unmap;
@@ -1120,33 +1081,3 @@ svga_destroy_sampler_view_priv(struct svga_sampler_view *v)
    pipe_texture_reference(&v->texture, NULL);
    FREE(v);
 }
-
-boolean
-svga_screen_buffer_from_texture(struct pipe_texture *texture,
-				struct pipe_buffer **buffer,
-				unsigned *stride)
-{
-   struct svga_texture *stex = svga_texture(texture);
-
-   *buffer = svga_screen_buffer_wrap_surface
-      (texture->screen,
-       svga_translate_format(texture->format),
-       stex->handle);
-
-   *stride = util_format_get_stride(texture->format, texture->width0);
-
-   return *buffer != NULL;
-}
-
-
-struct svga_winsys_surface *
-svga_screen_texture_get_winsys_surface(struct pipe_texture *texture)
-{
-   struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
-   struct svga_winsys_surface *vsurf = NULL;
-
-   assert(svga_texture(texture)->key.cachable == 0);
-   svga_texture(texture)->key.cachable = 0;
-   sws->surface_reference(sws, &vsurf, svga_texture(texture)->handle);
-   return vsurf;
-}
diff --git a/src/gallium/drivers/svga/svga_screen_texture.h b/src/gallium/drivers/svga/svga_screen_texture.h
index 24c1f78ca55..ca6602b4369 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.h
+++ b/src/gallium/drivers/svga/svga_screen_texture.h
@@ -78,7 +78,7 @@ struct svga_texture
 {
    struct pipe_texture base;
 
-   boolean defined[6][PIPE_MAX_TEXTURE_LEVELS];
+   boolean defined[6][SVGA_MAX_TEXTURE_LEVELS];
    
    struct svga_sampler_view *cached_view;
 
diff --git a/src/gallium/drivers/svga/svga_state_need_swtnl.c b/src/gallium/drivers/svga/svga_state_need_swtnl.c
index d774e3e504d..dfaab53aef4 100644
--- a/src/gallium/drivers/svga/svga_state_need_swtnl.c
+++ b/src/gallium/drivers/svga/svga_state_need_swtnl.c
@@ -76,8 +76,13 @@ static int update_need_swvfetch( struct svga_context *svga,
    unsigned i;
    boolean need_swvfetch = FALSE;
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
-      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.ve[i].src_format);
+   if (!svga->curr.velems) {
+      /* No vertex elements bound. */
+      return 0;
+   }
+
+   for (i = 0; i < svga->curr.velems->count; i++) {
+      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.velems->velem[i].src_format);
       if (svga->state.sw.ve_format[i] == SVGA3D_DECLTYPE_MAX) {
          need_swvfetch = TRUE;
          break;
diff --git a/src/gallium/drivers/svga/svga_state_rss.c b/src/gallium/drivers/svga/svga_state_rss.c
index 107cc403b4d..b7195d246bc 100644
--- a/src/gallium/drivers/svga/svga_state_rss.c
+++ b/src/gallium/drivers/svga/svga_state_rss.c
@@ -191,15 +191,24 @@ static int emit_rss( struct svga_context *svga,
       EMIT_RS( svga, svga->curr.stencil_ref.ref_value[0], STENCILREF, fail );
    }
 
-   if (dirty & SVGA_NEW_RAST)
+   if (dirty & (SVGA_NEW_RAST | SVGA_NEW_NEED_PIPELINE))
    {
       const struct svga_rasterizer_state *curr = svga->curr.rast; 
+      unsigned cullmode = curr->cullmode;
 
       /* Shademode: still need to rearrange index list to move
        * flat-shading PV first vertex.
        */
       EMIT_RS( svga, curr->shademode, SHADEMODE, fail );
-      EMIT_RS( svga, curr->cullmode, CULLMODE, fail );
+
+      /* Don't do culling while the software pipeline is active.  It
+       * does it for us, and additionally introduces potentially
+       * back-facing triangles.
+       */
+      if (svga->state.sw.need_pipeline)
+         cullmode = SVGA3D_FACE_NONE;
+
+      EMIT_RS( svga, cullmode, CULLMODE, fail );
       EMIT_RS( svga, curr->scissortestenable, SCISSORTESTENABLE, fail );
       EMIT_RS( svga, curr->multisampleantialias, MULTISAMPLEANTIALIAS, fail );
       EMIT_RS( svga, curr->lastpixel, LASTPIXEL, fail );
diff --git a/src/gallium/drivers/svga/svga_state_vdecl.c b/src/gallium/drivers/svga/svga_state_vdecl.c
index ded903170b5..f531e223048 100644
--- a/src/gallium/drivers/svga/svga_state_vdecl.c
+++ b/src/gallium/drivers/svga/svga_state_vdecl.c
@@ -95,17 +95,17 @@ upload_user_buffers( struct svga_context *svga )
 static int emit_hw_vs_vdecl( struct svga_context *svga,
                              unsigned dirty )
 {
-   const struct pipe_vertex_element *ve = svga->curr.ve;
+   const struct pipe_vertex_element *ve = svga->curr.velems->velem;
    SVGA3dVertexDecl decl;
    unsigned i;
 
-   assert(svga->curr.num_vertex_elements >=
+   assert(svga->curr.velems->count >=
           svga->curr.vs->base.info.file_count[TGSI_FILE_INPUT]);
 
    svga_hwtnl_reset_vdecl( svga->hwtnl, 
-                           svga->curr.num_vertex_elements );
+                           svga->curr.velems->count );
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
+   for (i = 0; i < svga->curr.velems->count; i++) {
       const struct pipe_vertex_buffer *vb = &svga->curr.vb[ve[i].vertex_buffer_index];
       unsigned usage, index;
 
diff --git a/src/gallium/drivers/svga/svga_state_vs.c b/src/gallium/drivers/svga/svga_state_vs.c
index d7999fe53d2..781f7bf5339 100644
--- a/src/gallium/drivers/svga/svga_state_vs.c
+++ b/src/gallium/drivers/svga/svga_state_vs.c
@@ -186,8 +186,8 @@ static int update_zero_stride( struct svga_context *svga,
    svga->curr.zero_stride_vertex_elements = 0;
    svga->curr.num_zero_stride_vertex_elements = 0;
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
-      const struct pipe_vertex_element *vel = &svga->curr.ve[i];
+   for (i = 0; i < svga->curr.velems->count; i++) {
+      const struct pipe_vertex_element *vel = &svga->curr.velems->velem[i];
       const struct pipe_vertex_buffer *vbuffer = &svga->curr.vb[
          vel->vertex_buffer_index];
       if (vbuffer->stride == 0) {
diff --git a/src/gallium/drivers/svga/svga_swtnl_state.c b/src/gallium/drivers/svga/svga_swtnl_state.c
index 35f36a828fd..246d34e649e 100644
--- a/src/gallium/drivers/svga/svga_swtnl_state.c
+++ b/src/gallium/drivers/svga/svga_swtnl_state.c
@@ -99,8 +99,8 @@ static int update_swtnl_draw( struct svga_context *svga,
 
    if (dirty & SVGA_NEW_VELEMENT)
       draw_set_vertex_elements(svga->swtnl.draw, 
-                               svga->curr.num_vertex_elements, 
-                               svga->curr.ve );
+                               svga->curr.velems->count, 
+                               svga->curr.velems->velem );
 
    if (dirty & SVGA_NEW_CLIP)
       draw_set_clip_state(svga->swtnl.draw, 
diff --git a/src/gallium/drivers/svga/svga_winsys.h b/src/gallium/drivers/svga/svga_winsys.h
index b4e3af0eafc..d4bb176f9a8 100644
--- a/src/gallium/drivers/svga/svga_winsys.h
+++ b/src/gallium/drivers/svga/svga_winsys.h
@@ -51,6 +51,7 @@ struct pipe_context;
 struct pipe_fence_handle;
 struct pipe_texture;
 struct svga_region;
+struct winsys_handle;
 
 
 #define SVGA_BUFFER_USAGE_PINNED  (PIPE_BUFFER_USAGE_CUSTOM << 0)
@@ -186,6 +187,25 @@ struct svga_winsys_screen
                      uint32 numFaces,
                      uint32 numMipLevels);
 
+   /**
+    * Creates a surface from a winsys handle.
+    * Used to implement pipe_screen::texture_from_handle.
+    */
+   struct svga_winsys_surface *
+   (*surface_from_handle)(struct svga_winsys_screen *sws,
+                          struct winsys_handle *whandle,
+                          SVGA3dSurfaceFormat *format);
+
+   /**
+    * Get a winsys_handle from a surface.
+    * Used to implement pipe_screen::texture_get_handle.
+    */
+   boolean
+   (*surface_get_handle)(struct svga_winsys_screen *sws,
+                         struct svga_winsys_surface *surface,
+                         unsigned stride,
+                         struct winsys_handle *whandle);
+
    /**
     * Whether this surface is sitting in a validate list
     */
@@ -283,20 +303,7 @@ svga_screen_buffer_wrap_surface(struct pipe_screen *screen,
 				enum SVGA3dSurfaceFormat format,
 				struct svga_winsys_surface *srf);
 
-struct svga_winsys_surface *
-svga_screen_texture_get_winsys_surface(struct pipe_texture *texture);
 struct svga_winsys_surface *
 svga_screen_buffer_get_winsys_surface(struct pipe_buffer *buffer);
 
-boolean
-svga_screen_buffer_from_texture(struct pipe_texture *texture,
-				struct pipe_buffer **buffer,
-				unsigned *stride);
-
-struct pipe_texture *
-svga_screen_texture_wrap_surface(struct pipe_screen *screen,
-				 struct pipe_texture *base,
-				 enum SVGA3dSurfaceFormat format,
-				 struct svga_winsys_surface *srf);
-
 #endif /* SVGA_WINSYS_H_ */
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index df40fbade6c..133521f45e2 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -773,6 +773,70 @@ trace_context_delete_vs_state(struct pipe_context *_pipe,
 }
 
 
+static INLINE void *
+trace_context_create_vertex_elements_state(struct pipe_context *_pipe,
+                                           unsigned num_elements,
+                                           const struct  pipe_vertex_element *elements)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+   void * result;
+
+   trace_dump_call_begin("pipe_context", "create_vertex_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(uint, num_elements);
+
+   trace_dump_arg_begin("elements");
+   trace_dump_struct_array(vertex_element, elements, num_elements);
+   trace_dump_arg_end();
+
+   result = pipe->create_vertex_elements_state(pipe, num_elements, elements);
+
+   trace_dump_ret(ptr, result);
+
+   trace_dump_call_end();
+
+   return result;
+}
+
+
+static INLINE void
+trace_context_bind_vertex_elements_state(struct pipe_context *_pipe,
+                                         void *state)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+
+   trace_dump_call_begin("pipe_context", "bind_vertex_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, state);
+
+   pipe->bind_vertex_elements_state(pipe, state);
+
+   trace_dump_call_end();
+}
+
+
+static INLINE void
+trace_context_delete_vertex_elements_state(struct pipe_context *_pipe,
+                                           void *state)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+
+   trace_dump_call_begin("pipe_context", "delete_verte_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, state);
+
+   pipe->delete_vertex_elements_state(pipe, state);
+
+   trace_dump_call_end();
+}
+
+
 static INLINE void
 trace_context_set_blend_color(struct pipe_context *_pipe,
                               const struct pipe_blend_color *state)
@@ -1047,29 +1111,6 @@ trace_context_set_vertex_buffers(struct pipe_context *_pipe,
 }
 
 
-static INLINE void
-trace_context_set_vertex_elements(struct pipe_context *_pipe,
-                                  unsigned num_elements,
-                                  const struct pipe_vertex_element *elements)
-{
-   struct trace_context *tr_ctx = trace_context(_pipe);
-   struct pipe_context *pipe = tr_ctx->pipe;
-
-   trace_dump_call_begin("pipe_context", "set_vertex_elements");
-
-   trace_dump_arg(ptr, pipe);
-   trace_dump_arg(uint, num_elements);
-
-   trace_dump_arg_begin("elements");
-   trace_dump_struct_array(vertex_element, elements, num_elements);
-   trace_dump_arg_end();
-
-   pipe->set_vertex_elements(pipe, num_elements, elements);
-
-   trace_dump_call_end();
-}
-
-
 static INLINE void
 trace_context_surface_copy(struct pipe_context *_pipe,
                            struct pipe_surface *dest,
@@ -1303,6 +1344,9 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.create_vs_state = trace_context_create_vs_state;
    tr_ctx->base.bind_vs_state = trace_context_bind_vs_state;
    tr_ctx->base.delete_vs_state = trace_context_delete_vs_state;
+   tr_ctx->base.create_vertex_elements_state = trace_context_create_vertex_elements_state;
+   tr_ctx->base.bind_vertex_elements_state = trace_context_bind_vertex_elements_state;
+   tr_ctx->base.delete_vertex_elements_state = trace_context_delete_vertex_elements_state;
    tr_ctx->base.set_blend_color = trace_context_set_blend_color;
    tr_ctx->base.set_stencil_ref = trace_context_set_stencil_ref;
    tr_ctx->base.set_clip_state = trace_context_set_clip_state;
@@ -1314,7 +1358,6 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.set_fragment_sampler_textures = trace_context_set_fragment_sampler_textures;
    tr_ctx->base.set_vertex_sampler_textures = trace_context_set_vertex_sampler_textures;
    tr_ctx->base.set_vertex_buffers = trace_context_set_vertex_buffers;
-   tr_ctx->base.set_vertex_elements = trace_context_set_vertex_elements;
    if (pipe->surface_copy)
       tr_ctx->base.surface_copy = trace_context_surface_copy;
    if (pipe->surface_fill)
diff --git a/src/gallium/drivers/trace/tr_drm.c b/src/gallium/drivers/trace/tr_drm.c
index 2b4915003e2..906b3262e4b 100644
--- a/src/gallium/drivers/trace/tr_drm.c
+++ b/src/gallium/drivers/trace/tr_drm.c
@@ -33,6 +33,7 @@
 #include "tr_context.h"
 #include "tr_buffer.h"
 #include "tr_texture.h"
+#include "tr_public.h"
 
 struct trace_drm_api
 {
@@ -62,69 +63,8 @@ trace_drm_create_screen(struct drm_api *_api, int fd,
 
    screen = api->create_screen(api, fd, arg);
 
-   return trace_screen_create(screen);
-}
-
-
-static struct pipe_texture *
-trace_drm_texture_from_shared_handle(struct drm_api *_api,
-                                     struct pipe_screen *_screen,
-                                     struct pipe_texture *templ,
-                                     const char *name,
-                                     unsigned stride,
-                                     unsigned handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct drm_api *api = tr_api->api;
-   struct pipe_texture *result;
-
-   /* TODO trace call */
-
-   result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle);
-
-   result = trace_texture_create(trace_screen(_screen), result);
-
-   return result;
-}
-
-static boolean
-trace_drm_shared_handle_from_texture(struct drm_api *_api,
-                                     struct pipe_screen *_screen,
-                                     struct pipe_texture *_texture,
-                                     unsigned *stride,
-                                     unsigned *handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_texture *tr_texture = trace_texture(_texture);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct pipe_texture *texture = tr_texture->texture;
-   struct drm_api *api = tr_api->api;
-
-   /* TODO trace call */
-
-   return api->shared_handle_from_texture(api, screen, texture, stride, handle);
-}
 
-static boolean
-trace_drm_local_handle_from_texture(struct drm_api *_api,
-                                    struct pipe_screen *_screen,
-                                    struct pipe_texture *_texture,
-                                    unsigned *stride,
-                                    unsigned *handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_texture *tr_texture = trace_texture(_texture);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct pipe_texture *texture = tr_texture->texture;
-   struct drm_api *api = tr_api->api;
-
-   /* TODO trace call */
-
-   return api->local_handle_from_texture(api, screen, texture, stride, handle);
+   return trace_screen_create(screen);
 }
 
 static void
@@ -158,9 +98,6 @@ trace_drm_create(struct drm_api *api)
    tr_api->base.name = api->name;
    tr_api->base.driver_name = api->driver_name;
    tr_api->base.create_screen = trace_drm_create_screen;
-   tr_api->base.texture_from_shared_handle = trace_drm_texture_from_shared_handle;
-   tr_api->base.shared_handle_from_texture = trace_drm_shared_handle_from_texture;
-   tr_api->base.local_handle_from_texture = trace_drm_local_handle_from_texture;
    tr_api->base.destroy = trace_drm_destroy;
    tr_api->api = api;
 
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index f97d963dba6..f82dd01c697 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -479,7 +479,6 @@ void trace_dump_vertex_element(const struct pipe_vertex_element *state)
    trace_dump_member(uint, state, src_offset);
 
    trace_dump_member(uint, state, vertex_buffer_index);
-   trace_dump_member(uint, state, nr_components);
 
    trace_dump_member(format, state, src_format);
 
diff --git a/src/gallium/drivers/trace/tr_public.h b/src/gallium/drivers/trace/tr_public.h
new file mode 100644
index 00000000000..62e217097d0
--- /dev/null
+++ b/src/gallium/drivers/trace/tr_public.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ **************************************************************************/
+
+#ifndef TR_PUBLIC_H
+#define TR_PUBLIC_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct pipe_screen;
+struct pipe_context;
+
+struct pipe_screen *
+trace_screen_create(struct pipe_screen *screen);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* TR_PUBLIC_H */
diff --git a/src/gallium/drivers/trace/tr_screen.c b/src/gallium/drivers/trace/tr_screen.c
index 388d83eb5c2..86ddb995405 100644
--- a/src/gallium/drivers/trace/tr_screen.c
+++ b/src/gallium/drivers/trace/tr_screen.c
@@ -35,6 +35,7 @@
 #include "tr_texture.h"
 #include "tr_context.h"
 #include "tr_screen.h"
+#include "tr_public.h"
 
 #include "util/u_inlines.h"
 #include "pipe/p_format.h"
@@ -236,39 +237,41 @@ trace_screen_texture_create(struct pipe_screen *_screen,
    return result;
 }
 
-
 static struct pipe_texture *
-trace_screen_texture_blanket(struct pipe_screen *_screen,
-                             const struct pipe_texture *templat,
-                             const unsigned *ppitch,
-                             struct pipe_buffer *_buffer)
+trace_screen_texture_from_handle(struct pipe_screen *_screen,
+                                 const struct pipe_texture *templ,
+                                 struct winsys_handle *handle)
 {
-   struct trace_screen *tr_scr = trace_screen(_screen);
-   struct trace_buffer *tr_buf = trace_buffer(_buffer);
-   struct pipe_screen *screen = tr_scr->screen;
-   struct pipe_buffer *buffer = tr_buf->buffer;
-   unsigned pitch = *ppitch;
+   struct trace_screen *tr_screen = trace_screen(_screen);
+   struct pipe_screen *screen = tr_screen->screen;
    struct pipe_texture *result;
 
-   trace_dump_call_begin("pipe_screen", "texture_blanket");
+   /* TODO trace call */
 
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(template, templat);
-   trace_dump_arg(uint, pitch);
-   trace_dump_arg(ptr, buffer);
+   result = screen->texture_from_handle(screen, templ, handle);
 
-   result = screen->texture_blanket(screen, templat, ppitch, buffer);
+   result = trace_texture_create(trace_screen(_screen), result);
 
-   trace_dump_ret(ptr, result);
+   return result;
+}
 
-   trace_dump_call_end();
+static boolean
+trace_screen_texture_get_handle(struct pipe_screen *_screen,
+                                struct pipe_texture *_texture,
+                                struct winsys_handle *handle)
+{
+   struct trace_screen *tr_screen = trace_screen(_screen);
+   struct trace_texture *tr_texture = trace_texture(_texture);
+   struct pipe_screen *screen = tr_screen->screen;
+   struct pipe_texture *texture = tr_texture->texture;
 
-   result = trace_texture_create(tr_scr, result);
+   /* TODO trace call */
 
-   return result;
+   return screen->texture_get_handle(screen, texture, handle);
 }
 
 
+
 static void
 trace_screen_texture_destroy(struct pipe_texture *_texture)
 {
@@ -484,45 +487,7 @@ trace_screen_transfer_unmap(struct pipe_screen *_screen,
  */
 
 
-static struct pipe_buffer *
-trace_screen_surface_buffer_create(struct pipe_screen *_screen,
-                                   unsigned width, unsigned height,
-                                   enum pipe_format format,
-                                   unsigned usage,
-                                   unsigned tex_usage,
-                                   unsigned *pstride)
-{
-   struct trace_screen *tr_scr = trace_screen(_screen);
-   struct pipe_screen *screen = tr_scr->screen;
-   unsigned stride;
-   struct pipe_buffer *result;
-
-   trace_dump_call_begin("pipe_screen", "surface_buffer_create");
-
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(uint, width);
-   trace_dump_arg(uint, height);
-   trace_dump_arg(format, format);
-   trace_dump_arg(uint, usage);
-   trace_dump_arg(uint, tex_usage);
 
-   result = screen->surface_buffer_create(screen,
-                                          width, height,
-                                          format,
-                                          usage,
-                                          tex_usage,
-                                          pstride);
-
-   stride = *pstride;
-
-   trace_dump_arg(uint, stride);
-
-   trace_dump_ret(ptr, result);
-
-   trace_dump_call_end();
-
-   return trace_buffer_create(tr_scr, result);
-}
 
 
 static struct pipe_buffer *
@@ -931,7 +896,8 @@ trace_screen_create(struct pipe_screen *screen)
    assert(screen->context_create);
    tr_scr->base.context_create = trace_screen_context_create;
    tr_scr->base.texture_create = trace_screen_texture_create;
-   tr_scr->base.texture_blanket = trace_screen_texture_blanket;
+   tr_scr->base.texture_from_handle = trace_screen_texture_from_handle;
+   tr_scr->base.texture_get_handle = trace_screen_texture_get_handle;
    tr_scr->base.texture_destroy = trace_screen_texture_destroy;
    tr_scr->base.get_tex_surface = trace_screen_get_tex_surface;
    tr_scr->base.tex_surface_destroy = trace_screen_tex_surface_destroy;
@@ -941,7 +907,6 @@ trace_screen_create(struct pipe_screen *screen)
    tr_scr->base.transfer_unmap = trace_screen_transfer_unmap;
    tr_scr->base.buffer_create = trace_screen_buffer_create;
    tr_scr->base.user_buffer_create = trace_screen_user_buffer_create;
-   tr_scr->base.surface_buffer_create = trace_screen_surface_buffer_create;
    if (screen->buffer_map)
       tr_scr->base.buffer_map = trace_screen_buffer_map;
    if (screen->buffer_map_range)
diff --git a/src/gallium/drivers/trace/tr_screen.h b/src/gallium/drivers/trace/tr_screen.h
index fe5a0fa1909..597e2fc2650 100644
--- a/src/gallium/drivers/trace/tr_screen.h
+++ b/src/gallium/drivers/trace/tr_screen.h
@@ -99,9 +99,6 @@ trace_enabled(void);
 struct trace_screen *
 trace_screen(struct pipe_screen *screen);
 
-struct pipe_screen *
-trace_screen_create(struct pipe_screen *screen);
-
 void
 trace_screen_user_buffer_update(struct pipe_screen *screen,
                                 struct pipe_buffer *buffer);
diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h
index b93b38310ac..e2766d15cd1 100644
--- a/src/gallium/include/pipe/p_compiler.h
+++ b/src/gallium/include/pipe/p_compiler.h
@@ -31,13 +31,8 @@
 
 #include "p_config.h"
 
-#ifndef XFree86Server
 #include <stdlib.h>
 #include <string.h>
-#else
-#include "xf86_ansic.h"
-#include "xf86_libc.h"
-#endif
 #include <stddef.h>
 #include <stdarg.h>
 
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index f82b77903e9..376b01aa696 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -177,6 +177,12 @@ struct pipe_context {
    void   (*bind_gs_state)(struct pipe_context *, void *);
    void   (*delete_gs_state)(struct pipe_context *, void *);
 
+   void * (*create_vertex_elements_state)(struct pipe_context *,
+                                          unsigned num_elements,
+                                          const struct pipe_vertex_element *);
+   void   (*bind_vertex_elements_state)(struct pipe_context *, void *);
+   void   (*delete_vertex_elements_state)(struct pipe_context *, void *);
+
    /*@}*/
 
    /**
@@ -220,9 +226,6 @@ struct pipe_context {
                                unsigned num_buffers,
                                const struct pipe_vertex_buffer * );
 
-   void (*set_vertex_elements)( struct pipe_context *,
-                                unsigned num_elements,
-                                const struct pipe_vertex_element * );
    /*@}*/
 
 
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 5cebd43ace2..5c97dc87e82 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -176,11 +176,12 @@ enum pipe_texture_target {
 #define PIPE_TEX_COMPARE_R_TO_TEXTURE  1
 
 #define PIPE_TEXTURE_USAGE_RENDER_TARGET   0x1
-#define PIPE_TEXTURE_USAGE_DISPLAY_TARGET  0x2 /* ie a backbuffer */
-#define PIPE_TEXTURE_USAGE_PRIMARY         0x4 /* ie a frontbuffer */
+#define PIPE_TEXTURE_USAGE_DISPLAY_TARGET  0x2  /* windows presentable buffer, ie a backbuffer */
+#define PIPE_TEXTURE_USAGE_SCANOUT         0x4  /* ie a frontbuffer */
 #define PIPE_TEXTURE_USAGE_DEPTH_STENCIL   0x8
 #define PIPE_TEXTURE_USAGE_SAMPLER         0x10
 #define PIPE_TEXTURE_USAGE_DYNAMIC         0x20
+#define PIPE_TEXTURE_USAGE_SHARED          0x40
 /** Pipe driver custom usage flags should be greater or equal to this value */
 #define PIPE_TEXTURE_USAGE_CUSTOM          (1 << 16)
 
diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h
index e4a92228093..b771bfe85ec 100644
--- a/src/gallium/include/pipe/p_screen.h
+++ b/src/gallium/include/pipe/p_screen.h
@@ -49,6 +49,8 @@ extern "C" {
 #endif
 
 
+/** Opaque type */
+struct winsys_handle;
 /** Opaque type */
 struct pipe_fence_handle;
 struct pipe_winsys;
@@ -108,16 +110,23 @@ struct pipe_screen {
                                            const struct pipe_texture *templat);
 
    /**
-    * Create a new texture object, using the given template info, but on top of
-    * existing memory.
-    * 
-    * It is assumed that the buffer data is layed out according to the expected
-    * by the hardware. NULL will be returned if any inconsistency is found.  
+    * Create a texture from a winsys_handle. The handle is often created in
+    * another process by first creating a pipe texture and then calling
+    * texture_get_handle.
+    */
+   struct pipe_texture * (*texture_from_handle)(struct pipe_screen *,
+                                                const struct pipe_texture *templat,
+                                                struct winsys_handle *handle);
+
+   /**
+    * Get a winsys_handle from a texture. Some platforms/winsys requires
+    * that the texture is created with a special usage flag like
+    * DISPLAYTARGET or PRIMARY.
     */
-   struct pipe_texture * (*texture_blanket)(struct pipe_screen *,
-                                            const struct pipe_texture *templat,
-                                            const unsigned *stride,
-                                            struct pipe_buffer *buffer);
+   boolean (*texture_get_handle)(struct pipe_screen *,
+                                 struct pipe_texture *tex,
+                                 struct winsys_handle *handle);
+
 
    void (*texture_destroy)(struct pipe_texture *pt);
 
@@ -187,23 +196,6 @@ struct pipe_screen {
                                              void *ptr,
                                              unsigned bytes);
 
-   /**
-    * Allocate storage for a display target surface.
-    *
-    * Often surfaces which are meant to be blitted to the front screen (i.e.,
-    * display targets) must be allocated with special characteristics, memory
-    * pools, or obtained directly from the windowing system.
-    *
-    * This callback is invoked by the pipe_screenwhen creating a texture marked
-    * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag  to get the underlying
-    * buffer storage.
-    */
-   struct pipe_buffer *(*surface_buffer_create)(struct pipe_screen *screen,
-						unsigned width, unsigned height,
-						enum pipe_format format,
-						unsigned usage,
-						unsigned tex_usage,
-						unsigned *stride);
 
 
    /**
@@ -273,6 +265,7 @@ struct pipe_screen {
 
    /**
     * Do any special operations to ensure buffer size is correct
+    * \param context_private  the private data of the calling context
     */
    void (*update_buffer)( struct pipe_screen *ws,
                           void *context_private );
@@ -280,10 +273,12 @@ struct pipe_screen {
    /**
     * Do any special operations to ensure frontbuffer contents are
     * displayed, eg copy fake frontbuffer.
+    * \param winsys_drawable_handle  an opaque handle that the calling context
+    *                                gets out-of-band
     */
    void (*flush_frontbuffer)( struct pipe_screen *screen,
                               struct pipe_surface *surf,
-                              void *context_private );
+                              void *winsys_drawable_handle );
 
 
 
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 02558520bfe..3a97d888ce6 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -249,7 +249,7 @@ struct pipe_framebuffer_state
 {
    unsigned width, height;
 
-   /** multiple colorbuffers for multiple render targets */
+   /** multiple color buffers for multiple render targets */
    unsigned nr_cbufs;
    struct pipe_surface *cbufs[PIPE_MAX_COLOR_BUFS];
 
@@ -285,12 +285,12 @@ struct pipe_sampler_state
 struct pipe_surface
 {
    struct pipe_reference reference;
-   enum pipe_format format;      /**< PIPE_FORMAT_x */
+   enum pipe_format format;
    unsigned width;               /**< logical width in pixels */
    unsigned height;              /**< logical height in pixels */
    unsigned layout;              /**< PIPE_SURFACE_LAYOUT_x */
    unsigned offset;              /**< offset from start of buffer, in bytes */
-   unsigned usage;               /**< PIPE_BUFFER_USAGE_*  */
+   unsigned usage;               /**< bitmask of PIPE_BUFFER_USAGE_x */
 
    unsigned zslice;
    struct pipe_texture *texture; /**< texture into which this is a view  */
@@ -336,7 +336,7 @@ struct pipe_texture
 
    unsigned nr_samples:8;    /**< for multisampled surfaces, nr of samples */
 
-   unsigned tex_usage;       /* PIPE_TEXTURE_USAGE_* */
+   unsigned tex_usage;       /**< bitmask of PIPE_TEXTURE_USAGE_* */
 
    struct pipe_screen *screen; /**< screen that this texture belongs to */
 };
@@ -373,9 +373,8 @@ struct pipe_vertex_element
     * this attribute live in?
     */
    unsigned vertex_buffer_index:8;
-   unsigned nr_components:8;
  
-   enum pipe_format src_format; 	   /**< PIPE_FORMAT_* */
+   enum pipe_format src_format;
 };
 
 
diff --git a/src/gallium/include/state_tracker/drm_api.h b/src/gallium/include/state_tracker/drm_api.h
index e9fa9b4d2a3..fe7ef253ef0 100644
--- a/src/gallium/include/state_tracker/drm_api.h
+++ b/src/gallium/include/state_tracker/drm_api.h
@@ -17,6 +17,32 @@ enum drm_create_screen_mode {
 	DRM_CREATE_MAX
 };
 
+#define DRM_API_HANDLE_TYPE_SHARED 0
+#define DRM_API_HANDLE_TYPE_KMS    1
+
+/**
+ * For use with pipe_screen::{texture_from_handle|texture_get_handle}.
+ */
+struct winsys_handle
+{
+	/**
+	 * Unused for texture_from_handle, always
+	 * DRM_API_HANDLE_TYPE_SHARED.  Input to texture_get_handle,
+	 * use TEXTURE_USAGE to select handle for kms or ipc.
+	 */
+	unsigned type;
+	/**
+	 * Input to texture_from_handle.
+	 * Output for texture_get_handle.
+	 */
+	unsigned handle;
+	/**
+	 * Input to texture_from_handle.
+	 * Output for texture_get_handle.
+	 */
+	unsigned stride;
+};
+
 /**
  * Modes other than DRM_CREATE_NORMAL derive from this struct.
  */
@@ -28,6 +54,8 @@ struct drm_create_screen_arg {
 
 struct drm_api
 {
+	void (*destroy)(struct drm_api *api);
+
         const char *name;
 
 	/**
@@ -36,37 +64,10 @@ struct drm_api
 	const char *driver_name;
 
 	/**
-	 * Special buffer functions
+	 * Create a pipe srcreen.
 	 */
-	/*@{*/
 	struct pipe_screen*  (*create_screen)(struct drm_api *api, int drm_fd,
 	                                      struct drm_create_screen_arg *arg);
-	/*@}*/
-
-	/**
-	 * Special buffer functions
-	 */
-	/*@{*/
-	struct pipe_texture*
-	    (*texture_from_shared_handle)(struct drm_api *api,
-	                                  struct pipe_screen *screen,
-	                                  struct pipe_texture *templ,
-	                                  const char *name,
-	                                  unsigned stride,
-	                                  unsigned handle);
-	boolean (*shared_handle_from_texture)(struct drm_api *api,
-	                                      struct pipe_screen *screen,
-	                                      struct pipe_texture *texture,
-	                                      unsigned *stride,
-	                                      unsigned *handle);
-	boolean (*local_handle_from_texture)(struct drm_api *api,
-	                                     struct pipe_screen *screen,
-	                                     struct pipe_texture *texture,
-	                                     unsigned *stride,
-	                                     unsigned *handle);
-	/*@}*/
-
-	void (*destroy)(struct drm_api *api);
 };
 
 extern struct drm_api * drm_api_create(void);
diff --git a/src/gallium/include/state_tracker/sw_winsys.h b/src/gallium/include/state_tracker/sw_winsys.h
new file mode 100644
index 00000000000..0de98bbc1c9
--- /dev/null
+++ b/src/gallium/include/state_tracker/sw_winsys.h
@@ -0,0 +1,124 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2009 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Software rasterizer winsys.
+ */
+
+
+#ifndef SW_WINSYS_H
+#define SW_WINSYS_H
+
+
+#include "pipe/p_compiler.h" /* for boolean */
+#include "pipe/p_format.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+struct pipe_screen;
+struct pipe_context;
+
+
+/**
+ * Opaque pointer.
+ */
+struct sw_displaytarget;
+
+
+/**
+ * This is the interface that sw expects any window system
+ * hosting it to implement.
+ * 
+ * sw is for the most part a self sufficient driver. The only thing it
+ * does not know is how to display a surface.
+ */
+struct sw_winsys
+{
+   void 
+   (*destroy)( struct sw_winsys *ws );
+
+   boolean
+   (*is_displaytarget_format_supported)( struct sw_winsys *ws,
+                                         enum pipe_format format );
+   
+   /**
+    * Allocate storage for a render target.
+    * 
+    * Often surfaces which are meant to be blitted to the front screen (i.e.,
+    * display targets) must be allocated with special characteristics, memory 
+    * pools, or obtained directly from the windowing system.
+    *  
+    * This callback is invoked by the pipe_screen when creating a texture marked
+    * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag to get the underlying 
+    * storage.
+    */
+   struct sw_displaytarget *
+   (*displaytarget_create)( struct sw_winsys *ws,
+                            enum pipe_format format,
+                            unsigned width, unsigned height,
+                            unsigned alignment,
+                            unsigned *stride );
+
+   /**
+    * \param flags  bitmask of PIPE_BUFFER_USAGE_x flags
+    */
+   void *
+   (*displaytarget_map)( struct sw_winsys *ws, 
+                         struct sw_displaytarget *dt,
+                         unsigned flags );
+
+   void
+   (*displaytarget_unmap)( struct sw_winsys *ws,
+                           struct sw_displaytarget *dt );
+
+   /**
+    * @sa pipe_screen:flush_frontbuffer.
+    *
+    * This call will likely become asynchronous eventually.
+    */
+   void
+   (*displaytarget_display)( struct sw_winsys *ws, 
+                             struct sw_displaytarget *dt,
+                             void *context_private );
+
+   void 
+   (*displaytarget_destroy)( struct sw_winsys *ws, 
+                             struct sw_displaytarget *dt );
+};
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* SW_WINSYS_H */
diff --git a/src/gallium/include/state_tracker/xlib_sw_winsys.h b/src/gallium/include/state_tracker/xlib_sw_winsys.h
new file mode 100644
index 00000000000..f22c22bb620
--- /dev/null
+++ b/src/gallium/include/state_tracker/xlib_sw_winsys.h
@@ -0,0 +1,29 @@
+#ifndef XLIB_SW_WINSYS_H
+#define XLIB_SW_WINSYS_H
+
+#include "state_tracker/sw_winsys.h"
+#include <X11/Xlib.h>
+
+
+struct pipe_screen;
+struct pipe_surface;
+
+/* This is what the xlib software winsys expects to find in the
+ * "private" field of flush_frontbuffers().
+ *
+ * Xlib-based state trackers somehow need to know this.
+ */
+struct xlib_drawable {
+   Visual *visual;
+   int depth;
+   Drawable drawable;
+};
+
+
+/* This is the public interface to the ws/xlib module.  Why isn't it
+ * being defined in that directory?
+ */
+struct sw_winsys *xlib_create_sw_winsys( Display *display );
+
+
+#endif
diff --git a/src/gallium/state_trackers/dri/dri_context.c b/src/gallium/state_trackers/dri/dri_context.c
index 908cef454e9..2f991c39e33 100644
--- a/src/gallium/state_trackers/dri/dri_context.c
+++ b/src/gallium/state_trackers/dri/dri_context.c
@@ -128,7 +128,7 @@ dri_unbind_context(__DRIcontext * cPriv)
       if (--ctx->bind_count == 0) {
 	 if (ctx->st && ctx->st == st_get_current()) {
 	    st_flush(ctx->st, PIPE_FLUSH_RENDER_CACHE, NULL);
-	    st_make_current(NULL, NULL, NULL);
+	    st_make_current(NULL, NULL, NULL, NULL);
 	 }
       }
    }
@@ -161,7 +161,13 @@ dri_make_current(__DRIcontext * cPriv,
 	 ctx->r_stamp = driReadPriv->lastStamp - 1;
       }
 
-      st_make_current(ctx->st, draw->stfb, read->stfb);
+      /* DRI co-state tracker currently overrides flush_frontbuffer.
+       * When this is fixed, will need to pass the drawable in the
+       * fourth parameter here so that when Mesa calls
+       * flush_frontbuffer directly (in front-buffer rendering), it
+       * will have access to the drawable argument:
+       */
+      st_make_current(ctx->st, draw->stfb, read->stfb, NULL);
 
       if (__dri1_api_hooks) {
 	 dri1_update_drawables(ctx, draw, read);
@@ -170,7 +176,7 @@ dri_make_current(__DRIcontext * cPriv,
 			   ctx->pipe->priv);
       }
    } else {
-      st_make_current(NULL, NULL, NULL);
+      st_make_current(NULL, NULL, NULL, NULL);
    }
 
    return GL_TRUE;
diff --git a/src/gallium/state_trackers/dri/dri_drawable.c b/src/gallium/state_trackers/dri/dri_drawable.c
index 173f4041c8c..458473853cf 100644
--- a/src/gallium/state_trackers/dri/dri_drawable.c
+++ b/src/gallium/state_trackers/dri/dri_drawable.c
@@ -58,6 +58,7 @@ dri_surface_from_handle(struct drm_api *api,
    struct pipe_surface *surface = NULL;
    struct pipe_texture *texture = NULL;
    struct pipe_texture templat;
+   struct winsys_handle whandle;
 
    memset(&templat, 0, sizeof(templat));
    templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET;
@@ -68,8 +69,11 @@ dri_surface_from_handle(struct drm_api *api,
    templat.width0 = width;
    templat.height0 = height;
 
-   texture = api->texture_from_shared_handle(api, screen, &templat,
-                                             "dri2 buffer", pitch, handle);
+   memset(&whandle, 0, sizeof(whandle));
+   whandle.handle = handle;
+   whandle.stride = pitch;
+
+   texture = screen->texture_from_handle(screen, &templat, &whandle);
 
    if (!texture) {
       debug_printf("%s: Failed to blanket the buffer with a texture\n", __func__);
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d.c b/src/gallium/state_trackers/egl/common/egl_g3d.c
index 50774b03f35..e4972d493d6 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.c
@@ -508,20 +508,6 @@ egl_g3d_add_configs(_EGLDriver *drv, _EGLDisplay *dpy, EGLint id)
    return id;
 }
 
-/**
- * Flush the front buffer of the context's draw surface.
- */
-static void
-egl_g3d_flush_frontbuffer(struct pipe_screen *screen,
-                          struct pipe_surface *surf, void *context_private)
-{
-   struct egl_g3d_context *gctx = egl_g3d_context(context_private);
-   struct egl_g3d_surface *gsurf = egl_g3d_surface(gctx->base.DrawSurface);
-
-   if (gsurf)
-      gsurf->native->flush_frontbuffer(gsurf->native);
-}
-
 /**
  * Re-validate the context.
  */
@@ -602,7 +588,6 @@ egl_g3d_initialize(_EGLDriver *drv, _EGLDisplay *dpy,
    }
 
    gdpy->native->user_data = (void *) dpy;
-   gdpy->native->screen->flush_frontbuffer = egl_g3d_flush_frontbuffer;
    gdpy->native->screen->update_buffer = egl_g3d_update_buffer;
 
    egl_g3d_init_st(&gdrv->base);
@@ -907,8 +892,13 @@ egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
    if (gctx) {
       ok = egl_g3d_realloc_context(dpy, &gctx->base);
       if (ok) {
+         /* XXX: need to pass the winsys argument for
+          * flush_frontbuffer in the fourth parameter here:
+          */
          ok = gctx->stapi->st_make_current(gctx->st_ctx,
-               gctx->draw.st_fb, gctx->read.st_fb);
+                                           gctx->draw.st_fb,
+                                           gctx->read.st_fb,
+                                           NULL);
          if (ok) {
             egl_g3d_validate_context(dpy, &gctx->base);
             if (gdraw->base.Type == EGL_WINDOW_BIT) {
@@ -920,7 +910,7 @@ egl_g3d_make_current(_EGLDriver *drv, _EGLDisplay *dpy,
       }
    }
    else if (old_gctx) {
-      ok = old_gctx->stapi->st_make_current(NULL, NULL, NULL);
+      ok = old_gctx->stapi->st_make_current(NULL, NULL, NULL, NULL);
       old_gctx->base.WindowRenderBuffer = EGL_NONE;
    }
 
diff --git a/src/gallium/state_trackers/egl/common/native.h b/src/gallium/state_trackers/egl/common/native.h
index 9c22ff3e432..93c81b26e10 100644
--- a/src/gallium/state_trackers/egl/common/native.h
+++ b/src/gallium/state_trackers/egl/common/native.h
@@ -140,9 +140,6 @@ struct native_config {
 struct native_display {
    /**
     * The pipe screen of the native display.
-    *
-    * Note that the "flush_frontbuffer" and "update_buffer" callbacks will be
-    * overridden.
     */
    struct pipe_screen *screen;
 
diff --git a/src/gallium/state_trackers/egl/common/st_public_tmp.h b/src/gallium/state_trackers/egl/common/st_public_tmp.h
index 507a0ec4027..562dd68c150 100644
--- a/src/gallium/state_trackers/egl/common/st_public_tmp.h
+++ b/src/gallium/state_trackers/egl/common/st_public_tmp.h
@@ -9,7 +9,7 @@ ST_PUBLIC(st_get_framebuffer_surface,     int,                       struct st_f
 ST_PUBLIC(st_get_framebuffer_texture,     int,                       struct st_framebuffer *stfb, uint surfIndex, struct pipe_texture **texture)
 ST_PUBLIC(st_framebuffer_private,         void *,                    struct st_framebuffer *stfb)
 ST_PUBLIC(st_unreference_framebuffer,     void,                      struct st_framebuffer *stfb)
-ST_PUBLIC(st_make_current,                GLboolean,                 struct st_context *st, struct st_framebuffer *draw, struct st_framebuffer *read)
+ST_PUBLIC(st_make_current,                GLboolean,                 struct st_context *st, struct st_framebuffer *draw, struct st_framebuffer *read, void *winsys_drawable_handle)
 ST_PUBLIC(st_get_current,                 struct st_context *,       void)
 ST_PUBLIC(st_flush,                       void,                      struct st_context *st, uint pipeFlushFlags, struct pipe_fence_handle **fence)
 ST_PUBLIC(st_finish,                      void,                      struct st_context *st)
diff --git a/src/gallium/state_trackers/egl/kms/native_kms.c b/src/gallium/state_trackers/egl/kms/native_kms.c
index 94588bfa743..73222408565 100644
--- a/src/gallium/state_trackers/egl/kms/native_kms.c
+++ b/src/gallium/state_trackers/egl/kms/native_kms.c
@@ -55,7 +55,7 @@ kms_surface_validate(struct native_surface *nsurf, uint attachment_mask,
       templ.format = ksurf->color_format;
       templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
       if (ksurf->type == KMS_SURFACE_TYPE_SCANOUT)
-         templ.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY;
+         templ.tex_usage |= PIPE_TEXTURE_USAGE_SCANOUT;
    }
 
    /* create textures */
@@ -100,7 +100,7 @@ kms_surface_init_framebuffers(struct native_surface *nsurf, boolean need_back)
    for (i = 0; i < num_framebuffers; i++) {
       struct kms_framebuffer *fb;
       enum native_attachment natt;
-      unsigned int handle, stride;
+      struct winsys_handle whandle;
       uint block_bits;
 
       if (i == 0) {
@@ -128,13 +128,17 @@ kms_surface_init_framebuffers(struct native_surface *nsurf, boolean need_back)
       /* TODO detect the real value */
       fb->is_passive = TRUE;
 
-      if (!kdpy->api->local_handle_from_texture(kdpy->api,
-               kdpy->base.screen, fb->texture, &stride, &handle))
+      memset(&whandle, 0, sizeof(whandle));
+      whandle.type = DRM_API_HANDLE_TYPE_KMS;
+
+      if (!kdpy->base.screen->texture_get_handle(kdpy->base.screen,
+               fb->texture, &whandle))
          return FALSE;
 
       block_bits = util_format_get_blocksizebits(ksurf->color_format);
       err = drmModeAddFB(kdpy->fd, ksurf->width, ksurf->height,
-            block_bits, block_bits, stride, handle, &fb->buffer_id);
+            block_bits, block_bits, whandle.stride, whandle.handle,
+            &fb->buffer_id);
       if (err) {
          fb->buffer_id = 0;
          return FALSE;
diff --git a/src/gallium/state_trackers/egl/x11/native_dri2.c b/src/gallium/state_trackers/egl/x11/native_dri2.c
index 8d2a8b1dffe..98399792311 100644
--- a/src/gallium/state_trackers/egl/x11/native_dri2.c
+++ b/src/gallium/state_trackers/egl/x11/native_dri2.c
@@ -114,6 +114,7 @@ dri2_surface_process_drawable_buffers(struct native_surface *nsurf,
    struct dri2_surface *dri2surf = dri2_surface(nsurf);
    struct dri2_display *dri2dpy = dri2surf->dri2dpy;
    struct pipe_texture templ;
+   struct winsys_handle whandle;
    uint valid_mask;
    int i;
 
@@ -171,9 +172,11 @@ dri2_surface_process_drawable_buffers(struct native_surface *nsurf,
          continue;
       }
 
-      dri2surf->textures[natt] =
-         dri2dpy->api->texture_from_shared_handle(dri2dpy->api,
-               dri2dpy->base.screen, &templ, desc, xbuf->pitch, xbuf->name);
+      memset(&whandle, 0, sizeof(whandle));
+      whandle.stride = xbuf->pitch;
+      whandle.handle = xbuf->name;
+      dri2surf->textures[natt] = dri2dpy->base.screen->texture_from_handle(
+         dri2dpy->base.screen, &templ, &whandle);
       if (dri2surf->textures[natt])
          valid_mask |= 1 << natt;
    }
diff --git a/src/gallium/state_trackers/egl/x11/native_x11.c b/src/gallium/state_trackers/egl/x11/native_x11.c
index 7b4fe63fa07..c6eb17ab1ac 100644
--- a/src/gallium/state_trackers/egl/x11/native_x11.c
+++ b/src/gallium/state_trackers/egl/x11/native_x11.c
@@ -142,16 +142,9 @@ native_create_display(EGLNativeDisplayType dpy,
 
    if (!ndpy) {
       EGLint level = (force_sw) ? _EGL_INFO : _EGL_WARNING;
-      boolean use_shm;
-
-      /*
-       * XXX st/mesa calls pipe_screen::update_buffer in st_validate_state.
-       * When SHM is used, there is a good chance that the shared memory
-       * segment is detached before the softpipe tile cache is flushed.
-       */
-      use_shm = FALSE;
-      _eglLog(level, "use software%s fallback", (use_shm) ? " (SHM)" : "");
-      ndpy = x11_create_ximage_display(dpy, event_handler, use_shm);
+
+      _eglLog(level, "use software fallback");
+      ndpy = x11_create_ximage_display(dpy, event_handler);
    }
 
    return ndpy;
diff --git a/src/gallium/state_trackers/egl/x11/native_x11.h b/src/gallium/state_trackers/egl/x11/native_x11.h
index 8c6a7d93497..1566524926a 100644
--- a/src/gallium/state_trackers/egl/x11/native_x11.h
+++ b/src/gallium/state_trackers/egl/x11/native_x11.h
@@ -30,8 +30,7 @@
 
 struct native_display *
 x11_create_ximage_display(EGLNativeDisplayType dpy,
-                          struct native_event_handler *event_handler,
-                          boolean use_xshm);
+                          struct native_event_handler *event_handler);
 
 struct native_display *
 x11_create_dri2_display(EGLNativeDisplayType dpy,
diff --git a/src/gallium/state_trackers/egl/x11/native_ximage.c b/src/gallium/state_trackers/egl/x11/native_ximage.c
index 3421c1951aa..e0d12acabe8 100644
--- a/src/gallium/state_trackers/egl/x11/native_ximage.c
+++ b/src/gallium/state_trackers/egl/x11/native_ximage.c
@@ -28,17 +28,19 @@
 #include <sys/shm.h>
 #include <X11/Xlib.h>
 #include <X11/Xutil.h>
-#include <X11/extensions/XShm.h>
 #include "util/u_memory.h"
 #include "util/u_math.h"
 #include "util/u_format.h"
 #include "pipe/p_compiler.h"
-#include "util/u_simple_screen.h"
 #include "util/u_inlines.h"
-#include "softpipe/sp_winsys.h"
+#include "state_tracker/xlib_sw_winsys.h"
+#include "target-helpers/wrap_screen.h"
+#include "util/u_debug.h"
+#include "softpipe/sp_public.h"
+#include "llvmpipe/lp_public.h"
+#include "cell/ppu/cell_public.h"
 #include "egllog.h"
 
-#include "sw_winsys.h"
 #include "native_x11.h"
 #include "x11_screen.h"
 
@@ -53,24 +55,18 @@ struct ximage_display {
    Display *dpy;
    boolean own_dpy;
 
-   struct x11_screen *xscr;
-   int xscr_number;
-
    struct native_event_handler *event_handler;
 
-   boolean use_xshm;
+   struct x11_screen *xscr;
+   int xscr_number;
 
-   struct pipe_winsys *winsys;
    struct ximage_config *configs;
    int num_configs;
 };
 
 struct ximage_buffer {
-   XImage *ximage;
-
    struct pipe_texture *texture;
-   XShmSegmentInfo *shm_info;
-   boolean xshm_attached;
+   struct xlib_drawable xdraw;
 };
 
 struct ximage_surface {
@@ -81,8 +77,6 @@ struct ximage_surface {
    XVisualInfo visual;
    struct ximage_display *xdpy;
 
-   GC gc;
-
    unsigned int server_stamp;
    unsigned int client_stamp;
    int width, height;
@@ -121,18 +115,6 @@ ximage_surface_free_buffer(struct native_surface *nsurf,
    struct ximage_buffer *xbuf = &xsurf->buffers[which];
 
    pipe_texture_reference(&xbuf->texture, NULL);
-
-   if (xbuf->shm_info) {
-      if (xbuf->xshm_attached)
-         XShmDetach(xsurf->xdpy->dpy, xbuf->shm_info);
-      if (xbuf->shm_info->shmaddr != (void *) -1)
-         shmdt(xbuf->shm_info->shmaddr);
-      if (xbuf->shm_info->shmid != -1)
-         shmctl(xbuf->shm_info->shmid, IPC_RMID, 0);
-
-      xbuf->shm_info->shmaddr = (void *) -1;
-      xbuf->shm_info->shmid = -1;
-   }
 }
 
 static boolean
@@ -156,40 +138,25 @@ ximage_surface_alloc_buffer(struct native_surface *nsurf,
    templ.depth0 = 1;
    templ.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET;
 
-   if (xbuf->shm_info) {
-      struct pipe_buffer *pbuf;
-      unsigned stride, size;
-      void *addr = NULL;
-
-      stride = util_format_get_stride(xsurf->color_format, xsurf->width);
-      /* alignment should depend on visual? */
-      stride = align(stride, 4);
-      size = stride * xsurf->height;
-
-      /* create and attach shm object */
-      xbuf->shm_info->shmid = shmget(IPC_PRIVATE, size, 0755);
-      if (xbuf->shm_info->shmid != -1) {
-         xbuf->shm_info->shmaddr =
-            shmat(xbuf->shm_info->shmid, NULL, 0);
-         if (xbuf->shm_info->shmaddr != (void *) -1) {
-            if (XShmAttach(xsurf->xdpy->dpy, xbuf->shm_info)) {
-               addr = xbuf->shm_info->shmaddr;
-               xbuf->xshm_attached = TRUE;
-            }
-         }
-      }
-
-      if (addr) {
-         pbuf = screen->user_buffer_create(screen, addr, size);
-         if (pbuf) {
-            xbuf->texture =
-               screen->texture_blanket(screen, &templ, &stride, pbuf);
-            pipe_buffer_reference(&pbuf, NULL);
-         }
+   if (xsurf->type != XIMAGE_SURFACE_TYPE_PBUFFER) {
+      switch (which) {
+      case NATIVE_ATTACHMENT_FRONT_LEFT:
+      case NATIVE_ATTACHMENT_FRONT_RIGHT:
+         templ.tex_usage |= PIPE_TEXTURE_USAGE_SCANOUT;
+         break;
+      case NATIVE_ATTACHMENT_BACK_LEFT:
+      case NATIVE_ATTACHMENT_BACK_RIGHT:
+         templ.tex_usage |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+         break;
+      default:
+         break;
       }
    }
-   else {
-      xbuf->texture = screen->texture_create(screen, &templ);
+   xbuf->texture = screen->texture_create(screen, &templ);
+   if (xbuf->texture) {
+      xbuf->xdraw.visual = xsurf->visual.visual;
+      xbuf->xdraw.depth = xsurf->visual.depth;
+      xbuf->xdraw.drawable = xsurf->drawable;
    }
 
    /* clean up the buffer if allocation failed */
@@ -269,18 +236,10 @@ ximage_surface_update_buffers(struct native_surface *nsurf, uint buffer_mask)
    new_valid = 0x0;
    for (att = 0; att < NUM_NATIVE_ATTACHMENTS; att++) {
       if (native_attachment_mask_test(buffer_mask, att)) {
-         struct ximage_buffer *xbuf = &xsurf->buffers[att];
-
          /* reallocate the texture */
          if (!ximage_surface_alloc_buffer(&xsurf->base, att))
             break;
 
-         /* update ximage */
-         if (xbuf->ximage) {
-            xbuf->ximage->width = xsurf->width;
-            xbuf->ximage->height = xsurf->height;
-         }
-
          new_valid |= (1 << att);
          if (buffer_mask == new_valid)
             break;
@@ -300,43 +259,22 @@ ximage_surface_draw_buffer(struct native_surface *nsurf,
    struct ximage_surface *xsurf = ximage_surface(nsurf);
    struct ximage_buffer *xbuf = &xsurf->buffers[which];
    struct pipe_screen *screen = xsurf->xdpy->base.screen;
-   struct pipe_transfer *transfer;
+   struct pipe_surface *psurf;
 
    if (xsurf->type == XIMAGE_SURFACE_TYPE_PBUFFER)
       return TRUE;
 
-   assert(xsurf->drawable && xbuf->ximage && xbuf->texture);
+   assert(xsurf->drawable && xbuf->texture);
 
-   transfer = screen->get_tex_transfer(screen, xbuf->texture,
-         0, 0, 0, PIPE_TRANSFER_READ, 0, 0, xsurf->width, xsurf->height);
-   if (!transfer)
+   /* what's the cost of surface creation? */
+   psurf = screen->get_tex_surface(screen,
+         xbuf->texture, 0, 0, 0, PIPE_BUFFER_USAGE_CPU_READ);
+   if (!psurf)
       return FALSE;
 
-   xbuf->ximage->bytes_per_line = transfer->stride;
-   xbuf->ximage->data = screen->transfer_map(screen, transfer);
-   if (!xbuf->ximage->data) {
-      screen->tex_transfer_destroy(transfer);
-      return FALSE;
-   }
-
+   screen->flush_frontbuffer(screen, psurf, &xbuf->xdraw);
 
-   if (xbuf->shm_info)
-      XShmPutImage(xsurf->xdpy->dpy, xsurf->drawable, xsurf->gc,
-            xbuf->ximage, 0, 0, 0, 0, xsurf->width, xsurf->height, False);
-   else
-      XPutImage(xsurf->xdpy->dpy, xsurf->drawable, xsurf->gc,
-            xbuf->ximage, 0, 0, 0, 0, xsurf->width, xsurf->height);
-
-   xbuf->ximage->data = NULL;
-   screen->transfer_unmap(screen, transfer);
-
-   /*
-    * softpipe allows the pipe transfer to be re-used, but we don't want to
-    * rely on that behavior.
-    */
-   screen->tex_transfer_destroy(transfer);
-
-   XSync(xsurf->xdpy->dpy, FALSE);
+   pipe_surface_reference(&psurf, NULL);
 
    return TRUE;
 }
@@ -364,7 +302,8 @@ ximage_surface_swap_buffers(struct native_surface *nsurf)
    boolean ret;
 
    /* display the back buffer first */
-   ret = ximage_surface_draw_buffer(nsurf, NATIVE_ATTACHMENT_BACK_LEFT);
+   ret = ximage_surface_draw_buffer(&xsurf->base,
+         NATIVE_ATTACHMENT_BACK_LEFT);
    /* force buffers to be updated in next validation call */
    xsurf->server_stamp++;
    ximage_surface_notify_invalid(&xsurf->base);
@@ -372,13 +311,12 @@ ximage_surface_swap_buffers(struct native_surface *nsurf)
    xfront = &xsurf->buffers[NATIVE_ATTACHMENT_FRONT_LEFT];
    xback = &xsurf->buffers[NATIVE_ATTACHMENT_BACK_LEFT];
 
-   /* skip swapping so that the front buffer is allocated only when needed */
-   if (!xfront->texture)
-      return ret;
-
-   xtmp = *xfront;
-   *xfront = *xback;
-   *xback = xtmp;
+   /* skip swapping unless there is a front buffer */
+   if (xfront->texture) {
+      xtmp = *xfront;
+      *xfront = *xback;
+      *xback = xtmp;
+   }
 
    return ret;
 }
@@ -433,18 +371,9 @@ ximage_surface_destroy(struct native_surface *nsurf)
    struct ximage_surface *xsurf = ximage_surface(nsurf);
    int i;
 
-   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
-      struct ximage_buffer *xbuf = &xsurf->buffers[i];
+   for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++)
       ximage_surface_free_buffer(&xsurf->base, i);
-      /* xbuf->shm_info is owned by xbuf->ximage? */
-      if (xbuf->ximage) {
-         XDestroyImage(xbuf->ximage);
-         xbuf->ximage = NULL;
-      }
-   }
 
-   if (xsurf->type != XIMAGE_SURFACE_TYPE_PBUFFER)
-      XFreeGC(xsurf->xdpy->dpy, xsurf->gc);
    free(xsurf);
 }
 
@@ -457,7 +386,6 @@ ximage_display_create_surface(struct native_display *ndpy,
    struct ximage_display *xdpy = ximage_display(ndpy);
    struct ximage_config *xconf = ximage_config(nconf);
    struct ximage_surface *xsurf;
-   int i;
 
    xsurf = CALLOC_STRUCT(ximage_surface);
    if (!xsurf)
@@ -471,52 +399,8 @@ ximage_display_create_surface(struct native_display *ndpy,
    if (xsurf->type != XIMAGE_SURFACE_TYPE_PBUFFER) {
       xsurf->drawable = drawable;
       xsurf->visual = *xconf->visual;
-
-      xsurf->gc = XCreateGC(xdpy->dpy, xsurf->drawable, 0, NULL);
-      if (!xsurf->gc) {
-         free(xsurf);
-         return NULL;
-      }
-
       /* initialize the geometry */
       ximage_surface_update_buffers(&xsurf->base, 0x0);
-
-      for (i = 0; i < NUM_NATIVE_ATTACHMENTS; i++) {
-         struct ximage_buffer *xbuf = &xsurf->buffers[i];
-
-         if (xdpy->use_xshm) {
-            xbuf->shm_info = calloc(1, sizeof(*xbuf->shm_info));
-            if (xbuf->shm_info) {
-               /* initialize shm info */
-               xbuf->shm_info->shmid = -1;
-               xbuf->shm_info->shmaddr = (void *) -1;
-               xbuf->shm_info->readOnly = TRUE;
-
-               xbuf->ximage = XShmCreateImage(xsurf->xdpy->dpy,
-                     xsurf->visual.visual,
-                     xsurf->visual.depth,
-                     ZPixmap, NULL,
-                     xbuf->shm_info,
-                     0, 0);
-            }
-         }
-         else {
-            xbuf->ximage = XCreateImage(xsurf->xdpy->dpy,
-                  xsurf->visual.visual,
-                  xsurf->visual.depth,
-                  ZPixmap, 0,   /* format, offset */
-                  NULL,         /* data */
-                  0, 0,         /* size */
-                  8,            /* bitmap_pad */
-                  0);           /* bytes_per_line */
-         }
-
-         if (!xbuf->ximage) {
-            XFreeGC(xdpy->dpy, xsurf->gc);
-            free(xsurf);
-            return NULL;
-         }
-      }
    }
 
    xsurf->base.destroy = ximage_surface_destroy;
@@ -727,7 +611,6 @@ ximage_display_destroy(struct native_display *ndpy)
       free(xdpy->configs);
 
    xdpy->base.screen->destroy(xdpy->base.screen);
-   free(xdpy->winsys);
 
    x11_screen_destroy(xdpy->xscr);
    if (xdpy->own_dpy)
@@ -735,10 +618,63 @@ ximage_display_destroy(struct native_display *ndpy)
    free(xdpy);
 }
 
+
+/* Helper function to build a subset of a driver stack consisting of
+ * one of the software rasterizers (cell, llvmpipe, softpipe) and the
+ * xlib winsys.
+ *
+ * This function could be shared, but currently causes headaches for
+ * the build systems, particularly scons if we try.
+ *
+ * Long term, want to avoid having global #defines for things like
+ * GALLIUM_LLVMPIPE, GALLIUM_CELL, etc.  Scons already eliminates
+ * those #defines, so things that are painful for it now are likely to
+ * be painful for other build systems in the future.
+ */
+static struct pipe_screen *
+swrast_xlib_create_screen( Display *display )
+{
+   struct sw_winsys *winsys;
+   struct pipe_screen *screen = NULL;
+
+   /* Create the underlying winsys, which performs presents to Xlib
+    * drawables:
+    */
+   winsys = xlib_create_sw_winsys( display );
+   if (winsys == NULL)
+      return NULL;
+
+   /* Create a software rasterizer on top of that winsys.  Use
+    * llvmpipe if it is available.
+    */
+#if defined(GALLIUM_LLVMPIPE)
+   if (screen == NULL &&
+       !debug_get_bool_option("GALLIUM_NO_LLVM", FALSE))
+      screen = llvmpipe_create_screen( winsys );
+#endif
+
+   if (screen == NULL)
+      screen = softpipe_create_screen( winsys );
+
+   if (screen == NULL)
+      goto fail;
+
+   /* Inject any wrapping layers we want to here:
+    */
+   return gallium_wrap_screen( screen );
+
+fail:
+   if (winsys)
+      winsys->destroy( winsys );
+
+   return NULL;
+}
+
+
+
 struct native_display *
 x11_create_ximage_display(EGLNativeDisplayType dpy,
-                          struct native_event_handler *event_handler,
-                          boolean use_xshm)
+                          struct native_event_handler *event_handler)
 {
    struct ximage_display *xdpy;
 
@@ -756,6 +692,8 @@ x11_create_ximage_display(EGLNativeDisplayType dpy,
       xdpy->own_dpy = TRUE;
    }
 
+   xdpy->event_handler = event_handler;
+
    xdpy->xscr_number = DefaultScreen(xdpy->dpy);
    xdpy->xscr = x11_screen_create(xdpy->dpy, xdpy->xscr_number);
    if (!xdpy->xscr) {
@@ -763,13 +701,7 @@ x11_create_ximage_display(EGLNativeDisplayType dpy,
       return NULL;
    }
 
-   xdpy->event_handler = event_handler;
-
-   xdpy->use_xshm =
-      (use_xshm && x11_screen_support(xdpy->xscr, X11_SCREEN_EXTENSION_XSHM));
-
-   xdpy->winsys = create_sw_winsys();
-   xdpy->base.screen = softpipe_create_screen(xdpy->winsys);
+   xdpy->base.screen = swrast_xlib_create_screen(xdpy->dpy);
 
    xdpy->base.destroy = ximage_display_destroy;
    xdpy->base.get_param = ximage_display_get_param;
diff --git a/src/gallium/state_trackers/egl/x11/sw_winsys.c b/src/gallium/state_trackers/egl/x11/sw_winsys.c
deleted file mode 100644
index 33328aadf26..00000000000
--- a/src/gallium/state_trackers/egl/x11/sw_winsys.c
+++ /dev/null
@@ -1,231 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-/**
- * Totally software-based winsys layer.
- * Note that the one winsys function that we can't implement here
- * is flush_frontbuffer().
- * Whoever uses this code will have to provide that.
- *
- * Authors: Brian Paul
- */
-
-
-#include "util/u_simple_screen.h"
-#include "pipe/p_state.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-
-#include "sw_winsys.h"
-
-
-
-/** Subclass of pipe_winsys */
-struct sw_pipe_winsys
-{
-   struct pipe_winsys Base;
-   /* no extra fields for now */
-};
-
-
-/** subclass of pipe_buffer */
-struct sw_pipe_buffer
-{
-   struct pipe_buffer Base;
-   boolean UserBuffer;  /** Is this a user-space buffer? */
-   void *Data;
-   void *Mapped;
-};
-
-
-/** cast wrapper */
-static INLINE struct sw_pipe_buffer *
-sw_pipe_buffer(struct pipe_buffer *b)
-{
-   return (struct sw_pipe_buffer *) b;
-}
-
-
-static const char *
-get_name(struct pipe_winsys *pws)
-{
-   return "software";
-}
-
-
-/** Create new pipe_buffer and allocate storage of given size */
-static struct pipe_buffer *
-buffer_create(struct pipe_winsys *pws, 
-              unsigned alignment, 
-              unsigned usage,
-              unsigned size)
-{
-   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
-   if (!buffer)
-      return NULL;
-
-   pipe_reference_init(&buffer->Base.reference, 1);
-   buffer->Base.alignment = alignment;
-   buffer->Base.usage = usage;
-   buffer->Base.size = size;
-
-   /* align to 16-byte multiple for Cell */
-   buffer->Data = align_malloc(size, MAX2(alignment, 16));
-
-   return &buffer->Base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
-{
-   struct sw_pipe_buffer *buffer = CALLOC_STRUCT(sw_pipe_buffer);
-   if (!buffer)
-      return NULL;
-
-   pipe_reference_init(&buffer->Base.reference, 1);
-   buffer->Base.size = bytes;
-   buffer->UserBuffer = TRUE;
-   buffer->Data = ptr;
-
-   return &buffer->Base;
-}
-
-
-static void *
-buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf, unsigned flags)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-   buffer->Mapped = buffer->Data;
-   return buffer->Mapped;
-}
-
-
-static void
-buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-   buffer->Mapped = NULL;
-}
-
-
-static void
-buffer_destroy(struct pipe_buffer *buf)
-{
-   struct sw_pipe_buffer *buffer = sw_pipe_buffer(buf);
-
-   if (buffer->Data && !buffer->UserBuffer) {
-      align_free(buffer->Data);
-      buffer->Data = NULL;
-   }
-
-   free(buffer);
-}
-
-
-static struct pipe_buffer *
-surface_buffer_create(struct pipe_winsys *winsys,
-                      unsigned width, unsigned height,
-                      enum pipe_format format, 
-                      unsigned usage,
-                      unsigned tex_usage,
-                      unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-fence_reference(struct pipe_winsys *sws, struct pipe_fence_handle **ptr,
-                struct pipe_fence_handle *fence)
-{
-   /* no-op */
-}
-
-
-static int
-fence_signalled(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                unsigned flag)
-{
-   /* no-op */
-   return 0;
-}
-
-
-static int
-fence_finish(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-             unsigned flag)
-{
-   /* no-op */
-   return 0;
-}
-
-
-/**
- * Create/return a new pipe_winsys object.
- */
-struct pipe_winsys *
-create_sw_winsys(void)
-{
-   struct sw_pipe_winsys *ws = CALLOC_STRUCT(sw_pipe_winsys);
-   if (!ws)
-      return NULL;
-
-   /* Fill in this struct with callbacks that pipe will need to
-    * communicate with the window system, buffer manager, etc. 
-    */
-   ws->Base.buffer_create = buffer_create;
-   ws->Base.user_buffer_create = user_buffer_create;
-   ws->Base.buffer_map = buffer_map;
-   ws->Base.buffer_unmap = buffer_unmap;
-   ws->Base.buffer_destroy = buffer_destroy;
-
-   ws->Base.surface_buffer_create = surface_buffer_create;
-
-   ws->Base.fence_reference = fence_reference;
-   ws->Base.fence_signalled = fence_signalled;
-   ws->Base.fence_finish = fence_finish;
-
-   ws->Base.flush_frontbuffer = NULL; /* not implemented here! */
-
-   ws->Base.get_name = get_name;
-
-   return &ws->Base;
-}
diff --git a/src/gallium/state_trackers/egl/x11/sw_winsys.h b/src/gallium/state_trackers/egl/x11/sw_winsys.h
deleted file mode 100644
index f96c5a14b0a..00000000000
--- a/src/gallium/state_trackers/egl/x11/sw_winsys.h
+++ /dev/null
@@ -1,40 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-
-#ifndef SW_WINSYS_H
-#define SW_WINSYS_H
-
-
-struct pipe_winsys;
-
-
-extern struct pipe_winsys *
-create_sw_winsys(void);
-
-
-#endif /* SW_WINSYS_H */
diff --git a/src/gallium/state_trackers/glx/xlib/SConscript b/src/gallium/state_trackers/glx/xlib/SConscript
index fa96df357d5..bb202351509 100644
--- a/src/gallium/state_trackers/glx/xlib/SConscript
+++ b/src/gallium/state_trackers/glx/xlib/SConscript
@@ -13,8 +13,6 @@ if env['platform'] == 'linux' \
         '#/src/mesa/main',
     ])
 
-    env.Append(CPPDEFINES = ['USE_XSHM'])
-
     st_xlib = env.ConvenienceLibrary(
 	target = 'st_xlib',
 	source = [
diff --git a/src/gallium/state_trackers/glx/xlib/glx_api.c b/src/gallium/state_trackers/glx/xlib/glx_api.c
index 08bf624b5c1..24545858500 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_api.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_api.c
@@ -689,7 +689,7 @@ choose_visual( Display *dpy, int screen, const int *list, GLboolean fbConfig )
    int desiredVisualID = -1;
    int numAux = 0;
 
-   xmesa_init();
+   xmesa_init( dpy );
 
    parselist = list;
 
diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.c b/src/gallium/state_trackers/glx/xlib/xm_api.c
index 217bdeff75e..f4d7133d2ff 100644
--- a/src/gallium/state_trackers/glx/xlib/xm_api.c
+++ b/src/gallium/state_trackers/glx/xlib/xm_api.c
@@ -35,10 +35,6 @@
  * corner of the window.  Therefore, most drawing functions in this
  * file have to flip Y coordinates.
  *
- * Define USE_XSHM in the Makefile with -DUSE_XSHM if you want to compile
- * in support for the MIT Shared Memory extension.  If enabled, when you
- * use an Ximage for the back buffer in double buffered mode, the "swap"
- * operation will be faster.  You must also link with -lXext.
  *
  * Byte swapping:  If the Mesa host and the X display use a different
  * byte order then there's some trickiness to be aware of when using
@@ -67,11 +63,7 @@
 #include "pipe/p_screen.h"
 #include "pipe/p_context.h"
 
-#include "trace/tr_screen.h"
-#include "trace/tr_context.h"
-#include "trace/tr_texture.h"
-
-#include "xm_winsys.h"
+#include "xm_public.h"
 #include <GL/glx.h>
 
 
@@ -91,7 +83,6 @@ void xmesa_set_driver( const struct xm_driver *templ )
  */
 pipe_mutex _xmesa_lock;
 
-static struct pipe_screen *_screen = NULL;
 static struct pipe_screen *screen = NULL;
 
 
@@ -111,41 +102,6 @@ static int host_byte_order( void )
 }
 
 
-/**
- * Check if the X Shared Memory extension is available.
- * Return:  0 = not available
- *          1 = shared XImage support available
- *          2 = shared Pixmap support available also
- */
-int xmesa_check_for_xshm( Display *display )
-{
-#if defined(USE_XSHM)
-   int major, minor, ignore;
-   Bool pixmaps;
-
-   if (getenv("SP_NO_RAST")) 
-      return 0;
-
-   if (getenv("MESA_NOSHM")) {
-      return 0;
-   }
-
-   if (XQueryExtension( display, "MIT-SHM", &ignore, &ignore, &ignore )) {
-      if (XShmQueryVersion( display, &major, &minor, &pixmaps )==True) {
-	 return (pixmaps==True) ? 2 : 1;
-      }
-      else {
-	 return 0;
-      }
-   }
-   else {
-      return 0;
-   }
-#else
-   /* No  XSHM support */
-   return 0;
-#endif
-}
 
 
 /**
@@ -242,7 +198,7 @@ xmesa_get_window_size(Display *dpy, XMesaBuffer b,
 
    pipe_mutex_lock(_xmesa_lock);
    XSync(b->xm_visual->display, 0); /* added for Chromium */
-   stat = get_drawable_size(dpy, b->drawable, width, height);
+   stat = get_drawable_size(dpy, b->ws.drawable, width, height);
    pipe_mutex_unlock(_xmesa_lock);
 
    if (!stat) {
@@ -397,7 +353,9 @@ create_xmesa_buffer(Drawable d, BufferType type,
    if (!b)
       return NULL;
 
-   b->drawable = d;
+   b->ws.drawable = d;
+   b->ws.visual = vis->visinfo->visual;
+   b->ws.depth = vis->visinfo->depth;
 
    b->xm_visual = vis;
    b->type = type;
@@ -422,18 +380,6 @@ create_xmesa_buffer(Drawable d, BufferType type,
                                    (void *) b);
    fb = &b->stfb->Base;
 
-   /*
-    * Create scratch XImage for xmesa_display_surface()
-    */
-   b->tempImage = XCreateImage(vis->display,
-                               vis->visinfo->visual,
-                               vis->visinfo->depth,
-                               ZPixmap, 0,   /* format, offset */
-                               NULL,         /* data */
-                               0, 0,         /* size */
-                               32,           /* bitmap_pad */
-                               0);           /* bytes_per_line */
-
    /* GLX_EXT_texture_from_pixmap */
    b->TextureTarget = 0;
    b->TextureFormat = GLX_TEXTURE_FORMAT_NONE_EXT;
@@ -490,16 +436,11 @@ xmesa_free_buffer(XMesaBuffer buffer)
          /* Since the X window for the XMesaBuffer is going away, we don't
           * want to dereference this pointer in the future.
           */
-         b->drawable = 0;
-
-         buffer->tempImage->data = NULL;
-         XDestroyImage(buffer->tempImage);
+         b->ws.drawable = 0;
 
          /* Unreference.  If count = zero we'll really delete the buffer */
          _mesa_reference_framebuffer(&fb, NULL);
 
-         XFreeGC(b->xm_visual->display, b->gc);
-
          free(buffer);
 
          return;
@@ -576,21 +517,6 @@ initialize_visual_and_buffer(XMesaVisual v, XMesaBuffer b,
       printf("X/Mesa bits per pixel = %d\n", v->BitsPerPixel);
    }
 
-   if (b && window) {
-      /* these should have been set in create_xmesa_buffer */
-      ASSERT(b->drawable == window);
-
-      /* Setup for single/double buffering */
-      if (v->mesa_visual.doubleBufferMode) {
-         /* Double buffered */
-         b->shm = xmesa_check_for_xshm( v->display );
-      }
-
-      /* X11 graphics context */
-      b->gc = XCreateGC( v->display, window, 0, NULL );
-      XSetFunction( v->display, b->gc, GXcopy );
-   }
-
    return GL_TRUE;
 }
 
@@ -673,7 +599,7 @@ XMesaVisual XMesaCreateVisual( Display *display,
    XMesaVisual v;
    GLint red_bits, green_bits, blue_bits, alpha_bits;
 
-   xmesa_init();
+   xmesa_init( display );
 
    /* For debugging only */
    if (_mesa_getenv("MESA_XSYNC")) {
@@ -773,13 +699,12 @@ void XMesaDestroyVisual( XMesaVisual v )
  * Do one-time initializations.
  */
 void
-xmesa_init(void)
+xmesa_init( Display *display )
 {
    static GLboolean firstTime = GL_TRUE;
    if (firstTime) {
       pipe_mutex_init(_xmesa_lock);
-      _screen = driver.create_pipe_screen();
-      screen = trace_screen_create( _screen );
+      screen = driver.create_pipe_screen( display );
       firstTime = GL_FALSE;
    }
 }
@@ -800,7 +725,7 @@ XMesaContext XMesaCreateContext( XMesaVisual v, XMesaContext share_list )
    GLcontext *mesaCtx;
    uint pf;
 
-   xmesa_init();
+   xmesa_init( v->display );
 
    /* Note: the XMesaContext contains a Mesa GLcontext struct (inheritance) */
    c = (XMesaContext) CALLOC_STRUCT(xmesa_context);
@@ -1095,7 +1020,8 @@ GLboolean XMesaMakeCurrent2( XMesaContext c, XMesaBuffer drawBuffer,
       c->xm_buffer = drawBuffer;
       c->xm_read_buffer = readBuffer;
 
-      st_make_current(c->st, drawBuffer->stfb, readBuffer->stfb);
+      st_make_current(c->st, drawBuffer->stfb, readBuffer->stfb, 
+                      &drawBuffer->ws);
 
       xmesa_check_and_update_buffer_size(c, drawBuffer);
       if (readBuffer != drawBuffer)
@@ -1106,7 +1032,7 @@ GLboolean XMesaMakeCurrent2( XMesaContext c, XMesaBuffer drawBuffer,
    }
    else {
       /* Detach */
-      st_make_current( NULL, NULL, NULL );
+      st_make_current( NULL, NULL, NULL, NULL );
 
    }
    return GL_TRUE;
@@ -1149,13 +1075,9 @@ void XMesaSwapBuffers( XMesaBuffer b )
    st_swapbuffers(b->stfb, &frontLeftSurf, NULL);
 
    if (frontLeftSurf) {
-      if (_screen != screen) {
-         struct trace_surface *tr_surf = trace_surface( frontLeftSurf );
-         struct pipe_surface *surf = tr_surf->surface;
-         frontLeftSurf = surf;
-      }
-
-      driver.display_surface(b, frontLeftSurf);
+      screen->flush_frontbuffer( screen,
+                                 frontLeftSurf, 
+                                 &b->ws );
    }
 
    xmesa_check_and_update_buffer_size(NULL, b);
@@ -1203,7 +1125,7 @@ XMesaBuffer XMesaFindBuffer( Display *dpy, Drawable d )
 {
    XMesaBuffer b;
    for (b = XMesaBufferList; b; b = b->Next) {
-      if (b->drawable == d && b->xm_visual->display == dpy) {
+      if (b->ws.drawable == d && b->xm_visual->display == dpy) {
          return b;
       }
    }
@@ -1237,10 +1159,10 @@ void XMesaGarbageCollect( void )
       next = b->Next;
       if (b->xm_visual &&
           b->xm_visual->display &&
-          b->drawable &&
+          b->ws.drawable &&
           b->type == WINDOW) {
          XSync(b->xm_visual->display, False);
-         if (!window_exists( b->xm_visual->display, b->drawable )) {
+         if (!window_exists( b->xm_visual->display, b->ws.drawable )) {
             /* found a dead window, free the ancillary info */
             XMesaDestroyBuffer( b );
          }
diff --git a/src/gallium/state_trackers/glx/xlib/xm_api.h b/src/gallium/state_trackers/glx/xlib/xm_api.h
index 004cb260dcd..de47064b410 100644
--- a/src/gallium/state_trackers/glx/xlib/xm_api.h
+++ b/src/gallium/state_trackers/glx/xlib/xm_api.h
@@ -62,15 +62,11 @@ and create a window, you must do the following to use the X/Mesa interface:
 #include "state_tracker/st_public.h"
 #include "os/os_thread.h"
 
+#include "state_tracker/xlib_sw_winsys.h"
 
 # include <X11/Xlib.h>
 # include <X11/Xlibint.h>
 # include <X11/Xutil.h>
-# ifdef USE_XSHM  /* was SHM */
-#  include <sys/ipc.h>
-#  include <sys/shm.h>
-#  include <X11/extensions/XShm.h>
-# endif
 
 typedef struct xmesa_buffer *XMesaBuffer;
 typedef struct xmesa_context *XMesaContext;
@@ -316,6 +312,7 @@ typedef enum {
  */
 struct xmesa_buffer {
    struct st_framebuffer *stfb;
+   struct xlib_drawable ws;
 
    GLboolean wasCurrent;	/* was ever the current buffer? */
    XMesaVisual xm_visual;	/* the X/Mesa visual */
@@ -329,13 +326,6 @@ struct xmesa_buffer {
    XImage *tempImage;
    unsigned long selectedEvents;/* for pbuffers only */
 
-   GLuint shm;			/* X Shared Memory extension status:	*/
-				/*    0 = not available			*/
-				/*    1 = XImage support available	*/
-				/*    2 = Pixmap support available too	*/
-#if defined(USE_XSHM)
-   XShmSegmentInfo shminfo;
-#endif
 
    GC gc;			/* scratch GC for span, line, tri drawing */
 
@@ -367,7 +357,7 @@ xmesa_buffer(GLframebuffer *fb)
 
 
 extern void
-xmesa_init(void);
+xmesa_init(Display *dpy);
 
 extern void
 xmesa_delete_framebuffer(struct gl_framebuffer *fb);
@@ -397,8 +387,6 @@ xmesa_buffer_height(XMesaBuffer b)
    return b->stfb->Base.Height;
 }
 
-extern int
-xmesa_check_for_xshm(Display *display);
 
 
 #endif
diff --git a/src/gallium/state_trackers/glx/xlib/xm_public.h b/src/gallium/state_trackers/glx/xlib/xm_public.h
new file mode 100644
index 00000000000..ac6a8ffb27a
--- /dev/null
+++ b/src/gallium/state_trackers/glx/xlib/xm_public.h
@@ -0,0 +1,44 @@
+
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef XM_WINSYS_H
+#define XM_WINSYS_H
+
+struct xm_driver;
+
+/* This is the driver interface required by the glx/xlib state tracker. 
+ */
+struct xm_driver {
+   struct pipe_screen *(*create_pipe_screen)( Display *display );
+};
+
+extern void
+xmesa_set_driver( const struct xm_driver *driver );
+
+
+#endif
diff --git a/src/gallium/state_trackers/glx/xlib/xm_winsys.h b/src/gallium/state_trackers/glx/xlib/xm_winsys.h
deleted file mode 100644
index 4bd5b5c8d3b..00000000000
--- a/src/gallium/state_trackers/glx/xlib/xm_winsys.h
+++ /dev/null
@@ -1,52 +0,0 @@
-
-/**************************************************************************
- * 
- * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-#ifndef XM_WINSYS_H
-#define XM_WINSYS_H
-
-struct pipe_context;
-struct pipe_screen;
-struct pipe_surface;
-struct xmesa_buffer;
-
-
-struct xm_driver {
-
-   struct pipe_screen *(*create_pipe_screen)( void );
-
-   void (*display_surface)( struct xmesa_buffer *, 
-                            struct pipe_surface * );
-
-};
-
-
-extern void
-xmesa_set_driver( const struct xm_driver *driver );
-
-
-#endif
diff --git a/src/gallium/state_trackers/python/SConscript b/src/gallium/state_trackers/python/SConscript
index 527e065cd91..781f54bf2b9 100644
--- a/src/gallium/state_trackers/python/SConscript
+++ b/src/gallium/state_trackers/python/SConscript
@@ -24,6 +24,7 @@ if 'python' in env['statetrackers']:
             'ws2_32',
         ])
     else:
+        env.Append(CPPDEFINES = ['GCC_HASCLASSVISIBILITY'])
         env.Append(LIBS = [
         	'GL',
             'X11',
@@ -33,31 +34,27 @@ if 'python' in env['statetrackers']:
         'gallium.i',
         'st_device.c',
         'st_sample.c',
+        'st_hardpipe_winsys.c',
+        'st_softpipe_winsys.c',
     ]
 
-    drivers = [
-        trace
-    ]
+    env.Prepend(LIBS = [
+        ws_null,
+        trace,
+        gallium,
+    ])
 
     if 'llvmpipe' in env['drivers']:
+        env.Append(CPPDEFINES = ['HAVE_LLVMPIPE'])
         env.Tool('llvm')
-        sources += ['st_llvmpipe_winsys.c']
-        drivers += [llvmpipe]
-    else:
-        sources += ['st_softpipe_winsys.c']
-        drivers += [softpipe]
-
-    pyst = env.ConvenienceLibrary(
-        target = 'pyst',
-        source = sources,
-    )
+        env.Prepend(LIBS = [llvmpipe])
+    if 'softpipe' in env['drivers']:
+        env.Append(CPPDEFINES = ['HAVE_SOFTPIPE'])
+        env.Prepend(LIBS = [softpipe])
 
     env['no_import_lib'] = 1
 
     env.SharedLibrary(
         target = '_gallium',
-        source = [
-            'st_hardpipe_winsys.c',
-        ],
-        LIBS = [pyst] + drivers + gallium + env['LIBS'],
+        source = sources,
     )
diff --git a/src/gallium/state_trackers/python/p_context.i b/src/gallium/state_trackers/python/p_context.i
index 3f36ccb6217..5c44462e80f 100644
--- a/src/gallium/state_trackers/python/p_context.i
+++ b/src/gallium/state_trackers/python/p_context.i
@@ -51,7 +51,7 @@ struct st_context {
    void set_blend( const struct pipe_blend_state *state ) {
       cso_set_blend($self->cso, state);
    }
-   
+
    void set_fragment_sampler( unsigned index, const struct pipe_sampler_state *state ) {
       cso_single_sampler($self->cso, index, state);
       cso_single_sampler_done($self->cso);
@@ -222,9 +222,9 @@ struct st_context {
    void set_vertex_elements(unsigned num) 
    {
       $self->num_vertex_elements = num;
-      $self->pipe->set_vertex_elements($self->pipe, 
-                                       $self->num_vertex_elements, 
-                                       $self->vertex_elements);
+      cso_set_vertex_elements($self->cso,
+                              $self->num_vertex_elements, 
+                              $self->vertex_elements);
    }
 
    /*
diff --git a/src/gallium/state_trackers/python/st_device.c b/src/gallium/state_trackers/python/st_device.c
index 45e78417500..335e8e7f0d8 100644
--- a/src/gallium/state_trackers/python/st_device.c
+++ b/src/gallium/state_trackers/python/st_device.c
@@ -34,8 +34,7 @@
 #include "util/u_math.h"
 #include "util/u_memory.h"
 #include "util/u_simple_shaders.h"
-#include "trace/tr_screen.h"
-#include "trace/tr_context.h"
+#include "trace/tr_public.h"
 
 #include "st_device.h"
 #include "st_winsys.h"
@@ -75,43 +74,34 @@ st_device_destroy(struct st_device *st_dev)
 }
 
 
-static struct st_device *
-st_device_create_from_st_winsys(const struct st_winsys *st_ws) 
+struct st_device *
+st_device_create(boolean hardware)
 {
+   struct pipe_screen *screen;
    struct st_device *st_dev;
-   
-   if(!st_ws->screen_create)
-      return NULL;
-   
+
+   if (hardware)
+      screen = st_hardware_screen_create();
+   else
+      screen = st_software_screen_create();
+
+   screen = trace_screen_create(screen);
+   if (!screen)
+      goto no_screen;
+
    st_dev = CALLOC_STRUCT(st_device);
-   if(!st_dev)
-      return NULL;
+   if (!st_dev)
+      goto no_device;
    
    pipe_reference_init(&st_dev->reference, 1);
-   st_dev->st_ws = st_ws;
-   
-   st_dev->real_screen = st_ws->screen_create();
-   if(!st_dev->real_screen) {
-      st_device_destroy(st_dev);
-      return NULL;
-   }
-
-   st_dev->screen = trace_screen_create(st_dev->real_screen);
-   if(!st_dev->screen) {
-      st_device_destroy(st_dev);
-      return NULL;
-   }
+   st_dev->screen = screen;
    
    return st_dev;
-}
-
 
-struct st_device *
-st_device_create(boolean hardware) {
-   if(hardware)
-      return st_device_create_from_st_winsys(&st_hardpipe_winsys);
-   else
-      return st_device_create_from_st_winsys(&st_softpipe_winsys);
+no_device:
+   screen->destroy(screen);
+no_screen:
+   return NULL;
 }
 
 
diff --git a/src/gallium/state_trackers/python/st_device.h b/src/gallium/state_trackers/python/st_device.h
index de9e0215d8e..6ec7409b11d 100644
--- a/src/gallium/state_trackers/python/st_device.h
+++ b/src/gallium/state_trackers/python/st_device.h
@@ -47,7 +47,8 @@ struct st_surface
 };
 
 
-struct st_context {
+struct st_context
+{
    struct st_device *st_dev;
    
    struct pipe_context *pipe;
@@ -72,13 +73,11 @@ struct st_context {
 };
 
 
-struct st_device {
+struct st_device
+{
    /* FIXME: we also need to refcount for textures and surfaces... */
    struct pipe_reference reference;
 
-   const struct st_winsys *st_ws; 
-
-   struct pipe_screen *real_screen;
    struct pipe_screen *screen;
 };
 
diff --git a/src/gallium/state_trackers/python/st_hardpipe_winsys.c b/src/gallium/state_trackers/python/st_hardpipe_winsys.c
index a3110a19d5d..b141177b793 100644
--- a/src/gallium/state_trackers/python/st_hardpipe_winsys.c
+++ b/src/gallium/state_trackers/python/st_hardpipe_winsys.c
@@ -54,11 +54,6 @@ static PFNGETGALLIUMSCREENMESAPROC pfnGetGalliumScreenMESA = NULL;
 static PFNCREATEGALLIUMCONTEXTMESAPROC pfnCreateGalliumContextMESA = NULL;
 
 
-/* XXX: Force init_gallium symbol to be linked */
-extern void init_gallium(void);
-void (*force_init_gallium_linkage)(void) = &init_gallium;
-
-
 #ifdef PIPE_OS_WINDOWS
 
 static INLINE boolean
@@ -207,16 +202,11 @@ st_hardpipe_load(void)
 #endif
 
 
-static struct pipe_screen *
-st_hardpipe_screen_create(void)
+struct pipe_screen *
+st_hardware_screen_create(void)
 {
    if(st_hardpipe_load())
       return pfnGetGalliumScreenMESA();
    else
-      return st_softpipe_winsys.screen_create();
+      return st_software_screen_create();
 }
-
-
-const struct st_winsys st_hardpipe_winsys = {
-   &st_hardpipe_screen_create
-};
diff --git a/src/gallium/state_trackers/python/st_llvmpipe_winsys.c b/src/gallium/state_trackers/python/st_llvmpipe_winsys.c
deleted file mode 100644
index 5d83b5a9e15..00000000000
--- a/src/gallium/state_trackers/python/st_llvmpipe_winsys.c
+++ /dev/null
@@ -1,141 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2010 VMware, Inc.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * THE 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/**
- * @file
- * Llvmpipe support. 
- * 
- * @author Jose Fonseca
- */
-
-
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "llvmpipe/lp_winsys.h"
-#include "st_winsys.h"
-
-
-static boolean
-llvmpipe_ws_is_displaytarget_format_supported( struct llvmpipe_winsys *ws,
-      enum pipe_format format )
-{
-   return FALSE;
-}
-
-
-static void *
-llvmpipe_ws_displaytarget_map(struct llvmpipe_winsys *ws,
-                              struct llvmpipe_displaytarget *dt,
-                              unsigned flags )
-{
-   assert(0);
-   return NULL;
-}
-
-
-static void
-llvmpipe_ws_displaytarget_unmap(struct llvmpipe_winsys *ws,
-                                struct llvmpipe_displaytarget *dt )
-{
-   assert(0);
-}
-
-
-static void
-llvmpipe_ws_displaytarget_destroy(struct llvmpipe_winsys *winsys,
-                                  struct llvmpipe_displaytarget *dt)
-{
-   assert(0);
-}
-
-
-static struct llvmpipe_displaytarget *
-llvmpipe_ws_displaytarget_create(struct llvmpipe_winsys *winsys,
-                                 enum pipe_format format,
-                                 unsigned width, unsigned height,
-                                 unsigned alignment,
-                                 unsigned *stride)
-{
-   return NULL;
-}
-
-
-static void
-llvmpipe_ws_displaytarget_display(struct llvmpipe_winsys *winsys,
-                                  struct llvmpipe_displaytarget *dt,
-                                  void *context_private)
-{
-   assert(0);
-}
-
-
-static void
-llvmpipe_ws_destroy(struct llvmpipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-static struct pipe_screen *
-st_llvmpipe_screen_create(void)
-{
-   static struct llvmpipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(llvmpipe_winsys);
-   if (!winsys)
-      goto no_winsys;
-
-   winsys->destroy = llvmpipe_ws_destroy;
-   winsys->is_displaytarget_format_supported = llvmpipe_ws_is_displaytarget_format_supported;
-   winsys->displaytarget_create = llvmpipe_ws_displaytarget_create;
-   winsys->displaytarget_map = llvmpipe_ws_displaytarget_map;
-   winsys->displaytarget_unmap = llvmpipe_ws_displaytarget_unmap;
-   winsys->displaytarget_display = llvmpipe_ws_displaytarget_display;
-   winsys->displaytarget_destroy = llvmpipe_ws_displaytarget_destroy;
-
-   screen = llvmpipe_create_screen(winsys);
-   if (!screen)
-      goto no_screen;
-
-   return screen;
-
-no_screen:
-   FREE(winsys);
-no_winsys:
-   return NULL;
-}
-
-
-
-const struct st_winsys st_softpipe_winsys = {
-   &st_llvmpipe_screen_create
-};
diff --git a/src/gallium/state_trackers/python/st_softpipe_winsys.c b/src/gallium/state_trackers/python/st_softpipe_winsys.c
index 81676bc3a4f..985374190c3 100644
--- a/src/gallium/state_trackers/python/st_softpipe_winsys.c
+++ b/src/gallium/state_trackers/python/st_softpipe_winsys.c
@@ -26,18 +26,47 @@
  * 
  **************************************************************************/
 
-/**
- * @file
- * Softpipe support. 
- * 
- * @author Keith Whitwell
- * @author Brian Paul
- * @author Jose Fonseca
- */
-
-#include "softpipe/sp_winsys.h"
+#include "util/u_debug.h"
+#include "softpipe/sp_public.h"
+#include "llvmpipe/lp_public.h"
+#include "state_tracker/sw_winsys.h"
+#include "null/null_sw_winsys.h"
 #include "st_winsys.h"
 
-const struct st_winsys st_softpipe_winsys = {
-   &softpipe_create_screen_malloc
-};
+
+struct pipe_screen *
+st_software_screen_create(void)
+{
+   struct sw_winsys *ws;
+   const char *default_driver;
+   const char *driver;
+   struct pipe_screen *screen = NULL;
+
+#if defined(HAVE_LLVMPIPE)
+   default_driver = "llvmpipe";
+#elif defined(HAVE_SOFTPIPE)
+   default_driver = "softpipe";
+#else
+   default_driver = "";
+#endif
+
+   ws = null_sw_create();
+   if(!ws)
+      return NULL;
+
+   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
+
+#ifdef HAVE_LLVMPIPE
+   if (strcmp(driver, "llvmpipe") == 0) {
+      screen = llvmpipe_create_screen(ws);
+   }
+#endif
+
+#ifdef HAVE_SOFTPIPE
+   if (strcmp(driver, "softpipe") == 0) {
+      screen = softpipe_create_screen(ws);
+   }
+#endif
+
+   return screen;
+}
diff --git a/src/gallium/state_trackers/python/st_winsys.h b/src/gallium/state_trackers/python/st_winsys.h
index 0c7b6a200e1..e1a99383a41 100644
--- a/src/gallium/state_trackers/python/st_winsys.h
+++ b/src/gallium/state_trackers/python/st_winsys.h
@@ -31,19 +31,13 @@
 
 
 struct pipe_screen;
-struct pipe_context;
 
 
-struct st_winsys 
-{
-   struct pipe_screen *
-   (*screen_create)(void);
-};
+struct pipe_screen *
+st_hardware_screen_create(void);
 
-
-extern const struct st_winsys st_softpipe_winsys;
-
-extern const struct st_winsys st_hardpipe_winsys;
+struct pipe_screen *
+st_software_screen_create(void);
 
 
 #endif /* ST_WINSYS_H_ */
diff --git a/src/gallium/state_trackers/vega/Makefile b/src/gallium/state_trackers/vega/Makefile
index 037d8dc911a..7f04b2aa832 100644
--- a/src/gallium/state_trackers/vega/Makefile
+++ b/src/gallium/state_trackers/vega/Makefile
@@ -53,7 +53,7 @@ INCLUDE_DIRS = \
 
 
 .c.o:
-	$(CC) -c $(INCLUDE_DIRS) $(DEFINES) $(CFLAGS) $< -o $@
+	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
 
 default: depend $(TOP)/$(LIB_DIR)/$(VG_LIB_NAME)
 
diff --git a/src/gallium/state_trackers/vega/api_masks.c b/src/gallium/state_trackers/vega/api_masks.c
index 9c123a4cf95..7eb5ea1f078 100644
--- a/src/gallium/state_trackers/vega/api_masks.c
+++ b/src/gallium/state_trackers/vega/api_masks.c
@@ -86,6 +86,8 @@ draw_clear_quad(struct vg_context *st,
 
    /* draw */
    if (buf) {
+      cso_set_vertex_elements(st->cso_context, 2, st->velems);
+
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c
index c06dbf52069..eef2c1eb876 100644
--- a/src/gallium/state_trackers/vega/polygon.c
+++ b/src/gallium/state_trackers/vega/polygon.c
@@ -292,12 +292,12 @@ static void draw_polygon(struct vg_context *ctx,
    pipe->set_vertex_buffers(pipe, 1, &vbuffer);
 
    /* tell pipe about the vertex attributes */
+   memset(&velement, 0, sizeof(velement));
    velement.src_offset = 0;
    velement.instance_divisor = 0;
    velement.vertex_buffer_index = 0;
    velement.src_format = PIPE_FORMAT_R32G32_FLOAT;
-   velement.nr_components = COMPONENTS;
-   pipe->set_vertex_elements(pipe, 1, &velement);
+   cso_set_vertex_elements(ctx->cso_context, 1, &velement);
 
    /* draw */
    pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_FAN, 
diff --git a/src/gallium/state_trackers/vega/renderer.c b/src/gallium/state_trackers/vega/renderer.c
index 05620efa9c0..47e8b470a11 100644
--- a/src/gallium/state_trackers/vega/renderer.c
+++ b/src/gallium/state_trackers/vega/renderer.c
@@ -210,6 +210,7 @@ void renderer_draw_quad(struct renderer *r,
    buf = setup_vertex_data(r, x1, y1, x2, y2, depth);
 
    if (buf) {
+      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
       util_draw_vertex_buffer(r->pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
@@ -248,6 +249,7 @@ void renderer_draw_texture(struct renderer *r,
                                s0, t0, s1, t1, 0.0f);
 
    if (buf) {
+      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
       util_draw_vertex_buffer(pipe, buf, 0,
                            PIPE_PRIM_TRIANGLE_FAN,
                            4,  /* verts */
@@ -370,6 +372,7 @@ void renderer_copy_texture(struct renderer *ctx,
                          0.0f);
 
    if (buf) {
+      cso_set_vertex_elements(ctx->cso, 2, ctx->owner->velems);
       util_draw_vertex_buffer(ctx->pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
@@ -535,6 +538,7 @@ void renderer_copy_surface(struct renderer *ctx,
                            (float) dstX1, (float) dstY1, z);
 
    if (buf) {
+      cso_set_vertex_elements(ctx->cso, 2, ctx->owner->velems);
       util_draw_vertex_buffer(ctx->pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
@@ -587,6 +591,7 @@ void renderer_texture_quad(struct renderer *r,
                           s0, t0, s1, t1, 0.0f);
 
    if (buf) {
+      cso_set_vertex_elements(r->cso, 2, r->owner->velems);
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
diff --git a/src/gallium/state_trackers/vega/vg_context.c b/src/gallium/state_trackers/vega/vg_context.c
index 426bf9bc62b..170391ec031 100644
--- a/src/gallium/state_trackers/vega/vg_context.c
+++ b/src/gallium/state_trackers/vega/vg_context.c
@@ -72,6 +72,7 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
                                       struct vg_context *share)
 {
    struct vg_context *ctx;
+   unsigned i;
 
    ctx = CALLOC_STRUCT(vg_context);
 
@@ -103,6 +104,13 @@ struct vg_context * vg_create_context(struct pipe_context *pipe,
    ctx->blend_sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
    ctx->blend_sampler.normalized_coords = 0;
 
+   for (i = 0; i < 2; i++) {
+      ctx->velems[i].src_offset = i * 4 * sizeof(float);
+      ctx->velems[i].instance_divisor = 0;
+      ctx->velems[i].vertex_buffer_index = 0;
+      ctx->velems[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    vg_set_error(ctx, VG_NO_ERROR);
 
    ctx->owned_objects[VG_OBJECT_PAINT] = cso_hash_create();
diff --git a/src/gallium/state_trackers/vega/vg_context.h b/src/gallium/state_trackers/vega/vg_context.h
index bc88c8d139d..804e9e76d77 100644
--- a/src/gallium/state_trackers/vega/vg_context.h
+++ b/src/gallium/state_trackers/vega/vg_context.h
@@ -146,6 +146,7 @@ struct vg_context
    struct vg_shader *clear_vs;
    struct vg_shader *texture_vs;
    struct pipe_buffer *vs_const_buffer;
+   struct pipe_vertex_element velems[2];
 };
 
 struct vg_object {
diff --git a/src/gallium/state_trackers/vega/vg_tracker.c b/src/gallium/state_trackers/vega/vg_tracker.c
index 57d3baad7fb..ea5c2ce41f6 100644
--- a/src/gallium/state_trackers/vega/vg_tracker.c
+++ b/src/gallium/state_trackers/vega/vg_tracker.c
@@ -376,12 +376,12 @@ void st_unreference_framebuffer(struct st_framebuffer *stfb)
 
 boolean st_make_current(struct vg_context *st,
                         struct st_framebuffer *draw,
-                        struct st_framebuffer *read)
+                        struct st_framebuffer *read,
+                        void *winsys_drawable_handle)
 {
    vg_set_current_context(st);
-   if (st) {
+   if (st)
       st->draw_buffer = draw;
-   }
    return VG_TRUE;
 }
 
diff --git a/src/gallium/state_trackers/vega/vg_tracker.h b/src/gallium/state_trackers/vega/vg_tracker.h
index c1196954a76..165a6b7a332 100644
--- a/src/gallium/state_trackers/vega/vg_tracker.h
+++ b/src/gallium/state_trackers/vega/vg_tracker.h
@@ -101,7 +101,8 @@ void st_unreference_framebuffer(struct st_framebuffer *stfb);
 PUBLIC
 boolean st_make_current(struct vg_context *st,
                         struct st_framebuffer *draw,
-                        struct st_framebuffer *read);
+                        struct st_framebuffer *read,
+                        void *winsys_drawable_handle);
 
 PUBLIC
 struct vg_context *st_get_current(void);
diff --git a/src/gallium/state_trackers/wgl/stw_context.c b/src/gallium/state_trackers/wgl/stw_context.c
index 05ccd5febcf..1f11b649c3e 100644
--- a/src/gallium/state_trackers/wgl/stw_context.c
+++ b/src/gallium/state_trackers/wgl/stw_context.c
@@ -226,7 +226,7 @@ DrvDeleteContext(
       
       /* Unbind current if deleting current context. */
       if (curctx == ctx)
-         st_make_current( NULL, NULL, NULL );
+         st_make_current( NULL, NULL, NULL, NULL );
 
       st_destroy_context(ctx->st);
       FREE(ctx);
@@ -317,7 +317,7 @@ stw_make_current(
    }
 
    if (hdc == NULL || dhglrc == 0) {
-      return st_make_current( NULL, NULL, NULL );
+      return st_make_current( NULL, NULL, NULL, NULL );
    }
 
    pipe_mutex_lock( stw_dev->ctx_mutex ); 
@@ -352,7 +352,7 @@ stw_make_current(
    /* pass to stw_flush_frontbuffer as context_private */
    ctx->st->pipe->priv = hdc;
    
-   if(!st_make_current( ctx->st, fb->stfb, fb->stfb ))
+   if(!st_make_current( ctx->st, fb->stfb, fb->stfb, hdc ))
       goto fail;
 
 success:
@@ -367,7 +367,7 @@ success:
 fail:
    if(fb)
       stw_framebuffer_release(fb);
-   st_make_current( NULL, NULL, NULL );
+   st_make_current( NULL, NULL, NULL, NULL );
    return FALSE;
 }
 
diff --git a/src/gallium/state_trackers/wgl/stw_device.c b/src/gallium/state_trackers/wgl/stw_device.c
index 472a2a5379a..ea300f27cb5 100644
--- a/src/gallium/state_trackers/wgl/stw_device.c
+++ b/src/gallium/state_trackers/wgl/stw_device.c
@@ -33,11 +33,6 @@
 #include "pipe/p_screen.h"
 #include "state_tracker/st_public.h"
 
-#ifdef DEBUG
-#include "trace/tr_screen.h"
-#include "trace/tr_texture.h"
-#endif
-
 #include "stw_device.h"
 #include "stw_winsys.h"
 #include "stw_pixelformat.h"
@@ -107,13 +102,10 @@ stw_init(const struct stw_winsys *stw_winsys)
    if(stw_winsys->get_adapter_luid)
       stw_winsys->get_adapter_luid(screen, &stw_dev->AdapterLuid);
 
-#ifdef DEBUG
-   stw_dev->screen = trace_screen_create(screen);
-   stw_dev->trace_running = stw_dev->screen != screen ? TRUE : FALSE;
-#else
    stw_dev->screen = screen;
-#endif
-   
+
+   /* XXX
+    */
    stw_dev->screen->flush_frontbuffer = &stw_flush_frontbuffer;
    
    pipe_mutex_init( stw_dev->ctx_mutex );
diff --git a/src/gallium/state_trackers/wgl/stw_device.h b/src/gallium/state_trackers/wgl/stw_device.h
index a83841f6b7d..2e9ba197dfa 100644
--- a/src/gallium/state_trackers/wgl/stw_device.h
+++ b/src/gallium/state_trackers/wgl/stw_device.h
@@ -48,10 +48,6 @@ struct stw_device
    
    struct pipe_screen *screen;
    
-#ifdef DEBUG
-   boolean trace_running;
-#endif
-
    LUID AdapterLuid;
 
    struct stw_pixelformat_info pixelformats[STW_MAX_PIXELFORMATS];
diff --git a/src/gallium/state_trackers/wgl/stw_ext_gallium.c b/src/gallium/state_trackers/wgl/stw_ext_gallium.c
index 8dd63f124ad..5ecbd8048de 100644
--- a/src/gallium/state_trackers/wgl/stw_ext_gallium.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_gallium.c
@@ -31,11 +31,6 @@
 #include "stw_winsys.h"
 #include "stw_ext_gallium.h"
 
-#ifdef DEBUG
-#include "trace/tr_screen.h"
-#include "trace/tr_context.h"
-#endif
-
 
 struct pipe_screen * APIENTRY
 wglGetGalliumScreenMESA(void)
diff --git a/src/gallium/state_trackers/wgl/stw_framebuffer.c b/src/gallium/state_trackers/wgl/stw_framebuffer.c
index 02de21ccb2b..4f1629de2f2 100644
--- a/src/gallium/state_trackers/wgl/stw_framebuffer.c
+++ b/src/gallium/state_trackers/wgl/stw_framebuffer.c
@@ -34,11 +34,6 @@
 #include "state_tracker/st_context.h"
 #include "state_tracker/st_public.h"
 
-#ifdef DEBUG
-#include "trace/tr_screen.h"
-#include "trace/tr_texture.h"
-#endif
-
 #include "stw_icd.h"
 #include "stw_framebuffer.h"
 #include "stw_device.h"
@@ -495,13 +490,6 @@ DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 
    surface = (struct pipe_surface *)data->pPrivateData;
 
-#ifdef DEBUG
-   if(stw_dev->trace_running) {
-      screen = trace_screen(screen)->screen;
-      surface = trace_surface(surface)->surface;
-   }
-#endif
-
    if(data->hSharedSurface != fb->hSharedSurface) {
       if(fb->shared_surface) {
          stw_dev->stw_winsys->shared_surface_close(screen, fb->shared_surface);
@@ -563,13 +551,6 @@ stw_framebuffer_present_locked(HDC hdc,
    else {
       struct pipe_screen *screen = stw_dev->screen;
 
-#ifdef DEBUG
-      if(stw_dev->trace_running) {
-         screen = trace_screen(screen)->screen;
-         surface = trace_surface(surface)->surface;
-      }
-#endif
-
       stw_dev->stw_winsys->present( screen, surface, hdc );
 
       stw_framebuffer_update(fb);
diff --git a/src/gallium/state_trackers/xorg/xorg_crtc.c b/src/gallium/state_trackers/xorg/xorg_crtc.c
index 53a3c5c2b7d..7268a79a23d 100644
--- a/src/gallium/state_trackers/xorg/xorg_crtc.c
+++ b/src/gallium/state_trackers/xorg/xorg_crtc.c
@@ -197,11 +197,11 @@ crtc_load_cursor_argb_ga3d(xf86CrtcPtr crtc, CARD32 * image)
 
     if (!crtcp->cursor_tex) {
 	struct pipe_texture templat;
-	unsigned pitch;
+	struct winsys_handle whandle;
 
 	memset(&templat, 0, sizeof(templat));
 	templat.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET;
-	templat.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY;
+	templat.tex_usage |= PIPE_TEXTURE_USAGE_SCANOUT;
 	templat.target = PIPE_TEXTURE_2D;
 	templat.last_level = 0;
 	templat.depth0 = 1;
@@ -209,13 +209,14 @@ crtc_load_cursor_argb_ga3d(xf86CrtcPtr crtc, CARD32 * image)
 	templat.width0 = 64;
 	templat.height0 = 64;
 
+	memset(&whandle, 0, sizeof(whandle));
+	whandle.type = DRM_API_HANDLE_TYPE_KMS;
+
 	crtcp->cursor_tex = ms->screen->texture_create(ms->screen,
 						       &templat);
-	ms->api->local_handle_from_texture(ms->api,
-					   ms->screen,
-					   crtcp->cursor_tex,
-					   &pitch,
-					   &crtcp->cursor_handle);
+	ms->screen->texture_get_handle(ms->screen, crtcp->cursor_tex, &whandle);
+
+	crtcp->cursor_handle = whandle.handle;
     }
 
     transfer = ms->screen->get_tex_transfer(ms->screen, crtcp->cursor_tex,
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index 5fc85c0e98c..f23e4c6cc7c 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -67,7 +67,7 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
     struct exa_pixmap_priv *exa_priv;
     BufferPrivatePtr private = buffer->driverPrivate;
     PixmapPtr pPixmap;
-    unsigned stride, handle;
+    struct winsys_handle whandle;
 
     if (pDraw->type == DRAWABLE_PIXMAP)
 	pPixmap = (PixmapPtr) pDraw;
@@ -75,6 +75,7 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
 	pPixmap = (*pScreen->GetWindowPixmap)((WindowPtr) pDraw);
     exa_priv = exaGetPixmapDriverPrivate(pPixmap);
 
+
     switch (buffer->attachment) {
     default:
 	if (buffer->attachment != DRI2BufferFakeFrontLeft ||
@@ -128,7 +129,7 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
 	    template.depth0 = 1;
 	    template.last_level = 0;
 	    template.tex_usage = PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
-		PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+		PIPE_TEXTURE_USAGE_SHARED;
 	    tex = ms->screen->texture_create(ms->screen, &template);
 	    pipe_texture_reference(&exa_priv->depth_stencil_tex, tex);
 	}
@@ -153,10 +154,13 @@ dri2_do_create_buffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int form
     if (!tex)
 	FatalError("NO TEXTURE IN DRI2\n");
 
-    ms->api->shared_handle_from_texture(ms->api, ms->screen, tex, &stride, &handle);
+    memset(&whandle, 0, sizeof(whandle));
+    whandle.type = DRM_API_HANDLE_TYPE_SHARED;
+
+    ms->screen->texture_get_handle(ms->screen, tex, &whandle);
 
-    buffer->name = handle;
-    buffer->pitch = stride;
+    buffer->name = whandle.handle;
+    buffer->pitch = whandle.stride;
     buffer->cpp = 4;
     buffer->driverPrivate = private;
     buffer->flags = 0; /* not tiled */
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index d7c67463d21..8ac5179545a 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -994,8 +994,9 @@ static Bool
 drv_create_front_buffer_ga3d(ScrnInfoPtr pScrn)
 {
     modesettingPtr ms = modesettingPTR(pScrn);
-    unsigned handle, stride, fb_id;
     struct pipe_texture *tex;
+    struct winsys_handle whandle;
+    unsigned fb_id;
     int ret;
 
     ms->noEvict = TRUE;
@@ -1006,10 +1007,10 @@ drv_create_front_buffer_ga3d(ScrnInfoPtr pScrn)
     if (!tex)
 	return FALSE;
 
-    if (!ms->api->local_handle_from_texture(ms->api, ms->screen,
-					    tex,
-					    &stride,
-					    &handle))
+    memset(&whandle, 0, sizeof(whandle));
+    whandle.type = DRM_API_HANDLE_TYPE_KMS;
+
+    if (!ms->screen->texture_get_handle(ms->screen, tex, &whandle))
 	goto err_destroy;
 
     ret = drmModeAddFB(ms->fd,
@@ -1017,8 +1018,8 @@ drv_create_front_buffer_ga3d(ScrnInfoPtr pScrn)
 		       pScrn->virtualY,
 		       pScrn->depth,
 		       pScrn->bitsPerPixel,
-		       stride,
-		       handle,
+		       whandle.stride,
+		       whandle.handle,
 		       &fb_id);
     if (ret) {
 	debug_printf("%s: failed to create framebuffer (%i, %s)\n",
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index a242e02ee77..a7ffe3f4992 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -789,7 +789,7 @@ xorg_exa_set_displayed_usage(PixmapPtr pPixmap)
 	return 0;
     }
 
-    priv->flags |= PIPE_TEXTURE_USAGE_PRIMARY;
+    priv->flags |= PIPE_TEXTURE_USAGE_SCANOUT;
 
     return 0;
 }
@@ -805,7 +805,7 @@ xorg_exa_set_shared_usage(PixmapPtr pPixmap)
 	return 0;
     }
 
-    priv->flags |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+    priv->flags |= PIPE_TEXTURE_USAGE_SHARED;
 
     return 0;
 }
@@ -943,7 +943,7 @@ xorg_exa_set_texture(PixmapPtr pPixmap, struct  pipe_texture *tex)
 {
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap);
 
-    int mask = PIPE_TEXTURE_USAGE_PRIMARY | PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+    int mask = PIPE_TEXTURE_USAGE_SHARED | PIPE_TEXTURE_USAGE_SCANOUT;
 
     if (!priv)
 	return FALSE;
@@ -976,8 +976,8 @@ xorg_exa_create_root_texture(ScrnInfoPtr pScrn,
     template.depth0 = 1;
     template.last_level = 0;
     template.tex_usage |= PIPE_TEXTURE_USAGE_RENDER_TARGET;
-    template.tex_usage |= PIPE_TEXTURE_USAGE_PRIMARY;
-    template.tex_usage |= PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+    template.tex_usage |= PIPE_TEXTURE_USAGE_SCANOUT;
+    template.tex_usage |= PIPE_TEXTURE_USAGE_SHARED;
 
     return exa->scrn->texture_create(exa->scrn, &template);
 }
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.c b/src/gallium/state_trackers/xorg/xorg_renderer.c
index 83b0d31e38d..1eb926360b9 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.c
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.c
@@ -68,6 +68,8 @@ renderer_draw(struct xorg_renderer *r)
 
 
    if (buf) {
+      cso_set_vertex_elements(r->cso, r->attrs_per_vertex, r->velems);
+
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_QUADS,
                               num_verts,  /* verts */
@@ -92,6 +94,7 @@ renderer_init_state(struct xorg_renderer *r)
 {
    struct pipe_depth_stencil_alpha_state dsa;
    struct pipe_rasterizer_state raster;
+   unsigned i;
 
    /* set common initial clip state */
    memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
@@ -103,6 +106,14 @@ renderer_init_state(struct xorg_renderer *r)
    raster.gl_rasterization_rules = 1;
    cso_set_rasterizer(r->cso, &raster);
 
+   /* vertex elements state */
+   memset(&r->velems[0], 0, sizeof(r->velems[0]) * 3);
+   for (i = 0; i < 3; i++) {
+      r->velems[i].src_offset = i * 4 * sizeof(float);
+      r->velems[i].instance_divisor = 0;
+      r->velems[i].vertex_buffer_index = 0;
+      r->velems[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
 }
 
 
@@ -600,6 +611,8 @@ void renderer_draw_yuv(struct xorg_renderer *r,
    if (buf) {
       const int num_attribs = 2; /*pos + tex coord*/
 
+      cso_set_vertex_elements(r->cso, num_attribs, r->velems);
+
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_QUADS,
                               4,  /* verts */
diff --git a/src/gallium/state_trackers/xorg/xorg_renderer.h b/src/gallium/state_trackers/xorg/xorg_renderer.h
index af6aa0567d6..3d006287199 100644
--- a/src/gallium/state_trackers/xorg/xorg_renderer.h
+++ b/src/gallium/state_trackers/xorg/xorg_renderer.h
@@ -28,6 +28,7 @@ struct xorg_renderer {
 
    float buffer[BUF_SIZE];
    int buffer_size;
+   struct pipe_vertex_element velems[3];
 
    /* number of attributes per vertex for the current
     * draw operation */
diff --git a/src/gallium/state_trackers/xorg/xvmc/surface.c b/src/gallium/state_trackers/xorg/xvmc/surface.c
index 87d1dfaace2..12d94e0c5ca 100644
--- a/src/gallium/state_trackers/xorg/xvmc/surface.c
+++ b/src/gallium/state_trackers/xorg/xvmc/surface.c
@@ -106,7 +106,7 @@ CreateOrResizeBackBuffer(struct pipe_video_context *vpipe, unsigned int width, u
    template.width0 = width;
    template.height0 = height;
    template.depth0 = 1;
-   template.tex_usage = PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+   template.tex_usage = PIPE_TEXTURE_USAGE_SHARED;
 
    tex = vpipe->screen->texture_create(vpipe->screen, &template);
    if (!tex)
diff --git a/src/gallium/targets/Makefile b/src/gallium/targets/Makefile
new file mode 100644
index 00000000000..a0bc5eb14fa
--- /dev/null
+++ b/src/gallium/targets/Makefile
@@ -0,0 +1,12 @@
+# src/gallium/winsys/Makefile
+TOP = ../../..
+include $(TOP)/configs/current
+
+SUBDIRS = $(GALLIUM_TARGET_DIRS)
+
+default install clean:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE) $@) || exit 1; \
+		fi \
+	done
diff --git a/src/gallium/targets/SConscript b/src/gallium/targets/SConscript
new file mode 100644
index 00000000000..df62fc65fb7
--- /dev/null
+++ b/src/gallium/targets/SConscript
@@ -0,0 +1,16 @@
+Import('*')
+
+#if env['dri']:
+#	SConscript([
+#		'drm/SConscript',
+#	])
+	
+if 'xlib' in env['winsys']:
+	SConscript([
+		'libgl-xlib/SConscript',
+	])
+
+if 'gdi' in env['winsys']:
+	SConscript([
+		'libgl-gdi/SConscript',
+	])
diff --git a/src/gallium/targets/libgl-gdi/SConscript b/src/gallium/targets/libgl-gdi/SConscript
new file mode 100644
index 00000000000..57704440ce7
--- /dev/null
+++ b/src/gallium/targets/libgl-gdi/SConscript
@@ -0,0 +1,51 @@
+#######################################################################
+# SConscript for gdi winsys
+
+Import('*')
+
+if env['platform'] == 'windows':
+
+    env = env.Clone()
+
+    env.Append(CPPPATH = [
+        '#src/gallium/state_trackers/wgl',
+    ])
+
+    env.Append(LIBS = [
+        'gdi32',
+        'user32',
+        'kernel32',
+        'ws2_32',
+    ])
+
+    sources = []
+    drivers = []
+
+    if 'softpipe' in env['drivers']:
+        sources = ['gdi_softpipe_winsys.c']
+        drivers = [softpipe]
+
+    if 'llvmpipe' in env['drivers']:
+        env.Tool('llvm')
+        if 'LLVM_VERSION' in env:
+            sources = ['gdi_llvmpipe_winsys.c']
+            drivers = [llvmpipe]
+
+    if not sources or not drivers:
+        print 'warning: softpipe or llvmpipe not selected, gdi winsys disabled'
+        Return()
+    
+    if env['gcc']:
+        sources += ['#src/gallium/state_trackers/wgl/opengl32.mingw.def']
+    else:
+        sources += ['#src/gallium/state_trackers/wgl/opengl32.def']
+        
+    drivers += [trace]
+
+    env['no_import_lib'] = 1
+
+    env.SharedLibrary(
+        target ='opengl32',
+        source = sources,
+        LIBS = wgl + ws_gdi + glapi + mesa + drivers + gallium + glsl + env['LIBS'],
+    )
diff --git a/src/gallium/targets/libgl-gdi/gdi_llvmpipe_winsys.c b/src/gallium/targets/libgl-gdi/gdi_llvmpipe_winsys.c
new file mode 100644
index 00000000000..29316a1ce28
--- /dev/null
+++ b/src/gallium/targets/libgl-gdi/gdi_llvmpipe_winsys.c
@@ -0,0 +1,124 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * LLVMpipe support.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ */
+
+
+#include <windows.h>
+
+#include "stw_winsys.h"
+#include "gdi/gdi_sw_winsys.h"
+#include "llvmpipe/lp_texture.h"
+#include "llvmpipe/lp_screen.h"
+#include "llvmpipe/lp_public.h"
+
+
+static struct pipe_screen *
+gdi_llvmpipe_screen_create(void)
+{
+   static struct sw_winsys *winsys;
+   struct pipe_screen *screen;
+
+   winsys = gdi_create_sw_winsys();
+   if(!winsys)
+      goto no_winsys;
+
+   screen = llvmpipe_create_screen(winsys);
+   if(!screen)
+      goto no_screen;
+
+   return screen;
+   
+no_screen:
+   winsys->destroy(winsys);
+no_winsys:
+   return NULL;
+}
+
+
+
+
+static void
+gdi_llvmpipe_present(struct pipe_screen *screen,
+                     struct pipe_surface *surface,
+                     HDC hDC)
+{
+   /* This will fail if any interposing layer (trace, debug, etc) has
+    * been introduced between the state-trackers and llvmpipe.
+    *
+    * Ideally this would get replaced with a call to
+    * pipe_screen::flush_frontbuffer().
+    *
+    * Failing that, it may be necessary for intervening layers to wrap
+    * other structs such as this stw_winsys as well...
+    */
+   gdi_sw_display(llvmpipe_screen(screen)->winsys,
+                  llvmpipe_texture(surface->texture)->dt,
+                  hDC);
+}
+
+
+static const struct stw_winsys stw_winsys = {
+   &gdi_llvmpipe_screen_create,
+   &gdi_llvmpipe_present,
+   NULL, /* get_adapter_luid */
+   NULL, /* shared_surface_open */
+   NULL, /* shared_surface_close */
+   NULL  /* compose */
+};
+
+
+BOOL WINAPI
+DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
+{
+   switch (fdwReason) {
+   case DLL_PROCESS_ATTACH:
+      stw_init(&stw_winsys);
+      stw_init_thread();
+      break;
+
+   case DLL_THREAD_ATTACH:
+      stw_init_thread();
+      break;
+
+   case DLL_THREAD_DETACH:
+      stw_cleanup_thread();
+      break;
+
+   case DLL_PROCESS_DETACH:
+      stw_cleanup_thread();
+      stw_cleanup();
+      break;
+   }
+   return TRUE;
+}
diff --git a/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c b/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c
new file mode 100644
index 00000000000..dfe60195d94
--- /dev/null
+++ b/src/gallium/targets/libgl-gdi/gdi_softpipe_winsys.c
@@ -0,0 +1,124 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * LLVMpipe support.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ */
+
+
+#include <windows.h>
+
+#include "stw_winsys.h"
+#include "gdi/gdi_sw_winsys.h"
+#include "softpipe/sp_texture.h"
+#include "softpipe/sp_screen.h"
+#include "softpipe/sp_public.h"
+
+
+static struct pipe_screen *
+gdi_softpipe_screen_create(void)
+{
+   static struct sw_winsys *winsys;
+   struct pipe_screen *screen;
+
+   winsys = gdi_create_sw_winsys();
+   if(!winsys)
+      goto no_winsys;
+
+   screen = softpipe_create_screen(winsys);
+   if(!screen)
+      goto no_screen;
+
+   return screen;
+   
+no_screen:
+   winsys->destroy(winsys);
+no_winsys:
+   return NULL;
+}
+
+
+
+
+static void
+gdi_softpipe_present(struct pipe_screen *screen,
+                     struct pipe_surface *surface,
+                     HDC hDC)
+{
+   /* This will fail if any interposing layer (trace, debug, etc) has
+    * been introduced between the state-trackers and softpipe.
+    *
+    * Ideally this would get replaced with a call to
+    * pipe_screen::flush_frontbuffer().
+    *
+    * Failing that, it may be necessary for intervening layers to wrap
+    * other structs such as this stw_winsys as well...
+    */
+   gdi_sw_display(softpipe_screen(screen)->winsys,
+                  softpipe_texture(surface->texture)->dt,
+                  hDC);
+}
+
+
+static const struct stw_winsys stw_winsys = {
+   &gdi_softpipe_screen_create,
+   &gdi_softpipe_present,
+   NULL, /* get_adapter_luid */
+   NULL, /* shared_surface_open */
+   NULL, /* shared_surface_close */
+   NULL  /* compose */
+};
+
+
+BOOL WINAPI
+DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
+{
+   switch (fdwReason) {
+   case DLL_PROCESS_ATTACH:
+      stw_init(&stw_winsys);
+      stw_init_thread();
+      break;
+
+   case DLL_THREAD_ATTACH:
+      stw_init_thread();
+      break;
+
+   case DLL_THREAD_DETACH:
+      stw_cleanup_thread();
+      break;
+
+   case DLL_PROCESS_DETACH:
+      stw_cleanup_thread();
+      stw_cleanup();
+      break;
+   }
+   return TRUE;
+}
diff --git a/src/gallium/targets/libgl-xlib/Makefile b/src/gallium/targets/libgl-xlib/Makefile
new file mode 100644
index 00000000000..5a4e035c2eb
--- /dev/null
+++ b/src/gallium/targets/libgl-xlib/Makefile
@@ -0,0 +1,100 @@
+# src/gallium/targets/libgl-xlib/Makefile
+
+# This makefile produces a "stand-alone" libGL.so which is based on
+# Xlib (no DRI HW acceleration)
+
+
+TOP = ../../../..
+include $(TOP)/configs/current
+
+
+GL_MAJOR = 1
+GL_MINOR = 5
+GL_TINY = 0$(MESA_MAJOR)0$(MESA_MINOR)0$(MESA_TINY)
+
+
+INCLUDE_DIRS = \
+	-I$(TOP)/include \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/src/mesa/main \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/src/gallium/state_trackers/glx/xlib \
+	-I$(TOP)/src/gallium/auxiliary
+
+DEFINES += \
+	-DGALLIUM_SOFTPIPE
+#-DGALLIUM_CELL will be defined by the config */
+
+XLIB_TARGET_SOURCES = \
+	xlib.c
+
+
+XLIB_TARGET_OBJECTS = $(XLIB_TARGET_SOURCES:.c=.o)
+
+
+# Note: CELL_SPU_LIB is only defined for cell configs
+
+LIBS = \
+	$(GALLIUM_DRIVERS) \
+	$(TOP)/src/gallium/state_trackers/glx/xlib/libxlib.a \
+	$(TOP)/src/gallium/winsys/xlib/libws_xlib.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/identity/libidentity.a \
+	$(TOP)/src/mesa/libglapi.a \
+	$(TOP)/src/mesa/libmesagallium.a \
+	$(GALLIUM_AUXILIARIES) \
+	$(CELL_SPU_LIB) \
+
+
+.SUFFIXES : .cpp
+
+.c.o:
+	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDE_DIRS) $(CXXFLAGS) $< -o $@
+
+
+
+default: $(TOP)/$(LIB_DIR)/gallium $(TOP)/$(LIB_DIR)/gallium/$(GL_LIB_NAME)
+
+$(TOP)/$(LIB_DIR)/gallium:
+	@ mkdir -p $(TOP)/$(LIB_DIR)/gallium
+
+# Make the libGL.so library
+$(TOP)/$(LIB_DIR)/gallium/$(GL_LIB_NAME): $(XLIB_TARGET_OBJECTS) $(LIBS) Makefile
+	$(TOP)/bin/mklib -o $(GL_LIB) \
+		-linker "$(CC)" \
+		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
+		-install $(TOP)/$(LIB_DIR)/gallium \
+		$(MKLIB_OPTIONS) $(XLIB_TARGET_OBJECTS) \
+		-Wl,--start-group $(LIBS) -Wl,--end-group $(GL_LIB_DEPS)
+
+
+depend: $(XLIB_TARGET_SOURCES)
+	@ echo "running $(MKDEP)"
+	@ rm -f depend  # workaround oops on gutsy?!?
+	@ touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(XLIB_TARGET_SOURCES) \
+		> /dev/null 2>/dev/null
+
+
+install: default
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/include/GL
+	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
+	$(INSTALL) -m 644 $(TOP)/include/GL/*.h $(DESTDIR)$(INSTALL_DIR)/include/GL
+	@if [ -e $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME) ]; then \
+		$(MINSTALL) $(TOP)/$(LIB_DIR)/libGL* $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR); \
+	fi
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` $(TOP)/include/GL/*.h
+
+clean:
+	-rm -f *.o
+
+
+include depend
diff --git a/src/gallium/targets/libgl-xlib/SConscript b/src/gallium/targets/libgl-xlib/SConscript
new file mode 100644
index 00000000000..efa7e797d1f
--- /dev/null
+++ b/src/gallium/targets/libgl-xlib/SConscript
@@ -0,0 +1,69 @@
+#######################################################################
+# SConscript for xlib winsys
+
+Import('*')
+
+if env['platform'] != 'linux':
+    Return()
+
+if 'mesa' not in env['statetrackers']:
+    print 'warning: Mesa state tracker disabled: skipping build of xlib libGL.so'
+    Return()
+
+if env['dri']:
+    print 'warning: DRI enabled: skipping build of xlib libGL.so'
+    Return()
+
+if not set(('softpipe', 'llvmpipe', 'cell')).intersection(env['drivers']):
+    print 'warning: no supported pipe driver: skipping build of xlib libGL.so'
+    Return()
+
+env = env.Clone()
+
+env.Append(CPPPATH = [
+    '#/src/mesa',
+    '#/src/mesa/main',
+    '#src/gallium/state_trackers/glx/xlib',
+])
+
+env.Append(CPPDEFINES = ['USE_XSHM'])
+
+env.Prepend(LIBS = [
+    st_xlib,
+    ws_xlib,
+    trace,
+    identity,
+    glapi,
+    mesa,
+    glsl,
+    gallium,
+])
+
+sources = [
+    'xlib.c',
+]
+
+if 'softpipe' in env['drivers']:
+    env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
+    env.Prepend(LIBS = [softpipe])
+
+if 'llvmpipe' in env['drivers']:
+    env.Tool('llvm')
+    if 'LLVM_VERSION' in env:
+        env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
+        env.Tool('udis86')
+        env.Prepend(LIBS = [llvmpipe])
+    
+if 'cell' in env['drivers']:
+    env.Append(CPPDEFINES = 'GALLIUM_CELL')
+    env.Prepend(LIBS = [cell])
+
+# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions
+libgl = env.SharedLibrary(
+    target ='GL',
+    source = sources,
+)
+
+if not env['dri']:
+    # Only install this libGL.so if DRI not enabled
+    env.InstallSharedLibrary(libgl, version=(1, 5))
diff --git a/src/gallium/targets/libgl-xlib/xlib.c b/src/gallium/targets/libgl-xlib/xlib.c
new file mode 100644
index 00000000000..05dc8db57d9
--- /dev/null
+++ b/src/gallium/targets/libgl-xlib/xlib.c
@@ -0,0 +1,160 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * 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 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell
+ */
+#include "pipe/p_compiler.h"
+#include "state_tracker/xlib_sw_winsys.h"
+#include "util/u_debug.h"
+#include "softpipe/sp_public.h"
+#include "llvmpipe/lp_public.h"
+#include "cell/ppu/cell_public.h"
+#include "target-helpers/wrap_screen.h"
+#include "xm_public.h"
+
+/* advertise OpenGL support */
+PUBLIC const int st_api_OpenGL = 1;
+
+
+/* Helper function to build a subset of a driver stack consisting of
+ * one of the software rasterizers (cell, llvmpipe, softpipe) and the
+ * xlib winsys.
+ *
+ * This function could be shared, but currently causes headaches for
+ * the build systems, particularly scons if we try.  Long term, want
+ * to avoid having global #defines for things like GALLIUM_LLVMPIPE,
+ * GALLIUM_CELL, etc.  Scons already eliminates those #defines, so
+ * things that are painful for it now are likely to be painful for
+ * other build systems in the future.
+ */
+static struct pipe_screen *
+swrast_xlib_create_screen( Display *display )
+{
+   struct sw_winsys *winsys;
+   struct pipe_screen *screen = NULL;
+
+   /* Create the underlying winsys, which performs presents to Xlib
+    * drawables:
+    */
+   winsys = xlib_create_sw_winsys( display );
+   if (winsys == NULL)
+      return NULL;
+
+   /* Create a software rasterizer on top of that winsys:
+    */
+#if defined(GALLIUM_CELL)
+   if (screen == NULL &&
+       !debug_get_bool_option("GALLIUM_NO_CELL", FALSE))
+      screen = cell_create_screen( winsys );
+#endif
+
+#if defined(GALLIUM_LLVMPIPE)
+   if (screen == NULL &&
+       !debug_get_bool_option("GALLIUM_NO_LLVM", FALSE))
+      screen = llvmpipe_create_screen( winsys );
+#endif
+
+   if (screen == NULL)
+      screen = softpipe_create_screen( winsys );
+
+   if (screen == NULL)
+      goto fail;
+
+   /* Inject any wrapping layers we want to here:
+    */
+   return gallium_wrap_screen( screen );
+
+fail:
+   if (winsys)
+      winsys->destroy( winsys );
+
+   return NULL;
+}
+
+struct xm_driver xlib_driver = 
+{
+   .create_pipe_screen = swrast_xlib_create_screen,
+};
+
+
+/* Build the rendering stack.
+ */
+static void _init( void ) __attribute__((constructor));
+static void _init( void )
+{
+   /* Initialize the xlib libgl code, pass in the winsys:
+    */
+   xmesa_set_driver( &xlib_driver );
+}
+
+
+/***********************************************************************
+ *
+ * Butt-ugly hack to convince the linker not to throw away public GL
+ * symbols (they are all referenced from getprocaddress, I guess).
+ */
+extern void (*linker_foo(const unsigned char *procName))();
+extern void (*glXGetProcAddress(const unsigned char *procName))();
+
+extern void (*linker_foo(const unsigned char *procName))()
+{
+   return glXGetProcAddress(procName);
+}
+
+
+/**
+ * When GLX_INDIRECT_RENDERING is defined, some symbols are missing in
+ * libglapi.a.  We need to define them here.
+ */
+#ifdef GLX_INDIRECT_RENDERING
+
+#define GL_GLEXT_PROTOTYPES
+#include "GL/gl.h"
+#include "glapi/glapi.h"
+#include "glapi/glapitable.h"
+#include "glapi/glapidispatch.h"
+
+#if defined(USE_MGL_NAMESPACE)
+#define NAME(func)  mgl##func
+#else
+#define NAME(func)  gl##func
+#endif
+
+#define DISPATCH(FUNC, ARGS, MESSAGE)		\
+   CALL_ ## FUNC(GET_DISPATCH(), ARGS);
+
+#define RETURN_DISPATCH(FUNC, ARGS, MESSAGE) 	\
+   return CALL_ ## FUNC(GET_DISPATCH(), ARGS);
+
+/* skip normal ones */
+#define _GLAPI_SKIP_NORMAL_ENTRY_POINTS
+#include "glapi/glapitemp.h"
+
+#endif /* GLX_INDIRECT_RENDERING */
diff --git a/src/gallium/winsys/drm/Makefile.egl b/src/gallium/winsys/drm/Makefile.egl
index 8363de6e975..bc5dd3a53b8 100644
--- a/src/gallium/winsys/drm/Makefile.egl
+++ b/src/gallium/winsys/drm/Makefile.egl
@@ -13,7 +13,8 @@ EGL_DRIVER_OBJECTS = $(EGL_DRIVER_SOURCES:.c=.o)
 
 common_LIBS = -ldrm -lm -ldl
 
-x11_ST = $(TOP)/src/gallium/state_trackers/egl/libeglx11.a
+x11_ST = $(TOP)/src/gallium/state_trackers/egl/libeglx11.a \
+	 $(TOP)/src/gallium/winsys/xlib/libws_xlib.a
 x11_LIBS = $(common_LIBS) -lX11 -lXext -lXfixes
 
 kms_ST = $(TOP)/src/gallium/state_trackers/egl/libeglkms.a
@@ -38,8 +39,10 @@ $(EGL_DISPLAY_LIBS): $(TOP)/$(LIB_DIR)/%.so: %.so
 
 define mklib-egl
 $(MKLIB) -o $@ -noprefix -linker '$(CC)' -ldflags '$(LDFLAGS)' \
-	$(MKLIB_OPTIONS) $(EGL_DRIVER_OBJECTS) $($(1)_ST) \
-	$(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) $($(1)_LIBS) $(EGL_DRIVER_LIBS)
+	$(MKLIB_OPTIONS) $(EGL_DRIVER_OBJECTS) \
+	-Wl,--start-group $($(1)_ST) $(EGL_DRIVER_PIPES) \
+	$(GALLIUM_AUXILIARIES) -Wl,--end-group \
+	$($(1)_LIBS) $(EGL_DRIVER_LIBS)
 endef
 
 egl_x11_$(EGL_DRIVER_NAME).so: $(EGL_DRIVER_OBJECTS) $(x11_ST) $(EGL_DRIVER_PIPES) $(GALLIUM_AUXILIARIES) Makefile
diff --git a/src/gallium/winsys/drm/i965/gem/i965_drm_api.c b/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
index a061eef0beb..21e82303f72 100644
--- a/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
+++ b/src/gallium/winsys/drm/i965/gem/i965_drm_api.c
@@ -37,129 +37,6 @@ i965_libdrm_get_device_id(unsigned int *device_id)
    fclose(file);
 }
 
-static struct i965_libdrm_buffer *
-i965_libdrm_buffer_from_handle(struct i965_libdrm_winsys *idws,
-                               const char* name, unsigned handle)
-{
-   struct i965_libdrm_buffer *buf = CALLOC_STRUCT(i965_libdrm_buffer);
-   uint32_t swizzle = 0;
-
-   if (BRW_DUMP)
-      debug_printf("%s\n", __FUNCTION__);
-
-   if (!buf)
-      return NULL;
-   pipe_reference_init(&buf->base.reference, 1);
-   buf->bo = drm_intel_bo_gem_create_from_name(idws->gem, name, handle);
-   buf->base.size = buf->bo->size;
-   buf->base.sws = &idws->base;
-   buf->flinked = TRUE;
-   buf->flink = handle;
-
-
-   if (!buf->bo)
-      goto err;
-
-   drm_intel_bo_get_tiling(buf->bo, &buf->tiling, &swizzle);
-   if (buf->tiling != 0)
-      buf->map_gtt = TRUE;
-
-   return buf;
-
-err:
-   FREE(buf);
-   return NULL;
-}
-
-
-/*
- * Exported functions
- */
-
-
-static struct pipe_texture *
-i965_libdrm_texture_from_shared_handle(struct drm_api *api,
-                                       struct pipe_screen *screen,
-                                       struct pipe_texture *template,
-                                       const char* name,
-                                       unsigned pitch,
-                                       unsigned handle)
-{
-   /* XXX: this is silly -- there should be a way to get directly from
-    * the "drm_api" struct to ourselves, without peering into
-    * unrelated code:
-    */
-   struct i965_libdrm_winsys *idws = i965_libdrm_winsys(brw_screen(screen)->sws);
-   struct i965_libdrm_buffer *buffer;
-
-   if (BRW_DUMP)
-      debug_printf("%s %s pitch %d handle 0x%x\n", __FUNCTION__,
-		   name, pitch, handle);
-
-   buffer = i965_libdrm_buffer_from_handle(idws, name, handle);
-   if (!buffer)
-      return NULL;
-
-   return brw_texture_blanket_winsys_buffer(screen, template, pitch,
-					    buffer->tiling,
-					    &buffer->base);
-}
-
-
-static boolean
-i965_libdrm_shared_handle_from_texture(struct drm_api *api,
-                                       struct pipe_screen *screen,
-                                       struct pipe_texture *texture,
-                                       unsigned *pitch,
-                                       unsigned *handle)
-{
-   struct i965_libdrm_buffer *buf = NULL;
-   struct brw_winsys_buffer *buffer = NULL;
-
-   if (BRW_DUMP)
-      debug_printf("%s\n", __FUNCTION__);
-
-   if (!brw_texture_get_winsys_buffer(texture, &buffer, pitch))
-      return FALSE;
-
-   buf = i965_libdrm_buffer(buffer);
-   if (!buf->flinked) {
-      if (drm_intel_bo_flink(buf->bo, &buf->flink))
-         return FALSE;
-      buf->flinked = TRUE;
-   }
-
-   *handle = buf->flink;
-
-   if (BRW_DUMP)
-      debug_printf("   -> pitch %d handle 0x%x\n", *pitch, *handle);
-
-   return TRUE;
-}
-
-static boolean
-i965_libdrm_local_handle_from_texture(struct drm_api *api,
-                                      struct pipe_screen *screen,
-                                      struct pipe_texture *texture,
-                                      unsigned *pitch,
-                                      unsigned *handle)
-{
-   struct brw_winsys_buffer *buffer = NULL;
-
-   if (BRW_DUMP)
-      debug_printf("%s\n", __FUNCTION__);
-
-   if (!brw_texture_get_winsys_buffer(texture, &buffer, pitch))
-      return FALSE;
-
-   *handle = i965_libdrm_buffer(buffer)->bo->handle;
-
-   if (BRW_DUMP)
-      debug_printf("   -> pitch %d handle 0x%x\n", *pitch, *handle);
-
-   return TRUE;
-}
-
 static void
 i965_libdrm_winsys_destroy(struct brw_winsys_screen *iws)
 {
@@ -225,9 +102,6 @@ struct drm_api i965_libdrm_api =
 {
    .name = "i965",
    .create_screen = i965_libdrm_create_screen,
-   .texture_from_shared_handle = i965_libdrm_texture_from_shared_handle,
-   .shared_handle_from_texture = i965_libdrm_shared_handle_from_texture,
-   .local_handle_from_texture = i965_libdrm_local_handle_from_texture,
    .destroy = destroy,
 };
 
diff --git a/src/gallium/winsys/drm/i965/gem/i965_drm_buffer.c b/src/gallium/winsys/drm/i965/gem/i965_drm_buffer.c
index 07be1df87f0..33a17496b2b 100644
--- a/src/gallium/winsys/drm/i965/gem/i965_drm_buffer.c
+++ b/src/gallium/winsys/drm/i965/gem/i965_drm_buffer.c
@@ -1,4 +1,5 @@
 
+#include "state_tracker/drm_api.h"
 #include "i965_drm_winsys.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
@@ -122,6 +123,78 @@ err:
    return PIPE_ERROR_OUT_OF_MEMORY;
 }
 
+static enum pipe_error
+i965_libdrm_bo_from_handle(struct brw_winsys_screen *sws,
+                           struct winsys_handle *whandle,
+                           unsigned *stride,
+                           unsigned *tile,
+                           struct brw_winsys_buffer **bo_out)
+{
+   struct i965_libdrm_winsys *idws = i965_libdrm_winsys(sws);
+   struct i965_libdrm_buffer *buf = CALLOC_STRUCT(i965_libdrm_buffer);
+   uint32_t swizzle = 0;
+
+   if (BRW_DUMP)
+      debug_printf("%s\n", __FUNCTION__);
+
+   if (!buf)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+
+   pipe_reference_init(&buf->base.reference, 1);
+   buf->bo = drm_intel_bo_gem_create_from_name(idws->gem, "FROM_HANDLE", whandle->handle);
+   buf->base.size = buf->bo->size;
+   buf->base.sws = &idws->base;
+   buf->flinked = TRUE;
+   buf->flink = whandle->handle;
+
+
+   if (!buf->bo)
+      goto err;
+
+   drm_intel_bo_get_tiling(buf->bo, &buf->tiling, &swizzle);
+   if (buf->tiling != 0)
+      buf->map_gtt = TRUE;
+
+   *tile = buf->tiling;
+   *stride = whandle->stride;
+
+   *bo_out = &buf->base;
+   return PIPE_OK;
+
+err:
+   FREE(buf);
+   return PIPE_ERROR_OUT_OF_MEMORY;
+}
+
+static enum pipe_error
+i965_libdrm_bo_get_handle(struct brw_winsys_buffer *buffer,
+                          struct winsys_handle *whandle,
+                          unsigned stride)
+{
+   struct i965_libdrm_buffer *buf = i965_libdrm_buffer(buffer);
+
+   if (BRW_DUMP)
+      debug_printf("%s\n", __FUNCTION__);
+
+   if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+      if (!buf->flinked) {
+         if (drm_intel_bo_flink(buf->bo, &buf->flink))
+            return PIPE_ERROR_BAD_INPUT;
+         buf->flinked = TRUE;
+      }
+
+      whandle->handle = buf->flink;
+   } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+      whandle->handle = buf->bo->handle;
+   } else {
+      assert(!"unknown usage");
+      return PIPE_ERROR_BAD_INPUT;
+   }
+
+   whandle->stride = stride;
+   return PIPE_OK;
+}
+
 static void 
 i965_libdrm_bo_destroy(struct brw_winsys_buffer *buffer)
 {
@@ -415,6 +488,8 @@ void
 i965_libdrm_winsys_init_buffer_functions(struct i965_libdrm_winsys *idws)
 {
    idws->base.bo_alloc             = i965_libdrm_bo_alloc;
+   idws->base.bo_from_handle       = i965_libdrm_bo_from_handle;
+   idws->base.bo_get_handle        = i965_libdrm_bo_get_handle;
    idws->base.bo_destroy           = i965_libdrm_bo_destroy;
    idws->base.bo_emit_reloc        = i965_libdrm_bo_emit_reloc;
    idws->base.bo_exec              = i965_libdrm_bo_exec;
diff --git a/src/gallium/winsys/drm/i965/xlib/xlib_i965.c b/src/gallium/winsys/drm/i965/xlib/xlib_i965.c
index 74501eeb16f..063e9f600b9 100644
--- a/src/gallium/winsys/drm/i965/xlib/xlib_i965.c
+++ b/src/gallium/winsys/drm/i965/xlib/xlib_i965.c
@@ -38,7 +38,7 @@
 #include "pipe/p_error.h"
 #include "pipe/p_context.h"
 
-#include "xm_winsys.h"
+#include "xm_public.h"
 
 #include "i965/brw_winsys.h"
 #include "i965/brw_screen.h"
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
index 377ed255131..e3b980a832b 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
@@ -38,99 +38,6 @@ intel_drm_get_device_id(unsigned int *device_id)
    fclose(file);
 }
 
-static struct intel_buffer *
-intel_drm_buffer_from_handle(struct intel_drm_winsys *idws,
-                             const char* name, unsigned handle)
-{
-   struct intel_drm_buffer *buf = CALLOC_STRUCT(intel_drm_buffer);
-   uint32_t tile = 0, swizzle = 0;
-
-   if (!buf)
-      return NULL;
-
-   buf->magic = 0xDEAD1337;
-   buf->bo = drm_intel_bo_gem_create_from_name(idws->pools.gem, name, handle);
-   buf->flinked = TRUE;
-   buf->flink = handle;
-
-   if (!buf->bo)
-      goto err;
-
-   drm_intel_bo_get_tiling(buf->bo, &tile, &swizzle);
-   if (tile != INTEL_TILE_NONE)
-      buf->map_gtt = TRUE;
-
-   return (struct intel_buffer *)buf;
-
-err:
-   FREE(buf);
-   return NULL;
-}
-
-
-/*
- * Exported functions
- */
-
-
-static struct pipe_texture *
-intel_drm_texture_from_shared_handle(struct drm_api *api,
-                                     struct pipe_screen *screen,
-                                     struct pipe_texture *templ,
-                                     const char* name,
-                                     unsigned pitch,
-                                     unsigned handle)
-{
-   struct intel_drm_winsys *idws = intel_drm_winsys(i915_screen(screen)->iws);
-   struct intel_buffer *buffer;
-
-   buffer = intel_drm_buffer_from_handle(idws, name, handle);
-   if (!buffer)
-      return NULL;
-
-   return i915_texture_blanket_intel(screen, templ, pitch, buffer);
-}
-
-static boolean
-intel_drm_shared_handle_from_texture(struct drm_api *api,
-                                     struct pipe_screen *screen,
-                                     struct pipe_texture *texture,
-                                     unsigned *pitch,
-                                     unsigned *handle)
-{
-   struct intel_drm_buffer *buf = NULL;
-   struct intel_buffer *buffer = NULL;
-   if (!i915_get_texture_buffer_intel(texture, &buffer, pitch))
-      return FALSE;
-
-   buf = intel_drm_buffer(buffer);
-   if (!buf->flinked) {
-      if (drm_intel_bo_flink(buf->bo, &buf->flink))
-         return FALSE;
-      buf->flinked = TRUE;
-   }
-
-   *handle = buf->flink;
-
-   return TRUE;
-}
-
-static boolean
-intel_drm_local_handle_from_texture(struct drm_api *api,
-                                    struct pipe_screen *screen,
-                                    struct pipe_texture *texture,
-                                    unsigned *pitch,
-                                    unsigned *handle)
-{
-   struct intel_buffer *buffer = NULL;
-   if (!i915_get_texture_buffer_intel(texture, &buffer, pitch))
-      return FALSE;
-
-   *handle = intel_drm_buffer(buffer)->bo->handle;
-
-   return TRUE;
-}
-
 static void
 intel_drm_winsys_destroy(struct intel_winsys *iws)
 {
@@ -192,9 +99,6 @@ struct drm_api intel_drm_api =
    .name = "i915",
    .driver_name = "i915",
    .create_screen = intel_drm_create_screen,
-   .texture_from_shared_handle = intel_drm_texture_from_shared_handle,
-   .shared_handle_from_texture = intel_drm_shared_handle_from_texture,
-   .local_handle_from_texture = intel_drm_local_handle_from_texture,
    .destroy = destroy,
 };
 
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
index ac4dd6e00e9..cb4f92a3b17 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
@@ -1,4 +1,5 @@
 
+#include "state_tracker/drm_api.h"
 #include "intel_drm_winsys.h"
 #include "util/u_memory.h"
 
@@ -52,6 +53,66 @@ err:
    return NULL;
 }
 
+static struct intel_buffer *
+intel_drm_buffer_from_handle(struct intel_winsys *iws,
+                             struct winsys_handle *whandle,
+                             unsigned *stride)
+{
+   struct intel_drm_winsys *idws = intel_drm_winsys(iws);
+   struct intel_drm_buffer *buf = CALLOC_STRUCT(intel_drm_buffer);
+   uint32_t tile = 0, swizzle = 0;
+
+   if (!buf)
+      return NULL;
+
+   buf->magic = 0xDEAD1337;
+   buf->bo = drm_intel_bo_gem_create_from_name(idws->pools.gem, "gallium3d_from_handle", whandle->handle);
+   buf->flinked = TRUE;
+   buf->flink = whandle->handle;
+
+   if (!buf->bo)
+      goto err;
+
+   drm_intel_bo_get_tiling(buf->bo, &tile, &swizzle);
+   if (tile != INTEL_TILE_NONE)
+      buf->map_gtt = TRUE;
+
+   *stride = whandle->stride;
+
+   return (struct intel_buffer *)buf;
+
+err:
+   FREE(buf);
+   return NULL;
+}
+
+static boolean
+intel_drm_buffer_get_handle(struct intel_winsys *iws,
+                            struct intel_buffer *buffer,
+                            struct winsys_handle *whandle,
+                            unsigned stride)
+{
+   struct intel_drm_buffer *buf = intel_drm_buffer(buffer);
+
+   if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+      if (!buf->flinked) {
+         if (drm_intel_bo_flink(buf->bo, &buf->flink))
+            return FALSE;
+         buf->flinked = TRUE;
+      }
+
+      whandle->handle = buf->flink;
+   } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+      whandle->handle = buf->bo->handle;
+   } else {
+      assert(!"unknown usage");
+      return FALSE;
+   }
+
+   whandle->stride = stride;
+   return TRUE;
+}
+
 static int
 intel_drm_buffer_set_fence_reg(struct intel_winsys *iws,
                                struct intel_buffer *buffer,
@@ -146,6 +207,8 @@ void
 intel_drm_winsys_init_buffer_functions(struct intel_drm_winsys *idws)
 {
    idws->base.buffer_create = intel_drm_buffer_create;
+   idws->base.buffer_from_handle = intel_drm_buffer_from_handle;
+   idws->base.buffer_get_handle = intel_drm_buffer_get_handle;
    idws->base.buffer_set_fence_reg = intel_drm_buffer_set_fence_reg;
    idws->base.buffer_map = intel_drm_buffer_map;
    idws->base.buffer_unmap = intel_drm_buffer_unmap;
diff --git a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
index 0d05f316c42..21517b4bb54 100644
--- a/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
+++ b/src/gallium/winsys/drm/nouveau/drm/nouveau_drm_api.c
@@ -21,9 +21,10 @@ dri_surface_from_handle(struct drm_api *api, struct pipe_screen *pscreen,
 	struct pipe_surface *ps = NULL;
 	struct pipe_texture *pt = NULL;
 	struct pipe_texture tmpl;
+	struct winsys_handle whandle;
 
 	memset(&tmpl, 0, sizeof(tmpl));
-	tmpl.tex_usage = PIPE_TEXTURE_USAGE_PRIMARY;
+	tmpl.tex_usage = PIPE_TEXTURE_USAGE_SCANOUT;
 	tmpl.target = PIPE_TEXTURE_2D;
 	tmpl.last_level = 0;
 	tmpl.depth0 = 1;
@@ -31,8 +32,11 @@ dri_surface_from_handle(struct drm_api *api, struct pipe_screen *pscreen,
 	tmpl.width0 = width;
 	tmpl.height0 = height;
 
-	pt = api->texture_from_shared_handle(api, pscreen, &tmpl,
-					     "front buffer", pitch, handle);
+	memset(&whandle, 0, sizeof(whandle));
+	whandle.stride = pitch;
+	whandle.handle = handle;
+
+	pt = pscreen->texture_from_handle(pscreen, &tmpl, &whandle);
 	if (!pt)
 		return NULL;
 
@@ -142,74 +146,10 @@ nouveau_drm_create_screen(struct drm_api *api, int fd,
 	return nvws->pscreen;
 }
 
-static struct pipe_texture *
-nouveau_drm_pt_from_name(struct drm_api *api, struct pipe_screen *pscreen,
-			 struct pipe_texture *templ, const char *name,
-			 unsigned stride, unsigned handle)
-{
-	struct nouveau_device *dev = nouveau_screen(pscreen)->device;
-	struct pipe_texture *pt;
-	struct pipe_buffer *pb;
-	int ret;
-
-	pb = CALLOC(1, sizeof(struct pipe_buffer) + sizeof(struct nouveau_bo*));
-	if (!pb)
-		return NULL;
-
-	ret = nouveau_bo_handle_ref(dev, handle, (struct nouveau_bo**)(pb+1));
-	if (ret) {
-		debug_printf("%s: ref name 0x%08x failed with %d\n",
-			     __func__, handle, ret);
-		FREE(pb);
-		return NULL;
-	}
-
-	pipe_reference_init(&pb->reference, 1);
-	pb->screen = pscreen;
-	pb->alignment = 0;
-	pb->usage = PIPE_BUFFER_USAGE_GPU_READ_WRITE |
-		    PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-	pb->size = nouveau_bo(pb)->size;
-	pt = pscreen->texture_blanket(pscreen, templ, &stride, pb);
-	pipe_buffer_reference(&pb, NULL);
-	return pt;
-}
-
-static boolean
-nouveau_drm_name_from_pt(struct drm_api *api, struct pipe_screen *pscreen,
-			 struct pipe_texture *pt, unsigned *stride,
-			 unsigned *handle)
-{
-	struct nouveau_miptree *mt = nouveau_miptree(pt);
-
-	if (!mt || !mt->bo)
-		return false;
-
-	return nouveau_bo_handle_get(mt->bo, handle) == 0;
-}
-
-static boolean
-nouveau_drm_handle_from_pt(struct drm_api *api, struct pipe_screen *pscreen,
-			   struct pipe_texture *pt, unsigned *stride,
-			   unsigned *handle)
-{
-	struct nouveau_miptree *mt = nouveau_miptree(pt);
-
-	if (!mt || !mt->bo)
-		return false;
-
-	*handle = mt->bo->handle;
-	*stride = util_format_get_stride(mt->base.format, mt->base.width0);
-	return true;
-}
-
 struct drm_api drm_api_hooks = {
 	.name = "nouveau",
 	.driver_name = "nouveau",
 	.create_screen = nouveau_drm_create_screen,
-	.texture_from_shared_handle = nouveau_drm_pt_from_name,
-	.shared_handle_from_texture = nouveau_drm_name_from_pt,
-	.local_handle_from_texture = nouveau_drm_handle_from_pt,
 };
 
 struct drm_api *
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
index e5c69199330..daa032af6f2 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_buffer.c
@@ -264,6 +264,75 @@ static int radeon_fence_finish(struct pipe_winsys *ws,
     return 0;
 }
 
+/* Create a buffer from a handle. */
+static struct pipe_buffer* radeon_buffer_from_handle(struct radeon_winsys *radeon_ws,
+                                                     struct pipe_screen *screen,
+                                                     struct winsys_handle *whandle,
+                                                     unsigned *stride)
+{
+    struct radeon_bo_manager* bom = radeon_ws->priv->bom;
+    struct radeon_pipe_buffer* radeon_buffer;
+    struct radeon_bo* bo = NULL;
+
+    bo = radeon_bo_open(bom, whandle->handle, 0, 0, 0, 0);
+    if (bo == NULL) {
+        return NULL;
+    }
+
+    radeon_buffer = CALLOC_STRUCT(radeon_pipe_buffer);
+    if (radeon_buffer == NULL) {
+        radeon_bo_unref(bo);
+        return NULL;
+    }
+
+    pipe_reference_init(&radeon_buffer->base.reference, 1);
+    radeon_buffer->base.screen = screen;
+    radeon_buffer->base.usage = PIPE_BUFFER_USAGE_PIXEL;
+    radeon_buffer->bo = bo;
+
+    *stride = whandle->stride;
+
+    return &radeon_buffer->base;
+}
+
+static boolean radeon_buffer_get_handle(struct radeon_winsys *radeon_ws,
+                                        struct pipe_buffer *buffer,
+                                        unsigned stride,
+                                        struct winsys_handle *whandle)
+{
+    int retval, fd;
+    struct drm_gem_flink flink;
+    struct radeon_pipe_buffer* radeon_buffer;
+
+    radeon_buffer = (struct radeon_pipe_buffer*)buffer;
+
+
+    if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+        if (!radeon_buffer->flinked) {
+            fd = radeon_ws->priv->fd;
+
+            flink.handle = radeon_buffer->bo->handle;
+
+            retval = ioctl(fd, DRM_IOCTL_GEM_FLINK, &flink);
+            if (retval) {
+                debug_printf("radeon: DRM_IOCTL_GEM_FLINK failed, error %d\n",
+                             retval);
+                return FALSE;
+            }
+
+            radeon_buffer->flink = flink.name;
+            radeon_buffer->flinked = TRUE;
+        }
+
+        whandle->handle = radeon_buffer->flink;
+    } else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+        whandle->handle = ((struct radeon_pipe_buffer*)buffer)->bo->handle;
+    }
+    whandle->stride = stride;
+
+    return TRUE;
+}
+
 struct radeon_winsys* radeon_pipe_winsys(int fd)
 {
     struct radeon_winsys* radeon_ws;
@@ -298,6 +367,8 @@ struct radeon_winsys* radeon_pipe_winsys(int fd)
     radeon_ws->base.get_name = radeon_get_name;
 
     radeon_ws->buffer_set_tiling = radeon_buffer_set_tiling;
+    radeon_ws->buffer_from_handle = radeon_buffer_from_handle;
+    radeon_ws->buffer_get_handle = radeon_buffer_get_handle;
 
     return radeon_ws;
 }
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.c b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
index e817a26da6d..97edb6a47e3 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
@@ -148,124 +148,6 @@ struct pipe_screen* radeon_create_screen(struct drm_api* api,
     }
 }
 
-
-boolean radeon_buffer_from_texture(struct drm_api* api,
-                                   struct pipe_screen* screen,
-                                   struct pipe_texture* texture,
-                                   struct pipe_buffer** buffer,
-                                   unsigned* stride)
-{
-    /* XXX fix this */
-    return r300_get_texture_buffer(screen, texture, buffer, stride);
-}
-
-/* Create a buffer from a handle. */
-/* XXX what's up with name? */
-struct pipe_buffer* radeon_buffer_from_handle(struct drm_api* api,
-                                              struct pipe_screen* screen,
-                                              const char* name,
-                                              unsigned handle)
-{
-    struct radeon_bo_manager* bom =
-        ((struct radeon_winsys*)screen->winsys)->priv->bom;
-    struct radeon_pipe_buffer* radeon_buffer;
-    struct radeon_bo* bo = NULL;
-
-    bo = radeon_bo_open(bom, handle, 0, 0, 0, 0);
-    if (bo == NULL) {
-        return NULL;
-    }
-
-    radeon_buffer = CALLOC_STRUCT(radeon_pipe_buffer);
-    if (radeon_buffer == NULL) {
-        radeon_bo_unref(bo);
-        return NULL;
-    }
-
-    pipe_reference_init(&radeon_buffer->base.reference, 1);
-    radeon_buffer->base.screen = screen;
-    radeon_buffer->base.usage = PIPE_BUFFER_USAGE_PIXEL;
-    radeon_buffer->bo = bo;
-    return &radeon_buffer->base;
-}
-
-static struct pipe_texture*
-radeon_texture_from_shared_handle(struct drm_api *api,
-                                  struct pipe_screen *screen,
-                                  struct pipe_texture *templ,
-                                  const char *name,
-                                  unsigned stride,
-                                  unsigned handle)
-{
-    struct pipe_buffer *buffer;
-    struct pipe_texture *blanket;
-
-    buffer = radeon_buffer_from_handle(api, screen, name, handle);
-    if (!buffer) {
-        return NULL;
-    }
-
-    blanket = screen->texture_blanket(screen, templ, &stride, buffer);
-
-    pipe_buffer_reference(&buffer, NULL);
-
-    return blanket;
-}
-
-static boolean radeon_shared_handle_from_texture(struct drm_api *api,
-                                                 struct pipe_screen *screen,
-                                                 struct pipe_texture *texture,
-                                                 unsigned *stride,
-                                                 unsigned *handle)
-{
-    int retval, fd;
-    struct drm_gem_flink flink;
-    struct radeon_pipe_buffer* radeon_buffer;
-    struct pipe_buffer *buffer = NULL;
-
-    if (!radeon_buffer_from_texture(api, screen, texture, &buffer, stride)) {
-        return FALSE;
-    }
-
-    radeon_buffer = (struct radeon_pipe_buffer*)buffer;
-    if (!radeon_buffer->flinked) {
-        fd = ((struct radeon_winsys*)screen->winsys)->priv->fd;
-
-        flink.handle = radeon_buffer->bo->handle;
-
-        retval = ioctl(fd, DRM_IOCTL_GEM_FLINK, &flink);
-        if (retval) {
-            debug_printf("radeon: DRM_IOCTL_GEM_FLINK failed, error %d\n",
-                    retval);
-            return FALSE;
-        }
-
-        radeon_buffer->flink = flink.name;
-        radeon_buffer->flinked = TRUE;
-    }
-
-    *handle = radeon_buffer->flink;
-    return TRUE;
-}
-
-static boolean radeon_local_handle_from_texture(struct drm_api *api,
-                                                struct pipe_screen *screen,
-                                                struct pipe_texture *texture,
-                                                unsigned *stride,
-                                                unsigned *handle)
-{
-    struct pipe_buffer *buffer = NULL;
-    if (!radeon_buffer_from_texture(api, screen, texture, &buffer, stride)) {
-        return FALSE;
-    }
-
-    *handle = ((struct radeon_pipe_buffer*)buffer)->bo->handle;
-
-    pipe_buffer_reference(&buffer, NULL);
-
-    return TRUE;
-}
-
 static void radeon_drm_api_destroy(struct drm_api *api)
 {
     return;
@@ -275,9 +157,6 @@ struct drm_api drm_api_hooks = {
     .name = "radeon",
     .driver_name = "radeon",
     .create_screen = radeon_create_screen,
-    .texture_from_shared_handle = radeon_texture_from_shared_handle,
-    .shared_handle_from_texture = radeon_shared_handle_from_texture,
-    .local_handle_from_texture = radeon_local_handle_from_texture,
     .destroy = radeon_drm_api_destroy,
 };
 
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.h b/src/gallium/winsys/drm/radeon/core/radeon_drm.h
index f62a9b80485..78451b6f011 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.h
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.h
@@ -37,28 +37,6 @@ struct pipe_screen* radeon_create_screen(struct drm_api* api,
                                          int drmFB,
 					 struct drm_create_screen_arg *arg);
 
-
-boolean radeon_buffer_from_texture(struct drm_api* api,
-                                   struct pipe_screen* screen,
-                                   struct pipe_texture* texture,
-                                   struct pipe_buffer** buffer,
-                                   unsigned* stride);
-
-struct pipe_buffer* radeon_buffer_from_handle(struct drm_api* api,
-                                              struct pipe_screen* screen,
-                                              const char* name,
-                                              unsigned handle);
-
-boolean radeon_handle_from_buffer(struct drm_api* api,
-                                  struct pipe_screen* screen,
-                                  struct pipe_buffer* buffer,
-                                  unsigned* handle);
-
-boolean radeon_global_handle_from_buffer(struct drm_api* api,
-                                         struct pipe_screen* screen,
-                                         struct pipe_buffer* buffer,
-                                         unsigned* handle);
-
 void radeon_destroy_drm_api(struct drm_api* api);
 
 /* Guess at whether this chipset should use r300g.
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_winsys.h b/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
index 4901080ca7b..37eeb459791 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
+++ b/src/gallium/winsys/drm/radeon/core/radeon_winsys.h
@@ -56,6 +56,18 @@ struct radeon_winsys {
     /* VRAM size. */
     uint32_t vram_size;
 
+    /* Create a buffer from a winsys handle. */
+    struct pipe_buffer *(*buffer_from_handle)(struct radeon_winsys *winsys,
+                                              struct pipe_screen *screen,
+                                              struct winsys_handle *whandle,
+                                              unsigned *stride);
+
+    /* Get the handle from a buffer. */
+    boolean (*buffer_get_handle)(struct radeon_winsys *winsys,
+                                 struct pipe_buffer *buffer,
+                                 unsigned stride,
+                                 struct winsys_handle *whandle);
+
     /* Add a pipe_buffer to the list of buffer objects to validate. */
     boolean (*add_buffer)(struct radeon_winsys* winsys,
                           struct pipe_buffer* pbuffer,
diff --git a/src/gallium/winsys/drm/radeon/python/README b/src/gallium/winsys/drm/radeon/python/README
deleted file mode 100644
index 339836a592b..00000000000
--- a/src/gallium/winsys/drm/radeon/python/README
+++ /dev/null
@@ -1,15 +0,0 @@
-Python bindings for the radeon gallium driver.
-
-
-See gallium/src/gallium/state_trackers/python/README for more information.
-
-
-Build as:
-
-  scons debug=1 statetrackers=python winsys=drm/radeon/python
-  
-Run as:
-
-  export PYTHONPATH=$PWD/build/linux-x86-debug/gallium/winsys/drm/radeon/python:$PWD/build/linux-x86-debug/gallium/state_trackers/python 
- 
-  python progs/gallium/python/samples/tri.py
diff --git a/src/gallium/winsys/drm/radeon/python/SConscript b/src/gallium/winsys/drm/radeon/python/SConscript
deleted file mode 100644
index 91cae986975..00000000000
--- a/src/gallium/winsys/drm/radeon/python/SConscript
+++ /dev/null
@@ -1,33 +0,0 @@
-import os.path
-
-Import('*')
-
-if env['platform'] == 'linux':
-
-    env = env.Clone()
-    
-    env.Tool('python')
-    
-    env.ParseConfig('pkg-config --cflags --libs libdrm')
-    
-    env.Prepend(CPPPATH = [
-        '#src/gallium/state_trackers/python',
-        '../core',
-    ])
-    
-    drivers = [
-        softpipe,
-        radeon,
-        trace,
-    ]
-    
-    sources = [
-        'radeon_hardpipe_winsys.c',
-        'xf86dri.c',
-    ]
-    
-    env.SharedLibrary(
-        target ='_gallium',
-        source = sources,
-        LIBS = [pyst] + drivers + gallium + env['LIBS'],
-    )
diff --git a/src/gallium/winsys/drm/radeon/python/radeon_hardpipe_winsys.c b/src/gallium/winsys/drm/radeon/python/radeon_hardpipe_winsys.c
deleted file mode 100644
index fc63081a4cc..00000000000
--- a/src/gallium/winsys/drm/radeon/python/radeon_hardpipe_winsys.c
+++ /dev/null
@@ -1,132 +0,0 @@
- /**************************************************************************
- * 
- * Copyright 2009 VMware, Inc.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL VMWARE 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 <stdio.h>
-#include <string.h>
-#include <stdlib.h>
-
-#include <X11/X.h>
-#include <X11/Xlib.h>
-#include <drm/drm.h>
-
-#include "pipe/p_screen.h"
-#include "pipe/p_context.h"
-
-#include "st_winsys.h"
-
-#include "radeon_winsys.h"
-
-#include "xf86dri.h"
-
-
-/* XXX: Force init_gallium symbol to be linked */
-extern void init_gallium(void);
-void (*force_init_gallium_linkage)(void) = &init_gallium;
-
-
-static struct pipe_screen *
-radeon_hardpipe_screen_create(void)
-{
-   Display *dpy;
-   Window rootWin;
-   XWindowAttributes winAttr;
-   int isCapable;
-   int screen;
-   char *driverName;
-   char *curBusID;
-   unsigned magic;
-   int ddxDriverMajor;
-   int ddxDriverMinor;
-   int ddxDriverPatch;
-   drm_handle_t sAreaOffset;
-   int ret;
-   int drmFD;
-   drm_context_t hHWContext;
-   XID id;
-
-   dpy = XOpenDisplay(":0");
-   if (!dpy) {
-      fprintf(stderr, "Open Display Failed\n");
-      return NULL;
-   }
-
-   screen = DefaultScreen(dpy);
-   rootWin = RootWindow(dpy, screen);
-   XGetWindowAttributes(dpy, rootWin, &winAttr);
-
-   ret = uniDRIQueryDirectRenderingCapable(dpy, screen, &isCapable);
-   if (!ret || !isCapable) {
-      fprintf(stderr, "No DRI on this display:sceen\n");
-      goto error;
-   }
-
-   if (!uniDRIOpenConnection(dpy, screen, &sAreaOffset,
-                             &curBusID)) {
-      fprintf(stderr, "Could not open DRI connection.\n");
-      goto error;
-   }
-
-   if (!uniDRIGetClientDriverName(dpy, screen, &ddxDriverMajor,
-                                  &ddxDriverMinor, &ddxDriverPatch,
-                                  &driverName)) {
-      fprintf(stderr, "Could not get DRI driver name.\n");
-      goto error;
-   }
-
-   if ((drmFD = drmOpen(NULL, curBusID)) < 0) {
-      perror("DRM Device could not be opened");
-      goto error;
-   }
-
-   drmGetMagic(drmFD, &magic);
-   if (!uniDRIAuthConnection(dpy, screen, magic)) {
-      fprintf(stderr, "Could not get X server to authenticate us.\n");
-      goto error;
-   }
-
-   if (!uniDRICreateContext(dpy, screen, winAttr.visual,
-                            &id, &hHWContext)) {
-      fprintf(stderr, "Could not create DRI context.\n");
-      goto error;
-   }
-
-   /* FIXME: create a radeon pipe_screen from drmFD and hHWContext */
-
-   return NULL;
-   
-error:
-   return NULL;
-}
-
-
-
-
-const struct st_winsys st_hardpipe_winsys = {
-   &radeon_hardpipe_screen_create,
-};
-
diff --git a/src/gallium/winsys/drm/radeon/python/xf86dri.c b/src/gallium/winsys/drm/radeon/python/xf86dri.c
deleted file mode 100644
index 1736f1e54f8..00000000000
--- a/src/gallium/winsys/drm/radeon/python/xf86dri.c
+++ /dev/null
@@ -1,605 +0,0 @@
-/**************************************************************************
-
-Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
-Copyright 2000 VA Linux Systems, 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 PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
-ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
-TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
-SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <martin@valinux.com>
- *   Jens Owen <jens@tungstengraphics.com>
- *   Rickard E. (Rik) Faith <faith@valinux.com>
- *
- */
-
-/* THIS IS NOT AN X CONSORTIUM STANDARD */
-
-#ifdef HAVE_CONFIG_H
-#include "config.h"
-#endif
-
-#define NEED_REPLIES
-#include <X11/Xlibint.h>
-#include <X11/extensions/Xext.h>
-#include <X11/extensions/extutil.h>
-#include "xf86dristr.h"
-
-static XExtensionInfo _xf86dri_info_data;
-static XExtensionInfo *xf86dri_info = &_xf86dri_info_data;
-static char xf86dri_extension_name[] = XF86DRINAME;
-
-#define uniDRICheckExtension(dpy,i,val) \
-  XextCheckExtension (dpy, i, xf86dri_extension_name, val)
-
-/*****************************************************************************
- *                                                                           *
- *			   private utility routines                          *
- *                                                                           *
- *****************************************************************************/
-
-static int close_display(Display * dpy, XExtCodes * extCodes);
-static /* const */ XExtensionHooks xf86dri_extension_hooks = {
-    NULL,			       /* create_gc */
-    NULL,			       /* copy_gc */
-    NULL,			       /* flush_gc */
-    NULL,			       /* free_gc */
-    NULL,			       /* create_font */
-    NULL,			       /* free_font */
-    close_display,		       /* close_display */
-    NULL,			       /* wire_to_event */
-    NULL,			       /* event_to_wire */
-    NULL,			       /* error */
-    NULL,			       /* error_string */
-};
-
-static
-XEXT_GENERATE_FIND_DISPLAY(find_display, xf86dri_info,
-			   xf86dri_extension_name, &xf86dri_extension_hooks,
-			   0, NULL)
-
-    static XEXT_GENERATE_CLOSE_DISPLAY(close_display, xf86dri_info)
-
-/*****************************************************************************
- *                                                                           *
- *		    public XFree86-DRI Extension routines                    *
- *                                                                           *
- *****************************************************************************/
-#if 0
-#include <stdio.h>
-#define TRACE(msg)  fprintf(stderr,"uniDRI%s\n", msg);
-#else
-#define TRACE(msg)
-#endif
-    Bool uniDRIQueryExtension(dpy, event_basep, error_basep)
-    Display *dpy;
-    int *event_basep, *error_basep;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-
-    TRACE("QueryExtension...");
-    if (XextHasExtension(info)) {
-	*event_basep = info->codes->first_event;
-	*error_basep = info->codes->first_error;
-	TRACE("QueryExtension... return True");
-	return True;
-    } else {
-	TRACE("QueryExtension... return False");
-	return False;
-    }
-}
-
-Bool
-uniDRIQueryVersion(dpy, majorVersion, minorVersion, patchVersion)
-    Display *dpy;
-    int *majorVersion;
-    int *minorVersion;
-    int *patchVersion;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIQueryVersionReply rep;
-    xXF86DRIQueryVersionReq *req;
-
-    TRACE("QueryVersion...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIQueryVersion, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIQueryVersion;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("QueryVersion... return False");
-	return False;
-    }
-    *majorVersion = rep.majorVersion;
-    *minorVersion = rep.minorVersion;
-    *patchVersion = rep.patchVersion;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("QueryVersion... return True");
-    return True;
-}
-
-Bool
-uniDRIQueryDirectRenderingCapable(dpy, screen, isCapable)
-    Display *dpy;
-    int screen;
-    Bool *isCapable;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIQueryDirectRenderingCapableReply rep;
-    xXF86DRIQueryDirectRenderingCapableReq *req;
-
-    TRACE("QueryDirectRenderingCapable...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIQueryDirectRenderingCapable, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIQueryDirectRenderingCapable;
-    req->screen = screen;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("QueryDirectRenderingCapable... return False");
-	return False;
-    }
-    *isCapable = rep.isCapable;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("QueryDirectRenderingCapable... return True");
-    return True;
-}
-
-Bool
-uniDRIOpenConnection(dpy, screen, hSAREA, busIdString)
-    Display *dpy;
-    int screen;
-    drm_handle_t *hSAREA;
-    char **busIdString;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIOpenConnectionReply rep;
-    xXF86DRIOpenConnectionReq *req;
-
-    TRACE("OpenConnection...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIOpenConnection, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIOpenConnection;
-    req->screen = screen;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("OpenConnection... return False");
-	return False;
-    }
-
-    *hSAREA = rep.hSAREALow;
-#ifdef LONG64
-    if (sizeof(drm_handle_t) == 8) {
-	*hSAREA |= ((unsigned long)rep.hSAREAHigh) << 32;
-    }
-#endif
-    if (rep.length) {
-	if (!(*busIdString = (char *)Xcalloc(rep.busIdStringLength + 1, 1))) {
-	    _XEatData(dpy, ((rep.busIdStringLength + 3) & ~3));
-	    UnlockDisplay(dpy);
-	    SyncHandle();
-	    TRACE("OpenConnection... return False");
-	    return False;
-	}
-	_XReadPad(dpy, *busIdString, rep.busIdStringLength);
-    } else {
-	*busIdString = NULL;
-    }
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("OpenConnection... return True");
-    return True;
-}
-
-Bool
-uniDRIAuthConnection(dpy, screen, magic)
-    Display *dpy;
-    int screen;
-    drm_magic_t magic;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIAuthConnectionReq *req;
-    xXF86DRIAuthConnectionReply rep;
-
-    TRACE("AuthConnection...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIAuthConnection, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIAuthConnection;
-    req->screen = screen;
-    req->magic = magic;
-    rep.authenticated = 0;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse) || !rep.authenticated) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("AuthConnection... return False");
-	return False;
-    }
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("AuthConnection... return True");
-    return True;
-}
-
-Bool
-uniDRICloseConnection(dpy, screen)
-    Display *dpy;
-    int screen;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRICloseConnectionReq *req;
-
-    TRACE("CloseConnection...");
-
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRICloseConnection, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRICloseConnection;
-    req->screen = screen;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("CloseConnection... return True");
-    return True;
-}
-
-Bool
-uniDRIGetClientDriverName(dpy, screen, ddxDriverMajorVersion,
-			  ddxDriverMinorVersion, ddxDriverPatchVersion,
-			  clientDriverName)
-    Display *dpy;
-    int screen;
-    int *ddxDriverMajorVersion;
-    int *ddxDriverMinorVersion;
-    int *ddxDriverPatchVersion;
-    char **clientDriverName;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIGetClientDriverNameReply rep;
-    xXF86DRIGetClientDriverNameReq *req;
-
-    TRACE("GetClientDriverName...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIGetClientDriverName, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIGetClientDriverName;
-    req->screen = screen;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("GetClientDriverName... return False");
-	return False;
-    }
-
-    *ddxDriverMajorVersion = rep.ddxDriverMajorVersion;
-    *ddxDriverMinorVersion = rep.ddxDriverMinorVersion;
-    *ddxDriverPatchVersion = rep.ddxDriverPatchVersion;
-
-    if (rep.length) {
-	if (!(*clientDriverName =
-	      (char *)Xcalloc(rep.clientDriverNameLength + 1, 1))) {
-	    _XEatData(dpy, ((rep.clientDriverNameLength + 3) & ~3));
-	    UnlockDisplay(dpy);
-	    SyncHandle();
-	    TRACE("GetClientDriverName... return False");
-	    return False;
-	}
-	_XReadPad(dpy, *clientDriverName, rep.clientDriverNameLength);
-    } else {
-	*clientDriverName = NULL;
-    }
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("GetClientDriverName... return True");
-    return True;
-}
-
-Bool
-uniDRICreateContextWithConfig(dpy, screen, configID, context, hHWContext)
-    Display *dpy;
-    int screen;
-    int configID;
-    XID *context;
-    drm_context_t *hHWContext;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRICreateContextReply rep;
-    xXF86DRICreateContextReq *req;
-
-    TRACE("CreateContext...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRICreateContext, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRICreateContext;
-    req->visual = configID;
-    req->screen = screen;
-    *context = XAllocID(dpy);
-    req->context = *context;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("CreateContext... return False");
-	return False;
-    }
-    *hHWContext = rep.hHWContext;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("CreateContext... return True");
-    return True;
-}
-
-Bool
-uniDRICreateContext(dpy, screen, visual, context, hHWContext)
-    Display *dpy;
-    int screen;
-    Visual *visual;
-    XID *context;
-    drm_context_t *hHWContext;
-{
-    return uniDRICreateContextWithConfig(dpy, screen, visual->visualid,
-					 context, hHWContext);
-}
-
-Bool
-uniDRIDestroyContext(Display * ndpy, int screen, XID context)
-{
-    Display *const dpy = (Display *) ndpy;
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIDestroyContextReq *req;
-
-    TRACE("DestroyContext...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIDestroyContext, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIDestroyContext;
-    req->screen = screen;
-    req->context = context;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("DestroyContext... return True");
-    return True;
-}
-
-Bool
-uniDRICreateDrawable(Display * ndpy, int screen,
-		     Drawable drawable, drm_drawable_t * hHWDrawable)
-{
-    Display *const dpy = (Display *) ndpy;
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRICreateDrawableReply rep;
-    xXF86DRICreateDrawableReq *req;
-
-    TRACE("CreateDrawable...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRICreateDrawable, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRICreateDrawable;
-    req->screen = screen;
-    req->drawable = drawable;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("CreateDrawable... return False");
-	return False;
-    }
-    *hHWDrawable = rep.hHWDrawable;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("CreateDrawable... return True");
-    return True;
-}
-
-Bool
-uniDRIDestroyDrawable(Display * ndpy, int screen, Drawable drawable)
-{
-    Display *const dpy = (Display *) ndpy;
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIDestroyDrawableReq *req;
-
-    TRACE("DestroyDrawable...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIDestroyDrawable, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIDestroyDrawable;
-    req->screen = screen;
-    req->drawable = drawable;
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("DestroyDrawable... return True");
-    return True;
-}
-
-Bool
-uniDRIGetDrawableInfo(Display * dpy, int screen, Drawable drawable,
-		      unsigned int *index, unsigned int *stamp,
-		      int *X, int *Y, int *W, int *H,
-		      int *numClipRects, drm_clip_rect_t ** pClipRects,
-		      int *backX, int *backY,
-		      int *numBackClipRects,
-		      drm_clip_rect_t ** pBackClipRects)
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIGetDrawableInfoReply rep;
-    xXF86DRIGetDrawableInfoReq *req;
-    int total_rects;
-
-    TRACE("GetDrawableInfo...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIGetDrawableInfo, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIGetDrawableInfo;
-    req->screen = screen;
-    req->drawable = drawable;
-
-    if (!_XReply(dpy, (xReply *) & rep, 1, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("GetDrawableInfo... return False");
-	return False;
-    }
-    *index = rep.drawableTableIndex;
-    *stamp = rep.drawableTableStamp;
-    *X = (int)rep.drawableX;
-    *Y = (int)rep.drawableY;
-    *W = (int)rep.drawableWidth;
-    *H = (int)rep.drawableHeight;
-    *numClipRects = rep.numClipRects;
-    total_rects = *numClipRects;
-
-    *backX = rep.backX;
-    *backY = rep.backY;
-    *numBackClipRects = rep.numBackClipRects;
-    total_rects += *numBackClipRects;
-
-#if 0
-    /* Because of the fix in Xserver/GL/dri/xf86dri.c, this check breaks
-     * backwards compatibility (Because of the >> 2 shift) but the fix
-     * enables multi-threaded apps to work.
-     */
-    if (rep.length != ((((SIZEOF(xXF86DRIGetDrawableInfoReply) -
-			  SIZEOF(xGenericReply) +
-			  total_rects * sizeof(drm_clip_rect_t)) +
-			 3) & ~3) >> 2)) {
-	_XEatData(dpy, rep.length);
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("GetDrawableInfo... return False");
-	return False;
-    }
-#endif
-
-    if (*numClipRects) {
-	int len = sizeof(drm_clip_rect_t) * (*numClipRects);
-
-	*pClipRects = (drm_clip_rect_t *) Xcalloc(len, 1);
-	if (*pClipRects)
-	    _XRead(dpy, (char *)*pClipRects, len);
-    } else {
-	*pClipRects = NULL;
-    }
-
-    if (*numBackClipRects) {
-	int len = sizeof(drm_clip_rect_t) * (*numBackClipRects);
-
-	*pBackClipRects = (drm_clip_rect_t *) Xcalloc(len, 1);
-	if (*pBackClipRects)
-	    _XRead(dpy, (char *)*pBackClipRects, len);
-    } else {
-	*pBackClipRects = NULL;
-    }
-
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("GetDrawableInfo... return True");
-    return True;
-}
-
-Bool
-uniDRIGetDeviceInfo(dpy, screen, hFrameBuffer,
-		    fbOrigin, fbSize, fbStride, devPrivateSize, pDevPrivate)
-    Display *dpy;
-    int screen;
-    drm_handle_t *hFrameBuffer;
-    int *fbOrigin;
-    int *fbSize;
-    int *fbStride;
-    int *devPrivateSize;
-    void **pDevPrivate;
-{
-    XExtDisplayInfo *info = find_display(dpy);
-    xXF86DRIGetDeviceInfoReply rep;
-    xXF86DRIGetDeviceInfoReq *req;
-
-    TRACE("GetDeviceInfo...");
-    uniDRICheckExtension(dpy, info, False);
-
-    LockDisplay(dpy);
-    GetReq(XF86DRIGetDeviceInfo, req);
-    req->reqType = info->codes->major_opcode;
-    req->driReqType = X_XF86DRIGetDeviceInfo;
-    req->screen = screen;
-    if (!_XReply(dpy, (xReply *) & rep, 0, xFalse)) {
-	UnlockDisplay(dpy);
-	SyncHandle();
-	TRACE("GetDeviceInfo... return False");
-	return False;
-    }
-
-    *hFrameBuffer = rep.hFrameBufferLow;
-#ifdef LONG64
-    if (sizeof(drm_handle_t) == 8) {
-	*hFrameBuffer |= ((unsigned long)rep.hFrameBufferHigh) << 32;
-    }
-#endif
-
-    *fbOrigin = rep.framebufferOrigin;
-    *fbSize = rep.framebufferSize;
-    *fbStride = rep.framebufferStride;
-    *devPrivateSize = rep.devPrivateSize;
-
-    if (rep.length) {
-	if (!(*pDevPrivate = (void *)Xcalloc(rep.devPrivateSize, 1))) {
-	    _XEatData(dpy, ((rep.devPrivateSize + 3) & ~3));
-	    UnlockDisplay(dpy);
-	    SyncHandle();
-	    TRACE("GetDeviceInfo... return False");
-	    return False;
-	}
-	_XRead(dpy, (char *)*pDevPrivate, rep.devPrivateSize);
-    } else {
-	*pDevPrivate = NULL;
-    }
-
-    UnlockDisplay(dpy);
-    SyncHandle();
-    TRACE("GetDeviceInfo... return True");
-    return True;
-}
diff --git a/src/gallium/winsys/drm/radeon/python/xf86dri.h b/src/gallium/winsys/drm/radeon/python/xf86dri.h
deleted file mode 100644
index bf6de37d9dd..00000000000
--- a/src/gallium/winsys/drm/radeon/python/xf86dri.h
+++ /dev/null
@@ -1,123 +0,0 @@
-/**************************************************************************
-
-Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
-Copyright 2000 VA Linux Systems, 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 PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
-ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
-TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
-SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/**
- * \file xf86dri.h
- * Protocol numbers and function prototypes for DRI X protocol.
- *
- * \author Kevin E. Martin <martin@valinux.com>
- * \author Jens Owen <jens@tungstengraphics.com>
- * \author Rickard E. (Rik) Faith <faith@valinux.com>
- */
-
-#ifndef _XF86DRI_H_
-#define _XF86DRI_H_
-
-#include <stdint.h>
-#include <X11/Xfuncproto.h>
-#include <drm/drm.h>
-
-#define X_XF86DRIQueryVersion			0
-#define X_XF86DRIQueryDirectRenderingCapable	1
-#define X_XF86DRIOpenConnection			2
-#define X_XF86DRICloseConnection		3
-#define X_XF86DRIGetClientDriverName		4
-#define X_XF86DRICreateContext			5
-#define X_XF86DRIDestroyContext			6
-#define X_XF86DRICreateDrawable			7
-#define X_XF86DRIDestroyDrawable		8
-#define X_XF86DRIGetDrawableInfo		9
-#define X_XF86DRIGetDeviceInfo			10
-#define X_XF86DRIAuthConnection                 11
-#define X_XF86DRIOpenFullScreen                 12	/* Deprecated */
-#define X_XF86DRICloseFullScreen                13	/* Deprecated */
-
-#define XF86DRINumberEvents		0
-
-#define XF86DRIClientNotLocal		0
-#define XF86DRIOperationNotSupported	1
-#define XF86DRINumberErrors		(XF86DRIOperationNotSupported + 1)
-
-#ifndef _XF86DRI_SERVER_
-
-_XFUNCPROTOBEGIN
-    Bool uniDRIQueryExtension(Display * dpy, int *event_base,
-			      int *error_base);
-
-Bool uniDRIQueryVersion(Display * dpy, int *majorVersion, int *minorVersion,
-			int *patchVersion);
-
-Bool uniDRIQueryDirectRenderingCapable(Display * dpy, int screen,
-				       Bool * isCapable);
-
-Bool uniDRIOpenConnection(Display * dpy, int screen, drm_handle_t * hSAREA,
-			  char **busIDString);
-
-Bool uniDRIAuthConnection(Display * dpy, int screen, drm_magic_t magic);
-
-Bool uniDRICloseConnection(Display * dpy, int screen);
-
-Bool uniDRIGetClientDriverName(Display * dpy, int screen,
-			       int *ddxDriverMajorVersion,
-			       int *ddxDriverMinorVersion,
-			       int *ddxDriverPatchVersion,
-			       char **clientDriverName);
-
-Bool uniDRICreateContext(Display * dpy, int screen, Visual * visual,
-			 XID * ptr_to_returned_context_id,
-			 drm_context_t * hHWContext);
-
-Bool uniDRICreateContextWithConfig(Display * dpy, int screen, int configID,
-				   XID * ptr_to_returned_context_id,
-				   drm_context_t * hHWContext);
-
-extern Bool uniDRIDestroyContext(Display * dpy, int screen, XID context_id);
-
-extern Bool uniDRICreateDrawable(Display * dpy, int screen,
-				 Drawable drawable,
-				 drm_drawable_t * hHWDrawable);
-
-extern Bool uniDRIDestroyDrawable(Display * dpy, int screen,
-				  Drawable drawable);
-
-Bool uniDRIGetDrawableInfo(Display * dpy, int screen, Drawable drawable,
-			   unsigned int *index, unsigned int *stamp,
-			   int *X, int *Y, int *W, int *H,
-			   int *numClipRects, drm_clip_rect_t ** pClipRects,
-			   int *backX, int *backY,
-			   int *numBackClipRects,
-			   drm_clip_rect_t ** pBackClipRects);
-
-Bool uniDRIGetDeviceInfo(Display * dpy, int screen,
-			 drm_handle_t * hFrameBuffer, int *fbOrigin,
-			 int *fbSize, int *fbStride, int *devPrivateSize,
-			 void **pDevPrivate);
-
-_XFUNCPROTOEND
-#endif /* _XF86DRI_SERVER_ */
-#endif /* _XF86DRI_H_ */
diff --git a/src/gallium/winsys/drm/radeon/python/xf86dristr.h b/src/gallium/winsys/drm/radeon/python/xf86dristr.h
deleted file mode 100644
index d8989963609..00000000000
--- a/src/gallium/winsys/drm/radeon/python/xf86dristr.h
+++ /dev/null
@@ -1,389 +0,0 @@
-/**************************************************************************
-
-Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
-Copyright 2000 VA Linux Systems, 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 PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
-ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
-TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
-SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <martin@valinux.com>
- *   Jens Owen <jens@tungstengraphics.com>
- *   Rickard E. (Rik) Fiath <faith@valinux.com>
- *
- */
-
-#ifndef _XF86DRISTR_H_
-#define _XF86DRISTR_H_
-
-#include "xf86dri.h"
-
-#define XF86DRINAME "XFree86-DRI"
-
-/* The DRI version number.  This was originally set to be the same of the
- * XFree86 version number.  However, this version is really indepedent of
- * the XFree86 version.
- *
- * Version History:
- *    4.0.0: Original
- *    4.0.1: Patch to bump clipstamp when windows are destroyed, 28 May 02
- *    4.1.0: Add transition from single to multi in DRMInfo rec, 24 Jun 02
- */
-#define XF86DRI_MAJOR_VERSION	4
-#define XF86DRI_MINOR_VERSION	1
-#define XF86DRI_PATCH_VERSION	0
-
-typedef struct _XF86DRIQueryVersion
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIQueryVersion */
-    CARD16 length B16;
-} xXF86DRIQueryVersionReq;
-
-#define sz_xXF86DRIQueryVersionReq	4
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD16 majorVersion B16;	       /* major version of DRI protocol */
-    CARD16 minorVersion B16;	       /* minor version of DRI protocol */
-    CARD32 patchVersion B32;	       /* patch version of DRI protocol */
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRIQueryVersionReply;
-
-#define sz_xXF86DRIQueryVersionReply	32
-
-typedef struct _XF86DRIQueryDirectRenderingCapable
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* X_DRIQueryDirectRenderingCapable */
-    CARD16 length B16;
-    CARD32 screen B32;
-} xXF86DRIQueryDirectRenderingCapableReq;
-
-#define sz_xXF86DRIQueryDirectRenderingCapableReq	8
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    BOOL isCapable;
-    BOOL pad2;
-    BOOL pad3;
-    BOOL pad4;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-    CARD32 pad7 B32;
-    CARD32 pad8 B32;
-    CARD32 pad9 B32;
-} xXF86DRIQueryDirectRenderingCapableReply;
-
-#define sz_xXF86DRIQueryDirectRenderingCapableReply	32
-
-typedef struct _XF86DRIOpenConnection
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIOpenConnection */
-    CARD16 length B16;
-    CARD32 screen B32;
-} xXF86DRIOpenConnectionReq;
-
-#define sz_xXF86DRIOpenConnectionReq	8
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 hSAREALow B32;
-    CARD32 hSAREAHigh B32;
-    CARD32 busIdStringLength B32;
-    CARD32 pad6 B32;
-    CARD32 pad7 B32;
-    CARD32 pad8 B32;
-} xXF86DRIOpenConnectionReply;
-
-#define sz_xXF86DRIOpenConnectionReply	32
-
-typedef struct _XF86DRIAuthConnection
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRICloseConnection */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 magic B32;
-} xXF86DRIAuthConnectionReq;
-
-#define sz_xXF86DRIAuthConnectionReq	12
-
-typedef struct
-{
-    BYTE type;
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 authenticated B32;
-    CARD32 pad2 B32;
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRIAuthConnectionReply;
-
-#define zx_xXF86DRIAuthConnectionReply  32
-
-typedef struct _XF86DRICloseConnection
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRICloseConnection */
-    CARD16 length B16;
-    CARD32 screen B32;
-} xXF86DRICloseConnectionReq;
-
-#define sz_xXF86DRICloseConnectionReq	8
-
-typedef struct _XF86DRIGetClientDriverName
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIGetClientDriverName */
-    CARD16 length B16;
-    CARD32 screen B32;
-} xXF86DRIGetClientDriverNameReq;
-
-#define sz_xXF86DRIGetClientDriverNameReq	8
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 ddxDriverMajorVersion B32;
-    CARD32 ddxDriverMinorVersion B32;
-    CARD32 ddxDriverPatchVersion B32;
-    CARD32 clientDriverNameLength B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRIGetClientDriverNameReply;
-
-#define sz_xXF86DRIGetClientDriverNameReply	32
-
-typedef struct _XF86DRICreateContext
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRICreateContext */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 visual B32;
-    CARD32 context B32;
-} xXF86DRICreateContextReq;
-
-#define sz_xXF86DRICreateContextReq	16
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 hHWContext B32;
-    CARD32 pad2 B32;
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRICreateContextReply;
-
-#define sz_xXF86DRICreateContextReply	32
-
-typedef struct _XF86DRIDestroyContext
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIDestroyContext */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 context B32;
-} xXF86DRIDestroyContextReq;
-
-#define sz_xXF86DRIDestroyContextReq	12
-
-typedef struct _XF86DRICreateDrawable
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRICreateDrawable */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 drawable B32;
-} xXF86DRICreateDrawableReq;
-
-#define sz_xXF86DRICreateDrawableReq	12
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 hHWDrawable B32;
-    CARD32 pad2 B32;
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRICreateDrawableReply;
-
-#define sz_xXF86DRICreateDrawableReply	32
-
-typedef struct _XF86DRIDestroyDrawable
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIDestroyDrawable */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 drawable B32;
-} xXF86DRIDestroyDrawableReq;
-
-#define sz_xXF86DRIDestroyDrawableReq	12
-
-typedef struct _XF86DRIGetDrawableInfo
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIGetDrawableInfo */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 drawable B32;
-} xXF86DRIGetDrawableInfoReq;
-
-#define sz_xXF86DRIGetDrawableInfoReq	12
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 drawableTableIndex B32;
-    CARD32 drawableTableStamp B32;
-    INT16 drawableX B16;
-    INT16 drawableY B16;
-    INT16 drawableWidth B16;
-    INT16 drawableHeight B16;
-    CARD32 numClipRects B32;
-    INT16 backX B16;
-    INT16 backY B16;
-    CARD32 numBackClipRects B32;
-} xXF86DRIGetDrawableInfoReply;
-
-#define sz_xXF86DRIGetDrawableInfoReply	36
-
-typedef struct _XF86DRIGetDeviceInfo
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIGetDeviceInfo */
-    CARD16 length B16;
-    CARD32 screen B32;
-} xXF86DRIGetDeviceInfoReq;
-
-#define sz_xXF86DRIGetDeviceInfoReq	8
-
-typedef struct
-{
-    BYTE type;			       /* X_Reply */
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 hFrameBufferLow B32;
-    CARD32 hFrameBufferHigh B32;
-    CARD32 framebufferOrigin B32;
-    CARD32 framebufferSize B32;
-    CARD32 framebufferStride B32;
-    CARD32 devPrivateSize B32;
-} xXF86DRIGetDeviceInfoReply;
-
-#define sz_xXF86DRIGetDeviceInfoReply	32
-
-typedef struct _XF86DRIOpenFullScreen
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRIOpenFullScreen */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 drawable B32;
-} xXF86DRIOpenFullScreenReq;
-
-#define sz_xXF86DRIOpenFullScreenReq    12
-
-typedef struct
-{
-    BYTE type;
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 isFullScreen B32;
-    CARD32 pad2 B32;
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-} xXF86DRIOpenFullScreenReply;
-
-#define sz_xXF86DRIOpenFullScreenReply  32
-
-typedef struct _XF86DRICloseFullScreen
-{
-    CARD8 reqType;		       /* always DRIReqCode */
-    CARD8 driReqType;		       /* always X_DRICloseFullScreen */
-    CARD16 length B16;
-    CARD32 screen B32;
-    CARD32 drawable B32;
-} xXF86DRICloseFullScreenReq;
-
-#define sz_xXF86DRICloseFullScreenReq   12
-
-typedef struct
-{
-    BYTE type;
-    BOOL pad1;
-    CARD16 sequenceNumber B16;
-    CARD32 length B32;
-    CARD32 pad2 B32;
-    CARD32 pad3 B32;
-    CARD32 pad4 B32;
-    CARD32 pad5 B32;
-    CARD32 pad6 B32;
-    CARD32 pad7 B32;
-} xXF86DRICloseFullScreenReply;
-
-#define sz_xXF86DRICloseFullScreenReply  32
-
-#endif /* _XF86DRISTR_H_ */
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
index a7c6e88b9eb..657544dcb21 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_screen_dri.c
@@ -44,6 +44,16 @@
 
 #include <stdio.h>
 
+static struct svga_winsys_surface *
+vmw_drm_surface_from_handle(struct svga_winsys_screen *sws,
+			    struct winsys_handle *whandle,
+			    SVGA3dSurfaceFormat *format);
+static boolean
+vmw_drm_surface_get_handle(struct svga_winsys_screen *sws,
+			   struct svga_winsys_surface *surface,
+			   unsigned stride,
+			   struct winsys_handle *whandle);
+
 static struct dri1_api dri1_api_hooks;
 static struct dri1_api_version ddx_required = { 0, 1, 0 };
 static struct dri1_api_version ddx_compat = { 0, 0, 0 };
@@ -129,7 +139,12 @@ vmw_drm_create_screen(struct drm_api *drm_api,
 				     &drm_compat, "use old scanout field (not a error)"))
 	    use_old_scanout_flag = TRUE;
 	 dri1->api = &dri1_api_hooks;
+#if 0
 	 break;
+#else
+	 assert(!"No dri 1 support for now\n");
+	 return NULL;
+#endif
       default:
 	 return NULL;
       }
@@ -139,6 +154,10 @@ vmw_drm_create_screen(struct drm_api *drm_api,
    if (!vws)
       goto out_no_vws;
 
+   /* XXX do this properly */
+   vws->base.surface_from_handle = vmw_drm_surface_from_handle;
+   vws->base.surface_get_handle = vmw_drm_surface_get_handle;
+
    screen = svga_screen_create( &vws->base );
    if (!screen)
       goto out_no_screen;
@@ -200,6 +219,7 @@ vmw_dri1_present_locked(struct pipe_context *locked_pipe,
 			const struct drm_clip_rect *bbox,
 			struct pipe_fence_handle **p_fence)
 {
+#if 0
    struct svga_winsys_surface *srf =
       svga_screen_texture_get_winsys_surface(surf->texture);
    struct vmw_svga_winsys_surface *vsrf = vmw_svga_winsys_surface(srf);
@@ -246,21 +266,19 @@ vmw_dri1_present_locked(struct pipe_context *locked_pipe,
 
    *p_fence = (visible) ? vmw_pipe_fence(fence_seq) : NULL;
    vmw_svga_winsys_surface_reference(&vsrf, NULL);
+#else
+   assert(!"No dri 1 support for now\n");
+#endif
 }
 
-static struct pipe_texture *
-vmw_drm_texture_from_handle(struct drm_api *drm_api,
-			    struct pipe_screen *screen,
-			    struct pipe_texture *templat,
-			    const char *name,
-			    unsigned stride,
-			    unsigned handle)
+static struct svga_winsys_surface *
+vmw_drm_surface_from_handle(struct svga_winsys_screen *sws,
+			    struct winsys_handle *whandle,
+			    SVGA3dSurfaceFormat *format)
 {
     struct vmw_svga_winsys_surface *vsrf;
     struct svga_winsys_surface *ssrf;
-    struct vmw_winsys_screen *vws =
-	vmw_winsys_screen(svga_winsys_screen(screen));
-    struct pipe_texture *tex;
+    struct vmw_winsys_screen *vws = vmw_winsys_screen(sws);
     union drm_vmw_surface_reference_arg arg;
     struct drm_vmw_surface_arg *req = &arg.req;
     struct drm_vmw_surface_create_req *rep = &arg.rep;
@@ -273,7 +291,7 @@ vmw_drm_texture_from_handle(struct drm_api *drm_api,
      */
 
     memset(&arg, 0, sizeof(arg));
-    req->sid = handle;
+    req->sid = whandle->handle;
 
     ret = drmCommandWriteRead(vws->ioctl.drm_fd, DRM_VMW_REF_SURFACE,
 			      &arg, sizeof(arg));
@@ -281,14 +299,14 @@ vmw_drm_texture_from_handle(struct drm_api *drm_api,
     if (ret) {
 	fprintf(stderr, "Failed referencing shared surface. SID %d.\n"
 		"Error %d (%s).\n",
-		handle, ret, strerror(-ret));
+		whandle->handle, ret, strerror(-ret));
 	return NULL;
     }
 
     if (rep->mip_levels[0] != 1) {
 	fprintf(stderr, "Incorrect number of mipmap levels on shared surface."
 		" SID %d, levels %d\n",
-		handle, rep->mip_levels[0]);
+		whandle->handle, rep->mip_levels[0]);
 	goto out_mip;
     }
 
@@ -296,7 +314,7 @@ vmw_drm_texture_from_handle(struct drm_api *drm_api,
 	if (rep->mip_levels[i] != 0) {
 	    fprintf(stderr, "Incorrect number of faces levels on shared surface."
 		    " SID %d, face %d present.\n",
-		    handle, i);
+		    whandle->handle, i);
 	    goto out_mip;
 	}
     }
@@ -308,38 +326,32 @@ vmw_drm_texture_from_handle(struct drm_api *drm_api,
     pipe_reference_init(&vsrf->refcnt, 1);
     p_atomic_set(&vsrf->validated, 0);
     vsrf->screen = vws;
-    vsrf->sid = handle;
+    vsrf->sid = whandle->handle;
     ssrf = svga_winsys_surface(vsrf);
-    tex = svga_screen_texture_wrap_surface(screen, templat, rep->format, ssrf);
-    if (!tex)
-	vmw_svga_winsys_surface_reference(&vsrf, NULL);
+    *format = rep->format;
+
+    return ssrf;
 
-    return tex;
-  out_mip:
-    vmw_ioctl_surface_destroy(vws, handle);
+out_mip:
+    vmw_ioctl_surface_destroy(vws, whandle->handle);
     return NULL;
 }
 
 static boolean
-vmw_drm_handle_from_texture(struct drm_api *drm_api,
-                           struct pipe_screen *screen,
-			   struct pipe_texture *texture,
-			   unsigned *stride,
-			   unsigned *handle)
+vmw_drm_surface_get_handle(struct svga_winsys_screen *sws,
+			   struct svga_winsys_surface *surface,
+			   unsigned stride,
+			   struct winsys_handle *whandle)
 {
-    struct svga_winsys_surface *surface =
-	svga_screen_texture_get_winsys_surface(texture);
     struct vmw_svga_winsys_surface *vsrf;
 
     if (!surface)
 	return FALSE;
 
     vsrf = vmw_svga_winsys_surface(surface);
-    *handle = vsrf->sid;
-    *stride = util_format_get_nblocksx(texture->format, texture->width0) *
-       util_format_get_blocksize(texture->format);
+    whandle->handle = vsrf->sid;
+    whandle->stride = stride;
 
-    vmw_svga_winsys_surface_reference(&vsrf, NULL);
     return TRUE;
 }
 
@@ -353,9 +365,6 @@ static struct drm_api vmw_drm_api_hooks = {
    .name = "vmwgfx",
    .driver_name = "vmwgfx",
    .create_screen = vmw_drm_create_screen,
-   .texture_from_shared_handle = vmw_drm_texture_from_handle,
-   .shared_handle_from_texture = vmw_drm_handle_from_texture,
-   .local_handle_from_texture = vmw_drm_handle_from_texture,
 };
 
 struct drm_api* drm_api_create()
diff --git a/src/gallium/winsys/gdi/SConscript b/src/gallium/winsys/gdi/SConscript
index 4cbc86f3311..1267fc6eea4 100644
--- a/src/gallium/winsys/gdi/SConscript
+++ b/src/gallium/winsys/gdi/SConscript
@@ -1,5 +1,6 @@
 #######################################################################
-# SConscript for gdi winsys
+# SConscript for xlib winsys
+
 
 Import('*')
 
@@ -8,44 +9,15 @@ if env['platform'] == 'windows':
     env = env.Clone()
 
     env.Append(CPPPATH = [
-        '#src/gallium/state_trackers/wgl',
-    ])
-
-    env.Append(LIBS = [
-        'gdi32',
-        'user32',
-        'kernel32',
-        'ws2_32',
+        '#/src/gallium/include',
+        '#/src/gallium/auxiliary',
+        '#/src/gallium/drivers',
     ])
 
-    sources = []
-    drivers = []
-
-    if 'softpipe' in env['drivers']:
-        sources = ['gdi_softpipe_winsys.c']
-        drivers = [softpipe]
-
-    if 'llvmpipe' in env['drivers']:
-        env.Tool('llvm')
-        if 'LLVM_VERSION' in env:
-            sources = ['gdi_llvmpipe_winsys.c']
-            drivers = [llvmpipe]
-
-    if not sources or not drivers:
-        print 'warning: softpipe or llvmpipe not selected, gdi winsys disabled'
-        Return()
-    
-    if env['gcc']:
-        sources += ['#src/gallium/state_trackers/wgl/opengl32.mingw.def']
-    else:
-        sources += ['#src/gallium/state_trackers/wgl/opengl32.def']
-        
-    drivers += [trace]
-
-    env['no_import_lib'] = 1
-
-    env.SharedLibrary(
-        target ='opengl32',
-        source = sources,
-        LIBS = wgl + glapi + mesa + drivers + gallium + glsl + env['LIBS'],
+    ws_gdi = env.ConvenienceLibrary(
+        target = 'ws_gdi',
+        source = [
+           'gdi_sw_winsys.c',
+        ]
     )
+    Export('ws_gdi')
diff --git a/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c b/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c
deleted file mode 100644
index a9fa03b8e57..00000000000
--- a/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c
+++ /dev/null
@@ -1,272 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2009 VMware, Inc.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * THE 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- *
- **************************************************************************/
-
-/**
- * @file
- * LLVMpipe support.
- *
- * @author Jose Fonseca <jfonseca@vmware.com>
- */
-
-
-#include <windows.h>
-
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "llvmpipe/lp_winsys.h"
-#include "llvmpipe/lp_texture.h"
-#include "stw_winsys.h"
-
-
-struct gdi_llvmpipe_displaytarget
-{
-   enum pipe_format format;
-   unsigned width;
-   unsigned height;
-   unsigned stride;
-
-   unsigned size;
-
-   void *data;
-
-   BITMAPINFO bmi;
-};
-
-
-/** Cast wrapper */
-static INLINE struct gdi_llvmpipe_displaytarget *
-gdi_llvmpipe_displaytarget( struct llvmpipe_displaytarget *buf )
-{
-   return (struct gdi_llvmpipe_displaytarget *)buf;
-}
-
-
-static boolean
-gdi_llvmpipe_is_displaytarget_format_supported( struct llvmpipe_winsys *ws,
-                                                enum pipe_format format )
-{
-   switch(format) {
-   case PIPE_FORMAT_B8G8R8X8_UNORM:
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
-      return TRUE;
-
-   /* TODO: Support other formats possible with BMPs, as described in 
-    * http://msdn.microsoft.com/en-us/library/dd183376(VS.85).aspx */
-      
-   default:
-      return FALSE;
-   }
-}
-
-
-static void *
-gdi_llvmpipe_displaytarget_map(struct llvmpipe_winsys *ws,
-                               struct llvmpipe_displaytarget *dt,
-                               unsigned flags )
-{
-   struct gdi_llvmpipe_displaytarget *gdt = gdi_llvmpipe_displaytarget(dt);
-
-   return gdt->data;
-}
-
-
-static void
-gdi_llvmpipe_displaytarget_unmap(struct llvmpipe_winsys *ws,
-                                 struct llvmpipe_displaytarget *dt )
-{
-
-}
-
-
-static void
-gdi_llvmpipe_displaytarget_destroy(struct llvmpipe_winsys *winsys,
-                                   struct llvmpipe_displaytarget *dt)
-{
-   struct gdi_llvmpipe_displaytarget *gdt = gdi_llvmpipe_displaytarget(dt);
-
-   align_free(gdt->data);
-   FREE(gdt);
-}
-
-
-static struct llvmpipe_displaytarget *
-gdi_llvmpipe_displaytarget_create(struct llvmpipe_winsys *winsys,
-                                  enum pipe_format format,
-                                  unsigned width, unsigned height,
-                                  unsigned alignment,
-                                  unsigned *stride)
-{
-   struct gdi_llvmpipe_displaytarget *gdt;
-   unsigned cpp;
-   unsigned bpp;
-   
-   gdt = CALLOC_STRUCT(gdi_llvmpipe_displaytarget);
-   if(!gdt)
-      goto no_gdt;
-
-   gdt->format = format;
-   gdt->width = width;
-   gdt->height = height;
-
-   bpp = util_format_get_blocksizebits(format);
-   cpp = util_format_get_blocksize(format);
-   
-   gdt->stride = align(width * cpp, alignment);
-   gdt->size = gdt->stride * height;
-   
-   gdt->data = align_malloc(gdt->size, alignment);
-   if(!gdt->data)
-      goto no_data;
-
-   gdt->bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
-   gdt->bmi.bmiHeader.biWidth = gdt->stride / cpp;
-   gdt->bmi.bmiHeader.biHeight= -(long)height;
-   gdt->bmi.bmiHeader.biPlanes = 1;
-   gdt->bmi.bmiHeader.biBitCount = bpp;
-   gdt->bmi.bmiHeader.biCompression = BI_RGB;
-   gdt->bmi.bmiHeader.biSizeImage = 0;
-   gdt->bmi.bmiHeader.biXPelsPerMeter = 0;
-   gdt->bmi.bmiHeader.biYPelsPerMeter = 0;
-   gdt->bmi.bmiHeader.biClrUsed = 0;
-   gdt->bmi.bmiHeader.biClrImportant = 0;
-
-   *stride = gdt->stride;
-   return (struct llvmpipe_displaytarget *)gdt;
-
-no_data:
-   FREE(gdt);
-no_gdt:
-   return NULL;
-}
-
-
-static void
-gdi_llvmpipe_displaytarget_display(struct llvmpipe_winsys *winsys, 
-                                   struct llvmpipe_displaytarget *dt,
-                                   void *context_private)
-{
-   assert(0);
-}
-
-
-static void
-gdi_llvmpipe_destroy(struct llvmpipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-static struct pipe_screen *
-gdi_llvmpipe_screen_create(void)
-{
-   static struct llvmpipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(llvmpipe_winsys);
-   if(!winsys)
-      goto no_winsys;
-
-   winsys->destroy = gdi_llvmpipe_destroy;
-   winsys->is_displaytarget_format_supported = gdi_llvmpipe_is_displaytarget_format_supported;
-   winsys->displaytarget_create = gdi_llvmpipe_displaytarget_create;
-   winsys->displaytarget_map = gdi_llvmpipe_displaytarget_map;
-   winsys->displaytarget_unmap = gdi_llvmpipe_displaytarget_unmap;
-   winsys->displaytarget_display = gdi_llvmpipe_displaytarget_display;
-   winsys->displaytarget_destroy = gdi_llvmpipe_displaytarget_destroy;
-
-   screen = llvmpipe_create_screen(winsys);
-   if(!screen)
-      goto no_screen;
-
-   return screen;
-   
-no_screen:
-   FREE(winsys);
-no_winsys:
-   return NULL;
-}
-
-
-
-
-static void
-gdi_llvmpipe_present(struct pipe_screen *screen,
-                     struct pipe_surface *surface,
-                     HDC hDC)
-{
-    struct llvmpipe_texture *texture;
-    struct gdi_llvmpipe_displaytarget *gdt;
-
-    texture = llvmpipe_texture(surface->texture);
-    gdt = gdi_llvmpipe_displaytarget(texture->dt);
-
-    StretchDIBits(hDC,
-                  0, 0, gdt->width, gdt->height,
-                  0, 0, gdt->width, gdt->height,
-                  gdt->data, &gdt->bmi, 0, SRCCOPY);
-}
-
-
-static const struct stw_winsys stw_winsys = {
-   &gdi_llvmpipe_screen_create,
-   &gdi_llvmpipe_present,
-   NULL, /* get_adapter_luid */
-   NULL, /* shared_surface_open */
-   NULL, /* shared_surface_close */
-   NULL  /* compose */
-};
-
-
-BOOL WINAPI
-DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
-{
-   switch (fdwReason) {
-   case DLL_PROCESS_ATTACH:
-      stw_init(&stw_winsys);
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_ATTACH:
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_DETACH:
-      stw_cleanup_thread();
-      break;
-
-   case DLL_PROCESS_DETACH:
-      stw_cleanup_thread();
-      stw_cleanup();
-      break;
-   }
-   return TRUE;
-}
diff --git a/src/gallium/winsys/gdi/gdi_softpipe_winsys.c b/src/gallium/winsys/gdi/gdi_softpipe_winsys.c
deleted file mode 100644
index 71360e55aac..00000000000
--- a/src/gallium/winsys/gdi/gdi_softpipe_winsys.c
+++ /dev/null
@@ -1,318 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2008 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- *
- **************************************************************************/
-
-/**
- * @file
- * Softpipe support.
- *
- * @author Keith Whitwell
- * @author Brian Paul
- * @author Jose Fonseca
- */
-
-
-#include <windows.h>
-
-#include "util/u_simple_screen.h"
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "softpipe/sp_winsys.h"
-#include "softpipe/sp_texture.h"
-#include "stw_winsys.h"
-
-
-struct gdi_softpipe_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-};
-
-
-/** Cast wrapper */
-static INLINE struct gdi_softpipe_buffer *
-gdi_softpipe_buffer( struct pipe_buffer *buf )
-{
-   return (struct gdi_softpipe_buffer *)buf;
-}
-
-
-static void *
-gdi_softpipe_buffer_map(struct pipe_winsys *winsys,
-                        struct pipe_buffer *buf,
-                        unsigned flags)
-{
-   struct gdi_softpipe_buffer *gdi_softpipe_buf = gdi_softpipe_buffer(buf);
-   gdi_softpipe_buf->mapped = gdi_softpipe_buf->data;
-   return gdi_softpipe_buf->mapped;
-}
-
-
-static void
-gdi_softpipe_buffer_unmap(struct pipe_winsys *winsys,
-                          struct pipe_buffer *buf)
-{
-   struct gdi_softpipe_buffer *gdi_softpipe_buf = gdi_softpipe_buffer(buf);
-   gdi_softpipe_buf->mapped = NULL;
-}
-
-
-static void
-gdi_softpipe_buffer_destroy(struct pipe_buffer *buf)
-{
-   struct gdi_softpipe_buffer *oldBuf = gdi_softpipe_buffer(buf);
-
-   if (oldBuf->data) {
-      if (!oldBuf->userBuffer)
-         align_free(oldBuf->data);
-
-      oldBuf->data = NULL;
-   }
-
-   FREE(oldBuf);
-}
-
-
-static const char *
-gdi_softpipe_get_name(struct pipe_winsys *winsys)
-{
-   return "softpipe";
-}
-
-
-static struct pipe_buffer *
-gdi_softpipe_buffer_create(struct pipe_winsys *winsys,
-                           unsigned alignment,
-                           unsigned usage,
-                           unsigned size)
-{
-   struct gdi_softpipe_buffer *buffer = CALLOC_STRUCT(gdi_softpipe_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-   buffer->data = align_malloc(size, alignment);
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-gdi_softpipe_user_buffer_create(struct pipe_winsys *winsys,
-                               void *ptr,
-                               unsigned bytes)
-{
-   struct gdi_softpipe_buffer *buffer;
-
-   buffer = CALLOC_STRUCT(gdi_softpipe_buffer);
-   if(!buffer)
-      return NULL;
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-
-   return &buffer->base;
-}
-
-
-static struct pipe_buffer *
-gdi_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
-                                   unsigned width, unsigned height,
-                                   enum pipe_format format,
-                                   unsigned usage,
-                                   unsigned tex_usage,
-                                   unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-gdi_softpipe_dummy_flush_frontbuffer(struct pipe_winsys *winsys,
-                                     struct pipe_surface *surface,
-                                     void *context_private)
-{
-   assert(0);
-}
-
-
-static void
-gdi_softpipe_fence_reference(struct pipe_winsys *winsys,
-                             struct pipe_fence_handle **ptr,
-                             struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-gdi_softpipe_fence_signalled(struct pipe_winsys *winsys,
-                             struct pipe_fence_handle *fence,
-                             unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-gdi_softpipe_fence_finish(struct pipe_winsys *winsys,
-                          struct pipe_fence_handle *fence,
-                          unsigned flag)
-{
-   return 0;
-}
-
-
-static void
-gdi_softpipe_destroy(struct pipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-static struct pipe_screen *
-gdi_softpipe_screen_create(void)
-{
-   static struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(pipe_winsys);
-   if(!winsys)
-      return NULL;
-
-   winsys->destroy = gdi_softpipe_destroy;
-
-   winsys->buffer_create = gdi_softpipe_buffer_create;
-   winsys->user_buffer_create = gdi_softpipe_user_buffer_create;
-   winsys->buffer_map = gdi_softpipe_buffer_map;
-   winsys->buffer_unmap = gdi_softpipe_buffer_unmap;
-   winsys->buffer_destroy = gdi_softpipe_buffer_destroy;
-
-   winsys->surface_buffer_create = gdi_softpipe_surface_buffer_create;
-
-   winsys->fence_reference = gdi_softpipe_fence_reference;
-   winsys->fence_signalled = gdi_softpipe_fence_signalled;
-   winsys->fence_finish = gdi_softpipe_fence_finish;
-
-   winsys->flush_frontbuffer = gdi_softpipe_dummy_flush_frontbuffer;
-   winsys->get_name = gdi_softpipe_get_name;
-
-   screen = softpipe_create_screen(winsys);
-   if(!screen)
-      gdi_softpipe_destroy(winsys);
-
-   return screen;
-}
-
-
-static void
-gdi_softpipe_present(struct pipe_screen *screen,
-                     struct pipe_surface *surface,
-                     HDC hDC)
-{
-    struct softpipe_texture *texture;
-    struct gdi_softpipe_buffer *buffer;
-    BITMAPINFO bmi;
-
-    texture = softpipe_texture(surface->texture);
-                                               
-    buffer = gdi_softpipe_buffer(texture->buffer);
-
-    memset(&bmi, 0, sizeof(BITMAPINFO));
-    bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
-    bmi.bmiHeader.biWidth = texture->stride[surface->level] / util_format_get_blocksize(surface->format);
-    bmi.bmiHeader.biHeight= -(long)surface->height;
-    bmi.bmiHeader.biPlanes = 1;
-    bmi.bmiHeader.biBitCount = util_format_get_blocksizebits(surface->format);
-    bmi.bmiHeader.biCompression = BI_RGB;
-    bmi.bmiHeader.biSizeImage = 0;
-    bmi.bmiHeader.biXPelsPerMeter = 0;
-    bmi.bmiHeader.biYPelsPerMeter = 0;
-    bmi.bmiHeader.biClrUsed = 0;
-    bmi.bmiHeader.biClrImportant = 0;
-
-    StretchDIBits(hDC,
-                  0, 0, surface->width, surface->height,
-                  0, 0, surface->width, surface->height,
-                  buffer->data, &bmi, 0, SRCCOPY);
-}
-
-
-static const struct stw_winsys stw_winsys = {
-   &gdi_softpipe_screen_create,
-   &gdi_softpipe_present,
-   NULL, /* get_adapter_luid */
-   NULL, /* shared_surface_open */
-   NULL, /* shared_surface_close */
-   NULL  /* compose */
-};
-
-
-BOOL WINAPI
-DllMain(HINSTANCE hinstDLL, DWORD fdwReason, LPVOID lpReserved)
-{
-   switch (fdwReason) {
-   case DLL_PROCESS_ATTACH:
-      stw_init(&stw_winsys);
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_ATTACH:
-      stw_init_thread();
-      break;
-
-   case DLL_THREAD_DETACH:
-      stw_cleanup_thread();
-      break;
-
-   case DLL_PROCESS_DETACH:
-      stw_cleanup_thread();
-      stw_cleanup();
-      break;
-   }
-   return TRUE;
-}
diff --git a/src/gallium/winsys/gdi/gdi_sw_winsys.c b/src/gallium/winsys/gdi/gdi_sw_winsys.c
new file mode 100644
index 00000000000..f5c0b7d56ec
--- /dev/null
+++ b/src/gallium/winsys/gdi/gdi_sw_winsys.c
@@ -0,0 +1,222 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * GDI software rasterizer support.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ */
+
+
+#include <windows.h>
+
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "state_tracker/sw_winsys.h"
+#include "gdi_sw_winsys.h"
+
+
+struct gdi_sw_displaytarget
+{
+   enum pipe_format format;
+   unsigned width;
+   unsigned height;
+   unsigned stride;
+
+   unsigned size;
+
+   void *data;
+
+   BITMAPINFO bmi;
+};
+
+
+/** Cast wrapper */
+static INLINE struct gdi_sw_displaytarget *
+gdi_sw_displaytarget( struct sw_displaytarget *buf )
+{
+   return (struct gdi_sw_displaytarget *)buf;
+}
+
+
+static boolean
+gdi_sw_is_displaytarget_format_supported( struct sw_winsys *ws,
+                                                enum pipe_format format )
+{
+   switch(format) {
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      return TRUE;
+
+   /* TODO: Support other formats possible with BMPs, as described in 
+    * http://msdn.microsoft.com/en-us/library/dd183376(VS.85).aspx */
+      
+   default:
+      return FALSE;
+   }
+}
+
+
+static void *
+gdi_sw_displaytarget_map(struct sw_winsys *ws,
+                               struct sw_displaytarget *dt,
+                               unsigned flags )
+{
+   struct gdi_sw_displaytarget *gdt = gdi_sw_displaytarget(dt);
+
+   return gdt->data;
+}
+
+
+static void
+gdi_sw_displaytarget_unmap(struct sw_winsys *ws,
+                                 struct sw_displaytarget *dt )
+{
+
+}
+
+
+static void
+gdi_sw_displaytarget_destroy(struct sw_winsys *winsys,
+                                   struct sw_displaytarget *dt)
+{
+   struct gdi_sw_displaytarget *gdt = gdi_sw_displaytarget(dt);
+
+   align_free(gdt->data);
+   FREE(gdt);
+}
+
+
+static struct sw_displaytarget *
+gdi_sw_displaytarget_create(struct sw_winsys *winsys,
+                                  enum pipe_format format,
+                                  unsigned width, unsigned height,
+                                  unsigned alignment,
+                                  unsigned *stride)
+{
+   struct gdi_sw_displaytarget *gdt;
+   unsigned cpp;
+   unsigned bpp;
+   
+   gdt = CALLOC_STRUCT(gdi_sw_displaytarget);
+   if(!gdt)
+      goto no_gdt;
+
+   gdt->format = format;
+   gdt->width = width;
+   gdt->height = height;
+
+   bpp = util_format_get_blocksizebits(format);
+   cpp = util_format_get_blocksize(format);
+   
+   gdt->stride = align(width * cpp, alignment);
+   gdt->size = gdt->stride * height;
+   
+   gdt->data = align_malloc(gdt->size, alignment);
+   if(!gdt->data)
+      goto no_data;
+
+   gdt->bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
+   gdt->bmi.bmiHeader.biWidth = gdt->stride / cpp;
+   gdt->bmi.bmiHeader.biHeight= -(long)height;
+   gdt->bmi.bmiHeader.biPlanes = 1;
+   gdt->bmi.bmiHeader.biBitCount = bpp;
+   gdt->bmi.bmiHeader.biCompression = BI_RGB;
+   gdt->bmi.bmiHeader.biSizeImage = 0;
+   gdt->bmi.bmiHeader.biXPelsPerMeter = 0;
+   gdt->bmi.bmiHeader.biYPelsPerMeter = 0;
+   gdt->bmi.bmiHeader.biClrUsed = 0;
+   gdt->bmi.bmiHeader.biClrImportant = 0;
+
+   *stride = gdt->stride;
+   return (struct sw_displaytarget *)gdt;
+
+no_data:
+   FREE(gdt);
+no_gdt:
+   return NULL;
+}
+
+
+void
+gdi_sw_display( struct sw_winsys *winsys,
+                struct sw_displaytarget *dt,
+                HDC hDC )
+{
+    struct gdi_sw_displaytarget *gdt = gdi_sw_displaytarget(dt);
+
+    StretchDIBits(hDC,
+                  0, 0, gdt->width, gdt->height,
+                  0, 0, gdt->width, gdt->height,
+                  gdt->data, &gdt->bmi, 0, SRCCOPY);
+}
+
+static void
+gdi_sw_displaytarget_display(struct sw_winsys *winsys, 
+                             struct sw_displaytarget *dt,
+                             void *context_private)
+{
+    /* nasty:
+     */
+    HDC hDC = (HDC)context_private;
+
+    gdi_sw_display(winsys, dt, hDC);
+}
+
+
+static void
+gdi_sw_destroy(struct sw_winsys *winsys)
+{
+   FREE(winsys);
+}
+
+struct sw_winsys *
+gdi_create_sw_winsys(void)
+{
+   static struct sw_winsys *winsys;
+
+   winsys = CALLOC_STRUCT(sw_winsys);
+   if(!winsys)
+      return NULL;
+
+   winsys->destroy = gdi_sw_destroy;
+   winsys->is_displaytarget_format_supported = gdi_sw_is_displaytarget_format_supported;
+   winsys->displaytarget_create = gdi_sw_displaytarget_create;
+   winsys->displaytarget_map = gdi_sw_displaytarget_map;
+   winsys->displaytarget_unmap = gdi_sw_displaytarget_unmap;
+   winsys->displaytarget_display = gdi_sw_displaytarget_display;
+   winsys->displaytarget_destroy = gdi_sw_displaytarget_destroy;
+
+   return winsys;
+}
+
diff --git a/src/gallium/winsys/gdi/gdi_sw_winsys.h b/src/gallium/winsys/gdi/gdi_sw_winsys.h
new file mode 100644
index 00000000000..4bbcb47848b
--- /dev/null
+++ b/src/gallium/winsys/gdi/gdi_sw_winsys.h
@@ -0,0 +1,16 @@
+#ifndef GDI_SW_WINSYS_H
+#define GDI_SW_WINSYS_H
+
+#include <windows.h>
+
+#include "pipe/p_compiler.h"
+#include "state_tracker/sw_winsys.h"
+
+void gdi_sw_display( struct sw_winsys *winsys,
+                     struct sw_displaytarget *dt,
+                     HDC hDC );
+
+struct sw_winsys *
+gdi_create_sw_winsys(void);
+
+#endif
diff --git a/src/gallium/winsys/null/Makefile b/src/gallium/winsys/null/Makefile
new file mode 100644
index 00000000000..3a3fb75ab36
--- /dev/null
+++ b/src/gallium/winsys/null/Makefile
@@ -0,0 +1,16 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = ws_null
+
+LIBRARY_INCLUDES = \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/src/gallium/auxiliary
+
+C_SOURCES = \
+	null_sw_winsys.c 
+
+include ../../Makefile.template
+
+
diff --git a/src/gallium/winsys/null/SConscript b/src/gallium/winsys/null/SConscript
new file mode 100644
index 00000000000..21837dc60c2
--- /dev/null
+++ b/src/gallium/winsys/null/SConscript
@@ -0,0 +1,21 @@
+#######################################################################
+# SConscript for xlib winsys
+
+
+Import('*')
+
+env = env.Clone()
+
+env.Append(CPPPATH = [
+    '#/src/gallium/include',
+    '#/src/gallium/auxiliary',
+    '#/src/gallium/drivers',
+])
+
+ws_null = env.ConvenienceLibrary(
+    target = 'ws_null',
+    source = [
+       'null_sw_winsys.c',
+    ]
+)
+Export('ws_null')
diff --git a/src/gallium/winsys/null/null_sw_winsys.c b/src/gallium/winsys/null/null_sw_winsys.c
new file mode 100644
index 00000000000..d961d34860e
--- /dev/null
+++ b/src/gallium/winsys/null/null_sw_winsys.c
@@ -0,0 +1,124 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Null software rasterizer winsys.
+ * 
+ * There is no present support. Framebuffer data needs to be obtained via
+ * transfers.
+ *
+ * @author Jose Fonseca
+ */
+
+
+#include "pipe/p_format.h"
+#include "util/u_memory.h"
+#include "state_tracker/sw_winsys.h"
+#include "null_sw_winsys.h"
+
+
+static boolean
+null_sw_is_displaytarget_format_supported(struct sw_winsys *ws,
+                                          enum pipe_format format )
+{
+   return FALSE;
+}
+
+
+static void *
+null_sw_displaytarget_map(struct sw_winsys *ws,
+                          struct sw_displaytarget *dt,
+                          unsigned flags )
+{
+   assert(0);
+   return NULL;
+}
+
+
+static void
+null_sw_displaytarget_unmap(struct sw_winsys *ws,
+                            struct sw_displaytarget *dt )
+{
+   assert(0);
+}
+
+
+static void
+null_sw_displaytarget_destroy(struct sw_winsys *winsys,
+                              struct sw_displaytarget *dt)
+{
+   assert(0);
+}
+
+
+static struct sw_displaytarget *
+null_sw_displaytarget_create(struct sw_winsys *winsys,
+                             enum pipe_format format,
+                             unsigned width, unsigned height,
+                             unsigned alignment,
+                             unsigned *stride)
+{
+   return NULL;
+}
+
+
+static void
+null_sw_displaytarget_display(struct sw_winsys *winsys,
+                              struct sw_displaytarget *dt,
+                              void *context_private)
+{
+   assert(0);
+}
+
+
+static void
+null_sw_destroy(struct sw_winsys *winsys)
+{
+   FREE(winsys);
+}
+
+
+struct sw_winsys *
+null_sw_create(void)
+{
+   static struct sw_winsys *winsys;
+
+   winsys = CALLOC_STRUCT(sw_winsys);
+   if (!winsys)
+      return NULL;
+
+   winsys->destroy = null_sw_destroy;
+   winsys->is_displaytarget_format_supported = null_sw_is_displaytarget_format_supported;
+   winsys->displaytarget_create = null_sw_displaytarget_create;
+   winsys->displaytarget_map = null_sw_displaytarget_map;
+   winsys->displaytarget_unmap = null_sw_displaytarget_unmap;
+   winsys->displaytarget_display = null_sw_displaytarget_display;
+   winsys->displaytarget_destroy = null_sw_displaytarget_destroy;
+
+   return winsys;
+}
diff --git a/src/gallium/winsys/null/null_sw_winsys.h b/src/gallium/winsys/null/null_sw_winsys.h
new file mode 100644
index 00000000000..1986186febe
--- /dev/null
+++ b/src/gallium/winsys/null/null_sw_winsys.h
@@ -0,0 +1,40 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ **************************************************************************/
+
+
+#ifndef NULL_SW_WINSYS_H_
+#define NULL_SW_WINSYS_H_
+
+
+struct sw_winsys;
+
+
+struct sw_winsys *
+null_sw_create(void);
+
+
+#endif /* NULL_SW_WINSYS_H_ */
diff --git a/src/gallium/winsys/xlib/Makefile b/src/gallium/winsys/xlib/Makefile
index 824c666ae30..68542b488df 100644
--- a/src/gallium/winsys/xlib/Makefile
+++ b/src/gallium/winsys/xlib/Makefile
@@ -1,100 +1,16 @@
-# src/gallium/winsys/xlib/Makefile
-
-# This makefile produces a "stand-alone" libGL.so which is based on
-# Xlib (no DRI HW acceleration)
-
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
+LIBNAME = ws_xlib
 
-GL_MAJOR = 1
-GL_MINOR = 5
-GL_TINY = 0$(MESA_MAJOR)0$(MESA_MINOR)0$(MESA_TINY)
-
-
-INCLUDE_DIRS = \
-	-I$(TOP)/include \
-	-I$(TOP)/src/mesa \
-	-I$(TOP)/src/mesa/main \
+LIBRARY_INCLUDES = \
 	-I$(TOP)/src/gallium/include \
 	-I$(TOP)/src/gallium/drivers \
-	-I$(TOP)/src/gallium/state_trackers/glx/xlib \
 	-I$(TOP)/src/gallium/auxiliary
 
-DEFINES += \
-	-DGALLIUM_SOFTPIPE
-#-DGALLIUM_CELL will be defined by the config */
-
-XLIB_WINSYS_SOURCES = \
-	xlib.c \
-	xlib_cell.c \
-	xlib_llvmpipe.c \
-	xlib_softpipe.c
-
-
-XLIB_WINSYS_OBJECTS = $(XLIB_WINSYS_SOURCES:.c=.o)
-
-
-# Note: CELL_SPU_LIB is only defined for cell configs
-
-LIBS = \
-	$(GALLIUM_DRIVERS) \
-	$(TOP)/src/gallium/state_trackers/glx/xlib/libxlib.a \
-	$(TOP)/src/mesa/libglapi.a \
-	$(TOP)/src/mesa/libmesagallium.a \
-	$(GALLIUM_AUXILIARIES) \
-	$(CELL_SPU_LIB) \
-
-
-.SUFFIXES : .cpp
-
-.c.o:
-	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
-
-.cpp.o:
-	$(CXX) -c $(INCLUDE_DIRS) $(CXXFLAGS) $< -o $@
-
-
-
-default: $(TOP)/$(LIB_DIR)/gallium $(TOP)/$(LIB_DIR)/gallium/$(GL_LIB_NAME)
-
-$(TOP)/$(LIB_DIR)/gallium:
-	@ mkdir -p $(TOP)/$(LIB_DIR)/gallium
-
-# Make the libGL.so library
-$(TOP)/$(LIB_DIR)/gallium/$(GL_LIB_NAME): $(XLIB_WINSYS_OBJECTS) $(LIBS) Makefile
-	$(TOP)/bin/mklib -o $(GL_LIB) \
-		-linker "$(CC)" \
-		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
-		-install $(TOP)/$(LIB_DIR)/gallium \
-		$(MKLIB_OPTIONS) $(XLIB_WINSYS_OBJECTS) \
-		-Wl,--start-group $(LIBS) -Wl,--end-group $(GL_LIB_DEPS)
-
-
-depend: $(XLIB_WINSYS_SOURCES)
-	@ echo "running $(MKDEP)"
-	@ rm -f depend  # workaround oops on gutsy?!?
-	@ touch depend
-	$(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(XLIB_WINSYS_SOURCES) \
-		> /dev/null 2>/dev/null
-
-
-install: default
-	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/include/GL
-	$(INSTALL) -d $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR)
-	$(INSTALL) -m 644 $(TOP)/include/GL/*.h $(DESTDIR)$(INSTALL_DIR)/include/GL
-	@if [ -e $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME) ]; then \
-		$(MINSTALL) $(TOP)/$(LIB_DIR)/libGL* $(DESTDIR)$(INSTALL_DIR)/$(LIB_DIR); \
-	fi
-
-
-# Emacs tags
-tags:
-	etags `find . -name \*.[ch]` $(TOP)/include/GL/*.h
+C_SOURCES = \
+	xlib_sw_winsys.c 
 
-clean:
-	-rm -f *.o
+include ../../Makefile.template
 
 
-include depend
diff --git a/src/gallium/winsys/xlib/SConscript b/src/gallium/winsys/xlib/SConscript
index 8c9d318af2b..2af6153b4c7 100644
--- a/src/gallium/winsys/xlib/SConscript
+++ b/src/gallium/winsys/xlib/SConscript
@@ -1,64 +1,23 @@
 #######################################################################
 # SConscript for xlib winsys
 
-Import('*')
-
-if env['platform'] != 'linux':
-    Return()
-
-if 'mesa' not in env['statetrackers']:
-    print 'warning: Mesa state tracker disabled: skipping build of xlib libGL.so'
-    Return()
-
-if env['dri']:
-    print 'warning: DRI enabled: skipping build of xlib libGL.so'
-    Return()
-
-if not set(('softpipe', 'llvmpipe', 'cell')).intersection(env['drivers']):
-    print 'warning: no supported pipe driver: skipping build of xlib libGL.so'
-    Return()
 
-env = env.Clone()
-
-env.Append(CPPPATH = [
-    '#/src/mesa',
-    '#/src/mesa/main',
-    '#src/gallium/state_trackers/glx/xlib',
-])
-
-env.Append(CPPDEFINES = ['USE_XSHM'])
-
-sources = [
-    'xlib.c',
-]
+Import('*')
 
-drivers = [trace]
-    
-if 'softpipe' in env['drivers']:
-    env.Append(CPPDEFINES = 'GALLIUM_SOFTPIPE')
-    sources += ['xlib_softpipe.c']
-    drivers += [softpipe]
+if env['platform'] == 'linux':
 
-if 'llvmpipe' in env['drivers']:
-    env.Tool('llvm')
-    if 'LLVM_VERSION' in env:
-        env.Append(CPPDEFINES = 'GALLIUM_LLVMPIPE')
-        env.Tool('udis86')
-        sources += ['xlib_llvmpipe.c']
-        drivers += [llvmpipe]
-    
-if 'cell' in env['drivers']:
-    env.Append(CPPDEFINES = 'GALLIUM_CELL')
-    sources += ['xlib_cell.c']
-    drivers += [cell]
+    env = env.Clone()
 
-# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions
-libgl = env.SharedLibrary(
-    target ='GL',
-    source = sources,
-    LIBS = st_xlib + glapi + mesa + glsl + drivers + gallium + env['LIBS'],
-)
+    env.Append(CPPPATH = [
+        '#/src/gallium/include',
+        '#/src/gallium/auxiliary',
+        '#/src/gallium/drivers',
+    ])
 
-if not env['dri']:
-    # Only install this libGL.so if DRI not enabled
-    env.InstallSharedLibrary(libgl, version=(1, 5))
+    ws_xlib = env.ConvenienceLibrary(
+        target = 'ws_xlib',
+        source = [
+           'xlib_sw_winsys.c',
+        ]
+    )
+    Export('ws_xlib')
diff --git a/src/gallium/winsys/xlib/xlib.c b/src/gallium/winsys/xlib/xlib.c
deleted file mode 100644
index 67617a470d6..00000000000
--- a/src/gallium/winsys/xlib/xlib.c
+++ /dev/null
@@ -1,138 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- */
-
-#include "xlib.h"
-#include "xm_winsys.h"
-
-#include <stdlib.h>
-#include <assert.h>
-
-/* Todo, replace all this with callback-structs provided by the
- * individual implementations.
- */
-
-enum mode {
-   MODE_CELL,
-   MODE_LLVMPIPE,
-   MODE_SOFTPIPE
-};
-
-/* advertise OpenGL support */
-PUBLIC const int st_api_OpenGL = 1;
-
-static enum mode get_mode()
-{
-#ifdef GALLIUM_CELL
-   if (!getenv("GALLIUM_NOCELL")) 
-      return MODE_CELL;
-#endif
-
-#if defined(GALLIUM_LLVMPIPE)
-   return MODE_LLVMPIPE;
-#else
-   return MODE_SOFTPIPE;
-#endif
-}
-
-static void _init( void ) __attribute__((constructor));
-
-static void _init( void )
-{
-   enum mode xlib_mode = get_mode();
-
-   switch (xlib_mode) {
-   case MODE_CELL:
-#if defined(GALLIUM_CELL)
-      xmesa_set_driver( &xlib_cell_driver );
-#endif
-      break;
-   case MODE_LLVMPIPE:
-#if defined(GALLIUM_LLVMPIPE)
-      xmesa_set_driver( &xlib_llvmpipe_driver );
-#endif
-      break;
-   case MODE_SOFTPIPE:
-#if defined(GALLIUM_SOFTPIPE)
-      xmesa_set_driver( &xlib_softpipe_driver );
-#endif
-      break;
-   default:
-      assert(0);
-      break;
-   }
-}
-
-
-/***********************************************************************
- *
- * Butt-ugly hack to convince the linker not to throw away public GL
- * symbols (they are all referenced from getprocaddress, I guess).
- */
-extern void (*linker_foo(const unsigned char *procName))();
-extern void (*glXGetProcAddress(const unsigned char *procName))();
-
-extern void (*linker_foo(const unsigned char *procName))()
-{
-   return glXGetProcAddress(procName);
-}
-
-
-/**
- * When GLX_INDIRECT_RENDERING is defined, some symbols are missing in
- * libglapi.a.  We need to define them here.
- */
-#ifdef GLX_INDIRECT_RENDERING
-
-#define GL_GLEXT_PROTOTYPES
-#include "GL/gl.h"
-#include "glapi/glapi.h"
-#include "glapi/glapitable.h"
-#include "glapi/glapidispatch.h"
-
-#if defined(USE_MGL_NAMESPACE)
-#define NAME(func)  mgl##func
-#else
-#define NAME(func)  gl##func
-#endif
-
-#define DISPATCH(FUNC, ARGS, MESSAGE)		\
-   CALL_ ## FUNC(GET_DISPATCH(), ARGS);
-
-#define RETURN_DISPATCH(FUNC, ARGS, MESSAGE) 	\
-   return CALL_ ## FUNC(GET_DISPATCH(), ARGS);
-
-/* skip normal ones */
-#define _GLAPI_SKIP_NORMAL_ENTRY_POINTS
-#include "glapi/glapitemp.h"
-
-#endif /* GLX_INDIRECT_RENDERING */
diff --git a/src/gallium/winsys/xlib/xlib.h b/src/gallium/winsys/xlib/xlib.h
deleted file mode 100644
index 8e091d0c084..00000000000
--- a/src/gallium/winsys/xlib/xlib.h
+++ /dev/null
@@ -1,13 +0,0 @@
-
-#ifndef XLIB_H
-#define XLIB_H
-
-#include "pipe/p_compiler.h"
-#include "xm_winsys.h"
-
-extern struct xm_driver xlib_softpipe_driver;
-extern struct xm_driver xlib_llvmpipe_driver;
-extern struct xm_driver xlib_cell_driver;
-
-
-#endif
diff --git a/src/gallium/winsys/xlib/xlib_brw_context.c b/src/gallium/winsys/xlib/xlib_brw_context.c
deleted file mode 100644
index 22bf41a46f7..00000000000
--- a/src/gallium/winsys/xlib/xlib_brw_context.c
+++ /dev/null
@@ -1,209 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- *   Brian Paul
- */
-
-
-/* #include "glxheader.h" */
-/* #include "xmesaP.h" */
-
-#include "util/u_simple_screen.h"
-#include "util/u_inlines.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "i965simple/brw_winsys.h"
-#include "xlib_brw_aub.h"
-#include "xlib_brw.h"
-
-
-
-
-#define XBCWS_BATCHBUFFER_SIZE 1024
-
-
-/* The backend to the brw driver (ie struct brw_winsys) is actually a
- * per-context entity.
- */
-struct xlib_brw_context_winsys {
-   struct brw_winsys brw_context_winsys;   /**< batch buffer funcs */
-   struct aub_context *aub;
-                         
-   struct pipe_winsys *pipe_winsys;
-
-   unsigned batch_data[XBCWS_BATCHBUFFER_SIZE];
-   unsigned batch_nr;
-   unsigned batch_size;
-   unsigned batch_alloc;
-};
-
-
-/* Turn a brw_winsys into an xlib_brw_context_winsys:
- */
-static inline struct xlib_brw_context_winsys *
-xlib_brw_context_winsys( struct brw_winsys *sws )
-{
-   return (struct xlib_brw_context_winsys *)sws;
-}
-
-
-/* Simple batchbuffer interface:
- */
-
-static unsigned *xbcws_batch_start( struct brw_winsys *sws,
-					 unsigned dwords,
-					 unsigned relocs )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   if (xbcws->batch_size < xbcws->batch_nr + dwords)
-      return NULL;
-
-   xbcws->batch_alloc = xbcws->batch_nr + dwords;
-   return (void *)1;			/* not a valid pointer! */
-}
-
-static void xbcws_batch_dword( struct brw_winsys *sws,
-				    unsigned dword )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   assert(xbcws->batch_nr < xbcws->batch_alloc);
-   xbcws->batch_data[xbcws->batch_nr++] = dword;
-}
-
-static void xbcws_batch_reloc( struct brw_winsys *sws,
-			     struct pipe_buffer *buf,
-			     unsigned access_flags,
-			     unsigned delta )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   assert(xbcws->batch_nr < xbcws->batch_alloc);
-   xbcws->batch_data[xbcws->batch_nr++] = 
-      ( xlib_brw_get_buffer_offset( NULL, buf, access_flags ) +
-        delta );
-}
-
-static void xbcws_batch_end( struct brw_winsys *sws )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   assert(xbcws->batch_nr <= xbcws->batch_alloc);
-   xbcws->batch_alloc = 0;
-}
-
-static void xbcws_batch_flush( struct brw_winsys *sws,
-				    struct pipe_fence_handle **fence )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-   assert(xbcws->batch_nr <= xbcws->batch_size);
-
-   if (xbcws->batch_nr) {
-      xlib_brw_commands_aub( xbcws->pipe_winsys,
-                             xbcws->batch_data,
-                             xbcws->batch_nr );
-   }
-
-   xbcws->batch_nr = 0;
-}
-
-  
-
-/* Really a per-device function, just pass through:
- */
-static unsigned xbcws_get_buffer_offset( struct brw_winsys *sws,
-                                         struct pipe_buffer *buf,
-                                         unsigned access_flags )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   return xlib_brw_get_buffer_offset( xbcws->pipe_winsys,
-                                      buf,
-                                      access_flags );
-}
-
-
-/* Really a per-device function, just pass through:
- */
-static void xbcws_buffer_subdata_typed( struct brw_winsys *sws,
-                                       struct pipe_buffer *buf,
-                                       unsigned long offset, 
-                                       unsigned long size, 
-                                       const void *data,
-                                       unsigned data_type )
-{
-   struct xlib_brw_context_winsys *xbcws = xlib_brw_context_winsys(sws);
-
-   xlib_brw_buffer_subdata_typed( xbcws->pipe_winsys,
-                                  buf,
-                                  offset,
-                                  size,
-                                  data,
-                                  data_type );
-}
-
-
-/**
- * Create i965 hardware rendering context, but plugged into a
- * dump-to-aubfile backend.
- */
-struct pipe_context *
-xlib_create_brw_context( struct pipe_screen *screen,
-                         void *unused )
-{
-   struct xlib_brw_context_winsys *xbcws = CALLOC_STRUCT( xlib_brw_context_winsys );
-   
-   /* Fill in this struct with callbacks that i965simple will need to
-    * communicate with the window system, buffer manager, etc. 
-    */
-   xbcws->brw_context_winsys.batch_start = xbcws_batch_start;
-   xbcws->brw_context_winsys.batch_dword = xbcws_batch_dword;
-   xbcws->brw_context_winsys.batch_reloc = xbcws_batch_reloc;
-   xbcws->brw_context_winsys.batch_end = xbcws_batch_end;
-   xbcws->brw_context_winsys.batch_flush = xbcws_batch_flush;
-   xbcws->brw_context_winsys.buffer_subdata_typed = xbcws_buffer_subdata_typed;
-   xbcws->brw_context_winsys.get_buffer_offset = xbcws_get_buffer_offset;
-
-   xbcws->pipe_winsys = screen->winsys; /* redundant */
-
-   xbcws->batch_size = XBCWS_BATCHBUFFER_SIZE;
-
-   /* Create the i965simple context:
-    */
-#ifdef GALLIUM_CELL
-   return NULL;
-#else
-   return brw_create( screen,
-		      &xbcws->brw_context_winsys,
-		      0 );
-#endif
-}
diff --git a/src/gallium/winsys/xlib/xlib_cell.c b/src/gallium/winsys/xlib/xlib_cell.c
deleted file mode 100644
index 1dc9e8fa11f..00000000000
--- a/src/gallium/winsys/xlib/xlib_cell.c
+++ /dev/null
@@ -1,401 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- *   Brian Paul
- */
-
-#include "xlib.h"
-
-#ifdef GALLIUM_CELL
-
-#include "xm_api.h"
-
-#undef ASSERT
-#undef Elements
-
-#include "util/u_simple_screen.h"
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-
-#include "cell/ppu/cell_context.h"
-#include "cell/ppu/cell_screen.h"
-#include "cell/ppu/cell_winsys.h"
-#include "cell/ppu/cell_texture.h"
-
-
-/**
- * Subclass of pipe_buffer for Xlib winsys.
- * Low-level OS/window system memory buffer
- */
-struct xm_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-   
-   XImage *tempImage;
-   int shm;
-};
-
-
-/**
- * Subclass of pipe_winsys for Xlib winsys
- */
-struct xmesa_pipe_winsys
-{
-   struct pipe_winsys base;
-};
-
-
-
-/** Cast wrapper */
-static INLINE struct xm_buffer *
-xm_buffer( struct pipe_buffer *buf )
-{
-   return (struct xm_buffer *)buf;
-}
-
-
-/* Most callbacks map direcly onto dri_bufmgr operations:
- */
-static void *
-xm_buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
-              unsigned flags)
-{
-   struct xm_buffer *xm_buf = xm_buffer(buf);
-   xm_buf->mapped = xm_buf->data;
-   return xm_buf->mapped;
-}
-
-static void
-xm_buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
-{
-   struct xm_buffer *xm_buf = xm_buffer(buf);
-   xm_buf->mapped = NULL;
-}
-
-static void
-xm_buffer_destroy(/*struct pipe_winsys *pws,*/
-                  struct pipe_buffer *buf)
-{
-   struct xm_buffer *oldBuf = xm_buffer(buf);
-
-   if (oldBuf) {
-      if (oldBuf->data) {
-         if (!oldBuf->userBuffer) {
-            align_free(oldBuf->data);
-         }
-
-         oldBuf->data = NULL;
-      }
-      free(oldBuf);
-   }
-}
-
-
-/**
- * For Cell.  Basically, rearrange the pixels/quads from this layout:
- *  +--+--+--+--+
- *  |p0|p1|p2|p3|....
- *  +--+--+--+--+
- *
- * to this layout:
- *  +--+--+
- *  |p0|p1|....
- *  +--+--+
- *  |p2|p3|
- *  +--+--+
- */
-static void
-twiddle_tile(const uint *tileIn, uint *tileOut)
-{
-   int y, x;
-
-   for (y = 0; y < TILE_SIZE; y+=2) {
-      for (x = 0; x < TILE_SIZE; x+=2) {
-         int k = 4 * (y/2 * TILE_SIZE/2 + x/2);
-         tileOut[y * TILE_SIZE + (x + 0)] = tileIn[k];
-         tileOut[y * TILE_SIZE + (x + 1)] = tileIn[k+1];
-         tileOut[(y + 1) * TILE_SIZE + (x + 0)] = tileIn[k+2];
-         tileOut[(y + 1) * TILE_SIZE + (x + 1)] = tileIn[k+3];
-      }
-   }
-}
-
-
-
-/**
- * Display a surface that's in a tiled configuration.  That is, all the
- * pixels for a TILE_SIZExTILE_SIZE block are contiguous in memory.
- */
-static void
-xlib_cell_display_surface(struct xmesa_buffer *b, struct pipe_surface *surf)
-{
-   XImage *ximage;
-   struct xm_buffer *xm_buf = xm_buffer(
-      cell_texture(surf->texture)->buffer);
-   const uint tilesPerRow = (surf->width + TILE_SIZE - 1) / TILE_SIZE;
-   uint x, y;
-
-   ximage = b->tempImage;
-
-   /* check that the XImage has been previously initialized */
-   assert(ximage->format);
-   assert(ximage->bitmap_unit);
-
-   /* update XImage's fields */
-   ximage->width = TILE_SIZE;
-   ximage->height = TILE_SIZE;
-   ximage->bytes_per_line = TILE_SIZE * 4;
-
-   for (y = 0; y < surf->height; y += TILE_SIZE) {
-      for (x = 0; x < surf->width; x += TILE_SIZE) {
-         uint tmpTile[TILE_SIZE * TILE_SIZE];
-         int tx = x / TILE_SIZE;
-         int ty = y / TILE_SIZE;
-         int offset = ty * tilesPerRow + tx;
-         int w = TILE_SIZE;
-         int h = TILE_SIZE;
-
-         if (y + h > surf->height)
-            h = surf->height - y;
-         if (x + w > surf->width)
-            w = surf->width - x;
-
-         /* offset in pixels */
-         offset *= TILE_SIZE * TILE_SIZE;
-
-         /* twiddle from ximage buffer to temp tile */
-         twiddle_tile((uint *) xm_buf->data + offset, tmpTile);
-         /* display temp tile data */
-         ximage->data = (char *) tmpTile;
-         XPutImage(b->xm_visual->display, b->drawable, b->gc,
-                   ximage, 0, 0, x, y, w, h);
-      }
-   }
-}
-
-
-
-
-
-static void
-xm_flush_frontbuffer(struct pipe_winsys *pws,
-                     struct pipe_surface *surf,
-                     void *context_private)
-{
-   /*
-    * The front color buffer is actually just another XImage buffer.
-    * This function copies that XImage to the actual X Window.
-    */
-   XMesaContext xmctx = (XMesaContext) context_private;
-   if (xmctx)
-      xlib_cell_display_surface(xmctx->xm_buffer, surf);
-}
-
-
-
-static const char *
-xm_get_name(struct pipe_winsys *pws)
-{
-   return "Xlib/Cell";
-}
-
-
-static struct pipe_buffer *
-xm_buffer_create(struct pipe_winsys *pws, 
-                 unsigned alignment, 
-                 unsigned usage,
-                 unsigned size)
-{
-   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-
-   if (buffer->data == NULL) {
-      buffer->shm = 0;
-
-      /* align to 16-byte multiple for Cell */
-      buffer->data = align_malloc(size, max(alignment, 16));
-   }
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-xm_user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
-{
-   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-   buffer->shm = 0;
-
-   return &buffer->base;
-}
-
-
-
-static struct pipe_buffer *
-xm_surface_buffer_create(struct pipe_winsys *winsys,
-                         unsigned width, unsigned height,
-                         enum pipe_format format,
-                         unsigned usage,
-                         unsigned tex_usage,
-                         unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                /* XXX a bit of a hack */
-                                *stride * align(nblocksy, TILE_SIZE));
-}
-
-
-/*
- * Fence functions - basically nothing to do, as we don't create any actual
- * fence objects.
- */
-
-static void
-xm_fence_reference(struct pipe_winsys *sws, struct pipe_fence_handle **ptr,
-                   struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-xm_fence_signalled(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                   unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-xm_fence_finish(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                unsigned flag)
-{
-   return 0;
-}
-
-
-
-static struct pipe_winsys *
-xlib_create_cell_winsys( void )
-{
-   static struct xmesa_pipe_winsys *ws = NULL;
-
-   if (!ws) {
-      ws = CALLOC_STRUCT(xmesa_pipe_winsys);
-
-      /* Fill in this struct with callbacks that pipe will need to
-       * communicate with the window system, buffer manager, etc. 
-       */
-      ws->base.buffer_create = xm_buffer_create;
-      ws->base.user_buffer_create = xm_user_buffer_create;
-      ws->base.buffer_map = xm_buffer_map;
-      ws->base.buffer_unmap = xm_buffer_unmap;
-      ws->base.buffer_destroy = xm_buffer_destroy;
-
-      ws->base.surface_buffer_create = xm_surface_buffer_create;
-
-      ws->base.fence_reference = xm_fence_reference;
-      ws->base.fence_signalled = xm_fence_signalled;
-      ws->base.fence_finish = xm_fence_finish;
-
-      ws->base.flush_frontbuffer = xm_flush_frontbuffer;
-      ws->base.get_name = xm_get_name;
-   }
-
-   return &ws->base;
-}
-
-
-static struct pipe_screen *
-xlib_create_cell_screen( void )
-{
-   struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = xlib_create_cell_winsys();
-   if (winsys == NULL)
-      return NULL;
-
-   screen = cell_create_screen(winsys);
-   if (screen == NULL)
-      goto fail;
-
-   return screen;
-
-fail:
-   if (winsys)
-      winsys->destroy( winsys );
-
-   return NULL;
-}
-
-
-
-struct xm_driver xlib_cell_driver = 
-{
-   .create_pipe_screen = xlib_create_cell_screen,
-   .display_surface = xlib_cell_display_surface,
-};
-
-#else
-
-struct xm_driver xlib_cell_driver = 
-{
-   .create_pipe_screen = NULL,
-   .display_surface = NULL,
-};
-
-#endif
diff --git a/src/gallium/winsys/xlib/xlib_llvmpipe.c b/src/gallium/winsys/xlib/xlib_llvmpipe.c
deleted file mode 100644
index 6cebd4c2012..00000000000
--- a/src/gallium/winsys/xlib/xlib_llvmpipe.c
+++ /dev/null
@@ -1,442 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- *   Brian Paul
- */
-
-
-#if defined(GALLIUM_LLVMPIPE)
-
-#include "xm_api.h"
-
-#undef ASSERT
-#undef Elements
-
-#include "util/u_simple_screen.h"
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "llvmpipe/lp_winsys.h"
-#include "llvmpipe/lp_texture.h"
-
-#include "xlib.h"
-
-/**
- * Subclass of pipe_buffer for Xlib winsys.
- * Low-level OS/window system memory buffer
- */
-struct xm_displaytarget
-{
-   enum pipe_format format;
-   unsigned width;
-   unsigned height;
-   unsigned stride;
-
-   void *data;
-   void *mapped;
-
-   XImage *tempImage;
-#ifdef USE_XSHM
-   int shm;
-   XShmSegmentInfo shminfo;
-#endif
-};
-
-
-/**
- * Subclass of llvmpipe_winsys for Xlib winsys
- */
-struct xmesa_llvmpipe_winsys
-{
-   struct llvmpipe_winsys base;
-/*   struct xmesa_visual *xm_visual; */
-};
-
-
-
-/** Cast wrapper */
-static INLINE struct xm_displaytarget *
-xm_displaytarget( struct llvmpipe_displaytarget *dt )
-{
-   return (struct xm_displaytarget *)dt;
-}
-
-
-/**
- * X Shared Memory Image extension code
- */
-
-#ifdef USE_XSHM
-
-static volatile int mesaXErrorFlag = 0;
-
-/**
- * Catches potential Xlib errors.
- */
-static int
-mesaHandleXError(Display *dpy, XErrorEvent *event)
-{
-   (void) dpy;
-   (void) event;
-   mesaXErrorFlag = 1;
-   return 0;
-}
-
-
-static char *alloc_shm(struct xm_displaytarget *buf, unsigned size)
-{
-   XShmSegmentInfo *const shminfo = & buf->shminfo;
-
-   shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777);
-   if (shminfo->shmid < 0) {
-      return NULL;
-   }
-
-   shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0);
-   if (shminfo->shmaddr == (char *) -1) {
-      shmctl(shminfo->shmid, IPC_RMID, 0);
-      return NULL;
-   }
-
-   shminfo->readOnly = False;
-   return shminfo->shmaddr;
-}
-
-
-/**
- * Allocate a shared memory XImage back buffer for the given XMesaBuffer.
- */
-static void
-alloc_shm_ximage(struct xm_displaytarget *xm_buffer,
-                 struct xmesa_buffer *xmb,
-                 unsigned width, unsigned height)
-{
-   /*
-    * We have to do a _lot_ of error checking here to be sure we can
-    * really use the XSHM extension.  It seems different servers trigger
-    * errors at different points if the extension won't work.  Therefore
-    * we have to be very careful...
-    */
-   int (*old_handler)(Display *, XErrorEvent *);
-
-   xm_buffer->tempImage = XShmCreateImage(xmb->xm_visual->display,
-                                  xmb->xm_visual->visinfo->visual,
-                                  xmb->xm_visual->visinfo->depth,
-                                  ZPixmap,
-                                  NULL,
-                                  &xm_buffer->shminfo,
-                                  width, height);
-   if (xm_buffer->tempImage == NULL) {
-      xm_buffer->shm = 0;
-      return;
-   }
-
-
-   mesaXErrorFlag = 0;
-   old_handler = XSetErrorHandler(mesaHandleXError);
-   /* This may trigger the X protocol error we're ready to catch: */
-   XShmAttach(xmb->xm_visual->display, &xm_buffer->shminfo);
-   XSync(xmb->xm_visual->display, False);
-
-   if (mesaXErrorFlag) {
-      /* we are on a remote display, this error is normal, don't print it */
-      XFlush(xmb->xm_visual->display);
-      mesaXErrorFlag = 0;
-      XDestroyImage(xm_buffer->tempImage);
-      xm_buffer->tempImage = NULL;
-      xm_buffer->shm = 0;
-      (void) XSetErrorHandler(old_handler);
-      return;
-   }
-
-   xm_buffer->shm = 1;
-}
-
-#endif /* USE_XSHM */
-
-static boolean
-xm_is_displaytarget_format_supported( struct llvmpipe_winsys *ws,
-                                      enum pipe_format format )
-{
-   /* TODO: check visuals or other sensible thing here */
-   return TRUE;
-}
-
-
-static void *
-xm_displaytarget_map(struct llvmpipe_winsys *ws,
-                     struct llvmpipe_displaytarget *dt,
-                     unsigned flags)
-{
-   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
-   xm_dt->mapped = xm_dt->data;
-   return xm_dt->mapped;
-}
-
-static void
-xm_displaytarget_unmap(struct llvmpipe_winsys *ws,
-                       struct llvmpipe_displaytarget *dt)
-{
-   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
-   xm_dt->mapped = NULL;
-}
-
-static void
-xm_displaytarget_destroy(struct llvmpipe_winsys *ws,
-                         struct llvmpipe_displaytarget *dt)
-{
-   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
-
-   if (xm_dt->data) {
-#ifdef USE_XSHM
-      if (xm_dt->shminfo.shmid >= 0) {
-         shmdt(xm_dt->shminfo.shmaddr);
-         shmctl(xm_dt->shminfo.shmid, IPC_RMID, 0);
-         
-         xm_dt->shminfo.shmid = -1;
-         xm_dt->shminfo.shmaddr = (char *) -1;
-      }
-      else
-#endif
-         FREE(xm_dt->data);
-   }
-
-   FREE(xm_dt);
-}
-
-
-/**
- * Display/copy the image in the surface into the X window specified
- * by the XMesaBuffer.
- */
-static void
-xm_llvmpipe_display(struct xmesa_buffer *xm_buffer,
-                    struct llvmpipe_displaytarget *dt)
-{
-   XImage *ximage;
-   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
-   static boolean no_swap = 0;
-   static boolean firsttime = 1;
-
-   if (firsttime) {
-      no_swap = getenv("SP_NO_RAST") != NULL;
-      firsttime = 0;
-   }
-
-   if (no_swap)
-      return;
-
-#ifdef USE_XSHM
-   if (xm_dt->shm)
-   {
-      if (xm_dt->tempImage == NULL)
-      {
-         assert(util_format_get_blockwidth(xm_dt->format) == 1);
-         assert(util_format_get_blockheight(xm_dt->format) == 1);
-         alloc_shm_ximage(xm_dt, xm_buffer,
-                          xm_dt->stride / util_format_get_blocksize(xm_dt->format),
-                          xm_dt->height);
-      }
-
-      ximage = xm_dt->tempImage;
-      ximage->data = xm_dt->data;
-
-      /* _debug_printf("XSHM\n"); */
-      XShmPutImage(xm_buffer->xm_visual->display, xm_buffer->drawable, xm_buffer->gc,
-                   ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height, False);
-   }
-   else
-#endif
-   {
-      /* display image in Window */
-      ximage = xm_dt->tempImage;
-      ximage->data = xm_dt->data;
-
-      /* check that the XImage has been previously initialized */
-      assert(ximage->format);
-      assert(ximage->bitmap_unit);
-
-      /* update XImage's fields */
-      ximage->width = xm_dt->width;
-      ximage->height = xm_dt->height;
-      ximage->bytes_per_line = xm_dt->stride;
-
-      /* _debug_printf("XPUT\n"); */
-      XPutImage(xm_buffer->xm_visual->display, xm_buffer->drawable, xm_buffer->gc,
-                ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height);
-   }
-}
-
-/**
- * Display/copy the image in the surface into the X window specified
- * by the XMesaBuffer.
- */
-static void
-xm_displaytarget_display(struct llvmpipe_winsys *ws,
-                         struct llvmpipe_displaytarget *dt,
-                         void *context_private)
-{
-   XMesaContext xmctx = (XMesaContext) context_private;
-   struct xmesa_buffer *xm_buffer = xmctx->xm_buffer;
-   xm_llvmpipe_display(xm_buffer, dt);
-}
-
-
-static struct llvmpipe_displaytarget *
-xm_displaytarget_create(struct llvmpipe_winsys *winsys,
-                        enum pipe_format format,
-                        unsigned width, unsigned height,
-                        unsigned alignment,
-                        unsigned *stride)
-{
-   struct xm_displaytarget *xm_dt = CALLOC_STRUCT(xm_displaytarget);
-   unsigned nblocksy, size;
-
-   xm_dt = CALLOC_STRUCT(xm_displaytarget);
-   if(!xm_dt)
-      goto no_xm_dt;
-
-   xm_dt->format = format;
-   xm_dt->width = width;
-   xm_dt->height = height;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   xm_dt->stride = align(util_format_get_stride(format, width), alignment);
-   size = xm_dt->stride * nblocksy;
-
-#ifdef USE_XSHM
-   if (!debug_get_bool_option("XLIB_NO_SHM", FALSE))
-   {
-      xm_dt->shminfo.shmid = -1;
-      xm_dt->shminfo.shmaddr = (char *) -1;
-      xm_dt->shm = TRUE;
-         
-      xm_dt->data = alloc_shm(xm_dt, size);
-      if(!xm_dt->data)
-         goto no_data;
-   }
-#endif
-
-   if(!xm_dt->data) {
-      xm_dt->data = align_malloc(size, alignment);
-      if(!xm_dt->data)
-         goto no_data;
-   }
-
-   *stride = xm_dt->stride;
-   return (struct llvmpipe_displaytarget *)xm_dt;
-
-no_data:
-   FREE(xm_dt);
-no_xm_dt:
-   return NULL;
-}
-
-
-static void
-xm_destroy( struct llvmpipe_winsys *ws )
-{
-   FREE(ws);
-}
-
-
-static struct llvmpipe_winsys *
-xlib_create_llvmpipe_winsys( void )
-{
-   struct xmesa_llvmpipe_winsys *ws;
-
-   ws = CALLOC_STRUCT(xmesa_llvmpipe_winsys);
-   if (!ws)
-      return NULL;
-
-   ws->base.destroy = xm_destroy;
-
-   ws->base.is_displaytarget_format_supported = xm_is_displaytarget_format_supported;
-
-   ws->base.displaytarget_create = xm_displaytarget_create;
-   ws->base.displaytarget_map = xm_displaytarget_map;
-   ws->base.displaytarget_unmap = xm_displaytarget_unmap;
-   ws->base.displaytarget_destroy = xm_displaytarget_destroy;
-
-   ws->base.displaytarget_display = xm_displaytarget_display;
-
-   return &ws->base;
-}
-
-
-static struct pipe_screen *
-xlib_create_llvmpipe_screen( void )
-{
-   struct llvmpipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = xlib_create_llvmpipe_winsys();
-   if (winsys == NULL)
-      return NULL;
-
-   screen = llvmpipe_create_screen(winsys);
-   if (screen == NULL)
-      goto fail;
-
-   return screen;
-
-fail:
-   if (winsys)
-      winsys->destroy( winsys );
-
-   return NULL;
-}
-
-
-static void
-xlib_llvmpipe_display_surface(struct xmesa_buffer *xm_buffer,
-                              struct pipe_surface *surf)
-{
-   struct llvmpipe_texture *texture = llvmpipe_texture(surf->texture);
-
-   assert(texture->dt);
-   if (texture->dt)
-      xm_llvmpipe_display(xm_buffer, texture->dt);
-}
-
-
-struct xm_driver xlib_llvmpipe_driver = 
-{
-   .create_pipe_screen = xlib_create_llvmpipe_screen,
-   .display_surface = xlib_llvmpipe_display_surface
-};
-
-
-
-#endif /* GALLIUM_LLVMPIPE */
diff --git a/src/gallium/winsys/xlib/xlib_softpipe.c b/src/gallium/winsys/xlib/xlib_softpipe.c
deleted file mode 100644
index 716338aef47..00000000000
--- a/src/gallium/winsys/xlib/xlib_softpipe.c
+++ /dev/null
@@ -1,508 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
- * 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 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
- * THE COPYRIGHT HOLDERS, AUTHORS 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.
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * 
- **************************************************************************/
-
-/*
- * Authors:
- *   Keith Whitwell
- *   Brian Paul
- */
-
-
-#include "xm_api.h"
-
-#undef ASSERT
-#undef Elements
-
-#include "util/u_simple_screen.h"
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_inlines.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "softpipe/sp_winsys.h"
-#include "softpipe/sp_texture.h"
-
-#include "xlib.h"
-
-/**
- * Subclass of pipe_buffer for Xlib winsys.
- * Low-level OS/window system memory buffer
- */
-struct xm_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-
-   XImage *tempImage;
-#ifdef USE_XSHM
-   boolean shm;         /** Is this a shared memory buffer? */
-   XShmSegmentInfo shminfo;
-#endif
-};
-
-
-/**
- * Subclass of pipe_winsys for Xlib winsys
- */
-struct xmesa_pipe_winsys
-{
-   struct pipe_winsys base;
-/*   struct xmesa_visual *xm_visual; */
-};
-
-
-
-/** Cast wrapper */
-static INLINE struct xm_buffer *
-xm_buffer( struct pipe_buffer *buf )
-{
-   return (struct xm_buffer *)buf;
-}
-
-
-/**
- * X Shared Memory Image extension code
- */
-
-#ifdef USE_XSHM
-
-static volatile int mesaXErrorFlag = 0;
-
-/**
- * Catches potential Xlib errors.
- */
-static int
-mesaHandleXError(Display *dpy, XErrorEvent *event)
-{
-   (void) dpy;
-   (void) event;
-   mesaXErrorFlag = 1;
-   return 0;
-}
-
-
-static char *alloc_shm(struct xm_buffer *buf, unsigned size)
-{
-   XShmSegmentInfo *const shminfo = & buf->shminfo;
-
-   shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777);
-   if (shminfo->shmid < 0) {
-      return NULL;
-   }
-
-   shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0);
-   if (shminfo->shmaddr == (char *) -1) {
-      shmctl(shminfo->shmid, IPC_RMID, 0);
-      return NULL;
-   }
-
-   shminfo->readOnly = False;
-   return shminfo->shmaddr;
-}
-
-
-/**
- * Allocate a shared memory XImage back buffer for the given XMesaBuffer.
- */
-static void
-alloc_shm_ximage(struct xm_buffer *b, struct xmesa_buffer *xmb,
-                 unsigned width, unsigned height)
-{
-   /*
-    * We have to do a _lot_ of error checking here to be sure we can
-    * really use the XSHM extension.  It seems different servers trigger
-    * errors at different points if the extension won't work.  Therefore
-    * we have to be very careful...
-    */
-   int (*old_handler)(Display *, XErrorEvent *);
-
-   b->tempImage = XShmCreateImage(xmb->xm_visual->display,
-                                  xmb->xm_visual->visinfo->visual,
-                                  xmb->xm_visual->visinfo->depth,
-                                  ZPixmap,
-                                  NULL,
-                                  &b->shminfo,
-                                  width, height);
-   if (b->tempImage == NULL) {
-      b->shm = FALSE;
-      return;
-   }
-
-
-   mesaXErrorFlag = 0;
-   old_handler = XSetErrorHandler(mesaHandleXError);
-   /* This may trigger the X protocol error we're ready to catch: */
-   XShmAttach(xmb->xm_visual->display, &b->shminfo);
-   XSync(xmb->xm_visual->display, False);
-
-   if (mesaXErrorFlag) {
-      /* we are on a remote display, this error is normal, don't print it */
-      XFlush(xmb->xm_visual->display);
-      mesaXErrorFlag = 0;
-      XDestroyImage(b->tempImage);
-      b->tempImage = NULL;
-      b->shm = FALSE;
-      (void) XSetErrorHandler(old_handler);
-      return;
-   }
-
-   b->shm = TRUE;
-}
-
-#endif /* USE_XSHM */
-
-
-
-/* Most callbacks map direcly onto dri_bufmgr operations:
- */
-static void *
-xm_buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
-              unsigned flags)
-{
-   struct xm_buffer *xm_buf = xm_buffer(buf);
-   xm_buf->mapped = xm_buf->data;
-   return xm_buf->mapped;
-}
-
-static void
-xm_buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buf)
-{
-   struct xm_buffer *xm_buf = xm_buffer(buf);
-   xm_buf->mapped = NULL;
-}
-
-static void
-xm_buffer_destroy(struct pipe_buffer *buf)
-{
-   struct xm_buffer *oldBuf = xm_buffer(buf);
-
-   /*
-    * Note oldBuf->data may point to one of three things:
-    * 1. XShm shared memory image data
-    * 2. User-provided (wrapped) memory, see xm_user_buffer_create()
-    * 3. Regular, malloc'd memory
-    * We need to be careful with freeing that data now.
-    */
-
-   if (oldBuf->data) {
-#ifdef USE_XSHM
-      if (oldBuf->shminfo.shmid >= 0) {
-         shmdt(oldBuf->shminfo.shmaddr);
-         shmctl(oldBuf->shminfo.shmid, IPC_RMID, 0);
-         
-         oldBuf->shminfo.shmid = -1;
-         oldBuf->shminfo.shmaddr = (char *) -1;
-      }
-
-      if (oldBuf->shm) {
-         oldBuf->data = NULL;
-      }
-
-      if (oldBuf->tempImage) {
-         XDestroyImage(oldBuf->tempImage);
-         oldBuf->tempImage = NULL;
-      }
-#endif
-
-      if (oldBuf->data && !oldBuf->userBuffer) {
-         /* this was regular malloc'd memory */
-         align_free(oldBuf->data);
-      }
-
-      oldBuf->data = NULL;
-   }
-
-   free(oldBuf);
-}
-
-
-/**
- * Display/copy the image in the surface into the X window specified
- * by the XMesaBuffer.
- */
-static void
-xlib_softpipe_display_surface(struct xmesa_buffer *b,
-                              struct pipe_surface *surf)
-{
-   XImage *ximage;
-   struct softpipe_texture *spt = softpipe_texture(surf->texture);
-   struct xm_buffer *xm_buf = xm_buffer(spt->buffer);
-   static boolean no_swap = 0;
-   static boolean firsttime = 1;
-
-   if (firsttime) {
-      no_swap = getenv("SP_NO_RAST") != NULL;
-      firsttime = 0;
-   }
-
-   if (no_swap)
-      return;
-
-#ifdef USE_XSHM
-   if (xm_buf->shm)
-   {
-      if (xm_buf->tempImage == NULL) 
-      {
-         assert(util_format_get_blockwidth(surf->texture->format) == 1);
-         assert(util_format_get_blockheight(surf->texture->format) == 1);
-         alloc_shm_ximage(xm_buf, b, spt->stride[surf->level] /
-                          util_format_get_blocksize(surf->texture->format), surf->height);
-      }
-
-      ximage = xm_buf->tempImage;
-      ximage->data = xm_buf->data;
-
-      /* _debug_printf("XSHM\n"); */
-      XShmPutImage(b->xm_visual->display, b->drawable, b->gc,
-                   ximage, 0, 0, 0, 0, surf->width, surf->height, False);
-   }
-   else
-#endif
-   {
-      /* display image in Window */
-      ximage = b->tempImage;
-      ximage->data = xm_buf->data;
-
-      /* check that the XImage has been previously initialized */
-      assert(ximage->format);
-      assert(ximage->bitmap_unit);
-
-      /* update XImage's fields */
-      ximage->width = surf->width;
-      ximage->height = surf->height;
-      ximage->bytes_per_line = spt->stride[surf->level];
-
-      /* _debug_printf("XPUT\n"); */
-      XPutImage(b->xm_visual->display, b->drawable, b->gc,
-                ximage, 0, 0, 0, 0, surf->width, surf->height);
-   }
-}
-
-
-static void
-xm_flush_frontbuffer(struct pipe_winsys *pws,
-                     struct pipe_surface *surf,
-                     void *context_private)
-{
-   /*
-    * The front color buffer is actually just another XImage buffer.
-    * This function copies that XImage to the actual X Window.
-    */
-   XMesaContext xmctx = (XMesaContext) context_private;
-   xlib_softpipe_display_surface(xmctx->xm_buffer, surf);
-   xmesa_check_and_update_buffer_size(xmctx, xmctx->xm_buffer);
-}
-
-
-
-static const char *
-xm_get_name(struct pipe_winsys *pws)
-{
-   return "Xlib";
-}
-
-
-static struct pipe_buffer *
-xm_buffer_create(struct pipe_winsys *pws, 
-                 unsigned alignment, 
-                 unsigned usage,
-                 unsigned size)
-{
-   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-   /* align to 16-byte multiple for Cell */
-   buffer->data = align_malloc(size, max(alignment, 16));
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-xm_user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
-{
-   struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-
-   return &buffer->base;
-}
-
-
-static struct pipe_buffer *
-xm_surface_buffer_create(struct pipe_winsys *winsys,
-                         unsigned width, unsigned height,
-                         enum pipe_format format,
-                         unsigned usage,
-                         unsigned tex_usage,
-                         unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy, size;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-   size = *stride * nblocksy;
-
-#ifdef USE_XSHM
-   if (!debug_get_bool_option("XLIB_NO_SHM", FALSE))
-   {
-      struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-
-      pipe_reference_init(&buffer->base.reference, 1);
-      buffer->base.alignment = alignment;
-      buffer->base.usage = usage;
-      buffer->base.size = size;
-      buffer->userBuffer = FALSE;
-      buffer->shminfo.shmid = -1;
-      buffer->shminfo.shmaddr = (char *) -1;
-      buffer->shm = TRUE;
-         
-      buffer->data = alloc_shm(buffer, size);
-      if (!buffer->data)
-         goto out;
-
-      return &buffer->base;
-         
-   out:
-      if (buffer)
-         FREE(buffer);
-   }
-#endif
-   
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                size);
-}
-
-
-/*
- * Fence functions - basically nothing to do, as we don't create any actual
- * fence objects.
- */
-
-static void
-xm_fence_reference(struct pipe_winsys *sws, struct pipe_fence_handle **ptr,
-                   struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-xm_fence_signalled(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                   unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-xm_fence_finish(struct pipe_winsys *sws, struct pipe_fence_handle *fence,
-                unsigned flag)
-{
-   return 0;
-}
-
-
-
-static struct pipe_winsys *
-xlib_create_softpipe_winsys( void )
-{
-   static struct xmesa_pipe_winsys *ws = NULL;
-
-   if (!ws) {
-      ws = CALLOC_STRUCT(xmesa_pipe_winsys);
-
-      /* Fill in this struct with callbacks that pipe will need to
-       * communicate with the window system, buffer manager, etc. 
-       */
-      ws->base.buffer_create = xm_buffer_create;
-      ws->base.user_buffer_create = xm_user_buffer_create;
-      ws->base.buffer_map = xm_buffer_map;
-      ws->base.buffer_unmap = xm_buffer_unmap;
-      ws->base.buffer_destroy = xm_buffer_destroy;
-
-      ws->base.surface_buffer_create = xm_surface_buffer_create;
-
-      ws->base.fence_reference = xm_fence_reference;
-      ws->base.fence_signalled = xm_fence_signalled;
-      ws->base.fence_finish = xm_fence_finish;
-
-      ws->base.flush_frontbuffer = xm_flush_frontbuffer;
-      ws->base.get_name = xm_get_name;
-   }
-
-   return &ws->base;
-}
-
-
-static struct pipe_screen *
-xlib_create_softpipe_screen( void )
-{
-   struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = xlib_create_softpipe_winsys();
-   if (winsys == NULL)
-      return NULL;
-
-   screen = softpipe_create_screen(winsys);
-   if (screen == NULL)
-      goto fail;
-
-   return screen;
-
-fail:
-   if (winsys)
-      winsys->destroy( winsys );
-
-   return NULL;
-}
-
-
-struct xm_driver xlib_softpipe_driver = 
-{
-   .create_pipe_screen = xlib_create_softpipe_screen,
-   .display_surface = xlib_softpipe_display_surface
-};
-
-
-
diff --git a/src/gallium/winsys/xlib/xlib_sw_winsys.c b/src/gallium/winsys/xlib/xlib_sw_winsys.c
new file mode 100644
index 00000000000..29c642706a4
--- /dev/null
+++ b/src/gallium/winsys/xlib/xlib_sw_winsys.c
@@ -0,0 +1,439 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * 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 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell
+ *   Brian Paul
+ */
+
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "state_tracker/xlib_sw_winsys.h"
+
+#include <X11/Xlib.h>
+#include <X11/Xlibint.h>
+#include <X11/Xutil.h>
+#include <sys/ipc.h>
+#include <sys/shm.h>
+#include <X11/extensions/XShm.h>
+
+/**
+ * Subclass of pipe_buffer for Xlib winsys.
+ * Low-level OS/window system memory buffer
+ */
+struct xm_displaytarget
+{
+   enum pipe_format format;
+   unsigned width;
+   unsigned height;
+   unsigned stride;
+
+   void *data;
+   void *mapped;
+
+   Display *display;
+   Visual *visual;
+   XImage *tempImage;
+   GC gc;
+
+   /* This is the last drawable that this display target was presented
+    * against.  May need to recreate gc, tempImage when this changes??
+    */
+   Drawable drawable;
+
+   XShmSegmentInfo shminfo;
+   int shm;
+};
+
+
+/**
+ * Subclass of sw_winsys for Xlib winsys
+ */
+struct xlib_sw_winsys
+{
+   struct sw_winsys base;
+
+
+
+   Display *display;
+};
+
+
+
+/** Cast wrapper */
+static INLINE struct xm_displaytarget *
+xm_displaytarget( struct sw_displaytarget *dt )
+{
+   return (struct xm_displaytarget *)dt;
+}
+
+
+/**
+ * X Shared Memory Image extension code
+ */
+
+static volatile int mesaXErrorFlag = 0;
+
+/**
+ * Catches potential Xlib errors.
+ */
+static int
+mesaHandleXError(Display *dpy, XErrorEvent *event)
+{
+   (void) dpy;
+   (void) event;
+   mesaXErrorFlag = 1;
+   return 0;
+}
+
+
+static char *alloc_shm(struct xm_displaytarget *buf, unsigned size)
+{
+   XShmSegmentInfo *const shminfo = & buf->shminfo;
+
+   shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777);
+   if (shminfo->shmid < 0) {
+      return NULL;
+   }
+
+   shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0);
+   if (shminfo->shmaddr == (char *) -1) {
+      shmctl(shminfo->shmid, IPC_RMID, 0);
+      return NULL;
+   }
+
+   shminfo->readOnly = False;
+   return shminfo->shmaddr;
+}
+
+
+/**
+ * Allocate a shared memory XImage back buffer for the given XMesaBuffer.
+ */
+static void
+alloc_shm_ximage(struct xm_displaytarget *xm_dt,
+                 struct xlib_drawable *xmb,
+                 unsigned width, unsigned height)
+{
+   /*
+    * We have to do a _lot_ of error checking here to be sure we can
+    * really use the XSHM extension.  It seems different servers trigger
+    * errors at different points if the extension won't work.  Therefore
+    * we have to be very careful...
+    */
+   int (*old_handler)(Display *, XErrorEvent *);
+
+   xm_dt->tempImage = XShmCreateImage(xm_dt->display,
+                                      xmb->visual,
+                                      xmb->depth,
+                                      ZPixmap,
+                                      NULL,
+                                      &xm_dt->shminfo,
+                                      width, height);
+   if (xm_dt->tempImage == NULL) {
+      xm_dt->shm = 0;
+      return;
+   }
+
+
+   mesaXErrorFlag = 0;
+   old_handler = XSetErrorHandler(mesaHandleXError);
+   /* This may trigger the X protocol error we're ready to catch: */
+   XShmAttach(xm_dt->display, &xm_dt->shminfo);
+   XSync(xm_dt->display, False);
+
+   if (mesaXErrorFlag) {
+      /* we are on a remote display, this error is normal, don't print it */
+      XFlush(xm_dt->display);
+      mesaXErrorFlag = 0;
+      XDestroyImage(xm_dt->tempImage);
+      xm_dt->tempImage = NULL;
+      xm_dt->shm = 0;
+      (void) XSetErrorHandler(old_handler);
+      return;
+   }
+
+   xm_dt->shm = 1;
+}
+
+
+static void
+alloc_ximage(struct xm_displaytarget *xm_dt,
+             struct xlib_drawable *xmb,
+             unsigned width, unsigned height)
+{
+   if (xm_dt->shm) {
+      alloc_shm_ximage(xm_dt, xmb, width, height);
+      return;
+   }
+
+   xm_dt->tempImage = XCreateImage(xm_dt->display,
+                                   xmb->visual,
+                                   xmb->depth,
+                                   ZPixmap, 0,
+                                   NULL, width, height,
+                                   8, 0);
+}
+
+static boolean
+xm_is_displaytarget_format_supported( struct sw_winsys *ws,
+                                      enum pipe_format format )
+{
+   /* TODO: check visuals or other sensible thing here */
+   return TRUE;
+}
+
+
+static void *
+xm_displaytarget_map(struct sw_winsys *ws,
+                     struct sw_displaytarget *dt,
+                     unsigned flags)
+{
+   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
+   xm_dt->mapped = xm_dt->data;
+   return xm_dt->mapped;
+}
+
+static void
+xm_displaytarget_unmap(struct sw_winsys *ws,
+                       struct sw_displaytarget *dt)
+{
+   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
+   xm_dt->mapped = NULL;
+}
+
+static void
+xm_displaytarget_destroy(struct sw_winsys *ws,
+                         struct sw_displaytarget *dt)
+{
+   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
+
+   if (xm_dt->data) {
+      if (xm_dt->shminfo.shmid >= 0) {
+         shmdt(xm_dt->shminfo.shmaddr);
+         shmctl(xm_dt->shminfo.shmid, IPC_RMID, 0);
+         
+         xm_dt->shminfo.shmid = -1;
+         xm_dt->shminfo.shmaddr = (char *) -1;
+      }
+      else {
+         FREE(xm_dt->data);
+      }
+   }
+
+   if (xm_dt->tempImage)
+      XDestroyImage(xm_dt->tempImage);
+
+   if (xm_dt->gc)
+      XFreeGC(xm_dt->display, xm_dt->gc);
+
+   FREE(xm_dt);
+}
+
+
+/**
+ * Display/copy the image in the surface into the X window specified
+ * by the XMesaBuffer.
+ */
+static void
+xlib_sw_display(struct xlib_drawable *xlib_drawable,
+                struct sw_displaytarget *dt)
+{
+   static boolean no_swap = 0;
+   static boolean firsttime = 1;
+   struct xm_displaytarget *xm_dt = xm_displaytarget(dt);
+   Display *display = xm_dt->display;
+   XImage *ximage;
+
+   if (firsttime) {
+      no_swap = getenv("SP_NO_RAST") != NULL;
+      firsttime = 0;
+   }
+
+   if (no_swap)
+      return;
+
+   if (xm_dt->drawable != xlib_drawable->drawable) {
+      if (xm_dt->gc) {
+         XFreeGC( display, xm_dt->gc );
+         xm_dt->gc = NULL;
+      }
+
+      if (xm_dt->tempImage) {
+         XDestroyImage( xm_dt->tempImage );
+         xm_dt->tempImage = NULL;
+      }
+
+      xm_dt->drawable = xlib_drawable->drawable;
+   }
+
+   if (xm_dt->tempImage == NULL) {
+      assert(util_format_get_blockwidth(xm_dt->format) == 1);
+      assert(util_format_get_blockheight(xm_dt->format) == 1);
+      alloc_ximage(xm_dt, xlib_drawable,
+                   xm_dt->stride / util_format_get_blocksize(xm_dt->format),
+                   xm_dt->height);
+      if (!xm_dt->tempImage)
+         return;
+   }
+
+   if (xm_dt->gc == NULL) {
+      xm_dt->gc = XCreateGC( display, xlib_drawable->drawable, 0, NULL );
+      XSetFunction( display, xm_dt->gc, GXcopy );
+   }
+
+   if (xm_dt->shm)
+   {
+      ximage = xm_dt->tempImage;
+      ximage->data = xm_dt->data;
+
+      /* _debug_printf("XSHM\n"); */
+      XShmPutImage(xm_dt->display, xlib_drawable->drawable, xm_dt->gc,
+                   ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height, False);
+   }
+   else {
+      /* display image in Window */
+      ximage = xm_dt->tempImage;
+      ximage->data = xm_dt->data;
+
+      /* check that the XImage has been previously initialized */
+      assert(ximage->format);
+      assert(ximage->bitmap_unit);
+
+      /* update XImage's fields */
+      ximage->width = xm_dt->width;
+      ximage->height = xm_dt->height;
+      ximage->bytes_per_line = xm_dt->stride;
+
+      /* _debug_printf("XPUT\n"); */
+      XPutImage(xm_dt->display, xlib_drawable->drawable, xm_dt->gc,
+                ximage, 0, 0, 0, 0, xm_dt->width, xm_dt->height);
+   }
+}
+
+/**
+ * Display/copy the image in the surface into the X window specified
+ * by the XMesaBuffer.
+ */
+static void
+xm_displaytarget_display(struct sw_winsys *ws,
+                         struct sw_displaytarget *dt,
+                         void *context_private)
+{
+   struct xlib_drawable *xlib_drawable = (struct xlib_drawable *)context_private;
+   xlib_sw_display(xlib_drawable, dt);
+}
+
+
+static struct sw_displaytarget *
+xm_displaytarget_create(struct sw_winsys *winsys,
+                        enum pipe_format format,
+                        unsigned width, unsigned height,
+                        unsigned alignment,
+                        unsigned *stride)
+{
+   struct xm_displaytarget *xm_dt = CALLOC_STRUCT(xm_displaytarget);
+   unsigned nblocksy, size;
+
+   xm_dt = CALLOC_STRUCT(xm_displaytarget);
+   if(!xm_dt)
+      goto no_xm_dt;
+
+   xm_dt->display = ((struct xlib_sw_winsys *)winsys)->display;
+   xm_dt->format = format;
+   xm_dt->width = width;
+   xm_dt->height = height;
+
+   nblocksy = util_format_get_nblocksy(format, height);
+   xm_dt->stride = align(util_format_get_stride(format, width), alignment);
+   size = xm_dt->stride * nblocksy;
+
+   if (!debug_get_bool_option("XLIB_NO_SHM", FALSE))
+   {
+      xm_dt->shminfo.shmid = -1;
+      xm_dt->shminfo.shmaddr = (char *) -1;
+      xm_dt->shm = TRUE;
+         
+      xm_dt->data = alloc_shm(xm_dt, size);
+      if(!xm_dt->data)
+         goto no_data;
+   }
+
+   if(!xm_dt->data) {
+      xm_dt->data = align_malloc(size, alignment);
+      if(!xm_dt->data)
+         goto no_data;
+   }
+
+   *stride = xm_dt->stride;
+   return (struct sw_displaytarget *)xm_dt;
+
+no_data:
+   FREE(xm_dt);
+no_xm_dt:
+   return NULL;
+}
+
+
+static void
+xm_destroy( struct sw_winsys *ws )
+{
+   FREE(ws);
+}
+
+
+struct sw_winsys *
+xlib_create_sw_winsys( Display *display )
+{
+   struct xlib_sw_winsys *ws;
+
+   ws = CALLOC_STRUCT(xlib_sw_winsys);
+   if (!ws)
+      return NULL;
+
+   ws->display = display;
+   ws->base.destroy = xm_destroy;
+
+   ws->base.is_displaytarget_format_supported = xm_is_displaytarget_format_supported;
+
+   ws->base.displaytarget_create = xm_displaytarget_create;
+   ws->base.displaytarget_map = xm_displaytarget_map;
+   ws->base.displaytarget_unmap = xm_displaytarget_unmap;
+   ws->base.displaytarget_destroy = xm_displaytarget_destroy;
+
+   ws->base.displaytarget_display = xm_displaytarget_display;
+
+   return &ws->base;
+}
+
diff --git a/src/gallium/winsys/xlib/xmesa.h b/src/gallium/winsys/xlib/xmesa.h
deleted file mode 100644
index 98139af8336..00000000000
--- a/src/gallium/winsys/xlib/xmesa.h
+++ /dev/null
@@ -1,424 +0,0 @@
-/*
- * Mesa 3-D graphics library
- * Version:  7.1
- * 
- * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- * 
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
- * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
-
-
-/*
- * Mesa/X11 interface.  This header file serves as the documentation for
- * the Mesa/X11 interface functions.
- *
- * Note: this interface isn't intended for user programs.  It's primarily
- * just for implementing the pseudo-GLX interface.
- */
-
-
-/* Sample Usage:
-
-In addition to the usual X calls to select a visual, create a colormap
-and create a window, you must do the following to use the X/Mesa interface:
-
-1. Call XMesaCreateVisual() to make an XMesaVisual from an XVisualInfo.
-
-2. Call XMesaCreateContext() to create an X/Mesa rendering context, given
-   the XMesaVisual.
-
-3. Call XMesaCreateWindowBuffer() to create an XMesaBuffer from an X window
-   and XMesaVisual.
-
-4. Call XMesaMakeCurrent() to bind the XMesaBuffer to an XMesaContext and
-   to make the context the current one.
-
-5. Make gl* calls to render your graphics.
-
-6. Use XMesaSwapBuffers() when double buffering to swap front/back buffers.
-
-7. Before the X window is destroyed, call XMesaDestroyBuffer().
-
-8. Before exiting, call XMesaDestroyVisual and XMesaDestroyContext.
-
-*/
-
-
-
-
-#ifndef XMESA_H
-#define XMESA_H
-
-#ifdef __VMS
-#include <GL/vms_x_fix.h>
-#endif
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#ifdef XFree86Server
-#include "xmesa_xf86.h"
-#else
-#include <X11/Xlib.h>
-#include <X11/Xutil.h>
-#include "xmesa_x.h"
-#endif
-#include "GL/gl.h"
-
-#ifdef AMIWIN
-#include <pragmas/xlib_pragmas.h>
-extern struct Library *XLibBase;
-#endif
-
-
-#define XMESA_MAJOR_VERSION 6
-#define XMESA_MINOR_VERSION 3
-
-
-
-/*
- * Values passed to XMesaGetString:
- */
-#define XMESA_VERSION 1
-#define XMESA_EXTENSIONS 2
-
-
-/*
- * Values passed to XMesaSetFXmode:
- */
-#define XMESA_FX_WINDOW       1
-#define XMESA_FX_FULLSCREEN   2
-
-
-
-typedef struct xmesa_context *XMesaContext;
-
-typedef struct xmesa_visual *XMesaVisual;
-
-typedef struct xmesa_buffer *XMesaBuffer;
-
-
-
-/*
- * Create a new X/Mesa visual.
- * Input:  display - X11 display
- *         visinfo - an XVisualInfo pointer
- *         rgb_flag - GL_TRUE = RGB mode,
- *                    GL_FALSE = color index mode
- *         alpha_flag - alpha buffer requested?
- *         db_flag - GL_TRUE = double-buffered,
- *                   GL_FALSE = single buffered
- *         stereo_flag - stereo visual?
- *         ximage_flag - GL_TRUE = use an XImage for back buffer,
- *                       GL_FALSE = use an off-screen pixmap for back buffer
- *         depth_size - requested bits/depth values, or zero
- *         stencil_size - requested bits/stencil values, or zero
- *         accum_red_size - requested bits/red accum values, or zero
- *         accum_green_size - requested bits/green accum values, or zero
- *         accum_blue_size - requested bits/blue accum values, or zero
- *         accum_alpha_size - requested bits/alpha accum values, or zero
- *         num_samples - number of samples/pixel if multisampling, or zero
- *         level - visual level, usually 0
- *         visualCaveat - ala the GLX extension, usually GLX_NONE_EXT
- * Return;  a new XMesaVisual or 0 if error.
- */
-extern XMesaVisual XMesaCreateVisual( XMesaDisplay *display,
-                                      XMesaVisualInfo visinfo,
-                                      GLboolean rgb_flag,
-                                      GLboolean alpha_flag,
-                                      GLboolean db_flag,
-                                      GLboolean stereo_flag,
-                                      GLboolean ximage_flag,
-                                      GLint depth_size,
-                                      GLint stencil_size,
-                                      GLint accum_red_size,
-                                      GLint accum_green_size,
-                                      GLint accum_blue_size,
-                                      GLint accum_alpha_size,
-                                      GLint num_samples,
-                                      GLint level,
-                                      GLint visualCaveat );
-
-/*
- * Destroy an XMesaVisual, but not the associated XVisualInfo.
- */
-extern void XMesaDestroyVisual( XMesaVisual v );
-
-
-
-/*
- * Create a new XMesaContext for rendering into an X11 window.
- *
- * Input:  visual - an XMesaVisual
- *         share_list - another XMesaContext with which to share display
- *                      lists or NULL if no sharing is wanted.
- * Return:  an XMesaContext or NULL if error.
- */
-extern XMesaContext XMesaCreateContext( XMesaVisual v,
-					XMesaContext share_list );
-
-
-/*
- * Destroy a rendering context as returned by XMesaCreateContext()
- */
-extern void XMesaDestroyContext( XMesaContext c );
-
-
-#ifdef XFree86Server
-/*
- * These are the extra routines required for integration with XFree86.
- * None of these routines should be user visible. -KEM
- */
-extern GLboolean XMesaForceCurrent( XMesaContext c );
-
-extern GLboolean XMesaLoseCurrent( XMesaContext c );
-
-extern GLboolean XMesaCopyContext( XMesaContext src,
-				   XMesaContext dst,
-				   GLuint mask );
-#endif /* XFree86Server */
-
-
-/*
- * Create an XMesaBuffer from an X window.
- */
-extern XMesaBuffer XMesaCreateWindowBuffer( XMesaVisual v, XMesaWindow w );
-
-
-/*
- * Create an XMesaBuffer from an X pixmap.
- */
-extern XMesaBuffer XMesaCreatePixmapBuffer( XMesaVisual v,
-					    XMesaPixmap p,
-					    XMesaColormap cmap );
-
-
-/*
- * Destroy an XMesaBuffer, but not the corresponding window or pixmap.
- */
-extern void XMesaDestroyBuffer( XMesaBuffer b );
-
-
-/*
- * Return the XMesaBuffer handle which corresponds to an X drawable, if any.
- *
- * New in Mesa 2.3.
- */
-extern XMesaBuffer XMesaFindBuffer( XMesaDisplay *dpy,
-				    XMesaDrawable d );
-
-
-
-/*
- * Bind a buffer to a context and make the context the current one.
- */
-extern GLboolean XMesaMakeCurrent( XMesaContext c,
-				   XMesaBuffer b );
-
-
-/*
- * Bind two buffers (read and draw) to a context and make the
- * context the current one.
- * New in Mesa 3.3
- */
-extern GLboolean XMesaMakeCurrent2( XMesaContext c,
-                                    XMesaBuffer drawBuffer,
-                                    XMesaBuffer readBuffer );
-
-
-/*
- * Unbind the current context from its buffer.
- */
-extern GLboolean XMesaUnbindContext( XMesaContext c );
-
-
-/*
- * Return a handle to the current context.
- */
-extern XMesaContext XMesaGetCurrentContext( void );
-
-
-/*
- * Return handle to the current (draw) buffer.
- */
-extern XMesaBuffer XMesaGetCurrentBuffer( void );
-
-
-/*
- * Return handle to the current read buffer.
- * New in Mesa 3.3
- */
-extern XMesaBuffer XMesaGetCurrentReadBuffer( void );
-
-
-/*
- * Swap the front and back buffers for the given buffer.  No action is
- * taken if the buffer is not double buffered.
- */
-extern void XMesaSwapBuffers( XMesaBuffer b );
-
-
-/*
- * Copy a sub-region of the back buffer to the front buffer.
- *
- * New in Mesa 2.6
- */
-extern void XMesaCopySubBuffer( XMesaBuffer b,
-				int x,
-				int y,
-				int width,
-				int height );
-
-
-/*
- * Return a pointer to the the Pixmap or XImage being used as the back
- * color buffer of an XMesaBuffer.  This function is a way to get "under
- * the hood" of X/Mesa so one can manipulate the back buffer directly.
- * Input:  b - the XMesaBuffer
- * Output:  pixmap - pointer to back buffer's Pixmap, or 0
- *          ximage - pointer to back buffer's XImage, or NULL
- * Return:  GL_TRUE = context is double buffered
- *          GL_FALSE = context is single buffered
- */
-extern GLboolean XMesaGetBackBuffer( XMesaBuffer b,
-				     XMesaPixmap *pixmap,
-				     XMesaImage **ximage );
-
-
-
-/*
- * Return the depth buffer associated with an XMesaBuffer.
- * Input:  b - the XMesa buffer handle
- * Output:  width, height - size of buffer in pixels
- *          bytesPerValue - bytes per depth value (2 or 4)
- *          buffer - pointer to depth buffer values
- * Return:  GL_TRUE or GL_FALSE to indicate success or failure.
- *
- * New in Mesa 2.4.
- */
-extern GLboolean XMesaGetDepthBuffer( XMesaBuffer b,
-				      GLint *width,
-				      GLint *height,
-				      GLint *bytesPerValue,
-				      void **buffer );
-
-
-
-/*
- * Flush/sync a context
- */
-extern void XMesaFlush( XMesaContext c );
-
-
-
-/*
- * Get an X/Mesa-specific string.
- * Input:  name - either XMESA_VERSION or XMESA_EXTENSIONS
- */
-extern const char *XMesaGetString( XMesaContext c, int name );
-
-
-
-/*
- * Scan for XMesaBuffers whose window/pixmap has been destroyed, then free
- * any memory used by that buffer.
- *
- * New in Mesa 2.3.
- */
-extern void XMesaGarbageCollect( void );
-
-
-
-/*
- * Return a dithered pixel value.
- * Input:  c - XMesaContext
- *         x, y - window coordinate
- *         red, green, blue, alpha - color components in [0,1]
- * Return:  pixel value
- *
- * New in Mesa 2.3.
- */
-extern unsigned long XMesaDitherColor( XMesaContext xmesa,
-				       GLint x,
-				       GLint y,
-				       GLfloat red,
-				       GLfloat green,
-				       GLfloat blue,
-				       GLfloat alpha );
-
-
-
-/*
- * 3Dfx Glide driver only!
- * Set 3Dfx/Glide full-screen or window rendering mode.
- * Input:  mode - either XMESA_FX_WINDOW (window rendering mode) or
- *                XMESA_FX_FULLSCREEN (full-screen rendering mode)
- * Return:  GL_TRUE if success
- *          GL_FALSE if invalid mode or if not using 3Dfx driver
- *
- * New in Mesa 2.6.
- */
-extern GLboolean XMesaSetFXmode( GLint mode );
-
-
-
-/*
- * Reallocate the back/depth/stencil/accum/etc/ buffers associated with
- * buffer <b> if its size has changed.
- *
- * New in Mesa 4.0.2
- */
-extern void XMesaResizeBuffers( XMesaBuffer b );
-
-
-
-/*
- * Create a pbuffer.
- * New in Mesa 4.1
- */
-extern XMesaBuffer XMesaCreatePBuffer(XMesaVisual v, XMesaColormap cmap,
-                                      unsigned int width, unsigned int height);
-
-
-
-/*
- * Texture from Pixmap
- * New in Mesa 7.1
- */
-extern void
-XMesaBindTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer,
-                  const int *attrib_list);
-
-extern void
-XMesaReleaseTexImage(XMesaDisplay *dpy, XMesaBuffer drawable, int buffer);
-
-
-extern XMesaBuffer
-XMesaCreatePixmapTextureBuffer(XMesaVisual v, XMesaPixmap p,
-                               XMesaColormap cmap,
-                               int format, int target, int mipmap);
-
-
-
-#ifdef __cplusplus
-}
-#endif
-
-
-#endif
diff --git a/src/gallium/winsys/xlib/xmesa_x.h b/src/gallium/winsys/xlib/xmesa_x.h
deleted file mode 100644
index 865bab4313b..00000000000
--- a/src/gallium/winsys/xlib/xmesa_x.h
+++ /dev/null
@@ -1,86 +0,0 @@
-
-/**************************************************************************
-
-Copyright 1998-1999 Precision Insight, Inc., Cedar Park, Texas.
-All Rights Reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining a
-copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation 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 PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR
-ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
-TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
-SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
-**************************************************************************/
-
-/*
- * Authors:
- *   Kevin E. Martin <kevin@precisioninsight.com>
- *
- * When we're building the XMesa driver for stand-alone Mesa we
- * include this file when building the xm_*.c files.
- * We need to define some types and macros differently when building
- * in the Xserver vs. stand-alone Mesa.
- */
-
-#ifndef _XMESA_X_H_
-#define _XMESA_X_H_
-
-typedef Display      XMesaDisplay;
-typedef Pixmap       XMesaPixmap;
-typedef Colormap     XMesaColormap;
-typedef Drawable     XMesaDrawable;
-typedef Window       XMesaWindow;
-typedef GC           XMesaGC;
-typedef XVisualInfo *XMesaVisualInfo;
-typedef XImage       XMesaImage;
-typedef XPoint       XMesaPoint;
-typedef XColor       XMesaColor;
-
-#define XMesaDestroyImage      XDestroyImage
-
-#define XMesaPutPixel          XPutPixel
-#define XMesaGetPixel          XGetPixel
-
-#define XMesaSetForeground     XSetForeground
-#define XMesaSetBackground     XSetBackground
-#define XMesaSetPlaneMask      XSetPlaneMask
-#define XMesaSetFunction       XSetFunction
-#define XMesaSetFillStyle      XSetFillStyle
-#define XMesaSetTile           XSetTile
-
-#define XMesaDrawPoint         XDrawPoint
-#define XMesaDrawPoints        XDrawPoints
-#define XMesaDrawLine          XDrawLine
-#define XMesaFillRectangle     XFillRectangle
-#define XMesaGetImage          XGetImage
-#define XMesaPutImage          XPutImage
-#define XMesaCopyArea          XCopyArea
-
-#define XMesaCreatePixmap      XCreatePixmap
-#define XMesaFreePixmap        XFreePixmap
-#define XMesaFreeGC            XFreeGC
-
-#define GET_COLORMAP_SIZE(__v)  __v->visinfo->colormap_size
-#define GET_REDMASK(__v)        __v->mesa_visual.redMask
-#define GET_GREENMASK(__v)      __v->mesa_visual.greenMask
-#define GET_BLUEMASK(__v)       __v->mesa_visual.blueMask
-#define GET_VISUAL_DEPTH(__v)   __v->visinfo->depth
-#define GET_BLACK_PIXEL(__v)    BlackPixel(__v->display, __v->mesa_visual.screen)
-#define CHECK_BYTE_ORDER(__v)   host_byte_order()==ImageByteOrder(__v->display)
-#define CHECK_FOR_HPCR(__v)     XInternAtom(__v->display, "_HP_RGB_SMOOTH_MAP_LIST", True)
-
-#endif
diff --git a/src/mesa/SConscript b/src/mesa/SConscript
index 0a25dccde56..e80ec5ee880 100644
--- a/src/mesa/SConscript
+++ b/src/mesa/SConscript
@@ -249,6 +249,7 @@ if env['platform'] != 'winddk':
 	glapi_sources = [
 		'glapi/glapi.c',
 		'glapi/glapi_dispatch.c',
+		'glapi/glapi_entrypoint.c',
 		'glapi/glapi_getproc.c',
 		'glapi/glapi_nop.c',
 		'glapi/glthread.c',
diff --git a/src/mesa/drivers/dri/i810/Makefile b/src/mesa/drivers/dri/i810/Makefile
index 3874faee518..54a837d5ea9 100644
--- a/src/mesa/drivers/dri/i810/Makefile
+++ b/src/mesa/drivers/dri/i810/Makefile
@@ -5,9 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = i810_dri.so
 
-# Not yet
-# MINIGLX_SOURCES = server/i810_dri.c 
-
 DRIVER_SOURCES = \
 	i810context.c \
 	i810ioctl.c \
diff --git a/src/mesa/drivers/dri/i810/server/i810_dri.c b/src/mesa/drivers/dri/i810/server/i810_dri.c
deleted file mode 100644
index f52797c5ed2..00000000000
--- a/src/mesa/drivers/dri/i810/server/i810_dri.c
+++ /dev/null
@@ -1,975 +0,0 @@
-/**
- * \file server/i810_dri.c
- * \brief File to perform the device-specific initialization tasks typically
- * done in the X server.
- *
- * Here they are converted to run in the client (or perhaps a standalone
- * process), and to work with the frame buffer device rather than the X
- * server infrastructure.
- * 
- * Copyright (C) 2004 Dave Airlie (airlied@linux.ie)
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
-
-#include "driver.h"
-#include "drm.h"
-
-#include "i810.h"
-#include "i810_dri.h"
-#include "i810_reg.h"
-
-
-static int i810_pitches[] = {
-   512,
-   1024,
-   2048,
-   4096,
-   0
-};
-
-static int i810_pitch_flags[] = {
-   0x0,
-   0x1,
-   0x2,
-   0x3,
-   0
-};
-
-static unsigned int i810_drm_version = 0;
-
-static int
-I810AllocLow(I810MemRange * result, I810MemRange * pool, int size)
-{
-   if (size > pool->Size)
-      return 0;
-
-   pool->Size -= size;
-   result->Size = size;
-   result->Start = pool->Start;
-   result->End = pool->Start += size;
-
-   return 1;
-}
-
-static int
-I810AllocHigh(I810MemRange * result, I810MemRange * pool, int size)
-{
-   if (size > pool->Size)
-      return 0;
-
-   pool->Size -= size;
-   result->Size = size;
-   result->End = pool->End;
-   result->Start = pool->End -= size;
-
-   return 1;
-}
-
-
-/**
- * \brief Wait for free FIFO entries.
- *
- * \param ctx display handle.
- * \param entries number of free entries to wait.
- *
- * It polls the free entries from the chip until it reaches the requested value
- * or a timeout (3000 tries) occurs. Aborts the program if the FIFO times out.
- */
-static void I810WaitForFifo( const DRIDriverContext *ctx,
-			       int entries )
-{
-}
-
-/**
- * \brief Reset graphics card to known state.
- *
- * \param ctx display handle.
- *
- * Resets the values of several I810 registers.
- */
-static void I810EngineReset( const DRIDriverContext *ctx )
-{
-   unsigned char *I810MMIO = ctx->MMIOAddress;
-}
-
-/**
- * \brief Restore the drawing engine.
- *
- * \param ctx display handle
- *
- * Resets the graphics card and sets initial values for several registers of
- * the card's drawing engine.
- *
- * Turns on the i810 command processor engine (i.e., the ringbuffer).
- */
-static int I810EngineRestore( const DRIDriverContext *ctx )
-{
-   I810Ptr info = ctx->driverPrivate;
-   unsigned char *I810MMIO = ctx->MMIOAddress;
-
-   fprintf(stderr, "%s\n", __FUNCTION__);
-
-   return 1;
-}
-
-
-/**
- * \brief Shutdown the drawing engine.
- *
- * \param ctx display handle
- *
- * Turns off the command processor engine & restores the graphics card
- * to a state that fbdev understands.
- */
-static int I810EngineShutdown( const DRIDriverContext *ctx )
-{
-  drmI810Init info;
-  int ret;
-
-  memset(&info, 0, sizeof(drmI810Init));
-  info.func = I810_CLEANUP_DMA;
-  
-  ret = drmCommandWrite(ctx->drmFD, DRM_I810_INIT, &info, sizeof(drmI810Init));
-  if (ret>0)
-  {
-    fprintf(stderr,"[dri] I810 DMA Cleanup failed\n");
-    return -errno;
-  }
-  return 0;
-}
-
-/**
- * \brief Compute base 2 logarithm.
- *
- * \param val value.
- * 
- * \return base 2 logarithm of \p val.
- */
-static int I810MinBits(int val)
-{
-   int  bits;
-
-   if (!val) return 1;
-   for (bits = 0; val; val >>= 1, ++bits);
-   return bits;
-}
-
-static int I810DRIAgpPreInit( const DRIDriverContext *ctx, I810Ptr info)
-{
-
-  if (drmAgpAcquire(ctx->drmFD) < 0) {
-    fprintf(stderr, "[gart] AGP not available\n");
-    return 0;
-  }
-  
-  
-  if (drmAgpEnable(ctx->drmFD, 0) < 0) {
-    fprintf(stderr, "[gart] AGP not enabled\n");
-    drmAgpRelease(ctx->drmFD);
-    return 0;
-  }
-}
-
-/**
- * \brief Initialize the AGP state
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return one on success, or zero on failure.
- * 
- * Acquires and enables the AGP device. Reserves memory in the AGP space for
- * the ring buffer, vertex buffers and textures. Initialize the I810
- * registers to point to that memory and add client mappings.
- */
-static int I810DRIAgpInit( const DRIDriverContext *ctx, I810Ptr info)
-{
-   unsigned char *I810MMIO = ctx->MMIOAddress;
-   int            ret;
-   int            s, l;
-   unsigned long dcacheHandle;
-   unsigned long agpHandle;
-   int pitch_idx = 0;
-   int back_size = 0;
-   int sysmem_size = 0;
-   int width = ctx->shared.virtualWidth * ctx->cpp;
-
-
-   info->backHandle = DRM_AGP_NO_HANDLE;
-   info->zHandle = DRM_AGP_NO_HANDLE;
-   info->sysmemHandle = DRM_AGP_NO_HANDLE;
-   info->dcacheHandle = DRM_AGP_NO_HANDLE;
-
-   memset(&info->DcacheMem, 0, sizeof(I810MemRange));
-   memset(&info->BackBuffer, 0, sizeof(I810MemRange));
-   memset(&info->DepthBuffer, 0, sizeof(I810MemRange));
-   
-   drmAgpAlloc(ctx->drmFD, 4096 * 1024, 1, NULL, &dcacheHandle);
-   info->dcacheHandle = dcacheHandle;
-   
-   fprintf(stderr, "[agp] dcacheHandle : 0x%x\n", dcacheHandle);
-
-#define Elements(x) sizeof(x)/sizeof(*x)
-   for (pitch_idx = 0; pitch_idx < Elements(i810_pitches); pitch_idx++)
-     if (width <= i810_pitches[pitch_idx])
-       break;
-   
-   if (pitch_idx == Elements(i810_pitches)) {
-     fprintf(stderr,"[dri] Couldn't find depth/back buffer pitch\n");
-     exit(-1);
-   }
-   else
-   {
-     int lines = (ctx->shared.virtualWidth + 15) / 16 * 16;
-     back_size = i810_pitches[pitch_idx] * lines;
-     back_size = ((back_size + 4096 - 1) / 4096) * 4096;
-   }
-
-   sysmem_size = ctx->shared.fbSize;
-   fprintf(stderr,"sysmem_size is %lu back_size is %lu\n", sysmem_size, back_size);
-   if (dcacheHandle != DRM_AGP_NO_HANDLE) {
-     if (back_size > 4 * 1024 * 1024) {
-       fprintf(stderr,"[dri] Backsize is larger then 4 meg\n");
-       sysmem_size = sysmem_size - 2 * back_size;
-       drmAgpFree(ctx->drmFD, dcacheHandle);
-       info->dcacheHandle = dcacheHandle = DRM_AGP_NO_HANDLE;
-     } else {
-       sysmem_size = sysmem_size - back_size;
-     }
-   } else {
-     sysmem_size = sysmem_size - 2 * back_size;
-   }
-   
-   info->SysMem.Start=0;
-   info->SysMem.Size = sysmem_size;
-   info->SysMem.End = sysmem_size;
-   
-   if (dcacheHandle != DRM_AGP_NO_HANDLE) {
-      if (drmAgpBind(ctx->drmFD, dcacheHandle, info->DepthOffset) == 0) {
-	memset(&info->DcacheMem, 0, sizeof(I810MemRange));
-	fprintf(stderr,"[agp] GART: Found 4096K Z buffer memory\n");
-	info->DcacheMem.Start = info->DepthOffset;
-	 info->DcacheMem.Size = 1024 * 4096;
-	 info->DcacheMem.End =  info->DcacheMem.Start + info->DcacheMem.Size;
-      } else {
-	fprintf(stderr, "[agp] GART: dcache bind failed\n");
-	drmAgpFree(ctx->drmFD, dcacheHandle);
-	info->dcacheHandle = dcacheHandle = DRM_AGP_NO_HANDLE;
-      }
-   } else {
-     fprintf(stderr, "[agp] GART: no dcache memory found\n");
-   }
-   
-   drmAgpAlloc(ctx->drmFD, back_size, 0, NULL, &agpHandle);
-   info->backHandle = agpHandle;
-
-   if (agpHandle != DRM_AGP_NO_HANDLE) {
-      if (drmAgpBind(ctx->drmFD, agpHandle, info->BackOffset) == 0) {
-	fprintf(stderr, "[agp] Bound backbuffer memory\n");
-
-	info->BackBuffer.Start = info->BackOffset;
-	info->BackBuffer.Size = back_size;
-	info->BackBuffer.End = (info->BackBuffer.Start +
-				 info->BackBuffer.Size);
-      } else {
-	fprintf(stderr,"[agp] Unable to bind backbuffer.  Disabling DRI.\n");
-	return 0;
-      }
-   } else {
-     fprintf(stderr, "[dri] Unable to allocate backbuffer memory.  Disabling DRI.\n");
-     return 0;
-   }
-
-   if (dcacheHandle == DRM_AGP_NO_HANDLE) {
-     drmAgpAlloc(ctx->drmFD, back_size, 0, NULL, &agpHandle);
-
-     info->zHandle = agpHandle;
-
-     if (agpHandle != DRM_AGP_NO_HANDLE) {
-       if (drmAgpBind(ctx->drmFD, agpHandle, info->DepthOffset) == 0) {
-	 fprintf(stderr,"[agp] Bound depthbuffer memory\n");
-	 info->DepthBuffer.Start = info->DepthOffset;
-	 info->DepthBuffer.Size = back_size;
-	 info->DepthBuffer.End = (info->DepthBuffer.Start +
-				   info->DepthBuffer.Size);
-       } else {
-	 fprintf(stderr,"[agp] Unable to bind depthbuffer.  Disabling DRI.\n");
-	 return 0;
-       }
-     } else {
-       fprintf(stderr,"[agp] Unable to allocate depthbuffer memory.  Disabling DRI.\n");
-       return 0;
-     }
-   }
-
-   /* Now allocate and bind the agp space.  This memory will include the
-    * regular framebuffer as well as texture memory.
-    */
-   drmAgpAlloc(ctx->drmFD, sysmem_size, 0, NULL, &agpHandle);
-   info->sysmemHandle = agpHandle;
-   
-   if (agpHandle != DRM_AGP_NO_HANDLE) {
-     if (drmAgpBind(ctx->drmFD, agpHandle, 0) == 0) {
-       fprintf(stderr, "[agp] Bound System Texture Memory\n");
-     } else {
-       fprintf(stderr, "[agp] Unable to bind system texture memory. Disabling DRI.\n");
-       return 0;
-     }
-   } else {
-     fprintf(stderr, "[agp] Unable to allocate system texture memory. Disabling DRI.\n");
-     return 0;
-   }
-   
-   info->auxPitch = i810_pitches[pitch_idx];
-   info->auxPitchBits = i810_pitch_flags[pitch_idx];
-   
-   return 1;
-}
-
-
-/**
- * \brief Initialize the kernel data structures and enable the CP engine.
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return non-zero on success, or zero on failure.
- *
- * This function is a wrapper around the DRM_I810_CP_INIT command, passing
- * all the parameters in a drmI810Init structure.
- */
-static int I810DRIKernelInit( const DRIDriverContext *ctx,
-			       I810Ptr info)
-{
-   int cpp = ctx->bpp / 8;
-   drmI810Init  drmInfo;
-   int ret;
-   I810RingBuffer *ring = &(info->LpRing);
-
-   /* This is the struct passed to the kernel module for its initialization */
-   memset(&drmInfo, 0, sizeof(drmI810Init));
-   
-   /* make sure we have at least 1.4 */
-   drmInfo.func             = I810_INIT_DMA_1_4;
-
-   drmInfo.ring_start = ring->mem.Start;
-   drmInfo.ring_end = ring->mem.End;
-   drmInfo.ring_size = ring->mem.Size;
-
-   drmInfo.mmio_offset         = (unsigned int)info->regs;
-   drmInfo.buffers_offset      = (unsigned int)info->buffer_map;
-   drmInfo.sarea_priv_offset   = sizeof(drm_sarea_t);
-
-   drmInfo.front_offset        = 0;
-   drmInfo.back_offset         = info->BackBuffer.Start;
-   drmInfo.depth_offset        = info->DepthBuffer.Start;
-
-   drmInfo.w                   = ctx->shared.virtualWidth;
-   drmInfo.h                   = ctx->shared.virtualHeight;
-   drmInfo.pitch               = info->auxPitch;
-   drmInfo.pitch_bits          = info->auxPitchBits;
-   
-
-   ret = drmCommandWrite(ctx->drmFD, DRM_I810_INIT, &drmInfo, 
-			 sizeof(drmI810Init));
-
-   return ret >= 0;
-}
-
-
-/**
- * \brief Add a map for the vertex buffers that will be accessed by any
- * DRI-based clients.
- * 
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return one on success, or zero on failure.
- *
- * Calls drmAddBufs() with the previously allocated vertex buffers.
- */
-static int I810DRIBufInit( const DRIDriverContext *ctx, I810Ptr info )
-{
-   /* Initialize vertex buffers */
-   info->bufNumBufs = drmAddBufs(ctx->drmFD,
-				 I810_DMA_BUF_NR,
-				 I810_DMA_BUF_SZ,
-				 DRM_AGP_BUFFER,
-				 info->BufferMem.Start);
-
-   if (info->bufNumBufs <= 0) {
-      fprintf(stderr,
-	      "[drm] Could not create vertex/indirect buffers list\n");
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] Added %d %d byte vertex/indirect buffers\n",
-	   info->bufNumBufs, I810_DMA_BUF_SZ);
-   
-   return 1;
-}
-
-/**
- * \brief Install an IRQ handler.
- * 
- * \param ctx display handle.
- * \param info driver private data.
- *
- * Attempts to install an IRQ handler via drmCtlInstHandler(), falling back to
- * IRQ-free operation on failure.
- */
-static void I810DRIIrqInit(const DRIDriverContext *ctx,
-			     I810Ptr info)
-{
-   if (!info->irq) {
-      info->irq = drmGetInterruptFromBusID(ctx->drmFD,
-					   ctx->pciBus,
-					   ctx->pciDevice,
-					   ctx->pciFunc);
-
-      if ((drmCtlInstHandler(ctx->drmFD, info->irq)) != 0) {
-	 fprintf(stderr,
-		 "[drm] failure adding irq handler, "
-		 "there is a device already using that irq\n"
-		 "[drm] falling back to irq-free operation\n");
-	 info->irq = 0;
-      }
-   }
-
-   if (info->irq)
-      fprintf(stderr,
-	      "[drm] dma control initialized, using IRQ %d\n",
-	      info->irq);
-}
-
-static int I810CheckDRMVersion( const DRIDriverContext *ctx,
-				  I810Ptr info )
-{
-   drmVersionPtr  version;
-
-   version = drmGetVersion(ctx->drmFD);
-   if (version) {
-      int req_minor, req_patch;
-
-      req_minor = 4;
-      req_patch = 0;	
-
-      i810_drm_version = (version->version_major<<16) | version->version_minor;
-      if (version->version_major != 1 ||
-	  version->version_minor < req_minor ||
-	  (version->version_minor == req_minor && 
-	   version->version_patchlevel < req_patch)) {
-	 /* Incompatible drm version */
-	 fprintf(stderr,
-		 "[dri] I810DRIScreenInit failed because of a version "
-		 "mismatch.\n"
-		 "[dri] i810.o kernel module version is %d.%d.%d "
-		 "but version 1.%d.%d or newer is needed.\n"
-		 "[dri] Disabling DRI.\n",
-		 version->version_major,
-		 version->version_minor,
-		 version->version_patchlevel,
-		 req_minor,
-		 req_patch);
-	 drmFreeVersion(version);
-	 return 0;
-      }
-
-      info->drmMinor = version->version_minor;
-      drmFreeVersion(version);
-   }
-
-   return 1;
-}
-
-static int I810MemoryInit( const DRIDriverContext *ctx, I810Ptr info )
-{
-   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
-   int        cpp         = ctx->cpp;
-   int        bufferSize  = (ctx->shared.virtualHeight * width_bytes);
-   int        depthSize   = (((ctx->shared.virtualHeight+15) & ~15) * width_bytes);
-   int        l;
-
-   if (drmAddMap(ctx->drmFD, (drm_handle_t) info->BackBuffer.Start,
-		 info->BackBuffer.Size, DRM_AGP, 0,
-		 &info->backbuffer) < 0) {
-     fprintf(stderr, "[drm] drmAddMap(backbuffer) failed.  Disabling DRI\n");
-     return 0;
-   }
-   
-   if (drmAddMap(ctx->drmFD, (drm_handle_t) info->DepthBuffer.Start,
-		 info->DepthBuffer.Size, DRM_AGP, 0,
-		 &info->depthbuffer) < 0) {
-     fprintf(stderr, "[drm] drmAddMap(depthbuffer) failed.  Disabling DRI.\n");
-      return 0;
-   }
-
-   if (!I810AllocLow(&(info->FrontBuffer), &(info->SysMem), (((ctx->shared.virtualHeight * width_bytes) + 4095) & ~4095)))
-   {
-     fprintf(stderr,"Framebuffer allocation failed\n");
-     return 0;
-   }
-   else
-     fprintf(stderr,"Frame buffer at 0x%.8x (%luk, %lu bytes)\n",
-	     info->FrontBuffer.Start,
-	     info->FrontBuffer.Size / 1024, info->FrontBuffer.Size);
-   
-   memset(&(info->LpRing), 0, sizeof(I810RingBuffer));
-   if (I810AllocLow(&(info->LpRing.mem), &(info->SysMem), 16 * 4096)) {
-     fprintf(stderr,
-	    "Ring buffer at 0x%.8x (%luk, %lu bytes)\n",
-	     info->LpRing.mem.Start,
-	     info->LpRing.mem.Size / 1024, info->LpRing.mem.Size);
-     
-     info->LpRing.tail_mask = info->LpRing.mem.Size - 1;
-     info->LpRing.virtual_start = info->LpRing.mem.Start;
-     info->LpRing.head = 0;
-     info->LpRing.tail = 0;
-     info->LpRing.space = 0;
-   } else {
-     fprintf(stderr, "Ring buffer allocation failed\n");
-     return (0);
-   }
-
-   /* Allocate buffer memory */
-   I810AllocHigh(&(info->BufferMem), &(info->SysMem),
-		 I810_DMA_BUF_NR * I810_DMA_BUF_SZ);
-   
-
-   fprintf(stderr, "[dri] Buffer map : %lx\n",
-	      info->BufferMem.Start);
-
-   if (info->BufferMem.Start == 0 ||
-       info->BufferMem.End - info->BufferMem.Start >
-       I810_DMA_BUF_NR * I810_DMA_BUF_SZ) {
-     fprintf(stderr,"[dri] Not enough memory for dma buffers.  Disabling DRI.\n");
-     return 0;
-   }
-
-   if (drmAddMap(ctx->drmFD, (drm_handle_t) info->BufferMem.Start,
-		 info->BufferMem.Size, DRM_AGP, 0, &info->buffer_map) < 0) {
-     fprintf(stderr, "[drm] drmAddMap(buffer_map) failed.  Disabling DRI.\n");
-     return 0;
-   }
-
-   if (drmAddMap(ctx->drmFD, (drm_handle_t) info->LpRing.mem.Start,
-		 info->LpRing.mem.Size, DRM_AGP, 0, &info->ring_map) < 0) {
-     fprintf(stderr, "[drm] drmAddMap(ring_map) failed.  Disabling DRI. \n");
-     return 0;
-   }
-
-   /* Front, back and depth buffers - everything else texture??
-    */
-   info->textureSize = info->SysMem.Size;
-
-   if (info->textureSize < 0) 
-      return 0;
-
-   
-   l = I810MinBits((info->textureSize-1) / I810_NR_TEX_REGIONS);
-   if (l < I810_LOG_MIN_TEX_REGION_SIZE) l = I810_LOG_MIN_TEX_REGION_SIZE;
-
-   /* Round the texture size up to the nearest whole number of
-    * texture regions.  Again, be greedy about this, don't
-    * round down.
-    */
-   info->logTextureGranularity = l;
-   info->textureSize = (info->textureSize >> l) << l;
-
-   /* Set a minimum usable local texture heap size.  This will fit
-    * two 256x256x32bpp textures.
-    */
-   if (info->textureSize < 512 * 1024) {
-      info->textureOffset = 0;
-      info->textureSize = 0;
-   }
-
-   I810AllocLow(&(info->TexMem), &(info->SysMem), info->textureSize);
-
-   if (drmAddMap(ctx->drmFD, (drm_handle_t) info->TexMem.Start,
-		 info->TexMem.Size, DRM_AGP, 0, &info->textures) < 0) {
-     fprintf(stderr,
-		 "[drm] drmAddMap(textures) failed.  Disabling DRI.\n");
-      return 0;
-   }
-
-   /* Reserve space for textures */
-   fprintf(stderr, 
-	   "Will use back buffer at offset 0x%x\n",
-	   info->BackOffset);
-   fprintf(stderr, 
-	   "Will use depth buffer at offset 0x%x\n",
-	   info->DepthOffset);
-   fprintf(stderr, 
-	   "Will use %d kb for textures at offset 0x%x\n",
-	   info->TexMem.Size/1024, info->TexMem.Start);
-
-   return 1;
-} 
-
-
-
-/**
- * Called at the start of each server generation.
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return non-zero on success, or zero on failure.
- *
- * Performs static frame buffer allocation. Opens the DRM device and add maps
- * to the SAREA, framebuffer and MMIO regions. Fills in \p info with more
- * information. Creates a \e server context to grab the lock for the
- * initialization ioctls and calls the other initilization functions in this
- * file. Starts the CP engine via the DRM_I810_CP_START command.
- *
- * Setups a I810DRIRec structure to be passed to i810_dri.so for its
- * initialization.
- */
-static int I810ScreenInit( DRIDriverContext *ctx, I810Ptr info )
-{
-   I810DRIPtr   pI810DRI;
-   int err;
-
-   usleep(100);
-   /*assert(!ctx->IsClient);*/
-
-   /* from XFree86 driver */
-   info->DepthOffset = 0x3000000;
-   info->BackOffset = 0x3800000;
-   {
-      int  width_bytes = (ctx->shared.virtualWidth * ctx->cpp);
-      int  maxy        = ctx->shared.fbSize / width_bytes;
-
-
-      if (maxy <= ctx->shared.virtualHeight * 3) {
-	 fprintf(stderr, 
-		 "Static buffer allocation failed -- "
-		 "need at least %d kB video memory (have %d kB)\n",
-		 (ctx->shared.virtualWidth * ctx->shared.virtualHeight *
-		  ctx->cpp * 3 + 1023) / 1024,
-		 ctx->shared.fbSize / 1024);
-	 return 0;
-      } 
-   }
-
-
-   info->regsSize = ctx->MMIOSize;
-   ctx->shared.SAREASize = 0x2000;
-
-   /* Note that drmOpen will try to load the kernel module, if needed. */
-   ctx->drmFD = drmOpen("i810", NULL );
-   if (ctx->drmFD < 0) {
-      fprintf(stderr, "[drm] drmOpen failed\n");
-      return 0;
-   }
-
-   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-	      ctx->drmFD, ctx->pciBusID, strerror(-err));
-      return 0;
-   }
-
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   }
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-
-   if (drmAddMap(ctx->drmFD, 
-		 ctx->MMIOStart,
-		 ctx->MMIOSize,
-		 DRM_REGISTERS, 
-		 DRM_READ_ONLY, 
-		 &info->regs) < 0) {
-      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] register handle = 0x%08x\n", info->regs);
-
-   I810DRIAgpPreInit(ctx, info);
-   /* Need to AddMap the framebuffer and mmio regions here:
-    */
-   if (drmAddMap( ctx->drmFD,
-		  (drm_handle_t)ctx->FBStart,
-		  ctx->FBSize,
-		  DRM_FRAME_BUFFER,
-#ifndef _EMBEDDED
-		  0,
-#else
-		  DRM_READ_ONLY,
-#endif
-		  &ctx->shared.hFrameBuffer) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-      return 0;
-   }
-
-   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-	   ctx->shared.hFrameBuffer);
-
-   /* Check the i810 DRM version */
-   if (!I810CheckDRMVersion(ctx, info)) {
-      return 0;
-   }
-
-   /* Initialize AGP */
-   if (!I810DRIAgpInit(ctx, info)) {
-      return 0;
-   }
-
-
-   /* Memory manager setup */
-   if (!I810MemoryInit(ctx, info)) {
-      return 0;
-   }
-
-   /* Initialize the SAREA private data structure */
-   {
-      I810SAREAPtr pSAREAPriv;
-      pSAREAPriv = (I810SAREAPtr)(((char*)ctx->pSAREA) + 
-					sizeof(drm_sarea_t));
-      memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
-      //      pSAREAPriv->pf_enabled=1;
-   }
-
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-   /* Initialize the vertex buffers list */
-   if (!I810DRIBufInit(ctx, info)) {
-      fprintf(stderr, "I810DRIBufInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   /* Initialize the kernel data structures */
-   if (!I810DRIKernelInit(ctx, info)) {
-      fprintf(stderr, "I810DRIKernelInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   /* Initialize IRQ */
-   I810DRIIrqInit(ctx, info);
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-#if 0
-   memset((char *)ctx->FBAddress,
-	  0,
-	  info->auxPitch * ctx->cpp * ctx->shared.virtualHeight );
-
-   memset((char *)info->backbuffer,
-	  0,
-	  info->auxPitch * ctx->cpp * ctx->shared.virtualHeight );
-#endif
-
-   /* This is the struct passed to i810_dri.so for its initialization */
-   ctx->driverClientMsg = malloc(sizeof(I810DRIRec));
-   ctx->driverClientMsgSize = sizeof(I810DRIRec);
-   pI810DRI                    = (I810DRIPtr)ctx->driverClientMsg;
-
-   pI810DRI->regs              = info->regs;
-   pI810DRI->regsSize          = info->regsSize;
-   // regsMap is unused
-
-   pI810DRI->backbufferSize    = info->BackBuffer.Size;
-   pI810DRI->backbuffer        = info->backbuffer;
-
-   pI810DRI->depthbufferSize   = info->DepthBuffer.Size;
-   pI810DRI->depthbuffer       = info->depthbuffer;
-
-   pI810DRI->textures          = info->textures;
-   pI810DRI->textureSize       = info->textureSize;
-
-   pI810DRI->agp_buffers       = info->buffer_map;
-   pI810DRI->agp_buf_size      = info->BufferMem.Size;
-
-   pI810DRI->deviceID          = info->Chipset;
-   pI810DRI->width             = ctx->shared.virtualWidth;
-   pI810DRI->height            = ctx->shared.virtualHeight;
-   pI810DRI->mem               = ctx->shared.fbSize;
-   pI810DRI->cpp               = ctx->bpp / 8;
-   pI810DRI->bitsPerPixel      = ctx->bpp;
-   pI810DRI->fbOffset          = info->FrontBuffer.Start;
-   pI810DRI->fbStride          = info->auxPitch;
-   
-   pI810DRI->backOffset        = info->BackBuffer.Start;
-   pI810DRI->depthOffset       = info->DepthBuffer.Start;
-
-   pI810DRI->auxPitch          = info->auxPitch;
-   pI810DRI->auxPitchBits      = info->auxPitchBits;
-
-   pI810DRI->logTextureGranularity = info->logTextureGranularity;
-   pI810DRI->textureOffset     = info->TexMem.Start;
-  
-   pI810DRI->ringOffset        = info->LpRing.mem.Start;
-   pI810DRI->ringSize          = info->LpRing.mem.Size;
-
-   // drmBufs looks unused 
-   pI810DRI->irq               = info->irq;
-   pI810DRI->sarea_priv_offset = sizeof(drm_sarea_t);
-
-   /* Don't release the lock now - let the VT switch handler do it. */
-   return 1;
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa i810ValidateMode().
- */
-static int i810ValidateMode( const DRIDriverContext *ctx )
-{
-   unsigned char *I810MMIO = ctx->MMIOAddress;
-   I810Ptr info = ctx->driverPrivate;
-
-   return 1;
-}
-
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa i810ValidateMode().
- */
-static int i810PostValidateMode( const DRIDriverContext *ctx )
-{
-   unsigned char *I810MMIO = ctx->MMIOAddress;
-   I810Ptr info = ctx->driverPrivate;
-
-   return 1;
-}
-
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Fills in \p info with some default values and some information from \p ctx
- * and then calls I810ScreenInit() for the screen initialization.
- * 
- * Before exiting clears the framebuffer memory accessing it directly.
- */
-static int i810InitFBDev( DRIDriverContext *ctx )
-{
-  I810Ptr info = calloc(1, sizeof(*info));
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      switch (ctx->bpp / 8) {
-      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-      case 3:
-      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-      }
-
-      ctx->shared.virtualWidth = dummy;
-   }
-
-   ctx->driverPrivate = (void *)info;
-   
-   info->Chipset = ctx->chipset;
-
-   if (!I810ScreenInit( ctx, info ))
-      return 0;
-
-
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void i810HaltFBDev( DRIDriverContext *ctx )
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-       free(ctx->driverPrivate);
-       ctx->driverPrivate = 0;
-    }
-}
-
-
-extern void i810NotifyFocus( int );
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-const struct DRIDriverRec __driDriver = {
-   i810ValidateMode,
-   i810PostValidateMode,
-   i810InitFBDev,
-   i810HaltFBDev,
-   I810EngineShutdown,
-   I810EngineRestore,  
-#ifndef _EMBEDDED
-   0,
-#else
-   i810NotifyFocus, 
-#endif
-};
diff --git a/src/mesa/drivers/dri/i915/Makefile b/src/mesa/drivers/dri/i915/Makefile
index dc15ae425c9..5b49d0c77c6 100644
--- a/src/mesa/drivers/dri/i915/Makefile
+++ b/src/mesa/drivers/dri/i915/Makefile
@@ -4,8 +4,6 @@ include $(TOP)/configs/current
 
 LIBNAME = i915_dri.so
 
-MINIGLX_SOURCES = server/intel_dri.c
-
 DRIVER_SOURCES = \
 	i830_context.c \
 	i830_state.c \
diff --git a/src/mesa/drivers/dri/i915/server/intel_dri.c b/src/mesa/drivers/dri/i915/server/intel_dri.c
deleted file mode 120000
index effdd26448a..00000000000
--- a/src/mesa/drivers/dri/i915/server/intel_dri.c
+++ /dev/null
@@ -1 +0,0 @@
-../../intel/server/intel_dri.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile
index f98a1a27db2..a242580273f 100644
--- a/src/mesa/drivers/dri/i965/Makefile
+++ b/src/mesa/drivers/dri/i965/Makefile
@@ -99,7 +99,6 @@ DRIVER_SOURCES = \
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \
-	$(MINIGLX_SOURCES) \
 	$(DRIVER_SOURCES)
 
 ASM_SOURCES = 
diff --git a/src/mesa/drivers/dri/i965/brw_disasm.c b/src/mesa/drivers/dri/i965/brw_disasm.c
index a8f6b993ac3..54699cf8d34 100644
--- a/src/mesa/drivers/dri/i965/brw_disasm.c
+++ b/src/mesa/drivers/dri/i965/brw_disasm.c
@@ -74,9 +74,9 @@ struct {
     [BRW_OPCODE_JMPI] = { .name = "jmpi", .nsrc = 1, .ndst = 0 },
     [BRW_OPCODE_IF] = { .name = "if", .nsrc = 2, .ndst = 0 },
     [BRW_OPCODE_IFF] = { .name = "iff", .nsrc = 1, .ndst = 01 },
-    [BRW_OPCODE_WHILE] = { .name = "while", .nsrc = 1, .ndst = 0 },
+    [BRW_OPCODE_WHILE] = { .name = "while", .nsrc = 2, .ndst = 0 },
     [BRW_OPCODE_ELSE] = { .name = "else", .nsrc = 2, .ndst = 0 },
-    [BRW_OPCODE_BREAK] = { .name = "break", .nsrc = 1, .ndst = 0 },
+    [BRW_OPCODE_BREAK] = { .name = "break", .nsrc = 2, .ndst = 0 },
     [BRW_OPCODE_CONTINUE] = { .name = "cont", .nsrc = 1, .ndst = 0 },
     [BRW_OPCODE_HALT] = { .name = "halt", .nsrc = 1, .ndst = 0 },
     [BRW_OPCODE_MSAVE] = { .name = "msave", .nsrc = 1, .ndst = 1 },
diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c
index c78f7b38aee..1fd957b3ad6 100644
--- a/src/mesa/drivers/dri/i965/brw_program.c
+++ b/src/mesa/drivers/dri/i965/brw_program.c
@@ -95,9 +95,17 @@ static void brwDeleteProgram( GLcontext *ctx,
 			      struct gl_program *prog )
 {
    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
-      struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog;
-      struct brw_fragment_program *brw_fprog = brw_fragment_program(fprog);
-      dri_bo_unreference(brw_fprog->const_buffer);
+      struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
+      struct brw_fragment_program *brw_fp = brw_fragment_program(fp);
+
+      dri_bo_unreference(brw_fp->const_buffer);
+   }
+
+   if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
+      struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
+      struct brw_vertex_program *brw_vp = brw_vertex_program(vp);
+
+      dri_bo_unreference(brw_vp->const_buffer);
    }
 
    _mesa_delete_program( ctx, prog );
diff --git a/src/mesa/drivers/dri/i965/brw_sf.c b/src/mesa/drivers/dri/i965/brw_sf.c
index 8e6839b8120..57d1c29ade1 100644
--- a/src/mesa/drivers/dri/i965/brw_sf.c
+++ b/src/mesa/drivers/dri/i965/brw_sf.c
@@ -46,7 +46,6 @@
 static void compile_sf_prog( struct brw_context *brw,
 			     struct brw_sf_prog_key *key )
 {
-   GLcontext *ctx = &brw->intel.ctx;
    struct brw_sf_compile c;
    const GLuint *program;
    GLuint program_size;
@@ -69,20 +68,14 @@ static void compile_sf_prog( struct brw_context *brw,
 
    /* Construct map from attribute number to position in the vertex.
     */
-   for (i = idx = 0; i < VERT_RESULT_MAX; i++) 
+   for (i = idx = 0; i < VERT_RESULT_MAX; i++) {
       if (c.key.attrs & BITFIELD64_BIT(i)) {
 	 c.attr_to_idx[i] = idx;
 	 c.idx_to_attr[idx] = i;
-	 if (i >= VERT_RESULT_TEX0 && i <= VERT_RESULT_TEX7) {
-            c.point_attrs[i].CoordReplace = 
-               ctx->Point.CoordReplace[i - VERT_RESULT_TEX0];
-	 }
-         else {
-            c.point_attrs[i].CoordReplace = GL_FALSE;
-         }
 	 idx++;
       }
-   
+   }
+
    /* Which primitive?  Or all three? 
     */
    switch (key->primitive) {
@@ -162,6 +155,14 @@ static void upload_sf_prog(struct brw_context *brw)
    }
 
    key.do_point_sprite = ctx->Point.PointSprite;
+   if (key.do_point_sprite) {
+      int i;
+
+      for (i = 0; i < 8; i++) {
+	 if (ctx->Point.CoordReplace[i])
+	    key.point_sprite_coord_replace |= (1 << i);
+      }
+   }
    key.sprite_origin_lower_left = (ctx->Point.SpriteOrigin == GL_LOWER_LEFT);
    /* _NEW_LIGHT */
    key.do_flat_shading = (ctx->Light.ShadeModel == GL_FLAT);
diff --git a/src/mesa/drivers/dri/i965/brw_sf.h b/src/mesa/drivers/dri/i965/brw_sf.h
index 0ba731fac99..a0680a56f2c 100644
--- a/src/mesa/drivers/dri/i965/brw_sf.h
+++ b/src/mesa/drivers/dri/i965/brw_sf.h
@@ -46,6 +46,7 @@
 
 struct brw_sf_prog_key {
    GLbitfield64 attrs;
+   uint8_t point_sprite_coord_replace;
    GLuint primitive:2;
    GLuint do_twoside_color:1;
    GLuint do_flat_shading:1;
@@ -56,10 +57,6 @@ struct brw_sf_prog_key {
    GLuint pad:24;
 };
 
-struct brw_sf_point_tex {
-   GLboolean CoordReplace;	
-};
-
 struct brw_sf_compile {
    struct brw_compile func;
    struct brw_sf_prog_key key;
@@ -100,7 +97,6 @@ struct brw_sf_compile {
 
    GLubyte attr_to_idx[VERT_RESULT_MAX];   
    GLubyte idx_to_attr[VERT_RESULT_MAX];   
-   struct brw_sf_point_tex point_attrs[VERT_RESULT_MAX];
 };
 
  
diff --git a/src/mesa/drivers/dri/i965/brw_sf_emit.c b/src/mesa/drivers/dri/i965/brw_sf_emit.c
index bb08055e3bb..56f7c986e78 100644
--- a/src/mesa/drivers/dri/i965/brw_sf_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_sf_emit.c
@@ -354,6 +354,33 @@ static GLboolean calculate_masks( struct brw_sf_compile *c,
    return is_last_attr;
 }
 
+/* Calculates the predicate control for which channels of a reg
+ * (containing 2 attrs) to do point sprite coordinate replacement on.
+ */
+static uint16_t
+calculate_point_sprite_mask(struct brw_sf_compile *c, GLuint reg)
+{
+   int attr1, attr2;
+   uint16_t pc = 0;
+
+   attr1 = c->idx_to_attr[reg * 2];
+   if (attr1 >= VERT_RESULT_TEX0 && attr1 <= VERT_RESULT_TEX7) {
+      if (c->key.point_sprite_coord_replace & (1 << (attr1 - VERT_RESULT_TEX0)))
+	 pc |= 0x0f;
+   }
+
+   if (reg * 2 + 1 < c->nr_setup_attrs) {
+       attr2 = c->idx_to_attr[reg * 2 + 1];
+       if (attr2 >= VERT_RESULT_TEX0 && attr2 <= VERT_RESULT_TEX7) {
+	  if (c->key.point_sprite_coord_replace & (1 << (attr2 -
+							 VERT_RESULT_TEX0)))
+	     pc |= 0xf0;
+       }
+   }
+
+   return pc;
+}
+
 
 
 void brw_emit_tri_setup( struct brw_sf_compile *c, GLboolean allocate)
@@ -529,22 +556,27 @@ void brw_emit_point_sprite_setup( struct brw_sf_compile *c, GLboolean allocate)
    copy_z_inv_w(c);
    for (i = 0; i < c->nr_setup_regs; i++)
    {
-      struct brw_sf_point_tex *tex = &c->point_attrs[c->idx_to_attr[2*i]];
       struct brw_reg a0 = offset(c->vert[0], i);
-      GLushort pc, pc_persp, pc_linear;
+      GLushort pc, pc_persp, pc_linear, pc_coord_replace;
       GLboolean last = calculate_masks(c, i, &pc, &pc_persp, &pc_linear);
-            
-      if (pc_persp)
-      {				
-	  if (!tex->CoordReplace) {
-	      brw_set_predicate_control_flag_value(p, pc_persp);
-	      brw_MUL(p, a0, a0, c->inv_w[0]);
-	  }
+
+      pc_coord_replace = calculate_point_sprite_mask(c, i);
+      pc_persp &= ~pc_coord_replace;
+
+      if (pc_persp) {
+	 brw_set_predicate_control_flag_value(p, pc_persp);
+	 brw_MUL(p, a0, a0, c->inv_w[0]);
       }
 
-      if (tex->CoordReplace) {
-	  /* Caculate 1.0/PointWidth */
-	  brw_math(&c->func,
+      /* Point sprite coordinate replacement: A texcoord with this
+       * enabled gets replaced with the value (x, y, 0, 1) where x and
+       * y vary from 0 to 1 across the horizontal and vertical of the
+       * point.
+       */
+      if (pc_coord_replace) {
+	 brw_set_predicate_control_flag_value(p, pc_coord_replace);
+	 /* Caculate 1.0/PointWidth */
+	 brw_math(&c->func,
 		  c->tmp,
 		  BRW_MATH_FUNCTION_INV,
 		  BRW_MATH_SATURATE_NONE,
@@ -553,50 +585,51 @@ void brw_emit_point_sprite_setup( struct brw_sf_compile *c, GLboolean allocate)
 		  BRW_MATH_DATA_SCALAR,
 		  BRW_MATH_PRECISION_FULL);
 
-	  if (c->key.sprite_origin_lower_left) {
-	   	brw_MUL(p, c->m1Cx, c->tmp, c->inv_w[0]);
-		brw_MOV(p, vec1(suboffset(c->m1Cx, 1)), brw_imm_f(0.0));
-	  	brw_MUL(p, c->m2Cy, c->tmp, negate(c->inv_w[0]));
-		brw_MOV(p, vec1(suboffset(c->m2Cy, 0)), brw_imm_f(0.0));
-	  } else {
-	   	brw_MUL(p, c->m1Cx, c->tmp, c->inv_w[0]);
-		brw_MOV(p, vec1(suboffset(c->m1Cx, 1)), brw_imm_f(0.0));
-	  	brw_MUL(p, c->m2Cy, c->tmp, c->inv_w[0]);
-		brw_MOV(p, vec1(suboffset(c->m2Cy, 0)), brw_imm_f(0.0));
-	  }
-      } else {
-	  brw_MOV(p, c->m1Cx, brw_imm_ud(0));
-	  brw_MOV(p, c->m2Cy, brw_imm_ud(0));
-      }
+	 brw_set_access_mode(p, BRW_ALIGN_16);
 
-      {
-	 brw_set_predicate_control_flag_value(p, pc); 
-	 if (tex->CoordReplace) {
-	     if (c->key.sprite_origin_lower_left) {
-		 brw_MUL(p, c->m3C0, c->inv_w[0], brw_imm_f(1.0));
-		 brw_MOV(p, vec1(suboffset(c->m3C0, 0)), brw_imm_f(0.0));
-	     }
-	     else
-		 brw_MOV(p, c->m3C0, brw_imm_f(0.0));
+	 /* dA/dx, dA/dy */
+	 brw_MOV(p, c->m1Cx, brw_imm_f(0.0));
+	 brw_MOV(p, c->m2Cy, brw_imm_f(0.0));
+	 brw_MOV(p, brw_writemask(c->m1Cx, WRITEMASK_X), c->tmp);
+	 if (c->key.sprite_origin_lower_left) {
+	    brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), negate(c->tmp));
 	 } else {
-	 	brw_MOV(p, c->m3C0, a0); /* constant value */
+	    brw_MOV(p, brw_writemask(c->m2Cy, WRITEMASK_Y), c->tmp);
 	 }
 
-	 /* Copy m0..m3 to URB. 
-	  */
-	 brw_urb_WRITE(p, 
-		       brw_null_reg(),
-		       0,
-		       brw_vec8_grf(0, 0),
-		       0, 	/* allocate */
-		       1,	/* used */
-		       4, 	/* msg len */
-		       0,	/* response len */
-		       last, 	/* eot */
-		       last, 	/* writes complete */
-		       i*4,	/* urb destination offset */
-		       BRW_URB_SWIZZLE_TRANSPOSE);
+	 /* attribute constant offset */
+	 brw_MOV(p, c->m3C0, brw_imm_f(0.0));
+	 if (c->key.sprite_origin_lower_left) {
+	    brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_YW), brw_imm_f(1.0));
+	 } else {
+	    brw_MOV(p, brw_writemask(c->m3C0, WRITEMASK_W), brw_imm_f(1.0));
+	 }
+
+	 brw_set_access_mode(p, BRW_ALIGN_1);
       }
+
+      if (pc & ~pc_coord_replace) {
+	 brw_set_predicate_control_flag_value(p, pc & ~pc_coord_replace);
+	 brw_MOV(p, c->m1Cx, brw_imm_ud(0));
+	 brw_MOV(p, c->m2Cy, brw_imm_ud(0));
+	 brw_MOV(p, c->m3C0, a0); /* constant value */
+      }
+
+
+      brw_set_predicate_control_flag_value(p, pc);
+      /* Copy m0..m3 to URB. */
+      brw_urb_WRITE(p,
+		    brw_null_reg(),
+		    0,
+		    brw_vec8_grf(0, 0),
+		    0, 	/* allocate */
+		    1,	/* used */
+		    4, 	/* msg len */
+		    0,	/* response len */
+		    last, 	/* eot */
+		    last, 	/* writes complete */
+		    i*4,	/* urb destination offset */
+		    BRW_URB_SWIZZLE_TRANSPOSE);
    }
 }
 
diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c
index a7c4b589727..a48804a660f 100644
--- a/src/mesa/drivers/dri/i965/brw_vs_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c
@@ -1717,11 +1717,13 @@ void brw_vs_emit(struct brw_vs_compile *c )
             /* patch all the BREAK/CONT instructions from last BEGINLOOP */
             while (inst0 > loop_inst[loop_depth]) {
                inst0--;
-               if (inst0->header.opcode == BRW_OPCODE_BREAK) {
+               if (inst0->header.opcode == BRW_OPCODE_BREAK &&
+		   inst0->bits3.if_else.jump_count == 0) {
                   inst0->bits3.if_else.jump_count = br * (inst1 - inst0 + 1);
                   inst0->bits3.if_else.pop_count = 0;
                }
-               else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
+               else if (inst0->header.opcode == BRW_OPCODE_CONTINUE &&
+			inst0->bits3.if_else.jump_count == 0) {
                   inst0->bits3.if_else.jump_count = br * (inst1 - inst0);
                   inst0->bits3.if_else.pop_count = 0;
                }
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index 562608e2ecd..ea3c2405af9 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -614,112 +614,6 @@ static void invoke_subroutine( struct brw_wm_compile *c,
     }
 }
 
-/* Workaround for using brw_wm_emit.c's emit functions, which expect
- * destination regs to be uniquely written.  Moves arguments out to
- * temporaries as necessary for instructions which use their destination as
- * a temporary.
- */
-static void
-unalias3(struct brw_wm_compile *c,
-	 void (*func)(struct brw_compile *c,
-		      const struct brw_reg *dst,
-		      GLuint mask,
-		      const struct brw_reg *arg0,
-		      const struct brw_reg *arg1,
-		      const struct brw_reg *arg2),
-	 const struct brw_reg *dst,
-	 GLuint mask,
-	 const struct brw_reg *arg0,
-	 const struct brw_reg *arg1,
-	 const struct brw_reg *arg2)
-{
-    struct brw_compile *p = &c->func;
-    struct brw_reg tmp_arg0[4], tmp_arg1[4], tmp_arg2[4];
-    int i, j;
-    int mark = mark_tmps(c);
-
-    for (j = 0; j < 4; j++) {
-	tmp_arg0[j] = arg0[j];
-	tmp_arg1[j] = arg1[j];
-	tmp_arg2[j] = arg2[j];
-    }
-
-    for (i = 0; i < 4; i++) {
-	if (mask & (1<<i)) {
-	    for (j = 0; j < 4; j++) {
-		if (arg0[j].file == dst[i].file &&
-		    dst[i].nr == arg0[j].nr) {
-		    tmp_arg0[j] = alloc_tmp(c);
-		    brw_MOV(p, tmp_arg0[j], arg0[j]);
-		}
-		if (arg1[j].file == dst[i].file &&
-		    dst[i].nr == arg1[j].nr) {
-		    tmp_arg1[j] = alloc_tmp(c);
-		    brw_MOV(p, tmp_arg1[j], arg1[j]);
-		}
-		if (arg2[j].file == dst[i].file &&
-		    dst[i].nr == arg2[j].nr) {
-		    tmp_arg2[j] = alloc_tmp(c);
-		    brw_MOV(p, tmp_arg2[j], arg2[j]);
-		}
-	    }
-	}
-    }
-
-    func(p, dst, mask, tmp_arg0, tmp_arg1, tmp_arg2);
-
-    release_tmps(c, mark);
-}
-
-/* Workaround for using brw_wm_emit.c's emit functions, which expect
- * destination regs to be uniquely written.  Moves arguments out to
- * temporaries as necessary for instructions which use their destination as
- * a temporary.
- */
-static void
-unalias2(struct brw_wm_compile *c,
-	 void (*func)(struct brw_compile *c,
-		      const struct brw_reg *dst,
-		      GLuint mask,
-		      const struct brw_reg *arg0,
-		      const struct brw_reg *arg1),
-	 const struct brw_reg *dst,
-	 GLuint mask,
-	 const struct brw_reg *arg0,
-	 const struct brw_reg *arg1)
-{
-    struct brw_compile *p = &c->func;
-    struct brw_reg tmp_arg0[4], tmp_arg1[4];
-    int i, j;
-    int mark = mark_tmps(c);
-
-    for (j = 0; j < 4; j++) {
-	tmp_arg0[j] = arg0[j];
-	tmp_arg1[j] = arg1[j];
-    }
-
-    for (i = 0; i < 4; i++) {
-	if (mask & (1<<i)) {
-	    for (j = 0; j < 4; j++) {
-		if (arg0[j].file == dst[i].file &&
-		    dst[i].nr == arg0[j].nr) {
-		    tmp_arg0[j] = alloc_tmp(c);
-		    brw_MOV(p, tmp_arg0[j], arg0[j]);
-		}
-		if (arg1[j].file == dst[i].file &&
-		    dst[i].nr == arg1[j].nr) {
-		    tmp_arg1[j] = alloc_tmp(c);
-		    brw_MOV(p, tmp_arg1[j], arg1[j]);
-		}
-	    }
-	}
-    }
-
-    func(p, dst, mask, tmp_arg0, tmp_arg1);
-
-    release_tmps(c, mark);
-}
-
 static void emit_arl(struct brw_wm_compile *c,
                      const struct prog_instruction *inst)
 {
@@ -1813,14 +1707,29 @@ static void
 get_argument_regs(struct brw_wm_compile *c,
 		  const struct prog_instruction *inst,
 		  int index,
+		  struct brw_reg *dst,
 		  struct brw_reg *regs,
 		  int mask)
 {
-    int i;
+    struct brw_compile *p = &c->func;
+    int i, j;
 
     for (i = 0; i < 4; i++) {
-	if (mask & (1 << i))
+	if (mask & (1 << i)) {
 	    regs[i] = get_src_reg(c, inst, index, i);
+
+	    /* Unalias destination registers from our sources. */
+	    if (regs[i].file == BRW_GENERAL_REGISTER_FILE) {
+	       for (j = 0; j < 4; j++) {
+		   if (memcmp(&regs[i], &dst[j], sizeof(regs[0])) == 0) {
+		       struct brw_reg tmp = alloc_tmp(c);
+		       brw_MOV(p, tmp, regs[i]);
+		       regs[i] = tmp;
+		       break;
+		   }
+	       }
+	    }
+	}
     }
 }
 
@@ -1845,6 +1754,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 	int dst_flags;
 	struct brw_reg args[3][4], dst[4];
 	int j;
+	int mark = mark_tmps( c );
 
         c->cur_inst = i;
 
@@ -1866,7 +1776,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 	   }
 	}
 	for (j = 0; j < brw_wm_nr_args(inst->Opcode); j++)
-	    get_argument_regs(c, inst, j, args[j], WRITEMASK_XYZW);
+	    get_argument_regs(c, inst, j, dst, args[j], WRITEMASK_XYZW);
 
 	dst_flags = inst->DstReg.WriteMask;
 	if (inst->SaturateMode == SATURATE_ZERO_ONE)
@@ -1920,8 +1830,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 		emit_alu1(p, brw_RNDD, dst, dst_flags, args[0]);
 		break;
 	    case OPCODE_LRP:
-		unalias3(c, emit_lrp,
-			 dst, dst_flags, args[0], args[1], args[2]);
+		emit_lrp(p, dst, dst_flags, args[0], args[1], args[2]);
 		break;
 	    case OPCODE_TRUNC:
 		emit_alu1(p, brw_RNDZ, dst, dst_flags, args[0]);
@@ -1961,10 +1870,10 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 		emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
 		break;
 	    case OPCODE_MIN:	
-		unalias2(c, emit_min, dst, dst_flags, args[0], args[1]);
+		emit_min(p, dst, dst_flags, args[0], args[1]);
 		break;
 	    case OPCODE_MAX:	
-		unalias2(c, emit_max, dst, dst_flags, args[0], args[1]);
+		emit_max(p, dst, dst_flags, args[0], args[1]);
 		break;
 	    case OPCODE_DDX:
 	    case OPCODE_DDY:
@@ -2103,11 +2012,13 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
                   /* patch all the BREAK/CONT instructions from last BGNLOOP */
                   while (inst0 > loop_inst[loop_depth]) {
                      inst0--;
-                     if (inst0->header.opcode == BRW_OPCODE_BREAK) {
+                     if (inst0->header.opcode == BRW_OPCODE_BREAK &&
+			 inst0->bits3.if_else.jump_count == 0) {
 			inst0->bits3.if_else.jump_count = br * (inst1 - inst0 + 1);
 			inst0->bits3.if_else.pop_count = 0;
                      }
-                     else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
+                     else if (inst0->header.opcode == BRW_OPCODE_CONTINUE &&
+			      inst0->bits3.if_else.jump_count == 0) {
                         inst0->bits3.if_else.jump_count = br * (inst1 - inst0);
                         inst0->bits3.if_else.pop_count = 0;
                      }
@@ -2119,6 +2030,9 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
 			inst->Opcode);
 	}
 
+	/* Release temporaries containing any unaliased source regs. */
+	release_tmps( c, mark );
+
 	if (inst->CondUpdate)
 	    brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
 	else
diff --git a/src/mesa/drivers/dri/i965/server/intel_dri.c b/src/mesa/drivers/dri/i965/server/intel_dri.c
deleted file mode 120000
index effdd26448a..00000000000
--- a/src/mesa/drivers/dri/i965/server/intel_dri.c
+++ /dev/null
@@ -1 +0,0 @@
-../../intel/server/intel_dri.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/intel/server/intel_dri.c b/src/mesa/drivers/dri/intel/server/intel_dri.c
deleted file mode 100644
index e49c4214ad4..00000000000
--- a/src/mesa/drivers/dri/intel/server/intel_dri.c
+++ /dev/null
@@ -1,1306 +0,0 @@
-/**
- * \file server/intel_dri.c
- * \brief File to perform the device-specific initialization tasks typically
- * done in the X server.
- *
- * Here they are converted to run in the client (or perhaps a standalone
- * process), and to work with the frame buffer device rather than the X
- * server infrastructure.
- * 
- * Copyright (C) 2006 Dave Airlie (airlied@linux.ie)
-
- 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 THE COPYRIGHT HOLDERS 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.
-*/
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
-
-#include "driver.h"
-#include "drm.h"
-
-#include "intel.h"
-#include "i830_dri.h"
-
-#include "memops.h"
-#include "pciaccess.h"
-
-static size_t drm_page_size;
-static int nextTile = 0;
-#define xf86DrvMsg(...) do {} while(0)
-
-static const int pitches[] = {
-  128 * 8,
-  128 * 16,
-  128 * 32,
-  128 * 64,
-  0
-};
-
-static Bool I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea);
-
-static unsigned long
-GetBestTileAlignment(unsigned long size)
-{
-   unsigned long i;
-
-   for (i = KB(512); i < size; i <<= 1)
-      ;
-
-   if (i > MB(64))
-      i = MB(64);
-
-   return i;
-}
-
-static void SetFenceRegs(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  int i;
-  unsigned char *MMIO = ctx->MMIOAddress;
-
-  for (i = 0; i < 8; i++) {
-    OUTREG(FENCE + i * 4, pI830->Fence[i]);
-    //    if (I810_DEBUG & DEBUG_VERBOSE_VGA)
-    fprintf(stderr,"Fence Register : %x\n", pI830->Fence[i]);
-  }
-}
-
-/* Tiled memory is good... really, really good...
- *
- * Need to make it less likely that we miss out on this - probably
- * need to move the frontbuffer away from the 'guarenteed' alignment
- * of the first memory segment, or perhaps allocate a discontigous
- * framebuffer to get more alignment 'sweet spots'.
- */
-static void
-SetFence(const DRIDriverContext *ctx, I830Rec *pI830,
-	 int nr, unsigned int start, unsigned int pitch,
-         unsigned int size)
-{
-   unsigned int val;
-   unsigned int fence_mask = 0;
-   unsigned int fence_pitch;
-
-   if (nr < 0 || nr > 7) {
-      fprintf(stderr,
-		 "SetFence: fence %d out of range\n",nr);
-      return;
-   }
-
-   pI830->Fence[nr] = 0;
-
-   if (IS_I9XX(pI830))
-   	fence_mask = ~I915G_FENCE_START_MASK;
-   else
-   	fence_mask = ~I830_FENCE_START_MASK;
-
-   if (start & fence_mask) {
-      fprintf(stderr,
-		 "SetFence: %d: start (0x%08x) is not %s aligned\n",
-		 nr, start, (IS_I9XX(pI830)) ? "1MB" : "512k");
-      return;
-   }
-
-   if (start % size) {
-      fprintf(stderr,
-		 "SetFence: %d: start (0x%08x) is not size (%dk) aligned\n",
-		 nr, start, size / 1024);
-      return;
-   }
-
-   if (pitch & 127) {
-      fprintf(stderr,
-		 "SetFence: %d: pitch (%d) not a multiple of 128 bytes\n",
-		 nr, pitch);
-      return;
-   }
-
-   val = (start | FENCE_X_MAJOR | FENCE_VALID);
-
-   if (IS_I9XX(pI830)) {
-   	switch (size) {
-	   case MB(1):
-      		val |= I915G_FENCE_SIZE_1M;
-      		break;
-   	   case MB(2):
-      		val |= I915G_FENCE_SIZE_2M;
-      		break;
-   	   case MB(4):
-      		val |= I915G_FENCE_SIZE_4M;
-      		break;
-   	   case MB(8):
-      		val |= I915G_FENCE_SIZE_8M;
-      		break;
-   	   case MB(16):
-      		val |= I915G_FENCE_SIZE_16M;
-      		break;
-   	   case MB(32):
-      		val |= I915G_FENCE_SIZE_32M;
-      		break;
-   	   case MB(64):
-      		val |= I915G_FENCE_SIZE_64M;
-      		break;
-   	   default:
-      		fprintf(stderr,
-		 "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024);
-      		return;
-   	}
-    } else {
-   	switch (size) {
-	   case KB(512):
-      		val |= FENCE_SIZE_512K;
-      		break;
-	   case MB(1):
-      		val |= FENCE_SIZE_1M;
-      		break;
-   	   case MB(2):
-      		val |= FENCE_SIZE_2M;
-      		break;
-   	   case MB(4):
-      		val |= FENCE_SIZE_4M;
-      		break;
-   	   case MB(8):
-      		val |= FENCE_SIZE_8M;
-      		break;
-   	   case MB(16):
-      		val |= FENCE_SIZE_16M;
-      		break;
-   	   case MB(32):
-      		val |= FENCE_SIZE_32M;
-      		break;
-   	   case MB(64):
-      		val |= FENCE_SIZE_64M;
-      		break;
-   	   default:
-      		fprintf(stderr,
-		 "SetFence: %d: illegal size (%d kByte)\n", nr, size / 1024);
-      		return;
-   	}
-   }
-
-   if (IS_I9XX(pI830))
-	fence_pitch = pitch / 512;
-   else
-	fence_pitch = pitch / 128;
-
-   switch (fence_pitch) {
-   case 1:
-      val |= FENCE_PITCH_1;
-      break;
-   case 2:
-      val |= FENCE_PITCH_2;
-      break;
-   case 4:
-      val |= FENCE_PITCH_4;
-      break;
-   case 8:
-      val |= FENCE_PITCH_8;
-      break;
-   case 16:
-      val |= FENCE_PITCH_16;
-      break;
-   case 32:
-      val |= FENCE_PITCH_32;
-      break;
-   case 64:
-      val |= FENCE_PITCH_64;
-      break;
-   default:
-      fprintf(stderr,
-		 "SetFence: %d: illegal pitch (%d)\n", nr, pitch);
-      return;
-   }
-
-   pI830->Fence[nr] = val;
-}
-
-static Bool
-MakeTiles(const DRIDriverContext *ctx, I830Rec *pI830, I830MemRange *pMem)
-{
-   int pitch, ntiles, i;
-
-   pitch = pMem->Pitch * ctx->cpp;
-   /*
-    * Simply try to break the region up into at most four pieces of size
-    * equal to the alignment.
-    */
-   ntiles = ROUND_TO(pMem->Size, pMem->Alignment) / pMem->Alignment;
-   if (ntiles >= 4) {
-      return FALSE;
-   }
-
-   for (i = 0; i < ntiles; i++, nextTile++) {
-     SetFence(ctx, pI830, nextTile, pMem->Start + i * pMem->Alignment,
-	       pitch, pMem->Alignment);
-   }
-   return TRUE;
-}
-
-static void I830SetupMemoryTiling(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  int i;
-
-  /* Clear out */
-  for (i = 0; i < 8; i++)
-    pI830->Fence[i] = 0;
-  
-  nextTile = 0;
-
-  if (pI830->BackBuffer.Alignment >= KB(512)) {
-    if (MakeTiles(ctx, pI830, &(pI830->BackBuffer))) {
-      fprintf(stderr,
-		 "Activating tiled memory for the back buffer.\n");
-    } else {
-      fprintf(stderr,
-		 "MakeTiles failed for the back buffer.\n");
-      pI830->allowPageFlip = FALSE;
-    }
-  }
-  
-  if (pI830->DepthBuffer.Alignment >= KB(512)) {
-    if (MakeTiles(ctx, pI830, &(pI830->DepthBuffer))) {
-      fprintf(stderr,
-		 "Activating tiled memory for the depth buffer.\n");
-    } else {
-      fprintf(stderr,
-		 "MakeTiles failed for the depth buffer.\n");
-    }
-  }
-
-  return;
-}
-
-static int I830DetectMemory(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  struct pci_device host_bridge, ig_dev;
-  uint32_t gmch_ctrl;
-  int memsize = 0;
-  int range;
-  uint32_t aper_size;
-  uint32_t membase2 = 0;
-      
-  memset(&host_bridge, 0, sizeof(host_bridge));
-  memset(&ig_dev, 0, sizeof(ig_dev));
-
-  ig_dev.dev = 2;
-
-  pci_device_cfg_read_u32(&host_bridge, &gmch_ctrl, I830_GMCH_CTRL);
-
-  if (IS_I830(pI830) || IS_845G(pI830)) {
-    if ((gmch_ctrl & I830_GMCH_MEM_MASK) == I830_GMCH_MEM_128M) {
-      aper_size = 0x80000000;
-    } else {
-      aper_size = 0x40000000;
-    }
-  } else {
-    if (IS_I9XX(pI830)) {
-      int ret;
-      ret = pci_device_cfg_read_u32(&ig_dev, &membase2, 0x18);
-      if (membase2 & 0x08000000)
-	aper_size = 0x8000000;
-      else
-	aper_size = 0x10000000;
-
-      fprintf(stderr,"aper size is %08X %08x %d\n", aper_size, membase2, ret);
-    } else
-      aper_size = 0x8000000;
-  }
-
-  pI830->aper_size = aper_size;
-
-
-  /* We need to reduce the stolen size, by the GTT and the popup.
-   * The GTT varying according the the FbMapSize and the popup is 4KB */
-  range = (ctx->shared.fbSize / (1024*1024)) + 4;
-
-   if (IS_I85X(pI830) || IS_I865G(pI830) || IS_I9XX(pI830)) {
-      switch (gmch_ctrl & I830_GMCH_GMS_MASK) {
-      case I855_GMCH_GMS_STOLEN_1M:
-	 memsize = MB(1) - KB(range);
-	 break;
-      case I855_GMCH_GMS_STOLEN_4M:
-	 memsize = MB(4) - KB(range);
-	 break;
-      case I855_GMCH_GMS_STOLEN_8M:
-	 memsize = MB(8) - KB(range);
-	 break;
-      case I855_GMCH_GMS_STOLEN_16M:
-	 memsize = MB(16) - KB(range);
-	 break;
-      case I855_GMCH_GMS_STOLEN_32M:
-	 memsize = MB(32) - KB(range);
-	 break;
-      case I915G_GMCH_GMS_STOLEN_48M:
-	 if (IS_I9XX(pI830))
-	    memsize = MB(48) - KB(range);
-	 break;
-      case I915G_GMCH_GMS_STOLEN_64M:
-	 if (IS_I9XX(pI830))
-	    memsize = MB(64) - KB(range);
-	 break;
-      }
-   } else {
-      switch (gmch_ctrl & I830_GMCH_GMS_MASK) {
-      case I830_GMCH_GMS_STOLEN_512:
-	 memsize = KB(512) - KB(range);
-	 break;
-      case I830_GMCH_GMS_STOLEN_1024:
-	 memsize = MB(1) - KB(range);
-	 break;
-      case I830_GMCH_GMS_STOLEN_8192:
-	 memsize = MB(8) - KB(range);
-	 break;
-      case I830_GMCH_GMS_LOCAL:
-	 memsize = 0;
-	 xf86DrvMsg(pScrn->scrnIndex, X_WARNING,
-		    "Local memory found, but won't be used.\n");
-	 break;
-      }
-   }
-   if (memsize > 0) {
-     fprintf(stderr,
-		 "detected %d kB stolen memory.\n", memsize / 1024);
-   } else {
-     fprintf(stderr,
-		 "no video memory detected.\n");
-   }
-   return memsize;
-}
-
-static int AgpInit(const DRIDriverContext *ctx, I830Rec *info)
-{
-  unsigned long mode = 0x4;
-
-  if (drmAgpAcquire(ctx->drmFD) < 0) {
-    fprintf(stderr, "[gart] AGP not available\n");
-    return 0;
-  }
-  
-  if (drmAgpEnable(ctx->drmFD, mode) < 0) {
-    fprintf(stderr, "[gart] AGP not enabled\n");
-    drmAgpRelease(ctx->drmFD);
-    return 0;
-  }
-  else
-    fprintf(stderr, "[gart] AGP enabled at %dx\n", ctx->agpmode);
-
-  return 1;
-}
-
-/*
- * Allocate memory from the given pool.  Grow the pool if needed and if
- * possible.
- */
-static unsigned long
-AllocFromPool(const DRIDriverContext *ctx, I830Rec *pI830, 
-	      I830MemRange *result, I830MemPool *pool,
-	      long size, unsigned long alignment, int flags)
-{
-   long needed, start, end;
-
-   if (!result || !pool || !size)
-      return 0;
-
-   /* Calculate how much space is needed. */
-   if (alignment <= GTT_PAGE_SIZE)
-      needed = size;
-   else {
-	 start = ROUND_TO(pool->Free.Start, alignment);
-	 end = ROUND_TO(start + size, alignment);
-	 needed = end - pool->Free.Start;
-   }
-   if (needed > pool->Free.Size) {
-     return 0;
-   }
-
-   result->Start = ROUND_TO(pool->Free.Start, alignment);
-   pool->Free.Start += needed;
-   result->End = pool->Free.Start;
-
-   pool->Free.Size = pool->Free.End - pool->Free.Start;
-   result->Size = result->End - result->Start;
-   result->Pool = pool;
-   result->Alignment = alignment;
-   return needed;
-}
-
-static unsigned long AllocFromAGP(const DRIDriverContext *ctx, I830Rec *pI830, long size, unsigned long alignment, I830MemRange  *result)
-{
-   unsigned long start, end;
-   unsigned long newApStart, newApEnd;
-   int ret;
-   if (!result || !size)
-      return 0;
-   
-   if (!alignment)
-     alignment = 4;
-
-   start = ROUND_TO(pI830->MemoryAperture.Start, alignment);
-   end = ROUND_TO(start + size, alignment);
-   newApStart = end;
-   newApEnd = pI830->MemoryAperture.End;
-
-   ret=drmAgpAlloc(ctx->drmFD, size, 0, &(result->Physical), (drm_handle_t *)&(result->Key));
-   
-   if (ret)
-   {
-     fprintf(stderr,"drmAgpAlloc failed %d\n", ret);
-     return 0;
-   }
-   pI830->allocatedMemory += size;
-   pI830->MemoryAperture.Start = newApStart;
-   pI830->MemoryAperture.End = newApEnd;
-   pI830->MemoryAperture.Size = newApEnd - newApStart;
-   //   pI830->FreeMemory -= size;
-   result->Start = start;
-   result->End = start + size;
-   result->Size = size;
-   result->Offset = start;
-   result->Alignment = alignment;
-   result->Pool = NULL;
-  
-   return size;
-}
-
-unsigned long
-I830AllocVidMem(const DRIDriverContext *ctx, I830Rec *pI830,
-                I830MemRange *result, I830MemPool *pool, long size,
-                unsigned long alignment, int flags)
-{
-   unsigned long ret;
-
-   if (!result)
-      return 0;
-
-   /* Make sure these are initialised. */
-   result->Size = 0;
-   result->Key = -1;
-
-   if (!size) {
-      return 0;
-   }
-
-   if (pool->Free.Size < size) {
-      ret = AllocFromAGP(ctx, pI830, size, alignment, result);
-   }
-   else {
-      ret = AllocFromPool(ctx, pI830, result, pool, size, alignment, flags);
-      if (ret == 0)
-         ret = AllocFromAGP(ctx, pI830, size, alignment, result);
-   }
-   return ret;
-}
-
-static Bool BindAgpRange(const DRIDriverContext *ctx, I830MemRange *mem)
-{
-  if (!mem)
-    return FALSE;
-  
-  if (mem->Key == -1)
-    return TRUE;
-
-  return !drmAgpBind(ctx->drmFD, mem->Key, mem->Offset);
-}
-
-/* simple memory allocation routines needed */
-/* put ring buffer in low memory */
-/* need to allocate front, back, depth buffers aligned correctly,
-   allocate ring buffer, 
-*/
-
-/* */
-static Bool
-I830AllocateMemory(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  unsigned long size, ret;
-  unsigned long lines, lineSize, align;
-
-  /* allocate ring buffer */
-  memset(pI830->LpRing, 0, sizeof(I830RingBuffer));
-  pI830->LpRing->mem.Key = -1;
-
-  size = PRIMARY_RINGBUFFER_SIZE;
-  
-  ret = I830AllocVidMem(ctx, pI830, &pI830->LpRing->mem, &pI830->StolenPool, size, 0x1000, 0);
-  
-  if (ret != size)
-  {
-    fprintf(stderr,"unable to allocate ring buffer %ld\n", ret);
-    return FALSE;
-  }
-
-  pI830->LpRing->tail_mask = pI830->LpRing->mem.Size - 1;
-
-  
-  /* allocate front buffer */
-  memset(&(pI830->FrontBuffer), 0, sizeof(pI830->FrontBuffer));
-  pI830->FrontBuffer.Key = -1;
-  pI830->FrontBuffer.Pitch = ctx->shared.virtualWidth;
-
-  align = KB(512);  
-
-  lineSize = ctx->shared.virtualWidth * ctx->cpp;
-  lines = (ctx->shared.virtualHeight + 15) / 16 * 16;
-  size = lineSize * lines;
-  size = ROUND_TO_PAGE(size);
-
-  align = GetBestTileAlignment(size);
-
-  ret = I830AllocVidMem(ctx, pI830, &pI830->FrontBuffer, &pI830->StolenPool, size, align, 0);
-  if (ret < size)
-  {
-    fprintf(stderr,"unable to allocate front buffer %ld\n", ret);
-    return FALSE;
-  }
-
-  memset(&(pI830->BackBuffer), 0, sizeof(pI830->BackBuffer));
-  pI830->BackBuffer.Key = -1;
-  pI830->BackBuffer.Pitch = ctx->shared.virtualWidth;
-
-  ret = I830AllocVidMem(ctx, pI830, &pI830->BackBuffer, &pI830->StolenPool, size, align, 0);
-  if (ret < size)
-  {
-    fprintf(stderr,"unable to allocate back buffer %ld\n", ret);
-    return FALSE;
-  }
-  
-  memset(&(pI830->DepthBuffer), 0, sizeof(pI830->DepthBuffer));
-  pI830->DepthBuffer.Key = -1;
-  pI830->DepthBuffer.Pitch = ctx->shared.virtualWidth;
-
-  ret = I830AllocVidMem(ctx, pI830, &pI830->DepthBuffer, &pI830->StolenPool, size, align, 0);
-  if (ret < size)
-  {
-    fprintf(stderr,"unable to allocate depth buffer %ld\n", ret);
-    return FALSE;
-  }
-
-  memset(&(pI830->ContextMem), 0, sizeof(pI830->ContextMem));
-  pI830->ContextMem.Key = -1;
-  size = KB(32);
-
-  ret = I830AllocVidMem(ctx, pI830, &pI830->ContextMem, &pI830->StolenPool, size, align, 0);
-  if (ret < size)
-  {
-    fprintf(stderr,"unable to allocate context buffer %ld\n", ret);
-    return FALSE;
-  }
-
-#if 0  
-  memset(&(pI830->TexMem), 0, sizeof(pI830->TexMem));
-  pI830->TexMem.Key = -1;
-
-  size = 32768 * 1024;
-  ret = AllocFromAGP(ctx, pI830, size, align, &pI830->TexMem);
-  if (ret < size)
-  {
-    fprintf(stderr,"unable to allocate texture memory %ld\n", ret);
-    return FALSE;
-  }
-#endif
-
-  return TRUE;
-}
-
-static Bool
-I830BindMemory(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  if (!BindAgpRange(ctx, &pI830->LpRing->mem))
-    return FALSE;
-  if (!BindAgpRange(ctx, &pI830->FrontBuffer))
-    return FALSE;
-  if (!BindAgpRange(ctx, &pI830->BackBuffer))
-    return FALSE;
-  if (!BindAgpRange(ctx, &pI830->DepthBuffer))
-    return FALSE;
-  if (!BindAgpRange(ctx, &pI830->ContextMem))
-    return FALSE;
-#if 0
-  if (!BindAgpRange(ctx, &pI830->TexMem))
-    return FALSE;
-#endif
-  return TRUE;
-}
-
-static void SetupDRIMM(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  unsigned long aperEnd = ROUND_DOWN_TO(pI830->aper_size, GTT_PAGE_SIZE) / GTT_PAGE_SIZE;
-  unsigned long aperStart = ROUND_TO(pI830->aper_size - KB(32768), GTT_PAGE_SIZE) / GTT_PAGE_SIZE;
-
-  fprintf(stderr, "aper size is %08X\n", ctx->shared.fbSize);
-  if (drmMMInit(ctx->drmFD, aperStart, aperEnd - aperStart, DRM_BO_MEM_TT)) {
-      fprintf(stderr,
-	      "DRM MM Initialization Failed\n");
-  } else {
-    fprintf(stderr,
-	    "DRM MM Initialized at offset 0x%lx length %d page\n", aperStart, aperEnd-aperStart);
-  }
-
-}
-
-static Bool
-I830CleanupDma(const DRIDriverContext *ctx)
-{
-   drmI830Init info;
-
-   memset(&info, 0, sizeof(drmI830Init));
-   info.func = I830_CLEANUP_DMA;
-
-   if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT,
-		       &info, sizeof(drmI830Init))) {
-     fprintf(stderr, "I830 Dma Cleanup Failed\n");
-      return FALSE;
-   }
-
-   return TRUE;
-}
-
-static Bool
-I830InitDma(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-   I830RingBuffer *ring = pI830->LpRing;
-   drmI830Init info;
-
-   memset(&info, 0, sizeof(drmI830Init));
-   info.func = I830_INIT_DMA;
-
-   info.ring_start = ring->mem.Start + pI830->LinearAddr;
-   info.ring_end = ring->mem.End + pI830->LinearAddr;
-   info.ring_size = ring->mem.Size;
-
-   info.mmio_offset = (unsigned int)ctx->MMIOStart;
-
-   info.sarea_priv_offset = sizeof(drm_sarea_t);
-
-   info.front_offset = pI830->FrontBuffer.Start;
-   info.back_offset = pI830->BackBuffer.Start;
-   info.depth_offset = pI830->DepthBuffer.Start;
-   info.w = ctx->shared.virtualWidth;
-   info.h = ctx->shared.virtualHeight;
-   info.pitch = ctx->shared.virtualWidth;
-   info.back_pitch = pI830->BackBuffer.Pitch;
-   info.depth_pitch = pI830->DepthBuffer.Pitch;
-   info.cpp = ctx->cpp;
-
-   if (drmCommandWrite(ctx->drmFD, DRM_I830_INIT,
-		       &info, sizeof(drmI830Init))) {
-      fprintf(stderr,
-		 "I830 Dma Initialization Failed\n");
-      return FALSE;
-   }
-
-   return TRUE;
-}
-
-static int I830CheckDRMVersion( const DRIDriverContext *ctx,
-				  I830Rec *pI830 )
-{
-   drmVersionPtr  version;
-
-   version = drmGetVersion(ctx->drmFD);
-
-   if (version) {
-     int req_minor, req_patch;
-
-     req_minor = 4;
-     req_patch = 0;	
-
-     if (version->version_major != 1 ||
-	 version->version_minor < req_minor ||
-	 (version->version_minor == req_minor && 
-	  version->version_patchlevel < req_patch)) {
-       /* Incompatible drm version */
-       fprintf(stderr,
-	       "[dri] I830DRIScreenInit failed because of a version "
-	       "mismatch.\n"
-	       "[dri] i915.o kernel module version is %d.%d.%d "
-	       "but version 1.%d.%d or newer is needed.\n"
-	       "[dri] Disabling DRI.\n",
-	       version->version_major,
-	       version->version_minor,
-	       version->version_patchlevel,
-	       req_minor,
-	       req_patch);
-       drmFreeVersion(version);
-       return 0;
-     }
-     
-     pI830->drmMinor = version->version_minor;
-     drmFreeVersion(version);
-   }
-   return 1;
-}
-
-static void
-I830SetRingRegs(const DRIDriverContext *ctx, I830Rec *pI830)
-{
-  unsigned int itemp;
-  unsigned char *MMIO = ctx->MMIOAddress;
-
-   OUTREG(LP_RING + RING_LEN, 0);
-   OUTREG(LP_RING + RING_TAIL, 0);
-   OUTREG(LP_RING + RING_HEAD, 0);
-
-   if ((long)(pI830->LpRing->mem.Start & I830_RING_START_MASK) !=
-       pI830->LpRing->mem.Start) {
-      fprintf(stderr,
-		 "I830SetRingRegs: Ring buffer start (%lx) violates its "
-		 "mask (%x)\n", pI830->LpRing->mem.Start, I830_RING_START_MASK);
-   }
-   /* Don't care about the old value.  Reserved bits must be zero anyway. */
-   itemp = pI830->LpRing->mem.Start & I830_RING_START_MASK;
-   OUTREG(LP_RING + RING_START, itemp);
-
-   if (((pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES) !=
-       pI830->LpRing->mem.Size - 4096) {
-      fprintf(stderr,
-		 "I830SetRingRegs: Ring buffer size - 4096 (%lx) violates its "
-		 "mask (%x)\n", pI830->LpRing->mem.Size - 4096,
-		 I830_RING_NR_PAGES);
-   }
-   /* Don't care about the old value.  Reserved bits must be zero anyway. */
-   itemp = (pI830->LpRing->mem.Size - 4096) & I830_RING_NR_PAGES;
-   itemp |= (RING_NO_REPORT | RING_VALID);
-   OUTREG(LP_RING + RING_LEN, itemp);
-
-   pI830->LpRing->head = INREG(LP_RING + RING_HEAD) & I830_HEAD_MASK;
-   pI830->LpRing->tail = INREG(LP_RING + RING_TAIL);
-   pI830->LpRing->space = pI830->LpRing->head - (pI830->LpRing->tail + 8);
-   if (pI830->LpRing->space < 0)
-      pI830->LpRing->space += pI830->LpRing->mem.Size;
-
-   SetFenceRegs(ctx, pI830);
-   
-   /* RESET THE DISPLAY PIPE TO POINT TO THE FRONTBUFFER - hacky
-      hacky hacky */
-   OUTREG(DSPABASE, pI830->FrontBuffer.Start + pI830->LinearAddr);
-
-}
-
-static Bool
-I830SetParam(const DRIDriverContext *ctx, int param, int value)
-{
-   drmI830SetParam sp;
-
-   memset(&sp, 0, sizeof(sp));
-   sp.param = param;
-   sp.value = value;
-
-   if (drmCommandWrite(ctx->drmFD, DRM_I830_SETPARAM, &sp, sizeof(sp))) {
-      fprintf(stderr, "I830 SetParam Failed\n");
-      return FALSE;
-   }
-
-   return TRUE;
-}
-
-static Bool
-I830DRIMapScreenRegions(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
-{
-   fprintf(stderr,
-              "[drm] Mapping front buffer\n");
-
-   if (drmAddMap(ctx->drmFD,
-                 (drm_handle_t)(sarea->front_offset + pI830->LinearAddr),
-                 sarea->front_size,
-                 DRM_FRAME_BUFFER,  /*DRM_AGP,*/
-                 0,
-                 &sarea->front_handle) < 0) {
-     fprintf(stderr,
-	     "[drm] drmAddMap(front_handle) failed. Disabling DRI\n");
-      return FALSE;
-   }
-   ctx->shared.hFrameBuffer = sarea->front_handle;
-   ctx->shared.fbSize = sarea->front_size;
-   fprintf(stderr, "[drm] Front Buffer = 0x%08x\n",
-	   sarea->front_handle);
-
-   if (drmAddMap(ctx->drmFD,
-                 (drm_handle_t)(sarea->back_offset),
-                 sarea->back_size, DRM_AGP, 0,
-                 &sarea->back_handle) < 0) {
-      fprintf(stderr,
-                 "[drm] drmAddMap(back_handle) failed. Disabling DRI\n");
-      return FALSE;
-   }
-   fprintf(stderr, "[drm] Back Buffer = 0x%08x\n",
-              sarea->back_handle);
-
-   if (drmAddMap(ctx->drmFD,
-                 (drm_handle_t)sarea->depth_offset,
-                 sarea->depth_size, DRM_AGP, 0,
-                 &sarea->depth_handle) < 0) {
-      fprintf(stderr,
-                 "[drm] drmAddMap(depth_handle) failed. Disabling DRI\n");
-      return FALSE;
-   }
-   fprintf(stderr, "[drm] Depth Buffer = 0x%08x\n",
-              sarea->depth_handle);
-
-#if 0
-   if (drmAddMap(ctx->drmFD,
-		 (drm_handle_t)sarea->tex_offset,
-		 sarea->tex_size, DRM_AGP, 0,
-		 &sarea->tex_handle) < 0) {
-      fprintf(stderr,
-		 "[drm] drmAddMap(tex_handle) failed. Disabling DRI\n");
-      return FALSE;
-   }
-   fprintf(stderr, "[drm] textures = 0x%08x\n",
-	      sarea->tex_handle);
-#endif
-   return TRUE;
-}
-
-
-static void
-I830DRIUnmapScreenRegions(const DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
-{
-#if 1
-   if (sarea->front_handle) {
-      drmRmMap(ctx->drmFD, sarea->front_handle);
-      sarea->front_handle = 0;
-   }
-#endif
-   if (sarea->back_handle) {
-      drmRmMap(ctx->drmFD, sarea->back_handle);
-      sarea->back_handle = 0;
-   }
-   if (sarea->depth_handle) {
-      drmRmMap(ctx->drmFD, sarea->depth_handle);
-      sarea->depth_handle = 0;
-   }
-   if (sarea->tex_handle) {
-      drmRmMap(ctx->drmFD, sarea->tex_handle);
-      sarea->tex_handle = 0;
-   }
-}
-
-static Bool
-I830DRIDoMappings(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
-{
-  if (drmAddMap(ctx->drmFD,
-		(drm_handle_t)pI830->LpRing->mem.Start,
-		pI830->LpRing->mem.Size, DRM_AGP, 0,
-		&pI830->ring_map) < 0) {
-    fprintf(stderr,
-	    "[drm] drmAddMap(ring_map) failed. Disabling DRI\n");
-    return FALSE;
-  }
-  fprintf(stderr, "[drm] ring buffer = 0x%08x\n",
-	  pI830->ring_map);
-
-  if (I830InitDma(ctx, pI830) == FALSE) {
-    return FALSE;
-  }
-  
-   /* init to zero to be safe */
-
-  I830DRIMapScreenRegions(ctx, pI830, sarea);
-  SetupDRIMM(ctx, pI830);
-
-   if (ctx->pciDevice != PCI_CHIP_845_G &&
-       ctx->pciDevice != PCI_CHIP_I830_M) {
-      I830SetParam(ctx, I830_SETPARAM_USE_MI_BATCHBUFFER_START, 1 );
-   }
-
-   /* Okay now initialize the dma engine */
-   {
-      pI830->irq = drmGetInterruptFromBusID(ctx->drmFD,
-					    ctx->pciBus,
-					    ctx->pciDevice,
-					    ctx->pciFunc);
-
-      if (drmCtlInstHandler(ctx->drmFD, pI830->irq)) {
-	 fprintf(stderr,
-		    "[drm] failure adding irq handler\n");
-	 pI830->irq = 0;
-	 return FALSE;
-      }
-      else
-	 fprintf(stderr,
-		    "[drm] dma control initialized, using IRQ %d\n",
-		    pI830->irq);
-   }
-
-   fprintf(stderr, "[dri] visual configs initialized\n");
-
-   return TRUE;
-}
-
-static Bool
-I830ClearScreen(DRIDriverContext *ctx, I830Rec *pI830, drmI830Sarea *sarea)
-{
-  /* need to drmMap front and back buffers and zero them */
-  drmAddress map_addr;
-  int ret;
-
-  ret = drmMap(ctx->drmFD,
-	       sarea->front_handle,
-	       sarea->front_size,
-	       &map_addr);
-
-  if (ret)
-  {
-    fprintf(stderr, "Unable to map front buffer\n");
-    return FALSE;
-  }
-
-  drimemsetio((char *)map_addr,
-	      0,
-	      sarea->front_size);
-  drmUnmap(map_addr, sarea->front_size);
-
-
-  ret = drmMap(ctx->drmFD,
-	       sarea->back_handle,
-	       sarea->back_size,
-	       &map_addr);
-
-  if (ret)
-  {
-    fprintf(stderr, "Unable to map back buffer\n");
-    return FALSE;
-  }
-
-  drimemsetio((char *)map_addr,
-	      0,
-	      sarea->back_size);
-  drmUnmap(map_addr, sarea->back_size);
-
-  return TRUE;
-}
-
-static Bool
-I830ScreenInit(DRIDriverContext *ctx, I830Rec *pI830)
-		  
-{
-   I830DRIPtr pI830DRI;
-   drmI830Sarea *pSAREAPriv;
-   int err;
-      
-   drm_page_size = getpagesize();   
-
-   pI830->registerSize = ctx->MMIOSize;
-   /* This is a hack for now.  We have to have more than a 4k page here
-    * because of the size of the state.  However, the state should be
-    * in a per-context mapping.  This will be added in the Mesa 3.5 port
-    * of the I830 driver.
-    */
-   ctx->shared.SAREASize = SAREA_MAX;
-
-   /* Note that drmOpen will try to load the kernel module, if needed. */
-   ctx->drmFD = drmOpen("i915", NULL );
-   if (ctx->drmFD < 0) {
-      fprintf(stderr, "[drm] drmOpen failed\n");
-      return 0;
-   }
-
-   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-	      ctx->drmFD, ctx->pciBusID, strerror(-err));
-      return 0;
-   }
-
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-     fprintf(stderr, "[drm] drmAddMap failed\n");
-     return 0;
-   }
-
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08x\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-   
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   
-   }
-   
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08x to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-   
-
-   if (drmAddMap(ctx->drmFD, 
-		 ctx->MMIOStart,
-		 ctx->MMIOSize,
-		 DRM_REGISTERS, 
-		 DRM_READ_ONLY, 
-		 &pI830->registerHandle) < 0) {
-      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] register handle = 0x%08x\n", pI830->registerHandle);
-
-
-   if (!I830CheckDRMVersion(ctx, pI830)) {
-     return FALSE;
-   }
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-   /* Initialize the SAREA private data structure */
-   pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + 
-				 sizeof(drm_sarea_t));
-   memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
-
-   pI830->StolenMemory.Size = I830DetectMemory(ctx, pI830);
-   pI830->StolenMemory.Start = 0;
-   pI830->StolenMemory.End = pI830->StolenMemory.Size;
-
-   pI830->MemoryAperture.Start = pI830->StolenMemory.End;
-   pI830->MemoryAperture.End = KB(40000);
-   pI830->MemoryAperture.Size = pI830->MemoryAperture.End - pI830->MemoryAperture.Start;
-
-   pI830->StolenPool.Fixed = pI830->StolenMemory;
-   pI830->StolenPool.Total = pI830->StolenMemory;
-   pI830->StolenPool.Free = pI830->StolenPool.Total;
-   pI830->FreeMemory = pI830->StolenPool.Total.Size;
-
-   if (!AgpInit(ctx, pI830))
-     return FALSE;
-
-   if (I830AllocateMemory(ctx, pI830) == FALSE)
-   {
-     return FALSE;
-   }
-
-   if (I830BindMemory(ctx, pI830) == FALSE)
-   {
-     return FALSE;
-   }
-
-   pSAREAPriv->rotated_offset = -1;
-   pSAREAPriv->rotated_size = 0;
-   pSAREAPriv->rotated_pitch = ctx->shared.virtualWidth;
-
-   pSAREAPriv->front_offset = pI830->FrontBuffer.Start;
-   pSAREAPriv->front_size = pI830->FrontBuffer.Size;
-   pSAREAPriv->width = ctx->shared.virtualWidth;
-   pSAREAPriv->height = ctx->shared.virtualHeight;
-   pSAREAPriv->pitch = ctx->shared.virtualWidth;
-   pSAREAPriv->virtualX = ctx->shared.virtualWidth;
-   pSAREAPriv->virtualY = ctx->shared.virtualHeight;
-   pSAREAPriv->back_offset = pI830->BackBuffer.Start;
-   pSAREAPriv->back_size = pI830->BackBuffer.Size;
-   pSAREAPriv->depth_offset = pI830->DepthBuffer.Start;
-   pSAREAPriv->depth_size = pI830->DepthBuffer.Size;
-#if 0
-   pSAREAPriv->tex_offset = pI830->TexMem.Start;
-   pSAREAPriv->tex_size = pI830->TexMem.Size;
-#endif
-   pSAREAPriv->log_tex_granularity = pI830->TexGranularity;
-
-   ctx->driverClientMsg = malloc(sizeof(I830DRIRec));
-   ctx->driverClientMsgSize = sizeof(I830DRIRec);
-   pI830DRI = (I830DRIPtr)ctx->driverClientMsg;
-   pI830DRI->deviceID = pI830->Chipset;
-   pI830DRI->regsSize = I830_REG_SIZE;
-   pI830DRI->width = ctx->shared.virtualWidth;
-   pI830DRI->height = ctx->shared.virtualHeight;
-   pI830DRI->mem = ctx->shared.fbSize;
-   pI830DRI->cpp = ctx->cpp;
-
-   pI830DRI->bitsPerPixel = ctx->bpp;
-   pI830DRI->sarea_priv_offset = sizeof(drm_sarea_t);
-   
-   err = I830DRIDoMappings(ctx, pI830, pSAREAPriv);
-   if (err == FALSE)
-       return FALSE;
-
-   I830SetupMemoryTiling(ctx, pI830);
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-   I830ClearScreen(ctx, pI830, pSAREAPriv);
-
-   I830SetRingRegs(ctx, pI830);
-
-   return TRUE;
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa radeonValidateMode().
- */
-static int i830ValidateMode( const DRIDriverContext *ctx )
-{
-  return 1;
-}
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa i810ValidateMode().
- */
-static int i830PostValidateMode( const DRIDriverContext *ctx )
-{
-  I830Rec *pI830 = ctx->driverPrivate;
-
-  I830SetRingRegs(ctx, pI830);
-  return 1;
-}
-
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Fills in \p info with some default values and some information from \p ctx
- * and then calls I810ScreenInit() for the screen initialization.
- * 
- * Before exiting clears the framebuffer memory accessing it directly.
- */
-static int i830InitFBDev( DRIDriverContext *ctx )
-{
-  I830Rec *pI830 = calloc(1, sizeof(I830Rec));
-  int i;
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      switch (ctx->bpp / 8) {
-      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-      case 3:
-      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-      }
-
-      ctx->shared.virtualWidth = dummy;
-      ctx->shared.Width = ctx->shared.virtualWidth;
-   }
-
-
-   for (i = 0; pitches[i] != 0; i++) {
-     if (pitches[i] >= ctx->shared.virtualWidth) {
-       ctx->shared.virtualWidth = pitches[i];
-       break;
-     }
-   }
-
-   ctx->driverPrivate = (void *)pI830;
-   
-   pI830->LpRing = calloc(1, sizeof(I830RingBuffer));
-   pI830->Chipset = ctx->chipset;
-   pI830->LinearAddr = ctx->FBStart;
-
-   if (!I830ScreenInit( ctx, pI830 ))
-      return 0;
-
-   
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void i830HaltFBDev( DRIDriverContext *ctx )
-{
-  drmI830Sarea *pSAREAPriv;
-  I830Rec *pI830 = ctx->driverPrivate;
-
-   if (pI830->irq) {
-       drmCtlUninstHandler(ctx->drmFD);
-       pI830->irq = 0;   }
-
-   I830CleanupDma(ctx);
-
-  pSAREAPriv = (drmI830Sarea *)(((char*)ctx->pSAREA) + 
-				sizeof(drm_sarea_t));
-
-  I830DRIUnmapScreenRegions(ctx, pI830, pSAREAPriv);
-  drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-  drmClose(ctx->drmFD);
-  
-  if (ctx->driverPrivate) {
-    free(ctx->driverPrivate);
-    ctx->driverPrivate = 0;
-  }
-}
-
-
-extern void i810NotifyFocus( int );
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-const struct DRIDriverRec __driDriver = {
-   i830ValidateMode,
-   i830PostValidateMode,
-   i830InitFBDev,
-   i830HaltFBDev,
-   NULL,//I830EngineShutdown,
-   NULL, //I830EngineRestore,  
-#ifndef _EMBEDDED
-   0,
-#else
-   i810NotifyFocus, 
-#endif
-};
diff --git a/src/mesa/drivers/dri/mach64/Makefile b/src/mesa/drivers/dri/mach64/Makefile
index a8f463e9fd0..c20fdece297 100644
--- a/src/mesa/drivers/dri/mach64/Makefile
+++ b/src/mesa/drivers/dri/mach64/Makefile
@@ -5,9 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = mach64_dri.so
 
-# Not yet
-# MINIGLX_SOURCES = server/mach64_dri.c 
-
 DRIVER_SOURCES = \
 	mach64_context.c \
 	mach64_ioctl.c \
diff --git a/src/mesa/drivers/dri/mga/Makefile b/src/mesa/drivers/dri/mga/Makefile
index 0cc329fb22d..92533bccc29 100644
--- a/src/mesa/drivers/dri/mga/Makefile
+++ b/src/mesa/drivers/dri/mga/Makefile
@@ -5,8 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = mga_dri.so
 
-MINIGLX_SOURCES = server/mga_dri.c 
-
 DRIVER_SOURCES = \
 	mgadd.c \
 	mgaioctl.c \
diff --git a/src/mesa/drivers/dri/mga/server/mga_dri.c b/src/mesa/drivers/dri/mga/server/mga_dri.c
deleted file mode 100644
index bc575e62ee8..00000000000
--- a/src/mesa/drivers/dri/mga/server/mga_dri.c
+++ /dev/null
@@ -1,1088 +0,0 @@
-
-/*
- * Copyright 2000 VA Linux Systems Inc., Fremont, California.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * VA LINUX SYSTEMS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES
- * OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Keith Whitwell <keith@tungstengraphics.com>
- *    Gareth Hughes <gareth@valinux.com>
- */
-
-#include <errno.h>
-#include <unistd.h>
-#include <string.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <inttypes.h>
-
-#include "driver.h"
-#include "drm.h"
-#include "memops.h"
-
-#include "mga_reg.h"
-#include "mga.h"
-#include "mga_macros.h"
-#include "mga_dri.h"
-
-
-/* Quiescence, locking
- */
-#define MGA_TIMEOUT		2048
-
-static void MGAWaitForIdleDMA( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-   drm_lock_t lock;
-   int ret;
-   int i = 0;
-
-   memset( &lock, 0, sizeof(lock) );
-
-   for (;;) {
-      do {
-         /* first ask for quiescent and flush */
-         lock.flags = DRM_LOCK_QUIESCENT | DRM_LOCK_FLUSH;
-         do {
-	    ret = drmCommandWrite( ctx->drmFD, DRM_MGA_FLUSH,
-                                   &lock, sizeof( lock ) );
-         } while ( ret == -EBUSY && i++ < DRM_MGA_IDLE_RETRY );
-
-         /* if it's still busy just try quiescent */
-         if ( ret == -EBUSY ) { 
-            lock.flags = DRM_LOCK_QUIESCENT;
-            do {
-	       ret = drmCommandWrite( ctx->drmFD, DRM_MGA_FLUSH,
-                                      &lock, sizeof( lock ) );
-            } while ( ret == -EBUSY && i++ < DRM_MGA_IDLE_RETRY );
-         }
-      } while ( ( ret == -EBUSY ) && ( i++ < MGA_TIMEOUT ) );
-
-      if ( ret == 0 )
-	 return;
-
-      fprintf( stderr,
-               "[dri] Idle timed out, resetting engine...\n" );
-
-      drmCommandNone( ctx->drmFD, DRM_MGA_RESET );
-   }
-}
-
-static unsigned int mylog2( unsigned int n )
-{
-   unsigned int log2 = 1;
-   while ( n > 1 ) n >>= 1, log2++;
-   return log2;
-}
-
-static int MGADRIAgpInit(struct DRIDriverContextRec *ctx, MGAPtr pMga)
-{
-   unsigned long mode;
-   unsigned int vendor, device;
-   int ret, count, i;
-
-   if(pMga->agpSize < 12)pMga->agpSize = 12;
-   if(pMga->agpSize > 64)pMga->agpSize = 64; /* cap */
-
-   /* FIXME: Make these configurable...
-    */
-   pMga->agp.size = pMga->agpSize * 1024 * 1024;
-
-   pMga->warp.offset = 0;
-   pMga->warp.size = MGA_WARP_UCODE_SIZE;
-
-   pMga->primary.offset = (pMga->warp.offset +
-				    pMga->warp.size);
-   pMga->primary.size = 1024 * 1024;
-
-   pMga->buffers.offset = (pMga->primary.offset +
-				    pMga->primary.size);
-   pMga->buffers.size = MGA_NUM_BUFFERS * MGA_BUFFER_SIZE;
-
-
-   pMga->agpTextures.offset = (pMga->buffers.offset +
-                                    pMga->buffers.size);
-
-   pMga->agpTextures.size = pMga->agp.size -
-                                     pMga->agpTextures.offset;
-
-   if ( drmAgpAcquire( ctx->drmFD ) < 0 ) {
-     fprintf( stderr, "[agp] AGP not available\n" );
-      return 0;
-   }
-
-   mode   = drmAgpGetMode( ctx->drmFD );        /* Default mode */
-   vendor = drmAgpVendorId( ctx->drmFD );
-   device = drmAgpDeviceId( ctx->drmFD );
-
-   mode &= ~MGA_AGP_MODE_MASK;
-   switch ( pMga->agpMode ) {
-   case 4:
-      mode |= MGA_AGP_4X_MODE;
-   case 2:
-      mode |= MGA_AGP_2X_MODE;
-   case 1:
-   default:
-      mode |= MGA_AGP_1X_MODE;
-   }
-
-#if 0
-   fprintf( stderr,
-            "[agp] Mode 0x%08lx [AGP 0x%04x/0x%04x; Card 0x%04x/0x%04x]\n",
-            mode, vendor, device,
-            ctx->pciVendor,
-            ctx->pciChipType );
-#endif
-
-   if ( drmAgpEnable( ctx->drmFD, mode ) < 0 ) {
-     fprintf( stderr, "[agp] AGP not enabled\n" );
-      drmAgpRelease( ctx->drmFD );
-      return 0;
-   }
-
-   if ( pMga->Chipset == PCI_CHIP_MGAG200 ) {
-      switch ( pMga->agpMode ) {
-      case 2:
-	 fprintf( stderr,
-		     "[drm] Enabling AGP 2x PLL encoding\n" );
-	 OUTREG( MGAREG_AGP_PLL, MGA_AGP2XPLL_ENABLE );
-	 break;
-
-      case 1:
-      default:
-	 fprintf( stderr,
-		     "[drm] Disabling AGP 2x PLL encoding\n" );
-	 OUTREG( MGAREG_AGP_PLL, MGA_AGP2XPLL_DISABLE );
-	 pMga->agpMode = 1;
-	 break;
-      }
-   }
-
-   ret = drmAgpAlloc( ctx->drmFD, pMga->agp.size,
-		      0, NULL, &pMga->agp.handle );
-   if ( ret < 0 ) {
-      fprintf( stderr, "[agp] Out of memory (%d)\n", ret );
-      drmAgpRelease( ctx->drmFD );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[agp] %d kB allocated with handle 0x%08x\n",
-	       pMga->agp.size/1024, (unsigned int)pMga->agp.handle );
-
-   if ( drmAgpBind( ctx->drmFD, pMga->agp.handle, 0 ) < 0 ) {
-      fprintf( stderr, "[agp] Could not bind memory\n" );
-      drmAgpFree( ctx->drmFD, pMga->agp.handle );
-      drmAgpRelease( ctx->drmFD );
-      return 0;
-   }
-
-   /* WARP microcode space
-    */
-   if ( drmAddMap( ctx->drmFD,
-		   pMga->warp.offset,
-		   pMga->warp.size,
-		   DRM_AGP, DRM_READ_ONLY,
-		   &pMga->warp.handle ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not add WARP microcode mapping\n" );
-      return 0;
-   }
-   fprintf( stderr,
- 	       "[agp] WARP microcode handle = 0x%08x\n",
-	       pMga->warp.handle );
-
-   if ( drmMap( ctx->drmFD,
-		pMga->warp.handle,
-		pMga->warp.size,
-		&pMga->warp.map ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not map WARP microcode\n" );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[agp] WARP microcode mapped at 0x%08lx\n",
-	       (unsigned long)pMga->warp.map );
-
-   /* Primary DMA space
-    */
-   if ( drmAddMap( ctx->drmFD,
-		   pMga->primary.offset,
-		   pMga->primary.size,
-		   DRM_AGP, DRM_READ_ONLY,
-		   &pMga->primary.handle ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not add primary DMA mapping\n" );
-      return 0;
-   }
-   fprintf( stderr,
- 	       "[agp] Primary DMA handle = 0x%08x\n",
-	       pMga->primary.handle );
-
-   if ( drmMap( ctx->drmFD,
-		pMga->primary.handle,
-		pMga->primary.size,
-		&pMga->primary.map ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not map primary DMA\n" );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[agp] Primary DMA mapped at 0x%08lx\n",
-	       (unsigned long)pMga->primary.map );
-
-   /* DMA buffers
-    */
-   if ( drmAddMap( ctx->drmFD,
-		   pMga->buffers.offset,
-		   pMga->buffers.size,
-		   DRM_AGP, 0,
-		   &pMga->buffers.handle ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not add DMA buffers mapping\n" );
-      return 0;
-   }
-   fprintf( stderr,
- 	       "[agp] DMA buffers handle = 0x%08x\n",
-	       pMga->buffers.handle );
-
-   if ( drmMap( ctx->drmFD,
-		pMga->buffers.handle,
-		pMga->buffers.size,
-		&pMga->buffers.map ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not map DMA buffers\n" );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[agp] DMA buffers mapped at 0x%08lx\n",
-	       (unsigned long)pMga->buffers.map );
-
-   count = drmAddBufs( ctx->drmFD,
-		       MGA_NUM_BUFFERS, MGA_BUFFER_SIZE,
-		       DRM_AGP_BUFFER, pMga->buffers.offset );
-   if ( count <= 0 ) {
-      fprintf( stderr,
-		  "[drm] failure adding %d %d byte DMA buffers\n",
-		  MGA_NUM_BUFFERS, MGA_BUFFER_SIZE );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[drm] Added %d %d byte DMA buffers\n",
-	       count, MGA_BUFFER_SIZE );
-
-   i = mylog2(pMga->agpTextures.size / MGA_NR_TEX_REGIONS);
-   if(i < MGA_LOG_MIN_TEX_REGION_SIZE)
-      i = MGA_LOG_MIN_TEX_REGION_SIZE;
-   pMga->agpTextures.size = (pMga->agpTextures.size >> i) << i;
-
-   if ( drmAddMap( ctx->drmFD,
-                   pMga->agpTextures.offset,
-                   pMga->agpTextures.size,
-                   DRM_AGP, 0,
-                   &pMga->agpTextures.handle ) < 0 ) {
-      fprintf( stderr,
-                  "[agp] Could not add agpTexture mapping\n" );
-      return 0;
-   }
-/* should i map it ? */
-   fprintf( stderr,
-               "[agp] agpTexture handle = 0x%08x\n",
-               pMga->agpTextures.handle );
-   fprintf( stderr,
-               "[agp] agpTexture size: %d kb\n", pMga->agpTextures.size/1024 );
-
-   return 1;
-}
-
-static int MGADRIMapInit( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-   pMga->registers.size = MGAIOMAPSIZE;
-
-   if ( drmAddMap( ctx->drmFD,
-		   (drm_handle_t)pMga->IOAddress,
-		   pMga->registers.size,
-		   DRM_REGISTERS, DRM_READ_ONLY,
-		   &pMga->registers.handle ) < 0 ) {
-      fprintf( stderr,
-		  "[drm] Could not add MMIO registers mapping\n" );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[drm] Registers handle = 0x%08lx\n",
-	       pMga->registers.handle );
-
-   pMga->status.size = SAREA_MAX;
-
-   if ( drmAddMap( ctx->drmFD, 0, pMga->status.size,
-		   DRM_SHM, DRM_READ_ONLY | DRM_LOCKED | DRM_KERNEL,
-		   &pMga->status.handle ) < 0 ) {
-      fprintf( stderr,
-		  "[drm] Could not add status page mapping\n" );
-      return 0;
-   }
-   fprintf( stderr,
- 	       "[drm] Status handle = 0x%08x\n",
-	       pMga->status.handle );
-
-   if ( drmMap( ctx->drmFD,
-		pMga->status.handle,
-		pMga->status.size,
-		&pMga->status.map ) < 0 ) {
-      fprintf( stderr,
-		  "[agp] Could not map status page\n" );
-      return 0;
-   }
-   fprintf( stderr,
-	       "[agp] Status page mapped at 0x%08lx\n",
-	       (unsigned long)pMga->status.map );
-
-   return 1;
-}
-
-static int MGADRIKernelInit( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-   drm_mga_init_t init;
-   int ret;
-
-   memset( &init, 0, sizeof(init) );
-
-   init.func = MGA_INIT_DMA;
-   init.sarea_priv_offset = sizeof(drm_sarea_t);
-
-   switch ( pMga->Chipset ) {
-   case PCI_CHIP_MGAG550:
-   case PCI_CHIP_MGAG400:
-      init.chipset = MGA_CARD_TYPE_G400;
-      break;
-   case PCI_CHIP_MGAG200:
-   case PCI_CHIP_MGAG200_PCI:
-      init.chipset = MGA_CARD_TYPE_G200;
-      break;
-   default:
-      return 0;
-   }
-
-   init.sgram = 0; /* FIXME !pMga->HasSDRAM; */
-
-
-   switch (ctx->bpp)
-     {
-     case 16:
-       init.maccess = MGA_MACCESS_PW16;
-       break;
-     case 32:
-       init.maccess = MGA_MACCESS_PW32;
-       break;
-     default:
-       fprintf( stderr, "[mga] invalid bpp (%d)\n", ctx->bpp );
-       return 0;
-     }
-
-
-   init.fb_cpp		= ctx->bpp / 8;
-   init.front_offset	= pMga->frontOffset;
-   init.front_pitch	= pMga->frontPitch / init.fb_cpp;
-   init.back_offset	= pMga->backOffset;
-   init.back_pitch	= pMga->backPitch / init.fb_cpp;
-
-   init.depth_cpp	= ctx->bpp / 8;
-   init.depth_offset	= pMga->depthOffset;
-   init.depth_pitch	= pMga->depthPitch / init.depth_cpp;
-
-   init.texture_offset[0] = pMga->textureOffset;
-   init.texture_size[0] = pMga->textureSize;
-
-   init.fb_offset = ctx->shared.hFrameBuffer;
-   init.mmio_offset = pMga->registers.handle;
-   init.status_offset = pMga->status.handle;
-
-   init.warp_offset = pMga->warp.handle;
-   init.primary_offset = pMga->primary.handle;
-   init.buffers_offset = pMga->buffers.handle;
-
-   init.texture_offset[1] = pMga->agpTextures.handle;
-   init.texture_size[1] = pMga->agpTextures.size;
-
-   ret = drmCommandWrite( ctx->drmFD, DRM_MGA_INIT, &init, sizeof(init));
-   if ( ret < 0 ) {
-      fprintf( stderr,
-		  "[drm] Failed to initialize DMA! (%d)\n", ret );
-      return 0;
-   }
-
-   return 1;
-}
-
-static void MGADRIIrqInit(struct DRIDriverContextRec *ctx, MGAPtr pMga)
-{
-  if (!pMga->irq)
-    {
-      pMga->irq = drmGetInterruptFromBusID(ctx->drmFD,
-                                           ctx->pciBus,
-                                           ctx->pciDevice,
-                                           ctx->pciFunc);
-
-      fprintf(stderr, "[drm] got IRQ %d\n", pMga->irq);
-
-    if((drmCtlInstHandler(ctx->drmFD, pMga->irq)) != 0)
-      {
-        fprintf(stderr,
-                "[drm] failure adding irq handler, "
-                "there is a device already using that irq\n"
-                "[drm] falling back to irq-free operation\n");
-        pMga->irq = 0;
-      }
-    else
-      {
-        pMga->reg_ien = INREG( MGAREG_IEN );
-      }
-    }
-
-  if (pMga->irq)
-    fprintf(stderr,
-            "[drm] dma control initialized, using IRQ %d\n",
-            pMga->irq);
-}
-
-static int MGADRIBuffersInit( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-   pMga->drmBuffers = drmMapBufs( ctx->drmFD );
-   if ( !pMga->drmBuffers )
-     {
-       fprintf( stderr,
-                "[drm] Failed to map DMA buffers list\n" );
-       return 0;
-     }
-   
-   fprintf( stderr,
-            "[drm] Mapped %d DMA buffers\n",
-            pMga->drmBuffers->count );
-
-   return 1;
-}
-
-static int MGAMemoryInit( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
-   int        bufferSize  = ((ctx->shared.virtualHeight * width_bytes
-			      + MGA_BUFFER_ALIGN)
-			     & ~MGA_BUFFER_ALIGN);
-   int        depthSize   = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes
-			      + MGA_BUFFER_ALIGN)
-			     & ~MGA_BUFFER_ALIGN);
-   int        l;
-
-   pMga->frontOffset = 0;
-   pMga->frontPitch = ctx->shared.virtualWidth * ctx->cpp;
-
-   fprintf(stderr, 
-	   "Using %d MB AGP aperture\n", pMga->agpSize);
-   fprintf(stderr, 
-	   "Using %d MB for vertex/indirect buffers\n", pMga->buffers.size>>20);
-   fprintf(stderr, 
-	   "Using %d MB for AGP textures\n", pMga->agpTextures.size>>20);
-
-   /* Front, back and depth buffers - everything else texture??
-    */
-   pMga->textureSize = ctx->shared.fbSize - 2 * bufferSize - depthSize;
-
-   if (pMga->textureSize < 0) 
-      return 0;
-
-   l = mylog2( pMga->textureSize / MGA_NR_TEX_REGIONS );
-   if ( l < MGA_LOG_MIN_TEX_REGION_SIZE )
-      l = MGA_LOG_MIN_TEX_REGION_SIZE;
-
-   /* Round the texture size up to the nearest whole number of
-    * texture regions.  Again, be greedy about this, don't
-    * round down.
-    */
-   pMga->logTextureGranularity = l;
-   pMga->textureSize = (pMga->textureSize >> l) << l;
-
-   /* Set a minimum usable local texture heap size.  This will fit
-    * two 256x256x32bpp textures.
-    */
-   if (pMga->textureSize < 512 * 1024) {
-      pMga->textureOffset = 0;
-      pMga->textureSize = 0;
-   }
-
-   /* Reserve space for textures */
-   pMga->textureOffset = ((ctx->shared.fbSize - pMga->textureSize +
-			   MGA_BUFFER_ALIGN) &
-			  ~MGA_BUFFER_ALIGN);
-
-   /* Reserve space for the shared depth
-    * buffer.
-    */
-   pMga->depthOffset = ((pMga->textureOffset - depthSize +
-			 MGA_BUFFER_ALIGN) &
-			~MGA_BUFFER_ALIGN);
-   pMga->depthPitch = ctx->shared.virtualWidth * ctx->cpp;
-
-   pMga->backOffset = ((pMga->depthOffset - bufferSize +
-			MGA_BUFFER_ALIGN) &
-                        ~MGA_BUFFER_ALIGN);
-   pMga->backPitch = ctx->shared.virtualWidth * ctx->cpp;
-
-
-   fprintf(stderr, 
-	   "Will use back buffer at offset 0x%x\n",
-	   pMga->backOffset);
-   fprintf(stderr, 
-	   "Will use depth buffer at offset 0x%x\n",
-	   pMga->depthOffset);
-   fprintf(stderr, 
-	   "Will use %d kb for textures at offset 0x%x\n",
-	   pMga->textureSize/1024, pMga->textureOffset);
-
-   return 1;
-} 
-
-static int MGACheckDRMVersion( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-  drmVersionPtr version;
-
-  /* Check the MGA DRM version */
-  version = drmGetVersion(ctx->drmFD);
-  if ( version ) {
-    if ( version->version_major != 3 ||
-         version->version_minor < 0 ) {
-            /* incompatible drm version */
-      fprintf( stderr,
-               "[dri] MGADRIScreenInit failed because of a version mismatch.\n"
-               "[dri] mga.o kernel module version is %d.%d.%d but version 3.0.x is needed.\n"
-               "[dri] Disabling DRI.\n",
-               version->version_major,
-               version->version_minor,
-               version->version_patchlevel );
-      drmFreeVersion( version );
-      return 0;
-    }
-    drmFreeVersion( version );
-  }
-
-  return 1;
-}
-
-static void print_client_msg( MGADRIPtr pMGADRI )
-{
-  fprintf( stderr, "chipset:                  %d\n", pMGADRI->chipset );
-
-  fprintf( stderr, "width:                    %d\n", pMGADRI->width );
-  fprintf( stderr, "height:                   %d\n", pMGADRI->height );
-  fprintf( stderr, "mem:                      %d\n", pMGADRI->mem );
-  fprintf( stderr, "cpp:                      %d\n", pMGADRI->cpp );
-
-  fprintf( stderr, "agpMode:                  %d\n", pMGADRI->agpMode );
-
-  fprintf( stderr, "frontOffset:              %d\n", pMGADRI->frontOffset );
-  fprintf( stderr, "frontPitch:               %d\n", pMGADRI->frontPitch );
-
-  fprintf( stderr, "backOffset:               %d\n", pMGADRI->backOffset );
-  fprintf( stderr, "backPitch:                %d\n", pMGADRI->backPitch );
-
-  fprintf( stderr, "depthOffset:              %d\n", pMGADRI->depthOffset );
-  fprintf( stderr, "depthPitch:               %d\n", pMGADRI->depthPitch );
-
-  fprintf( stderr, "textureOffset:            %d\n", pMGADRI->textureOffset );
-  fprintf( stderr, "textureSize:              %d\n", pMGADRI->textureSize );
-
-  fprintf( stderr, "logTextureGranularity:    %d\n", pMGADRI->logTextureGranularity );
-  fprintf( stderr, "logAgpTextureGranularity: %d\n", pMGADRI->logAgpTextureGranularity );
-
-  fprintf( stderr, "agpTextureHandle:         %u\n", (unsigned int)pMGADRI->agpTextureOffset );
-  fprintf( stderr, "agpTextureSize:           %u\n", (unsigned int)pMGADRI->agpTextureSize );
-
-#if 0
-   pMGADRI->registers.handle	= pMga->registers.handle;
-   pMGADRI->registers.size	= pMga->registers.size;
-   pMGADRI->status.handle	= pMga->status.handle;
-   pMGADRI->status.size		= pMga->status.size;
-   pMGADRI->primary.handle	= pMga->primary.handle;
-   pMGADRI->primary.size	= pMga->primary.size;
-   pMGADRI->buffers.handle	= pMga->buffers.handle;
-   pMGADRI->buffers.size	= pMga->buffers.size;
-   pMGADRI->sarea_priv_offset = sizeof(drm_sarea_t);
-#endif
-}
-
-static int MGAScreenInit( struct DRIDriverContextRec *ctx, MGAPtr pMga )
-{
-  int       i;
-  int       err;
-  MGADRIPtr pMGADRI;
-
-  usleep(100);
-  /*assert(!ctx->IsClient);*/
-
-   {
-      int  width_bytes = (ctx->shared.virtualWidth * ctx->cpp);
-      int  maxy        = ctx->shared.fbSize / width_bytes;
-
-
-      if (maxy <= ctx->shared.virtualHeight * 3) {
-	 fprintf(stderr, 
-		 "Static buffer allocation failed -- "
-		 "need at least %d kB video memory (have %d kB)\n",
-		 (ctx->shared.virtualWidth * ctx->shared.virtualHeight *
-		  ctx->cpp * 3 + 1023) / 1024,
-		 ctx->shared.fbSize / 1024);
-	 return 0;
-      } 
-   }
-
-   switch(pMga->Chipset) {
-   case PCI_CHIP_MGAG550:
-   case PCI_CHIP_MGAG400:
-   case PCI_CHIP_MGAG200:
-#if 0
-   case PCI_CHIP_MGAG200_PCI:
-#endif
-      break;
-   default:
-      fprintf(stderr, "[drm] Direct rendering only supported with G200/G400/G550 AGP\n");
-      return 0;
-   }
-
-   fprintf( stderr,
-	       "[drm] bpp: %d depth: %d\n",
-            ctx->bpp, ctx->bpp /* FIXME: depth */ );
-
-   if ( (ctx->bpp / 8) != 2 &&
-	(ctx->bpp / 8) != 4 ) {
-      fprintf( stderr,
-		  "[dri] Direct rendering only supported in 16 and 32 bpp modes\n" );
-      return 0;
-   }
-
-   ctx->shared.SAREASize = SAREA_MAX;
-
-
-   /* Note that drmOpen will try to load the kernel module, if needed. */
-   ctx->drmFD = drmOpen("mga", NULL );
-   if (ctx->drmFD < 0) {
-      fprintf(stderr, "[drm] drmOpen failed\n");
-      return 0;
-   }
-
-   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-	      ctx->drmFD, ctx->pciBusID, strerror(-err));
-      return 0;
-   }
-
-     
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   }
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-   
-   /* Need to AddMap the framebuffer and mmio regions here:
-    */
-   if (drmAddMap( ctx->drmFD,
-		  (drm_handle_t)ctx->FBStart,
-		  ctx->FBSize,
-		  DRM_FRAME_BUFFER,
-		  0,
-		  &ctx->shared.hFrameBuffer) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-	   ctx->shared.hFrameBuffer);
-
-
-#if 0 /* will be done in MGADRIMapInit */
-   if (drmAddMap(ctx->drmFD, 
-		 ctx->FixedInfo.mmio_start,
-		 ctx->FixedInfo.mmio_len,
-		 DRM_REGISTERS, 
-		 DRM_READ_ONLY, 
-		 &pMga->registers.handle) < 0) {
-      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] register handle = 0x%08lx\n", pMga->registers.handle);
-#endif
-
-
-   /* Check the mga DRM version */
-   if (!MGACheckDRMVersion(ctx, pMga)) {
-      return 0;
-   }
-
-   if ( !MGADRIAgpInit( ctx, pMga ) ) {
-      return 0;
-   }
-
-   if ( !MGADRIMapInit( ctx, pMga ) ) {
-      return 0;
-   }
-
-   /* Memory manager setup */
-   if (!MGAMemoryInit(ctx, pMga)) {
-      return 0;
-   }
-
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-   /* Initialize the kernel data structures */
-   if (!MGADRIKernelInit(ctx, pMga)) {
-      fprintf(stderr, "MGADRIKernelInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   /* Initialize the vertex buffers list */
-   if (!MGADRIBuffersInit(ctx, pMga)) {
-      fprintf(stderr, "MGADRIBuffersInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   /* Initialize IRQ */
-   MGADRIIrqInit(ctx, pMga);
-
-
-   /* Initialize the SAREA private data structure */
-   {
-      drm_mga_sarea_t *pSAREAPriv;
-      pSAREAPriv = (drm_mga_sarea_t *)(((char*)ctx->pSAREA) + 
-					sizeof(drm_sarea_t));
-      memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
-   }
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-   drimemsetio((char *)ctx->FBAddress + pMga->frontOffset,
-	  0,
-	  pMga->frontPitch * ctx->shared.virtualHeight );
-
-   drimemsetio((char *)ctx->FBAddress + pMga->backOffset,
-	  0,
-	  pMga->backPitch * ctx->shared.virtualHeight );
-
-   /* Can release the lock now */
-/*   DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);*/
-
-   /* This is the struct passed to radeon_dri.so for its initialization */
-   ctx->driverClientMsg = malloc(sizeof(MGADRIRec));
-   ctx->driverClientMsgSize = sizeof(MGADRIRec);
-
-   pMGADRI                    = (MGADRIPtr)ctx->driverClientMsg;
-
-
-   switch(pMga->Chipset) {
-   case PCI_CHIP_MGAG550:
-   case PCI_CHIP_MGAG400:
-      pMGADRI->chipset = MGA_CARD_TYPE_G400;
-      break;
-   case PCI_CHIP_MGAG200:
-   case PCI_CHIP_MGAG200_PCI:
-      pMGADRI->chipset = MGA_CARD_TYPE_G200;
-      break;
-   default:
-      return 0;
-   }
-   pMGADRI->width		= ctx->shared.virtualWidth;
-   pMGADRI->height		= ctx->shared.virtualHeight;
-   pMGADRI->mem			= ctx->shared.fbSize;
-   pMGADRI->cpp			= ctx->bpp / 8;
-
-   pMGADRI->agpMode		= pMga->agpMode;
-
-   pMGADRI->frontOffset		= pMga->frontOffset;
-   pMGADRI->frontPitch		= pMga->frontPitch;
-   pMGADRI->backOffset		= pMga->backOffset;
-   pMGADRI->backPitch		= pMga->backPitch;
-   pMGADRI->depthOffset		= pMga->depthOffset;
-   pMGADRI->depthPitch		= pMga->depthPitch;
-   pMGADRI->textureOffset	= pMga->textureOffset;
-   pMGADRI->textureSize		= pMga->textureSize;
-   pMGADRI->logTextureGranularity = pMga->logTextureGranularity;
-
-   i = mylog2( pMga->agpTextures.size / MGA_NR_TEX_REGIONS );
-   if ( i < MGA_LOG_MIN_TEX_REGION_SIZE )
-      i = MGA_LOG_MIN_TEX_REGION_SIZE;
-
-   pMGADRI->logAgpTextureGranularity = i;
-   pMGADRI->agpTextureOffset = (unsigned int)pMga->agpTextures.handle;
-   pMGADRI->agpTextureSize = (unsigned int)pMga->agpTextures.size;
-
-   pMGADRI->registers.handle	= pMga->registers.handle;
-   pMGADRI->registers.size	= pMga->registers.size;
-   pMGADRI->status.handle	= pMga->status.handle;
-   pMGADRI->status.size		= pMga->status.size;
-   pMGADRI->primary.handle	= pMga->primary.handle;
-   pMGADRI->primary.size	= pMga->primary.size;
-   pMGADRI->buffers.handle	= pMga->buffers.handle;
-   pMGADRI->buffers.size	= pMga->buffers.size;
-   pMGADRI->sarea_priv_offset = sizeof(drm_sarea_t);
-
-   print_client_msg( pMGADRI );
-
-   return 1;
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa mgaValidateMode().
- */
-static int mgaValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa mgaValidateMode().
- */
-static int mgaPostValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Fills in \p info with some default values and some information from \p ctx
- * and then calls MGAScreenInit() for the screen initialization.
- * 
- * Before exiting clears the framebuffer memomry accessing it directly.
- */
-static int mgaInitFBDev( struct DRIDriverContextRec *ctx )
-{
-   MGAPtr pMga = calloc(1, sizeof(*pMga));
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      switch (ctx->bpp / 8) {
-      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-      case 3:
-      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-      }
-
-      ctx->shared.virtualWidth = dummy;
-   }
-
-   ctx->driverPrivate = (void *)pMga;
-   
-   pMga->agpMode       = MGA_DEFAULT_AGP_MODE;
-   pMga->agpSize       = MGA_DEFAULT_AGP_SIZE;
-  
-   pMga->Chipset = ctx->chipset;
-
-   pMga->IOAddress = ctx->MMIOStart;
-   pMga->IOBase    = ctx->MMIOAddress;
-
-   pMga->frontPitch = ctx->shared.virtualWidth * ctx->cpp;
-
-   if (!MGAScreenInit( ctx, pMga ))
-      return 0;
-
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void mgaHaltFBDev( struct DRIDriverContextRec *ctx )
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-       free(ctx->driverPrivate);
-       ctx->driverPrivate = NULL;
-    }
-}
-
-
-static int mgaEngineShutdown( const DRIDriverContext *ctx )
-{
-   fprintf(stderr, "%s() is not yet implemented!\n", __FUNCTION__);
-
-   return 1;
-}
-
-static int mgaEngineRestore( const DRIDriverContext *ctx )
-{
-   fprintf(stderr, "%s() is not yet implemented!\n", __FUNCTION__);
-
-   return 1;
-}
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-struct DRIDriverRec __driDriver = {
-   mgaValidateMode,
-   mgaPostValidateMode,
-   mgaInitFBDev,
-   mgaHaltFBDev,
-   mgaEngineShutdown,
-   mgaEngineRestore,
-   0
-};
-
-
-
-
-#if 0
-void MGADRICloseScreen( ScreenPtr pScreen )
-{
-   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
-   MGAPtr pMga = MGAPTR(pScrn);
-   MGADRIServerPrivatePtr pMga = pMga->DRIServerInfo;
-   drmMGAInit init;
-
-   if ( pMga->drmBuffers ) {
-      drmUnmapBufs( pMga->drmBuffers );
-      pMga->drmBuffers = NULL;
-   }
-
-   if (pMga->irq) {
-      drmCtlUninstHandler(ctx->drmFD);
-      pMga->irq = 0;
-   }
-
-   /* Cleanup DMA */
-   memset( &init, 0, sizeof(drmMGAInit) );
-   init.func = MGA_CLEANUP_DMA;
-   drmCommandWrite( ctx->drmFD, DRM_MGA_INIT, &init, sizeof(drmMGAInit) );
-
-   if ( pMga->status.map ) {
-      drmUnmap( pMga->status.map, pMga->status.size );
-      pMga->status.map = NULL;
-   }
-   if ( pMga->buffers.map ) {
-      drmUnmap( pMga->buffers.map, pMga->buffers.size );
-      pMga->buffers.map = NULL;
-   }
-   if ( pMga->primary.map ) {
-      drmUnmap( pMga->primary.map, pMga->primary.size );
-      pMga->primary.map = NULL;
-   }
-   if ( pMga->warp.map ) {
-      drmUnmap( pMga->warp.map, pMga->warp.size );
-      pMga->warp.map = NULL;
-   }
-
-   if ( pMga->agpTextures.map ) {
-      drmUnmap( pMga->agpTextures.map, pMga->agpTextures.size );
-      pMga->agpTextures.map = NULL;
-   }
-
-   if ( pMga->agp.handle ) {
-      drmAgpUnbind( ctx->drmFD, pMga->agp.handle );
-      drmAgpFree( ctx->drmFD, pMga->agp.handle );
-      pMga->agp.handle = 0;
-      drmAgpRelease( ctx->drmFD );
-   }
-
-   DRICloseScreen( pScreen );
-
-   if ( pMga->pDRIInfo ) {
-      if ( pMga->pDRIpMga->devPrivate ) {
-	 xfree( pMga->pDRIpMga->devPrivate );
-	 pMga->pDRIpMga->devPrivate = 0;
-      }
-      DRIDestroyInfoRec( pMga->pDRIInfo );
-      pMga->pDRIInfo = 0;
-   }
-   if ( pMga->DRIServerInfo ) {
-      xfree( pMga->DRIServerInfo );
-      pMga->DRIServerInfo = 0;
-   }
-   if ( pMga->pVisualConfigs ) {
-      xfree( pMga->pVisualConfigs );
-   }
-   if ( pMga->pVisualConfigsPriv ) {
-      xfree( pMga->pVisualConfigsPriv );
-   }
-}
-#endif
diff --git a/src/mesa/drivers/dri/nouveau/Makefile b/src/mesa/drivers/dri/nouveau/Makefile
index 49e89335614..7be19b26fda 100644
--- a/src/mesa/drivers/dri/nouveau/Makefile
+++ b/src/mesa/drivers/dri/nouveau/Makefile
@@ -8,8 +8,6 @@ DRI_LIB_DEPS += $(shell pkg-config libdrm_nouveau --libs)
 
 LIBNAME = nouveau_vieux_dri.so
 
-MINIGLX_SOURCES =
-
 DRIVER_SOURCES = \
 	nouveau_screen.c \
 	nouveau_context.c \
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_context.c b/src/mesa/drivers/dri/nouveau/nouveau_context.c
index 52185a2fb90..502e01255cb 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_context.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_context.c
@@ -43,19 +43,22 @@
 
 #define need_GL_EXT_framebuffer_object
 #define need_GL_EXT_fog_coord
+#define need_GL_EXT_secondary_color
 
 #include "main/remap_helper.h"
 
 static const struct dri_extension nouveau_extensions[] = {
 	{ "GL_ARB_multitexture",	NULL },
+	{ "GL_ARB_texture_env_add",	NULL },
 	{ "GL_ARB_texture_env_combine",	NULL },
 	{ "GL_ARB_texture_env_dot3",	NULL },
-	{ "GL_ARB_texture_env_add",	NULL },
-	{ "GL_EXT_texture_lod_bias",	NULL },
-	{ "GL_EXT_framebuffer_object",	GL_EXT_framebuffer_object_functions },
 	{ "GL_ARB_texture_mirrored_repeat", NULL },
-	{ "GL_EXT_stencil_wrap",	NULL },
 	{ "GL_EXT_fog_coord",		GL_EXT_fog_coord_functions },
+	{ "GL_EXT_framebuffer_object",	GL_EXT_framebuffer_object_functions },
+	{ "GL_EXT_secondary_color",	GL_EXT_secondary_color_functions },
+	{ "GL_EXT_stencil_wrap",	NULL },
+	{ "GL_EXT_texture_lod_bias",	NULL },
+	{ "GL_NV_blend_square",         NULL },
 	{ "GL_SGIS_generate_mipmap",	NULL },
 	{ NULL,				NULL }
 };
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_state.c b/src/mesa/drivers/dri/nouveau/nouveau_state.c
index e1871db0eb9..bc610451b40 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_state.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_state.c
@@ -150,6 +150,7 @@ nouveau_enable(GLcontext *ctx, GLenum cap, GLboolean state)
 		break;
 	case GL_COLOR_SUM_EXT:
 		context_dirty(ctx, FRAG);
+		context_dirty(ctx, LIGHT_MODEL);
 		break;
 	case GL_CULL_FACE:
 		context_dirty(ctx, CULL_FACE);
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
index 69a9b96f0ca..a365b977f29 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
@@ -281,14 +281,13 @@ vbo_maybe_split(GLcontext *ctx, const struct gl_client_array **arrays,
 	int stride;
 
 	/* Try to keep client buffers smaller than the scratch BOs. */
-	if (!ib && render->mode == VBO &&
+	if (render->mode == VBO &&
 	    (stride = get_max_client_stride(ctx)))
 		    vert_avail = MIN2(vert_avail,
 				      RENDER_SCRATCH_SIZE / stride);
 
-
-	if ((ib && ib->count > idx_avail) ||
-	    (!ib && max_index - min_index > vert_avail)) {
+	if (max_index - min_index > vert_avail ||
+	    (ib && ib->count > idx_avail)) {
 		struct split_limits limits = {
 			.max_verts = vert_avail,
 			.max_indices = idx_avail,
diff --git a/src/mesa/drivers/dri/nouveau/nv04_state_fb.c b/src/mesa/drivers/dri/nouveau/nv04_state_fb.c
index aad1e491d2f..5e5e0c58742 100644
--- a/src/mesa/drivers/dri/nouveau/nv04_state_fb.c
+++ b/src/mesa/drivers/dri/nouveau/nv04_state_fb.c
@@ -63,7 +63,7 @@ nv04_emit_framebuffer(GLcontext *ctx, int emit)
 		return;
 
 	/* Render target */
-	if (fb->_NumColorDrawBuffers) {
+	if (fb->_ColorDrawBuffers[0]) {
 		s = &to_nouveau_renderbuffer(
 			fb->_ColorDrawBuffers[0])->surface;
 
diff --git a/src/mesa/drivers/dri/nouveau/nv04_state_raster.c b/src/mesa/drivers/dri/nouveau/nv04_state_raster.c
index 4314fc33cf4..c191571a5f8 100644
--- a/src/mesa/drivers/dri/nouveau/nv04_state_raster.c
+++ b/src/mesa/drivers/dri/nouveau/nv04_state_raster.c
@@ -275,6 +275,10 @@ nv04_emit_blend(GLcontext *ctx, int emit)
 		else
 			blend |= NV04_MULTITEX_TRIANGLE_BLEND_SHADE_MODE_FLAT;
 
+		/* Secondary color */
+		if (NEED_SECONDARY_COLOR(ctx))
+			blend |= NV04_MULTITEX_TRIANGLE_BLEND_SPECULAR_ENABLE;
+
 		/* Fog. */
 		if (ctx->Fog.Enabled)
 			blend |= NV04_MULTITEX_TRIANGLE_BLEND_FOG_ENABLE;
@@ -309,6 +313,10 @@ nv04_emit_blend(GLcontext *ctx, int emit)
 		else
 			blend |= get_texenv_mode(GL_MODULATE);
 
+		/* Secondary color */
+		if (NEED_SECONDARY_COLOR(ctx))
+			blend |= NV04_TEXTURED_TRIANGLE_BLEND_SPECULAR_ENABLE;
+
 		/* Fog. */
 		if (ctx->Fog.Enabled)
 			blend |= NV04_TEXTURED_TRIANGLE_BLEND_FOG_ENABLE;
diff --git a/src/mesa/drivers/dri/nouveau/nv10_state_fb.c b/src/mesa/drivers/dri/nouveau/nv10_state_fb.c
index 05c36b4f8f5..6bd383ebcd3 100644
--- a/src/mesa/drivers/dri/nouveau/nv10_state_fb.c
+++ b/src/mesa/drivers/dri/nouveau/nv10_state_fb.c
@@ -111,7 +111,7 @@ nv10_emit_framebuffer(GLcontext *ctx, int emit)
 	}
 
 	/* Render target */
-	if (fb->_NumColorDrawBuffers) {
+	if (fb->_ColorDrawBuffers[0]) {
 		s = &to_nouveau_renderbuffer(
 			fb->_ColorDrawBuffers[0])->surface;
 
diff --git a/src/mesa/drivers/dri/nouveau/nv10_state_tnl.c b/src/mesa/drivers/dri/nouveau/nv10_state_tnl.c
index 6db14d83b83..406e24c455d 100644
--- a/src/mesa/drivers/dri/nouveau/nv10_state_tnl.c
+++ b/src/mesa/drivers/dri/nouveau/nv10_state_tnl.c
@@ -201,8 +201,10 @@ nv10_emit_light_model(GLcontext *ctx, int emit)
 	BEGIN_RING(chan, celsius, NV10TCL_LIGHT_MODEL, 1);
 	OUT_RING(chan, ((m->LocalViewer ?
 			 NV10TCL_LIGHT_MODEL_LOCAL_VIEWER : 0) |
-			(m->ColorControl == GL_SEPARATE_SPECULAR_COLOR ?
-			 NV10TCL_LIGHT_MODEL_SEPARATE_SPECULAR : 0)));
+			(NEED_SECONDARY_COLOR(ctx) ?
+			 NV10TCL_LIGHT_MODEL_SEPARATE_SPECULAR : 0) |
+			(!ctx->Light.Enabled && ctx->Fog.ColorSumEnabled ?
+			 NV10TCL_LIGHT_MODEL_VERTEX_SPECULAR : 0)));
 }
 
 static float
diff --git a/src/mesa/drivers/dri/nouveau/nv20_state_fb.c b/src/mesa/drivers/dri/nouveau/nv20_state_fb.c
index 869acd6e317..d638541df9e 100644
--- a/src/mesa/drivers/dri/nouveau/nv20_state_fb.c
+++ b/src/mesa/drivers/dri/nouveau/nv20_state_fb.c
@@ -67,7 +67,7 @@ nv20_emit_framebuffer(GLcontext *ctx, int emit)
 		return;
 
 	/* Render target */
-	if (fb->_NumColorDrawBuffers) {
+	if (fb->_ColorDrawBuffers[0]) {
 		s = &to_nouveau_renderbuffer(
 			fb->_ColorDrawBuffers[0])->surface;
 
diff --git a/src/mesa/drivers/dri/nouveau/nv20_state_tnl.c b/src/mesa/drivers/dri/nouveau/nv20_state_tnl.c
index 0d566064f67..43f8c723122 100644
--- a/src/mesa/drivers/dri/nouveau/nv20_state_tnl.c
+++ b/src/mesa/drivers/dri/nouveau/nv20_state_tnl.c
@@ -158,7 +158,7 @@ nv20_emit_light_model(GLcontext *ctx, int emit)
 	OUT_RING(chan, ((m->LocalViewer ?
 			 NV20TCL_LIGHT_MODEL_VIEWER_LOCAL :
 			 NV20TCL_LIGHT_MODEL_VIEWER_NONLOCAL) |
-			(m->ColorControl == GL_SEPARATE_SPECULAR_COLOR ?
+			(NEED_SECONDARY_COLOR(ctx) ?
 			 NV20TCL_LIGHT_MODEL_SEPARATE_SPECULAR :
 			 0)));
 
diff --git a/src/mesa/drivers/dri/r128/Makefile b/src/mesa/drivers/dri/r128/Makefile
index 52c5a38a705..8144c9b43ff 100644
--- a/src/mesa/drivers/dri/r128/Makefile
+++ b/src/mesa/drivers/dri/r128/Makefile
@@ -5,8 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = r128_dri.so
 
-MINIGLX_SOURCES = server/r128_dri.c 
-
 DRIVER_SOURCES = \
 	r128_context.c \
 	r128_lock.c \
diff --git a/src/mesa/drivers/dri/r128/server/r128_dri.c b/src/mesa/drivers/dri/r128/server/r128_dri.c
deleted file mode 100644
index 6e3db948af7..00000000000
--- a/src/mesa/drivers/dri/r128/server/r128_dri.c
+++ /dev/null
@@ -1,1112 +0,0 @@
-/*
- * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
- *                      Precision Insight, Inc., Cedar Park, Texas, and
- *                      VA Linux Systems Inc., Fremont, California.
- *
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining
- * a copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation on the rights to use, copy, modify, merge,
- * publish, distribute, sublicense, and/or sell copies of the Software,
- * and to permit persons to whom the Software is furnished to do so,
- * subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial
- * portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
- * SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- */
-
-/*
- * Authors:
- *   Kevin E. Martin <martin@valinux.com>
- *   Rickard E. Faith <faith@valinux.com>
- *   Daryll Strauss <daryll@valinux.com>
- *   Gareth Hughes <gareth@valinux.com>
- *
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
-// Fix this to use kernel pci_ids.h when all of these IDs make it into the kernel 
-#include "pci_ids.h"
-
-#include "driver.h"
-#include "drm.h"
-#include "memops.h"
-
-#include "r128.h"
-#include "r128_dri.h"
-#include "r128_macros.h"
-#include "r128_reg.h"
-#include "r128_version.h"
-#include "r128_drm.h"
-
-static size_t r128_drm_page_size;
-
-/* Compute log base 2 of val. */
-static int R128MinBits(int val)
-{
-    int bits;
-
-    if (!val) return 1;
-    for (bits = 0; val; val >>= 1, ++bits);
-    return bits;
-}
-
-/* Initialize the AGP state.  Request memory for use in AGP space, and
-   initialize the Rage 128 registers to point to that memory. */
-static GLboolean R128DRIAgpInit(const DRIDriverContext *ctx)
-{
-    unsigned char *R128MMIO = ctx->MMIOAddress;
-    R128InfoPtr info = ctx->driverPrivate;
-    unsigned long mode;
-    unsigned int  vendor, device;
-    int           ret;
-    unsigned long cntl, chunk;
-    int           s, l;
-    int           flags;
-    unsigned long agpBase;
-
-    if (drmAgpAcquire(ctx->drmFD) < 0) {
-	fprintf(stderr, "[agp] AGP not available\n");
-	return GL_FALSE;
-    }
-
-				/* Modify the mode if the default mode is
-				   not appropriate for this particular
-				   combination of graphics card and AGP
-				   chipset. */
-
-    mode   = drmAgpGetMode(ctx->drmFD);        /* Default mode */
-    vendor = drmAgpVendorId(ctx->drmFD);
-    device = drmAgpDeviceId(ctx->drmFD);
-
-    mode &= ~R128_AGP_MODE_MASK;
-    switch (info->agpMode) {
-    case 4:          mode |= R128_AGP_4X_MODE;
-    case 2:          mode |= R128_AGP_2X_MODE;
-    case 1: default: mode |= R128_AGP_1X_MODE;
-    }
-
-    fprintf(stderr,
-	       "[agp] Mode 0x%08lx [AGP 0x%04x/0x%04x; Card 0x%04x/0x%04x]\n",
-	       mode, vendor, device,
-	       0x1002,
-	       info->Chipset);
-
-    if (drmAgpEnable(ctx->drmFD, mode) < 0) {
-	fprintf(stderr, "[agp] AGP not enabled\n");
-	drmAgpRelease(ctx->drmFD);
-	return GL_FALSE;
-    }
-
-    info->agpOffset = 0;
-
-    if ((ret = drmAgpAlloc(ctx->drmFD, info->agpSize*1024*1024, 0, NULL,
-			   &info->agpMemHandle)) < 0) {
-	fprintf(stderr, "[agp] Out of memory (%d)\n", ret);
-	drmAgpRelease(ctx->drmFD);
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] %d kB allocated with handle 0x%08x\n",
-	       info->agpSize*1024, info->agpMemHandle);
-
-    if (drmAgpBind(ctx->drmFD, info->agpMemHandle, info->agpOffset) < 0) {
-	fprintf(stderr, "[agp] Could not bind\n");
-	drmAgpFree(ctx->drmFD, info->agpMemHandle);
-	drmAgpRelease(ctx->drmFD);
-	return GL_FALSE;
-    }
-
-				/* Initialize the CCE ring buffer data */
-    info->ringStart       = info->agpOffset;
-    info->ringMapSize     = info->ringSize*1024*1024 + r128_drm_page_size;
-    info->ringSizeLog2QW  = R128MinBits(info->ringSize*1024*1024/8) - 1;
-
-    info->ringReadOffset  = info->ringStart + info->ringMapSize;
-    info->ringReadMapSize = r128_drm_page_size;
-
-				/* Reserve space for vertex/indirect buffers */
-    info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
-    info->bufMapSize      = info->bufSize*1024*1024;
-
-				/* Reserve the rest for AGP textures */
-    info->agpTexStart     = info->bufStart + info->bufMapSize;
-    s = (info->agpSize*1024*1024 - info->agpTexStart);
-    l = R128MinBits((s-1) / R128_NR_TEX_REGIONS);
-    if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
-    info->agpTexMapSize   = (s >> l) << l;
-    info->log2AGPTexGran  = l;
-
-    if (info->CCESecure) flags = DRM_READ_ONLY;
-    else                  flags = 0;
-
-    if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
-		  DRM_AGP, flags, &info->ringHandle) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not add ring mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] ring handle = 0x%08x\n", info->ringHandle);
-
-    if (drmMap(ctx->drmFD, info->ringHandle, info->ringMapSize,
-	       (drmAddressPtr)&info->ring) < 0) {
-	fprintf(stderr, "[agp] Could not map ring\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] Ring mapped at 0x%08lx\n",
-	       (unsigned long)info->ring);
-
-    if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
-		  DRM_AGP, flags, &info->ringReadPtrHandle) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not add ring read ptr mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
- 	       "[agp] ring read ptr handle = 0x%08x\n",
-	       info->ringReadPtrHandle);
-
-    if (drmMap(ctx->drmFD, info->ringReadPtrHandle, info->ringReadMapSize,
-	       (drmAddressPtr)&info->ringReadPtr) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not map ring read ptr\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] Ring read ptr mapped at 0x%08lx\n",
-	       (unsigned long)info->ringReadPtr);
-
-    if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
-		  DRM_AGP, 0, &info->bufHandle) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not add vertex/indirect buffers mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] vertex/indirect buffers handle = 0x%08lx\n",
-	       info->bufHandle);
-
-    if (drmMap(ctx->drmFD, info->bufHandle, info->bufMapSize,
-	       (drmAddressPtr)&info->buf) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not map vertex/indirect buffers\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] Vertex/indirect buffers mapped at 0x%08lx\n",
-	       (unsigned long)info->buf);
-
-    if (drmAddMap(ctx->drmFD, info->agpTexStart, info->agpTexMapSize,
-		  DRM_AGP, 0, &info->agpTexHandle) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not add AGP texture map mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] AGP texture map handle = 0x%08lx\n",
-	       info->agpTexHandle);
-
-    if (drmMap(ctx->drmFD, info->agpTexHandle, info->agpTexMapSize,
-	       (drmAddressPtr)&info->agpTex) < 0) {
-	fprintf(stderr,
-		   "[agp] Could not map AGP texture map\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[agp] AGP Texture map mapped at 0x%08lx\n",
-	       (unsigned long)info->agpTex);
-
-				/* Initialize Rage 128's AGP registers */
-    cntl  = INREG(R128_AGP_CNTL);
-    cntl &= ~R128_AGP_APER_SIZE_MASK;
-    switch (info->agpSize) {
-    case 256: cntl |= R128_AGP_APER_SIZE_256MB; break;
-    case 128: cntl |= R128_AGP_APER_SIZE_128MB; break;
-    case  64: cntl |= R128_AGP_APER_SIZE_64MB;  break;
-    case  32: cntl |= R128_AGP_APER_SIZE_32MB;  break;
-    case  16: cntl |= R128_AGP_APER_SIZE_16MB;  break;
-    case   8: cntl |= R128_AGP_APER_SIZE_8MB;   break;
-    case   4: cntl |= R128_AGP_APER_SIZE_4MB;   break;
-    default:
-	fprintf(stderr,
-		   "[agp] Illegal aperture size %d kB\n",
-		   info->agpSize*1024);
-	return GL_FALSE;
-    }
-    agpBase = drmAgpBase(ctx->drmFD);
-    OUTREG(R128_AGP_BASE, agpBase); 
-    OUTREG(R128_AGP_CNTL, cntl);
-
-				/* Disable Rage 128's PCIGART registers */
-    chunk = INREG(R128_BM_CHUNK_0_VAL);
-    chunk &= ~(R128_BM_PTR_FORCE_TO_PCI |
-	       R128_BM_PM4_RD_FORCE_TO_PCI |
-	       R128_BM_GLOBAL_FORCE_TO_PCI);
-    OUTREG(R128_BM_CHUNK_0_VAL, chunk);
-
-    OUTREG(R128_PCI_GART_PAGE, 1); /* Ensure AGP GART is used (for now) */
-
-    return GL_TRUE;
-}
-
-static GLboolean R128DRIPciInit(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    unsigned char *R128MMIO = ctx->MMIOAddress;
-    uint32_t chunk;
-    int ret;
-    int flags;
-
-    info->agpOffset = 0;
-
-    ret = drmScatterGatherAlloc(ctx->drmFD, info->agpSize*1024*1024,
-				&info->pciMemHandle);
-    if (ret < 0) {
-	fprintf(stderr, "[pci] Out of memory (%d)\n", ret);
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] %d kB allocated with handle 0x%08x\n",
-	       info->agpSize*1024, info->pciMemHandle);
-
-				/* Initialize the CCE ring buffer data */
-    info->ringStart       = info->agpOffset;
-    info->ringMapSize     = info->ringSize*1024*1024 + r128_drm_page_size;
-    info->ringSizeLog2QW  = R128MinBits(info->ringSize*1024*1024/8) - 1;
-
-    info->ringReadOffset  = info->ringStart + info->ringMapSize;
-    info->ringReadMapSize = r128_drm_page_size;
-
-				/* Reserve space for vertex/indirect buffers */
-    info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
-    info->bufMapSize      = info->bufSize*1024*1024;
-
-    flags = DRM_READ_ONLY | DRM_LOCKED | DRM_KERNEL;
-
-    if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
-		  DRM_SCATTER_GATHER, flags, &info->ringHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add ring mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] ring handle = 0x%08lx\n", info->ringHandle);
-
-    if (drmMap(ctx->drmFD, info->ringHandle, info->ringMapSize,
-	       (drmAddressPtr)&info->ring) < 0) {
-	fprintf(stderr, "[pci] Could not map ring\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] Ring mapped at 0x%08lx\n",
-	       (unsigned long)info->ring);
-    fprintf(stderr,
-	       "[pci] Ring contents 0x%08lx\n",
-	       *(unsigned long *)info->ring);
-
-    if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
-		  DRM_SCATTER_GATHER, flags, &info->ringReadPtrHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add ring read ptr mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] ring read ptr handle = 0x%08lx\n",
-	       info->ringReadPtrHandle);
-
-    if (drmMap(ctx->drmFD, info->ringReadPtrHandle, info->ringReadMapSize,
-	       (drmAddressPtr)&info->ringReadPtr) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not map ring read ptr\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] Ring read ptr mapped at 0x%08lx\n",
-	       (unsigned long)info->ringReadPtr);
-    fprintf(stderr,
-	       "[pci] Ring read ptr contents 0x%08lx\n",
-	       *(unsigned long *)info->ringReadPtr);
-
-    if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
-		  DRM_SCATTER_GATHER, 0, &info->bufHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add vertex/indirect buffers mapping\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] vertex/indirect buffers handle = 0x%08lx\n",
-	       info->bufHandle);
-
-    if (drmMap(ctx->drmFD, info->bufHandle, info->bufMapSize,
-	       (drmAddressPtr)&info->buf) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not map vertex/indirect buffers\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[pci] Vertex/indirect buffers mapped at 0x%08lx\n",
-	       (unsigned long)info->buf);
-    fprintf(stderr,
-	       "[pci] Vertex/indirect buffers contents 0x%08lx\n",
-	       *(unsigned long *)info->buf);
-
-    if (!info->IsPCI) {
-	/* This is really an AGP card, force PCI GART mode */
-        chunk = INREG(R128_BM_CHUNK_0_VAL);
-        chunk |= (R128_BM_PTR_FORCE_TO_PCI |
-		  R128_BM_PM4_RD_FORCE_TO_PCI |
-		  R128_BM_GLOBAL_FORCE_TO_PCI);
-        OUTREG(R128_BM_CHUNK_0_VAL, chunk);
-        OUTREG(R128_PCI_GART_PAGE, 0); /* Ensure PCI GART is used */
-    }
-
-    return GL_TRUE;
-}
-
-/* Add a map for the MMIO registers that will be accessed by any
-   DRI-based clients. */
-static GLboolean R128DRIMapInit(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    int flags;
-
-    if (info->CCESecure) flags = DRM_READ_ONLY;
-    else                 flags = 0;
-
-				/* Map registers */
-    if (drmAddMap(ctx->drmFD, ctx->MMIOStart, ctx->MMIOSize,
-		  DRM_REGISTERS, flags, &info->registerHandle) < 0) {
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[drm] register handle = 0x%08x\n", info->registerHandle);
-
-    return GL_TRUE;
-}
-
-/* Initialize the kernel data structures. */
-static int R128DRIKernelInit(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    drm_r128_init_t drmInfo;
-
-    memset( &drmInfo, 0, sizeof(&drmInfo) );
-
-    drmInfo.func                = R128_INIT_CCE;
-    drmInfo.sarea_priv_offset   = sizeof(drm_sarea_t);
-    drmInfo.is_pci              = info->IsPCI;
-    drmInfo.cce_mode            = info->CCEMode;
-    drmInfo.cce_secure          = info->CCESecure;
-    drmInfo.ring_size           = info->ringSize*1024*1024;
-    drmInfo.usec_timeout        = info->CCEusecTimeout;
-
-    drmInfo.fb_bpp              = ctx->bpp;
-    drmInfo.depth_bpp           = ctx->bpp;
-
-    drmInfo.front_offset        = info->frontOffset;
-    drmInfo.front_pitch         = info->frontPitch;
-
-    drmInfo.back_offset         = info->backOffset;
-    drmInfo.back_pitch          = info->backPitch;
-
-    drmInfo.depth_offset        = info->depthOffset;
-    drmInfo.depth_pitch         = info->depthPitch;
-    drmInfo.span_offset         = info->spanOffset;
-
-    drmInfo.fb_offset           = info->LinearAddr;
-    drmInfo.mmio_offset         = info->registerHandle;
-    drmInfo.ring_offset         = info->ringHandle;
-    drmInfo.ring_rptr_offset    = info->ringReadPtrHandle;
-    drmInfo.buffers_offset      = info->bufHandle;
-    drmInfo.agp_textures_offset = info->agpTexHandle;
-
-    if (drmCommandWrite(ctx->drmFD, DRM_R128_INIT,
-                        &drmInfo, sizeof(drmInfo)) < 0)
-        return GL_FALSE;
-
-    return GL_TRUE;
-}
-
-/* Add a map for the vertex buffers that will be accessed by any
-   DRI-based clients. */
-static GLboolean R128DRIBufInit(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-				/* Initialize vertex buffers */
-    if (info->IsPCI) {
-	info->bufNumBufs = drmAddBufs(ctx->drmFD,
-				      info->bufMapSize / R128_BUFFER_SIZE,
-				      R128_BUFFER_SIZE,
-				      DRM_SG_BUFFER,
-				      info->bufStart);
-    } else {
-	info->bufNumBufs = drmAddBufs(ctx->drmFD,
-				      info->bufMapSize / R128_BUFFER_SIZE,
-				      R128_BUFFER_SIZE,
-				      DRM_AGP_BUFFER,
-				      info->bufStart);
-    }
-    if (info->bufNumBufs <= 0) {
-	fprintf(stderr,
-		   "[drm] Could not create vertex/indirect buffers list\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[drm] Added %d %d byte vertex/indirect buffers\n",
-	       info->bufNumBufs, R128_BUFFER_SIZE);
-
-    if (!(info->buffers = drmMapBufs(ctx->drmFD))) {
-	fprintf(stderr,
-		   "[drm] Failed to map vertex/indirect buffers list\n");
-	return GL_FALSE;
-    }
-    fprintf(stderr,
-	       "[drm] Mapped %d vertex/indirect buffers\n",
-	       info->buffers->count);
-
-    return GL_TRUE;
-}
-
-static void R128DRIIrqInit(const DRIDriverContext *ctx)
-{
-   R128InfoPtr info = ctx->driverPrivate;
-   unsigned char *R128MMIO = ctx->MMIOAddress;
-   
-   if (!info->irq) {
-       info->irq = drmGetInterruptFromBusID(
-	   ctx->drmFD,
-	   ctx->pciBus,
-	   ctx->pciDevice,
-	   ctx->pciFunc);
-
-      if((drmCtlInstHandler(ctx->drmFD, info->irq)) != 0) {
-	 fprintf(stderr,
-		    "[drm] failure adding irq handler, "
-		    "there is a device already using that irq\n"
-		    "[drm] falling back to irq-free operation\n");
-	 info->irq = 0;
-      } else {
-          info->gen_int_cntl = INREG( R128_GEN_INT_CNTL );
-      }
-   }
-
-   if (info->irq)
-      fprintf(stderr,
-		 "[drm] dma control initialized, using IRQ %d\n",
-		 info->irq);
-}
-
-static int R128CCEStop(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    drm_r128_cce_stop_t stop;
-    int            ret, i;
-
-    stop.flush = 1;
-    stop.idle  = 1;
-
-    ret = drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
-                           &stop, sizeof(stop) );
-
-    if ( ret == 0 ) {
-        return 0;
-    } else if ( errno != EBUSY ) {
-        return -errno;
-    }
-
-    stop.flush = 0;
-
-    i = 0;
-    do {
-        ret = drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
-                               &stop, sizeof(stop) );
-    } while ( ret && errno == EBUSY && i++ < R128_IDLE_RETRY );
-
-    if ( ret == 0 ) {
-        return 0;
-    } else if ( errno != EBUSY ) {
-        return -errno;
-    }
-
-    stop.idle = 0;
-
-    if ( drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
-                          &stop, sizeof(stop) )) {
-        return -errno;
-    } else {
-        return 0;
-    }
-}
-
-/* Initialize the CCE state, and start the CCE (if used by the X server) */
-static void R128DRICCEInit(const DRIDriverContext *ctx)
-{
-   R128InfoPtr info = ctx->driverPrivate;
-
-				/* Turn on bus mastering */
-    info->BusCntl &= ~R128_BUS_MASTER_DIS;
-
-				/* CCEMode is initialized in r128_driver.c */
-    switch (info->CCEMode) {
-    case R128_PM4_NONPM4:                 info->CCEFifoSize = 0;   break;
-    case R128_PM4_192PIO:                 info->CCEFifoSize = 192; break;
-    case R128_PM4_192BM:                  info->CCEFifoSize = 192; break;
-    case R128_PM4_128PIO_64INDBM:         info->CCEFifoSize = 128; break;
-    case R128_PM4_128BM_64INDBM:          info->CCEFifoSize = 128; break;
-    case R128_PM4_64PIO_128INDBM:         info->CCEFifoSize = 64;  break;
-    case R128_PM4_64BM_128INDBM:          info->CCEFifoSize = 64;  break;
-    case R128_PM4_64PIO_64VCBM_64INDBM:   info->CCEFifoSize = 64;  break;
-    case R128_PM4_64BM_64VCBM_64INDBM:    info->CCEFifoSize = 64;  break;
-    case R128_PM4_64PIO_64VCPIO_64INDPIO: info->CCEFifoSize = 64;  break;
-    }
-
-    /* Make sure the CCE is on for the X server */
-    R128CCE_START(ctx, info);
-}
-
-
-static int R128MemoryInit(const DRIDriverContext *ctx)
-{
-   R128InfoPtr info = ctx->driverPrivate;
-   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
-   int        cpp         = ctx->cpp;
-   int        bufferSize  = ((ctx->shared.virtualHeight * width_bytes
-			      + R128_BUFFER_ALIGN)
-			     & ~R128_BUFFER_ALIGN);
-   int        depthSize   = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes
-			      + R128_BUFFER_ALIGN)
-			     & ~R128_BUFFER_ALIGN);
-   int        l;
-
-   info->frontOffset = 0;
-   info->frontPitch = ctx->shared.virtualWidth;
-
-   fprintf(stderr, 
-	   "Using %d MB AGP aperture\n", info->agpSize);
-   fprintf(stderr, 
-	   "Using %d MB for the ring buffer\n", info->ringSize);
-   fprintf(stderr, 
-	   "Using %d MB for vertex/indirect buffers\n", info->bufSize);
-   fprintf(stderr, 
-	   "Using %d MB for AGP textures\n", info->agpTexSize);
-
-   /* Front, back and depth buffers - everything else texture??
-    */
-   info->textureSize = ctx->shared.fbSize - 2 * bufferSize - depthSize;
-
-   if (info->textureSize < 0) 
-      return 0;
-
-   l = R128MinBits((info->textureSize-1) / R128_NR_TEX_REGIONS);
-   if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
-
-   /* Round the texture size up to the nearest whole number of
-    * texture regions.  Again, be greedy about this, don't
-    * round down.
-    */
-   info->log2TexGran = l;
-   info->textureSize = (info->textureSize >> l) << l;
-
-   /* Set a minimum usable local texture heap size.  This will fit
-    * two 256x256x32bpp textures.
-    */
-   if (info->textureSize < 512 * 1024) {
-      info->textureOffset = 0;
-      info->textureSize = 0;
-   }
-
-   /* Reserve space for textures */
-   info->textureOffset = ((ctx->shared.fbSize - info->textureSize +
-			   R128_BUFFER_ALIGN) &
-			  ~R128_BUFFER_ALIGN);
-
-   /* Reserve space for the shared depth
-    * buffer.
-    */
-   info->depthOffset = ((info->textureOffset - depthSize +
-			 R128_BUFFER_ALIGN) &
-			~R128_BUFFER_ALIGN);
-   info->depthPitch = ctx->shared.virtualWidth;
-
-   info->backOffset = ((info->depthOffset - bufferSize +
-			R128_BUFFER_ALIGN) &
-		       ~R128_BUFFER_ALIGN);
-   info->backPitch = ctx->shared.virtualWidth;
-
-
-   fprintf(stderr, 
-	   "Will use back buffer at offset 0x%x\n",
-	   info->backOffset);
-   fprintf(stderr, 
-	   "Will use depth buffer at offset 0x%x\n",
-	   info->depthOffset);
-   fprintf(stderr, 
-	   "Will use %d kb for textures at offset 0x%x\n",
-	   info->textureSize/1024, info->textureOffset);
-
-   return 1;
-} 
-
-
-/* Initialize the screen-specific data structures for the DRI and the
-   Rage 128.  This is the main entry point to the device-specific
-   initialization code.  It calls device-independent DRI functions to
-   create the DRI data structures and initialize the DRI state. */
-static GLboolean R128DRIScreenInit(DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    R128DRIPtr    pR128DRI;
-    int           err, major, minor, patch;
-    drmVersionPtr version;
-    drm_r128_sarea_t *pSAREAPriv;
-
-    switch (ctx->bpp) {
-    case 8:
-	/* These modes are not supported (yet). */
-    case 15:
-    case 24:
-	fprintf(stderr,
-		   "[dri] R128DRIScreenInit failed (depth %d not supported).  "
-		   "[dri] Disabling DRI.\n", ctx->bpp);
-	return GL_FALSE;
-
-	/* Only 16 and 32 color depths are supports currently. */
-    case 16:
-    case 32:
-	break;
-    }
-    r128_drm_page_size = getpagesize();
-    
-    info->registerSize = ctx->MMIOSize;
-    ctx->shared.SAREASize = SAREA_MAX;
-
-    /* Note that drmOpen will try to load the kernel module, if needed. */
-    ctx->drmFD = drmOpen("r128", NULL );
-    if (ctx->drmFD < 0) {
-	fprintf(stderr, "[drm] drmOpen failed\n");
-	return 0;
-    }
-    
-    /* Check the r128 DRM version */
-    version = drmGetVersion(ctx->drmFD);
-    if (version) {
-	if (version->version_major != 2 ||
-	    version->version_minor < 2) {
-	    /* incompatible drm version */
-	    fprintf(stderr,
-		"[dri] R128DRIScreenInit failed because of a version mismatch.\n"
-		"[dri] r128.o kernel module version is %d.%d.%d but version 2.2 or greater is needed.\n"
-		"[dri] Disabling the DRI.\n",
-		version->version_major,
-		version->version_minor,
-		version->version_patchlevel);
-	    drmFreeVersion(version);
-	    return GL_FALSE;
-	}
-	info->drmMinor = version->version_minor;
-	drmFreeVersion(version);
-    }
-    
-    if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-	fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-		ctx->drmFD, ctx->pciBusID, strerror(-err));
-	return 0;
-    }
-    
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   }
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-   
-   /* Need to AddMap the framebuffer and mmio regions here:
-    */
-   if (drmAddMap( ctx->drmFD,
-		  (drm_handle_t)ctx->FBStart,
-		  ctx->FBSize,
-		  DRM_FRAME_BUFFER,
-		  0,
-		  &ctx->shared.hFrameBuffer) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-      return 0;
-   }
-
-   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-	   ctx->shared.hFrameBuffer);
-
-   if (!R128MemoryInit(ctx))
-	return GL_FALSE;
-   
-				/* Initialize AGP */
-    if (!info->IsPCI && !R128DRIAgpInit(ctx)) {
-	info->IsPCI = GL_TRUE;
-	fprintf(stderr,
-		   "[agp] AGP failed to initialize -- falling back to PCI mode.\n");
-	fprintf(stderr,
-		   "[agp] Make sure you have the agpgart kernel module loaded.\n");
-    }
-
-				/* Initialize PCIGART */
-    if (info->IsPCI && !R128DRIPciInit(ctx)) {
-	return GL_FALSE;
-    }
-
-				/* DRIScreenInit doesn't add all the
-				   common mappings.  Add additional
-				   mappings here. */
-    if (!R128DRIMapInit(ctx)) {
-	return GL_FALSE;
-    }
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-    /* Initialize the kernel data structures */
-    if (!R128DRIKernelInit(ctx)) {
-	return GL_FALSE;
-    }
-
-    /* Initialize the vertex buffers list */
-    if (!R128DRIBufInit(ctx)) {
-	return GL_FALSE;
-    }
-
-    /* Initialize IRQ */
-    R128DRIIrqInit(ctx);
-
-    /* Initialize and start the CCE if required */
-    R128DRICCEInit(ctx);
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-   drimemsetio((char *)ctx->FBAddress + info->frontOffset,
-	  0,
-	  info->frontPitch * ctx->cpp * ctx->shared.virtualHeight );
-
-   drimemsetio((char *)ctx->FBAddress + info->backOffset,
-	  0,
-	  info->backPitch * ctx->cpp * ctx->shared.virtualHeight );
-    
-    pSAREAPriv = (drm_r128_sarea_t *)(((char*)ctx->pSAREA) + 
-					sizeof(drm_sarea_t));
-    memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
-
-   /* This is the struct passed to radeon_dri.so for its initialization */
-   ctx->driverClientMsg = malloc(sizeof(R128DRIRec));
-   ctx->driverClientMsgSize = sizeof(R128DRIRec);
-   
-    pR128DRI                    = (R128DRIPtr)ctx->driverClientMsg;
-    pR128DRI->deviceID          = info->Chipset;
-    pR128DRI->width             = ctx->shared.virtualWidth;
-    pR128DRI->height            = ctx->shared.virtualHeight;
-    pR128DRI->depth             = ctx->bpp;
-    pR128DRI->bpp               = ctx->bpp;
-
-    pR128DRI->IsPCI             = info->IsPCI;
-    pR128DRI->AGPMode           = info->agpMode;
-
-    pR128DRI->frontOffset       = info->frontOffset;
-    pR128DRI->frontPitch        = info->frontPitch;
-    pR128DRI->backOffset        = info->backOffset;
-    pR128DRI->backPitch         = info->backPitch;
-    pR128DRI->depthOffset       = info->depthOffset;
-    pR128DRI->depthPitch        = info->depthPitch;
-    pR128DRI->spanOffset        = info->spanOffset;
-    pR128DRI->textureOffset     = info->textureOffset;
-    pR128DRI->textureSize       = info->textureSize;
-    pR128DRI->log2TexGran       = info->log2TexGran;
-
-    pR128DRI->registerHandle    = info->registerHandle;
-    pR128DRI->registerSize      = info->registerSize;
-
-    pR128DRI->agpTexHandle      = info->agpTexHandle;
-    pR128DRI->agpTexMapSize     = info->agpTexMapSize;
-    pR128DRI->log2AGPTexGran    = info->log2AGPTexGran;
-    pR128DRI->agpTexOffset      = info->agpTexStart;
-    pR128DRI->sarea_priv_offset = sizeof(drm_sarea_t);
-
-    return GL_TRUE;
-}
-
-/* The screen is being closed, so clean up any state and free any
-   resources used by the DRI. */
-void R128DRICloseScreen(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-    drm_r128_init_t drmInfo;
-
-    /* Stop the CCE if it is still in use */
-    R128CCE_STOP(ctx, info);
-
-    if (info->irq) {
-	drmCtlUninstHandler(ctx->drmFD);
-	info->irq = 0;
-    }
-
-    /* De-allocate vertex buffers */
-    if (info->buffers) {
-	drmUnmapBufs(info->buffers);
-	info->buffers = NULL;
-    }
-
-    /* De-allocate all kernel resources */
-    memset(&drmInfo, 0, sizeof(drmInfo));
-    drmInfo.func = R128_CLEANUP_CCE;
-    drmCommandWrite(ctx->drmFD, DRM_R128_INIT,
-                    &drmInfo, sizeof(drmInfo));
-
-    /* De-allocate all AGP resources */
-    if (info->agpTex) {
-	drmUnmap(info->agpTex, info->agpTexMapSize);
-	info->agpTex = NULL;
-    }
-    if (info->buf) {
-	drmUnmap(info->buf, info->bufMapSize);
-	info->buf = NULL;
-    }
-    if (info->ringReadPtr) {
-	drmUnmap(info->ringReadPtr, info->ringReadMapSize);
-	info->ringReadPtr = NULL;
-    }
-    if (info->ring) {
-	drmUnmap(info->ring, info->ringMapSize);
-	info->ring = NULL;
-    }
-    if (info->agpMemHandle != DRM_AGP_NO_HANDLE) {
-	drmAgpUnbind(ctx->drmFD, info->agpMemHandle);
-	drmAgpFree(ctx->drmFD, info->agpMemHandle);
-	info->agpMemHandle = 0;
-	drmAgpRelease(ctx->drmFD);
-    }
-    if (info->pciMemHandle) {
-	drmScatterGatherFree(ctx->drmFD, info->pciMemHandle);
-	info->pciMemHandle = 0;
-    }
-}
-
-static GLboolean R128PreInitDRI(const DRIDriverContext *ctx)
-{
-    R128InfoPtr info = ctx->driverPrivate;
-
-    /*info->CCEMode = R128_DEFAULT_CCE_PIO_MODE;*/
-    info->CCEMode = R128_DEFAULT_CCE_BM_MODE;
-    info->CCESecure = GL_TRUE;
-
-    info->agpMode        = R128_DEFAULT_AGP_MODE;
-    info->agpSize        = R128_DEFAULT_AGP_SIZE;
-    info->ringSize       = R128_DEFAULT_RING_SIZE;
-    info->bufSize        = R128_DEFAULT_BUFFER_SIZE;
-    info->agpTexSize     = R128_DEFAULT_AGP_TEX_SIZE;
-
-    info->CCEusecTimeout = R128_DEFAULT_CCE_TIMEOUT;
-
-    return GL_TRUE;
-}
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Fills in \p info with some default values and some information from \p ctx
- * and then calls R128ScreenInit() for the screen initialization.
- * 
- * Before exiting clears the framebuffer memory accessing it directly.
- */
-static int R128InitFBDev( DRIDriverContext *ctx )
-{
-   R128InfoPtr info = calloc(1, sizeof(*info));
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      switch (ctx->bpp / 8) {
-      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-      case 3:
-      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-      }
-
-      ctx->shared.virtualWidth = dummy;
-   }
-
-   ctx->driverPrivate = (void *)info;
-   
-   info->Chipset = ctx->chipset;
-
-   switch (info->Chipset) {
-   case PCI_DEVICE_ID_ATI_RAGE128_LE:
-   case PCI_DEVICE_ID_ATI_RAGE128_RE:
-   case PCI_DEVICE_ID_ATI_RAGE128_RK:
-   case PCI_DEVICE_ID_ATI_RAGE128_PD:
-   case PCI_DEVICE_ID_ATI_RAGE128_PP:
-   case PCI_DEVICE_ID_ATI_RAGE128_PR:
-       /* This is a PCI card */
-       info->IsPCI = GL_TRUE;
-       break;
-   default:
-       /* This is an AGP card */
-       info->IsPCI = GL_FALSE;
-       break;
-   }
-
-   info->frontPitch = ctx->shared.virtualWidth;
-   info->LinearAddr = ctx->FBStart & 0xfc000000;
-   
-   if (!R128PreInitDRI(ctx))
-       return 0;
-
-   if (!R128DRIScreenInit(ctx))
-      return 0;
-
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void R128HaltFBDev( DRIDriverContext *ctx )
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-       free(ctx->driverPrivate);
-       ctx->driverPrivate = 0;
-    }
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa R128PostValidateMode().
- */
-static int R128ValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa R128ValidateMode().
- */
-static int R128PostValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Shutdown the drawing engine.
- *
- * \param ctx display handle
- *
- * Turns off the command processor engine & restores the graphics card
- * to a state that fbdev understands.
- */
-static int R128EngineShutdown( const DRIDriverContext *ctx )
-{
-    return 1;
-}
-
-/**
- * \brief Restore the drawing engine.
- *
- * \param ctx display handle
- *
- * Resets the graphics card and sets initial values for several registers of
- * the card's drawing engine.
- *
- * Turns on the R128 command processor engine (i.e., the ringbuffer).
- */
-static int R128EngineRestore( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-const struct DRIDriverRec __driDriver = {
-   R128ValidateMode,
-   R128PostValidateMode,
-   R128InitFBDev,
-   R128HaltFBDev,
-   R128EngineShutdown,
-   R128EngineRestore,  
-   0,
-};
diff --git a/src/mesa/drivers/dri/r200/Makefile b/src/mesa/drivers/dri/r200/Makefile
index 14eb96c1bab..c9c1346c3a2 100644
--- a/src/mesa/drivers/dri/r200/Makefile
+++ b/src/mesa/drivers/dri/r200/Makefile
@@ -7,8 +7,6 @@ CFLAGS += $(RADEON_CFLAGS)
 
 LIBNAME = r200_dri.so
 
-MINIGLX_SOURCES = server/radeon_dri.c 
-
 ifeq ($(RADEON_LDFLAGS),)
 CS_SOURCES = radeon_cs_space_drm.c radeon_bo.c radeon_cs.c
 endif
@@ -23,17 +21,19 @@ RADEON_COMMON_SOURCES = \
 	radeon_fbo.c \
 	radeon_lock.c \
 	radeon_mipmap_tree.c \
+	radeon_pixel_read.c \
 	radeon_queryobj.c \
 	radeon_span.c \
 	radeon_texture.c \
-	radeon_tex_copy.c
+	radeon_tex_copy.c \
+	radeon_tex_getimage.c \
+	radeon_tile.c
 
 DRIVER_SOURCES = r200_context.c \
 		 r200_ioctl.c \
 		 r200_state.c \
 		 r200_state_init.c \
 		 r200_cmdbuf.c \
-		 r200_pixel.c \
 		 r200_tex.c \
 		 r200_texstate.c \
 		 r200_tcl.c \
diff --git a/src/mesa/drivers/dri/r200/r200_blit.c b/src/mesa/drivers/dri/r200/r200_blit.c
index 30757600934..55e8d32512c 100644
--- a/src/mesa/drivers/dri/r200/r200_blit.c
+++ b/src/mesa/drivers/dri/r200/r200_blit.c
@@ -48,6 +48,8 @@ unsigned r200_check_blit(gl_format mesa_format)
     case MESA_FORMAT_ARGB4444:
     case MESA_FORMAT_ARGB1555:
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_L8:
+    case MESA_FORMAT_I8:
 	    break;
     default:
 	    return 0;
@@ -105,6 +107,12 @@ static void inline emit_tx_setup(struct r200_context *r200,
     case MESA_FORMAT_ARGB8888:
 	    txformat |= R200_TXFORMAT_ARGB8888 | R200_TXFORMAT_ALPHA_IN_MAP;
 	    break;
+    case MESA_FORMAT_RGBA8888:
+	    txformat |= R200_TXFORMAT_RGBA8888 | R200_TXFORMAT_ALPHA_IN_MAP;
+	    break;
+    case MESA_FORMAT_RGBA8888_REV:
+	    txformat |= R200_TXFORMAT_ABGR8888 | R200_TXFORMAT_ALPHA_IN_MAP;
+	    break;
     case MESA_FORMAT_XRGB8888:
 	    txformat |= R200_TXFORMAT_ARGB8888;
 	    break;
@@ -118,8 +126,15 @@ static void inline emit_tx_setup(struct r200_context *r200,
 	    txformat |= R200_TXFORMAT_ARGB1555 | R200_TXFORMAT_ALPHA_IN_MAP;
 	    break;
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_I8:
 	    txformat |= R200_TXFORMAT_I8 | R200_TXFORMAT_ALPHA_IN_MAP;
 	    break;
+    case MESA_FORMAT_L8:
+	    txformat |= R200_TXFORMAT_I8;
+	    break;
+    case MESA_FORMAT_AL88:
+	    txformat |= R200_TXFORMAT_AI88 | R200_TXFORMAT_ALPHA_IN_MAP;
+	    break;
     default:
 	    break;
     }
@@ -182,6 +197,8 @@ static inline void emit_cb_setup(struct r200_context *r200,
 	    dst_format = RADEON_COLOR_FORMAT_ARGB1555;
 	    break;
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_L8:
+    case MESA_FORMAT_I8:
 	    dst_format = RADEON_COLOR_FORMAT_RGB8;
 	    break;
     default:
diff --git a/src/mesa/drivers/dri/r200/r200_context.c b/src/mesa/drivers/dri/r200/r200_context.c
index dad2580e08b..5739443bdf2 100644
--- a/src/mesa/drivers/dri/r200/r200_context.c
+++ b/src/mesa/drivers/dri/r200/r200_context.c
@@ -51,7 +51,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_context.h"
 #include "r200_ioctl.h"
 #include "r200_state.h"
-#include "r200_pixel.h"
 #include "r200_tex.h"
 #include "r200_swtcl.h"
 #include "r200_tcl.h"
@@ -266,6 +265,7 @@ static void r200_init_vtbl(radeonContextPtr radeon)
    radeon->vtbl.emit_query_finish = r200_emit_query_finish;
    radeon->vtbl.check_blit = r200_check_blit;
    radeon->vtbl.blit = r200_blit;
+   radeon->vtbl.is_format_renderable = radeonIsFormatRenderable;
 }
 
 
@@ -473,7 +473,6 @@ GLboolean r200CreateContext( const __GLcontextModes *glVisual,
    /* XXX these should really go right after _mesa_init_driver_functions() */
    radeon_fbo_init(&rmesa->radeon);
    radeonInitSpanFuncs( ctx );
-   r200InitPixelFuncs( ctx );
    r200InitTnlFuncs( ctx );
    r200InitState( rmesa );
    r200InitSwtcl( ctx );
diff --git a/src/mesa/drivers/dri/r200/r200_pixel.c b/src/mesa/drivers/dri/r200/r200_pixel.c
deleted file mode 100644
index bfb7e2a2ed3..00000000000
--- a/src/mesa/drivers/dri/r200/r200_pixel.c
+++ /dev/null
@@ -1,494 +0,0 @@
-/*
-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>
- */
-
-#include "main/glheader.h"
-#include "main/enums.h"
-#include "main/mtypes.h"
-#include "main/macros.h"
-#include "swrast/swrast.h"
-
-#include "r200_context.h"
-#include "r200_ioctl.h"
-#include "r200_pixel.h"
-#include "r200_swtcl.h"
-
-#include "drirenderbuffer.h"
-
-
-static GLboolean
-check_color( const GLcontext *ctx, GLenum type, GLenum format,
-	     const struct gl_pixelstore_attrib *packing,
-	     const void *pixels, GLint sz, GLint pitch )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
-
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (	(pitch & 63) ||
-	ctx->_ImageTransferState ||
-	packing->SwapBytes ||
-	packing->LsbFirst) {
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "%s: failed 1\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if ( type == GL_UNSIGNED_INT_8_8_8_8_REV &&
-	cpp == 4 &&
-	format == GL_BGRA ) {
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
-      return GL_TRUE;
-   }
-
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s: failed\n", __FUNCTION__);
-
-   return GL_FALSE;
-}
-
-static GLboolean
-check_color_per_fragment_ops( const GLcontext *ctx )
-{
-   int result;
-   result = (!(     ctx->Color.AlphaEnabled ||
-		    ctx->Depth.Test ||
-		    ctx->Fog.Enabled ||
-		    ctx->Scissor.Enabled ||
-		    ctx->Stencil._Enabled ||
-		    !ctx->Color.ColorMask[0][0] ||
-		    !ctx->Color.ColorMask[0][1] ||
-		    !ctx->Color.ColorMask[0][2] ||
-		    !ctx->Color.ColorMask[0][3] ||
-		    ctx->Color.ColorLogicOpEnabled ||
-		    ctx->Texture._EnabledUnits
-           ) &&
-	   ctx->Current.RasterPosValid);
-
-   return result;
-}
-
-
-#if 0
-static GLboolean
-clip_pixelrect( const GLcontext *ctx,
-		const GLframebuffer *buffer,
-		GLint *x, GLint *y,
-		GLsizei *width, GLsizei *height,
-		GLint *size )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   /* left clipping */
-   if (*x < buffer->_Xmin) {
-      *width -= (buffer->_Xmin - *x);
-      *x = buffer->_Xmin;
-   }
-
-   /* right clipping */
-   if (*x + *width > buffer->_Xmax)
-      *width -= (*x + *width - buffer->_Xmax - 1);
-
-   if (*width <= 0)
-      return GL_FALSE;
-
-   /* bottom clipping */
-   if (*y < buffer->_Ymin) {
-      *height -= (buffer->_Ymin - *y);
-      *y = buffer->_Ymin;
-   }
-
-   /* top clipping */
-   if (*y + *height > buffer->_Ymax)
-      *height -= (*y + *height - buffer->_Ymax - 1);
-
-   if (*height <= 0)
-      return GL_FALSE;
-
-   *size = ((*y + *height - 1) * rmesa->radeon.radeonScreen->frontPitch +
-	    (*x + *width - 1) * rmesa->radeon.radeonScreen->cpp);
-
-   return GL_TRUE;
-}
-#endif
-
-static GLboolean
-r200TryReadPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *pack,
-		  GLvoid *pixels )
-{
-   return GL_FALSE;
-#if 0
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLint pitch = pack->RowLength ? pack->RowLength : width;
-   GLint blit_format;
-   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
-   GLint size = width * height * cpp;
-
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Only accelerate reading to GART buffers.
-    */
-   if ( !r200IsGartMemory(rmesa, pixels,
-			 pitch * height * rmesa->radeon.radeonScreen->cpp ) ) {
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "%s: dest not GART\n", __FUNCTION__);
-   }
-
-   /* Need GL_PACK_INVERT_MESA to cope with upsidedown results from
-    * blitter:
-    */
-   if (!pack->Invert) {
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "%s: MESA_PACK_INVERT not set\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if (!check_color(ctx, type, format, pack, pixels, size, pitch))
-      return GL_FALSE;
-
-   switch ( rmesa->radeon.radeonScreen->cpp ) {
-   case 4:
-      blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
-      break;
-   default:
-      return GL_FALSE;
-   }
-
-
-   /* Although the blits go on the command buffer, need to do this and
-    * fire with lock held to guarentee cliprects and drawOffset are
-    * correct.
-    *
-    * This is an unusual situation however, as the code which flushes
-    * a full command buffer expects to be called unlocked.  As a
-    * workaround, immediately flush the buffer on aquiring the lock.
-    */
-   LOCK_HARDWARE( &rmesa->radeon );
-
-   if (rmesa->store.cmd_used)
-      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
-
-   if (!clip_pixelrect(ctx, ctx->ReadBuffer, &x, &y, &width, &height,
-		       &size)) {
-      UNLOCK_HARDWARE( &rmesa->radeon );
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "%s totally clipped -- nothing to do\n",
-		 __FUNCTION__);
-      return GL_TRUE;
-   }
-
-   {
-      __DRIdrawable *dPriv = rmesa->radeon.dri.drawable;
-      driRenderbuffer *drb = (driRenderbuffer *) ctx->ReadBuffer->_ColorReadBuffer;
-      int nbox = dPriv->numClipRects;
-      int src_offset = drb->offset
-		     + rmesa->radeon.radeonScreen->fbLocation;
-      int src_pitch = drb->pitch * drb->cpp;
-      int dst_offset = r200GartOffsetFromVirtual( rmesa, pixels );
-      int dst_pitch = pitch * rmesa->radeon.radeonScreen->cpp;
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      int i;
-
-      r200EmitWait( rmesa, RADEON_WAIT_3D );
-
-      y = dPriv->h - y - height;
-      x += dPriv->x;
-      y += dPriv->y;
-
-
-      if (R200_DEBUG & RADEON_PIXEL)
-	 fprintf(stderr, "readpixel blit src_pitch %d dst_pitch %d\n",
-		 src_pitch, dst_pitch);
-
-      for (i = 0 ; i < nbox ; i++)
-      {
-	 GLint bx = box[i].x1;
-	 GLint by = box[i].y1;
-	 GLint bw = box[i].x2 - bx;
-	 GLint bh = box[i].y2 - by;
-
-	 if (bx < x) bw -= x - bx, bx = x;
-	 if (by < y) bh -= y - by, by = y;
-	 if (bx + bw > x + width) bw = x + width - bx;
-	 if (by + bh > y + height) bh = y + height - by;
-	 if (bw <= 0) continue;
-	 if (bh <= 0) continue;
-
-	 r200EmitBlit( rmesa,
-		       blit_format,
-		       src_pitch, src_offset,
-		       dst_pitch, dst_offset,
-		       bx, by,
-		       bx - x, by - y,
-		       bw, bh );
-      }
-
-      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
-   }
-   UNLOCK_HARDWARE( &rmesa->radeon );
-
-   radeonFinish( ctx ); /* required by GL */
-#endif
-   return GL_TRUE;
-}
-
-static void
-r200ReadPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *pack,
-		 GLvoid *pixels )
-{
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!r200TryReadPixels( ctx, x, y, width, height, format, type, pack,
-			   pixels))
-      _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack,
-			  pixels);
-}
-
-
-
-
-static void do_draw_pix( GLcontext *ctx,
-			 GLint x, GLint y, GLsizei width, GLsizei height,
-			 GLint pitch,
-			 const void *pixels,
-			 GLuint planemask)
-{
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-#if 0
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   __DRIdrawable *dPriv = radeon_get_drawable(&rmesa->radeon);
-   drm_clip_rect_t *box = dPriv->pClipRects;
-   struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorDrawBuffers[0];
-   driRenderbuffer *drb = (driRenderbuffer *) rb;
-   int nbox = dPriv->numClipRects;
-   int i;
-   int blit_format;
-   int size;
-   int src_offset = r200GartOffsetFromVirtual( rmesa, pixels );
-   int src_pitch = pitch * rmesa->radeon.radeonScreen->cpp;
-
-   switch ( rmesa->radeon.radeonScreen->cpp ) {
-   case 2:
-      blit_format = R200_CP_COLOR_FORMAT_RGB565;
-      break;
-   case 4:
-      blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
-      break;
-   default:
-      return;
-   }
-
-
-   LOCK_HARDWARE( &rmesa->radeon );
-
-   if (rmesa->store.cmd_used)
-      rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
-
-   y -= height;			/* cope with pixel zoom */
-
-   if (!clip_pixelrect(ctx, ctx->DrawBuffer,
-		       &x, &y, &width, &height,
-		       &size)) {
-      UNLOCK_HARDWARE( &rmesa->radeon );
-      return;
-   }
-
-   y = dPriv->h - y - height; 	/* convert from gl to hardware coords */
-   x += dPriv->x;
-   y += dPriv->y;
-
-
-   r200EmitWait( rmesa, RADEON_WAIT_3D );
-
-   for (i = 0 ; i < nbox ; i++ )
-   {
-      GLint bx = box[i].x1;
-      GLint by = box[i].y1;
-      GLint bw = box[i].x2 - bx;
-      GLint bh = box[i].y2 - by;
-
-      if (bx < x) bw -= x - bx, bx = x;
-      if (by < y) bh -= y - by, by = y;
-      if (bx + bw > x + width) bw = x + width - bx;
-      if (by + bh > y + height) bh = y + height - by;
-      if (bw <= 0) continue;
-      if (bh <= 0) continue;
-
-      r200EmitBlit( rmesa,
-		    blit_format,
-		    src_pitch, src_offset,
-		    drb->pitch * drb->cpp,
-		    drb->offset + rmesa->radeon.radeonScreen->fbLocation,
-		    bx - x, by - y,
-		    bx, by,
-		    bw, bh );
-   }
-
-   rcommonFlushCmdBufLocked( &rmesa->radeon, __FUNCTION__ );
-   radeonWaitForIdleLocked( &rmesa->radeon ); /* required by GL */
-   UNLOCK_HARDWARE( &rmesa->radeon );
-#endif
-}
-
-
-
-
-static GLboolean
-r200TryDrawPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *unpack,
-		  const GLvoid *pixels )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-   GLint pitch = unpack->RowLength ? unpack->RowLength : width;
-   GLuint planemask;
-   GLuint cpp = rmesa->radeon.radeonScreen->cpp;
-   GLint size = height * pitch * cpp;
-
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* check that we're drawing to exactly one color buffer */
-   if (ctx->DrawBuffer->_NumColorDrawBuffers != 1)
-     return GL_FALSE;
-
-   switch (format) {
-   case GL_RGB:
-   case GL_RGBA:
-   case GL_BGRA:
-      planemask = radeonPackColor(cpp,
-				ctx->Color.ColorMask[0][RCOMP],
-				ctx->Color.ColorMask[0][GCOMP],
-				ctx->Color.ColorMask[0][BCOMP],
-				ctx->Color.ColorMask[0][ACOMP]);
-
-      if (cpp == 2)
-	 planemask |= planemask << 16;
-
-      if (planemask != ~0)
-	 return GL_FALSE;	/* fix me -- should be possible */
-
-      /* Can't do conversions on GART reads/draws.
-       */
-      if ( !r200IsGartMemory( rmesa, pixels, size ) ) {
-	 if (R200_DEBUG & RADEON_PIXEL)
-	    fprintf(stderr, "%s: not GART memory\n", __FUNCTION__);
-	 return GL_FALSE;
-      }
-
-      if (!check_color(ctx, type, format, unpack, pixels, size, pitch)) {
-	 return GL_FALSE;
-      }
-      if (!check_color_per_fragment_ops(ctx)) {
-	 return GL_FALSE;
-      }
-
-      if (ctx->Pixel.ZoomX != 1.0F ||
-	  ctx->Pixel.ZoomY != -1.0F)
-	 return GL_FALSE;
-      break;
-
-   default:
-      return GL_FALSE;
-   }
-
-   if (0)// r200IsGartMemory(rmesa, pixels, size) )
-   {
-      do_draw_pix( ctx, x, y, width, height, pitch, pixels, planemask );
-      return GL_TRUE;
-   }
-   else if (0)
-   {
-      /* Pixels is in regular memory -- get dma buffers and perform
-       * upload through them.
-       */
-   }
-   else
-      return GL_FALSE;
-}
-
-static void
-r200DrawPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *unpack,
-		 const GLvoid *pixels )
-{
-   if (R200_DEBUG & RADEON_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!r200TryDrawPixels( ctx, x, y, width, height, format, type,
-			  unpack, pixels ))
-      _swrast_DrawPixels( ctx, x, y, width, height, format, type,
-			  unpack, pixels );
-}
-
-
-static void
-r200Bitmap( GLcontext *ctx, GLint px, GLint py,
-		  GLsizei width, GLsizei height,
-		  const struct gl_pixelstore_attrib *unpack,
-		  const GLubyte *bitmap )
-{
-   r200ContextPtr rmesa = R200_CONTEXT(ctx);
-
-   if (rmesa->radeon.Fallback)
-      _swrast_Bitmap( ctx, px, py, width, height, unpack, bitmap );
-   else
-      r200PointsBitmap( ctx, px, py, width, height, unpack, bitmap );
-}
-
-
-
-void r200InitPixelFuncs( GLcontext *ctx )
-{
-   if (!getenv("R200_NO_BLITS")) {
-      ctx->Driver.ReadPixels = r200ReadPixels;
-      ctx->Driver.DrawPixels = r200DrawPixels;
-      if (getenv("R200_HW_BITMAP"))
-	 ctx->Driver.Bitmap = r200Bitmap;
-   }
-}
diff --git a/src/mesa/drivers/dri/r200/r200_pixel.h b/src/mesa/drivers/dri/r200/r200_pixel.h
deleted file mode 100644
index e62aa05d749..00000000000
--- a/src/mesa/drivers/dri/r200/r200_pixel.h
+++ /dev/null
@@ -1,39 +0,0 @@
-/*
-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 __R200_PIXEL_H__
-#define __R200_PIXEL_H__
-
-extern void r200InitPixelFuncs( GLcontext *ctx );
-
-#endif
diff --git a/src/mesa/drivers/dri/r200/r200_state.c b/src/mesa/drivers/dri/r200/r200_state.c
index 050e5aa8770..71f764aaaef 100644
--- a/src/mesa/drivers/dri/r200/r200_state.c
+++ b/src/mesa/drivers/dri/r200/r200_state.c
@@ -46,6 +46,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/tnl.h"
 #include "tnl/t_pipeline.h"
 #include "swrast_setup/swrast_setup.h"
+#include "drivers/common/meta.h"
 
 #include "radeon_common.h"
 #include "radeon_mipmap_tree.h"
@@ -2494,6 +2495,9 @@ void r200InitStateFuncs( struct dd_function_table *functions )
 
    functions->DrawBuffer		= radeonDrawBuffer;
    functions->ReadBuffer		= radeonReadBuffer;
+   functions->CopyPixels                = _mesa_meta_CopyPixels;
+   functions->DrawPixels                = _mesa_meta_DrawPixels;
+   functions->ReadPixels                = radeonReadPixels;
 
    functions->AlphaFunc			= r200AlphaFunc;
    functions->BlendColor		= r200BlendColor;
diff --git a/src/mesa/drivers/dri/r200/radeon_pixel_read.c b/src/mesa/drivers/dri/r200/radeon_pixel_read.c
new file mode 120000
index 00000000000..3b03803126f
--- /dev/null
+++ b/src/mesa/drivers/dri/r200/radeon_pixel_read.c
@@ -0,0 +1 @@
+../radeon/radeon_pixel_read.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r200/radeon_tex_getimage.c b/src/mesa/drivers/dri/r200/radeon_tex_getimage.c
new file mode 120000
index 00000000000..d9836d7326e
--- /dev/null
+++ b/src/mesa/drivers/dri/r200/radeon_tex_getimage.c
@@ -0,0 +1 @@
+../radeon/radeon_tex_getimage.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r200/radeon_tile.c b/src/mesa/drivers/dri/r200/radeon_tile.c
new file mode 120000
index 00000000000..d4bfe27da64
--- /dev/null
+++ b/src/mesa/drivers/dri/r200/radeon_tile.c
@@ -0,0 +1 @@
+../radeon/radeon_tile.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r200/radeon_tile.h b/src/mesa/drivers/dri/r200/radeon_tile.h
new file mode 120000
index 00000000000..31074c581ea
--- /dev/null
+++ b/src/mesa/drivers/dri/r200/radeon_tile.h
@@ -0,0 +1 @@
+../radeon/radeon_tile.h
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r200/server/radeon_dri.c b/src/mesa/drivers/dri/r200/server/radeon_dri.c
deleted file mode 120000
index d05847d650f..00000000000
--- a/src/mesa/drivers/dri/r200/server/radeon_dri.c
+++ /dev/null
@@ -1 +0,0 @@
-../../radeon/server/radeon_dri.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile
index 04459c2ddfa..2245998c952 100644
--- a/src/mesa/drivers/dri/r300/Makefile
+++ b/src/mesa/drivers/dri/r300/Makefile
@@ -7,8 +7,6 @@ CFLAGS += $(RADEON_CFLAGS)
 
 LIBNAME = r300_dri.so
 
-MINIGLX_SOURCES = server/radeon_dri.c
-
 ifeq ($(RADEON_LDFLAGS),)
 CS_SOURCES = radeon_cs_space_drm.c radeon_bo.c radeon_cs.c
 endif
@@ -33,10 +31,13 @@ RADEON_COMMON_SOURCES = \
 	radeon_fbo.c \
 	radeon_lock.c \
 	radeon_mipmap_tree.c \
-	radeon_span.c \
+	radeon_pixel_read.c \
 	radeon_queryobj.c \
+	radeon_span.c \
 	radeon_texture.c \
-	radeon_tex_copy.c
+	radeon_tex_copy.c \
+	radeon_tex_getimage.c \
+	radeon_tile.c
 
 DRIVER_SOURCES = \
 		 radeon_screen.c \
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index 6cfa5686f4a..e2dbb1dbf40 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -332,36 +332,37 @@ void r300_emit_cb_setup(struct r300_context *r300,
     assert(offset % 32 == 0);
 
     switch (format) {
-        case MESA_FORMAT_RGB565:
-            assert(_mesa_little_endian());
-            cbpitch |= R300_COLOR_FORMAT_RGB565;
+        case MESA_FORMAT_SL8:
+        case MESA_FORMAT_A8:
+        case MESA_FORMAT_L8:
+        case MESA_FORMAT_I8:
+            cbpitch |= R300_COLOR_FORMAT_I8;
             break;
+        case MESA_FORMAT_RGB565:
         case MESA_FORMAT_RGB565_REV:
-            assert(!_mesa_little_endian());
             cbpitch |= R300_COLOR_FORMAT_RGB565;
             break;
         case MESA_FORMAT_ARGB4444:
-            assert(_mesa_little_endian());
-            cbpitch |= R300_COLOR_FORMAT_ARGB4444;
-            break;
         case MESA_FORMAT_ARGB4444_REV:
-            assert(!_mesa_little_endian());
             cbpitch |= R300_COLOR_FORMAT_ARGB4444;
             break;
+        case MESA_FORMAT_RGBA5551:
         case MESA_FORMAT_ARGB1555:
-            assert(_mesa_little_endian());
-            cbpitch |= R300_COLOR_FORMAT_ARGB1555;
-            break;
         case MESA_FORMAT_ARGB1555_REV:
-            assert(!_mesa_little_endian());
             cbpitch |= R300_COLOR_FORMAT_ARGB1555;
             break;
+        case MESA_FORMAT_RGBA8888:
+        case MESA_FORMAT_RGBA8888_REV:
+        case MESA_FORMAT_XRGB8888:
+        case MESA_FORMAT_ARGB8888:
+        case MESA_FORMAT_XRGB8888_REV:
+        case MESA_FORMAT_ARGB8888_REV:
+        case MESA_FORMAT_SRGBA8:
+        case MESA_FORMAT_SARGB8:
+            cbpitch |= R300_COLOR_FORMAT_ARGB8888;
+            break;
         default:
-            if (cpp == 4) {
-                cbpitch |= R300_COLOR_FORMAT_ARGB8888;
-            } else {
-                _mesa_problem(r300->radeon.glCtx, "unexpected format in emit_cb_offset()");;
-            }
+            _mesa_problem(r300->radeon.glCtx, "unexpected format in emit_cb_offset()");
             break;
     }
 
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index ff35cd52753..364e0ba6b61 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -321,6 +321,12 @@ static void r300_init_vtbl(radeonContextPtr radeon)
 
 	radeon->vtbl.check_blit = r300_check_blit;
 	radeon->vtbl.blit = r300_blit;
+
+	if (radeon->radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+		radeon->vtbl.is_format_renderable = r500IsFormatRenderable;
+	} else {
+		radeon->vtbl.is_format_renderable = r300IsFormatRenderable;
+	}
 }
 
 static void r300InitConstValues(GLcontext *ctx, radeonScreenPtr screen)
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 5979dedac4f..87489412419 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -46,6 +46,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/simple_list.h"
 #include "main/api_arrayelt.h"
 
+#include "drivers/common/meta.h"
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
 #include "shader/prog_parameter.h"
@@ -2237,6 +2238,68 @@ void r300UpdateShaderStates(r300ContextPtr rmesa)
 	}
 }
 
+#define EASY_US_OUT_FMT(comps, c0, c1, c2, c3) \
+	(R500_OUT_FMT_##comps | R500_C0_SEL_##c0 | R500_C1_SEL_##c1 | \
+	 R500_C2_SEL_##c2 | R500_C3_SEL_##c3)
+static void r300SetupUsOutputFormat(GLcontext *ctx)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	uint32_t hw_format;
+	struct radeon_renderbuffer *rrb = radeon_get_colorbuffer(&rmesa->radeon);
+
+	if (!rrb) {
+		return;
+	}
+	
+	switch (rrb->base.Format)
+	{
+		case MESA_FORMAT_RGBA5551:
+		case MESA_FORMAT_RGBA8888:
+			hw_format = EASY_US_OUT_FMT(C4_8, A, B, G, R);
+			break;
+		case MESA_FORMAT_RGB565_REV:
+		case MESA_FORMAT_RGBA8888_REV:
+			hw_format = EASY_US_OUT_FMT(C4_8, R, G, B, A);
+			break;
+		case MESA_FORMAT_RGB565:
+		case MESA_FORMAT_ARGB4444:
+		case MESA_FORMAT_ARGB1555:
+		case MESA_FORMAT_XRGB8888:
+		case MESA_FORMAT_ARGB8888:
+			hw_format = EASY_US_OUT_FMT(C4_8, B, G, R, A);
+			break;
+		case MESA_FORMAT_ARGB4444_REV:
+		case MESA_FORMAT_ARGB1555_REV:
+		case MESA_FORMAT_XRGB8888_REV:
+		case MESA_FORMAT_ARGB8888_REV:
+			hw_format = EASY_US_OUT_FMT(C4_8, A, R, G, B);
+			break;
+		case MESA_FORMAT_SRGBA8:
+			hw_format = EASY_US_OUT_FMT(C4_10_GAMMA, A, B, G, R);
+			break;
+		case MESA_FORMAT_SARGB8:
+			hw_format = EASY_US_OUT_FMT(C4_10_GAMMA, B, G, R, A);
+			break;
+		case MESA_FORMAT_SL8:
+			hw_format = EASY_US_OUT_FMT(C4_10_GAMMA, A, A, R, A);
+			break;
+		case MESA_FORMAT_A8:
+			hw_format = EASY_US_OUT_FMT(C4_8, A, A, A, A);
+			break;
+		case MESA_FORMAT_L8:
+		case MESA_FORMAT_I8:
+			hw_format = EASY_US_OUT_FMT(C4_8, A, A, R, A);
+			break;
+		default:
+			assert(!"Unsupported format");
+			break;
+	}
+
+	R300_STATECHANGE(rmesa, us_out_fmt);
+	rmesa->hw.us_out_fmt.cmd[1] = hw_format;
+}
+#undef EASY_US_OUT_FMT
+
 /**
  * Called by Mesa after an internal state update.
  */
@@ -2266,6 +2329,10 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state)
 			r300->hw.shade2.cmd[1] &= ~R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 	}
 
+	if (new_state & _NEW_BUFFERS) {
+		r300SetupUsOutputFormat(ctx);
+	}
+
 	r300->radeon.NewGLState |= new_state;
 }
 
@@ -2326,8 +2393,12 @@ void r300InitStateFuncs(struct dd_function_table *functions)
 	functions->ClipPlane = r300ClipPlane;
 	functions->Scissor = radeonScissor;
 
-	functions->DrawBuffer		= radeonDrawBuffer;
-	functions->ReadBuffer		= radeonReadBuffer;
+	functions->DrawBuffer = radeonDrawBuffer;
+	functions->ReadBuffer = radeonReadBuffer;
+
+	functions->CopyPixels = _mesa_meta_CopyPixels;
+	functions->DrawPixels = _mesa_meta_DrawPixels;
+	functions->ReadPixels = radeonReadPixels;
 }
 
 void r300InitShaderFunctions(r300ContextPtr r300)
diff --git a/src/mesa/drivers/dri/r300/r300_tex.c b/src/mesa/drivers/dri/r300/r300_tex.c
index 8dd85073954..baef206bc26 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.c
+++ b/src/mesa/drivers/dri/r300/r300_tex.c
@@ -308,6 +308,45 @@ static struct gl_texture_object *r300NewTextureObject(GLcontext * ctx,
 	return &t->base;
 }
 
+unsigned r300IsFormatRenderable(gl_format mesa_format)
+{
+	switch (mesa_format)
+	{
+		case MESA_FORMAT_RGB565:
+		case MESA_FORMAT_RGBA5551:
+		case MESA_FORMAT_RGBA8888:
+		case MESA_FORMAT_RGB565_REV:
+		case MESA_FORMAT_RGBA8888_REV:
+		case MESA_FORMAT_ARGB4444:
+		case MESA_FORMAT_ARGB1555:
+		case MESA_FORMAT_XRGB8888:
+		case MESA_FORMAT_ARGB8888:
+		case MESA_FORMAT_ARGB4444_REV:
+		case MESA_FORMAT_ARGB1555_REV:
+		case MESA_FORMAT_XRGB8888_REV:
+		case MESA_FORMAT_ARGB8888_REV:
+		case MESA_FORMAT_SRGBA8:
+		case MESA_FORMAT_SARGB8:
+		case MESA_FORMAT_SL8:
+		case MESA_FORMAT_A8:
+		case MESA_FORMAT_L8:
+		case MESA_FORMAT_I8:
+		case MESA_FORMAT_Z16:
+			return 1;
+		default:
+			return 0;
+	}
+}
+
+unsigned r500IsFormatRenderable(gl_format mesa_format)
+{
+	if (mesa_format == MESA_FORMAT_S8_Z24) {
+		return 1;
+	} else {
+		return r300IsFormatRenderable(mesa_format);
+	}
+}
+
 void r300InitTextureFuncs(radeonContextPtr radeon, struct dd_function_table *functions)
 {
 	/* Note: we only plug in the functions we implement in the driver
diff --git a/src/mesa/drivers/dri/r300/r300_tex.h b/src/mesa/drivers/dri/r300/r300_tex.h
index 9694e703b83..aca44cd7669 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.h
+++ b/src/mesa/drivers/dri/r300/r300_tex.h
@@ -53,4 +53,7 @@ extern void r300InitTextureFuncs(radeonContextPtr radeon, struct dd_function_tab
 
 int32_t r300TranslateTexFormat(gl_format mesaFormat);
 
+unsigned r300IsFormatRenderable(gl_format mesaFormat);
+unsigned r500IsFormatRenderable(gl_format mesaFormat);
+
 #endif				/* __r300_TEX_H__ */
diff --git a/src/mesa/drivers/dri/r300/radeon_pixel_read.c b/src/mesa/drivers/dri/r300/radeon_pixel_read.c
new file mode 120000
index 00000000000..3b03803126f
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_pixel_read.c
@@ -0,0 +1 @@
+../radeon/radeon_pixel_read.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/radeon_tex_getimage.c b/src/mesa/drivers/dri/r300/radeon_tex_getimage.c
new file mode 120000
index 00000000000..d9836d7326e
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_tex_getimage.c
@@ -0,0 +1 @@
+../radeon/radeon_tex_getimage.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/radeon_tile.c b/src/mesa/drivers/dri/r300/radeon_tile.c
new file mode 120000
index 00000000000..d4bfe27da64
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_tile.c
@@ -0,0 +1 @@
+../radeon/radeon_tile.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/radeon_tile.h b/src/mesa/drivers/dri/r300/radeon_tile.h
new file mode 120000
index 00000000000..31074c581ea
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_tile.h
@@ -0,0 +1 @@
+../radeon/radeon_tile.h
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/server/radeon_dri.c b/src/mesa/drivers/dri/r300/server/radeon_dri.c
deleted file mode 120000
index d05847d650f..00000000000
--- a/src/mesa/drivers/dri/r300/server/radeon_dri.c
+++ /dev/null
@@ -1 +0,0 @@
-../../radeon/server/radeon_dri.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r600/Makefile b/src/mesa/drivers/dri/r600/Makefile
index 5d509415393..17915621ee4 100644
--- a/src/mesa/drivers/dri/r600/Makefile
+++ b/src/mesa/drivers/dri/r600/Makefile
@@ -7,8 +7,6 @@ CFLAGS += $(RADEON_CFLAGS)
 
 LIBNAME = r600_dri.so
 
-MINIGLX_SOURCES = server/radeon_dri.c
-
 ifeq ($(RADEON_LDFLAGS),)
 CS_SOURCES = radeon_cs_space_drm.c radeon_bo.c radeon_cs.c
 endif
@@ -33,10 +31,13 @@ RADEON_COMMON_SOURCES = \
 	radeon_fbo.c \
 	radeon_lock.c \
 	radeon_mipmap_tree.c \
+	radeon_pixel_read.c \
+	radeon_queryobj.c \
 	radeon_span.c \
 	radeon_texture.c \
-	radeon_queryobj.c \
-	radeon_tex_copy.c
+	radeon_tex_copy.c \
+	radeon_tex_getimage.c \
+	radeon_tile.c
 
 DRIVER_SOURCES = \
 		 radeon_screen.c \
diff --git a/src/mesa/drivers/dri/r600/r600_context.c b/src/mesa/drivers/dri/r600/r600_context.c
index 134e97e7c33..76d5027649e 100644
--- a/src/mesa/drivers/dri/r600/r600_context.c
+++ b/src/mesa/drivers/dri/r600/r600_context.c
@@ -239,6 +239,7 @@ static void r600_init_vtbl(radeonContextPtr radeon)
 	radeon->vtbl.emit_query_finish = r600_emit_query_finish;
 	radeon->vtbl.check_blit = r600_check_blit;
 	radeon->vtbl.blit = r600_blit;
+	radeon->vtbl.is_format_renderable = radeonIsFormatRenderable;
 }
 
 static void r600InitConstValues(GLcontext *ctx, radeonScreenPtr screen)
diff --git a/src/mesa/drivers/dri/r600/r700_state.c b/src/mesa/drivers/dri/r600/r700_state.c
index 6f156b54096..8797f8059a9 100644
--- a/src/mesa/drivers/dri/r600/r700_state.c
+++ b/src/mesa/drivers/dri/r600/r700_state.c
@@ -39,6 +39,7 @@
 #include "swrast_setup/swrast_setup.h"
 #include "main/api_arrayelt.h"
 #include "main/framebuffer.h"
+#include "drivers/common/meta.h"
 
 #include "shader/prog_parameter.h"
 #include "shader/prog_statevars.h"
@@ -1857,8 +1858,12 @@ void r700InitStateFuncs(struct dd_function_table *functions) //-----------------
 
 	functions->Scissor = radeonScissor;
 
-	functions->DrawBuffer		= radeonDrawBuffer;
-	functions->ReadBuffer		= radeonReadBuffer;
+	functions->DrawBuffer = radeonDrawBuffer;
+	functions->ReadBuffer = radeonReadBuffer;
+
+	functions->CopyPixels = _mesa_meta_CopyPixels;
+	functions->DrawPixels = _mesa_meta_DrawPixels;
+	functions->ReadPixels = radeonReadPixels;
 
 }
 
diff --git a/src/mesa/drivers/dri/r600/radeon_pixel_read.c b/src/mesa/drivers/dri/r600/radeon_pixel_read.c
new file mode 120000
index 00000000000..3b03803126f
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/radeon_pixel_read.c
@@ -0,0 +1 @@
+../radeon/radeon_pixel_read.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r600/radeon_tex_getimage.c b/src/mesa/drivers/dri/r600/radeon_tex_getimage.c
new file mode 120000
index 00000000000..d9836d7326e
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/radeon_tex_getimage.c
@@ -0,0 +1 @@
+../radeon/radeon_tex_getimage.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r600/radeon_tile.c b/src/mesa/drivers/dri/r600/radeon_tile.c
new file mode 120000
index 00000000000..d4bfe27da64
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/radeon_tile.c
@@ -0,0 +1 @@
+../radeon/radeon_tile.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r600/radeon_tile.h b/src/mesa/drivers/dri/r600/radeon_tile.h
new file mode 120000
index 00000000000..31074c581ea
--- /dev/null
+++ b/src/mesa/drivers/dri/r600/radeon_tile.h
@@ -0,0 +1 @@
+../radeon/radeon_tile.h
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/r600/server/radeon_dri.c b/src/mesa/drivers/dri/r600/server/radeon_dri.c
deleted file mode 120000
index d05847d650f..00000000000
--- a/src/mesa/drivers/dri/r600/server/radeon_dri.c
+++ /dev/null
@@ -1 +0,0 @@
-../../radeon/server/radeon_dri.c
\ No newline at end of file
diff --git a/src/mesa/drivers/dri/radeon/Makefile b/src/mesa/drivers/dri/radeon/Makefile
index a54ea16ec63..19df62742ec 100644
--- a/src/mesa/drivers/dri/radeon/Makefile
+++ b/src/mesa/drivers/dri/radeon/Makefile
@@ -8,8 +8,6 @@ CFLAGS += $(RADEON_CFLAGS)
 
 LIBNAME = radeon_dri.so
 
-MINIGLX_SOURCES = server/radeon_dri.c 
-
 ifeq ($(RADEON_LDFLAGS),)
 CS_SOURCES = radeon_cs_space_drm.c radeon_bo.c radeon_cs.c
 endif
@@ -24,10 +22,13 @@ RADEON_COMMON_SOURCES = \
 	radeon_fbo.c \
 	radeon_lock.c \
 	radeon_mipmap_tree.c \
+	radeon_pixel_read.c \
 	radeon_queryobj.c \
 	radeon_span.c \
 	radeon_texture.c \
-	radeon_tex_copy.c
+	radeon_tex_copy.c \
+	radeon_tex_getimage.c \
+	radeon_tile.c
 
 DRIVER_SOURCES = \
 	radeon_context.c \
diff --git a/src/mesa/drivers/dri/radeon/radeon_blit.c b/src/mesa/drivers/dri/radeon/radeon_blit.c
index e1e1f215508..143822361e1 100644
--- a/src/mesa/drivers/dri/radeon/radeon_blit.c
+++ b/src/mesa/drivers/dri/radeon/radeon_blit.c
@@ -48,6 +48,8 @@ unsigned r100_check_blit(gl_format mesa_format)
     case MESA_FORMAT_ARGB4444:
     case MESA_FORMAT_ARGB1555:
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_L8:
+    case MESA_FORMAT_I8:
 	    break;
     default:
 	    return 0;
@@ -103,6 +105,9 @@ static void inline emit_tx_setup(struct r100_context *r100,
     case MESA_FORMAT_ARGB8888:
 	    txformat |= RADEON_TXFORMAT_ARGB8888 | RADEON_TXFORMAT_ALPHA_IN_MAP;
 	    break;
+    case MESA_FORMAT_RGBA8888:
+            txformat |= RADEON_TXFORMAT_RGBA8888 | RADEON_TXFORMAT_ALPHA_IN_MAP;
+            break;
     case MESA_FORMAT_XRGB8888:
 	    txformat |= RADEON_TXFORMAT_ARGB8888;
 	    break;
@@ -116,8 +121,15 @@ static void inline emit_tx_setup(struct r100_context *r100,
 	    txformat |= RADEON_TXFORMAT_ARGB1555 | RADEON_TXFORMAT_ALPHA_IN_MAP;
 	    break;
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_I8:
 	    txformat |= RADEON_TXFORMAT_I8 | RADEON_TXFORMAT_ALPHA_IN_MAP;
 	    break;
+    case MESA_FORMAT_L8:
+            txformat |= RADEON_TXFORMAT_I8;
+            break;
+    case MESA_FORMAT_AL88:
+            txformat |= RADEON_TXFORMAT_AI88 | RADEON_TXFORMAT_ALPHA_IN_MAP;
+            break;
     default:
 	    break;
     }
@@ -177,6 +189,8 @@ static inline void emit_cb_setup(struct r100_context *r100,
 	    dst_format = RADEON_COLOR_FORMAT_ARGB1555;
 	    break;
     case MESA_FORMAT_A8:
+    case MESA_FORMAT_L8:
+    case MESA_FORMAT_I8:
 	    dst_format = RADEON_COLOR_FORMAT_RGB8;
 	    break;
     default:
diff --git a/src/mesa/drivers/dri/radeon/radeon_common.h b/src/mesa/drivers/dri/radeon/radeon_common.h
index cd01c9984e3..35b3f08fff9 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common.h
+++ b/src/mesa/drivers/dri/radeon/radeon_common.h
@@ -44,6 +44,12 @@ radeon_renderbuffer_set_bo(struct radeon_renderbuffer *rb,
 struct radeon_renderbuffer *
 radeon_create_renderbuffer(gl_format format, __DRIdrawable *driDrawPriv);
 
+void
+radeonReadPixels(GLcontext * ctx,
+				GLint x, GLint y, GLsizei width, GLsizei height,
+				GLenum format, GLenum type,
+				const struct gl_pixelstore_attrib *pack, GLvoid * pixels);
+
 void radeon_check_front_buffer_rendering(GLcontext *ctx);
 static inline struct radeon_renderbuffer *radeon_renderbuffer(struct gl_renderbuffer *rb)
 {
diff --git a/src/mesa/drivers/dri/radeon/radeon_common_context.h b/src/mesa/drivers/dri/radeon/radeon_common_context.h
index d1a24e265f2..5156c5d0d0a 100644
--- a/src/mesa/drivers/dri/radeon/radeon_common_context.h
+++ b/src/mesa/drivers/dri/radeon/radeon_common_context.h
@@ -539,6 +539,7 @@ struct radeon_context {
                         unsigned reg_width,
                         unsigned reg_height,
                         unsigned flip_y);
+	   unsigned (*is_format_renderable)(gl_format mesa_format);
    } vtbl;
 };
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c
index 878a453bd53..56aba16e9e0 100644
--- a/src/mesa/drivers/dri/radeon/radeon_context.c
+++ b/src/mesa/drivers/dri/radeon/radeon_context.c
@@ -200,6 +200,7 @@ static void r100_init_vtbl(radeonContextPtr radeon)
    radeon->vtbl.emit_query_finish = r100_emit_query_finish;
    radeon->vtbl.check_blit = r100_check_blit;
    radeon->vtbl.blit = r100_blit;
+   radeon->vtbl.is_format_renderable = radeonIsFormatRenderable;
 }
 
 /* Create the device specific context.
diff --git a/src/mesa/drivers/dri/radeon/radeon_fbo.c b/src/mesa/drivers/dri/radeon/radeon_fbo.c
index 46664a17556..63986058356 100644
--- a/src/mesa/drivers/dri/radeon/radeon_fbo.c
+++ b/src/mesa/drivers/dri/radeon/radeon_fbo.c
@@ -409,82 +409,51 @@ radeon_framebuffer_renderbuffer(GLcontext * ctx,
    radeon_draw_buffer(ctx, fb);
 }
 
-
-/* TODO: According to EXT_fbo spec internal format of texture image
- * once set during glTexImage call, should be preserved when
- * attaching image to renderbuffer. When HW doesn't support
- * rendering to format of attached image, set framebuffer
- * completeness accordingly in radeon_validate_framebuffer (issue #79).
- */
 static GLboolean
 radeon_update_wrapper(GLcontext *ctx, struct radeon_renderbuffer *rrb, 
 		     struct gl_texture_image *texImage)
 {
-	int retry = 0;
-	gl_format texFormat;
-
 	radeon_print(RADEON_TEXTURE, RADEON_TRACE,
-		"%s(%p, rrb %p, texImage %p) \n",
-		__func__, ctx, rrb, texImage);
-
-restart:
-	if (texImage->TexFormat == _dri_texformat_argb8888) {
-		rrb->base.DataType = GL_UNSIGNED_BYTE;
-		DBG("Render to RGBA8 texture OK\n");
+		"%s(%p, rrb %p, texImage %p, texFormat %s) \n",
+		__func__, ctx, rrb, texImage, _mesa_get_format_name(texImage->TexFormat));
+
+	switch (texImage->TexFormat) {
+		case MESA_FORMAT_RGBA8888:
+		case MESA_FORMAT_RGBA8888_REV:
+		case MESA_FORMAT_ARGB8888:
+		case MESA_FORMAT_ARGB8888_REV:
+		case MESA_FORMAT_XRGB8888:
+		case MESA_FORMAT_XRGB8888_REV:
+		case MESA_FORMAT_RGB565:
+		case MESA_FORMAT_RGB565_REV:
+		case MESA_FORMAT_RGBA5551:
+		case MESA_FORMAT_ARGB1555:
+		case MESA_FORMAT_ARGB1555_REV:
+		case MESA_FORMAT_ARGB4444:
+		case MESA_FORMAT_ARGB4444_REV:
+			rrb->base.DataType = GL_UNSIGNED_BYTE;
+			break;
+		case MESA_FORMAT_Z16:
+			rrb->base.DataType = GL_UNSIGNED_SHORT;
+			break;
+		case MESA_FORMAT_X8_Z24:
+			rrb->base.DataType = GL_UNSIGNED_INT;
+			break;
+		case MESA_FORMAT_S8_Z24:
+			rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+			break;
 	}
-	else if (texImage->TexFormat == _dri_texformat_rgb565) {
-		rrb->base.DataType = GL_UNSIGNED_BYTE;
-		DBG("Render to RGB5 texture OK\n");
-	}
-	else if (texImage->TexFormat == _dri_texformat_argb1555) {
-		rrb->base.DataType = GL_UNSIGNED_BYTE;
-		DBG("Render to ARGB1555 texture OK\n");
-	}
-	else if (texImage->TexFormat == _dri_texformat_argb4444) {
-		rrb->base.DataType = GL_UNSIGNED_BYTE;
-		DBG("Render to ARGB4444 texture OK\n");
-	}
-	else if (texImage->TexFormat == MESA_FORMAT_Z16) {
-		rrb->base.DataType = GL_UNSIGNED_SHORT;
-		DBG("Render to DEPTH16 texture OK\n");
-	}
-	else if (texImage->TexFormat == MESA_FORMAT_S8_Z24) {
-		rrb->base.DataType = GL_UNSIGNED_INT_24_8_EXT;
-		DBG("Render to DEPTH_STENCIL texture OK\n");
-	}
-	else {
-		/* try redoing the FBO */
-		if (retry == 1) {
-			DBG("Render to texture BAD FORMAT %d\n",
-			    texImage->TexFormat);
-			return GL_FALSE;
-		}
-                /* XXX why is the tex format being set here?
-                 * I think this can be removed.
-                 */
-		texImage->TexFormat = radeonChooseTextureFormat(ctx, texImage->InternalFormat, 0,
-								_mesa_get_format_datatype(texImage->TexFormat),
-								1);
-
-		retry++;
-		goto restart;
-	}
-	
-	texFormat = texImage->TexFormat;
-
-	rrb->base.Format = texFormat;
-
-        rrb->cpp = _mesa_get_format_bytes(texFormat);
+		
+	rrb->cpp = _mesa_get_format_bytes(texImage->TexFormat);
 	rrb->pitch = texImage->Width * rrb->cpp;
+	rrb->base.Format = texImage->TexFormat;
 	rrb->base.InternalFormat = texImage->InternalFormat;
-        rrb->base._BaseFormat = _mesa_base_fbo_format(ctx, rrb->base.InternalFormat);
-
+	rrb->base._BaseFormat = _mesa_base_fbo_format(ctx, rrb->base.InternalFormat);
 	rrb->base.Width = texImage->Width;
 	rrb->base.Height = texImage->Height;
-	
 	rrb->base.Delete = radeon_delete_renderbuffer;
 	rrb->base.AllocStorage = radeon_nop_alloc_storage;
-	
+
 	return GL_TRUE;
 }
 
@@ -607,6 +576,35 @@ radeon_finish_render_texture(GLcontext * ctx,
 static void
 radeon_validate_framebuffer(GLcontext *ctx, struct gl_framebuffer *fb)
 {
+	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+	gl_format mesa_format;
+	int i;
+
+	for (i = -2; i < (GLint) ctx->Const.MaxColorAttachments; i++) {
+		struct gl_renderbuffer_attachment *att;
+		if (i == -2) {
+			att = &fb->Attachment[BUFFER_DEPTH];
+		} else if (i == -1) {
+			att = &fb->Attachment[BUFFER_STENCIL];
+		} else {
+			att = &fb->Attachment[BUFFER_COLOR0 + i];
+		}
+
+		if (att->Type == GL_TEXTURE) {
+			mesa_format = att->Texture->Image[att->CubeMapFace][att->TextureLevel]->TexFormat;
+		} else {
+			/* All renderbuffer formats are renderable, but not sampable */
+			continue;
+		}
+
+		if (!radeon->vtbl.is_format_renderable(mesa_format)){
+			fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED;
+			radeon_print(RADEON_TEXTURE, RADEON_TRACE,
+						"%s: HW doesn't support format %s as output format of attachment %d\n",
+						__FUNCTION__, _mesa_get_format_name(mesa_format), i);
+			return;
+		}
+	}
 }
 
 void radeon_fbo_init(struct radeon_context *radeon)
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
index cd843d965e5..78c5f5dd572 100644
--- a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.c
@@ -36,6 +36,7 @@
 #include "main/texobj.h"
 #include "main/enums.h"
 #include "radeon_texture.h"
+#include "radeon_tile.h"
 
 static unsigned get_aligned_compressed_row_stride(
 		gl_format format,
@@ -69,16 +70,51 @@ static unsigned get_aligned_compressed_row_stride(
 	return stride;
 }
 
-static unsigned get_compressed_image_size(
+unsigned get_texture_image_size(
 		gl_format format,
 		unsigned rowStride,
-		unsigned height)
+		unsigned height,
+		unsigned depth,
+		unsigned tiling)
 {
-	unsigned blockWidth, blockHeight;
+	if (_mesa_is_format_compressed(format)) {
+		unsigned blockWidth, blockHeight;
 
-	_mesa_get_format_block_size(format, &blockWidth, &blockHeight);
+		_mesa_get_format_block_size(format, &blockWidth, &blockHeight);
+
+		return rowStride * ((height + blockHeight - 1) / blockHeight) * depth;
+	} else if (tiling) {
+		/* Need to align height to tile height */
+		unsigned tileWidth, tileHeight;
 
-	return rowStride * ((height + blockHeight - 1) / blockHeight);
+		get_tile_size(format, &tileWidth, &tileHeight);
+		tileHeight--;
+
+		height = (height + tileHeight) & ~tileHeight;
+	}
+
+	return rowStride * height * depth;
+}
+
+unsigned get_texture_image_row_stride(radeonContextPtr rmesa, gl_format format, unsigned width, unsigned tiling)
+{
+	if (_mesa_is_format_compressed(format)) {
+		return get_aligned_compressed_row_stride(format, width, rmesa->texture_compressed_row_align);
+	} else {
+		unsigned row_align;
+
+		if (!_mesa_is_pow_two(width)) {
+			row_align = rmesa->texture_rect_row_align - 1;
+		} else if (tiling) {
+			unsigned tileWidth, tileHeight;
+			get_tile_size(format, &tileWidth, &tileHeight);
+			row_align = tileWidth * _mesa_get_format_bytes(format) - 1;
+		} else {
+			row_align = rmesa->texture_row_align - 1;
+		}
+
+		return (_mesa_format_row_stride(format, width) + row_align) & ~row_align;
+	}
 }
 
 /**
@@ -92,34 +128,15 @@ static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree
 	GLuint face, GLuint level, GLuint* curOffset)
 {
 	radeon_mipmap_level *lvl = &mt->levels[level];
-	uint32_t row_align;
 	GLuint height;
 
 	height = _mesa_next_pow_two_32(lvl->height);
 
-	/* Find image size in bytes */
-	if (_mesa_is_format_compressed(mt->mesaFormat)) {
-		lvl->rowstride = get_aligned_compressed_row_stride(mt->mesaFormat, lvl->width, rmesa->texture_compressed_row_align);
-		lvl->size = get_compressed_image_size(mt->mesaFormat, lvl->rowstride, height);
-	} else if (mt->target == GL_TEXTURE_RECTANGLE_NV) {
-		row_align = rmesa->texture_rect_row_align - 1;
-		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
-		lvl->size = lvl->rowstride * height;
-	} else if (mt->tilebits & RADEON_TXO_MICRO_TILE) {
-		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-		 * though the actual offset may be different (if texture is less than
-		 * 32 bytes width) to the untiled case */
-		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) * 2 + 31) & ~31;
-		lvl->size = lvl->rowstride * ((height + 1) / 2) * lvl->depth;
-	} else {
-		row_align = rmesa->texture_row_align - 1;
-		lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
-		lvl->size = lvl->rowstride * height * lvl->depth;
-	}
+	lvl->rowstride = get_texture_image_row_stride(rmesa, mt->mesaFormat, lvl->width, mt->tilebits);
+	lvl->size = get_texture_image_size(mt->mesaFormat, lvl->rowstride, lvl->height, lvl->depth, mt->tilebits);
+
 	assert(lvl->size > 0);
 
-	/* All images are aligned to a 32-byte offset */
-	*curOffset = (*curOffset + 0x1f) & ~0x1f;
 	lvl->faces[face].offset = *curOffset;
 	*curOffset += lvl->size;
 
@@ -451,12 +468,9 @@ static void migrate_image_to_miptree(radeon_mipmap_tree *mt,
 
 		radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel];
 
-		/* TODO: bring back these assertions once the FBOs are fixed */
-#if 0
 		assert(image->mtlevel == level);
 		assert(srclvl->size == dstlvl->size);
 		assert(srclvl->rowstride == dstlvl->rowstride);
-#endif
 
 		radeon_bo_map(image->mt->bo, GL_FALSE);
 
diff --git a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
index c911688c1a5..088f9701722 100644
--- a/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
+++ b/src/mesa/drivers/dri/radeon/radeon_mipmap_tree.h
@@ -89,4 +89,13 @@ void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t);
 GLuint radeon_miptree_image_offset(radeon_mipmap_tree *mt,
 				   GLuint face, GLuint level);
 uint32_t get_base_teximage_offset(radeonTexObj *texObj);
+
+unsigned get_texture_image_row_stride(radeonContextPtr rmesa, gl_format format, unsigned width, unsigned tiling);
+
+unsigned get_texture_image_size(
+		gl_format format,
+		unsigned rowStride,
+		unsigned height,
+		unsigned depth,
+		unsigned tiling);
 #endif /* __RADEON_MIPMAP_TREE_H_ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_pixel_read.c b/src/mesa/drivers/dri/radeon/radeon_pixel_read.c
new file mode 100644
index 00000000000..27841938e66
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_pixel_read.c
@@ -0,0 +1,188 @@
+/*
+ * Copyright (C) 2010 Maciej Cencora <m.cencora@gmail.com>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "stdint.h"
+#include "main/bufferobj.h"
+#include "main/enums.h"
+#include "main/image.h"
+#include "main/state.h"
+#include "swrast/swrast.h"
+
+#include "radeon_common_context.h"
+#include "radeon_debug.h"
+#include "radeon_mipmap_tree.h"
+
+static gl_format gl_format_and_type_to_mesa_format(GLenum format, GLenum type)
+{
+    switch (format)
+    {
+        case GL_RGB:
+            switch (type) {
+                case GL_UNSIGNED_SHORT_5_6_5:
+                    return MESA_FORMAT_RGB565;
+                case GL_UNSIGNED_SHORT_5_6_5_REV:
+                    return MESA_FORMAT_RGB565_REV;
+            }
+            break;
+        case GL_RGBA:
+            switch (type) {
+                case GL_UNSIGNED_BYTE:
+                    return MESA_FORMAT_RGBA8888_REV;
+                case GL_FLOAT:
+                    return MESA_FORMAT_RGBA_FLOAT32;
+                case GL_UNSIGNED_SHORT_4_4_4_4:
+                    return MESA_FORMAT_ARGB4444;
+                case GL_UNSIGNED_SHORT_4_4_4_4_REV:
+                    return MESA_FORMAT_ARGB4444;
+                case GL_UNSIGNED_SHORT_5_5_5_1:
+                    return MESA_FORMAT_RGBA5551;
+                case GL_UNSIGNED_SHORT_1_5_5_5_REV:
+                    return MESA_FORMAT_ARGB1555_REV;
+                case GL_UNSIGNED_INT_8_8_8_8:
+                    return MESA_FORMAT_ARGB8888;
+                case GL_UNSIGNED_INT_8_8_8_8_REV:
+                    return MESA_FORMAT_ARGB8888_REV;
+            }
+            break;
+    }
+
+    return MESA_FORMAT_NONE;
+}
+
+static GLboolean
+do_blit_readpixels(GLcontext * ctx,
+                   GLint x, GLint y, GLsizei width, GLsizei height,
+                   GLenum format, GLenum type,
+                   const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
+{
+    radeonContextPtr radeon = RADEON_CONTEXT(ctx);
+    const struct radeon_renderbuffer *rrb = radeon_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
+    const gl_format dst_format = gl_format_and_type_to_mesa_format(format, type);
+    unsigned dst_rowstride, dst_imagesize, aligned_rowstride, flip_y;
+    struct radeon_bo *dst_buffer;
+    GLint dst_x = 0, dst_y = 0;
+
+    /* It's not worth if number of pixels to copy is really small */
+    if (width * height < 100) {
+        return GL_FALSE;
+    }
+
+    if (dst_format == MESA_FORMAT_NONE ||
+        !radeon->vtbl.check_blit(dst_format) || !radeon->vtbl.blit) {
+        return GL_FALSE;
+    }
+
+    if (ctx->_ImageTransferState) {
+        return GL_FALSE;
+    }
+
+    if (pack->SwapBytes || pack->LsbFirst) {
+        return GL_FALSE;
+    }
+
+    if (pack->RowLength > 0) {
+        dst_rowstride = pack->RowLength;
+    } else {
+        dst_rowstride = width;
+    }
+
+    if (!_mesa_clip_copytexsubimage(ctx, &dst_x, &dst_y, &x, &y, &width, &height)) {
+        return GL_TRUE;
+    }
+    assert(x >= 0 && y >= 0);
+
+    aligned_rowstride = get_texture_image_row_stride(radeon, dst_format, dst_rowstride, 0);
+    dst_imagesize = get_texture_image_size(dst_format,
+                                           aligned_rowstride,
+                                           height, 1, 0);
+    dst_buffer = radeon_bo_open(radeon->radeonScreen->bom, 0, dst_imagesize, 1024, RADEON_GEM_DOMAIN_GTT, 0);
+
+    /* Disable source Y flipping for FBOs */
+    flip_y = (ctx->ReadBuffer->Name == 0);
+    if (pack->Invert) {
+        y = rrb->base.Height - height - y;
+        flip_y = !flip_y;
+    }
+
+    if (radeon->vtbl.blit(ctx,
+                          rrb->bo,
+                          rrb->draw_offset,
+                          rrb->base.Format,
+                          rrb->pitch / rrb->cpp,
+                          rrb->base.Width,
+                          rrb->base.Height,
+                          x,
+                          y,
+                          dst_buffer,
+                          0, /* dst_offset */
+                          dst_format,
+                          aligned_rowstride / _mesa_get_format_bytes(dst_format),
+                          width,
+                          height,
+                          0, /* dst_x */
+                          0, /* dst_y */
+                          width,
+                          height,
+                          flip_y))
+    {
+        radeon_bo_map(dst_buffer, 0);
+        dst_rowstride *= _mesa_get_format_bytes(dst_format);
+        copy_rows(pixels, dst_rowstride, dst_buffer->ptr,
+                  aligned_rowstride, height, dst_rowstride);
+        radeon_bo_unmap(dst_buffer);
+        radeon_bo_unref(dst_buffer);
+        return GL_TRUE;
+    } else {
+        radeon_bo_unref(dst_buffer);
+        return GL_FALSE;
+    }
+}
+
+void
+radeonReadPixels(GLcontext * ctx,
+                 GLint x, GLint y, GLsizei width, GLsizei height,
+                 GLenum format, GLenum type,
+                 const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
+{
+    if (do_blit_readpixels(ctx, x, y, width, height, format, type, pack, pixels))
+        return;
+
+    /* Update Mesa state before calling down into _swrast_ReadPixels, as
+     * the spans code requires the computed buffer states to be up to date,
+     * but _swrast_ReadPixels only updates Mesa state after setting up
+     * the spans code.
+     */
+
+    radeon_print(RADEON_FALLBACKS, RADEON_NORMAL,
+                 "Falling back to sw for ReadPixels (format %s, type %s)\n",
+                 _mesa_lookup_enum_by_nr(format), _mesa_lookup_enum_by_nr(type));
+
+    if (ctx->NewState)
+        _mesa_update_state(ctx);
+
+    _swrast_ReadPixels(ctx, x, y, width, height, format, type, pack, pixels);
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c
index 0ce97e86972..ec5612fdd7f 100644
--- a/src/mesa/drivers/dri/radeon/radeon_state.c
+++ b/src/mesa/drivers/dri/radeon/radeon_state.c
@@ -45,6 +45,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "tnl/tnl.h"
 #include "tnl/t_pipeline.h"
 #include "swrast_setup/swrast_setup.h"
+#include "drivers/common/meta.h"
 
 #include "radeon_context.h"
 #include "radeon_mipmap_tree.h"
@@ -2248,6 +2249,9 @@ void radeonInitStateFuncs( GLcontext *ctx , GLboolean dri2 )
 
    ctx->Driver.DrawBuffer		= radeonDrawBuffer;
    ctx->Driver.ReadBuffer		= radeonReadBuffer;
+   ctx->Driver.CopyPixels               = _mesa_meta_CopyPixels;
+   ctx->Driver.DrawPixels               = _mesa_meta_DrawPixels;
+   ctx->Driver.ReadPixels               = radeonReadPixels;
 
    ctx->Driver.AlphaFunc		= radeonAlphaFunc;
    ctx->Driver.BlendEquationSeparate	= radeonBlendEquationSeparate;
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex_getimage.c b/src/mesa/drivers/dri/radeon/radeon_tex_getimage.c
new file mode 100644
index 00000000000..7bf6dcc2e32
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tex_getimage.c
@@ -0,0 +1,95 @@
+/*
+ * Copyright (C) 2009 Maciej Cencora.
+ * Copyright (C) 2008 Nicolai Haehnle.
+ * 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.
+ *
+ */
+
+#include "radeon_common_context.h"
+#include "radeon_texture.h"
+#include "radeon_mipmap_tree.h"
+
+#include "main/texgetimage.h"
+
+/**
+ * Need to map texture image into memory before copying image data,
+ * then unmap it.
+ */
+static void
+radeon_get_tex_image(GLcontext * ctx, GLenum target, GLint level,
+             GLenum format, GLenum type, GLvoid * pixels,
+             struct gl_texture_object *texObj,
+             struct gl_texture_image *texImage, int compressed)
+{
+    radeon_texture_image *image = get_radeon_texture_image(texImage);
+
+    radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
+                 "%s(%p, tex %p, image %p) compressed %d.\n",
+                 __func__, ctx, texObj, image, compressed);
+
+    if (image->mt) {
+        /* Map the texture image read-only */
+        radeon_teximage_map(image, GL_FALSE);
+    } else {
+        /* Image hasn't been uploaded to a miptree yet */
+        assert(image->base.Data);
+    }
+
+    if (compressed) {
+        /* FIXME: this can't work for small textures (mips) which
+                 use different hw stride */
+        _mesa_get_compressed_teximage(ctx, target, level, pixels,
+                          texObj, texImage);
+    } else {
+        _mesa_get_teximage(ctx, target, level, format, type, pixels,
+                   texObj, texImage);
+    }
+
+    if (image->mt) {
+        radeon_teximage_unmap(image);
+    }
+}
+
+void
+radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+          GLenum format, GLenum type, GLvoid * pixels,
+          struct gl_texture_object *texObj,
+          struct gl_texture_image *texImage)
+{
+    radeon_get_tex_image(ctx, target, level, format, type, pixels,
+                 texObj, texImage, 0);
+}
+
+void
+radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level,
+                GLvoid *pixels,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+    radeon_get_tex_image(ctx, target, level, 0, 0, pixels,
+                 texObj, texImage, 1);
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.c b/src/mesa/drivers/dri/radeon/radeon_texture.c
index ff37fd3e86e..2b655fbd953 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texture.c
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.c
@@ -39,7 +39,6 @@
 #include "main/texstore.h"
 #include "main/teximage.h"
 #include "main/texobj.h"
-#include "main/texgetimage.h"
 
 #include "xmlpool.h"		/* for symbolic values of enum-type options */
 
@@ -559,6 +558,15 @@ gl_format radeonChooseTextureFormat(GLcontext * ctx,
 	case GL_COMPRESSED_SLUMINANCE_ALPHA:
 		return MESA_FORMAT_SLA8;
 
+	case GL_COMPRESSED_SRGB_S3TC_DXT1_EXT:
+		return MESA_FORMAT_SRGB_DXT1;
+	case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:
+		return MESA_FORMAT_SRGBA_DXT1;
+	case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
+		return MESA_FORMAT_SRGBA_DXT3;
+	case GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
+		return MESA_FORMAT_SRGBA_DXT5;
+
 	default:
 		_mesa_problem(ctx,
 			      "unexpected internalFormat 0x%x in %s",
@@ -664,6 +672,7 @@ static void radeon_store_teximage(GLcontext* ctx, int dims,
 		struct gl_texture_image *texImage,
 		int compressed)
 {
+	radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 	radeonTexObj *t = radeon_tex_obj(texObj);
 	radeon_texture_image* image = get_radeon_texture_image(texImage);
 
@@ -678,8 +687,7 @@ static void radeon_store_teximage(GLcontext* ctx, int dims,
 		dstRowStride = image->mt->levels[image->mtlevel].rowstride;
 	} else if (t->bo) {
 		/* TFP case */
-		/* TODO */
-		assert(0);
+		dstRowStride = get_texture_image_row_stride(rmesa, texImage->TexFormat, width, 0);
 	} else {
 		dstRowStride = _mesa_format_row_stride(texImage->TexFormat, texImage->Width);
 	}
@@ -999,61 +1007,18 @@ void radeonTexSubImage3D(GLcontext * ctx, GLenum target, GLint level,
 		format, type, pixels, packing, texObj, texImage, 0);
 }
 
-/**
- * Need to map texture image into memory before copying image data,
- * then unmap it.
- */
-static void
-radeon_get_tex_image(GLcontext * ctx, GLenum target, GLint level,
-		     GLenum format, GLenum type, GLvoid * pixels,
-		     struct gl_texture_object *texObj,
-		     struct gl_texture_image *texImage, int compressed)
+unsigned radeonIsFormatRenderable(gl_format mesa_format)
 {
-	radeon_texture_image *image = get_radeon_texture_image(texImage);
-
-	radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
-			"%s(%p, tex %p, image %p) compressed %d.\n",
-			__func__, ctx, texObj, image, compressed);
+	if (mesa_format == _dri_texformat_argb8888 || mesa_format == _dri_texformat_rgb565 ||
+		mesa_format == _dri_texformat_argb1555 || mesa_format == _dri_texformat_argb4444)
+		return 1;
 
-	if (image->mt) {
-		/* Map the texture image read-only */
-		radeon_teximage_map(image, GL_FALSE);
-	} else {
-		/* Image hasn't been uploaded to a miptree yet */
-		assert(image->base.Data);
-	}
-
-	if (compressed) {
-		/* FIXME: this can't work for small textures (mips) which
-		         use different hw stride */
-		_mesa_get_compressed_teximage(ctx, target, level, pixels,
-					      texObj, texImage);
-	} else {
-		_mesa_get_teximage(ctx, target, level, format, type, pixels,
-				   texObj, texImage);
-	}
-     
-	if (image->mt) {
-		radeon_teximage_unmap(image);
+	switch (mesa_format)
+	{
+		case MESA_FORMAT_Z16:
+		case MESA_FORMAT_S8_Z24:
+			return 1;
+		default:
+			return 0;
 	}
 }
-
-void
-radeonGetTexImage(GLcontext * ctx, GLenum target, GLint level,
-		  GLenum format, GLenum type, GLvoid * pixels,
-		  struct gl_texture_object *texObj,
-		  struct gl_texture_image *texImage)
-{
-	radeon_get_tex_image(ctx, target, level, format, type, pixels,
-			     texObj, texImage, 0);
-}
-
-void
-radeonGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level,
-			    GLvoid *pixels,
-			    struct gl_texture_object *texObj,
-			    struct gl_texture_image *texImage)
-{
-	radeon_get_tex_image(ctx, target, level, 0, 0, pixels,
-			     texObj, texImage, 1);
-}
diff --git a/src/mesa/drivers/dri/radeon/radeon_texture.h b/src/mesa/drivers/dri/radeon/radeon_texture.h
index f09dd652142..4ce639ea34e 100644
--- a/src/mesa/drivers/dri/radeon/radeon_texture.h
+++ b/src/mesa/drivers/dri/radeon/radeon_texture.h
@@ -135,4 +135,6 @@ void radeonCopyTexSubImage2D(GLcontext *ctx, GLenum target, GLint level,
 			GLint x, GLint y,
 			GLsizei width, GLsizei height);
 
+unsigned radeonIsFormatRenderable(gl_format mesa_format);
+
 #endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_tile.c b/src/mesa/drivers/dri/radeon/radeon_tile.c
new file mode 100644
index 00000000000..403da110106
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tile.c
@@ -0,0 +1,512 @@
+/*
+ * Copyright (C) 2010 Maciej Cencora <m.cencora@gmail.com>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include "radeon_tile.h"
+
+#include <stdint.h>
+#include <string.h>
+
+#include "main/macros.h"
+#include "radeon_debug.h"
+
+#define MICRO_TILE_SIZE 32
+
+static void micro_tile_8_x_4_8bit(const void * const src, unsigned src_pitch,
+                                  void * const dst, unsigned dst_pitch,
+                                  unsigned width, unsigned height)
+{
+    unsigned row; /* current source row */
+    unsigned col; /* current source column */
+    unsigned k; /* number of processed tiles */
+    const unsigned tile_width = 8, tile_height = 4;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint8_t *src2 = (uint8_t *)src + src_pitch * row + col;
+            uint8_t *dst2 = (uint8_t *)dst + row * dst_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint8_t);
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint8_t));
+                dst2 += tile_width;
+                src2 += src_pitch;
+            }
+        }
+    }
+}
+
+static void micro_tile_4_x_4_16bit(const void * const src, unsigned src_pitch,
+                                   void * const dst, unsigned dst_pitch,
+                                   unsigned width, unsigned height)
+{
+    unsigned row; /* current source row */
+    unsigned col; /* current source column */
+    unsigned k; /* number of processed tiles */
+    const unsigned tile_width = 4, tile_height = 4;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint16_t *src2 = (uint16_t *)src + src_pitch * row + col;
+            uint16_t *dst2 = (uint16_t *)dst + row * dst_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint16_t);
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint16_t));
+                dst2 += tile_width;
+                src2 += src_pitch;
+            }
+        }
+    }
+}
+
+static void micro_tile_8_x_2_16bit(const void * const src, unsigned src_pitch,
+                                   void * const dst, unsigned dst_pitch,
+                                   unsigned width, unsigned height)
+{
+    unsigned row; /* current source row */
+    unsigned col; /* current source column */
+    unsigned k; /* number of processed tiles */
+    const unsigned tile_width = 8, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint16_t *src2 = (uint16_t *)src + src_pitch * row + col;
+            uint16_t *dst2 = (uint16_t *)dst + row * dst_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint16_t);
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint16_t));
+                dst2 += tile_width;
+                src2 += src_pitch;
+            }
+        }
+    }
+}
+
+static void micro_tile_4_x_2_32bit(const void * const src, unsigned src_pitch,
+                                   void * const dst, unsigned dst_pitch,
+                                   unsigned width, unsigned height)
+{
+    unsigned row; /* current source row */
+    unsigned col; /* current source column */
+    unsigned k; /* number of processed tiles */
+    const unsigned tile_width = 4, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint32_t *src2 = (uint32_t *)src + src_pitch * row + col;
+            uint32_t *dst2 = (uint32_t *)dst + row * dst_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint32_t);
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint32_t));
+                dst2 += tile_width;
+                src2 += src_pitch;
+            }
+        }
+    }
+}
+
+static void micro_tile_2_x_2_64bit(const void * const src, unsigned src_pitch,
+                                   void * const dst, unsigned dst_pitch,
+                                   unsigned width, unsigned height)
+{
+    unsigned row; /* current source row */
+    unsigned col; /* current source column */
+    unsigned k; /* number of processed tiles */
+    const unsigned tile_width = 2, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint64_t *src2 = (uint64_t *)src + src_pitch * row + col;
+            uint64_t *dst2 = (uint64_t *)dst + row * dst_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint64_t);
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint64_t));
+                dst2 += tile_width;
+                src2 += src_pitch;
+            }
+        }
+    }
+}
+
+static void micro_tile_1_x_1_128bit(const void * src, unsigned src_pitch,
+                                    void * dst, unsigned dst_pitch,
+                                    unsigned width, unsigned height)
+{
+    unsigned i, j;
+    const unsigned elem_size = 16; /* sizeof(uint128_t) */
+
+    for (j = 0; j < height; ++j)
+    {
+        for (i = 0; i < width; ++i)
+        {
+            memcpy(dst, src, width * elem_size);
+            dst += dst_pitch * elem_size;
+            src += src_pitch * elem_size;
+        }
+    }
+}
+
+void tile_image(const void * src, unsigned src_pitch,
+                void *dst, unsigned dst_pitch,
+                gl_format format, unsigned width, unsigned height)
+{
+    assert(src_pitch >= width);
+    assert(dst_pitch >= width);
+
+    radeon_print(RADEON_TEXTURE, RADEON_TRACE,
+                 "Software tiling: src_pitch %d, dst_pitch %d, width %d, height %d, bpp %d\n",
+                 src_pitch, dst_pitch, width, height, _mesa_get_format_bytes(format));
+
+    switch (_mesa_get_format_bytes(format))
+    {
+        case 16:
+            micro_tile_1_x_1_128bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 8:
+            micro_tile_2_x_2_64bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 4:
+            micro_tile_4_x_2_32bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 2:
+            if (_mesa_get_format_bits(format, GL_DEPTH_BITS))
+            {
+                micro_tile_4_x_4_16bit(src, src_pitch, dst, dst_pitch, width, height);
+            }
+            else
+            {
+                micro_tile_8_x_2_16bit(src, src_pitch, dst, dst_pitch, width, height);
+            }
+            break;
+        case 1:
+            micro_tile_8_x_4_8bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        default:
+            assert(0);
+            break;
+    }
+}
+
+static void micro_untile_8_x_4_8bit(const void * const src, unsigned src_pitch,
+                                    void * const dst, unsigned dst_pitch,
+                                    unsigned width, unsigned height)
+{
+    unsigned row; /* current destination row */
+    unsigned col; /* current destination column */
+    unsigned k; /* current tile number */
+    const unsigned tile_width = 8, tile_height = 4;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    assert(src_pitch % tile_width == 0);
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint8_t *src2 = (uint8_t *)src + row * src_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint8_t);
+            uint8_t *dst2 = (uint8_t *)dst + dst_pitch * row + col;
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint8_t));
+                dst2 += dst_pitch;
+                src2 += tile_width;
+            }
+        }
+    }
+}
+
+static void micro_untile_8_x_2_16bit(const void * const src, unsigned src_pitch,
+                                     void * const dst, unsigned dst_pitch,
+                                     unsigned width, unsigned height)
+{
+    unsigned row; /* current destination row */
+    unsigned col; /* current destination column */
+    unsigned k; /* current tile number */
+    const unsigned tile_width = 8, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    assert(src_pitch % tile_width == 0);
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint16_t *src2 = (uint16_t *)src + row * src_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint16_t);
+            uint16_t *dst2 = (uint16_t *)dst + dst_pitch * row + col;
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint16_t));
+                dst2 += dst_pitch;
+                src2 += tile_width;
+            }
+        }
+    }
+}
+
+static void micro_untile_4_x_4_16bit(const void * const src, unsigned src_pitch,
+                                     void * const dst, unsigned dst_pitch,
+                                     unsigned width, unsigned height)
+{
+    unsigned row; /* current destination row */
+    unsigned col; /* current destination column */
+    unsigned k; /* current tile number */
+    const unsigned tile_width = 4, tile_height = 4;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    assert(src_pitch % tile_width == 0);
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint16_t *src2 = (uint16_t *)src + row * src_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint16_t);
+            uint16_t *dst2 = (uint16_t *)dst + dst_pitch * row + col;
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint16_t));
+                dst2 += dst_pitch;
+                src2 += tile_width;
+            }
+        }
+    }
+}
+
+static void micro_untile_4_x_2_32bit(const void * const src, unsigned src_pitch,
+                                     void * const dst, unsigned dst_pitch,
+                                     unsigned width, unsigned height)
+{
+    unsigned row; /* current destination row */
+    unsigned col; /* current destination column */
+    unsigned k; /* current tile number */
+    const unsigned tile_width = 4, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    assert(src_pitch % tile_width == 0);
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint32_t *src2 = (uint32_t *)src + row * src_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint32_t);
+            uint32_t *dst2 = (uint32_t *)dst + dst_pitch * row + col;
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint32_t));
+                dst2 += dst_pitch;
+                src2 += tile_width;
+            }
+        }
+    }
+}
+
+static void micro_untile_2_x_2_64bit(const void * const src, unsigned src_pitch,
+                                     void * const dst, unsigned dst_pitch,
+                                     unsigned width, unsigned height)
+{
+    unsigned row; /* current destination row */
+    unsigned col; /* current destination column */
+    unsigned k; /* current tile number */
+    const unsigned tile_width = 2, tile_height = 2;
+    const unsigned tiles_in_row = (width + (tile_width - 1)) / tile_width;
+
+    assert(src_pitch % tile_width == 0);
+
+    k = 0;
+    for (row = 0; row < height; row += tile_height)
+    {
+        for (col = 0; col < width; col += tile_width, ++k)
+        {
+            uint64_t *src2 = (uint64_t *)src + row * src_pitch +
+                             (k % tiles_in_row) * MICRO_TILE_SIZE / sizeof(uint64_t);
+            uint64_t *dst2 = (uint64_t *)dst + dst_pitch * row + col;
+            unsigned j;
+
+            for (j = 0; j < MIN2(tile_height, height - row); ++j)
+            {
+                unsigned columns = MIN2(tile_width, width - col);
+                memcpy(dst2, src2, columns * sizeof(uint64_t));
+                dst2 += dst_pitch;
+                src2 += tile_width;
+            }
+        }
+    }
+}
+
+static void micro_untile_1_x_1_128bit(const void * src, unsigned src_pitch,
+                                      void * dst, unsigned dst_pitch,
+                                      unsigned width, unsigned height)
+{
+    unsigned i, j;
+    const unsigned elem_size = 16; /* sizeof(uint128_t) */
+
+    for (j = 0; j < height; ++j)
+    {
+        for (i = 0; i < width; ++i)
+        {
+            memcpy(dst, src, width * elem_size);
+            dst += dst_pitch * elem_size;
+            src += src_pitch * elem_size;
+        }
+    }
+}
+
+void untile_image(const void * src, unsigned src_pitch,
+                  void *dst, unsigned dst_pitch,
+                  gl_format format, unsigned width, unsigned height)
+{
+    assert(src_pitch >= width);
+    assert(dst_pitch >= width);
+
+    radeon_print(RADEON_TEXTURE, RADEON_TRACE,
+                 "Software untiling: src_pitch %d, dst_pitch %d, width %d, height %d, bpp %d\n",
+                 src_pitch, dst_pitch, width, height, _mesa_get_format_bytes(format));
+
+    switch (_mesa_get_format_bytes(format))
+    {
+        case 16:
+            micro_untile_1_x_1_128bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 8:
+            micro_untile_2_x_2_64bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 4:
+            micro_untile_4_x_2_32bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        case 2:
+            if (_mesa_get_format_bits(format, GL_DEPTH_BITS))
+            {
+                micro_untile_4_x_4_16bit(src, src_pitch, dst, dst_pitch, width, height);
+            }
+            else
+            {
+                micro_untile_8_x_2_16bit(src, src_pitch, dst, dst_pitch, width, height);
+            }
+            break;
+        case 1:
+            micro_untile_8_x_4_8bit(src, src_pitch, dst, dst_pitch, width, height);
+            break;
+        default:
+            assert(0);
+            break;
+    }
+}
+
+void get_tile_size(gl_format format, unsigned *block_width, unsigned *block_height)
+{
+    switch (_mesa_get_format_bytes(format))
+    {
+        case 16:
+            *block_width = 1;
+            *block_height = 1;
+            break;
+        case 8:
+            *block_width = 2;
+            *block_height = 2;
+            break;
+        case 4:
+            *block_width = 4;
+            *block_height = 2;
+            break;
+        case 2:
+            if (_mesa_get_format_bits(format, GL_DEPTH_BITS))
+            {
+                *block_width = 4;
+                *block_height = 4;
+            }
+            else
+            {
+                *block_width = 8;
+                *block_height = 2;
+            }
+            break;
+        case 1:
+            *block_width = 8;
+            *block_height = 4;
+            break;
+        default:
+            assert(0);
+            break;
+    }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_tile.h b/src/mesa/drivers/dri/radeon/radeon_tile.h
new file mode 100644
index 00000000000..31d9c5611c3
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tile.h
@@ -0,0 +1,38 @@
+/*
+ * Copyright (C) 2010 Maciej Cencora <m.cencora@gmail.com>
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <main/formats.h>
+
+void tile_image(const void * src, unsigned src_pitch,
+                void *dst, unsigned dst_pitch,
+                gl_format format, unsigned width, unsigned height);
+
+void untile_image(const void * src, unsigned src_pitch,
+                  void *dst, unsigned dst_pitch,
+                  gl_format format, unsigned width, unsigned height);
+
+void get_tile_size(gl_format format, unsigned *block_width, unsigned *block_height);
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_dri.c b/src/mesa/drivers/dri/radeon/server/radeon_dri.c
deleted file mode 100644
index 7ead588dac8..00000000000
--- a/src/mesa/drivers/dri/radeon/server/radeon_dri.c
+++ /dev/null
@@ -1,1337 +0,0 @@
-/**
- * \file server/radeon_dri.c
- * \brief File to perform the device-specific initialization tasks typically
- * done in the X server.
- *
- * Here they are converted to run in the client (or perhaps a standalone
- * process), and to work with the frame buffer device rather than the X
- * server infrastructure.
- */
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
-
-#include "driver.h"
-#include "drm.h"
-#include "memops.h"
-
-#include "radeon.h"
-#include "radeon_dri.h"
-#include "radeon_macros.h"
-#include "radeon_reg.h"
-#include "drm_sarea.h"
-
-static size_t radeon_drm_page_size;
-
-static int RadeonSetParam(const DRIDriverContext *ctx, int param, int value)
-{
-   drm_radeon_setparam_t sp;
-
-   memset(&sp, 0, sizeof(sp));
-   sp.param = param;
-   sp.value = value;
-
-   if (drmCommandWrite(ctx->drmFD, DRM_RADEON_SETPARAM, &sp, sizeof(sp))) {
-     return -1;
-   }
-
-   return 0;
-}
-
-/**
- * \brief Wait for free FIFO entries.
- *
- * \param ctx display handle.
- * \param entries number of free entries to wait.
- *
- * It polls the free entries from the chip until it reaches the requested value
- * or a timeout (3000 tries) occurs. Aborts the program if the FIFO times out.
- */
-static void RADEONWaitForFifo( const DRIDriverContext *ctx,
-			       int entries )
-{
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   int i;
-
-   for (i = 0; i < 3000; i++) {
-      int fifo_slots =
-	 INREG(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
-      if (fifo_slots >= entries) return;
-   }
-
-   /* There are recoveries possible, but I haven't seen them work
-    * in practice:
-    */
-   fprintf(stderr, "FIFO timed out: %d entries, stat=0x%08x\n",
-	   INREG(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK,
-	   INREG(RADEON_RBBM_STATUS));
-   exit(1);
-}
-
-/**
- * \brief Read a PLL register.
- *
- * \param ctx display handle.
- * \param addr PLL register index.
- *
- * \return value of the PLL register.
- */
-static unsigned int RADEONINPLL( const DRIDriverContext *ctx, int addr)
-{
-    unsigned char *RADEONMMIO = ctx->MMIOAddress;
-    unsigned int data;
-
-    OUTREG8(RADEON_CLOCK_CNTL_INDEX, addr & 0x3f);
-    data = INREG(RADEON_CLOCK_CNTL_DATA);
-
-    return data;
-}
-
-/**
- * \brief Reset graphics card to known state.
- *
- * \param ctx display handle.
- *
- * Resets the values of several Radeon registers.
- */
-static void RADEONEngineReset( const DRIDriverContext *ctx )
-{
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   unsigned int clock_cntl_index;
-   unsigned int mclk_cntl;
-   unsigned int rbbm_soft_reset;
-   unsigned int host_path_cntl;
-   int i;
-
-   OUTREGP(RADEON_RB2D_DSTCACHE_CTLSTAT,
-	   RADEON_RB2D_DC_FLUSH_ALL,
-	   ~RADEON_RB2D_DC_FLUSH_ALL);
-   for (i = 0; i < 512; i++) {
-      if (!(INREG(RADEON_RB2D_DSTCACHE_CTLSTAT) & RADEON_RB2D_DC_BUSY))
-	 break;
-   }
-
-   clock_cntl_index = INREG(RADEON_CLOCK_CNTL_INDEX);
-
-   mclk_cntl = INPLL(ctx, RADEON_MCLK_CNTL);
-   OUTPLL(RADEON_MCLK_CNTL, (mclk_cntl |
-			     RADEON_FORCEON_MCLKA |
-			     RADEON_FORCEON_MCLKB |
-			     RADEON_FORCEON_YCLKA |
-			     RADEON_FORCEON_YCLKB |
-			     RADEON_FORCEON_MC |
-			     RADEON_FORCEON_AIC));
-
-   /* Soft resetting HDP thru RBBM_SOFT_RESET register can cause some
-    * unexpected behaviour on some machines.  Here we use
-    * RADEON_HOST_PATH_CNTL to reset it.
-    */
-   host_path_cntl = INREG(RADEON_HOST_PATH_CNTL);
-   rbbm_soft_reset = INREG(RADEON_RBBM_SOFT_RESET);
-
-   OUTREG(RADEON_RBBM_SOFT_RESET, (rbbm_soft_reset |
-				   RADEON_SOFT_RESET_CP |
-				   RADEON_SOFT_RESET_HI |
-				   RADEON_SOFT_RESET_SE |
-				   RADEON_SOFT_RESET_RE |
-				   RADEON_SOFT_RESET_PP |
-				   RADEON_SOFT_RESET_E2 |
-				   RADEON_SOFT_RESET_RB));
-   INREG(RADEON_RBBM_SOFT_RESET);
-   OUTREG(RADEON_RBBM_SOFT_RESET, (rbbm_soft_reset & 
-				   (unsigned int) ~(RADEON_SOFT_RESET_CP |
-						    RADEON_SOFT_RESET_HI |
-						    RADEON_SOFT_RESET_SE |
-						    RADEON_SOFT_RESET_RE |
-						    RADEON_SOFT_RESET_PP |
-						    RADEON_SOFT_RESET_E2 |
-						    RADEON_SOFT_RESET_RB)));
-   INREG(RADEON_RBBM_SOFT_RESET);
-
-   OUTREG(RADEON_HOST_PATH_CNTL, host_path_cntl | RADEON_HDP_SOFT_RESET);
-   INREG(RADEON_HOST_PATH_CNTL);
-   OUTREG(RADEON_HOST_PATH_CNTL, host_path_cntl);
-
-   OUTREG(RADEON_RBBM_SOFT_RESET, rbbm_soft_reset);
-
-   OUTREG(RADEON_CLOCK_CNTL_INDEX, clock_cntl_index);
-   OUTPLL(RADEON_MCLK_CNTL, mclk_cntl);
-}
-
-/**
- * \brief Restore the drawing engine.
- *
- * \param ctx display handle
- *
- * Resets the graphics card and sets initial values for several registers of
- * the card's drawing engine.
- *
- * Turns on the radeon command processor engine (i.e., the ringbuffer).
- */
-static int RADEONEngineRestore( const DRIDriverContext *ctx )
-{
-   RADEONInfoPtr info = ctx->driverPrivate;
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   int pitch64, datatype, dp_gui_master_cntl, err;
-
-   fprintf(stderr, "%s\n", __FUNCTION__);
-
-   OUTREG(RADEON_RB3D_CNTL, 0);
-   RADEONEngineReset( ctx );
-
-   switch (ctx->bpp) {
-   case 16: datatype = 4; break;
-   case 32: datatype = 6; break;
-   default: return 0;
-   }
-
-   dp_gui_master_cntl =
-      ((datatype << RADEON_GMC_DST_DATATYPE_SHIFT)
-       | RADEON_GMC_CLR_CMP_CNTL_DIS);
-
-   pitch64 = ((ctx->shared.virtualWidth * (ctx->bpp / 8) + 0x3f)) >> 6;
-
-   RADEONWaitForFifo(ctx, 1);
-   OUTREG(RADEON_DEFAULT_OFFSET, ((INREG(RADEON_DEFAULT_OFFSET) & 0xC0000000)
-				  | (pitch64 << 22)));
-
-   RADEONWaitForFifo(ctx, 1);
-   OUTREG(RADEON_SURFACE_CNTL, RADEON_SURF_TRANSLATION_DIS); 
-
-   RADEONWaitForFifo(ctx, 1);
-   OUTREG(RADEON_DEFAULT_SC_BOTTOM_RIGHT, (RADEON_DEFAULT_SC_RIGHT_MAX
-					   | RADEON_DEFAULT_SC_BOTTOM_MAX));
-
-   RADEONWaitForFifo(ctx, 1);
-   OUTREG(RADEON_DP_GUI_MASTER_CNTL, (dp_gui_master_cntl
-				      | RADEON_GMC_BRUSH_SOLID_COLOR
-				      | RADEON_GMC_SRC_DATATYPE_COLOR));
-
-   RADEONWaitForFifo(ctx, 7);
-   OUTREG(RADEON_DST_LINE_START,    0);
-   OUTREG(RADEON_DST_LINE_END,      0);
-   OUTREG(RADEON_DP_BRUSH_FRGD_CLR, 0xffffffff);
-   OUTREG(RADEON_DP_BRUSH_BKGD_CLR, 0);
-   OUTREG(RADEON_DP_SRC_FRGD_CLR,   0xffffffff);
-   OUTREG(RADEON_DP_SRC_BKGD_CLR,   0);
-   OUTREG(RADEON_DP_WRITE_MASK,     0xffffffff);
-   OUTREG(RADEON_AUX_SC_CNTL,       0);
-
-/*    RADEONWaitForIdleMMIO(ctx); */
-   usleep(100); 
-
-
-   OUTREG(RADEON_GEN_INT_CNTL, info->gen_int_cntl);
-   if (info->colorTiling)
-	   info->crtc_offset_cntl |= RADEON_CRTC_TILE_EN;
-   OUTREG(RADEON_CRTC_OFFSET_CNTL, info->crtc_offset_cntl);
-
-   /* Initialize and start the CP if required */
-   if ((err = drmCommandNone(ctx->drmFD, DRM_RADEON_CP_START)) != 0) {
-      fprintf(stderr, "%s: CP start %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   return 1;
-}
-
-
-/**
- * \brief Shutdown the drawing engine.
- *
- * \param ctx display handle
- *
- * Turns off the command processor engine & restores the graphics card
- * to a state that fbdev understands.
- */
-static int RADEONEngineShutdown( const DRIDriverContext *ctx )
-{
-   drm_radeon_cp_stop_t  stop;
-   int              ret, i;
-
-   stop.flush = 1;
-   stop.idle  = 1;
-
-   ret = drmCommandWrite(ctx->drmFD, DRM_RADEON_CP_STOP, &stop, 
-			 sizeof(drm_radeon_cp_stop_t));
-
-   if (ret == 0) {
-      return 0;
-   } else if (errno != EBUSY) {
-      return -errno;
-   }
-
-   stop.flush = 0;
- 
-   i = 0;
-   do {
-      ret = drmCommandWrite(ctx->drmFD, DRM_RADEON_CP_STOP, &stop, 
-			    sizeof(drm_radeon_cp_stop_t));
-   } while (ret && errno == EBUSY && i++ < 10);
-
-   if (ret == 0) {
-      return 0;
-   } else if (errno != EBUSY) {
-      return -errno;
-   }
-
-   stop.idle = 0;
-
-   if (drmCommandWrite(ctx->drmFD, DRM_RADEON_CP_STOP,
-		       &stop, sizeof(drm_radeon_cp_stop_t))) {
-      return -errno;
-   } else {
-      return 0;
-   }
-}
-
-/**
- * \brief Compute base 2 logarithm.
- *
- * \param val value.
- * 
- * \return base 2 logarithm of \p val.
- */
-static int RADEONMinBits(int val)
-{
-   int  bits;
-
-   if (!val) return 1;
-   for (bits = 0; val; val >>= 1, ++bits);
-   return bits;
-}
-
-/**
- * \brief Initialize the AGP state
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return one on success, or zero on failure.
- * 
- * Acquires and enables the AGP device. Reserves memory in the AGP space for
- * the ring buffer, vertex buffers and textures. Initialize the Radeon
- * registers to point to that memory and add client mappings.
- */
-static int RADEONDRIAgpInit( const DRIDriverContext *ctx, RADEONInfoPtr info)
-{
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   unsigned long  mode;
-   int            ret;
-   int            s, l;
-
-   if (drmAgpAcquire(ctx->drmFD) < 0) {
-      fprintf(stderr, "[gart] AGP not available\n");
-      return 0;
-   }
-    
-   /* Modify the mode if the default mode is not appropriate for this
-    * particular combination of graphics card and AGP chipset.
-    */
-   mode   = drmAgpGetMode(ctx->drmFD);	/* Default mode */
-
-   /* Disable fast write entirely - too many lockups.
-    */
-   mode &= ~RADEON_AGP_MODE_MASK;
-   switch (ctx->agpmode) {
-   case 4:          mode |= RADEON_AGP_4X_MODE;
-   case 2:          mode |= RADEON_AGP_2X_MODE;
-   case 1: default: mode |= RADEON_AGP_1X_MODE;
-   }
-
-   if (drmAgpEnable(ctx->drmFD, mode) < 0) {
-      fprintf(stderr, "[gart] AGP not enabled\n");
-      drmAgpRelease(ctx->drmFD);
-      return 0;
-   }
-   else
-     fprintf(stderr, "[gart] AGP enabled at %dx\n", ctx->agpmode);
-
-   /* Workaround for some hardware bugs */
-   if (info->ChipFamily < CHIP_FAMILY_R200)
-      OUTREG(RADEON_AGP_CNTL, INREG(RADEON_AGP_CNTL) | 0x000e0000);
-
-   info->gartOffset = 0;
-
-   if ((ret = drmAgpAlloc(ctx->drmFD, info->gartSize*1024*1024, 0, NULL,
-			  &info->gartMemHandle)) < 0) {
-      fprintf(stderr, "[gart] Out of memory (%d)\n", ret);
-      drmAgpRelease(ctx->drmFD);
-      return 0;
-   }
-   fprintf(stderr,
-	   "[gart] %d kB allocated with handle 0x%08x\n",
-	   info->gartSize*1024, (unsigned)info->gartMemHandle);
-    
-   if (drmAgpBind(ctx->drmFD,
-		  info->gartMemHandle, info->gartOffset) < 0) {
-      fprintf(stderr, "[gart] Could not bind\n");
-      drmAgpFree(ctx->drmFD, info->gartMemHandle);
-      drmAgpRelease(ctx->drmFD);
-      return 0;
-   }
-
-   /* Initialize the CP ring buffer data */
-   info->ringStart       = info->gartOffset;
-   info->ringMapSize     = info->ringSize*1024*1024 + radeon_drm_page_size;
-
-   info->ringReadOffset  = info->ringStart + info->ringMapSize;
-   info->ringReadMapSize = radeon_drm_page_size;
-
-   /* Reserve space for vertex/indirect buffers */
-   info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
-   info->bufMapSize      = info->bufSize*1024*1024;
-
-   /* Reserve the rest for AGP textures */
-   info->gartTexStart     = info->bufStart + info->bufMapSize;
-   s = (info->gartSize*1024*1024 - info->gartTexStart);
-   l = RADEONMinBits((s-1) / RADEON_NR_TEX_REGIONS);
-   if (l < RADEON_LOG_TEX_GRANULARITY) l = RADEON_LOG_TEX_GRANULARITY;
-   info->gartTexMapSize   = (s >> l) << l;
-   info->log2GARTTexGran  = l;
-
-   if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
-		 DRM_AGP, DRM_READ_ONLY, &info->ringHandle) < 0) {
-      fprintf(stderr, "[gart] Could not add ring mapping\n");
-      return 0;
-   }
-   fprintf(stderr, "[gart] ring handle = 0x%08x\n", info->ringHandle);
-    
-
-   if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
-		 DRM_AGP, DRM_READ_ONLY, &info->ringReadPtrHandle) < 0) {
-      fprintf(stderr,
-	      "[gart] Could not add ring read ptr mapping\n");
-      return 0;
-   }
-    
-   fprintf(stderr,
- 	   "[gart] ring read ptr handle = 0x%08lx\n",
-	   info->ringReadPtrHandle);
-    
-   if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
-		 DRM_AGP, 0, &info->bufHandle) < 0) {
-      fprintf(stderr,
-	      "[gart] Could not add vertex/indirect buffers mapping\n");
-      return 0;
-   }
-   fprintf(stderr,
- 	   "[gart] vertex/indirect buffers handle = 0x%08x\n",
-	   info->bufHandle);
-
-   if (drmAddMap(ctx->drmFD, info->gartTexStart, info->gartTexMapSize,
-		 DRM_AGP, 0, &info->gartTexHandle) < 0) {
-      fprintf(stderr,
-	      "[gart] Could not add AGP texture map mapping\n");
-      return 0;
-   }
-   fprintf(stderr,
- 	   "[gart] AGP texture map handle = 0x%08lx\n",
-	   info->gartTexHandle);
-
-   /* Initialize Radeon's AGP registers */
-   /* Ring buffer is at AGP offset 0 */
-   OUTREG(RADEON_AGP_BASE, info->ringHandle);
-
-   return 1;
-}
-
-/* Initialize the PCI GART state.  Request memory for use in PCI space,
- * and initialize the Radeon registers to point to that memory.
- */
-static int RADEONDRIPciInit(const DRIDriverContext *ctx, RADEONInfoPtr info)
-{
-    int  ret;
-    int  flags = DRM_READ_ONLY | DRM_LOCKED | DRM_KERNEL;
-    int            s, l;
-
-    ret = drmScatterGatherAlloc(ctx->drmFD, info->gartSize*1024*1024,
-				&info->gartMemHandle);
-    if (ret < 0) {
-	fprintf(stderr, "[pci] Out of memory (%d)\n", ret);
-	return 0;
-    }
-    fprintf(stderr,
-	       "[pci] %d kB allocated with handle 0x%08lx\n",
-	       info->gartSize*1024, info->gartMemHandle);
-
-   info->gartOffset = 0;
-   
-   /* Initialize the CP ring buffer data */
-   info->ringStart       = info->gartOffset;
-   info->ringMapSize     = info->ringSize*1024*1024 + radeon_drm_page_size;
-
-   info->ringReadOffset  = info->ringStart + info->ringMapSize;
-   info->ringReadMapSize = radeon_drm_page_size;
-
-   /* Reserve space for vertex/indirect buffers */
-   info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
-   info->bufMapSize      = info->bufSize*1024*1024;
-
-   /* Reserve the rest for AGP textures */
-   info->gartTexStart     = info->bufStart + info->bufMapSize;
-   s = (info->gartSize*1024*1024 - info->gartTexStart);
-   l = RADEONMinBits((s-1) / RADEON_NR_TEX_REGIONS);
-   if (l < RADEON_LOG_TEX_GRANULARITY) l = RADEON_LOG_TEX_GRANULARITY;
-   info->gartTexMapSize   = (s >> l) << l;
-   info->log2GARTTexGran  = l;
-
-    if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
-		  DRM_SCATTER_GATHER, flags, &info->ringHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add ring mapping\n");
-	return 0;
-    }
-    fprintf(stderr,
-	       "[pci] ring handle = 0x%08x\n", info->ringHandle);
-
-    if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
-		  DRM_SCATTER_GATHER, flags, &info->ringReadPtrHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add ring read ptr mapping\n");
-	return 0;
-    }
-    fprintf(stderr,
- 	       "[pci] ring read ptr handle = 0x%08lx\n",
-	       info->ringReadPtrHandle);
-
-    if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
-		  DRM_SCATTER_GATHER, 0, &info->bufHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add vertex/indirect buffers mapping\n");
-	return 0;
-    }
-    fprintf(stderr,
- 	       "[pci] vertex/indirect buffers handle = 0x%08lx\n",
-	       info->bufHandle);
-
-    if (drmAddMap(ctx->drmFD, info->gartTexStart, info->gartTexMapSize,
-		  DRM_SCATTER_GATHER, 0, &info->gartTexHandle) < 0) {
-	fprintf(stderr,
-		   "[pci] Could not add GART texture map mapping\n");
-	return 0;
-    }
-    fprintf(stderr,
- 	       "[pci] GART texture map handle = 0x%08x\n",
-	       info->gartTexHandle);
-
-    return 1;
-}
-
-
-/**
- * \brief Initialize the kernel data structures and enable the CP engine.
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return non-zero on success, or zero on failure.
- *
- * This function is a wrapper around the DRM_RADEON_CP_INIT command, passing
- * all the parameters in a drm_radeon_init_t structure.
- */
-static int RADEONDRIKernelInit( const DRIDriverContext *ctx,
-			       RADEONInfoPtr info)
-{
-   int cpp = ctx->bpp / 8;
-   drm_radeon_init_t  drmInfo;
-   int ret;
-
-   memset(&drmInfo, 0, sizeof(drm_radeon_init_t));
-
-   if ( (info->ChipFamily == CHIP_FAMILY_R200) ||
-	(info->ChipFamily == CHIP_FAMILY_RV250) ||
-	(info->ChipFamily == CHIP_FAMILY_M9) ||
-	(info->ChipFamily == CHIP_FAMILY_RV280) )
-      drmInfo.func             = RADEON_INIT_R200_CP;
-   else
-      drmInfo.func             = RADEON_INIT_CP;
-
-   /* This is the struct passed to the kernel module for its initialization */
-   drmInfo.sarea_priv_offset   = sizeof(drm_sarea_t);
-   drmInfo.is_pci              = ctx->isPCI;
-   drmInfo.cp_mode             = RADEON_DEFAULT_CP_BM_MODE;
-   drmInfo.gart_size            = info->gartSize*1024*1024;
-   drmInfo.ring_size           = info->ringSize*1024*1024;
-   drmInfo.usec_timeout        = 1000;
-   drmInfo.fb_bpp              = ctx->bpp;
-   drmInfo.depth_bpp           = ctx->bpp;
-   drmInfo.front_offset        = info->frontOffset;
-   drmInfo.front_pitch         = info->frontPitch * cpp;
-   drmInfo.back_offset         = info->backOffset;
-   drmInfo.back_pitch          = info->backPitch * cpp;
-   drmInfo.depth_offset        = info->depthOffset;
-   drmInfo.depth_pitch         = info->depthPitch * cpp;
-   drmInfo.fb_offset           = info->LinearAddr;
-   drmInfo.mmio_offset         = info->registerHandle;
-   drmInfo.ring_offset         = info->ringHandle;
-   drmInfo.ring_rptr_offset    = info->ringReadPtrHandle;
-   drmInfo.buffers_offset      = info->bufHandle;
-   drmInfo.gart_textures_offset = info->gartTexHandle;
-
-   ret = drmCommandWrite(ctx->drmFD, DRM_RADEON_CP_INIT, &drmInfo, 
-			 sizeof(drm_radeon_init_t));
-
-   return ret >= 0;
-}
-
-
-/**
- * \brief Initialize the AGP heap.
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * This function is a wrapper around the DRM_RADEON_INIT_HEAP command, passing
- * all the parameters in a drm_radeon_mem_init_heap structure.
- */
-static void RADEONDRIAgpHeapInit(const DRIDriverContext *ctx,
-				 RADEONInfoPtr info)
-{
-   drm_radeon_mem_init_heap_t drmHeap;
-
-   /* Start up the simple memory manager for gart space */
-   drmHeap.region = RADEON_MEM_REGION_GART;
-   drmHeap.start  = 0;
-   drmHeap.size   = info->gartTexMapSize;
-    
-   if (drmCommandWrite(ctx->drmFD, DRM_RADEON_INIT_HEAP,
-		       &drmHeap, sizeof(drmHeap))) {
-      fprintf(stderr,
-	      "[drm] Failed to initialized gart heap manager\n");
-   } else {
-      fprintf(stderr,
-	      "[drm] Initialized kernel gart heap manager, %d\n",
-	      info->gartTexMapSize);
-   }
-}
-
-/**
- * \brief Add a map for the vertex buffers that will be accessed by any
- * DRI-based clients.
- * 
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return one on success, or zero on failure.
- *
- * Calls drmAddBufs() with the previously allocated vertex buffers.
- */
-static int RADEONDRIBufInit( const DRIDriverContext *ctx, RADEONInfoPtr info )
-{
-   /* Initialize vertex buffers */
-   info->bufNumBufs = drmAddBufs(ctx->drmFD,
-				 info->bufMapSize / RADEON_BUFFER_SIZE,
-				 RADEON_BUFFER_SIZE,
-				 ctx->isPCI ? DRM_SG_BUFFER : DRM_AGP_BUFFER,
-				 info->bufStart);
-
-   if (info->bufNumBufs <= 0) {
-      fprintf(stderr,
-	      "[drm] Could not create vertex/indirect buffers list\n");
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] Added %d %d byte vertex/indirect buffers\n",
-	   info->bufNumBufs, RADEON_BUFFER_SIZE);
-   
-   return 1;
-}
-
-/**
- * \brief Install an IRQ handler.
- * 
- * \param ctx display handle.
- * \param info driver private data.
- *
- * Attempts to install an IRQ handler via drmCtlInstHandler(), falling back to
- * IRQ-free operation on failure.
- */
-static void RADEONDRIIrqInit(const DRIDriverContext *ctx,
-			     RADEONInfoPtr info)
-{
-   if (!info->irq) {
-      info->irq = drmGetInterruptFromBusID(ctx->drmFD,
-					   ctx->pciBus,
-					   ctx->pciDevice,
-					   ctx->pciFunc);
-
-      if ((drmCtlInstHandler(ctx->drmFD, info->irq)) != 0) {
-	 fprintf(stderr,
-		 "[drm] failure adding irq handler, "
-		 "there is a device already using that irq\n"
-		 "[drm] falling back to irq-free operation\n");
-	 info->irq = 0;
-      }
-   }
-
-   if (info->irq)
-      fprintf(stderr,
-	      "[drm] dma control initialized, using IRQ %d\n",
-	      info->irq);
-}
-
-static int RADEONCheckDRMVersion( const DRIDriverContext *ctx,
-				  RADEONInfoPtr info )
-{
-   drmVersionPtr  version;
-
-   version = drmGetVersion(ctx->drmFD);
-   if (version) {
-      int req_minor, req_patch;
-
-      /* Need 1.8.x for proper cleanup-on-client-exit behaviour.
-       */
-      req_minor = 8;
-      req_patch = 0;	
-
-      if (version->version_major != 1 ||
-	  version->version_minor < req_minor ||
-	  (version->version_minor == req_minor && 
-	   version->version_patchlevel < req_patch)) {
-	 /* Incompatible drm version */
-	 fprintf(stderr,
-		 "[dri] RADEONDRIScreenInit failed because of a version "
-		 "mismatch.\n"
-		 "[dri] radeon.o kernel module version is %d.%d.%d "
-		 "but version 1.%d.%d or newer is needed.\n"
-		 "[dri] Disabling DRI.\n",
-		 version->version_major,
-		 version->version_minor,
-		 version->version_patchlevel,
-		 req_minor,
-		 req_patch);
-	 drmFreeVersion(version);
-	 return 0;
-      }
-
-      info->drmMinor = version->version_minor;
-      drmFreeVersion(version);
-   }
-
-   return 1;
-}
-
-static int RADEONMemoryInit( const DRIDriverContext *ctx, RADEONInfoPtr info )
-{
-   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
-   int        cpp         = ctx->cpp;
-   int        bufferSize  = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes			     + RADEON_BUFFER_ALIGN) & ~RADEON_BUFFER_ALIGN);
-   int        depthSize   = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes
-			     + RADEON_BUFFER_ALIGN) & ~RADEON_BUFFER_ALIGN);
-   int        l;
-
-   info->frontOffset = 0;
-   info->frontPitch = ctx->shared.virtualWidth;
-
-   fprintf(stderr, 
-	   "Using %d MB AGP aperture\n", info->gartSize);
-   fprintf(stderr, 
-	   "Using %d MB for the ring buffer\n", info->ringSize);
-   fprintf(stderr, 
-	   "Using %d MB for vertex/indirect buffers\n", info->bufSize);
-   fprintf(stderr, 
-	   "Using %d MB for AGP textures\n", info->gartTexSize);
-
-   /* Front, back and depth buffers - everything else texture??
-    */
-   info->textureSize = ctx->shared.fbSize - 2 * bufferSize - depthSize;
-
-   if (ctx->colorTiling==1)
-   {
-	info->textureSize = ctx->shared.fbSize - ((ctx->shared.fbSize - info->textureSize + width_bytes * 16 - 1) / (width_bytes * 16)) * (width_bytes*16);
-   }
-
-   if (info->textureSize < 0) 
-      return 0;
-
-   l = RADEONMinBits((info->textureSize-1) / RADEON_NR_TEX_REGIONS);
-   if (l < RADEON_LOG_TEX_GRANULARITY) l = RADEON_LOG_TEX_GRANULARITY;
-
-   /* Round the texture size up to the nearest whole number of
-    * texture regions.  Again, be greedy about this, don't
-    * round down.
-    */
-   info->log2TexGran = l;
-   info->textureSize = (info->textureSize >> l) << l;
-
-   /* Set a minimum usable local texture heap size.  This will fit
-    * two 256x256x32bpp textures.
-    */
-   if (info->textureSize < 512 * 1024) {
-      info->textureOffset = 0;
-      info->textureSize = 0;
-   }
-
-   /* Reserve space for textures */
-   if (ctx->colorTiling==1)
-   {
-      info->textureOffset = ((ctx->shared.fbSize - info->textureSize) / 
-			(width_bytes * 16)) * (width_bytes*16);
-   }
-   else
-   {
-      info->textureOffset = ((ctx->shared.fbSize - info->textureSize +
-   	 		   RADEON_BUFFER_ALIGN) &
-			  ~RADEON_BUFFER_ALIGN);
-   }
-   /* Reserve space for the shared depth
-    * buffer.
-    */
-   info->depthOffset = ((info->textureOffset - depthSize +
-			 RADEON_BUFFER_ALIGN) &
-			~RADEON_BUFFER_ALIGN);
-   info->depthPitch = ctx->shared.virtualWidth;
-
-   info->backOffset = ((info->depthOffset - bufferSize +
-			RADEON_BUFFER_ALIGN) &
-		       ~RADEON_BUFFER_ALIGN);
-   info->backPitch = ctx->shared.virtualWidth;
-
-
-   fprintf(stderr, 
-	   "Will use back buffer at offset 0x%x\n",
-	   info->backOffset);
-   fprintf(stderr, 
-	   "Will use depth buffer at offset 0x%x\n",
-	   info->depthOffset);
-   fprintf(stderr, 
-	   "Will use %d kb for textures at offset 0x%x\n",
-	   info->textureSize/1024, info->textureOffset);
-
-   info->frontPitchOffset = (((info->frontPitch * cpp / 64) << 22) |
-			     (info->frontOffset >> 10));
-
-   info->backPitchOffset = (((info->backPitch * cpp / 64) << 22) |
-			    (info->backOffset >> 10));
-
-   info->depthPitchOffset = (((info->depthPitch * cpp / 64) << 22) |
-			     (info->depthOffset >> 10));
-
-   return 1;
-}
-
-static int RADEONColorTilingInit( const DRIDriverContext *ctx, RADEONInfoPtr info )
-{
-   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
-   int        bufferSize  = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes			     + RADEON_BUFFER_ALIGN)
-			     & ~RADEON_BUFFER_ALIGN);
-   /* Setup color tiling */
-   if (info->drmMinor<14)
-      info->colorTiling=0;
-
-   if (info->colorTiling)
-   {
-
-      int colorTilingFlag;
-      drm_radeon_surface_alloc_t front,back;
-
-      RadeonSetParam(ctx, RADEON_SETPARAM_SWITCH_TILING, info->colorTiling ? 1 : 0);
-      
-      /* Setup the surfaces */
-      if (info->ChipFamily < CHIP_FAMILY_R200)
-         colorTilingFlag=RADEON_SURF_TILE_COLOR_MACRO;
-      else
-         colorTilingFlag=R200_SURF_TILE_COLOR_MACRO;
-
-      front.address = info->frontOffset;
-      front.size = bufferSize;
-      front.flags = (width_bytes) | colorTilingFlag;
-      drmCommandWrite(ctx->drmFD, DRM_RADEON_SURF_ALLOC, &front,sizeof(front)); 
- 
-      back.address = info->backOffset;
-      back.size = bufferSize;
-      back.flags = (width_bytes) | colorTilingFlag;
-      drmCommandWrite(ctx->drmFD, DRM_RADEON_SURF_ALLOC, &back,sizeof(back)); 
-
-   }
-   return 1;
-} 
-
-
-
-/**
- * Called at the start of each server generation.
- *
- * \param ctx display handle.
- * \param info driver private data.
- *
- * \return non-zero on success, or zero on failure.
- *
- * Performs static frame buffer allocation. Opens the DRM device and add maps
- * to the SAREA, framebuffer and MMIO regions. Fills in \p info with more
- * information. Creates a \e server context to grab the lock for the
- * initialization ioctls and calls the other initilization functions in this
- * file. Starts the CP engine via the DRM_RADEON_CP_START command.
- *
- * Setups a RADEONDRIRec structure to be passed to radeon_dri.so for its
- * initialization.
- */
-static int RADEONScreenInit( DRIDriverContext *ctx, RADEONInfoPtr info )
-{
-   RADEONDRIPtr   pRADEONDRI;
-   int err;
-
-   usleep(100);
-   /*assert(!ctx->IsClient);*/
-
-   {
-      int  width_bytes = (ctx->shared.virtualWidth * ctx->cpp);
-      int  maxy        = ctx->shared.fbSize / width_bytes;
-
-
-      if (maxy <= ctx->shared.virtualHeight * 3) {
-	 fprintf(stderr, 
-		 "Static buffer allocation failed -- "
-		 "need at least %d kB video memory (have %d kB)\n",
-		 (ctx->shared.virtualWidth * ctx->shared.virtualHeight *
-		  ctx->cpp * 3 + 1023) / 1024,
-		 ctx->shared.fbSize / 1024);
-	 return 0;
-      } 
-   }
-
-
-   if (info->ChipFamily >= CHIP_FAMILY_R300) {
-      fprintf(stderr, 
-	      "Direct rendering not yet supported on "
-	      "Radeon 9700 and newer cards\n");
-      return 0;
-   }
-   
-   radeon_drm_page_size = getpagesize();   
-
-   info->registerSize = ctx->MMIOSize;
-   ctx->shared.SAREASize = SAREA_MAX;
-
-   /* Note that drmOpen will try to load the kernel module, if needed. */
-   ctx->drmFD = drmOpen("radeon", NULL );
-   if (ctx->drmFD < 0) {
-      fprintf(stderr, "[drm] drmOpen failed\n");
-      return 0;
-   }
-
-   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-	      ctx->drmFD, ctx->pciBusID, strerror(-err));
-      return 0;
-   }
-
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   }
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-   
-   /* Need to AddMap the framebuffer and mmio regions here:
-    */
-   if (drmAddMap( ctx->drmFD,
-		  (drm_handle_t)ctx->FBStart,
-		  ctx->FBSize,
-		  DRM_FRAME_BUFFER,
-#ifndef _EMBEDDED
-		  0,
-#else
-		  DRM_READ_ONLY,
-#endif
-		  &ctx->shared.hFrameBuffer) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-      return 0;
-   }
-
-   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-	   ctx->shared.hFrameBuffer);
-
-
-
-   if (drmAddMap(ctx->drmFD, 
-		 ctx->MMIOStart,
-		 ctx->MMIOSize,
-		 DRM_REGISTERS, 
-		 DRM_READ_ONLY, 
-		 &info->registerHandle) < 0) {
-      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] register handle = 0x%08lx\n", info->registerHandle);
-
-   /* Check the radeon DRM version */
-   if (!RADEONCheckDRMVersion(ctx, info)) {
-      return 0;
-   }
-
-   if (ctx->isPCI) {
-      /* Initialize PCI */
-      if (!RADEONDRIPciInit(ctx, info))
-         return 0;
-   }
-   else {
-      /* Initialize AGP */
-      if (!RADEONDRIAgpInit(ctx, info))
-         return 0;
-   }
-
-   /* Memory manager setup */
-   if (!RADEONMemoryInit(ctx, info)) {
-      return 0;
-   }
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-   /* Initialize the kernel data structures */
-   if (!RADEONDRIKernelInit(ctx, info)) {
-      fprintf(stderr, "RADEONDRIKernelInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   /* Initialize the vertex buffers list */
-   if (!RADEONDRIBufInit(ctx, info)) {
-      fprintf(stderr, "RADEONDRIBufInit failed\n");
-      DRM_UNLOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext);
-      return 0;
-   }
-
-   RADEONColorTilingInit(ctx, info);
-
-   /* Initialize IRQ */
-   RADEONDRIIrqInit(ctx, info);
-
-   /* Initialize kernel gart memory manager */
-   RADEONDRIAgpHeapInit(ctx, info);
-
-   fprintf(stderr,"color tiling %sabled\n", info->colorTiling?"en":"dis");
-   fprintf(stderr,"page flipping %sabled\n", info->page_flip_enable?"en":"dis");
-   /* Initialize the SAREA private data structure */
-   {
-      drm_radeon_sarea_t *pSAREAPriv;
-      pSAREAPriv = (drm_radeon_sarea_t *)(((char*)ctx->pSAREA) + 
-					sizeof(drm_sarea_t));
-      memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
-      pSAREAPriv->pfState = info->page_flip_enable;
-   }
-
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-   drimemsetio((char *)ctx->FBAddress + info->frontOffset,
-	  0,
-	  info->frontPitch * ctx->cpp * ctx->shared.virtualHeight );
-
-   drimemsetio((char *)ctx->FBAddress + info->backOffset,
-	  0,
-	  info->backPitch * ctx->cpp * ctx->shared.virtualHeight );
-
-   /* This is the struct passed to radeon_dri.so for its initialization */
-   ctx->driverClientMsg = malloc(sizeof(RADEONDRIRec));
-   ctx->driverClientMsgSize = sizeof(RADEONDRIRec);
-   pRADEONDRI                    = (RADEONDRIPtr)ctx->driverClientMsg;
-   pRADEONDRI->deviceID          = info->Chipset;
-   pRADEONDRI->width             = ctx->shared.virtualWidth;
-   pRADEONDRI->height            = ctx->shared.virtualHeight;
-   pRADEONDRI->depth             = ctx->bpp; /* XXX: depth */
-   pRADEONDRI->bpp               = ctx->bpp;
-   pRADEONDRI->IsPCI             = ctx->isPCI;
-   pRADEONDRI->AGPMode           = ctx->agpmode;
-   pRADEONDRI->frontOffset       = info->frontOffset;
-   pRADEONDRI->frontPitch        = info->frontPitch;
-   pRADEONDRI->backOffset        = info->backOffset;
-   pRADEONDRI->backPitch         = info->backPitch;
-   pRADEONDRI->depthOffset       = info->depthOffset;
-   pRADEONDRI->depthPitch        = info->depthPitch;
-   pRADEONDRI->textureOffset     = info->textureOffset;
-   pRADEONDRI->textureSize       = info->textureSize;
-   pRADEONDRI->log2TexGran       = info->log2TexGran;
-   pRADEONDRI->registerHandle    = info->registerHandle;
-   pRADEONDRI->registerSize      = info->registerSize; 
-   pRADEONDRI->statusHandle      = info->ringReadPtrHandle;
-   pRADEONDRI->statusSize        = info->ringReadMapSize;
-   pRADEONDRI->gartTexHandle      = info->gartTexHandle;
-   pRADEONDRI->gartTexMapSize     = info->gartTexMapSize;
-   pRADEONDRI->log2GARTTexGran    = info->log2GARTTexGran;
-   pRADEONDRI->gartTexOffset      = info->gartTexStart;
-   pRADEONDRI->sarea_priv_offset = sizeof(drm_sarea_t);
-
-   /* Don't release the lock now - let the VT switch handler do it. */
-
-   return 1;
-}
-
-
-/**
- * \brief Get Radeon chip family from chipset number.
- * 
- * \param info driver private data.
- *
- * \return non-zero on success, or zero on failure.
- *
- * Called by radeonInitFBDev() to set RADEONInfoRec::ChipFamily
- * according to the value of RADEONInfoRec::Chipset.  Fails if the
- * chipset is unrecognized or not appropriate for this driver (i.e., not
- * an r100 style radeon)
- */
-static int get_chipfamily_from_chipset( RADEONInfoPtr info )
-{
-    switch (info->Chipset) {
-    case PCI_CHIP_RADEON_LY:
-    case PCI_CHIP_RADEON_LZ:
-	info->ChipFamily = CHIP_FAMILY_M6;
-	break;
-
-    case PCI_CHIP_RADEON_QY:
-    case PCI_CHIP_RADEON_QZ:
-	info->ChipFamily = CHIP_FAMILY_VE;
-	break;
-
-    case PCI_CHIP_R200_QL:
-    case PCI_CHIP_R200_QN:
-    case PCI_CHIP_R200_QO:
-    case PCI_CHIP_R200_Ql:
-    case PCI_CHIP_R200_BB:
-	info->ChipFamily = CHIP_FAMILY_R200;
-	break;
-
-    case PCI_CHIP_RV200_QW: /* RV200 desktop */
-    case PCI_CHIP_RV200_QX:
-	info->ChipFamily = CHIP_FAMILY_RV200;
-	break;
-
-    case PCI_CHIP_RADEON_LW:
-    case PCI_CHIP_RADEON_LX:
-	info->ChipFamily = CHIP_FAMILY_M7;
-	break;
-
-    case PCI_CHIP_RV250_Id:
-    case PCI_CHIP_RV250_Ie:
-    case PCI_CHIP_RV250_If:
-    case PCI_CHIP_RV250_Ig:
-	info->ChipFamily = CHIP_FAMILY_RV250;
-	break;
-
-    case PCI_CHIP_RV250_Ld:
-    case PCI_CHIP_RV250_Le:
-    case PCI_CHIP_RV250_Lf:
-    case PCI_CHIP_RV250_Lg:
-	info->ChipFamily = CHIP_FAMILY_M9;
-	break;
-
-    case PCI_CHIP_RV280_Y_:
-    case PCI_CHIP_RV280_Ya:
-    case PCI_CHIP_RV280_Yb:
-    case PCI_CHIP_RV280_Yc:
-	info->ChipFamily = CHIP_FAMILY_RV280;
-        break;
-
-    case PCI_CHIP_R300_ND:
-    case PCI_CHIP_R300_NE:
-    case PCI_CHIP_R300_NF:
-    case PCI_CHIP_R300_NG:
-	info->ChipFamily = CHIP_FAMILY_R300;
-        break;
-
-    default:
-	/* Original Radeon/7200 */
-	info->ChipFamily = CHIP_FAMILY_RADEON;
-    }
-
-    return 1;
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa radeonValidateMode().
- */
-static int radeonValidateMode( const DRIDriverContext *ctx )
-{
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   RADEONInfoPtr info = ctx->driverPrivate;
-
-   info->gen_int_cntl = INREG(RADEON_GEN_INT_CNTL);
-   info->crtc_offset_cntl = INREG(RADEON_CRTC_OFFSET_CNTL);
-
-   if (info->colorTiling)
-	   info->crtc_offset_cntl |= RADEON_CRTC_TILE_EN;
-   return 1;
-}
-
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa radeonValidateMode().
- */
-static int radeonPostValidateMode( const DRIDriverContext *ctx )
-{
-   unsigned char *RADEONMMIO = ctx->MMIOAddress;
-   RADEONInfoPtr info = ctx->driverPrivate;
-
-   RADEONColorTilingInit( ctx, info);
-   OUTREG(RADEON_GEN_INT_CNTL, info->gen_int_cntl);
-   if (info->colorTiling)
-	   info->crtc_offset_cntl |= RADEON_CRTC_TILE_EN;
-   OUTREG(RADEON_CRTC_OFFSET_CNTL, info->crtc_offset_cntl);
-   
-   return 1;
-}
-
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Fills in \p info with some default values and some information from \p ctx
- * and then calls RADEONScreenInit() for the screen initialization.
- * 
- * Before exiting clears the framebuffer memory accessing it directly.
- */
-static int radeonInitFBDev( DRIDriverContext *ctx )
-{
-   RADEONInfoPtr info = calloc(1, sizeof(*info));
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      if (ctx->colorTiling==1)
-      {
-         switch (ctx->bpp / 8) {
-         case 1: dummy = (ctx->shared.virtualWidth + 255) & ~255; break;
-         case 2: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-         case 3:
-         case 4: dummy = (ctx->shared.virtualWidth +  63) &  ~63; break;
-         }
-      } else {
-	 switch (ctx->bpp / 8) {
-         case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-         case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-         case 3:
-         case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-         }
-      }
-
-      ctx->shared.virtualWidth = dummy;
-      ctx->shared.Width = dummy;
-   }
-
-   fprintf(stderr,"shared virtual width is %d\n", ctx->shared.virtualWidth);
-   ctx->driverPrivate = (void *)info;
-   
-   info->gartFastWrite  = RADEON_DEFAULT_AGP_FAST_WRITE;
-   info->gartSize       = RADEON_DEFAULT_AGP_SIZE;
-   info->gartTexSize    = RADEON_DEFAULT_AGP_TEX_SIZE;
-   info->bufSize       = RADEON_DEFAULT_BUFFER_SIZE;
-   info->ringSize      = RADEON_DEFAULT_RING_SIZE;
-   info->page_flip_enable = RADEON_DEFAULT_PAGE_FLIP;
-   info->colorTiling = ctx->colorTiling;
-  
-   info->Chipset = ctx->chipset;
-
-   if (!get_chipfamily_from_chipset( info )) {
-      fprintf(stderr, "Unknown or non-radeon chipset -- cannot continue\n");
-      fprintf(stderr, "==> Verify PCI BusID is correct in miniglx.conf\n");
-      return 0;
-   }
-
-   info->frontPitch = ctx->shared.virtualWidth;
-   info->LinearAddr = ctx->FBStart & 0xfc000000;
-    
-
-   if (!RADEONScreenInit( ctx, info ))
-      return 0;
-
-
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void radeonHaltFBDev( DRIDriverContext *ctx )
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-       free(ctx->driverPrivate);
-       ctx->driverPrivate = 0;
-    }
-}
-
-
-extern void radeonNotifyFocus( int );
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-const struct DRIDriverRec __driDriver = {
-   radeonValidateMode,
-   radeonPostValidateMode,
-   radeonInitFBDev,
-   radeonHaltFBDev,
-   RADEONEngineShutdown,
-   RADEONEngineRestore,  
-#ifndef _EMBEDDED
-   0,
-#else
-   radeonNotifyFocus, 
-#endif
-};
diff --git a/src/mesa/drivers/dri/savage/Makefile b/src/mesa/drivers/dri/savage/Makefile
index 2e5c40802c1..53511552c6d 100644
--- a/src/mesa/drivers/dri/savage/Makefile
+++ b/src/mesa/drivers/dri/savage/Makefile
@@ -5,9 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = savage_dri.so
 
-# Doesn't exist yet.
-#MINIGLX_SOURCES = server/savage_dri.c 
-
 DRIVER_SOURCES = \
 	savage_xmesa.c \
 	savagedd.c \
diff --git a/src/mesa/drivers/dri/sis/Makefile b/src/mesa/drivers/dri/sis/Makefile
index ad009fc2398..6b4f938bab3 100644
--- a/src/mesa/drivers/dri/sis/Makefile
+++ b/src/mesa/drivers/dri/sis/Makefile
@@ -5,10 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = sis_dri.so
 
-
-# Not yet
-# MINIGLX_SOURCES = server/sis_dri.c 
-
 DRIVER_SOURCES = \
 	sis6326_state.c \
 	sis6326_clear.c \
diff --git a/src/mesa/drivers/dri/tdfx/Makefile b/src/mesa/drivers/dri/tdfx/Makefile
index b9f25db4fe8..96bd8f8202f 100644
--- a/src/mesa/drivers/dri/tdfx/Makefile
+++ b/src/mesa/drivers/dri/tdfx/Makefile
@@ -5,9 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = tdfx_dri.so
 
-# not yet
-# MINIGLX_SOURCES = server/tdfx_dri.c 
-
 DRIVER_SOURCES = \
 	tdfx_context.c \
 	tdfx_dd.c \
diff --git a/src/mesa/drivers/dri/tdfx/server/tdfx_dri.c b/src/mesa/drivers/dri/tdfx/server/tdfx_dri.c
deleted file mode 100644
index 63fe875f59e..00000000000
--- a/src/mesa/drivers/dri/tdfx/server/tdfx_dri.c
+++ /dev/null
@@ -1,471 +0,0 @@
-/*
- * Mesa 3-D graphics library
- * Version:  5.1
- *
- * Copyright (C) 1999-2003  Brian Paul   All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
- * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- */
-
-/* Authors:
- *    Keith Whitwell
- *    Daniel Borca
- */
-
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
- 
-#include "driver.h"
-#include "drm.h"
-#include "imports.h"
-
-#include "dri_util.h"
-
-#include "tdfx_context.h"
-#include "tdfx_dri.h"
-#include "xf86drm.h"
-
-
-#define TILE_WIDTH 128
-#define TILE_HEIGHT 32
-
-#define CMDFIFO_PAGES 64
-
-
-static int
-calcBufferStride (int xres, int tiled, int cpp)
-{
-  int strideInTiles;
-
-  if (tiled) {
-    /* Calculate tile width stuff */
-    strideInTiles = (xres+TILE_WIDTH-1)/TILE_WIDTH;
-
-    return strideInTiles*cpp*TILE_WIDTH;
-  } else {
-    return xres*cpp;
-  }
-} /* calcBufferStride */
-
-
-static int
-calcBufferHeightInTiles (int yres)
-{
-  int heightInTiles;            /* Height of buffer in tiles */
-
-  /* Calculate tile height stuff */
-  heightInTiles = yres >> 5;
-
-  if (yres & (TILE_HEIGHT - 1))
-    heightInTiles++;
-
-  return heightInTiles;
-
-} /* calcBufferHeightInTiles */
-
-
-static int
-calcBufferSize (int xres, int yres, int tiled, int cpp)
-{
-  int stride, height, bufSize;
-
-  if (tiled) {
-    stride = calcBufferStride(xres, tiled, cpp);
-    height = TILE_HEIGHT * calcBufferHeightInTiles(yres);
-  } else {
-    stride = xres*cpp;
-    height = yres;
-  }
-
-  bufSize = stride * height;
-
-  return bufSize;
-} /* calcBufferSize */
-
-
-static void allocateMemory (const DRIDriverContext *ctx, TDFXDRIPtr pTDFX)
-{
-  int memRemaining, fifoSize, screenSizeInTiles;
-  int fbSize;
-  char *str;
-  int pixmapCacheLinesMin;
-  int cursorOffset, cursorSize;
-
-  pTDFX->stride = calcBufferStride(pTDFX->width, !0, pTDFX->cpp);
-
-  /* enough to do DVD */
-  pixmapCacheLinesMin = ((720*480*pTDFX->cpp) + 
-					pTDFX->stride - 1)/pTDFX->stride;
-
-  if (pTDFX->deviceID > PCI_CHIP_VOODOO3) {
-  	if ((pixmapCacheLinesMin + pTDFX->height) > 4095)
-		pixmapCacheLinesMin = 4095 - pTDFX->height;
-  } else {
-  	if ((pixmapCacheLinesMin + pTDFX->height) > 2047)
-		pixmapCacheLinesMin = 2047 - pTDFX->height;
-  }
-
-  if (pTDFX->cpp!=3) {
-    screenSizeInTiles=calcBufferSize(pTDFX->width, pTDFX->height,
-				     !0, pTDFX->cpp);
-  }
-  else {
-    /* cpp==3 needs to bump up to 4 */
-    screenSizeInTiles=calcBufferSize(pTDFX->width, pTDFX->height,
-				     !0, 4);
-  }
-
-  /*
-   * Layout is:
-   *    cursor, fifo, fb, tex, bb, db
-   */
-
-  fbSize = (pTDFX->height + pixmapCacheLinesMin) * pTDFX->stride;
-
-  memRemaining=(pTDFX->mem - 1) &~ 0xFFF;
-  /* Note that a page is 4096 bytes, and a  */
-  /* tile is 32 x 128 = 4096 bytes.  So,    */
-  /* page and tile boundaries are the same  */
-  /* Place the depth offset first, forcing  */
-  /* it to be on an *odd* page boundary.    */
-  pTDFX->depthOffset = (memRemaining - screenSizeInTiles) &~ 0xFFF;
-  if ((pTDFX->depthOffset & (0x1 << 12)) == 0) {
-      pTDFX->depthOffset -= (0x1 << 12);
-  }
-  /* Now, place the back buffer, forcing it */
-  /* to be on an *even* page boundary.      */
-  pTDFX->backOffset = (pTDFX->depthOffset - screenSizeInTiles) &~ 0xFFF;
-  if (pTDFX->backOffset & (0x1 << 12)) {
-      pTDFX->backOffset -= (0x1 << 12);
-  }
-  /* Give the cmd fifo at least             */
-  /* CMDFIFO_PAGES pages, but no more than  */
-  /* 64. NOTE: Don't go higher than 64, as  */
-  /* there is suspect code in Glide3 !      */
-  fifoSize = ((64 <= CMDFIFO_PAGES) ? 64 : CMDFIFO_PAGES) << 12;
-
-  /* We give 4096 bytes to the cursor  */
-  cursorSize = 0/*4096*/;
-  cursorOffset = 0;
-
-  pTDFX->fifoOffset = cursorOffset + cursorSize;
-  pTDFX->fifoSize = fifoSize;
-  /* Now, place the front buffer, forcing   */
-  /* it to be on a page boundary too, just  */
-  /* for giggles.                           */
-  pTDFX->fbOffset = pTDFX->fifoOffset + pTDFX->fifoSize;
-  pTDFX->textureOffset = pTDFX->fbOffset + fbSize;
-  if (pTDFX->depthOffset <= pTDFX->textureOffset ||
-	pTDFX->backOffset <= pTDFX->textureOffset) {
-    /*
-     * pTDFX->textureSize < 0 means that the DRI is disabled.  pTDFX->backOffset
-     * is used to calculate the maximum amount of memory available for
-     * 2D offscreen use.  With DRI disabled, set this to the top of memory.
-     */
-
-    pTDFX->textureSize = -1;
-    pTDFX->backOffset = pTDFX->mem;
-    pTDFX->depthOffset = -1;
-    fprintf(stderr, 
-        "Not enough video memory available for textures and depth buffer\n"
-	"\tand/or back buffer.  Disabling DRI.  To use DRI try lower\n"
-	"\tresolution modes and/or a smaller virtual screen size\n");
-  } else {
-    pTDFX->textureSize = pTDFX->backOffset - pTDFX->textureOffset;
-  }
-}
-
-
-static int createScreen (DRIDriverContext *ctx, TDFXDRIPtr pTDFX)
-{
-   int err;
-   
-   {
-      int  width_bytes = (ctx->shared.virtualWidth * ctx->cpp);
-      int  maxy        = ctx->shared.fbSize / width_bytes;
-
-
-      if (maxy <= ctx->shared.virtualHeight * 3) {
-	 fprintf(stderr, 
-		 "Static buffer allocation failed -- "
-		 "need at least %d kB video memory (have %d kB)\n",
-		 (ctx->shared.virtualWidth * ctx->shared.virtualHeight *
-		  ctx->cpp * 3 + 1023) / 1024,
-		 ctx->shared.fbSize / 1024);
-	 return 0;
-      } 
-   }
-
-   ctx->shared.SAREASize = SAREA_MAX;
-   pTDFX->regsSize = ctx->MMIOSize;
-
-   /* Note that drmOpen will try to load the kernel module, if needed. */
-   ctx->drmFD = drmOpen("tdfx", NULL );
-   if (ctx->drmFD < 0) {
-      fprintf(stderr, "[drm] drmOpen failed\n");
-      return 0;
-   }
-
-   if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
-      fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-	      ctx->drmFD, ctx->pciBusID, strerror(-err));
-      return 0;
-   }
-
-   if (drmAddMap( ctx->drmFD,
-		  0,
-		  ctx->shared.SAREASize,
-		  DRM_SHM,
-		  DRM_CONTAINS_LOCK,
-		  &ctx->shared.hSAREA) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap failed\n");
-      return 0;
-   }
-   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-	   ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-   if (drmMap( ctx->drmFD,
-	       ctx->shared.hSAREA,
-	       ctx->shared.SAREASize,
-	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
-   {
-      fprintf(stderr, "[drm] drmMap failed\n");
-      return 0;
-   }
-   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-   
-   /* Need to AddMap the framebuffer and mmio regions here:
-    */
-   if (drmAddMap( ctx->drmFD,
-		  (drm_handle_t)ctx->FBStart,
-		  ctx->FBSize,
-		  DRM_FRAME_BUFFER,
-#ifndef _EMBEDDED
-		  0,
-#else
-		  DRM_READ_ONLY,
-#endif
-		  &ctx->shared.hFrameBuffer) < 0)
-   {
-      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-      return 0;
-   }
-
-   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-	   ctx->shared.hFrameBuffer);
-
-
-   if (drmAddMap(ctx->drmFD, 
-		 ctx->MMIOStart,
-		 ctx->MMIOSize,
-		 DRM_REGISTERS, 
-		 DRM_READ_ONLY, 
-		 &pTDFX->regs) < 0) {
-      fprintf(stderr, "[drm] drmAddMap mmio failed\n");	
-      return 0;
-   }
-   fprintf(stderr,
-	   "[drm] register handle = 0x%08lx\n", pTDFX->regs);
-
-
-   /* Create a 'server' context so we can grab the lock for
-    * initialization ioctls.
-    */
-   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
-      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-      return 0;
-   }
-
-   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
-
-   /* Initialize the kernel data structures */
-
-   /* Initialize kernel gart memory manager */
-   allocateMemory(ctx, pTDFX);
-
-   /* Initialize the SAREA private data structure */
-
-
-   /* Quick hack to clear the front & back buffers.  Could also use
-    * the clear ioctl to do this, but would need to setup hw state
-    * first.
-    */
-
-
-   /* This is the struct passed to tdfx_dri.so for its initialization */
-   ctx->driverClientMsg = malloc(sizeof(TDFXDRIRec));
-   ctx->driverClientMsgSize = sizeof(TDFXDRIRec);
-   memcpy(ctx->driverClientMsg, pTDFX, ctx->driverClientMsgSize);
-   pTDFX = (TDFXDRIPtr)ctx->driverClientMsg;
-
-   /* Don't release the lock now - let the VT switch handler do it. */
-
-   return 1;
-}
-
-
-/**
- * \brief Validate the fbdev mode.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Saves some registers and returns 1.
- *
- * \sa tdfxValidateMode().
- */
-static int tdfxValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Examine mode returned by fbdev.
- * 
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Restores registers that fbdev has clobbered and returns 1.
- *
- * \sa tdfxValidateMode().
- */
-static int tdfxPostValidateMode( const DRIDriverContext *ctx )
-{
-   return 1;
-}
-
-
-/**
- * \brief Initialize the framebuffer device mode
- *
- * \param ctx display handle.
- *
- * \return one on success, or zero on failure.
- *
- * Before exiting clears the framebuffer memory accessing it directly.
- */
-static int tdfxInitFBDev( DRIDriverContext *ctx )
-{
-   TDFXDRIPtr pTDFX = calloc(1, sizeof(TDFXDRIRec));
-
-   {
-      int  dummy = ctx->shared.virtualWidth;
-
-      switch (ctx->bpp / 8) {
-      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
-      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
-      case 3:
-      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
-      }
-
-      ctx->shared.virtualWidth = dummy;
-   }
-
-   ctx->driverPrivate = (void *)pTDFX;
-
-   pTDFX->deviceID = ctx->chipset;
-   pTDFX->width    = ctx->shared.virtualWidth;
-   pTDFX->height   = ctx->shared.virtualHeight;
-   pTDFX->cpp      = ctx->cpp;
-   pTDFX->mem      = ctx->FBSize; /* ->shared.fbSize? mem probe? */
-   pTDFX->sarea_priv_offset = sizeof(drm_sarea_t);
-
-   if (!createScreen(ctx, pTDFX))
-      return 0;
-
-   return 1;
-}
-
-
-/**
- * \brief The screen is being closed, so clean up any state and free any
- * resources used by the DRI.
- *
- * \param ctx display handle.
- *
- * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
- * private data.
- */
-static void tdfxHaltFBDev( DRIDriverContext *ctx )
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-       free(ctx->driverPrivate);
-       ctx->driverPrivate = 0;
-    }
-}
-
-
-/**
- * \brief Shutdown the drawing engine.
- *
- * \param ctx display handle
- *
- * Turns off the 3D engine & restores the graphics card
- * to a state that fbdev understands.
- */
-static int tdfxEngineShutdown( const DRIDriverContext *ctx )
-{
-   fprintf(stderr, "%s: not implemented\n", __FUNCTION__);
-   return 1;
-}
-
-
-/**
- * \brief Restore the drawing engine.
- *
- * \param ctx display handle
- *
- * Resets the graphics card and sets initial values for several registers of
- * the card's drawing engine.
- *
- * Turns on 3dfx
- */
-static int tdfxEngineRestore( const DRIDriverContext *ctx )
-{
-   fprintf(stderr, "%s: not implemented\n", __FUNCTION__);
-   return 1;
-}
-
-
-/**
- * \brief Exported driver interface for Mini GLX.
- *
- * \sa DRIDriverRec.
- */
-struct DRIDriverRec __driDriver = {
-   tdfxValidateMode,
-   tdfxPostValidateMode,
-   tdfxInitFBDev,
-   tdfxHaltFBDev,
-   tdfxEngineShutdown,
-   tdfxEngineRestore,
-   0
-};
diff --git a/src/mesa/drivers/dri/unichrome/Makefile b/src/mesa/drivers/dri/unichrome/Makefile
index 344d34fce36..14cf9f30386 100644
--- a/src/mesa/drivers/dri/unichrome/Makefile
+++ b/src/mesa/drivers/dri/unichrome/Makefile
@@ -5,8 +5,6 @@ include $(TOP)/configs/current
 
 LIBNAME = unichrome_dri.so
 
-MINIGLX_SOURCES = server/via_dri.c 
-
 DRIVER_SOURCES = \
 	via_context.c \
 	via_fb.c \
diff --git a/src/mesa/drivers/dri/unichrome/server/via_dri.c b/src/mesa/drivers/dri/unichrome/server/via_dri.c
deleted file mode 100644
index 74034485e28..00000000000
--- a/src/mesa/drivers/dri/unichrome/server/via_dri.c
+++ /dev/null
@@ -1,1251 +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 <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <errno.h>
-#include <unistd.h>
- 
-#include "driver.h"
-#include "drm.h"
-#include "imports.h"
-
-#include "dri_util.h"
-
-#include "via_context.h"
-#include "via_dri.h"
-#include "via_driver.h"
-#include "xf86drm.h"
-
-static void VIAEnableMMIO(DRIDriverContext * ctx);
-static void VIADisableMMIO(DRIDriverContext * ctx);
-static void VIADisableExtendedFIFO(DRIDriverContext *ctx);
-static void VIAEnableExtendedFIFO(DRIDriverContext *ctx);
-static void VIAInitialize2DEngine(DRIDriverContext *ctx);
-static void VIAInitialize3DEngine(DRIDriverContext *ctx);
-
-static int VIADRIScreenInit(DRIDriverContext * ctx);
-static void VIADRICloseScreen(DRIDriverContext * ctx);
-static int VIADRIFinishScreenInit(DRIDriverContext * ctx);
-
-/* _SOLO : missing macros normally defined by X code */
-#define xf86DrvMsg(a, b, ...) fprintf(stderr, __VA_ARGS__)
-#define MMIO_IN8(base, addr) ((*(((volatile uint8_t*)base)+(addr)))+0)
-#define MMIO_OUT8(base, addr, val) ((*(((volatile uint8_t*)base)+(addr)))=((uint8_t)val))
-#define MMIO_OUT16(base, addr, val) ((*(volatile uint16_t*)(((uint8_t*)base)+(addr)))=((uint16_t)val))
-
-#define VIDEO	0 
-#define AGP		1
-#define AGP_PAGE_SIZE 4096
-#define AGP_PAGES 8192
-#define AGP_SIZE (AGP_PAGE_SIZE * AGP_PAGES)
-#define AGP_CMDBUF_PAGES 512
-#define AGP_CMDBUF_SIZE (AGP_PAGE_SIZE * AGP_CMDBUF_PAGES)
-
-static char VIAKernelDriverName[] = "via";
-static char VIAClientDriverName[] = "unichrome";
-
-static int VIADRIAgpInit(const DRIDriverContext *ctx, VIAPtr pVia);
-static int VIADRIPciInit(DRIDriverContext * ctx, VIAPtr pVia);
-static int VIADRIFBInit(DRIDriverContext * ctx, VIAPtr pVia);
-static int VIADRIKernelInit(DRIDriverContext * ctx, VIAPtr pVia);
-static int VIADRIMapInit(DRIDriverContext * ctx, VIAPtr pVia);
-
-static void VIADRIIrqInit( DRIDriverContext *ctx )
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI = pVia->devPrivate;
-
-    pVIADRI->irqEnabled = drmGetInterruptFromBusID(pVia->drmFD,
-					   ctx->pciBus,
-					   ctx->pciDevice,
-					   ctx->pciFunc);
-
-    if ((drmCtlInstHandler(pVia->drmFD, pVIADRI->irqEnabled))) {
-	xf86DrvMsg(pScreen->myNum, X_WARNING,
-		   "[drm] Failure adding irq handler. "
-		   "Falling back to irq-free operation.\n");
-	pVIADRI->irqEnabled = 0;
-    }
-
-    if (pVIADRI->irqEnabled)
-	xf86DrvMsg(pScreen->myNum, X_INFO,
-		   "[drm] Irq handler installed, using IRQ %d.\n",
-		   pVIADRI->irqEnabled);
-}
-
-static void VIADRIIrqExit( DRIDriverContext *ctx ) {
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI = pVia->devPrivate;
-
-    if (pVIADRI->irqEnabled) {
-	if (drmCtlUninstHandler(pVia->drmFD)) {
-	    xf86DrvMsg(pScreen-myNum, X_INFO,"[drm] Irq handler uninstalled.\n");
-	} else {
-	    xf86DrvMsg(pScreen->myNum, X_ERROR,
-		       "[drm] Could not uninstall irq handler.\n");
-	}
-    }
-}
-	    
-static void VIADRIRingBufferCleanup(DRIDriverContext *ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI = pVia->devPrivate;
-    drm_via_dma_init_t ringBufInit;
-
-    if (pVIADRI->ringBufActive) {
-	xf86DrvMsg(pScreen->myNum, X_INFO, 
-		   "[drm] Cleaning up DMA ring-buffer.\n");
-	ringBufInit.func = VIA_CLEANUP_DMA;
-	if (drmCommandWrite(pVia->drmFD, DRM_VIA_DMA_INIT, &ringBufInit,
-			    sizeof(ringBufInit))) {
-	    xf86DrvMsg(pScreen->myNum, X_WARNING, 
-		       "[drm] Failed to clean up DMA ring-buffer: %d\n", errno);
-	}
-	pVIADRI->ringBufActive = 0;
-    }
-}
-
-static int VIADRIRingBufferInit(DRIDriverContext *ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI = pVia->devPrivate;
-    drm_via_dma_init_t ringBufInit;
-    drmVersionPtr drmVer;
-
-    pVIADRI->ringBufActive = 0;
-
-    if (NULL == (drmVer = drmGetVersion(pVia->drmFD))) {
-	return GL_FALSE;
-    }
-
-    if (((drmVer->version_major <= 1) && (drmVer->version_minor <= 3))) {
-	return GL_FALSE;
-    } 
-
-    /*
-     * Info frome code-snippet on DRI-DEVEL list; Erdi Chen.
-     */
-
-    switch (pVia->ChipId) {
-    case PCI_CHIP_VT3259:
-    	ringBufInit.reg_pause_addr = 0x40c;
-	break;
-    default:
-    	ringBufInit.reg_pause_addr = 0x418;
-	break;
-    }
-   
-    ringBufInit.offset = pVia->agpSize;
-    ringBufInit.size = AGP_CMDBUF_SIZE;
-    ringBufInit.func = VIA_INIT_DMA;
-    if (drmCommandWrite(pVia->drmFD, DRM_VIA_DMA_INIT, &ringBufInit,
-			sizeof(ringBufInit))) {
-	xf86DrvMsg(pScreen->myNum, X_ERROR, 
-		   "[drm] Failed to initialize DMA ring-buffer: %d\n", errno);
-	return GL_FALSE;
-    }
-    xf86DrvMsg(pScreen->myNum, X_INFO, 
-	       "[drm] Initialized AGP ring-buffer, size 0x%lx at AGP offset 0x%lx.\n",
-	       ringBufInit.size, ringBufInit.offset);
-   
-    pVIADRI->ringBufActive = 1;
-    return GL_TRUE;
-}	    
-
-static int VIADRIAgpInit(const DRIDriverContext *ctx, VIAPtr pVia)
-{
-    unsigned long  agp_phys;
-    drmAddress agpaddr;
-    VIADRIPtr pVIADRI;
-    pVIADRI = pVia->devPrivate;
-    pVia->agpSize = 0;
-
-    if (drmAgpAcquire(pVia->drmFD) < 0) {
-        xf86DrvMsg(pScreen->myNum, X_ERROR, "[drm] drmAgpAcquire failed %d\n", errno);
-        return GL_FALSE;
-    }
-
-    if (drmAgpEnable(pVia->drmFD, drmAgpGetMode(pVia->drmFD)&~0x0) < 0) {
-         xf86DrvMsg(pScreen->myNum, X_ERROR, "[drm] drmAgpEnable failed\n");
-        return GL_FALSE;
-    }
-    
-    xf86DrvMsg(pScreen->myNum, X_INFO, "[drm] drmAgpEnabled succeeded\n");
-
-    if (drmAgpAlloc(pVia->drmFD, AGP_SIZE, 0, &agp_phys, &pVia->agpHandle) < 0) {
-        xf86DrvMsg(pScreen->myNum, X_ERROR,
-                 "[drm] drmAgpAlloc failed\n");
-        drmAgpRelease(pVia->drmFD);
-        return GL_FALSE;
-    }
-   
-    if (drmAgpBind(pVia->drmFD, pVia->agpHandle, 0) < 0) {
-        xf86DrvMsg(pScreen->myNum, X_ERROR,
-                 "[drm] drmAgpBind failed\n");
-        drmAgpFree(pVia->drmFD, pVia->agpHandle);
-        drmAgpRelease(pVia->drmFD);
-
-        return GL_FALSE;
-    }
-
-    /*
-     * Place the ring-buffer last in the AGP region, and restrict the
-     * public map not to include the buffer for security reasons.
-     */
-
-    pVia->agpSize = AGP_SIZE - AGP_CMDBUF_SIZE;
-    pVia->agpAddr = drmAgpBase(pVia->drmFD);
-    xf86DrvMsg(pScreen->myNum, X_INFO,
-                 "[drm] agpAddr = 0x%08lx\n",pVia->agpAddr);
-		 
-    pVIADRI->agp.size = pVia->agpSize;
-    if (drmAddMap(pVia->drmFD, (drm_handle_t)0,
-                 pVIADRI->agp.size, DRM_AGP, 0, 
-                 &pVIADRI->agp.handle) < 0) {
-	xf86DrvMsg(pScreen->myNum, X_ERROR,
-	    "[drm] Failed to map public agp area\n");
-        pVIADRI->agp.size = 0;
-        return GL_FALSE;
-    }  
-    /* Map AGP from kernel to Xserver - Not really needed */
-    drmMap(pVia->drmFD, pVIADRI->agp.handle,pVIADRI->agp.size, &agpaddr);
-
-    xf86DrvMsg(pScreen->myNum, X_INFO, 
-                "[drm] agpAddr = 0x%08lx\n", pVia->agpAddr);
-    xf86DrvMsg(pScreen->myNum, X_INFO, 
-                "[drm] agpSize = 0x%08lx\n", pVia->agpSize);
-    xf86DrvMsg(pScreen->myNum, X_INFO, 
-                "[drm] agp physical addr = 0x%08lx\n", agp_phys);
-
-    {
-	drm_via_agp_t agp;
-	agp.offset = 0;
-	agp.size = AGP_SIZE-AGP_CMDBUF_SIZE;
-	if (drmCommandWrite(pVia->drmFD, DRM_VIA_AGP_INIT, &agp,
-			    sizeof(drm_via_agp_t)) < 0) {
-	    drmUnmap(&agpaddr,pVia->agpSize);
-	    drmRmMap(pVia->drmFD,pVIADRI->agp.handle);
-	    drmAgpUnbind(pVia->drmFD, pVia->agpHandle);
-	    drmAgpFree(pVia->drmFD, pVia->agpHandle);
-	    drmAgpRelease(pVia->drmFD);
-	    return GL_FALSE;
-	}
-    }
-
-    return GL_TRUE;
-}
-
-static int VIADRIFBInit(DRIDriverContext * ctx, VIAPtr pVia)
-{   
-    int FBSize = pVia->FBFreeEnd-pVia->FBFreeStart;
-    int FBOffset = pVia->FBFreeStart; 
-    VIADRIPtr pVIADRI = pVia->devPrivate;
-    pVIADRI->fbOffset = FBOffset;
-    pVIADRI->fbSize = pVia->videoRambytes;
-
-    {
-	drm_via_fb_t fb;
-	fb.offset = FBOffset;
-	fb.size = FBSize;
-	
-	if (drmCommandWrite(pVia->drmFD, DRM_VIA_FB_INIT, &fb,
-			    sizeof(drm_via_fb_t)) < 0) {
-	    xf86DrvMsg(pScreen->myNum, X_ERROR,
-		       "[drm] failed to init frame buffer area\n");
-	    return GL_FALSE;
-	} else {
-	    xf86DrvMsg(pScreen->myNum, X_INFO,
-		       "[drm] FBFreeStart= 0x%08x FBFreeEnd= 0x%08x "
-		       "FBSize= 0x%08x\n",
-		       pVia->FBFreeStart, pVia->FBFreeEnd, FBSize);
-	    return GL_TRUE;	
-	}   
-    }
-}
-
-static int VIADRIPciInit(DRIDriverContext * ctx, VIAPtr pVia)
-{
-    return GL_TRUE;	
-}
-
-static int VIADRIScreenInit(DRIDriverContext * ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI;
-    int err;
-
-#if 0
-    ctx->shared.SAREASize = ((sizeof(drm_sarea_t) + 0xfff) & 0x1000);
-#else
-    if (sizeof(drm_sarea_t)+sizeof(drm_via_sarea_t) > SAREA_MAX) {
-	xf86DrvMsg(pScrn->scrnIndex, X_ERROR,
-			"Data does not fit in SAREA\n");
-	return GL_FALSE;
-    }
-    ctx->shared.SAREASize = SAREA_MAX;
-#endif
-
-    ctx->drmFD = drmOpen(VIAKernelDriverName, NULL);
-    if (ctx->drmFD < 0) {
-        fprintf(stderr, "[drm] drmOpen failed\n");
-        return 0;
-    }
-    pVia->drmFD = ctx->drmFD;
-
-    err = drmSetBusid(ctx->drmFD, ctx->pciBusID);
-    if (err < 0) {
-        fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
-                ctx->drmFD, ctx->pciBusID, strerror(-err));
-        return 0;
-    }
-
-    err = drmAddMap(ctx->drmFD, 0, ctx->shared.SAREASize, DRM_SHM,
-                  DRM_CONTAINS_LOCK, &ctx->shared.hSAREA);
-    if (err < 0) {
-        fprintf(stderr, "[drm] drmAddMap failed\n");
-        return 0;
-    }
-    fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
-            ctx->shared.SAREASize, ctx->shared.hSAREA);
-
-    if (drmMap(ctx->drmFD,
-               ctx->shared.hSAREA,
-               ctx->shared.SAREASize,
-               (drmAddressPtr)(&ctx->pSAREA)) < 0)
-    {
-        fprintf(stderr, "[drm] drmMap failed\n");
-        return 0;
-    }
-    memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
-    fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
-            ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
-
-    /* Need to AddMap the framebuffer and mmio regions here:
-     */
-    if (drmAddMap(ctx->drmFD,
-                  (drm_handle_t)ctx->FBStart,
-                  ctx->FBSize,
-                  DRM_FRAME_BUFFER,
-#ifndef _EMBEDDED
-                   0,
-#else
-                   DRM_READ_ONLY,
-#endif
-                   &ctx->shared.hFrameBuffer) < 0)
-    {
-        fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
-        return 0;
-    }
-
-    fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
-            ctx->shared.hFrameBuffer);
-
-    pVIADRI = (VIADRIPtr) CALLOC(sizeof(VIADRIRec));
-    if (!pVIADRI) {
-        drmClose(ctx->drmFD);
-        return GL_FALSE;
-    }
-    pVia->devPrivate = pVIADRI;
-    ctx->driverClientMsg = pVIADRI;
-    ctx->driverClientMsgSize = sizeof(*pVIADRI);
-
-    /* DRIScreenInit doesn't add all the common mappings.  Add additional mappings here. */
-    if (!VIADRIMapInit(ctx, pVia)) {
-	VIADRICloseScreen(ctx);
-	return GL_FALSE;
-    }
-
-    pVIADRI->regs.size = VIA_MMIO_REGSIZE;
-    pVIADRI->regs.handle = pVia->registerHandle;
-    xf86DrvMsg(pScreen->myNum, X_INFO, "[drm] mmio Registers = 0x%08lx\n",
-	pVIADRI->regs.handle);
-
-    if (drmMap(pVia->drmFD,
-               pVIADRI->regs.handle,
-               pVIADRI->regs.size,
-               (drmAddress *)&pVia->MapBase) != 0)
-    {
-        VIADRICloseScreen(ctx);
-        return GL_FALSE;
-    }
-
-    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[dri] mmio mapped.\n" );
-
-    VIAEnableMMIO(ctx);
-
-    /* Get video memory clock. */
-    VGAOUT8(0x3D4, 0x3D);
-    pVia->MemClk = (VGAIN8(0x3D5) & 0xF0) >> 4;
-    xf86DrvMsg(0, X_INFO, "[dri] MemClk (0x%x)\n", pVia->MemClk);
-
-    /* 3D rendering has noise if not enabled. */
-    VIAEnableExtendedFIFO(ctx);
-
-    VIAInitialize2DEngine(ctx);
-
-    /* Must disable MMIO or 3D won't work. */
-    VIADisableMMIO(ctx);
-
-    VIAInitialize3DEngine(ctx);
-
-    pVia->IsPCI = !VIADRIAgpInit(ctx, pVia);
-
-    if (pVia->IsPCI) {
-        VIADRIPciInit(ctx, pVia);
-	xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[dri] use pci.\n" );
-    }
-    else
-        xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[dri] use agp.\n" );
-
-    if (!(VIADRIFBInit(ctx, pVia))) {
-	VIADRICloseScreen(ctx);
-        xf86DrvMsg(pScrn->scrnIndex, X_ERROR, "[dri] frame buffer initialize fail .\n" );
-        return GL_FALSE;
-    }
-    
-    xf86DrvMsg(pScrn->scrnIndex, X_INFO, "[dri] frame buffer initialized.\n" );
- 
-    return VIADRIFinishScreenInit(ctx);
-}
-
-static void
-VIADRICloseScreen(DRIDriverContext * ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI=(VIADRIPtr)pVia->devPrivate;
-
-    VIADRIRingBufferCleanup(ctx);
-
-    if (pVia->MapBase) {
-	xf86DrvMsg(pScreen->myNum, X_INFO, "[drm] Unmapping MMIO registers\n");
-        drmUnmap(pVia->MapBase, pVIADRI->regs.size);
-    }
-
-    if (pVia->agpSize) {
-	xf86DrvMsg(pScreen->myNum, X_INFO, "[drm] Freeing agp memory\n");
-        drmAgpFree(pVia->drmFD, pVia->agpHandle);
-	xf86DrvMsg(pScreen->myNum, X_INFO, "[drm] Releasing agp module\n");
-    	drmAgpRelease(pVia->drmFD);
-    }
-
-#if 0
-    if (pVia->DRIIrqEnable) 
-#endif
-        VIADRIIrqExit(ctx);
-}
-
-static int
-VIADRIFinishScreenInit(DRIDriverContext * ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    VIADRIPtr pVIADRI;
-    int err;
-
-    err = drmCreateContext(ctx->drmFD, &ctx->serverContext);
-    if (err != 0) {
-        fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
-        return GL_FALSE;
-    }
-
-    DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0);
-
-
-    if (!VIADRIKernelInit(ctx, pVia)) {
-	VIADRICloseScreen(ctx);
-	return GL_FALSE;
-    }
-    xf86DrvMsg(pScreen->myNum, X_INFO, "[dri] kernel data initialized.\n");
-
-    /* set SAREA value */
-    {
-	drm_via_sarea_t *saPriv;
-
-	saPriv=(drm_via_sarea_t*)(((char*)ctx->pSAREA) +
-                               sizeof(drm_sarea_t));
-	assert(saPriv);
-	memset(saPriv, 0, sizeof(*saPriv));
-	saPriv->ctxOwner = -1;
-    }
-    pVIADRI=(VIADRIPtr)pVia->devPrivate;
-    pVIADRI->deviceID=pVia->Chipset;  
-    pVIADRI->width=ctx->shared.virtualWidth;
-    pVIADRI->height=ctx->shared.virtualHeight;
-    pVIADRI->mem=ctx->shared.fbSize;
-    pVIADRI->bytesPerPixel= (ctx->bpp+7) / 8; 
-    pVIADRI->sarea_priv_offset = sizeof(drm_sarea_t);
-    /* TODO */
-    pVIADRI->scrnX=pVIADRI->width;
-    pVIADRI->scrnY=pVIADRI->height;
-
-    /* Initialize IRQ */
-#if 0
-    if (pVia->DRIIrqEnable) 
-#endif
-	VIADRIIrqInit(ctx);
-    
-    pVIADRI->ringBufActive = 0;
-    VIADRIRingBufferInit(ctx);
-
-    return GL_TRUE;
-}
-
-/* Initialize the kernel data structures. */
-static int VIADRIKernelInit(DRIDriverContext * ctx, VIAPtr pVia)
-{
-    drm_via_init_t drmInfo;
-    memset(&drmInfo, 0, sizeof(drm_via_init_t));
-    drmInfo.sarea_priv_offset   = sizeof(drm_sarea_t);
-    drmInfo.func = VIA_INIT_MAP;
-    drmInfo.fb_offset           = pVia->FrameBufferBase;
-    drmInfo.mmio_offset         = pVia->registerHandle;
-    if (pVia->IsPCI)
-	drmInfo.agpAddr = (uint32_t)NULL;
-    else
-	drmInfo.agpAddr = (uint32_t)pVia->agpAddr;
-
-    if ((drmCommandWrite(pVia->drmFD, DRM_VIA_MAP_INIT,&drmInfo,
-			     sizeof(drm_via_init_t))) < 0)
-	    return GL_FALSE;
-
-    return GL_TRUE;
-}
-/* Add a map for the MMIO registers */
-static int VIADRIMapInit(DRIDriverContext * ctx, VIAPtr pVia)
-{
-    int flags = 0;
-
-    if (drmAddMap(pVia->drmFD, pVia->MmioBase, VIA_MMIO_REGSIZE,
-		  DRM_REGISTERS, flags, &pVia->registerHandle) < 0) {
-	return GL_FALSE;
-    }
-
-    xf86DrvMsg(pScreen->myNum, X_INFO,
-	"[drm] register handle = 0x%08lx\n", pVia->registerHandle);
-
-    return GL_TRUE;
-}
-
-static int viaValidateMode(const DRIDriverContext *ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-
-    return 1;
-}
-
-static int viaPostValidateMode(const DRIDriverContext *ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-
-    return 1;
-}
-
-static void VIAEnableMMIO(DRIDriverContext * ctx)
-{
-    /*vgaHWPtr hwp = VGAHWPTR(ctx);*/
-    VIAPtr pVia = VIAPTR(ctx);
-    unsigned char val;
-
-#if 0
-    if (xf86IsPrimaryPci(pVia->PciInfo)) {
-        /* If we are primary card, we still use std vga port. If we use
-         * MMIO, system will hang in vgaHWSave when our card used in
-         * PLE and KLE (integrated Trident MVP4)
-         */
-        vgaHWSetStdFuncs(hwp);
-    }
-    else {
-        vgaHWSetMmioFuncs(hwp, pVia->MapBase, 0x8000);
-    }
-#endif
-
-    val = VGAIN8(0x3c3);
-    VGAOUT8(0x3c3, val | 0x01);
-    val = VGAIN8(0x3cc);
-    VGAOUT8(0x3c2, val | 0x01);
-
-    /* Unlock Extended IO Space */
-    VGAOUT8(0x3c4, 0x10);
-    VGAOUT8(0x3c5, 0x01);
-
-    /* Enable MMIO */
-    if(!pVia->IsSecondary) {
-	VGAOUT8(0x3c4, 0x1a);
-	val = VGAIN8(0x3c5);
-#ifdef DEBUG
-	xf86DrvMsg(pScrn->scrnIndex, X_INFO, "primary val = %x\n", val);
-#endif
-	VGAOUT8(0x3c5, val | 0x68);
-    }
-    else {
-	VGAOUT8(0x3c4, 0x1a);
-	val = VGAIN8(0x3c5);
-#ifdef DEBUG
-	xf86DrvMsg(pScrn->scrnIndex, X_INFO, "secondary val = %x\n", val);
-#endif
-	VGAOUT8(0x3c5, val | 0x38);
-    }
-
-    /* Unlock CRTC registers */
-    VGAOUT8(0x3d4, 0x47);
-    VGAOUT8(0x3d5, 0x00);
-
-    return;
-}
-
-static void VIADisableMMIO(DRIDriverContext * ctx)
-{
-    VIAPtr pVia = VIAPTR(ctx);
-    unsigned char val;
-
-    VGAOUT8(0x3c4, 0x1a);
-    val = VGAIN8(0x3c5);
-    VGAOUT8(0x3c5, val & 0x97);
-
-    return;
-}
-
-static void VIADisableExtendedFIFO(DRIDriverContext *ctx)
-{
-    VIAPtr  pVia = VIAPTR(ctx);
-    uint32_t  dwGE230, dwGE298;
-
-    /* Cause of exit XWindow will dump back register value, others chipset no
-     * need to set extended fifo value */
-    if (pVia->Chipset == VIA_CLE266 && pVia->ChipRev < 15 &&
-        (ctx->shared.virtualWidth > 1024 || pVia->HasSecondary)) {
-        /* Turn off Extend FIFO */
-        /* 0x298[29] */
-        dwGE298 = VIAGETREG(0x298);
-        VIASETREG(0x298, dwGE298 | 0x20000000);
-        /* 0x230[21] */
-        dwGE230 = VIAGETREG(0x230);
-        VIASETREG(0x230, dwGE230 & ~0x00200000);
-        /* 0x298[29] */
-        dwGE298 = VIAGETREG(0x298);
-        VIASETREG(0x298, dwGE298 & ~0x20000000);
-    }
-}
-
-static void VIAEnableExtendedFIFO(DRIDriverContext *ctx)
-{
-    VIAPtr  pVia = VIAPTR(ctx);
-    uint8_t   bRegTemp;
-    uint32_t  dwGE230, dwGE298;
-
-    switch (pVia->Chipset) {
-    case VIA_CLE266:
-        if (pVia->ChipRev > 14) {  /* For 3123Cx */
-            if (pVia->HasSecondary) {  /* SAMM or DuoView case */
-                if (ctx->shared.virtualWidth >= 1024)
-    	        {
-    	            /* 3c5.16[0:5] */
-        	        VGAOUT8(0x3C4, 0x16);
-            	    bRegTemp = VGAIN8(0x3C5);
-    	            bRegTemp &= ~0x3F;
-        	        bRegTemp |= 0x1C;
-            	    VGAOUT8(0x3C5, bRegTemp);
-        	        /* 3c5.17[0:6] */
-            	    VGAOUT8(0x3C4, 0x17);
-                	bRegTemp = VGAIN8(0x3C5);
-    	            bRegTemp &= ~0x7F;
-        	        bRegTemp |= 0x3F;
-            	    VGAOUT8(0x3C5, bRegTemp);
-            	    pVia->EnableExtendedFIFO = GL_TRUE;
-    	        }
-            }
-            else   /* Single view or Simultaneoue case */
-            {
-                if (ctx->shared.virtualWidth > 1024)
-    	        {
-    	            /* 3c5.16[0:5] */
-        	        VGAOUT8(0x3C4, 0x16);
-            	    bRegTemp = VGAIN8(0x3C5);
-    	            bRegTemp &= ~0x3F;
-        	        bRegTemp |= 0x17;
-            	    VGAOUT8(0x3C5, bRegTemp);
-        	        /* 3c5.17[0:6] */
-            	    VGAOUT8(0x3C4, 0x17);
-                	bRegTemp = VGAIN8(0x3C5);
-    	            bRegTemp &= ~0x7F;
-        	        bRegTemp |= 0x2F;
-            	    VGAOUT8(0x3C5, bRegTemp);
-            	    pVia->EnableExtendedFIFO = GL_TRUE;
-    	        }
-            }
-            /* 3c5.18[0:5] */
-            VGAOUT8(0x3C4, 0x18);
-            bRegTemp = VGAIN8(0x3C5);
-            bRegTemp &= ~0x3F;
-            bRegTemp |= 0x17;
-            bRegTemp |= 0x40;  /* force the preq always higher than treq */
-            VGAOUT8(0x3C5, bRegTemp);
-        }
-        else {      /* for 3123Ax */
-            if (ctx->shared.virtualWidth > 1024 || pVia->HasSecondary) {
-                /* Turn on Extend FIFO */
-                /* 0x298[29] */
-                dwGE298 = VIAGETREG(0x298);
-                VIASETREG(0x298, dwGE298 | 0x20000000);
-                /* 0x230[21] */
-                dwGE230 = VIAGETREG(0x230);
-                VIASETREG(0x230, dwGE230 | 0x00200000);
-                /* 0x298[29] */
-                dwGE298 = VIAGETREG(0x298);
-                VIASETREG(0x298, dwGE298 & ~0x20000000);
-
-                /* 3c5.16[0:5] */
-                VGAOUT8(0x3C4, 0x16);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp |= 0x17;
-                /* bRegTemp |= 0x10; */
-                VGAOUT8(0x3C5, bRegTemp);
-                /* 3c5.17[0:6] */
-                VGAOUT8(0x3C4, 0x17);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x7F;
-                bRegTemp |= 0x2F;
-                /*bRegTemp |= 0x1F;*/
-                VGAOUT8(0x3C5, bRegTemp);
-                /* 3c5.18[0:5] */
-                VGAOUT8(0x3C4, 0x18);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp |= 0x17;
-                bRegTemp |= 0x40;  /* force the preq always higher than treq */
-                VGAOUT8(0x3C5, bRegTemp);
-          	    pVia->EnableExtendedFIFO = GL_TRUE;
-            }
-        }
-        break;
-    case VIA_KM400:
-        if (pVia->HasSecondary) {  /* SAMM or DuoView case */
-            if ((ctx->shared.virtualWidth >= 1600) &&
-                (pVia->MemClk <= VIA_MEM_DDR200)) {
-        	    /* enable CRT extendded FIFO */
-            	VGAOUT8(0x3C4, 0x17);
-                VGAOUT8(0x3C5, 0x1C);
-    	        /* revise second display queue depth and read threshold */
-        	    VGAOUT8(0x3C4, 0x16);
-            	bRegTemp = VGAIN8(0x3C5);
-    	        bRegTemp &= ~0x3F;
-    	        bRegTemp = (bRegTemp) | (0x09);
-                VGAOUT8(0x3C5, bRegTemp);
-            }
-            else {
-                /* enable CRT extendded FIFO */
-                VGAOUT8(0x3C4, 0x17);
-                VGAOUT8(0x3C5,0x3F);
-                /* revise second display queue depth and read threshold */
-                VGAOUT8(0x3C4, 0x16);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp = (bRegTemp) | (0x1C);
-                VGAOUT8(0x3C5, bRegTemp);
-            }
-            /* 3c5.18[0:5] */
-            VGAOUT8(0x3C4, 0x18);
-            bRegTemp = VGAIN8(0x3C5);
-            bRegTemp &= ~0x3F;
-            bRegTemp |= 0x17;
-            bRegTemp |= 0x40;  /* force the preq always higher than treq */
-            VGAOUT8(0x3C5, bRegTemp);
-       	    pVia->EnableExtendedFIFO = GL_TRUE;
-        }
-        else {
-            if ( (ctx->shared.virtualWidth > 1024) && (ctx->shared.virtualWidth <= 1280) )
-            {
-                /* enable CRT extendded FIFO */
-                VGAOUT8(0x3C4, 0x17);
-                VGAOUT8(0x3C5, 0x3F);
-                /* revise second display queue depth and read threshold */
-                VGAOUT8(0x3C4, 0x16);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp = (bRegTemp) | (0x17);
-                VGAOUT8(0x3C5, bRegTemp);
-           	    pVia->EnableExtendedFIFO = GL_TRUE;
-            }
-            else if ((ctx->shared.virtualWidth > 1280))
-            {
-                /* enable CRT extendded FIFO */
-                VGAOUT8(0x3C4, 0x17);
-                VGAOUT8(0x3C5, 0x3F);
-                /* revise second display queue depth and read threshold */
-                VGAOUT8(0x3C4, 0x16);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp = (bRegTemp) | (0x1C);
-                VGAOUT8(0x3C5, bRegTemp);
-           	    pVia->EnableExtendedFIFO = GL_TRUE;
-            }
-            else
-            {
-                /* enable CRT extendded FIFO */
-                VGAOUT8(0x3C4, 0x17);
-                VGAOUT8(0x3C5, 0x3F);
-                /* revise second display queue depth and read threshold */
-                VGAOUT8(0x3C4, 0x16);
-                bRegTemp = VGAIN8(0x3C5);
-                bRegTemp &= ~0x3F;
-                bRegTemp = (bRegTemp) | (0x10);
-                VGAOUT8(0x3C5, bRegTemp);
-            }
-            /* 3c5.18[0:5] */
-            VGAOUT8(0x3C4, 0x18);
-            bRegTemp = VGAIN8(0x3C5);
-            bRegTemp &= ~0x3F;
-            bRegTemp |= 0x17;
-            bRegTemp |= 0x40;  /* force the preq always higher than treq */
-            VGAOUT8(0x3C5, bRegTemp);
-        }
-        break;
-    case VIA_K8M800:
-        /*=* R1 Display FIFO depth (384 /8 -1 -> 0xbf) SR17[7:0] (8bits) *=*/
-        VGAOUT8(0x3c4, 0x17);
-        VGAOUT8(0x3c5, 0xbf);
-
-        /*=* R2 Display fetch datum threshold value (328/4 -> 0x52)
-             SR16[5:0], SR16[7] (7bits) *=*/
-        VGAOUT8(0x3c4, 0x16);
-        bRegTemp = VGAIN8(0x3c5) & ~0xBF;
-        bRegTemp |= (0x52 & 0x3F);
-        bRegTemp |= ((0x52 & 0x40) << 1);
-        VGAOUT8(0x3c5, bRegTemp);
-
-        /*=* R3 Switch to the highest agent threshold value (74 -> 0x4a)
-             SR18[5:0], SR18[7] (7bits) *=*/
-        VGAOUT8(0x3c4, 0x18);
-        bRegTemp = VGAIN8(0x3c5) & ~0xBF;
-        bRegTemp |= (0x4a & 0x3F);
-        bRegTemp |= ((0x4a & 0x40) << 1);
-        VGAOUT8(0x3c5, bRegTemp);
-#if 0
-        /*=* R4 Fetch Number for a scan line (unit: 8 bytes)
-             SR1C[7:0], SR1D[1:0] (10bits) *=*/
-        wRegTemp = (pBIOSInfo->offsetWidthByQWord >> 1) + 4;
-        VGAOUT8(0x3c4, 0x1c);
-        VGAOUT8(0x3c5, (uint8_t)(wRegTemp & 0xFF));
-        VGAOUT8(0x3c4, 0x1d);
-        bRegTemp = VGAIN8(0x3c5) & ~0x03;
-        VGAOUT8(0x3c5, bRegTemp | ((wRegTemp & 0x300) >> 8));
-#endif
-        if (ctx->shared.virtualWidth >= 1400 && ctx->bpp == 32)
-        {
-            /*=* Max. length for a request SR22[4:0] (64/4 -> 0x10) *=*/
-            VGAOUT8(0x3c4, 0x22);
-            bRegTemp = VGAIN8(0x3c5) & ~0x1F;
-            VGAOUT8(0x3c5, bRegTemp | 0x10);
-        }
-        else
-        {
-            /*=* Max. length for a request SR22[4:0]
-                 (128/4 -> over flow 0x0) *=*/
-            VGAOUT8(0x3c4, 0x22);
-            bRegTemp = VGAIN8(0x3c5) & ~0x1F;
-            VGAOUT8(0x3c5, bRegTemp);
-        }
-        break;
-    case VIA_PM800:
-        /*=* R1 Display FIFO depth (96-1 -> 0x5f) SR17[7:0] (8bits) *=*/
-        VGAOUT8(0x3c4, 0x17);
-        VGAOUT8(0x3c5, 0x5f);
-
-        /*=* R2 Display fetch datum threshold value (32 -> 0x20)
-             SR16[5:0], SR16[7] (7bits) *=*/
-        VGAOUT8(0x3c4, 0x16);
-        bRegTemp = VGAIN8(0x3c5) & ~0xBF;
-        bRegTemp |= (0x20 & 0x3F);
-        bRegTemp |= ((0x20 & 0x40) << 1);
-        VGAOUT8(0x3c5, bRegTemp);
-
-        /*=* R3 Switch to the highest agent threshold value (16 -> 0x10)
-             SR18[5:0], SR18[7] (7bits) *=*/
-        VGAOUT8(0x3c4, 0x18);
-        bRegTemp = VGAIN8(0x3c5) & ~0xBF;
-        bRegTemp |= (0x10 & 0x3F);
-        bRegTemp |= ((0x10 & 0x40) << 1);
-        VGAOUT8(0x3c5, bRegTemp);
-#if 0
-        /*=* R4 Fetch Number for a scan line (unit: 8 bytes)
-             SR1C[7:0], SR1D[1:0] (10bits) *=*/
-        wRegTemp = (pBIOSInfo->offsetWidthByQWord >> 1) + 4;
-        VGAOUT8(0x3c4, 0x1c);
-        VGAOUT8(0x3c5, (uint8_t)(wRegTemp & 0xFF));
-        VGAOUT8(0x3c4, 0x1d);
-        bRegTemp = VGAIN8(0x3c5) & ~0x03;
-        VGAOUT8(0x3c5, bRegTemp | ((wRegTemp & 0x300) >> 8));
-#endif
-        if (ctx->shared.virtualWidth >= 1400 && ctx->bpp == 32)
-        {
-            /*=* Max. length for a request SR22[4:0] (64/4 -> 0x10) *=*/
-            VGAOUT8(0x3c4, 0x22);
-            bRegTemp = VGAIN8(0x3c5) & ~0x1F;
-            VGAOUT8(0x3c5, bRegTemp | 0x10);
-        }
-        else
-        {
-            /*=* Max. length for a request SR22[4:0] (0x1F) *=*/
-            VGAOUT8(0x3c4, 0x22);
-            bRegTemp = VGAIN8(0x3c5) & ~0x1F;
-            VGAOUT8(0x3c5, bRegTemp | 0x1F);
-        }
-        break;
-    default:
-        break;
-    }
-}
-
-static void VIAInitialize2DEngine(DRIDriverContext *ctx)
-{
-    VIAPtr  pVia = VIAPTR(ctx);
-    uint32_t  dwVQStartAddr, dwVQEndAddr;
-    uint32_t  dwVQLen, dwVQStartL, dwVQEndL, dwVQStartEndH;
-    uint32_t  dwGEMode;
-
-    /* init 2D engine regs to reset 2D engine */
-    VIASETREG(0x04, 0x0);
-    VIASETREG(0x08, 0x0);
-    VIASETREG(0x0c, 0x0);
-    VIASETREG(0x10, 0x0);
-    VIASETREG(0x14, 0x0);
-    VIASETREG(0x18, 0x0);
-    VIASETREG(0x1c, 0x0);
-    VIASETREG(0x20, 0x0);
-    VIASETREG(0x24, 0x0);
-    VIASETREG(0x28, 0x0);
-    VIASETREG(0x2c, 0x0);
-    VIASETREG(0x30, 0x0);
-    VIASETREG(0x34, 0x0);
-    VIASETREG(0x38, 0x0);
-    VIASETREG(0x3c, 0x0);
-    VIASETREG(0x40, 0x0);
-
-    VIADisableMMIO(ctx);
-
-    /* Init AGP and VQ regs */
-    VIASETREG(0x43c, 0x00100000);
-    VIASETREG(0x440, 0x00000000);
-    VIASETREG(0x440, 0x00333004);
-    VIASETREG(0x440, 0x60000000);
-    VIASETREG(0x440, 0x61000000);
-    VIASETREG(0x440, 0x62000000);
-    VIASETREG(0x440, 0x63000000);
-    VIASETREG(0x440, 0x64000000);
-    VIASETREG(0x440, 0x7D000000);
-
-    VIASETREG(0x43c, 0xfe020000);
-    VIASETREG(0x440, 0x00000000);
-
-    if (pVia->VQStart != 0) {
-        /* Enable VQ */
-        dwVQStartAddr = pVia->VQStart;
-        dwVQEndAddr = pVia->VQEnd;
-        dwVQStartL = 0x50000000 | (dwVQStartAddr & 0xFFFFFF);
-        dwVQEndL = 0x51000000 | (dwVQEndAddr & 0xFFFFFF);
-        dwVQStartEndH = 0x52000000 | ((dwVQStartAddr & 0xFF000000) >> 24) |
-                        ((dwVQEndAddr & 0xFF000000) >> 16);
-        dwVQLen = 0x53000000 | (VIA_VQ_SIZE >> 3);
-
-        VIASETREG(0x43c, 0x00fe0000);
-        VIASETREG(0x440, 0x080003fe);
-        VIASETREG(0x440, 0x0a00027c);
-        VIASETREG(0x440, 0x0b000260);
-        VIASETREG(0x440, 0x0c000274);
-        VIASETREG(0x440, 0x0d000264);
-        VIASETREG(0x440, 0x0e000000);
-        VIASETREG(0x440, 0x0f000020);
-        VIASETREG(0x440, 0x1000027e);
-        VIASETREG(0x440, 0x110002fe);
-        VIASETREG(0x440, 0x200f0060);
-
-        VIASETREG(0x440, 0x00000006);
-        VIASETREG(0x440, 0x40008c0f);
-        VIASETREG(0x440, 0x44000000);
-        VIASETREG(0x440, 0x45080c04);
-        VIASETREG(0x440, 0x46800408);
-
-        VIASETREG(0x440, dwVQStartEndH);
-        VIASETREG(0x440, dwVQStartL);
-        VIASETREG(0x440, dwVQEndL);
-        VIASETREG(0x440, dwVQLen);
-    }
-    else {
-        /* Diable VQ */
-        VIASETREG(0x43c, 0x00fe0000);
-        VIASETREG(0x440, 0x00000004);
-        VIASETREG(0x440, 0x40008c0f);
-        VIASETREG(0x440, 0x44000000);
-        VIASETREG(0x440, 0x45080c04);
-        VIASETREG(0x440, 0x46800408);
-    }
-
-    dwGEMode = 0;
-
-    switch (ctx->bpp) {
-    case 16:
-        dwGEMode |= VIA_GEM_16bpp;
-        break;
-    case 32:
-        dwGEMode |= VIA_GEM_32bpp;
-        break;
-    default:
-        dwGEMode |= VIA_GEM_8bpp;
-        break;
-    }
-
-#if 0
-    switch (ctx->shared.virtualWidth) {
-    case 800:
-        dwGEMode |= VIA_GEM_800;
-        break;
-    case 1024:
-        dwGEMode |= VIA_GEM_1024;
-        break;
-    case 1280:
-        dwGEMode |= VIA_GEM_1280;
-        break;
-    case 1600:
-        dwGEMode |= VIA_GEM_1600;
-        break;
-    case 2048:
-        dwGEMode |= VIA_GEM_2048;
-        break;
-    default:
-        dwGEMode |= VIA_GEM_640;
-        break;
-    }
-#endif
-    
-    VIAEnableMMIO(ctx);
-
-    /* Set BPP and Pitch */
-    VIASETREG(VIA_REG_GEMODE, dwGEMode);
-
-    /* Set Src and Dst base address and pitch, pitch is qword */
-    VIASETREG(VIA_REG_SRCBASE, 0x0);
-    VIASETREG(VIA_REG_DSTBASE, 0x0);
-    VIASETREG(VIA_REG_PITCH, VIA_PITCH_ENABLE |
-              ((ctx->shared.virtualWidth * ctx->bpp >> 3) >> 3) |
-              (((ctx->shared.virtualWidth * ctx->bpp >> 3) >> 3) << 16));
-}
-
-static int b3DRegsInitialized = 0;
-
-static void VIAInitialize3DEngine(DRIDriverContext *ctx)
-{
-    VIAPtr  pVia = VIAPTR(ctx);
-    int i;
-
-    if (!b3DRegsInitialized)
-    {
-
-        VIASETREG(0x43C, 0x00010000);
-
-        for (i = 0; i <= 0x7D; i++)
-        {
-            VIASETREG(0x440, (uint32_t) i << 24);
-        }
-
-        VIASETREG(0x43C, 0x00020000);
-
-        for (i = 0; i <= 0x94; i++)
-        {
-            VIASETREG(0x440, (uint32_t) i << 24);
-        }
-
-        VIASETREG(0x440, 0x82400000);
-
-        VIASETREG(0x43C, 0x01020000);
-
-
-        for (i = 0; i <= 0x94; i++)
-        {
-            VIASETREG(0x440, (uint32_t) i << 24);
-        }
-
-        VIASETREG(0x440, 0x82400000);
-        VIASETREG(0x43C, 0xfe020000);
-
-        for (i = 0; i <= 0x03; i++)
-        {
-            VIASETREG(0x440, (uint32_t) i << 24);
-        }
-
-        VIASETREG(0x43C, 0x00030000);
-
-        for (i = 0; i <= 0xff; i++)
-        {
-            VIASETREG(0x440, 0);
-        }
-        VIASETREG(0x43C, 0x00100000);
-        VIASETREG(0x440, 0x00333004);
-        VIASETREG(0x440, 0x10000002);
-        VIASETREG(0x440, 0x60000000);
-        VIASETREG(0x440, 0x61000000);
-        VIASETREG(0x440, 0x62000000);
-        VIASETREG(0x440, 0x63000000);
-        VIASETREG(0x440, 0x64000000);
-
-        VIASETREG(0x43C, 0x00fe0000);
-
-        if (pVia->ChipRev >= 3 )
-            VIASETREG(0x440,0x40008c0f);
-        else
-            VIASETREG(0x440,0x4000800f);
-
-        VIASETREG(0x440,0x44000000);
-        VIASETREG(0x440,0x45080C04);
-        VIASETREG(0x440,0x46800408);
-        VIASETREG(0x440,0x50000000);
-        VIASETREG(0x440,0x51000000);
-        VIASETREG(0x440,0x52000000);
-        VIASETREG(0x440,0x53000000);
-
-        b3DRegsInitialized = 1;
-        xf86DrvMsg(pScrn->scrnIndex, X_INFO,
-                   "3D Engine has been initialized.\n");
-    }
-
-    VIASETREG(0x43C,0x00fe0000);
-    VIASETREG(0x440,0x08000001);
-    VIASETREG(0x440,0x0A000183);
-    VIASETREG(0x440,0x0B00019F);
-    VIASETREG(0x440,0x0C00018B);
-    VIASETREG(0x440,0x0D00019B);
-    VIASETREG(0x440,0x0E000000);
-    VIASETREG(0x440,0x0F000000);
-    VIASETREG(0x440,0x10000000);
-    VIASETREG(0x440,0x11000000);
-    VIASETREG(0x440,0x20000000);
-}
-
-static int
-WaitIdleCLE266(VIAPtr pVia)
-{
-    int loop = 0;
-
-    /*mem_barrier();*/
-
-    while (!(VIAGETREG(VIA_REG_STATUS) & VIA_VR_QUEUE_BUSY) && (loop++ < MAXLOOP))
-        ;
-
-    while ((VIAGETREG(VIA_REG_STATUS) &
-          (VIA_CMD_RGTR_BUSY | VIA_2D_ENG_BUSY | VIA_3D_ENG_BUSY)) &&
-          (loop++ < MAXLOOP))
-        ;
-
-    return loop >= MAXLOOP;
-}
-
-static int viaInitFBDev(DRIDriverContext *ctx)
-{
-    VIAPtr pVia = CALLOC(sizeof(*pVia));
-
-    ctx->driverPrivate = (void *)pVia;
-
-    switch (ctx->chipset) {
-    case PCI_CHIP_CLE3122:
-    case PCI_CHIP_CLE3022:
-        pVia->Chipset = VIA_CLE266;
-        break;
-    case PCI_CHIP_VT7205:
-    case PCI_CHIP_VT3205:
-        pVia->Chipset = VIA_KM400;
-        break;
-    case PCI_CHIP_VT3204:
-    case PCI_CHIP_VT3344:
-        pVia->Chipset = VIA_K8M800;
-        break;
-    case PCI_CHIP_VT3259:
-        pVia->Chipset = VIA_PM800;
-        break;
-    default:
-        xf86DrvMsg(0, X_ERROR, "VIA: Unknown device ID (0x%x)\n", ctx->chipset);
-    }
-
-    /* _SOLO TODO XXX need to read ChipRev too */
-    pVia->ChipRev = 0;
-
-    pVia->videoRambytes = ctx->shared.fbSize;
-    pVia->MmioBase = ctx->MMIOStart;
-    pVia->FrameBufferBase = ctx->FBStart & 0xfc000000;
-
-    pVia->FBFreeStart = ctx->shared.virtualWidth * ctx->cpp *
-        ctx->shared.virtualHeight;
-
-#if 1
-    /* Alloc a second framebuffer for the second head */
-    pVia->FBFreeStart += ctx->shared.virtualWidth * ctx->cpp *
-	ctx->shared.virtualHeight;
-#endif
-
-    pVia->VQStart = pVia->FBFreeStart;
-    pVia->VQEnd = pVia->FBFreeStart + VIA_VQ_SIZE - 1;
-
-    pVia->FBFreeStart += VIA_VQ_SIZE;
-
-    pVia->FBFreeEnd = pVia->videoRambytes;
-
-    if (!VIADRIScreenInit(ctx))
-        return 0;
-
-    return 1;
-}
-
-static void viaHaltFBDev(DRIDriverContext *ctx)
-{
-    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
-    drmClose(ctx->drmFD);
-
-    if (ctx->driverPrivate) {
-        free(ctx->driverPrivate);
-        ctx->driverPrivate = 0;
-    }
-}
-
-static int viaEngineShutdown(const DRIDriverContext *ctx)
-{
-    return 1;
-}
-
-static int viaEngineRestore(const DRIDriverContext *ctx)
-{
-    return 1;
-}
-
-const struct DRIDriverRec __driDriver =
-{
-    viaValidateMode,
-    viaPostValidateMode,
-    viaInitFBDev,
-    viaHaltFBDev,
-    viaEngineShutdown,
-    viaEngineRestore,  
-    0,
-};
-
diff --git a/src/mesa/glapi/glapi.c b/src/mesa/glapi/glapi.c
index 13de594aafb..ce85cf6a87b 100644
--- a/src/mesa/glapi/glapi.c
+++ b/src/mesa/glapi/glapi.c
@@ -59,7 +59,7 @@
 #endif
 
 #include "glapi/glapi.h"
-#include "glapi/glapitable.h"
+#include "glapi/glapi_priv.h"
 
 extern _glapi_proc __glapi_noop_table[];
 
@@ -291,45 +291,3 @@ _glapi_get_dispatch(void)
    return _glapi_Dispatch;
 #endif
 }
-
-
-
-
-/*
- * The dispatch table size (number of entries) is the size of the
- * _glapi_table struct plus the number of dynamic entries we can add.
- * The extra slots can be filled in by DRI drivers that register new extension
- * functions.
- */
-#define DISPATCH_TABLE_SIZE (sizeof(struct _glapi_table) / sizeof(void *) + MAX_EXTENSION_FUNCS)
-
-
-/**
- * Return size of dispatch table struct as number of functions (or
- * slots).
- */
-PUBLIC GLuint
-_glapi_get_dispatch_table_size(void)
-{
-   return DISPATCH_TABLE_SIZE;
-}
-
-
-/**
- * Make sure there are no NULL pointers in the given dispatch table.
- * Intended for debugging purposes.
- */
-void
-_glapi_check_table_not_null(const struct _glapi_table *table)
-{
-#if 0 /* enable this for extra DEBUG */
-   const GLuint entries = _glapi_get_dispatch_table_size();
-   const void **tab = (const void **) table;
-   GLuint i;
-   for (i = 1; i < entries; i++) {
-      assert(tab[i]);
-   }
-#else
-   (void) table;
-#endif
-}
diff --git a/src/mesa/glapi/glapi.h b/src/mesa/glapi/glapi.h
index 1ca2e4beff1..7dcf2e8910b 100644
--- a/src/mesa/glapi/glapi.h
+++ b/src/mesa/glapi/glapi.h
@@ -165,29 +165,8 @@ extern _glapi_proc
 _glapi_get_proc_address(const char *funcName);
 
 
-/**
- * GL API local functions and defines
- */
-
-extern void
-init_glapi_relocs_once(void);
-
-extern void
-_glapi_check_table_not_null(const struct _glapi_table *table);
-
-
-extern void
-_glapi_check_table(const struct _glapi_table *table);
-
-
 extern const char *
 _glapi_get_proc_name(unsigned int offset);
 
 
-/*
- * Number of extension functions which we can dynamically add at runtime.
- */
-#define MAX_EXTENSION_FUNCS 300
-
-
 #endif
diff --git a/src/mesa/glapi/glapi_entrypoint.c b/src/mesa/glapi/glapi_entrypoint.c
new file mode 100644
index 00000000000..5e6e5995f24
--- /dev/null
+++ b/src/mesa/glapi/glapi_entrypoint.c
@@ -0,0 +1,331 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file glapi_entrypoint.c
+ *
+ * Arch-specific code for manipulating GL API entrypoints (dispatch stubs).
+ */
+
+
+#ifdef HAVE_DIX_CONFIG_H
+#include <dix-config.h>
+#include "glapi/mesa.h"
+#else
+#include "main/glheader.h"
+#include "main/compiler.h"
+#endif
+
+#include "glapi/glapi.h"
+#include "glapi/glapi_priv.h"
+
+
+#ifdef USE_X86_ASM
+
+#if defined( GLX_USE_TLS )
+extern       GLubyte gl_dispatch_functions_start[];
+extern       GLubyte gl_dispatch_functions_end[];
+#else
+extern const GLubyte gl_dispatch_functions_start[];
+#endif
+
+#endif /* USE_X86_ASM */
+
+
+#if defined(DISPATCH_FUNCTION_SIZE)
+
+_glapi_proc
+get_entrypoint_address(GLuint functionOffset)
+{
+   return (_glapi_proc) (gl_dispatch_functions_start
+                         + (DISPATCH_FUNCTION_SIZE * functionOffset));
+}
+
+#endif
+
+
+#if defined(PTHREADS) || defined(GLX_USE_TLS)
+
+/**
+ * Perform platform-specific GL API entry-point fixups.
+ */
+static void
+init_glapi_relocs( void )
+{
+#if defined(USE_X86_ASM) && defined(GLX_USE_TLS) && !defined(GLX_X86_READONLY_TEXT)
+    extern unsigned long _x86_get_dispatch(void);
+    char run_time_patch[] = {
+       0x65, 0xa1, 0, 0, 0, 0 /* movl %gs:0,%eax */
+    };
+    GLuint *offset = (GLuint *) &run_time_patch[2]; /* 32-bits for x86/32 */
+    const GLubyte * const get_disp = (const GLubyte *) run_time_patch;
+    GLubyte * curr_func = (GLubyte *) gl_dispatch_functions_start;
+
+    *offset = _x86_get_dispatch();
+    while ( curr_func != (GLubyte *) gl_dispatch_functions_end ) {
+	(void) memcpy( curr_func, get_disp, sizeof(run_time_patch));
+	curr_func += DISPATCH_FUNCTION_SIZE;
+    }
+#endif
+#ifdef USE_SPARC_ASM
+    extern void __glapi_sparc_icache_flush(unsigned int *);
+    static const unsigned int template[] = {
+#ifdef GLX_USE_TLS
+	0x05000000, /* sethi %hi(_glapi_tls_Dispatch), %g2 */
+	0x8730e00a, /* srl %g3, 10, %g3 */
+	0x8410a000, /* or %g2, %lo(_glapi_tls_Dispatch), %g2 */
+#ifdef __arch64__
+	0xc259c002, /* ldx [%g7 + %g2], %g1 */
+	0xc2584003, /* ldx [%g1 + %g3], %g1 */
+#else
+	0xc201c002, /* ld [%g7 + %g2], %g1 */
+	0xc2004003, /* ld [%g1 + %g3], %g1 */
+#endif
+	0x81c04000, /* jmp %g1 */
+	0x01000000, /* nop  */
+#else
+#ifdef __arch64__
+	0x03000000, /* 64-bit 0x00 --> sethi %hh(_glapi_Dispatch), %g1 */
+	0x05000000, /* 64-bit 0x04 --> sethi %lm(_glapi_Dispatch), %g2 */
+	0x82106000, /* 64-bit 0x08 --> or %g1, %hm(_glapi_Dispatch), %g1 */
+	0x8730e00a, /* 64-bit 0x0c --> srl %g3, 10, %g3 */
+	0x83287020, /* 64-bit 0x10 --> sllx %g1, 32, %g1 */
+	0x82004002, /* 64-bit 0x14 --> add %g1, %g2, %g1 */
+	0xc2586000, /* 64-bit 0x18 --> ldx [%g1 + %lo(_glapi_Dispatch)], %g1 */
+#else
+	0x03000000, /* 32-bit 0x00 --> sethi %hi(_glapi_Dispatch), %g1 */
+	0x8730e00a, /* 32-bit 0x04 --> srl %g3, 10, %g3 */
+	0xc2006000, /* 32-bit 0x08 --> ld [%g1 + %lo(_glapi_Dispatch)], %g1 */
+#endif
+	0x80a06000, /*             --> cmp %g1, 0 */
+	0x02800005, /*             --> be +4*5 */
+	0x01000000, /*             -->  nop  */
+#ifdef __arch64__
+	0xc2584003, /* 64-bit      --> ldx [%g1 + %g3], %g1 */
+#else
+	0xc2004003, /* 32-bit      --> ld [%g1 + %g3], %g1 */
+#endif
+	0x81c04000, /*             --> jmp %g1 */
+	0x01000000, /*             --> nop  */
+#ifdef __arch64__
+	0x9de3bf80, /* 64-bit      --> save  %sp, -128, %sp */
+#else
+	0x9de3bfc0, /* 32-bit      --> save  %sp, -64, %sp */
+#endif
+	0xa0100003, /*             --> mov  %g3, %l0 */
+	0x40000000, /*             --> call _glapi_get_dispatch */
+	0x01000000, /*             -->  nop */
+	0x82100008, /*             --> mov %o0, %g1 */
+	0x86100010, /*             --> mov %l0, %g3 */
+	0x10bffff7, /*             --> ba -4*9 */
+	0x81e80000, /*             -->  restore  */
+#endif
+    };
+#ifdef GLX_USE_TLS
+    extern unsigned int __glapi_sparc_tls_stub;
+    extern unsigned long __glapi_sparc_get_dispatch(void);
+    unsigned int *code = &__glapi_sparc_tls_stub;
+    unsigned long dispatch = __glapi_sparc_get_dispatch();
+#else
+    extern unsigned int __glapi_sparc_pthread_stub;
+    unsigned int *code = &__glapi_sparc_pthread_stub;
+    unsigned long dispatch = (unsigned long) &_glapi_Dispatch;
+    unsigned long call_dest = (unsigned long ) &_glapi_get_dispatch;
+    int idx;
+#endif
+
+#if defined(GLX_USE_TLS)
+    code[0] = template[0] | (dispatch >> 10);
+    code[1] = template[1];
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | (dispatch & 0x3ff);
+    code[3] = template[3];
+    __glapi_sparc_icache_flush(&code[2]);
+    code[4] = template[4];
+    code[5] = template[5];
+    __glapi_sparc_icache_flush(&code[4]);
+    code[6] = template[6];
+    __glapi_sparc_icache_flush(&code[6]);
+#else
+#if defined(__arch64__)
+    code[0] = template[0] | (dispatch >> (32 + 10));
+    code[1] = template[1] | ((dispatch & 0xffffffff) >> 10);
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | ((dispatch >> 32) & 0x3ff);
+    code[3] = template[3];
+    __glapi_sparc_icache_flush(&code[2]);
+    code[4] = template[4];
+    code[5] = template[5];
+    __glapi_sparc_icache_flush(&code[4]);
+    code[6] = template[6] | (dispatch & 0x3ff);
+    idx = 7;
+#else
+    code[0] = template[0] | (dispatch >> 10);
+    code[1] = template[1];
+    __glapi_sparc_icache_flush(&code[0]);
+    code[2] = template[2] | (dispatch & 0x3ff);
+    idx = 3;
+#endif
+    code[idx + 0] = template[idx + 0];
+    __glapi_sparc_icache_flush(&code[idx - 1]);
+    code[idx + 1] = template[idx + 1];
+    code[idx + 2] = template[idx + 2];
+    __glapi_sparc_icache_flush(&code[idx + 1]);
+    code[idx + 3] = template[idx + 3];
+    code[idx + 4] = template[idx + 4];
+    __glapi_sparc_icache_flush(&code[idx + 3]);
+    code[idx + 5] = template[idx + 5];
+    code[idx + 6] = template[idx + 6];
+    __glapi_sparc_icache_flush(&code[idx + 5]);
+    code[idx + 7] = template[idx + 7];
+    code[idx + 8] = template[idx + 8] |
+	    (((call_dest - ((unsigned long) &code[idx + 8]))
+	      >> 2) & 0x3fffffff);
+    __glapi_sparc_icache_flush(&code[idx + 7]);
+    code[idx + 9] = template[idx + 9];
+    code[idx + 10] = template[idx + 10];
+    __glapi_sparc_icache_flush(&code[idx + 9]);
+    code[idx + 11] = template[idx + 11];
+    code[idx + 12] = template[idx + 12];
+    __glapi_sparc_icache_flush(&code[idx + 11]);
+    code[idx + 13] = template[idx + 13];
+    __glapi_sparc_icache_flush(&code[idx + 13]);
+#endif
+#endif
+}
+
+void
+init_glapi_relocs_once( void )
+{
+   static pthread_once_t once_control = PTHREAD_ONCE_INIT;
+   pthread_once( & once_control, init_glapi_relocs );
+}
+
+#else
+
+void
+init_glapi_relocs_once( void ) { }
+
+#endif /* defined(PTHREADS) || defined(GLX_USE_TLS) */
+
+
+#ifdef USE_SPARC_ASM
+extern void __glapi_sparc_icache_flush(unsigned int *);
+#endif
+
+/**
+ * Generate a dispatch function (entrypoint) which jumps through
+ * the given slot number (offset) in the current dispatch table.
+ * We need assembly language in order to accomplish this.
+ */
+_glapi_proc
+generate_entrypoint(GLuint functionOffset)
+{
+#if defined(USE_X86_ASM)
+   /* 32 is chosen as something of a magic offset.  For x86, the dispatch
+    * at offset 32 is the first one where the offset in the
+    * "jmp OFFSET*4(%eax)" can't be encoded in a single byte.
+    */
+   const GLubyte * const template_func = gl_dispatch_functions_start 
+     + (DISPATCH_FUNCTION_SIZE * 32);
+   GLubyte * const code = (GLubyte *) malloc(DISPATCH_FUNCTION_SIZE);
+
+
+   if ( code != NULL ) {
+      (void) memcpy(code, template_func, DISPATCH_FUNCTION_SIZE);
+      fill_in_entrypoint_offset( (_glapi_proc) code, functionOffset );
+   }
+
+   return (_glapi_proc) code;
+#elif defined(USE_SPARC_ASM)
+
+#if defined(PTHREADS) || defined(GLX_USE_TLS)
+   static const unsigned int template[] = {
+      0x07000000, /* sethi %hi(0), %g3 */
+      0x8210000f, /* mov  %o7, %g1 */
+      0x40000000, /* call */
+      0x9e100001, /* mov  %g1, %o7 */
+   };
+#ifdef GLX_USE_TLS
+   extern unsigned int __glapi_sparc_tls_stub;
+   unsigned long call_dest = (unsigned long ) &__glapi_sparc_tls_stub;
+#else
+   extern unsigned int __glapi_sparc_pthread_stub;
+   unsigned long call_dest = (unsigned long ) &__glapi_sparc_pthread_stub;
+#endif
+   unsigned int *code = (unsigned int *) malloc(sizeof(template));
+   if (code) {
+      code[0] = template[0] | (functionOffset & 0x3fffff);
+      code[1] = template[1];
+      __glapi_sparc_icache_flush(&code[0]);
+      code[2] = template[2] |
+         (((call_dest - ((unsigned long) &code[2]))
+	   >> 2) & 0x3fffffff);
+      code[3] = template[3];
+      __glapi_sparc_icache_flush(&code[2]);
+   }
+   return (_glapi_proc) code;
+#endif
+
+#else
+   (void) functionOffset;
+   return NULL;
+#endif /* USE_*_ASM */
+}
+
+
+/**
+ * This function inserts a new dispatch offset into the assembly language
+ * stub that was generated with the preceeding function.
+ */
+void
+fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset)
+{
+#if defined(USE_X86_ASM)
+   GLubyte * const code = (GLubyte *) entrypoint;
+
+#if DISPATCH_FUNCTION_SIZE == 32
+   *((unsigned int *)(code + 11)) = 4 * offset;
+   *((unsigned int *)(code + 22)) = 4 * offset;
+#elif DISPATCH_FUNCTION_SIZE == 16 && defined( GLX_USE_TLS )
+   *((unsigned int *)(code +  8)) = 4 * offset;
+#elif DISPATCH_FUNCTION_SIZE == 16
+   *((unsigned int *)(code +  7)) = 4 * offset;
+#else
+# error Invalid DISPATCH_FUNCTION_SIZE!
+#endif
+
+#elif defined(USE_SPARC_ASM)
+   unsigned int *code = (unsigned int *) entrypoint;
+   code[0] &= ~0x3fffff;
+   code[0] |= (offset * sizeof(void *)) & 0x3fffff;
+   __glapi_sparc_icache_flush(&code[0]);
+#else
+
+   /* an unimplemented architecture */
+   (void) entrypoint;
+   (void) offset;
+
+#endif /* USE_*_ASM */
+}
diff --git a/src/mesa/glapi/glapi_getproc.c b/src/mesa/glapi/glapi_getproc.c
index a6dbf173e82..295657875dc 100644
--- a/src/mesa/glapi/glapi_getproc.c
+++ b/src/mesa/glapi/glapi_getproc.c
@@ -39,25 +39,19 @@
 #endif
 
 #include "glapi/glapi.h"
-#include "glapi/glapioffsets.h"
+#include "glapi/glapi_priv.h"
 #include "glapi/glapitable.h"
+#include "glapi/glapioffsets.h"
 
 
-#if defined(USE_X64_64_ASM) && defined(GLX_USE_TLS)
-# define DISPATCH_FUNCTION_SIZE  16
-#elif defined(USE_X86_ASM)
-# if defined(THREADS) && !defined(GLX_USE_TLS)
-#  define DISPATCH_FUNCTION_SIZE  32
-# else
-#  define DISPATCH_FUNCTION_SIZE  16
-# endif
-#endif
+/**********************************************************************
+ * Static function management.
+ */
+
 
-#if !defined(DISPATCH_FUNCTION_SIZE) && !defined(XFree86Server) && !defined(XGLServer)
+#if !defined(DISPATCH_FUNCTION_SIZE) && !defined(XFree86Server)
 # define NEED_FUNCTION_POINTER
 #endif
-
-/* The code in this file is auto-generated with Python */
 #include "glapi/glprocs.h"
 
 
@@ -66,13 +60,13 @@
  * and return the corresponding glprocs_table_t entry.
  */
 static const glprocs_table_t *
-find_entry( const char * n )
+get_static_proc( const char * n )
 {
    GLuint i;
    for (i = 0; static_functions[i].Name_offset >= 0; i++) {
       const char *testName = gl_string_table + static_functions[i].Name_offset;
 #ifdef MANGLE
-      /* skip the "m" prefix on the name */
+      /* skip the prefix on the name */
       if (strcmp(testName, n + 1) == 0)
 #else
       if (strcmp(testName, n) == 0)
@@ -92,27 +86,16 @@ find_entry( const char * n )
 static GLint
 get_static_proc_offset(const char *funcName)
 {
-   const glprocs_table_t * const f = find_entry( funcName );
-   if (f) {
-      return f->Offset;
+   const glprocs_table_t * const f = get_static_proc( funcName );
+   if (f == NULL) {
+      return -1;
    }
-   return -1;
-}
-
 
-#ifdef USE_X86_ASM
-
-#if defined( GLX_USE_TLS )
-extern       GLubyte gl_dispatch_functions_start[];
-extern       GLubyte gl_dispatch_functions_end[];
-#else
-extern const GLubyte gl_dispatch_functions_start[];
-#endif
-
-#endif /* USE_X86_ASM */
+   return f->Offset;
+}
 
 
-#if !defined(XFree86Server) && !defined(XGLServer)
+#if !defined(XFree86Server)
 
 /**
  * Return dispatch function address for the named static (built-in) function.
@@ -121,27 +104,32 @@ extern const GLubyte gl_dispatch_functions_start[];
 static _glapi_proc
 get_static_proc_address(const char *funcName)
 {
-   const glprocs_table_t * const f = find_entry( funcName );
-   if (f) {
+   const glprocs_table_t * const f = get_static_proc( funcName );
+   if (f == NULL) {
+      return NULL;
+   }
+
 #if defined(DISPATCH_FUNCTION_SIZE) && defined(GLX_INDIRECT_RENDERING)
-      return (f->Address == NULL)
-	 ? (_glapi_proc) (gl_dispatch_functions_start
-			  + (DISPATCH_FUNCTION_SIZE * f->Offset))
-         : f->Address;
+   return (f->Address == NULL)
+      ? get_entrypoint_address(f->Offset)
+      : f->Address;
 #elif defined(DISPATCH_FUNCTION_SIZE)
-      return (_glapi_proc) (gl_dispatch_functions_start 
-                            + (DISPATCH_FUNCTION_SIZE * f->Offset));
+   return get_entrypoint_address(f->Offset);
 #else
-      return f->Address;
+   return f->Address;
 #endif
-   }
-   else {
-      return NULL;
-   }
 }
 
-#endif /* !defined(XFree86Server) && !defined(XGLServer) */
+#else
+
+static _glapi_proc
+get_static_proc_address(const char *funcName)
+{
+   (void) funcName;
+   return NULL;
+}
 
+#endif /* !defined(XFree86Server) */
 
 
 /**
@@ -162,172 +150,6 @@ get_static_proc_name( GLuint offset )
 
 
 
-#if defined(PTHREADS) || defined(GLX_USE_TLS)
-
-/**
- * Perform platform-specific GL API entry-point fixups.
- */
-static void
-init_glapi_relocs( void )
-{
-#if defined(USE_X86_ASM) && defined(GLX_USE_TLS) && !defined(GLX_X86_READONLY_TEXT)
-    extern unsigned long _x86_get_dispatch(void);
-    char run_time_patch[] = {
-       0x65, 0xa1, 0, 0, 0, 0 /* movl %gs:0,%eax */
-    };
-    GLuint *offset = (GLuint *) &run_time_patch[2]; /* 32-bits for x86/32 */
-    const GLubyte * const get_disp = (const GLubyte *) run_time_patch;
-    GLubyte * curr_func = (GLubyte *) gl_dispatch_functions_start;
-
-    *offset = _x86_get_dispatch();
-    while ( curr_func != (GLubyte *) gl_dispatch_functions_end ) {
-	(void) memcpy( curr_func, get_disp, sizeof(run_time_patch));
-	curr_func += DISPATCH_FUNCTION_SIZE;
-    }
-#endif
-#ifdef USE_SPARC_ASM
-    extern void __glapi_sparc_icache_flush(unsigned int *);
-    static const unsigned int template[] = {
-#ifdef GLX_USE_TLS
-	0x05000000, /* sethi %hi(_glapi_tls_Dispatch), %g2 */
-	0x8730e00a, /* srl %g3, 10, %g3 */
-	0x8410a000, /* or %g2, %lo(_glapi_tls_Dispatch), %g2 */
-#ifdef __arch64__
-	0xc259c002, /* ldx [%g7 + %g2], %g1 */
-	0xc2584003, /* ldx [%g1 + %g3], %g1 */
-#else
-	0xc201c002, /* ld [%g7 + %g2], %g1 */
-	0xc2004003, /* ld [%g1 + %g3], %g1 */
-#endif
-	0x81c04000, /* jmp %g1 */
-	0x01000000, /* nop  */
-#else
-#ifdef __arch64__
-	0x03000000, /* 64-bit 0x00 --> sethi %hh(_glapi_Dispatch), %g1 */
-	0x05000000, /* 64-bit 0x04 --> sethi %lm(_glapi_Dispatch), %g2 */
-	0x82106000, /* 64-bit 0x08 --> or %g1, %hm(_glapi_Dispatch), %g1 */
-	0x8730e00a, /* 64-bit 0x0c --> srl %g3, 10, %g3 */
-	0x83287020, /* 64-bit 0x10 --> sllx %g1, 32, %g1 */
-	0x82004002, /* 64-bit 0x14 --> add %g1, %g2, %g1 */
-	0xc2586000, /* 64-bit 0x18 --> ldx [%g1 + %lo(_glapi_Dispatch)], %g1 */
-#else
-	0x03000000, /* 32-bit 0x00 --> sethi %hi(_glapi_Dispatch), %g1 */
-	0x8730e00a, /* 32-bit 0x04 --> srl %g3, 10, %g3 */
-	0xc2006000, /* 32-bit 0x08 --> ld [%g1 + %lo(_glapi_Dispatch)], %g1 */
-#endif
-	0x80a06000, /*             --> cmp %g1, 0 */
-	0x02800005, /*             --> be +4*5 */
-	0x01000000, /*             -->  nop  */
-#ifdef __arch64__
-	0xc2584003, /* 64-bit      --> ldx [%g1 + %g3], %g1 */
-#else
-	0xc2004003, /* 32-bit      --> ld [%g1 + %g3], %g1 */
-#endif
-	0x81c04000, /*             --> jmp %g1 */
-	0x01000000, /*             --> nop  */
-#ifdef __arch64__
-	0x9de3bf80, /* 64-bit      --> save  %sp, -128, %sp */
-#else
-	0x9de3bfc0, /* 32-bit      --> save  %sp, -64, %sp */
-#endif
-	0xa0100003, /*             --> mov  %g3, %l0 */
-	0x40000000, /*             --> call _glapi_get_dispatch */
-	0x01000000, /*             -->  nop */
-	0x82100008, /*             --> mov %o0, %g1 */
-	0x86100010, /*             --> mov %l0, %g3 */
-	0x10bffff7, /*             --> ba -4*9 */
-	0x81e80000, /*             -->  restore  */
-#endif
-    };
-#ifdef GLX_USE_TLS
-    extern unsigned int __glapi_sparc_tls_stub;
-    extern unsigned long __glapi_sparc_get_dispatch(void);
-    unsigned int *code = &__glapi_sparc_tls_stub;
-    unsigned long dispatch = __glapi_sparc_get_dispatch();
-#else
-    extern unsigned int __glapi_sparc_pthread_stub;
-    unsigned int *code = &__glapi_sparc_pthread_stub;
-    unsigned long dispatch = (unsigned long) &_glapi_Dispatch;
-    unsigned long call_dest = (unsigned long ) &_glapi_get_dispatch;
-    int idx;
-#endif
-
-#if defined(GLX_USE_TLS)
-    code[0] = template[0] | (dispatch >> 10);
-    code[1] = template[1];
-    __glapi_sparc_icache_flush(&code[0]);
-    code[2] = template[2] | (dispatch & 0x3ff);
-    code[3] = template[3];
-    __glapi_sparc_icache_flush(&code[2]);
-    code[4] = template[4];
-    code[5] = template[5];
-    __glapi_sparc_icache_flush(&code[4]);
-    code[6] = template[6];
-    __glapi_sparc_icache_flush(&code[6]);
-#else
-#if defined(__arch64__)
-    code[0] = template[0] | (dispatch >> (32 + 10));
-    code[1] = template[1] | ((dispatch & 0xffffffff) >> 10);
-    __glapi_sparc_icache_flush(&code[0]);
-    code[2] = template[2] | ((dispatch >> 32) & 0x3ff);
-    code[3] = template[3];
-    __glapi_sparc_icache_flush(&code[2]);
-    code[4] = template[4];
-    code[5] = template[5];
-    __glapi_sparc_icache_flush(&code[4]);
-    code[6] = template[6] | (dispatch & 0x3ff);
-    idx = 7;
-#else
-    code[0] = template[0] | (dispatch >> 10);
-    code[1] = template[1];
-    __glapi_sparc_icache_flush(&code[0]);
-    code[2] = template[2] | (dispatch & 0x3ff);
-    idx = 3;
-#endif
-    code[idx + 0] = template[idx + 0];
-    __glapi_sparc_icache_flush(&code[idx - 1]);
-    code[idx + 1] = template[idx + 1];
-    code[idx + 2] = template[idx + 2];
-    __glapi_sparc_icache_flush(&code[idx + 1]);
-    code[idx + 3] = template[idx + 3];
-    code[idx + 4] = template[idx + 4];
-    __glapi_sparc_icache_flush(&code[idx + 3]);
-    code[idx + 5] = template[idx + 5];
-    code[idx + 6] = template[idx + 6];
-    __glapi_sparc_icache_flush(&code[idx + 5]);
-    code[idx + 7] = template[idx + 7];
-    code[idx + 8] = template[idx + 8] |
-	    (((call_dest - ((unsigned long) &code[idx + 8]))
-	      >> 2) & 0x3fffffff);
-    __glapi_sparc_icache_flush(&code[idx + 7]);
-    code[idx + 9] = template[idx + 9];
-    code[idx + 10] = template[idx + 10];
-    __glapi_sparc_icache_flush(&code[idx + 9]);
-    code[idx + 11] = template[idx + 11];
-    code[idx + 12] = template[idx + 12];
-    __glapi_sparc_icache_flush(&code[idx + 11]);
-    code[idx + 13] = template[idx + 13];
-    __glapi_sparc_icache_flush(&code[idx + 13]);
-#endif
-#endif
-}
-
-void
-init_glapi_relocs_once( void )
-{
-   static pthread_once_t once_control = PTHREAD_ONCE_INIT;
-   pthread_once( & once_control, init_glapi_relocs );
-}
-
-#else
-
-void
-init_glapi_relocs_once( void ) { }
-
-#endif /* defined(PTHREADS) || defined(GLX_USE_TLS) */
-
-
-
 /**********************************************************************
  * Extension function management.
  */
@@ -378,110 +200,63 @@ struct _glapi_function {
 };
 
 
+/*
+ * Number of extension functions which we can dynamically add at runtime.
+ */
+#define MAX_EXTENSION_FUNCS 300
+
+
 static struct _glapi_function ExtEntryTable[MAX_EXTENSION_FUNCS];
 static GLuint NumExtEntryPoints = 0;
 
-#ifdef USE_SPARC_ASM
-extern void __glapi_sparc_icache_flush(unsigned int *);
-#endif
 
-static void
-fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset);
-
-/**
- * Generate a dispatch function (entrypoint) which jumps through
- * the given slot number (offset) in the current dispatch table.
- * We need assembly language in order to accomplish this.
- */
-static _glapi_proc
-generate_entrypoint(GLuint functionOffset)
+static struct _glapi_function *
+get_extension_proc(const char *funcName)
 {
-#if defined(USE_X86_ASM)
-   /* 32 is chosen as something of a magic offset.  For x86, the dispatch
-    * at offset 32 is the first one where the offset in the
-    * "jmp OFFSET*4(%eax)" can't be encoded in a single byte.
-    */
-   const GLubyte * const template_func = gl_dispatch_functions_start 
-     + (DISPATCH_FUNCTION_SIZE * 32);
-   GLubyte * const code = (GLubyte *) malloc(DISPATCH_FUNCTION_SIZE);
-
-
-   if ( code != NULL ) {
-      (void) memcpy(code, template_func, DISPATCH_FUNCTION_SIZE);
-      fill_in_entrypoint_offset( (_glapi_proc) code, functionOffset );
+   GLuint i;
+   for (i = 0; i < NumExtEntryPoints; i++) {
+      if (strcmp(ExtEntryTable[i].name, funcName) == 0) {
+         return & ExtEntryTable[i];
+      }
    }
+   return NULL;
+}
 
-   return (_glapi_proc) code;
-#elif defined(USE_SPARC_ASM)
-
-#if defined(PTHREADS) || defined(GLX_USE_TLS)
-   static const unsigned int template[] = {
-      0x07000000, /* sethi %hi(0), %g3 */
-      0x8210000f, /* mov  %o7, %g1 */
-      0x40000000, /* call */
-      0x9e100001, /* mov  %g1, %o7 */
-   };
-#ifdef GLX_USE_TLS
-   extern unsigned int __glapi_sparc_tls_stub;
-   unsigned long call_dest = (unsigned long ) &__glapi_sparc_tls_stub;
-#else
-   extern unsigned int __glapi_sparc_pthread_stub;
-   unsigned long call_dest = (unsigned long ) &__glapi_sparc_pthread_stub;
-#endif
-   unsigned int *code = (unsigned int *) malloc(sizeof(template));
-   if (code) {
-      code[0] = template[0] | (functionOffset & 0x3fffff);
-      code[1] = template[1];
-      __glapi_sparc_icache_flush(&code[0]);
-      code[2] = template[2] |
-         (((call_dest - ((unsigned long) &code[2]))
-	   >> 2) & 0x3fffffff);
-      code[3] = template[3];
-      __glapi_sparc_icache_flush(&code[2]);
+
+static GLint
+get_extension_proc_offset(const char *funcName)
+{
+   const struct _glapi_function * const f = get_extension_proc( funcName );
+   if (f == NULL) {
+      return -1;
    }
-   return (_glapi_proc) code;
-#endif
 
-#else
-   (void) functionOffset;
-   return NULL;
-#endif /* USE_*_ASM */
+   return f->dispatch_offset;
 }
 
 
-/**
- * This function inserts a new dispatch offset into the assembly language
- * stub that was generated with the preceeding function.
- */
-static void
-fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset)
+static _glapi_proc
+get_extension_proc_address(const char *funcName)
 {
-#if defined(USE_X86_ASM)
-   GLubyte * const code = (GLubyte *) entrypoint;
-
-#if DISPATCH_FUNCTION_SIZE == 32
-   *((unsigned int *)(code + 11)) = 4 * offset;
-   *((unsigned int *)(code + 22)) = 4 * offset;
-#elif DISPATCH_FUNCTION_SIZE == 16 && defined( GLX_USE_TLS )
-   *((unsigned int *)(code +  8)) = 4 * offset;
-#elif DISPATCH_FUNCTION_SIZE == 16
-   *((unsigned int *)(code +  7)) = 4 * offset;
-#else
-# error Invalid DISPATCH_FUNCTION_SIZE!
-#endif
+   const struct _glapi_function * const f = get_extension_proc( funcName );
+   if (f == NULL) {
+      return NULL;
+   }
 
-#elif defined(USE_SPARC_ASM)
-   unsigned int *code = (unsigned int *) entrypoint;
-   code[0] &= ~0x3fffff;
-   code[0] |= (offset * sizeof(void *)) & 0x3fffff;
-   __glapi_sparc_icache_flush(&code[0]);
-#else
+   return f->dispatch_stub;
+}
 
-   /* an unimplemented architecture */
-   (void) entrypoint;
-   (void) offset;
 
-#endif /* USE_*_ASM */
+static const char *
+get_extension_proc_name(GLuint offset)
+{
+   GLuint i;
+   for (i = 0; i < NumExtEntryPoints; i++) {
+      if (ExtEntryTable[i].dispatch_offset == offset) {
+         return ExtEntryTable[i].name;
+      }
+   }
+   return NULL;
 }
 
 
@@ -518,20 +293,55 @@ static struct _glapi_function *
 add_function_name( const char * funcName )
 {
    struct _glapi_function * entry = NULL;
-   
-   if (NumExtEntryPoints < MAX_EXTENSION_FUNCS) {
-      _glapi_proc entrypoint = generate_entrypoint(~0);
-      if (entrypoint != NULL) {
-	 entry = & ExtEntryTable[NumExtEntryPoints];
-
-	 ExtEntryTable[NumExtEntryPoints].name = str_dup(funcName);
-	 ExtEntryTable[NumExtEntryPoints].parameter_signature = NULL;
-	 ExtEntryTable[NumExtEntryPoints].dispatch_offset = ~0;
-	 ExtEntryTable[NumExtEntryPoints].dispatch_stub = entrypoint;
-	 NumExtEntryPoints++;
-      }
+   _glapi_proc entrypoint = NULL;
+   char * name_dup = NULL;
+
+   if (NumExtEntryPoints >= MAX_EXTENSION_FUNCS)
+      return NULL;
+
+   if (funcName == NULL)
+      return NULL;
+
+   name_dup = str_dup(funcName);
+   if (name_dup == NULL)
+      return NULL;
+
+   entrypoint = generate_entrypoint(~0);
+
+   if (entrypoint == NULL) {
+      free(name_dup);
+      return NULL;
    }
 
+   entry = & ExtEntryTable[NumExtEntryPoints];
+   NumExtEntryPoints++;
+
+   entry->name = name_dup;
+   entry->parameter_signature = NULL;
+   entry->dispatch_offset = ~0;
+   entry->dispatch_stub = entrypoint;
+
+   return entry;
+}
+
+
+static struct _glapi_function *
+set_entry_info( struct _glapi_function * entry, const char * signature, unsigned offset )
+{
+   char * sig_dup = NULL;
+
+   if (signature == NULL)
+      return NULL;
+
+   sig_dup = str_dup(signature);
+   if (sig_dup == NULL)
+      return NULL;
+
+   fill_in_entrypoint_offset(entry->dispatch_stub, offset);
+
+   entry->parameter_signature = sig_dup;
+   entry->dispatch_offset = offset;
+
    return entry;
 }
 
@@ -593,88 +403,103 @@ _glapi_add_dispatch( const char * const * function_names,
    struct _glapi_function * entry[8];
    GLboolean is_static[8];
    unsigned i;
-   unsigned j;
    int offset = ~0;
-   int new_offset;
 
+   init_glapi_relocs_once();
 
    (void) memset( is_static, 0, sizeof( is_static ) );
    (void) memset( entry, 0, sizeof( entry ) );
 
+   /* Find the _single_ dispatch offset for all function names that already
+    * exist (and have a dispatch offset).
+    */
+
    for ( i = 0 ; function_names[i] != NULL ; i++ ) {
-      /* Do some trivial validation on the name of the function.
-       */
+      const char * funcName = function_names[i];
+      int static_offset;
+      int extension_offset;
 
-      if (!function_names[i] || function_names[i][0] != 'g' || function_names[i][1] != 'l')
+      if (funcName[0] != 'g' || funcName[1] != 'l')
          return -1;
-   
-      /* Determine if the named function already exists.  If the function does
-       * exist, it must have the same parameter signature as the function
-       * being added.
-       */
-
-      new_offset = get_static_proc_offset(function_names[i]);
-      if (new_offset >= 0) {
+
+      /* search built-in functions */
+      static_offset = get_static_proc_offset(funcName);
+
+      if (static_offset >= 0) {
+
+	 is_static[i] = GL_TRUE;
+
 	 /* FIXME: Make sure the parameter signatures match!  How do we get
 	  * FIXME: the parameter signature for static functions?
 	  */
 
-	 if ( (offset != ~0) && (new_offset != offset) ) {
+	 if ( (offset != ~0) && (static_offset != offset) ) {
 	    return -1;
 	 }
 
-	 is_static[i] = GL_TRUE;
-	 offset = new_offset;
+	 offset = static_offset;
+
+	 continue;
       }
-   
-   
-      for ( j = 0 ; j < NumExtEntryPoints ; j++ ) {
-	 if (strcmp(ExtEntryTable[j].name, function_names[i]) == 0) {
-	    /* The offset may be ~0 if the function name was added by
-	     * glXGetProcAddress but never filled in by the driver.
-	     */
-
-	    if (ExtEntryTable[j].dispatch_offset != ~0) {
-	       if (strcmp(real_sig, ExtEntryTable[j].parameter_signature) 
-		   != 0) {
-		  return -1;
-	       }
-
-	       if ( (offset != ~0) && (ExtEntryTable[j].dispatch_offset != offset) ) {
-		  return -1;
-	       }
-
-	       offset = ExtEntryTable[j].dispatch_offset;
-	    }
-	    
-	    entry[i] = & ExtEntryTable[j];
-	    break;
+
+      /* search added extension functions */
+      entry[i] = get_extension_proc(funcName);
+
+      if (entry[i] != NULL) {
+	 extension_offset = entry[i]->dispatch_offset;
+
+	 /* The offset may be ~0 if the function name was added by
+	  * glXGetProcAddress but never filled in by the driver.
+	  */
+
+	 if (extension_offset == ~0) {
+	    continue;
 	 }
+
+	 if (strcmp(real_sig, entry[i]->parameter_signature) != 0) {
+	    return -1;
+	 }
+
+	 if ( (offset != ~0) && (extension_offset != offset) ) {
+	    return -1;
+	 }
+
+	 offset = extension_offset;
       }
    }
 
+   /* If all function names are either new (or with no dispatch offset),
+    * allocate a new dispatch offset.
+    */
+
    if (offset == ~0) {
       offset = next_dynamic_offset;
       next_dynamic_offset++;
    }
 
+   /* Fill in the dispatch offset for the new function names (and those with
+    * no dispatch offset).
+    */
+
    for ( i = 0 ; function_names[i] != NULL ; i++ ) {
-      if (! is_static[i] ) {
+      if (is_static[i]) {
+	 continue;
+      }
+
+      /* generate entrypoints for new function names */
+      if (entry[i] == NULL) {
+	 entry[i] = add_function_name( function_names[i] );
 	 if (entry[i] == NULL) {
-	    entry[i] = add_function_name( function_names[i] );
-	    if (entry[i] == NULL) {
-	       /* FIXME: Possible memory leak here.
-		*/
-	       return -1;
-	    }
+	    /* FIXME: Possible memory leak here. */
+	    return -1;
 	 }
+      }
 
-	 entry[i]->parameter_signature = str_dup(real_sig);
-	 fill_in_entrypoint_offset(entry[i]->dispatch_stub, offset);
-	 entry[i]->dispatch_offset = offset;
+      if (entry[i]->dispatch_offset == ~0) {
+	 set_entry_info( entry[i], real_sig, offset );
       }
    }
-   
+
    return offset;
 }
 
@@ -685,13 +510,13 @@ _glapi_add_dispatch( const char * const * function_names,
 PUBLIC GLint
 _glapi_get_proc_offset(const char *funcName)
 {
+   GLint offset;
+
    /* search extension functions first */
-   GLuint i;
-   for (i = 0; i < NumExtEntryPoints; i++) {
-      if (strcmp(ExtEntryTable[i].name, funcName) == 0) {
-         return ExtEntryTable[i].dispatch_offset;
-      }
-   }
+   offset = get_extension_proc_offset(funcName);
+   if (offset >= 0)
+      return offset;
+
    /* search static functions */
    return get_static_proc_offset(funcName);
 }
@@ -706,11 +531,14 @@ _glapi_get_proc_offset(const char *funcName)
 PUBLIC _glapi_proc
 _glapi_get_proc_address(const char *funcName)
 {
+   _glapi_proc func;
    struct _glapi_function * entry;
-   GLuint i;
+
+   init_glapi_relocs_once();
 
 #ifdef MANGLE
-   if (funcName[0] != 'm' || funcName[1] != 'g' || funcName[2] != 'l')
+   /* skip the prefix on the name */
+   if (funcName[1] != 'g' || funcName[2] != 'l')
       return NULL;
 #else
    if (funcName[0] != 'g' || funcName[1] != 'l')
@@ -718,23 +546,21 @@ _glapi_get_proc_address(const char *funcName)
 #endif
 
    /* search extension functions first */
-   for (i = 0; i < NumExtEntryPoints; i++) {
-      if (strcmp(ExtEntryTable[i].name, funcName) == 0) {
-         return ExtEntryTable[i].dispatch_stub;
-      }
-   }
+   func = get_extension_proc_address(funcName);
+   if (func)
+      return func;
 
-#if !defined( XFree86Server ) && !defined( XGLServer )
    /* search static functions */
-   {
-      const _glapi_proc func = get_static_proc_address(funcName);
-      if (func)
-         return func;
-   }
-#endif /* !defined( XFree86Server ) */
+   func = get_static_proc_address(funcName);
+   if (func)
+      return func;
 
+   /* generate entrypoint, dispatch offset must be filled in by the driver */
    entry = add_function_name(funcName);
-   return (entry == NULL) ? NULL : entry->dispatch_stub;
+   if (entry == NULL)
+      return NULL;
+
+   return entry->dispatch_stub;
 }
 
 
@@ -746,7 +572,6 @@ _glapi_get_proc_address(const char *funcName)
 const char *
 _glapi_get_proc_name(GLuint offset)
 {
-   GLuint i;
    const char * n;
 
    /* search built-in functions */
@@ -756,16 +581,56 @@ _glapi_get_proc_name(GLuint offset)
    }
 
    /* search added extension functions */
-   for (i = 0; i < NumExtEntryPoints; i++) {
-      if (ExtEntryTable[i].dispatch_offset == offset) {
-         return ExtEntryTable[i].name;
-      }
-   }
-   return NULL;
+   return get_extension_proc_name(offset);
 }
 
 
 
+/**********************************************************************
+ * GL API table functions.
+ */
+
+
+/*
+ * The dispatch table size (number of entries) is the size of the
+ * _glapi_table struct plus the number of dynamic entries we can add.
+ * The extra slots can be filled in by DRI drivers that register new extension
+ * functions.
+ */
+#define DISPATCH_TABLE_SIZE (sizeof(struct _glapi_table) / sizeof(void *) + MAX_EXTENSION_FUNCS)
+
+
+/**
+ * Return size of dispatch table struct as number of functions (or
+ * slots).
+ */
+PUBLIC GLuint
+_glapi_get_dispatch_table_size(void)
+{
+   return DISPATCH_TABLE_SIZE;
+}
+
+
+/**
+ * Make sure there are no NULL pointers in the given dispatch table.
+ * Intended for debugging purposes.
+ */
+void
+_glapi_check_table_not_null(const struct _glapi_table *table)
+{
+#ifdef EXTRA_DEBUG /* set to DEBUG for extra DEBUG */
+   const GLuint entries = _glapi_get_dispatch_table_size();
+   const void **tab = (const void **) table;
+   GLuint i;
+   for (i = 1; i < entries; i++) {
+      assert(tab[i]);
+   }
+#else
+   (void) table;
+#endif
+}
+
+
 /**
  * Do some spot checks to be sure that the dispatch table
  * slots are assigned correctly. For debugging only.
@@ -773,7 +638,7 @@ _glapi_get_proc_name(GLuint offset)
 void
 _glapi_check_table(const struct _glapi_table *table)
 {
-#if 0 /* enable this for extra DEBUG */
+#ifdef EXTRA_DEBUG /* set to DEBUG for extra DEBUG */
    {
       GLuint BeginOffset = _glapi_get_proc_offset("glBegin");
       char *BeginFunc = (char*) &table->Begin;
diff --git a/src/mesa/glapi/glapi_priv.h b/src/mesa/glapi/glapi_priv.h
new file mode 100644
index 00000000000..7cd81ee8dca
--- /dev/null
+++ b/src/mesa/glapi/glapi_priv.h
@@ -0,0 +1,66 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.1
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef _GLAPI_PRIV_H
+#define _GLAPI_PRIV_H
+
+#include "glthread.h"
+
+extern void
+_glapi_check_table_not_null(const struct _glapi_table *table);
+
+
+extern void
+_glapi_check_table(const struct _glapi_table *table);
+
+
+extern void
+init_glapi_relocs_once(void);
+
+
+extern _glapi_proc
+generate_entrypoint(GLuint functionOffset);
+
+
+extern void
+fill_in_entrypoint_offset(_glapi_proc entrypoint, GLuint offset);
+
+
+extern _glapi_proc
+get_entrypoint_address(GLuint functionOffset);
+
+
+#if defined(USE_X64_64_ASM) && defined(GLX_USE_TLS)
+# define DISPATCH_FUNCTION_SIZE  16
+#elif defined(USE_X86_ASM)
+# if defined(THREADS) && !defined(GLX_USE_TLS)
+#  define DISPATCH_FUNCTION_SIZE  32
+# else
+#  define DISPATCH_FUNCTION_SIZE  16
+# endif
+#endif
+
+
+#endif
diff --git a/src/mesa/main/version.h b/src/mesa/main/version.h
index d521569f8d3..59f62ebd6c5 100644
--- a/src/mesa/main/version.h
+++ b/src/mesa/main/version.h
@@ -1,6 +1,6 @@
 /*
  * Mesa 3-D graphics library
- * Version:  7.8
+ * Version:  7.9
  *
  * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
  * Copyright (C) 2009  VMware, Inc.  All Rights Reserved.
@@ -33,9 +33,9 @@
 
 /* Mesa version */
 #define MESA_MAJOR 7
-#define MESA_MINOR 8
+#define MESA_MINOR 9
 #define MESA_PATCH 0
-#define MESA_VERSION_STRING "7.8-devel"
+#define MESA_VERSION_STRING "7.9-devel"
 
 /* To make version comparison easy */
 #define MESA_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
diff --git a/src/mesa/sources.mak b/src/mesa/sources.mak
index 9f2e4e51575..74885548e5a 100644
--- a/src/mesa/sources.mak
+++ b/src/mesa/sources.mak
@@ -88,6 +88,7 @@ MAIN_SOURCES = \
 GLAPI_SOURCES = \
 	glapi/glapi.c \
 	glapi/glapi_dispatch.c \
+	glapi/glapi_entrypoint.c \
 	glapi/glapi_getproc.c \
 	glapi/glapi_nop.c \
 	glapi/glthread.c
diff --git a/src/mesa/state_tracker/st_cb_bitmap.c b/src/mesa/state_tracker/st_cb_bitmap.c
index 0332d4dbdfe..f326601c3be 100644
--- a/src/mesa/state_tracker/st_cb_bitmap.c
+++ b/src/mesa/state_tracker/st_cb_bitmap.c
@@ -440,6 +440,7 @@ draw_bitmap_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    cso_save_viewport(cso);
    cso_save_fragment_shader(cso);
    cso_save_vertex_shader(cso);
+   cso_save_vertex_elements(cso);
 
    /* rasterizer state: just scissor */
    st->bitmap.rasterizer.scissor = ctx->Scissor.Enabled;
@@ -490,6 +491,8 @@ draw_bitmap_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
       cso_set_viewport(cso, &vp);
    }
 
+   cso_set_vertex_elements(cso, 3, st->velems_util_draw);
+
    /* convert Z from [0,1] to [-1,-1] to match viewport Z scale/bias */
    z = z * 2.0 - 1.0;
 
@@ -509,6 +512,7 @@ draw_bitmap_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    cso_restore_viewport(cso);
    cso_restore_fragment_shader(cso);
    cso_restore_vertex_shader(cso);
+   cso_restore_vertex_elements(cso);
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_clear.c b/src/mesa/state_tracker/st_cb_clear.c
index 9e66eed3634..de86062fc40 100644
--- a/src/mesa/state_tracker/st_cb_clear.c
+++ b/src/mesa/state_tracker/st_cb_clear.c
@@ -213,6 +213,7 @@ clear_with_quad(GLcontext *ctx,
    cso_save_clip(st->cso_context);
    cso_save_fragment_shader(st->cso_context);
    cso_save_vertex_shader(st->cso_context);
+   cso_save_vertex_elements(st->cso_context);
 
    /* blend state: RGBA masking */
    {
@@ -264,6 +265,8 @@ clear_with_quad(GLcontext *ctx,
       cso_set_depth_stencil_alpha(st->cso_context, &depth_stencil);
    }
 
+   cso_set_vertex_elements(st->cso_context, 2, st->velems_util_draw);
+
    cso_set_rasterizer(st->cso_context, &st->clear.raster);
 
    /* viewport state: viewport matching window dims */
@@ -297,6 +300,8 @@ clear_with_quad(GLcontext *ctx,
    cso_restore_clip(st->cso_context);
    cso_restore_fragment_shader(st->cso_context);
    cso_restore_vertex_shader(st->cso_context);
+   cso_restore_vertex_elements(st->cso_context);
+
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index 98453321f86..c609435a158 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -529,6 +529,7 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    cso_save_sampler_textures(cso);
    cso_save_fragment_shader(cso);
    cso_save_vertex_shader(cso);
+   cso_save_vertex_elements(cso);
 
    /* rasterizer state: just scissor */
    {
@@ -581,6 +582,8 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
       cso_set_viewport(cso, &vp);
    }
 
+   cso_set_vertex_elements(cso, 3, st->velems_util_draw);
+
    /* texture state: */
    if (st->pixel_xfer.pixelmap_enabled) {
       struct pipe_texture *textures[2];
@@ -615,6 +618,7 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    cso_restore_sampler_textures(cso);
    cso_restore_fragment_shader(cso);
    cso_restore_vertex_shader(cso);
+   cso_restore_vertex_elements(cso);
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_flush.c b/src/mesa/state_tracker/st_cb_flush.c
index 1329f807bc9..28a384ba49b 100644
--- a/src/mesa/state_tracker/st_cb_flush.c
+++ b/src/mesa/state_tracker/st_cb_flush.c
@@ -79,7 +79,7 @@ display_front_buffer(struct st_context *st)
       /* Hook for copying "fake" frontbuffer if necessary:
        */
       st->pipe->screen->flush_frontbuffer( st->pipe->screen, front_surf,
-                                           st->pipe->priv );
+                                           st->winsys_drawable_handle );
 
       /*
         st->frontbuffer_status = FRONT_STATUS_UNDEFINED;
diff --git a/src/mesa/state_tracker/st_context.c b/src/mesa/state_tracker/st_context.c
index de8beaf5e25..ca6d4dfb069 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -141,6 +141,14 @@ st_create_context_priv( GLcontext *ctx, struct pipe_context *pipe )
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
       st->state.sampler_list[i] = &st->state.samplers[i];
 
+   for (i = 0; i < 3; i++) {
+      memset(&st->velems_util_draw[i], 0, sizeof(struct pipe_vertex_element));
+      st->velems_util_draw[i].src_offset = i * 4 * sizeof(float);
+      st->velems_util_draw[i].instance_divisor = 0;
+      st->velems_util_draw[i].vertex_buffer_index = 0;
+      st->velems_util_draw[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+   }
+
    /* we want all vertex data to be placed in buffer objects */
    vbo_use_buffer_objects(ctx);
 
@@ -264,7 +272,8 @@ void st_destroy_context( struct st_context *st )
 GLboolean
 st_make_current(struct st_context *st,
                 struct st_framebuffer *draw,
-                struct st_framebuffer *read)
+                struct st_framebuffer *read,
+                void *winsys_drawable_handle )
 {
    /* Call this periodically to detect when the user has begun using
     * GL rendering from multiple threads.
@@ -272,10 +281,13 @@ st_make_current(struct st_context *st,
    _glapi_check_multithread();
 
    if (st) {
-      if (!_mesa_make_current(st->ctx, &draw->Base, &read->Base))
+      if (!_mesa_make_current(st->ctx, &draw->Base, &read->Base)) {
+         st->pipe->priv = NULL;
          return GL_FALSE;
+      }
 
       _mesa_check_init_viewport(st->ctx, draw->InitWidth, draw->InitHeight);
+      st->winsys_drawable_handle = winsys_drawable_handle;
 
       return GL_TRUE;
    }
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 045c029c305..e2d34fb3d10 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -174,6 +174,9 @@ struct st_context
       unsigned vbuf_slot;
    } clear;
 
+   /** used for anything using util_draw_vertex_buffer */
+   struct pipe_vertex_element velems_util_draw[3];
+
    void *passthrough_fs;  /**< simple pass-through frag shader */
 
    struct gen_mipmap_state *gen_mipmap;
@@ -182,6 +185,7 @@ struct st_context
    struct cso_context *cso_context;
 
    int force_msaa;
+   void *winsys_drawable_handle;
 };
 
 
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 32b9a473cfc..8a6e1ed4662 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -57,6 +57,7 @@
 #include "pipe/p_defines.h"
 #include "util/u_inlines.h"
 #include "util/u_format.h"
+#include "cso_cache/cso_context.h"
 
 
 static GLuint double_types[4] = {
@@ -368,7 +369,6 @@ setup_interleaved_attribs(GLcontext *ctx,
          (unsigned) (arrays[mesaAttr]->Ptr - offset0);
       velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = 0;
-      velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format =
          st_pipe_vertex_format(arrays[mesaAttr]->Type,
                                arrays[mesaAttr]->Size,
@@ -458,7 +458,6 @@ setup_non_interleaved_attribs(GLcontext *ctx,
       vbuffer[attr].max_index = max_index;
       velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = attr;
-      velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format
          = st_pipe_vertex_format(arrays[mesaAttr]->Type,
                                  arrays[mesaAttr]->Size,
@@ -564,6 +563,7 @@ st_draw_vbo(GLcontext *ctx,
    (void) check_uniforms;
 #endif
 
+   memset(velements, 0, sizeof(struct pipe_vertex_element) * vpv->num_inputs);
    /*
     * Setup the vbuffer[] and velements[] arrays.
     */
@@ -596,14 +596,13 @@ st_draw_vbo(GLcontext *ctx,
       for (i = 0; i < num_velements; i++) {
          printf("vlements[%d].vbuffer_index = %u\n", i, velements[i].vertex_buffer_index);
          printf("vlements[%d].src_offset = %u\n", i, velements[i].src_offset);
-         printf("vlements[%d].nr_comps = %u\n", i, velements[i].nr_components);
          printf("vlements[%d].format = %s\n", i, util_format_name(velements[i].src_format));
       }
    }
 #endif
 
    pipe->set_vertex_buffers(pipe, num_vbuffers, vbuffer);
-   pipe->set_vertex_elements(pipe, num_velements, velements);
+   cso_set_vertex_elements(ctx->st->cso_context, num_velements, velements);
 
    if (num_vbuffers == 0 || num_velements == 0)
       return;
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index 087f2f22bbf..26a5b3fcd63 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -178,7 +178,6 @@ st_feedback_draw_vbo(GLcontext *ctx,
       vbuffers[attr].max_index = max_index;
       velements[attr].instance_divisor = 0;
       velements[attr].vertex_buffer_index = attr;
-      velements[attr].nr_components = arrays[mesaAttr]->Size;
       velements[attr].src_format = 
          st_pipe_vertex_format(arrays[mesaAttr]->Type,
                                arrays[mesaAttr]->Size,
diff --git a/src/mesa/state_tracker/st_extensions.c b/src/mesa/state_tracker/st_extensions.c
index 79be833768f..e03dd30f0f5 100644
--- a/src/mesa/state_tracker/st_extensions.c
+++ b/src/mesa/state_tracker/st_extensions.c
@@ -137,6 +137,9 @@ void st_init_limits(struct st_context *st)
    /* XXX separate query for early function return? */
    st->ctx->Shader.EmitContReturn =
       screen->get_param(screen, PIPE_CAP_TGSI_CONT_SUPPORTED);
+
+   /* Quads always follow GL provoking rules. */
+   c->QuadsFollowProvokingVertexConvention = GL_FALSE;
 }
 
 
diff --git a/src/mesa/state_tracker/st_public.h b/src/mesa/state_tracker/st_public.h
index 0824356cecc..4b40d6d0448 100644
--- a/src/mesa/state_tracker/st_public.h
+++ b/src/mesa/state_tracker/st_public.h
@@ -105,7 +105,8 @@ void st_unreference_framebuffer( struct st_framebuffer *stfb );
 PUBLIC
 GLboolean st_make_current(struct st_context *st,
                           struct st_framebuffer *draw,
-                          struct st_framebuffer *read);
+                          struct st_framebuffer *read,
+                          void *winsys_drawable_handle);
 
 PUBLIC
 struct st_context *st_get_current(void);