Merge branch 'upstream-gallium-0.1' into nouveau-gallium-0.1

Conflicts:

	src/gallium/drivers/Makefile

190 files changed, 7463 insertions, 2044 deletions

diff --git a/Makefile b/Makefile
index 6e92858ad5..1ec8ce7cba 100644
--- a/Makefile
+++ b/Makefile
@@ -12,6 +12,8 @@ default: $(TOP)/configs/current
 		fi \
 	done
 
+all: default
+
 
 doxygen:
 	cd doxygen && $(MAKE)
diff --git a/SConstruct b/SConstruct
index 22a4072c93..0420195395 100644
--- a/SConstruct
+++ b/SConstruct
@@ -28,12 +28,22 @@ import sys
 # to get the full list of options. See scons manpage for more info.
 #  
 
+platform_map = {
+	'linux2': 'linux',
+	'win32': 'winddk',
+}
+
+platform = platform_map.get(sys.platform, sys.platform)
+
 # TODO: auto-detect defaults
 opts = Options('config.py')
 opts.Add(BoolOption('debug', 'build debug version', False))
 opts.Add(BoolOption('dri', 'build dri drivers', False))
+opts.Add(BoolOption('llvm', 'use llvm', False))
 opts.Add(EnumOption('machine', 'use machine-specific assembly code', 'x86',
                      allowed_values=('generic', 'x86', 'x86-64')))
+opts.Add(EnumOption('platform', 'target platform', platform,
+                     allowed_values=('linux', 'cell', 'winddk')))
 
 env = Environment(
 	options = opts, 
@@ -43,29 +53,23 @@ Help(opts.GenerateHelpText(env))
 # for debugging
 #print env.Dump()
 
-if 0:
-	# platform will be typically 'posix' or 'win32' 
-	platform = env['PLATFORM']
-else:
-	# platform will be one of 'linux', 'freebsd', 'win32', 'darwin', etc.
-	platform = sys.platform
-	if platform == 'linux2':
-		platform = 'linux' 
-
 # replicate options values in local variables
 debug = env['debug']
 dri = env['dri']
+llvm = env['llvm']
 machine = env['machine']
+platform = env['platform']
 
 # derived options
 x86 = machine == 'x86'
-gcc = platform in ('posix', 'linux', 'freebsd', 'darwin')
-msvc = platform == 'win32'
+gcc = platform in ('linux', 'freebsd', 'darwin')
+msvc = platform in ('win32', 'winddk')
 
 Export([
 	'debug', 
 	'x86', 
 	'dri', 
+	'llvm',
 	'platform',
 	'gcc',
 	'msvc',
@@ -75,10 +79,32 @@ Export([
 #######################################################################
 # Environment setup
 #
-# TODO: put the compiler specific settings in seperate files
+# TODO: put the compiler specific settings in separate files
 # TODO: auto-detect as much as possible
 
-         
+
+if platform == 'winddk':
+	import ntpath
+	escape = env['ESCAPE']
+	env.Tool('msvc')
+	if 'BASEDIR' in os.environ:
+		WINDDK = os.environ['BASEDIR']
+	else:
+		WINDDK = "C:\\WINDDK\\3790.1830"
+	# NOTE: We need this elaborate construct to get the absolute paths and
+	# forward slashes to msvc unharmed when cross compiling from posix platforms 
+	env.Append(CPPFLAGS = [
+		escape('/I' + ntpath.join(WINDDK, 'inc\\ddk\\wxp')),
+		escape('/I' + ntpath.join(WINDDK, 'inc\\ddk\\wdm\\wxp')),
+		escape('/I' + ntpath.join(WINDDK, 'inc\\crt')),
+	])
+	env.Append(CPPDEFINES = [
+		('i386', '1'),
+	])
+	if debug:
+		env.Append(CPPDEFINES = ['DBG'])
+	
+
 # Optimization flags
 if gcc:
 	if debug:
@@ -106,9 +132,9 @@ else:
 # Includes
 env.Append(CPPPATH = [
 	'#/include',
-	'#/src/mesa',
-	'#/src/mesa/main',
-	'#/src/mesa/pipe',
+	'#/src/gallium/include',
+	'#/src/gallium/auxiliary',
+	'#/src/gallium/drivers',
 ])
 
 
@@ -157,6 +183,14 @@ if dri:
 		'GLX_INDIRECT_RENDERING',
 	])
 
+# LLVM
+if llvm:
+	# See also http://www.scons.org/wiki/UsingPkgConfig
+	env.ParseConfig('llvm-config --cflags --ldflags --libs')
+	env.Append(CPPDEFINES = ['MESA_LLVM'])
+	env.Append(CXXFLAGS = ['-Wno-long-long'])
+	
+
 # libGL
 if 1:
 	env.Append(LIBS = [
@@ -212,6 +246,8 @@ build_topdir = 'build'
 build_subdir = platform
 if dri:
 	build_subdir += "-dri"
+if llvm:
+	build_subdir += "-llvm"
 if x86:
 	build_subdir += "-x86"
 if debug:
@@ -222,7 +258,7 @@ build_dir = os.path.join(build_topdir, build_subdir)
 # http://www.scons.org/wiki/SimultaneousVariantBuilds
 
 SConscript(
-	'src/mesa/SConscript',
+	'src/SConscript',
 	build_dir = build_dir,
 	duplicate = 0 # http://www.scons.org/doc/0.97/HTML/scons-user/x2261.html
 )
diff --git a/configs/beos b/configs/beos
index 2b74af739d..c6e972789a 100644
--- a/configs/beos
+++ b/configs/beos
@@ -26,8 +26,8 @@ ifeq ($(CPU), x86)
 		-DUSE_3DNOW_ASM \
 		-DUSE_SSE_ASM
 	
-	ASM_SOURCES = $(X86_SOURCES)
-	ASM_API = $(X86_API)
+	MESA_ASM_SOURCES = $(X86_SOURCES)
+	GLAPI_ASM_SOURCES = $(X86_API)
 
 	CC = gcc
 	CXX = g++
diff --git a/configs/default b/configs/default
index 25a87e66a1..48ddd29282 100644
--- a/configs/default
+++ b/configs/default
@@ -51,7 +51,7 @@ OSMESA_LIB_NAME = lib$(OSMESA_LIB).so
 
 
 # Optional assembly language optimization files for libGL
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # GLw widget sources (Append "GLwMDrawA.c" here and add -lXm to GLW_LIB_DEPS in
 # order to build the Motif widget too)
@@ -60,13 +60,21 @@ GLW_SOURCES = GLwDrawA.c
 
 # Directories to build
 LIB_DIR = lib
-SRC_DIRS = gallium mesa glu glut/glx glw
+SRC_DIRS = gallium mesa gallium/winsys glu glut/glx glw
 GLU_DIRS = sgi
-DRIVER_DIRS = x11 osmesa
+DRIVER_DIRS = 
 # Which subdirs under $(TOP)/progs/ to enter:
 PROGRAM_DIRS = demos redbook samples glsl xdemos
 
 
+# Gallium directories and 
+GALLIUM_AUXILIARY_DIRS = draw cso_cache pipebuffer tgsi rtasm util
+GALLIUM_AUXILIARIES = $(foreach DIR,$(GALLIUM_AUXILIARY_DIRS),$(TOP)/src/gallium/auxiliary/$(DIR)/lib$(DIR).a)
+GALLIUM_DRIVER_DIRS = softpipe i915simple i965simple failover
+GALLIUM_DRIVERS = $(foreach DIR,$(GALLIUM_DRIVER_DIRS),$(TOP)/src/gallium/drivers/$(DIR)/lib$(DIR).a)
+GALLIUM_WINSYS_DIRS = xlib
+
+
 # Library/program dependencies
 #EXTRA_LIB_PATH ?=
 GL_LIB_DEPS     = $(EXTRA_LIB_PATH) -lX11 -lXext -lm -lpthread
diff --git a/configs/freebsd-dri b/configs/freebsd-dri
index 67d253b869..6fc1abbc80 100644
--- a/configs/freebsd-dri
+++ b/configs/freebsd-dri
@@ -22,7 +22,7 @@ CFLAGS = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) -Wmissing-prototypes -std=c99 -
 
 CXXFLAGS = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(DEFINES) -Wall -ansi -pedantic $(ASM_FLAGS) $(X11_INCLUDES) 
 
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # Library/program dependencies
 LIBDRM_CFLAGS = `pkg-config --cflags libdrm`
diff --git a/configs/freebsd-dri-amd64 b/configs/freebsd-dri-amd64
index 39341b9701..bb6c361398 100644
--- a/configs/freebsd-dri-amd64
+++ b/configs/freebsd-dri-amd64
@@ -6,5 +6,5 @@ include $(TOP)/configs/freebsd-dri
 CONFIG_NAME = freebsd-dri-x86-64
 
 ASM_FLAGS = -DUSE_X86_64_ASM
-ASM_SOURCES = $(X86-64_SOURCES)
-ASM_API = $(X86-64_API)
+MESA_ASM_SOURCES = $(X86-64_SOURCES)
+GLAPI_ASM_SOURCES = $(X86-64_API)
diff --git a/configs/freebsd-dri-x86 b/configs/freebsd-dri-x86
index af0d27ff47..9475437fc5 100644
--- a/configs/freebsd-dri-x86
+++ b/configs/freebsd-dri-x86
@@ -9,5 +9,5 @@ CONFIG_NAME = freebsd-dri-x86
 PIC_FLAGS = 
 
 ASM_FLAGS = -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
diff --git a/configs/linux-cell b/configs/linux-cell
index 09f62fc1ff..bcf5f00acb 100644
--- a/configs/linux-cell
+++ b/configs/linux-cell
@@ -5,6 +5,9 @@ include $(TOP)/configs/default
 CONFIG_NAME = linux-cell
 
 
+GALLIUM_DRIVER_DIRS += cell
+
+
 # Compiler and flags
 CC = ppu32-gcc
 CXX = ppu32-g++
@@ -21,7 +24,7 @@ CFLAGS = $(OPT_FLAGS) -Wall -Winline -fPIC -m32 -mabi=altivec -maltivec -I. -I$(
 CXXFLAGS = $(CFLAGS)
 
 # Omitting glw here:
-SRC_DIRS = gallium mesa glu glut/glx
+SRC_DIRS = gallium mesa gallium/winsys glu glut/glx
 
 
 MKDEP_OPTIONS = -fdepend -Y
diff --git a/configs/linux-directfb b/configs/linux-directfb
index dff27f7850..2ed94fe275 100644
--- a/configs/linux-directfb
+++ b/configs/linux-directfb
@@ -17,8 +17,8 @@ HAVE_X86 = $(shell uname -m | grep 'i[3-6]86' >/dev/null && echo yes)
 ifeq ($(HAVE_X86), yes)
      CFLAGS   += -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
      CXXFLAGS += -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
-     ASM_SOURCES = $(X86_SOURCES)
-     ASM_API = $(X86_API)
+     MESA_ASM_SOURCES = $(X86_SOURCES)
+     GLAPI_ASM_SOURCES = $(X86_API)
 endif
 
 # Directories
diff --git a/configs/linux-dri b/configs/linux-dri
index e6135856fc..67e60cbd4c 100644
--- a/configs/linux-dri
+++ b/configs/linux-dri
@@ -33,7 +33,7 @@ CFLAGS = -Wall -Wmissing-prototypes -std=c99 -ffast-math \
 CXXFLAGS = -Wall $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES)
 
 
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # Library/program dependencies
 EXTRA_LIB_PATH=-L/usr/X11R6/lib
@@ -53,16 +53,15 @@ GL_LIB_DEPS   = $(EXTRA_LIB_PATH) -lX11 -lXext -lXxf86vm -lXdamage -lXfixes \
 USING_EGL=0
 
 # Directories
+SRC_DIRS := glx/x11 $(SRC_DIRS)
 ifeq ($(USING_EGL), 1)
-SRC_DIRS = egl glx/x11 gallium mesa glu glut/glx glw
+SRC_DIRS := egl $(SRC_DIRS)
 PROGRAM_DIRS = egl
-else
-SRC_DIRS = glx/x11 gallium mesa glu glut/glx glw
-PROGRAM_DIRS =
 endif
 
 DRIVER_DIRS = dri
 WINDOW_SYSTEM=dri
+GALLIUM_WINSYS_DIRS = dri
 
 # gamma are missing because they have not been converted to use the new
 # interface.
diff --git a/configs/linux-dri-ppc b/configs/linux-dri-ppc
index fb87688065..a3a3ca83cb 100644
--- a/configs/linux-dri-ppc
+++ b/configs/linux-dri-ppc
@@ -9,7 +9,7 @@ OPT_FLAGS = -Os -mcpu=603
 PIC_FLAGS = -fPIC
 
 ASM_FLAGS = -DUSE_PPC_ASM -DUSE_VMX_ASM
-ASM_SOURCES = $(PPC_SOURCES)
+MESA_ASM_SOURCES = $(PPC_SOURCES)
 
 # Build only the drivers for cards that exist on PowerPC.  At some point MGA
 # will be added, but not yet.
diff --git a/configs/linux-dri-x86 b/configs/linux-dri-x86
index b196004e58..ec8242dd68 100644
--- a/configs/linux-dri-x86
+++ b/configs/linux-dri-x86
@@ -12,6 +12,6 @@ PIC_FLAGS =
 ARCH_FLAGS = -m32
 
 ASM_FLAGS = -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
 
diff --git a/configs/linux-dri-x86-64 b/configs/linux-dri-x86-64
index 821ab3e336..bb56de375a 100644
--- a/configs/linux-dri-x86-64
+++ b/configs/linux-dri-x86-64
@@ -8,8 +8,8 @@ CONFIG_NAME = linux-dri-x86-64
 ARCH_FLAGS = -m64
 
 ASM_FLAGS = -DUSE_X86_64_ASM
-ASM_SOURCES = $(X86-64_SOURCES)
-ASM_API = $(X86-64_API)
+MESA_ASM_SOURCES = $(X86-64_SOURCES)
+GLAPI_ASM_SOURCES = $(X86-64_API)
 
 LIB_DIR = lib64
 
diff --git a/configs/linux-dri-xcb b/configs/linux-dri-xcb
index ea4bdf1864..fbf9b9b268 100644
--- a/configs/linux-dri-xcb
+++ b/configs/linux-dri-xcb
@@ -33,7 +33,7 @@ CFLAGS = -Wall -Wmissing-prototypes $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) \
 CXXFLAGS = -Wall $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES)
 
 
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # Library/program dependencies
 EXTRA_LIB_PATH=$(shell pkg-config --libs-only-L x11)
diff --git a/configs/linux-icc b/configs/linux-icc
index 978a45af70..d90a1dab3d 100644
--- a/configs/linux-icc
+++ b/configs/linux-icc
@@ -16,7 +16,7 @@ GL_LIB_DEPS = -L/usr/X11R6/lib -lX11 -lXext -lm -lpthread
 GLUT_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -l$(GLU_LIB) -l$(GL_LIB) -L/usr/X11R6/lib -lX11 -lXmu -lXt -lXi -lm
 APP_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -l$(GLUT_LIB) -l$(GLU_LIB) -l$(GL_LIB) -lm
 
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
 
 
diff --git a/configs/linux-icc-static b/configs/linux-icc-static
index 0c957568c2..384db3bfe4 100644
--- a/configs/linux-icc-static
+++ b/configs/linux-icc-static
@@ -23,5 +23,5 @@ GL_LIB_DEPS =
 GLUT_LIB_DEPS =
 APP_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -l$(GLUT_LIB) -l$(GLU_LIB) -l$(GL_LIB) -L/usr/X11R6/lib -lX11 -lXmu -lXt -lXi -lm -lpthread -lcxa -lunwind
 
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
diff --git a/configs/linux-indirect b/configs/linux-indirect
index bd33345ed7..0c4805ea87 100644
--- a/configs/linux-indirect
+++ b/configs/linux-indirect
@@ -34,7 +34,7 @@ CFLAGS   = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES) \
 CXXFLAGS = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES)
 
 
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # Library/program dependencies
 EXTRA_LIB_PATH=-L/usr/X11R6/lib
diff --git a/configs/linux-llvm b/configs/linux-llvm
index 915189f462..44e200e856 100644
--- a/configs/linux-llvm
+++ b/configs/linux-llvm
@@ -5,6 +5,8 @@ include $(TOP)/configs/linux
 
 CONFIG_NAME = linux-llvm
 
+GALLIUM_AUXILIARY_DIRS += gallivm
+
 OPT_FLAGS = -g -ansi -pedantic
 DEFINES += -DDEBUG -DDEBUG_MATH -DMESA_LLVM=1
 
diff --git a/configs/linux-solo b/configs/linux-solo
index d49b972228..3145e12775 100644
--- a/configs/linux-solo
+++ b/configs/linux-solo
@@ -33,7 +33,7 @@ CFLAGS   = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES) \
 CXXFLAGS = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES)
 
 
-ASM_SOURCES = 
+MESA_ASM_SOURCES = 
 
 # Library/program dependencies
 DRI_LIB_DEPS = -lm -lpthread -lexpat -ldl -L$(TOP)/$(LIB_DIR) $(PCIACCESS_LIB)
diff --git a/configs/linux-solo-x86 b/configs/linux-solo-x86
index 13cab37658..5f5aa09c82 100644
--- a/configs/linux-solo-x86
+++ b/configs/linux-solo-x86
@@ -9,5 +9,5 @@ CONFIG_NAME = linux-solo-x86
 PIC_FLAGS = 
 
 ASM_FLAGS = -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
diff --git a/configs/linux-sparc b/configs/linux-sparc
index 9925afc19b..346d438c28 100644
--- a/configs/linux-sparc
+++ b/configs/linux-sparc
@@ -5,5 +5,5 @@ include $(TOP)/configs/linux
 CONFIG_NAME = linux-sparc
 
 #ASM_FLAGS = -DUSE_SPARC_ASM
-#ASM_SOURCES = $(SPARC_SOURCES)
-#ASM_API = $(SPARC_API)
+#MESA_ASM_SOURCES = $(SPARC_SOURCES)
+#GLAPI_ASM_SOURCES = $(SPARC_API)
diff --git a/configs/linux-x86 b/configs/linux-x86
index 18fa06101d..a4cf4e8d62 100644
--- a/configs/linux-x86
+++ b/configs/linux-x86
@@ -5,5 +5,5 @@ include $(TOP)/configs/linux
 CONFIG_NAME = linux-x86
 
 ASM_FLAGS = -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
diff --git a/configs/linux-x86-64 b/configs/linux-x86-64
index 67c0391836..c2441e09d0 100644
--- a/configs/linux-x86-64
+++ b/configs/linux-x86-64
@@ -6,8 +6,8 @@ CONFIG_NAME = linux-x86-64
 
 ARCH_FLAGS = -m64
 
-ASM_SOURCES = $(X86-64_SOURCES)
-ASM_API = $(X86-64_API)
+MESA_ASM_SOURCES = $(X86-64_SOURCES)
+GLAPI_ASM_SOURCES = $(X86-64_API)
 ASM_FLAGS = -DUSE_X86_64_ASM
 
 LIB_DIR = lib64
diff --git a/configs/linux-x86-glide b/configs/linux-x86-glide
index f2f8aeea60..b963fbdc66 100644
--- a/configs/linux-x86-glide
+++ b/configs/linux-x86-glide
@@ -15,8 +15,8 @@ CXXFLAGS = -Wall -O3 -ansi -pedantic -fPIC -D_POSIX_SOURCE -D_POSIX_C_SOURCE=199
 GLUT_CFLAGS = -fexceptions
 
 
-ASM_SOURCES = $(X86_SOURCES)
-ASM_API = $(X86_API)
+MESA_ASM_SOURCES = $(X86_SOURCES)
+GLAPI_ASM_SOURCES = $(X86_API)
 
 # Library/program dependencies
 GL_LIB_DEPS = -L/usr/X11R6/lib -lX11 -lXext -L/usr/local/glide/lib -lglide3x -lm -lpthread
diff --git a/configs/sunos5-gcc b/configs/sunos5-gcc
index 77b293c545..3fa13d0496 100644
--- a/configs/sunos5-gcc
+++ b/configs/sunos5-gcc
@@ -16,8 +16,8 @@ ARCH_FLAGS ?=
 
 DEFINES = -D_REENTRANT -DUSE_XSHM
 
-ASM_SOURCES = $(SPARC_SOURCES)
-ASM_API = $(SPARC_API)
+MESA_ASM_SOURCES = $(SPARC_SOURCES)
+GLAPI_ASM_SOURCES = $(SPARC_API)
 ASM_FLAGS = -DUSE_SPARC_ASM
 
 CFLAGS   = $(WARN_FLAGS) $(OPT_FLAGS) $(PIC_FLAGS) $(ARCH_FLAGS) $(DEFINES) \
diff --git a/doxygen/gallium.doxy b/doxygen/gallium.doxy
index fbcec03ea6..3ef8c6c297 100644
--- a/doxygen/gallium.doxy
+++ b/doxygen/gallium.doxy
@@ -113,7 +113,7 @@ FULL_PATH_NAMES        = YES
 # If left blank the directory from which doxygen is run is used as the 
 # path to strip.
 
-STRIP_FROM_PATH        = ../src/mesa
+STRIP_FROM_PATH        = ../src
 
 # The STRIP_FROM_INC_PATH tag can be used to strip a user-defined part of 
 # the path mentioned in the documentation of a class, which tells 
@@ -478,9 +478,9 @@ WARN_LOGFILE           =
 # with spaces.
 
 INPUT                  = gallium.doc \
-                         ../src/mesa/pipe \
+                         ../src/gallium \
                          ../src/mesa/state_tracker \
-                         ../src/mesa/drivers/dri/intel_winsys
+                         ../src/mesa/x86/rtasm
 
 # This tag can be used to specify the character encoding of the source files that 
 # doxygen parses. Internally doxygen uses the UTF-8 encoding, which is also the default 
diff --git a/progs/tests/Makefile b/progs/tests/Makefile
index 00e58a53ce..8196c1b348 100644
--- a/progs/tests/Makefile
+++ b/progs/tests/Makefile
@@ -62,6 +62,7 @@ SOURCES = \
 	sharedtex.c \
 	stencilwrap.c \
 	stencil_wrap.c \
+	subtex \
 	subtexrate.c \
 	tex1d.c \
 	texcompress2.c \
diff --git a/progs/tests/subtex.c b/progs/tests/subtex.c
new file mode 100644
index 0000000000..81ceb085aa
--- /dev/null
+++ b/progs/tests/subtex.c
@@ -0,0 +1,223 @@
+/*
+ * Test glTexSubImage mid-way through a frame.
+ *
+ * The same texture is used for both quads but it gets redefined
+ * with glTexSubImage (or glTexImage) after the first quad.
+ */
+
+
+#include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include "GL/glut.h"
+
+static GLuint Window = 0;
+static GLboolean Anim = GL_FALSE;
+static GLfloat Angle = 0.0f;
+
+
+
+static void
+first_texture(void)
+{
+   static int width=8, height=8;
+   static GLubyte tex1[] = {
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 0, 1, 0, 0, 0,
+     0, 0, 0, 1, 1, 0, 0, 0,
+     0, 0, 0, 0, 1, 0, 0, 0,
+     0, 0, 0, 0, 1, 0, 0, 0,
+     0, 0, 0, 0, 1, 0, 0, 0,
+     0, 0, 0, 1, 1, 1, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0 };
+
+   GLubyte tex[64][3];
+   GLint i, j;
+
+   /* red on white */
+   for (i=0;i<height;i++) {
+      for (j=0;j<width;j++) {
+         int p = i*width+j;
+         if (tex1[(height-i-1)*width+j]) {
+            tex[p][0] = 255;   tex[p][1] = 0;     tex[p][2] = 0;
+         }
+         else {
+            tex[p][0] = 255;   tex[p][1] = 255;   tex[p][2] = 255;
+         }
+      }
+   }
+
+   glTexImage2D( GL_TEXTURE_2D, 0, 3, width, height, 0,
+                 GL_RGB, GL_UNSIGNED_BYTE, tex );
+}
+
+
+static void
+second_texture(void)
+{
+   static int width=8, height=8;
+
+   static GLubyte tex2[] = {
+     0, 0, 0, 0, 0, 0, 0, 0,
+     0, 0, 0, 2, 2, 0, 0, 0,
+     0, 0, 2, 0, 0, 2, 0, 0,
+     0, 0, 0, 0, 0, 2, 0, 0,
+     0, 0, 0, 0, 2, 0, 0, 0,
+     0, 0, 0, 2, 0, 0, 0, 0,
+     0, 0, 2, 2, 2, 2, 0, 0,
+     0, 0, 0, 0, 0, 0, 0, 0 };
+
+   GLubyte tex[64][3];
+   GLint i, j;
+
+   /* green on blue */
+   for (i=0;i<height;i++) {
+      for (j=0;j<width;j++) {
+         int p = i*width+j;
+         if (tex2[(height-i-1)*width+j]) {
+            tex[p][0] = 0;     tex[p][1] = 255;   tex[p][2] = 0;
+         }
+         else {
+            tex[p][0] = 0;     tex[p][1] = 0;     tex[p][2] = 255;
+         }
+      }
+   }
+#if 0
+   glTexImage2D( GL_TEXTURE_2D, 0, 3, width, height, 0,
+                 GL_RGB, GL_UNSIGNED_BYTE, tex );
+#else
+   glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height,
+                 GL_RGB, GL_UNSIGNED_BYTE, tex );
+#endif
+}
+
+
+
+static void draw( void )
+{
+   glClear( GL_COLOR_BUFFER_BIT );
+
+   glColor3f( 1.0, 1.0, 1.0 );
+
+   /* draw first polygon */
+   glPushMatrix();
+   glTranslatef( -1.0, 0.0, 0.0 );
+   glRotatef( Angle, 0.0, 0.0, 1.0 );
+
+   first_texture();
+
+   glBegin( GL_POLYGON );
+   glTexCoord2f( 0.0, 0.0 );   glVertex2f( -1.0, -1.0 );
+   glTexCoord2f( 1.0, 0.0 );   glVertex2f(  1.0, -1.0 );
+   glTexCoord2f( 1.0, 1.0 );   glVertex2f(  1.0,  1.0 );
+   glTexCoord2f( 0.0, 1.0 );   glVertex2f( -1.0,  1.0 );
+   glEnd();
+   glPopMatrix();
+
+   /* draw second polygon */
+   glPushMatrix();
+   glTranslatef( 1.0, 0.0, 0.0 );
+   glRotatef( Angle-90.0, 0.0, 1.0, 0.0 );
+
+   second_texture();
+
+   glBegin( GL_POLYGON );
+   glTexCoord2f( 0.0, 0.0 );   glVertex2f( -1.0, -1.0 );
+   glTexCoord2f( 1.0, 0.0 );   glVertex2f(  1.0, -1.0 );
+   glTexCoord2f( 1.0, 1.0 );   glVertex2f(  1.0,  1.0 );
+   glTexCoord2f( 0.0, 1.0 );   glVertex2f( -1.0,  1.0 );
+   glEnd();
+   glPopMatrix();
+
+   glutSwapBuffers();
+}
+
+
+
+static void idle( void )
+{
+   static double t0 = -1.;
+   double dt, t = glutGet(GLUT_ELAPSED_TIME) / 1000.0;
+   if (t0 < 0.0)
+      t0 = t;
+   dt = t - t0;
+   t0 = t;
+   Angle += 120.0*dt;
+   glutPostRedisplay();
+}
+
+
+
+/* change view Angle, exit upon ESC */
+static void key(unsigned char k, int x, int y)
+{
+   (void) x;
+   (void) y;
+   switch (k) {
+   case 'a':
+      Anim = !Anim;
+      if (Anim)
+         glutIdleFunc( idle );
+      else
+         glutIdleFunc( NULL );
+      break;
+   case 27:
+      glutDestroyWindow(Window);
+      exit(0);
+   }
+}
+
+
+
+/* new window size or exposure */
+static void reshape( int width, int height )
+{
+   glViewport(0, 0, (GLint)width, (GLint)height);
+   glMatrixMode(GL_PROJECTION);
+   glLoadIdentity();
+   /*   glOrtho( -3.0, 3.0, -3.0, 3.0, -10.0, 10.0 );*/
+   glFrustum( -2.0, 2.0, -2.0, 2.0, 6.0, 20.0 );
+   glMatrixMode(GL_MODELVIEW);
+   glLoadIdentity();
+   glTranslatef( 0.0, 0.0, -8.0 );
+}
+
+
+static void init( void )
+{
+   /* Setup texturing */
+   glEnable( GL_TEXTURE_2D );
+   glTexEnvi( GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_DECAL );
+
+
+   glBindTexture( GL_TEXTURE_2D, 0 );
+   glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST );
+   glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
+   glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT );
+   glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT );
+}
+
+
+
+int main( int argc, char *argv[] )
+{
+   glutInit(&argc, argv);
+   glutInitWindowPosition(0, 0);
+   glutInitWindowSize(300, 300);
+   glutInitDisplayMode( GLUT_RGB | GLUT_DEPTH | GLUT_DOUBLE );
+
+   Window = glutCreateWindow("Texture Objects");
+   if (!Window) {
+      exit(1);
+   }
+
+   init();
+
+   glutReshapeFunc( reshape );
+   glutKeyboardFunc( key );
+   if (Anim)
+      glutIdleFunc( idle );
+   glutDisplayFunc( draw );
+   glutMainLoop();
+   return 0;
+}
diff --git a/src/SConscript b/src/SConscript
new file mode 100644
index 0000000000..5b09943894
--- /dev/null
+++ b/src/SConscript
@@ -0,0 +1,7 @@
+Import('*')
+
+SConscript([
+	'gallium/SConscript',
+	'mesa/SConscript',
+	'gallium/winsys/SConscript',
+])
diff --git a/src/gallium/Makefile b/src/gallium/Makefile
index 89e068a449..568747e157 100644
--- a/src/gallium/Makefile
+++ b/src/gallium/Makefile
@@ -18,3 +18,4 @@ subdirs:
 
 clean:
 	rm -f `find . -name \*.[oa]`
+	rm -f `find . -name depend`
diff --git a/src/gallium/Makefile.template b/src/gallium/Makefile.template
index 83b25f9b47..4e462b5c97 100644
--- a/src/gallium/Makefile.template
+++ b/src/gallium/Makefile.template
@@ -16,10 +16,8 @@ OBJECTS = $(C_SOURCES:.c=.o) \
 INCLUDES = \
 	-I. \
 	-I$(TOP)/src/gallium/include \
-	-I$(TOP)/src/gallium/include/pipe \
 	-I$(TOP)/src/gallium/auxiliary \
 	-I$(TOP)/src/gallium/drivers \
-	-I$(TOP)/src/mesa \
 	-I$(TOP)/include \
         $(DRIVER_INCLUDES)
 
diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index d9c20e0100..a08b4b830e 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -1,9 +1,29 @@
 Import('*')
 
-#env = env.Clone()
+env = env.Clone()
+
+auxiliaries = []
+
+Export('auxiliaries')
+
+
+SConscript([
+	# NOTE: order matters!
+	'auxiliary/util/SConscript',
+	'auxiliary/rtasm/SConscript',
+	'auxiliary/tgsi/SConscript',
+	'auxiliary/cso_cache/SConscript',
+	'auxiliary/draw/SConscript',
+	'auxiliary/pipebuffer/SConscript',
+])
+
+if llvm:
+	SConscript(['auxiliary/gallivm/SConscript'])
 
 SConscript([
-	'softpipe/SConscript',
-	'i915simple/SConscript',
-	'i965simple/SConscript',
+	'drivers/softpipe/SConscript',
+	'drivers/i915simple/SConscript',
+	'drivers/i965simple/SConscript',
+	'drivers/failover/SConscript',
+	#'drivers/cell/SConscript',
 ])
diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index da68498aa1..eaa0f2fe4e 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -2,11 +2,7 @@ TOP = ../../..
 include $(TOP)/configs/current
 
 
-ifeq ($(CONFIG_NAME), linux-llvm)
-LLVM_DIR = llvm
-endif
-
-SUBDIRS = pipebuffer $(LLVM_DIR)
+SUBDIRS = $(GALLIUM_AUXILIARY_DIRS)
 
 
 default: subdirs
diff --git a/src/gallium/auxiliary/cso_cache/Makefile b/src/gallium/auxiliary/cso_cache/Makefile
new file mode 100644
index 0000000000..3e49266163
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/Makefile
@@ -0,0 +1,13 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = cso_cache
+
+C_SOURCES = \
+	cso_cache.c \
+	cso_hash.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/cso_cache/SConscript b/src/gallium/auxiliary/cso_cache/SConscript
new file mode 100644
index 0000000000..9751881613
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/SConscript
@@ -0,0 +1,10 @@
+Import('*')
+
+cso_cache = env.ConvenienceLibrary(
+	target = 'cso_cache',
+	source = [
+		'cso_cache.c',
+		'cso_hash.c',
+	])
+
+auxiliaries.insert(0, cso_cache)
diff --git a/src/gallium/auxiliary/cso_cache/cso_hash.c b/src/gallium/auxiliary/cso_cache/cso_hash.c
index 0338cb3b47..b40217c524 100644
--- a/src/gallium/auxiliary/cso_cache/cso_hash.c
+++ b/src/gallium/auxiliary/cso_cache/cso_hash.c
@@ -158,11 +158,14 @@ static void cso_data_rehash(struct cso_hash_data *hash, int hint)
          while (firstNode != e) {
             unsigned h = firstNode->key;
             struct cso_node *lastNode = firstNode;
+            struct cso_node *afterLastNode;
+            struct cso_node **beforeFirstNode;
+            
             while (lastNode->next != e && lastNode->next->key == h)
                lastNode = lastNode->next;
 
-            struct cso_node *afterLastNode = lastNode->next;
-            struct cso_node **beforeFirstNode = &hash->buckets[h % hash->numBuckets];
+            afterLastNode = lastNode->next;
+            beforeFirstNode = &hash->buckets[h % hash->numBuckets];
             while (*beforeFirstNode != e)
                beforeFirstNode = &(*beforeFirstNode)->next;
             lastNode->next = *beforeFirstNode;
@@ -222,10 +225,12 @@ struct cso_hash_iter cso_hash_insert(struct cso_hash *hash,
 {
    cso_data_might_grow(hash->data.d);
 
-   struct cso_node **nextNode = cso_hash_find_node(hash, key);
-   struct cso_node *node = cso_hash_create_node(hash, key, data, nextNode);
-   struct cso_hash_iter iter = {hash, node};
-   return iter;
+   {
+      struct cso_node **nextNode = cso_hash_find_node(hash, key);
+      struct cso_node *node = cso_hash_create_node(hash, key, data, nextNode);
+      struct cso_hash_iter iter = {hash, node};
+      return iter;
+   }
 }
 
 struct cso_hash * cso_hash_create(void)
@@ -290,6 +295,10 @@ static struct cso_node *cso_hash_data_next(struct cso_node *node)
       struct cso_node *e;
       struct cso_hash_data *d;
    } a;
+   int start;
+   struct cso_node **bucket;
+   int n;
+
    a.next = node->next;
    if (!a.next) {
       fprintf(stderr, "iterating beyond the last element\n");
@@ -298,9 +307,9 @@ static struct cso_node *cso_hash_data_next(struct cso_node *node)
    if (a.next->next)
       return a.next;
 
-   int start = (node->key % a.d->numBuckets) + 1;
-   struct cso_node **bucket = a.d->buckets + start;
-   int n = a.d->numBuckets - start;
+   start = (node->key % a.d->numBuckets) + 1;
+   bucket = a.d->buckets + start;
+   n = a.d->numBuckets - start;
    while (n--) {
       if (*bucket != a.e)
          return *bucket;
@@ -316,19 +325,21 @@ static struct cso_node *cso_hash_data_prev(struct cso_node *node)
       struct cso_node *e;
       struct cso_hash_data *d;
    } a;
+   int start;
+   struct cso_node *sentinel;
+   struct cso_node **bucket;
 
    a.e = node;
    while (a.e->next)
       a.e = a.e->next;
 
-   int start;
    if (node == a.e)
       start = a.d->numBuckets - 1;
    else
       start = node->key % a.d->numBuckets;
 
-   struct cso_node *sentinel = node;
-   struct cso_node **bucket = a.d->buckets + start;
+   sentinel = node;
+   bucket = a.d->buckets + start;
    while (start >= 0) {
       if (*bucket != sentinel) {
          struct cso_node *prev = *bucket;
diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
index fe9b150f30..c9980f0b83 100644
--- a/src/gallium/auxiliary/draw/Makefile
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -1,2 +1,37 @@
-default:
-	cd ../../../mesa ; make
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = draw
+
+C_SOURCES = \
+	draw_aaline.c \
+	draw_aapoint.c \
+	draw_clip.c \
+	draw_vs_exec.c \
+	draw_vs_sse.c \
+	draw_vs_llvm.c \
+	draw_context.c\
+	draw_cull.c \
+	draw_debug.c \
+	draw_flatshade.c \
+	draw_offset.c \
+	draw_prim.c \
+	draw_pstipple.c \
+	draw_stipple.c \
+	draw_twoside.c \
+	draw_unfilled.c \
+	draw_validate.c \
+	draw_vbuf.c \
+	draw_vertex.c \
+	draw_vertex_cache.c \
+	draw_vertex_fetch.c \
+	draw_vertex_shader.c \
+	draw_vf.c \
+	draw_vf_generic.c \
+	draw_vf_sse.c \
+	draw_wide_prims.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
new file mode 100644
index 0000000000..8e3a8caa74
--- /dev/null
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -0,0 +1,31 @@
+Import('*')
+
+draw = env.ConvenienceLibrary(
+	target = 'draw',
+	source = [
+		'draw_clip.c',
+		'draw_vs_exec.c',
+		'draw_vs_sse.c',
+		'draw_vs_llvm.c',
+		'draw_context.c',
+		'draw_cull.c',
+		'draw_debug.c',
+		'draw_flatshade.c',
+		'draw_offset.c',
+		'draw_prim.c',
+		'draw_stipple.c',
+		'draw_twoside.c',
+		'draw_unfilled.c',
+		'draw_validate.c',
+		'draw_vbuf.c',
+		'draw_vertex.c',
+		'draw_vertex_cache.c',
+		'draw_vertex_fetch.c',
+		'draw_vertex_shader.c',
+		'draw_vf.c',
+		'draw_vf_generic.c',
+		'draw_vf_sse.c',
+		'draw_wide_prims.c',
+	])
+
+auxiliaries.insert(0, draw)
diff --git a/src/gallium/auxiliary/draw/draw_aaline.c b/src/gallium/auxiliary/draw/draw_aaline.c
new file mode 100644
index 0000000000..73a02a32e4
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_aaline.c
@@ -0,0 +1,832 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/**
+ * AA line stage:  AA lines are converted to texture mapped triangles.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/util/tgsi_transform.h"
+#include "tgsi/util/tgsi_dump.h"
+
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+/**
+ * Max texture level for the alpha texture used for antialiasing
+ */
+#define MAX_TEXTURE_LEVEL  5   /* 32 x 32 */
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct aaline_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;
+   void *aaline_fs;
+   void *aapoint_fs; /* not yet */
+   void *sprite_fs; /* not yet */
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct aaline_stage
+{
+   struct draw_stage stage;
+
+   float half_line_width;
+
+   /** For AA lines, this is the vertex attrib slot for the new texcoords */
+   uint tex_slot;
+
+   void *sampler_cso;
+   struct pipe_texture *texture;
+   uint sampler_unit;
+
+
+   /*
+    * Currently bound state
+    */
+   struct aaline_fragment_shader *fs;
+   struct {
+      void *sampler[PIPE_MAX_SAMPLERS];
+      struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+   } state;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   void (*driver_bind_sampler_state)(struct pipe_context *, unsigned, void *);
+
+   void (*driver_set_sampler_texture)(struct pipe_context *,
+                                      unsigned sampler,
+                                      struct pipe_texture *);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct aa_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int colorOutput; /**< which output is the primary color */
+   int maxSampler;  /**< max sampler index found */
+   int maxInput, maxGeneric;  /**< max input index found */
+   int colorTemp, texTemp;  /**< temp registers */
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for a free sampler, a free input attrib, and two free temp regs.
+ */
+static void
+aa_transform_decl(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_declaration *decl)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
+       decl->Semantic.SemanticName == TGSI_SEMANTIC_COLOR &&
+       decl->Semantic.SemanticIndex == 0) {
+      aactx->colorOutput = decl->u.DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_SAMPLER) {
+      if ((int) decl->u.DeclarationRange.Last > aactx->maxSampler)
+         aactx->maxSampler = decl->u.DeclarationRange.Last + 1;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      if ((int) decl->u.DeclarationRange.Last > aactx->maxInput)
+         aactx->maxInput = decl->u.DeclarationRange.Last;
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_GENERIC &&
+           (int) decl->Semantic.SemanticIndex > aactx->maxGeneric) {
+         aactx->maxGeneric = decl->Semantic.SemanticIndex;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->u.DeclarationRange.First;
+           i <= decl->u.DeclarationRange.Last; i++) {
+         aactx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+aa_transform_inst(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_instruction *inst)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (aactx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      uint i;
+
+      /* find two free temp regs */
+      for (i = 0; i < 32; i++) {
+         if ((aactx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (aactx->colorTemp < 0)
+               aactx->colorTemp  = i;
+            else if (aactx->texTemp < 0)
+               aactx->texTemp  = i;
+            else
+               break;
+         }
+      }
+      assert(aactx->colorTemp >= 0);
+      assert(aactx->texTemp >= 0);
+
+      /* declare new generic input/texcoord */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+      decl.Semantic.SemanticIndex = aactx->maxGeneric + 1;
+      decl.Declaration.Interpolate = 1;
+      /* XXX this could be linear... */
+      decl.Interpolation.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = aactx->maxInput + 1;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new sampler */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_SAMPLER;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = aactx->maxSampler + 1;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = aactx->texTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = aactx->colorTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      aactx->firstInstruction = FALSE;
+   }
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END &&
+       aactx->colorOutput != -1) {
+      struct tgsi_full_instruction newInst;
+
+      /* TEX */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_TEX;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->maxInput + 1;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->maxSampler + 1;
+
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MOV rgb */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XYZ;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MUL alpha */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->texTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* END */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_END;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 0;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+   else {
+      /* Not an END instruction.
+       * Look for writes to result.color and replace with colorTemp reg.
+       */
+      uint i;
+
+      for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+         struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+         if (dst->DstRegister.File == TGSI_FILE_OUTPUT &&
+             dst->DstRegister.Index == aactx->colorOutput) {
+            dst->DstRegister.File = TGSI_FILE_TEMPORARY;
+            dst->DstRegister.Index = aactx->colorTemp;
+         }
+      }
+
+      ctx->emit_instruction(ctx, inst);
+   }
+}
+
+
+/**
+ * Generate the frag shader we'll use for drawing AA lines.
+ * This will be the user's shader plus some texture/modulate instructions.
+ */
+static void
+generate_aaline_fs(struct aaline_stage *aaline)
+{
+   const struct pipe_shader_state *orig_fs = &aaline->fs->state;
+   struct draw_context *draw = aaline->stage.draw;
+   struct pipe_shader_state aaline_fs;
+   struct aa_transform_context transform;
+
+#define MAX 1000
+
+   aaline_fs = *orig_fs; /* copy to init */
+   aaline_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+
+   memset(&transform, 0, sizeof(transform));
+   transform.colorOutput = -1;
+   transform.maxSampler = -1;
+   transform.maxInput = -1;
+   transform.maxGeneric = -1;
+   transform.colorTemp = -1;
+   transform.texTemp = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = aa_transform_inst;
+   transform.base.transform_declaration = aa_transform_decl;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) aaline_fs.tokens,
+                         MAX, &transform.base);
+
+#if 0 /* DEBUG */
+   tgsi_dump(orig_fs->tokens, 0);
+   tgsi_dump(aaline_fs.tokens, 0);
+#endif
+
+   aaline_fs.input_semantic_name[aaline_fs.num_inputs] = TGSI_SEMANTIC_GENERIC;
+   aaline_fs.input_semantic_index[aaline_fs.num_inputs] = transform.maxGeneric + 1;
+   aaline_fs.num_inputs++;
+
+   aaline->fs->aaline_fs
+      = aaline->driver_create_fs_state(aaline->pipe, &aaline_fs);
+
+   /* advertise the extra post-transform vertex attributes which will have
+    * the texcoords.
+    */
+   draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
+   draw->extra_vp_outputs.semantic_index = transform.maxGeneric + 1;
+}
+
+
+/**
+ * Create the texture map we'll use for antialiasing the lines.
+ */
+static void
+aaline_create_texture(struct aaline_stage *aaline)
+{
+   struct pipe_context *pipe = aaline->pipe;
+   struct pipe_texture texTemp;
+   uint level;
+
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = PIPE_FORMAT_U_A8; /* XXX verify supported by driver! */
+   texTemp.last_level = MAX_TEXTURE_LEVEL;
+   texTemp.width[0] = 1 << MAX_TEXTURE_LEVEL;
+   texTemp.height[0] = 1 << MAX_TEXTURE_LEVEL;
+   texTemp.depth[0] = 1;
+   texTemp.cpp = 1;
+
+   aaline->texture = pipe->texture_create(pipe, &texTemp);
+
+   /* Fill in mipmap images.
+    * Basically each level is solid opaque, except for the outermost
+    * texels which are zero.  Special case the 1x1 and 2x2 levels.
+    */
+   for (level = 0; level <= MAX_TEXTURE_LEVEL; level++) {
+      struct pipe_surface *surface;
+      const uint size = aaline->texture->width[level];
+      ubyte *data;
+      uint i, j;
+
+      assert(aaline->texture->width[level] == aaline->texture->height[level]);
+
+      surface = pipe->get_tex_surface(pipe, aaline->texture, 0, level, 0);
+      data = pipe_surface_map(surface);
+
+      for (i = 0; i < size; i++) {
+         for (j = 0; j < size; j++) {
+            ubyte d;
+            if (size == 1) {
+               d = 255;
+            }
+            else if (size == 2) {
+               d = 200; /* tuneable */
+            }
+            else if (i == 0 || j == 0 || i == size - 1 || j == size - 1) {
+               d = 0;
+            }
+            else {
+               d = 255;
+            }
+            data[i * surface->pitch + j] = d;
+         }
+      }
+
+      /* unmap */
+      pipe_surface_unmap(surface);
+      pipe_surface_reference(&surface, NULL);
+   }
+}
+
+
+/**
+ * Create the sampler CSO that'll be used for antialiasing.
+ * By using a mipmapped texture, we don't have to generate a different
+ * texture image for each line size.
+ */
+static void
+aaline_create_sampler(struct aaline_stage *aaline)
+{
+   struct pipe_sampler_state sampler;
+   struct pipe_context *pipe = aaline->pipe;
+
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_LINEAR;
+   sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.normalized_coords = 1;
+   sampler.min_lod = 0.0f;
+   sampler.max_lod = MAX_TEXTURE_LEVEL;
+
+   aaline->sampler_cso = pipe->create_sampler_state(pipe, &sampler);
+}
+
+
+/**
+ * When we're about to draw our first AA line in a batch, this function is
+ * called to tell the driver to bind our modified fragment shader.
+ */
+static void
+bind_aaline_fragment_shader(struct aaline_stage *aaline)
+{
+   if (!aaline->fs->aaline_fs) {
+      generate_aaline_fs(aaline);
+   }
+   aaline->driver_bind_fs_state(aaline->pipe, aaline->fs->aaline_fs);
+}
+
+
+
+static INLINE struct aaline_stage *
+aaline_stage( struct draw_stage *stage )
+{
+   return (struct aaline_stage *) stage;
+}
+
+
+static void
+passthrough_point(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->point(stage->next, header);
+}
+
+
+static void
+passthrough_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+/**
+ * Draw a wide line by drawing a quad, using geometry which will
+ * fullfill GL's antialiased line requirements.
+ */
+static void
+aaline_line(struct draw_stage *stage, struct prim_header *header)
+{
+   const struct aaline_stage *aaline = aaline_stage(stage);
+   const float half_width = aaline->half_line_width;
+   struct prim_header tri;
+   struct vertex_header *v[8];
+   uint texPos = aaline->tex_slot;
+   float *pos, *tex;
+   float dx = header->v[1]->data[0][0] - header->v[0]->data[0][0];
+   float dy = header->v[1]->data[0][1] - header->v[0]->data[0][1];
+   double a = atan2(dy, dx);
+   float c_a = (float) cos(a), s_a = (float) sin(a);
+   uint i;
+
+   /* XXX the ends of lines aren't quite perfect yet, but probably passable */
+   dx = 0.5F * half_width;
+   dy = half_width;
+
+   /* allocate/dup new verts */
+   for (i = 0; i < 8; i++) {
+      v[i] = dup_vert(stage, header->v[i/4], i);
+   }
+
+   /*
+    * Quad strip for line from v0 to v1 (*=endpoints):
+    *
+    *  1   3                     5   7
+    *  +---+---------------------+---+
+    *  |                             |
+    *  | *v0                     v1* |
+    *  |                             |
+    *  +---+---------------------+---+
+    *  0   2                     4   6
+    */
+
+   /* new verts */
+   pos = v[0]->data[0];
+   pos[0] += (-dx * c_a -  dy * s_a);
+   pos[1] += (-dx * s_a +  dy * c_a);
+
+   pos = v[1]->data[0];
+   pos[0] += (-dx * c_a - -dy * s_a);
+   pos[1] += (-dx * s_a + -dy * c_a);
+
+   pos = v[2]->data[0];
+   pos[0] += ( dx * c_a -  dy * s_a);
+   pos[1] += ( dx * s_a +  dy * c_a);
+
+   pos = v[3]->data[0];
+   pos[0] += ( dx * c_a - -dy * s_a);
+   pos[1] += ( dx * s_a + -dy * c_a);
+
+   pos = v[4]->data[0];
+   pos[0] += (-dx * c_a -  dy * s_a);
+   pos[1] += (-dx * s_a +  dy * c_a);
+
+   pos = v[5]->data[0];
+   pos[0] += (-dx * c_a - -dy * s_a);
+   pos[1] += (-dx * s_a + -dy * c_a);
+
+   pos = v[6]->data[0];
+   pos[0] += ( dx * c_a -  dy * s_a);
+   pos[1] += ( dx * s_a +  dy * c_a);
+
+   pos = v[7]->data[0];
+   pos[0] += ( dx * c_a - -dy * s_a);
+   pos[1] += ( dx * s_a + -dy * c_a);
+
+   /* new texcoords */
+   tex = v[0]->data[texPos];
+   ASSIGN_4V(tex, 0, 0, 0, 1);
+
+   tex = v[1]->data[texPos];
+   ASSIGN_4V(tex, 0, 1, 0, 1);
+
+   tex = v[2]->data[texPos];
+   ASSIGN_4V(tex, .5, 0, 0, 1);
+
+   tex = v[3]->data[texPos];
+   ASSIGN_4V(tex, .5, 1, 0, 1);
+
+   tex = v[4]->data[texPos];
+   ASSIGN_4V(tex, .5, 0, 0, 1);
+
+   tex = v[5]->data[texPos];
+   ASSIGN_4V(tex, .5, 1, 0, 1);
+
+   tex = v[6]->data[texPos];
+   ASSIGN_4V(tex, 1, 0, 0, 1);
+
+   tex = v[7]->data[texPos];
+   ASSIGN_4V(tex, 1, 1, 0, 1);
+
+   /* emit 6 tris for the quad strip */
+   tri.v[0] = v[2];  tri.v[1] = v[1];  tri.v[2] = v[0];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[3];  tri.v[1] = v[1];  tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[4];  tri.v[1] = v[3];  tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[5];  tri.v[1] = v[3];  tri.v[2] = v[4];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[6];  tri.v[1] = v[5];  tri.v[2] = v[4];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[7];  tri.v[1] = v[5];  tri.v[2] = v[6];
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void
+aaline_first_line(struct draw_stage *stage, struct prim_header *header)
+{
+   auto struct aaline_stage *aaline = aaline_stage(stage);
+   struct draw_context *draw = stage->draw;
+   struct pipe_context *pipe = aaline->pipe;
+
+   assert(draw->rasterizer->line_smooth);
+
+   if (draw->rasterizer->line_width <= 3.0)
+      aaline->half_line_width = 1.5f;
+   else
+      aaline->half_line_width = 0.5f * draw->rasterizer->line_width;
+
+   aaline->tex_slot = draw->num_vs_outputs;
+   assert(aaline->tex_slot > 0); /* output[0] is vertex pos */
+   draw->extra_vp_outputs.slot = aaline->tex_slot;
+
+   /*
+    * Bind our fragprog, sampler and texture
+    */
+   bind_aaline_fragment_shader(aaline);
+
+   aaline->driver_bind_sampler_state(pipe, aaline->sampler_unit, aaline->sampler_cso);
+   aaline->driver_set_sampler_texture(pipe, aaline->sampler_unit, aaline->texture);
+
+   /* now really draw first line */
+   stage->line = aaline_line;
+   stage->line(stage, header);
+}
+
+
+static void
+aaline_flush(struct draw_stage *stage, unsigned flags)
+{
+   struct draw_context *draw = stage->draw;
+   struct aaline_stage *aaline = aaline_stage(stage);
+   struct pipe_context *pipe = aaline->pipe;
+
+   stage->line = aaline_first_line;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader */
+   aaline->driver_bind_fs_state(pipe, aaline->fs->driver_fs);
+
+   /* XXX restore original texture, sampler state */
+   aaline->driver_bind_sampler_state(pipe, aaline->sampler_unit,
+                                 aaline->state.sampler[aaline->sampler_unit]);
+   aaline->driver_set_sampler_texture(pipe, aaline->sampler_unit,
+                                 aaline->state.texture[aaline->sampler_unit]);
+
+   draw->extra_vp_outputs.slot = 0;
+}
+
+
+static void
+aaline_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+aaline_destroy(struct draw_stage *stage)
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+static struct aaline_stage *
+draw_aaline_stage(struct draw_context *draw)
+{
+   struct aaline_stage *aaline = CALLOC_STRUCT(aaline_stage);
+
+   draw_alloc_temp_verts( &aaline->stage, 8 );
+
+   aaline->stage.draw = draw;
+   aaline->stage.next = NULL;
+   aaline->stage.point = passthrough_point;
+   aaline->stage.line = aaline_first_line;
+   aaline->stage.tri = passthrough_tri;
+   aaline->stage.flush = aaline_flush;
+   aaline->stage.reset_stipple_counter = aaline_reset_stipple_counter;
+   aaline->stage.destroy = aaline_destroy;
+
+   return aaline;
+}
+
+
+/*
+ * XXX temporary? solution to mapping a pipe_context to a aaline_stage.
+ */
+
+#define MAX_CONTEXTS 10
+
+static struct pipe_context *Pipe[MAX_CONTEXTS];
+static struct aaline_stage *Stage[MAX_CONTEXTS];
+static uint NumContexts;
+
+static void
+add_aa_pipe_context(struct pipe_context *pipe, struct aaline_stage *aa)
+{
+   assert(NumContexts < MAX_CONTEXTS);
+   Pipe[NumContexts] = pipe;
+   Stage[NumContexts] = aa;
+   NumContexts++;
+}
+
+static struct aaline_stage *
+aaline_stage_from_pipe(struct pipe_context *pipe)
+{
+   uint i;
+   for (i = 0; i < NumContexts; i++) {
+      if (Pipe[i] == pipe)
+         return Stage[i];
+   }
+   assert(0);
+   return NULL;
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+aaline_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = CALLOC_STRUCT(aaline_fragment_shader);
+
+   if (aafs) {
+      aafs->state = *fs;
+
+      /* pass-through */
+      aafs->driver_fs = aaline->driver_create_fs_state(aaline->pipe, fs);
+   }
+
+   return aafs;
+}
+
+
+static void
+aaline_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
+   /* save current */
+   aaline->fs = aafs;
+   /* pass-through */
+   aaline->driver_bind_fs_state(aaline->pipe, aafs->driver_fs);
+}
+
+
+static void
+aaline_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
+   /* pass-through */
+   aaline->driver_delete_fs_state(aaline->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+static void
+aaline_bind_sampler_state(struct pipe_context *pipe,
+                          unsigned unit, void *sampler)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   /* save current */
+   aaline->state.sampler[unit] = sampler;
+   /* pass-through */
+   aaline->driver_bind_sampler_state(aaline->pipe, unit, sampler);
+}
+
+
+static void
+aaline_set_sampler_texture(struct pipe_context *pipe,
+                           unsigned sampler, struct pipe_texture *texture)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   /* save current */
+   aaline->state.texture[sampler] = texture;
+   /* pass-through */
+   aaline->driver_set_sampler_texture(aaline->pipe, sampler, texture);
+}
+
+
+/**
+ * Called by drivers that want to install this AA line prim stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for AA lines.
+ */
+void
+draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
+{
+   struct aaline_stage *aaline;
+
+   /*
+    * Create / install AA line drawing / prim stage
+    */
+   aaline = draw_aaline_stage( draw );
+   assert(aaline);
+   draw->pipeline.aaline = &aaline->stage;
+
+   aaline->pipe = pipe;
+
+   /* create special texture, sampler state */
+   aaline_create_texture(aaline);
+   aaline_create_sampler(aaline);
+
+   /* save original driver functions */
+   aaline->driver_create_fs_state = pipe->create_fs_state;
+   aaline->driver_bind_fs_state = pipe->bind_fs_state;
+   aaline->driver_delete_fs_state = pipe->delete_fs_state;
+
+   aaline->driver_bind_sampler_state = pipe->bind_sampler_state;
+   aaline->driver_set_sampler_texture = pipe->set_sampler_texture;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = aaline_create_fs_state;
+   pipe->bind_fs_state = aaline_bind_fs_state;
+   pipe->delete_fs_state = aaline_delete_fs_state;
+
+   pipe->bind_sampler_state = aaline_bind_sampler_state;
+   pipe->set_sampler_texture = aaline_set_sampler_texture;
+
+   add_aa_pipe_context(pipe, aaline);
+}
diff --git a/src/gallium/auxiliary/draw/draw_aapoint.c b/src/gallium/auxiliary/draw/draw_aapoint.c
new file mode 100644
index 0000000000..43119cc70b
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_aapoint.c
@@ -0,0 +1,875 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/**
+ * AA point stage:  AA points are converted to quads and rendered with a
+ * special fragment shader.  Another approach would be to use a texture
+ * map image of a point, but experiments indicate the quality isn't nearly
+ * as good as this approach.
+ *
+ * Note: this looks a lot like draw_aaline.c but there's actually little
+ * if any code that can be shared.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/util/tgsi_transform.h"
+#include "tgsi/util/tgsi_dump.h"
+
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+/*
+ * Enabling NORMALIZE might give _slightly_ better results.
+ * Basically, it controls whether we compute distance as d=sqrt(x*x+y*y) or
+ * d=x*x+y*y.  Since we're working with a unit circle, the later seems
+ * close enough and saves some costly instructions.
+ */
+#define NORMALIZE 0
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct aapoint_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;   /**< the regular shader */
+   void *aapoint_fs;  /**< the aa point-augmented shader */
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct aapoint_stage
+{
+   struct draw_stage stage;
+
+   int psize_slot;
+   float radius;
+
+   /** this is the vertex attrib slot for the new texcoords */
+   uint tex_slot;
+
+   /*
+    * Currently bound state
+    */
+   struct aapoint_fragment_shader *fs;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct aa_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int colorOutput; /**< which output is the primary color */
+   int maxInput, maxGeneric;  /**< max input index found */
+   int tmp0, colorTemp;  /**< temp registers */
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for two free temp regs and available input reg for new texcoords.
+ */
+static void
+aa_transform_decl(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_declaration *decl)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
+       decl->Semantic.SemanticName == TGSI_SEMANTIC_COLOR &&
+       decl->Semantic.SemanticIndex == 0) {
+      aactx->colorOutput = decl->u.DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      if ((int) decl->u.DeclarationRange.Last > aactx->maxInput)
+         aactx->maxInput = decl->u.DeclarationRange.Last;
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_GENERIC &&
+           (int) decl->Semantic.SemanticIndex > aactx->maxGeneric) {
+         aactx->maxGeneric = decl->Semantic.SemanticIndex;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->u.DeclarationRange.First;
+           i <= decl->u.DeclarationRange.Last; i++) {
+         aactx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+aa_transform_inst(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_instruction *inst)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+   struct tgsi_full_instruction newInst;
+
+   if (aactx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      const int texInput = aactx->maxInput + 1;
+      int tmp0;
+      uint i;
+
+      /* find two free temp regs */
+      for (i = 0; i < 32; i++) {
+         if ((aactx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (aactx->tmp0 < 0)
+               aactx->tmp0 = i;
+            else if (aactx->colorTemp < 0)
+               aactx->colorTemp = i;
+            else
+               break;
+         }
+      }
+
+      assert(aactx->colorTemp != aactx->tmp0);
+
+      tmp0 = aactx->tmp0;
+
+      /* declare new generic input/texcoord */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+      decl.Semantic.SemanticIndex = aactx->maxGeneric + 1;
+      decl.Declaration.Interpolate = 1;
+      /* XXX this could be linear... */
+      decl.Interpolation.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = texInput;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = tmp0;
+      ctx->emit_declaration(ctx, &decl);
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = aactx->colorTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      aactx->firstInstruction = FALSE;
+
+
+      /*
+       * Emit code to compute fragment coverage, kill if outside point radius
+       *
+       * Temp reg0 usage:
+       *  t0.x = distance of fragment from center point
+       *  t0.y = boolean, is t0.x > 1 ?
+       *  t0.z = temporary for computing 1/(1-k) value
+       *  t0.w = final coverage value
+       */
+
+      /* MUL t0.xy, tex, tex;  # compute x^2, y^2 */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XY;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* ADD t0.x, t0.x, t0.y;  # x^2 + y^2 */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_ADD;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      ctx->emit_instruction(ctx, &newInst);
+
+#if NORMALIZE  /* OPTIONAL normalization of length */
+      /* RSQ t0.x, t0.x; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_RSQ;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* RCP t0.x, t0.x; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_RCP;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+#endif
+
+      /* SGT t0.y, t0.xxxx, t0.wwww;  # bool b = d > 1 (NOTE t0.w == 1) */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SGT;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Y;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* KILP -t0.yyyy;   # if b, KILL */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_KILP;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.Negate = 1;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* SGT t0.y, t0.x, tex.z;  # bool b = distance > k */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SGT;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Y;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Z;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* IF t0.y   # if b then */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_IF;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
+      ctx->emit_instruction(ctx, &newInst);
+
+      {
+         /* compute coverage factor = (1-d)/(1-k) */
+
+         /* SUB t0.z, tex.w, tex.z;  # m = 1 - k */
+         newInst = tgsi_default_full_instruction();
+         newInst.Instruction.Opcode = TGSI_OPCODE_SUB;
+         newInst.Instruction.NumDstRegs = 1;
+         newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+         newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Z;
+         newInst.Instruction.NumSrcRegs = 2;
+         newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+         newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+         newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
+         newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+         newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+         newInst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Z;
+         ctx->emit_instruction(ctx, &newInst);
+
+         /* RCP t0.z, t0.z;  # t0.z = 1 / m */
+         newInst = tgsi_default_full_instruction();
+         newInst.Instruction.Opcode = TGSI_OPCODE_RCP;
+         newInst.Instruction.NumDstRegs = 1;
+         newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+         newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Z;
+         newInst.Instruction.NumSrcRegs = 1;
+         newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+         newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Z;
+         ctx->emit_instruction(ctx, &newInst);
+
+         /* SUB t0.x, 1, t0.x;  # d = 1 - d */
+         newInst = tgsi_default_full_instruction();
+         newInst.Instruction.Opcode = TGSI_OPCODE_SUB;
+         newInst.Instruction.NumDstRegs = 1;
+         newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+         newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+         newInst.Instruction.NumSrcRegs = 2;
+         newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+         newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+         newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
+         newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+         newInst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+         ctx->emit_instruction(ctx, &newInst);
+
+         /* MUL t0.w, t0.x, t0.z;   # coverage = d * m */
+         newInst = tgsi_default_full_instruction();
+         newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+         newInst.Instruction.NumDstRegs = 1;
+         newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+         newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+         newInst.Instruction.NumSrcRegs = 2;
+         newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+         newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
+         newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+         newInst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Z;
+         ctx->emit_instruction(ctx, &newInst);
+      }
+
+      /* ELSE */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_ELSE;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 0;
+      ctx->emit_instruction(ctx, &newInst);
+
+      {
+         /* MOV t0.w, tex.w;  # coverage = 1.0 */
+         newInst = tgsi_default_full_instruction();
+         newInst.Instruction.Opcode = TGSI_OPCODE_MOV;
+         newInst.Instruction.NumDstRegs = 1;
+         newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+         newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+         newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+         newInst.Instruction.NumSrcRegs = 1;
+         newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+         newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+         ctx->emit_instruction(ctx, &newInst);
+      }
+
+      /* ENDIF */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_ENDIF;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 0;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
+      /* add alpha modulation code at tail of program */
+
+      /* MOV result.color.xyz, colorTemp; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XYZ;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MUL result.color.w, colorTemp, tmp0.w; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+   else {
+      /* Not an END instruction.
+       * Look for writes to result.color and replace with colorTemp reg.
+       */
+      uint i;
+
+      for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+         struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+         if (dst->DstRegister.File == TGSI_FILE_OUTPUT &&
+             dst->DstRegister.Index == aactx->colorOutput) {
+            dst->DstRegister.File = TGSI_FILE_TEMPORARY;
+            dst->DstRegister.Index = aactx->colorTemp;
+         }
+      }
+   }
+
+   ctx->emit_instruction(ctx, inst);
+}
+
+
+/**
+ * Generate the frag shader we'll use for drawing AA lines.
+ * This will be the user's shader plus some texture/modulate instructions.
+ */
+static void
+generate_aapoint_fs(struct aapoint_stage *aapoint)
+{
+   const struct pipe_shader_state *orig_fs = &aapoint->fs->state;
+   struct draw_context *draw = aapoint->stage.draw;
+   struct pipe_shader_state aapoint_fs;
+   struct aa_transform_context transform;
+
+#define MAX 1000
+
+   aapoint_fs = *orig_fs; /* copy to init */
+   aapoint_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+
+   memset(&transform, 0, sizeof(transform));
+   transform.colorOutput = -1;
+   transform.maxInput = -1;
+   transform.maxGeneric = -1;
+   transform.colorTemp = -1;
+   transform.tmp0 = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = aa_transform_inst;
+   transform.base.transform_declaration = aa_transform_decl;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) aapoint_fs.tokens,
+                         MAX, &transform.base);
+
+#if 0 /* DEBUG */
+   tgsi_dump(orig_fs->tokens, 0);
+   tgsi_dump(aapoint_fs.tokens, 0);
+#endif
+
+   aapoint_fs.input_semantic_name[aapoint_fs.num_inputs] = TGSI_SEMANTIC_GENERIC;
+   aapoint_fs.input_semantic_index[aapoint_fs.num_inputs] = transform.maxGeneric + 1;
+   aapoint_fs.num_inputs++;
+
+   aapoint->fs->aapoint_fs
+      = aapoint->driver_create_fs_state(aapoint->pipe, &aapoint_fs);
+
+   /* advertise the extra post-transform vertex attributes which will have
+    * the texcoords.
+    */
+   draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
+   draw->extra_vp_outputs.semantic_index = transform.maxGeneric + 1;
+}
+
+
+/**
+ * When we're about to draw our first AA line in a batch, this function is
+ * called to tell the driver to bind our modified fragment shader.
+ */
+static void
+bind_aapoint_fragment_shader(struct aapoint_stage *aapoint)
+{
+   if (!aapoint->fs->aapoint_fs) {
+      generate_aapoint_fs(aapoint);
+   }
+   aapoint->driver_bind_fs_state(aapoint->pipe, aapoint->fs->aapoint_fs);
+}
+
+
+
+static INLINE struct aapoint_stage *
+aapoint_stage( struct draw_stage *stage )
+{
+   return (struct aapoint_stage *) stage;
+}
+
+
+static void
+passthrough_line(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->line(stage->next, header);
+}
+
+
+static void
+passthrough_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+/**
+ * Draw an AA point by drawing a quad.
+ */
+static void
+aapoint_point(struct draw_stage *stage, struct prim_header *header)
+{
+   const struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct prim_header tri;
+   struct vertex_header *v[4];
+   uint texPos = aapoint->tex_slot;
+   float radius, *pos, *tex;
+   uint i;
+   float k;
+
+   if (aapoint->psize_slot >= 0) {
+      radius = 0.5f * header->v[0]->data[aapoint->psize_slot][0];
+   }
+   else {
+      radius = aapoint->radius;
+   }
+
+   /*
+    * Note: the texcoords (generic attrib, really) we use are special:
+    * The S and T components simply vary from -1 to +1.
+    * The R component is k, below.
+    * The Q component is 1.0 and will used as a handy constant in the
+    * fragment shader.
+    */
+
+   /*
+    * k is the threshold distance from the point's center at which
+    * we begin alpha attenuation (the coverage value).
+    * Operating within a unit circle, we'll compute the fragment's
+    * distance 'd' from the center point using the texcoords.
+    * IF d > 1.0 THEN
+    *    KILL fragment
+    * ELSE IF d > k THEN
+    *    compute coverage in [0,1] proportional to d in [k, 1].
+    * ELSE
+    *    coverage = 1.0;  // full coverage
+    * ENDIF
+    */
+
+#if !NORMALIZE
+   k = 1.0 / radius;
+   k = 1.0 - 2.0 * k + k * k;
+#else
+   k = 1.0 - 1.0 / radius;
+#endif
+
+   /* allocate/dup new verts */
+   for (i = 0; i < 4; i++) {
+      v[i] = dup_vert(stage, header->v[0], i);
+   }
+
+   /* new verts */
+   pos = v[0]->data[0];
+   pos[0] -= radius;
+   pos[1] -= radius;
+
+   pos = v[1]->data[0];
+   pos[0] += radius;
+   pos[1] -= radius;
+
+   pos = v[2]->data[0];
+   pos[0] += radius;
+   pos[1] += radius;
+
+   pos = v[3]->data[0];
+   pos[0] -= radius;
+   pos[1] += radius;
+
+   /* new texcoords */
+   tex = v[0]->data[texPos];
+   ASSIGN_4V(tex, -1, -1, k, 1);
+
+   tex = v[1]->data[texPos];
+   ASSIGN_4V(tex,  1, -1, k, 1);
+
+   tex = v[2]->data[texPos];
+   ASSIGN_4V(tex,  1,  1, k, 1);
+
+   tex = v[3]->data[texPos];
+   ASSIGN_4V(tex, -1,  1, k, 1);
+
+   /* emit 2 tris for the quad strip */
+   tri.v[0] = v[0];
+   tri.v[1] = v[1];
+   tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[0];
+   tri.v[1] = v[2];
+   tri.v[2] = v[3];
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void
+aapoint_first_point(struct draw_stage *stage, struct prim_header *header)
+{
+   auto struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   assert(draw->rasterizer->point_smooth);
+
+   if (draw->rasterizer->point_size <= 2.0)
+      aapoint->radius = 1.0;
+   else
+      aapoint->radius = 0.5f * draw->rasterizer->point_size;
+
+   aapoint->tex_slot = draw->num_vs_outputs;
+   assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */
+   draw->extra_vp_outputs.slot = aapoint->tex_slot;
+
+   /*
+    * Bind our fragprog.
+    */
+   bind_aapoint_fragment_shader(aapoint);
+
+   /* find psize slot in post-transform vertex */
+   aapoint->psize_slot = -1;
+   if (draw->rasterizer->point_size_per_vertex) {
+      /* find PSIZ vertex output */
+      const struct draw_vertex_shader *vs = draw->vertex_shader;
+      uint i;
+      for (i = 0; i < vs->state->num_outputs; i++) {
+         if (vs->state->output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) {
+            aapoint->psize_slot = i;
+            break;
+         }
+      }
+   }
+
+   /* now really draw first line */
+   stage->point = aapoint_point;
+   stage->point(stage, header);
+}
+
+
+static void
+aapoint_flush(struct draw_stage *stage, unsigned flags)
+{
+   struct draw_context *draw = stage->draw;
+   struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct pipe_context *pipe = aapoint->pipe;
+
+   stage->point = aapoint_first_point;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader */
+   aapoint->driver_bind_fs_state(pipe, aapoint->fs->driver_fs);
+
+   draw->extra_vp_outputs.slot = 0;
+}
+
+
+static void
+aapoint_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+aapoint_destroy(struct draw_stage *stage)
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+static struct aapoint_stage *
+draw_aapoint_stage(struct draw_context *draw)
+{
+   struct aapoint_stage *aapoint = CALLOC_STRUCT(aapoint_stage);
+
+   draw_alloc_temp_verts( &aapoint->stage, 4 );
+
+   aapoint->stage.draw = draw;
+   aapoint->stage.next = NULL;
+   aapoint->stage.point = aapoint_first_point;
+   aapoint->stage.line = passthrough_line;
+   aapoint->stage.tri = passthrough_tri;
+   aapoint->stage.flush = aapoint_flush;
+   aapoint->stage.reset_stipple_counter = aapoint_reset_stipple_counter;
+   aapoint->stage.destroy = aapoint_destroy;
+
+   return aapoint;
+}
+
+
+/*
+ * XXX temporary? solution to mapping a pipe_context to a aapoint_stage.
+ */
+
+#define MAX_CONTEXTS 10
+
+static struct pipe_context *Pipe[MAX_CONTEXTS];
+static struct aapoint_stage *Stage[MAX_CONTEXTS];
+static uint NumContexts;
+
+static void
+add_aa_pipe_context(struct pipe_context *pipe, struct aapoint_stage *aa)
+{
+   assert(NumContexts < MAX_CONTEXTS);
+   Pipe[NumContexts] = pipe;
+   Stage[NumContexts] = aa;
+   NumContexts++;
+}
+
+static struct aapoint_stage *
+aapoint_stage_from_pipe(struct pipe_context *pipe)
+{
+   uint i;
+   for (i = 0; i < NumContexts; i++) {
+      if (Pipe[i] == pipe)
+         return Stage[i];
+   }
+   assert(0);
+   return NULL;
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+aapoint_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = CALLOC_STRUCT(aapoint_fragment_shader);
+
+   if (aafs) {
+      aafs->state = *fs;
+
+      /* pass-through */
+      aafs->driver_fs = aapoint->driver_create_fs_state(aapoint->pipe, fs);
+   }
+
+   return aafs;
+}
+
+
+static void
+aapoint_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = (struct aapoint_fragment_shader *) fs;
+   /* save current */
+   aapoint->fs = aafs;
+   /* pass-through */
+   aapoint->driver_bind_fs_state(aapoint->pipe, aafs->driver_fs);
+}
+
+
+static void
+aapoint_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = (struct aapoint_fragment_shader *) fs;
+   /* pass-through */
+   aapoint->driver_delete_fs_state(aapoint->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+/**
+ * Called by drivers that want to install this AA point prim stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for AA points.
+ */
+void
+draw_install_aapoint_stage(struct draw_context *draw,
+                           struct pipe_context *pipe)
+{
+   struct aapoint_stage *aapoint;
+
+   /*
+    * Create / install AA point drawing / prim stage
+    */
+   aapoint = draw_aapoint_stage( draw );
+   assert(aapoint);
+   draw->pipeline.aapoint = &aapoint->stage;
+
+   aapoint->pipe = pipe;
+
+   /* save original driver functions */
+   aapoint->driver_create_fs_state = pipe->create_fs_state;
+   aapoint->driver_bind_fs_state = pipe->bind_fs_state;
+   aapoint->driver_delete_fs_state = pipe->delete_fs_state;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = aapoint_create_fs_state;
+   pipe->bind_fs_state = aapoint_bind_fs_state;
+   pipe->delete_fs_state = aapoint_delete_fs_state;
+
+   add_aa_pipe_context(pipe, aapoint);
+}
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 4be3830316..c28e78d33a 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -72,10 +72,10 @@ struct draw_context *draw_create( void )
    {
       uint i;
       const unsigned size = (MAX_VERTEX_SIZE + 0x0f) & ~0x0f;
-      char *tmp = align_malloc(Elements(draw->vcache.vertex) * size, 16);
+      char *tmp = align_malloc(Elements(draw->vs.queue) * size, 16);
 
-      for (i = 0; i < Elements(draw->vcache.vertex); i++)
-	 draw->vcache.vertex[i] = (struct vertex_header *)(tmp + i * size);
+      for (i = 0; i < Elements(draw->vs.queue); i++)
+	 draw->vs.queue[i].vertex = (struct vertex_header *)(tmp + i * size);
    }
 
    draw->shader_queue_flush = draw_vertex_shader_queue_flush;
@@ -103,10 +103,14 @@ void draw_destroy( struct draw_context *draw )
    draw->pipeline.flatshade->destroy( draw->pipeline.flatshade );
    draw->pipeline.cull->destroy( draw->pipeline.cull );
    draw->pipeline.validate->destroy( draw->pipeline.validate );
+   if (draw->pipeline.aaline)
+      draw->pipeline.aaline->destroy( draw->pipeline.aaline );
+   if (draw->pipeline.aapoint)
+      draw->pipeline.aapoint->destroy( draw->pipeline.aapoint );
    if (draw->pipeline.rasterize)
       draw->pipeline.rasterize->destroy( draw->pipeline.rasterize );
    tgsi_exec_machine_free_data(&draw->machine);
-   align_free( draw->vcache.vertex[0] ); /* Frees all the vertices. */
+   align_free( draw->vs.queue[0].vertex ); /* Frees all the vertices. */
    FREE( draw );
 }
 
@@ -240,6 +244,26 @@ draw_convert_wide_lines(struct draw_context *draw, boolean enable)
 
 
 /**
+ * The draw module may sometimes generate vertices with extra attributes
+ * (such as texcoords for AA lines).  The driver can call this function
+ * to find those attributes.
+ */
+int
+draw_find_vs_output(struct draw_context *draw,
+                    uint semantic_name, uint semantic_index)
+{
+   /* XXX there may be more than one extra vertex attrib.
+    * For example, simulated gl_FragCoord and gl_PointCoord.
+    */
+   if (draw->extra_vp_outputs.semantic_name == semantic_name &&
+       draw->extra_vp_outputs.semantic_index == semantic_index) {
+      return draw->extra_vp_outputs.slot;
+   }
+   return 0;
+}
+
+
+/**
  * Allocate space for temporary post-transform vertices, such as for clipping.
  */
 void draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr )
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index ddeb184497..c25301f71d 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -41,6 +41,7 @@
 #include "pipe/p_state.h"
 
 
+struct pipe_context;
 struct vertex_buffer;
 struct vertex_info;
 struct draw_context;
@@ -93,6 +94,26 @@ void draw_convert_wide_points(struct draw_context *draw, boolean enable);
 
 void draw_convert_wide_lines(struct draw_context *draw, boolean enable);
 
+boolean draw_use_sse(struct draw_context *draw);
+
+void
+draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+void
+draw_install_aapoint_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+void
+draw_install_pstipple_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+
+int
+draw_find_vs_output(struct draw_context *draw,
+                    uint semantic_name, uint semantic_index);
+
+
+/*
+ * Vertex shader functions
+ */
 
 struct draw_vertex_shader *
 draw_create_vertex_shader(struct draw_context *draw,
@@ -102,7 +123,11 @@ void draw_bind_vertex_shader(struct draw_context *draw,
 void draw_delete_vertex_shader(struct draw_context *draw,
                                struct draw_vertex_shader *dvs);
 
-boolean draw_use_sse(struct draw_context *draw);
+
+
+/*
+ * Vertex data functions
+ */
 
 void draw_set_vertex_buffer(struct draw_context *draw,
 			    unsigned attr,
diff --git a/src/gallium/auxiliary/draw/draw_prim.c b/src/gallium/auxiliary/draw/draw_prim.c
index 51e2242719..7d6cd43410 100644
--- a/src/gallium/auxiliary/draw/draw_prim.c
+++ b/src/gallium/auxiliary/draw/draw_prim.c
@@ -114,6 +114,7 @@ static void draw_prim_queue_flush( struct draw_context *draw )
    }
 
    draw->pq.queue_nr = 0;   
+   draw->vs.post_nr = 0;   
    draw_vertex_cache_unreference( draw );
 }
 
@@ -121,11 +122,15 @@ static void draw_prim_queue_flush( struct draw_context *draw )
 
 void draw_do_flush( struct draw_context *draw, unsigned flags )
 {
+   static boolean flushing = FALSE;
+
    if (0)
       debug_printf("Flushing with %d verts, %d prims\n",
                    draw->vs.queue_nr,
                    draw->pq.queue_nr );
 
+   if (!flushing) {
+      flushing = TRUE;
 
    if (flags >= DRAW_FLUSH_SHADER_QUEUE) {
       if (draw->vs.queue_nr)
@@ -146,6 +151,9 @@ void draw_do_flush( struct draw_context *draw, unsigned flags )
 	 }
       }    
    }
+
+      flushing = FALSE;
+   }
 }
 
 
@@ -219,6 +227,7 @@ static void do_triangle( struct draw_context *draw,
 {
    struct prim_header *prim = get_queued_prim( draw, 3 );
    
+//   _mesa_printf("tri %d %d %d\n", i0, i1, i2);
    prim->reset_line_stipple = 1;
    prim->edgeflags = ~0;
    prim->pad = 0;
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index bc11259cb2..6abced139b 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -44,10 +44,11 @@
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
 
-#include "x86/rtasm/x86sse.h"
+#include "rtasm/rtasm_x86sse.h"
 #include "tgsi/exec/tgsi_exec.h"
 
 
+struct pipe_context;
 struct gallivm_prog;
 struct gallivm_cpu_engine;
 
@@ -119,7 +120,7 @@ struct draw_stage
 };
 
 
-#define PRIM_QUEUE_LENGTH      16
+#define PRIM_QUEUE_LENGTH      32
 #define VCACHE_SIZE            32
 #define VCACHE_OVERFLOW        4
 #define VS_QUEUE_LENGTH        (VCACHE_SIZE + VCACHE_OVERFLOW + 1)	/* can never fill up */
@@ -179,6 +180,9 @@ struct draw_context
       struct draw_stage *offset;
       struct draw_stage *unfilled;
       struct draw_stage *stipple;
+      struct draw_stage *aapoint;
+      struct draw_stage *aaline;
+      struct draw_stage *pstipple;
       struct draw_stage *wide;
       struct draw_stage *rasterize;
    } pipeline;
@@ -212,9 +216,17 @@ struct draw_context
    unsigned nr_planes;
 
    boolean convert_wide_points; /**< convert wide points to tris? */
-   boolean convert_wide_lines;  /**< convert side lines to tris? */
+   boolean convert_wide_lines;  /**< convert wide lines to tris? */
    boolean use_sse;
 
+   /* If a prim stage introduces new vertex attributes, they'll be stored here
+    */
+   struct {
+      uint semantic_name;
+      uint semantic_index;
+      int slot;
+   } extra_vp_outputs;
+
    unsigned reduced_prim;
 
    /** TGSI program interpreter runtime state */
@@ -234,8 +246,12 @@ struct draw_context
     */
    struct {
       unsigned referenced;  /**< bitfield */
-      unsigned idx[VCACHE_SIZE + VCACHE_OVERFLOW];
-      struct vertex_header *vertex[VCACHE_SIZE + VCACHE_OVERFLOW];
+
+      struct {
+	 unsigned in;		/* client array element */
+	 unsigned out;		/* index in vs queue/array */
+      } idx[VCACHE_SIZE + VCACHE_OVERFLOW];
+
       unsigned overflow;
 
       /** To find space in the vertex cache: */
@@ -248,9 +264,10 @@ struct draw_context
    struct {
       struct {
 	 unsigned elt;   /**< index into the user's vertex arrays */
-	 struct vertex_header *dest; /**< points into vcache.vertex[] array */
+	 struct vertex_header *vertex;
       } queue[VS_QUEUE_LENGTH];
       unsigned queue_nr;
+      unsigned post_nr;
    } vs;
 
    /**
diff --git a/src/gallium/auxiliary/draw/draw_pstipple.c b/src/gallium/auxiliary/draw/draw_pstipple.c
new file mode 100644
index 0000000000..4048abf856
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pstipple.c
@@ -0,0 +1,717 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/**
+ * Polygon stipple stage:  implement polygon stipple with texture map and
+ * fragment program.  The fragment program samples the texture and does
+ * a fragment kill for the stipple-failing fragments.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/util/tgsi_transform.h"
+#include "tgsi/util/tgsi_dump.h"
+
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct pstip_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;
+   void *pstip_fs;
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct pstip_stage
+{
+   struct draw_stage stage;
+
+   void *sampler_cso;
+   struct pipe_texture *texture;
+   uint sampler_unit;
+
+   /*
+    * Currently bound state
+    */
+   struct pstip_fragment_shader *fs;
+   struct {
+      void *sampler[PIPE_MAX_SAMPLERS];
+      struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+      const struct pipe_poly_stipple *stipple;
+   } state;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   void (*driver_bind_sampler_state)(struct pipe_context *, unsigned, void *);
+
+   void (*driver_set_sampler_texture)(struct pipe_context *,
+                                      unsigned sampler,
+                                      struct pipe_texture *);
+
+   void (*driver_set_polygon_stipple)(struct pipe_context *,
+                                      const struct pipe_poly_stipple *);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct pstip_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int wincoordInput;
+   int maxInput;
+   int maxSampler;  /**< max sampler index found */
+   int texTemp;  /**< temp registers */
+   int numImmed;
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for a free sampler, a free input attrib, and two free temp regs.
+ */
+static void
+pstip_transform_decl(struct tgsi_transform_context *ctx,
+                     struct tgsi_full_declaration *decl)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_SAMPLER) {
+      if ((int) decl->u.DeclarationRange.Last > pctx->maxSampler)
+         pctx->maxSampler = (int) decl->u.DeclarationRange.Last;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      pctx->maxInput = MAX2(pctx->maxInput, decl->u.DeclarationRange.Last);
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_POSITION)
+         pctx->wincoordInput = (int) decl->u.DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->u.DeclarationRange.First;
+           i <= decl->u.DeclarationRange.Last; i++) {
+         pctx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+static void
+pstip_transform_immed(struct tgsi_transform_context *ctx,
+                      struct tgsi_full_immediate *immed)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+   pctx->numImmed++;
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+pstip_transform_inst(struct tgsi_transform_context *ctx,
+                     struct tgsi_full_instruction *inst)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+
+   if (pctx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      struct tgsi_full_instruction newInst;
+      uint i;
+      int wincoordInput;
+      const int sampler = pctx->maxSampler + 1;
+
+      if (pctx->wincoordInput < 0)
+         wincoordInput = pctx->maxInput + 1;
+      else
+         wincoordInput = pctx->wincoordInput;
+
+      /* find one free temp reg */
+      for (i = 0; i < 32; i++) {
+         if ((pctx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (pctx->texTemp < 0)
+               pctx->texTemp  = i;
+            else
+               break;
+         }
+      }
+      assert(pctx->texTemp >= 0);
+
+      if (pctx->wincoordInput < 0) {
+         /* declare new position input reg */
+         decl = tgsi_default_full_declaration();
+         decl.Declaration.File = TGSI_FILE_INPUT;
+         decl.Declaration.Semantic = 1;
+         decl.Semantic.SemanticName = TGSI_SEMANTIC_POSITION;
+         decl.Semantic.SemanticIndex = 0;
+         decl.Declaration.Interpolate = 1;
+         decl.Interpolation.Interpolate = TGSI_INTERPOLATE_LINEAR; /* XXX? */
+         decl.u.DeclarationRange.First = 
+            decl.u.DeclarationRange.Last = wincoordInput;
+         ctx->emit_declaration(ctx, &decl);
+      }
+
+      /* declare new sampler */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_SAMPLER;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = sampler;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.u.DeclarationRange.First = 
+      decl.u.DeclarationRange.Last = pctx->texTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* emit immediate = {1/32, 1/32, 1, 1}
+       * The index/position of this immediate will be pctx->numImmed
+       */
+      {
+         static const float value[4] = { 1.0/32, 1.0/32, 1.0, 1.0 };
+         struct tgsi_full_immediate immed;
+         uint size = 4;
+         immed = tgsi_default_full_immediate();
+         immed.Immediate.Size = 1 + size; /* one for the token itself */
+         immed.u.ImmediateFloat32 = (struct tgsi_immediate_float32 *) value;
+         ctx->emit_immediate(ctx, &immed);
+      }
+
+      pctx->firstInstruction = FALSE;
+
+
+      /* 
+       * Insert new MUL/TEX/KILP instructions at start of program
+       * Take gl_FragCoord, divide by 32 (stipple size), sample the
+       * texture and kill fragment if needed.
+       *
+       * We'd like to use non-normalized texcoords to index into a RECT
+       * texture, but we can only use GL_REPEAT wrap mode with normalized
+       * texcoords.  Darn.
+       */
+
+      /* MUL texTemp, INPUT[wincoord], 1/32; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = pctx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = wincoordInput;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_IMMEDIATE;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = pctx->numImmed;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* TEX texTemp, texTemp, sampler; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_TEX;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = pctx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = pctx->texTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = sampler;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* KILP texTemp;   # if texTemp < 0, KILL fragment */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_KILP;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = pctx->texTemp;
+      newInst.FullSrcRegisters[0].SrcRegister.Negate = 1;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+
+   /* emit this instruction */
+   ctx->emit_instruction(ctx, inst);
+}
+
+
+/**
+ * Generate the frag shader we'll use for doing polygon stipple.
+ * This will be the user's shader prefixed with a TEX and KIL instruction.
+ */
+static void
+generate_pstip_fs(struct pstip_stage *pstip)
+{
+   const struct pipe_shader_state *orig_fs = &pstip->fs->state;
+   /*struct draw_context *draw = pstip->stage.draw;*/
+   struct pipe_shader_state pstip_fs;
+   struct pstip_transform_context transform;
+
+#define MAX 1000
+
+   pstip_fs = *orig_fs; /* copy to init */
+   pstip_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+
+   memset(&transform, 0, sizeof(transform));
+   transform.wincoordInput = -1;
+   transform.maxInput = -1;
+   transform.maxSampler = -1;
+   transform.texTemp = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = pstip_transform_inst;
+   transform.base.transform_declaration = pstip_transform_decl;
+   transform.base.transform_immediate = pstip_transform_immed;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) pstip_fs.tokens,
+                         MAX, &transform.base);
+
+#if 1 /* DEBUG */
+   tgsi_dump(orig_fs->tokens, 0);
+   tgsi_dump(pstip_fs.tokens, 0);
+#endif
+
+   pstip->sampler_unit = transform.maxSampler + 1;
+
+   if (transform.wincoordInput < 0) {
+      pstip_fs.input_semantic_name[pstip_fs.num_inputs] = TGSI_SEMANTIC_POSITION;
+      pstip_fs.input_semantic_index[pstip_fs.num_inputs] = transform.maxInput;
+      pstip_fs.num_inputs++;
+   }
+
+   pstip->fs->pstip_fs = pstip->driver_create_fs_state(pstip->pipe, &pstip_fs);
+}
+
+
+/**
+ * Load texture image with current stipple pattern.
+ */
+static void
+pstip_update_texture(struct pstip_stage *pstip)
+{
+   static const uint bit31 = 1 << 31;
+   struct pipe_context *pipe = pstip->pipe;
+   struct pipe_surface *surface;
+   const uint *stipple = pstip->state.stipple->stipple;
+   uint i, j;
+   ubyte *data;
+
+   surface = pipe->get_tex_surface(pipe, pstip->texture, 0, 0, 0);
+   data = pipe_surface_map(surface);
+
+   /*
+    * Load alpha texture.
+    * Note: 0 means keep the fragment, 255 means kill it.
+    * We'll negate the texel value and use KILP which kills if value
+    * is negative.
+    */
+   for (i = 0; i < 32; i++) {
+      for (j = 0; j < 32; j++) {
+         if (stipple[i] & (bit31 >> j)) {
+            /* fragment "on" */
+            data[i * surface->pitch + j] = 0;
+         }
+         else {
+            /* fragment "off" */
+            data[i * surface->pitch + j] = 255;
+         }
+      }
+   }
+
+   /* unmap */
+   pipe_surface_unmap(surface);
+   pipe_surface_reference(&surface, NULL);
+   pipe->texture_update(pipe, pstip->texture);
+}
+
+
+/**
+ * Create the texture map we'll use for stippling.
+ */
+static void
+pstip_create_texture(struct pstip_stage *pstip)
+{
+   struct pipe_context *pipe = pstip->pipe;
+   struct pipe_texture texTemp;
+
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = PIPE_FORMAT_U_A8; /* XXX verify supported by driver! */
+   texTemp.last_level = 0;
+   texTemp.width[0] = 32;
+   texTemp.height[0] = 32;
+   texTemp.depth[0] = 1;
+   texTemp.cpp = 1;
+
+   pstip->texture = pipe->texture_create(pipe, &texTemp);
+
+   //pstip_update_texture(pstip);
+}
+
+
+/**
+ * Create the sampler CSO that'll be used for antialiasing.
+ * By using a mipmapped texture, we don't have to generate a different
+ * texture image for each line size.
+ */
+static void
+pstip_create_sampler(struct pstip_stage *pstip)
+{
+   struct pipe_sampler_state sampler;
+   struct pipe_context *pipe = pstip->pipe;
+
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
+   sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
+   sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+   sampler.normalized_coords = 1;
+   sampler.min_lod = 0.0f;
+   sampler.max_lod = 0.0f;
+
+   pstip->sampler_cso = pipe->create_sampler_state(pipe, &sampler);
+}
+
+
+/**
+ * When we're about to draw our first AA line in a batch, this function is
+ * called to tell the driver to bind our modified fragment shader.
+ */
+static void
+bind_pstip_fragment_shader(struct pstip_stage *pstip)
+{
+   if (!pstip->fs->pstip_fs) {
+      generate_pstip_fs(pstip);
+   }
+   pstip->driver_bind_fs_state(pstip->pipe, pstip->fs->pstip_fs);
+}
+
+
+
+static INLINE struct pstip_stage *
+pstip_stage( struct draw_stage *stage )
+{
+   return (struct pstip_stage *) stage;
+}
+
+
+static void
+passthrough_point(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->point(stage->next, header);
+}
+
+
+static void
+passthrough_line(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->line(stage->next, header);
+}
+
+
+static void
+passthrough_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+
+static void
+pstip_first_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   struct pstip_stage *pstip = pstip_stage(stage);
+   struct draw_context *draw = stage->draw;
+   struct pipe_context *pipe = pstip->pipe;
+
+   assert(draw->rasterizer->poly_stipple_enable);
+
+   /*
+    * Bind our fragprog, sampler and texture
+    */
+   bind_pstip_fragment_shader(pstip);
+
+   pstip->driver_bind_sampler_state(pipe, pstip->sampler_unit, pstip->sampler_cso);
+   pstip->driver_set_sampler_texture(pipe, pstip->sampler_unit, pstip->texture);
+
+   /* now really draw first line */
+   stage->tri = passthrough_tri;
+   stage->tri(stage, header);
+}
+
+
+static void
+pstip_flush(struct draw_stage *stage, unsigned flags)
+{
+   /*struct draw_context *draw = stage->draw;*/
+   struct pstip_stage *pstip = pstip_stage(stage);
+   struct pipe_context *pipe = pstip->pipe;
+
+   stage->tri = pstip_first_tri;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader */
+   pstip->driver_bind_fs_state(pipe, pstip->fs->driver_fs);
+
+   /* XXX restore original texture, sampler state */
+   pstip->driver_bind_sampler_state(pipe, pstip->sampler_unit,
+                                 pstip->state.sampler[pstip->sampler_unit]);
+   pstip->driver_set_sampler_texture(pipe, pstip->sampler_unit,
+                                 pstip->state.texture[pstip->sampler_unit]);
+}
+
+
+static void
+pstip_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+pstip_destroy(struct draw_stage *stage)
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+static struct pstip_stage *
+draw_pstip_stage(struct draw_context *draw)
+{
+   struct pstip_stage *pstip = CALLOC_STRUCT(pstip_stage);
+
+   draw_alloc_temp_verts( &pstip->stage, 8 );
+
+   pstip->stage.draw = draw;
+   pstip->stage.next = NULL;
+   pstip->stage.point = passthrough_point;
+   pstip->stage.line = passthrough_line;
+   pstip->stage.tri = pstip_first_tri;
+   pstip->stage.flush = pstip_flush;
+   pstip->stage.reset_stipple_counter = pstip_reset_stipple_counter;
+   pstip->stage.destroy = pstip_destroy;
+
+   return pstip;
+}
+
+
+/*
+ * XXX temporary? solution to mapping a pipe_context to a pstip_stage.
+ */
+
+#define MAX_CONTEXTS 10
+
+static struct pipe_context *Pipe[MAX_CONTEXTS];
+static struct pstip_stage *Stage[MAX_CONTEXTS];
+static uint NumContexts;
+
+static void
+add_pstip_pipe_context(struct pipe_context *pipe, struct pstip_stage *pstip)
+{
+   assert(NumContexts < MAX_CONTEXTS);
+   Pipe[NumContexts] = pipe;
+   Stage[NumContexts] = pstip;
+   NumContexts++;
+}
+
+static struct pstip_stage *
+pstip_stage_from_pipe(struct pipe_context *pipe)
+{
+   uint i;
+   for (i = 0; i < NumContexts; i++) {
+      if (Pipe[i] == pipe)
+         return Stage[i];
+   }
+   assert(0);
+   return NULL;
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+pstip_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = CALLOC_STRUCT(pstip_fragment_shader);
+
+   if (aafs) {
+      aafs->state = *fs;
+
+      /* pass-through */
+      aafs->driver_fs = pstip->driver_create_fs_state(pstip->pipe, fs);
+   }
+
+   return aafs;
+}
+
+
+static void
+pstip_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = (struct pstip_fragment_shader *) fs;
+   /* save current */
+   pstip->fs = aafs;
+   /* pass-through */
+   pstip->driver_bind_fs_state(pstip->pipe, aafs->driver_fs);
+}
+
+
+static void
+pstip_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = (struct pstip_fragment_shader *) fs;
+   /* pass-through */
+   pstip->driver_delete_fs_state(pstip->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+static void
+pstip_bind_sampler_state(struct pipe_context *pipe,
+                         unsigned unit, void *sampler)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   /* save current */
+   pstip->state.sampler[unit] = sampler;
+   /* pass-through */
+   pstip->driver_bind_sampler_state(pstip->pipe, unit, sampler);
+}
+
+
+static void
+pstip_set_sampler_texture(struct pipe_context *pipe,
+                          unsigned sampler, struct pipe_texture *texture)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   /* save current */
+   pstip->state.texture[sampler] = texture;
+   /* pass-through */
+   pstip->driver_set_sampler_texture(pstip->pipe, sampler, texture);
+}
+
+
+static void
+pstip_set_polygon_stipple(struct pipe_context *pipe,
+                          const struct pipe_poly_stipple *stipple)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   /* save current */
+   pstip->state.stipple = stipple;
+   /* pass-through */
+   pstip->driver_set_polygon_stipple(pstip->pipe, stipple);
+
+   pstip_update_texture(pstip);
+}
+
+
+
+/**
+ * Called by drivers that want to install this AA line prim stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for AA lines.
+ */
+void
+draw_install_pstipple_stage(struct draw_context *draw,
+                            struct pipe_context *pipe)
+{
+   struct pstip_stage *pstip;
+
+   /*
+    * Create / install AA line drawing / prim stage
+    */
+   pstip = draw_pstip_stage( draw );
+   assert(pstip);
+   draw->pipeline.pstipple = &pstip->stage;
+
+   pstip->pipe = pipe;
+
+   /* create special texture, sampler state */
+   pstip_create_texture(pstip);
+   pstip_create_sampler(pstip);
+
+   /* save original driver functions */
+   pstip->driver_create_fs_state = pipe->create_fs_state;
+   pstip->driver_bind_fs_state = pipe->bind_fs_state;
+   pstip->driver_delete_fs_state = pipe->delete_fs_state;
+
+   pstip->driver_bind_sampler_state = pipe->bind_sampler_state;
+   pstip->driver_set_sampler_texture = pipe->set_sampler_texture;
+   pstip->driver_set_polygon_stipple = pipe->set_polygon_stipple;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = pstip_create_fs_state;
+   pipe->bind_fs_state = pstip_bind_fs_state;
+   pipe->delete_fs_state = pstip_delete_fs_state;
+
+   pipe->bind_sampler_state = pstip_bind_sampler_state;
+   pipe->set_sampler_texture = pstip_set_sampler_texture;
+   pipe->set_polygon_stipple = pstip_set_polygon_stipple;
+
+   add_pstip_pipe_context(pipe, pstip);
+}
diff --git a/src/gallium/auxiliary/draw/draw_validate.c b/src/gallium/auxiliary/draw/draw_validate.c
index 4375ebabbc..efd6793f2b 100644
--- a/src/gallium/auxiliary/draw/draw_validate.c
+++ b/src/gallium/auxiliary/draw/draw_validate.c
@@ -58,7 +58,18 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
     * shorter pipelines for lines & points.
     */
 
-   if ((draw->rasterizer->line_width != 1.0 && draw->convert_wide_lines) ||
+   if (draw->rasterizer->line_smooth && draw->pipeline.aaline) {
+      draw->pipeline.aaline->next = next;
+      next = draw->pipeline.aaline;
+   }
+
+   if (draw->rasterizer->point_smooth && draw->pipeline.aapoint) {
+      draw->pipeline.aapoint->next = next;
+      next = draw->pipeline.aapoint;
+   }
+
+   if ((draw->rasterizer->line_width != 1.0 && draw->convert_wide_lines
+        && !draw->rasterizer->line_smooth) ||
        (draw->rasterizer->point_size != 1.0 && draw->convert_wide_points) ||
        draw->rasterizer->point_sprite) {
       draw->pipeline.wide->next = next;
@@ -71,6 +82,12 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
       precalc_flat = 1;		/* only needed for lines really */
    }
 
+   if (draw->rasterizer->poly_stipple_enable
+       && draw->pipeline.pstipple) {
+      draw->pipeline.pstipple->next = next;
+      next = draw->pipeline.pstipple;
+   }
+
    if (draw->rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
        draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL) {
       draw->pipeline.unfilled->next = next;
diff --git a/src/gallium/auxiliary/draw/draw_vertex_cache.c b/src/gallium/auxiliary/draw/draw_vertex_cache.c
index 44427999cc..53f8bbec44 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_cache.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_cache.c
@@ -41,7 +41,7 @@ void draw_vertex_cache_invalidate( struct draw_context *draw )
    assert(draw->vs.queue_nr == 0);
    assert(draw->vcache.referenced == 0);
 
-   memset(draw->vcache.idx, ~0, sizeof(draw->vcache.idx));
+//   memset(draw->vcache.idx, ~0, sizeof(draw->vcache.idx));
 }
 
 
@@ -62,43 +62,51 @@ static struct vertex_header *get_vertex( struct draw_context *draw,
    
    assert(slot < 32); /* so we don't exceed the bitfield size below */
 
-   /* Cache miss?
-    */
-   if (draw->vcache.idx[slot] != i) {
-
-      /* If slot is in use, use the overflow area:
+   if (draw->vcache.referenced & (1<<slot))
+   {
+      /* Cache hit?
        */
-      if (draw->vcache.referenced & (1 << slot)) {
-	 slot = VCACHE_SIZE + draw->vcache.overflow++;
+      if (draw->vcache.idx[slot].in == i) {
+//	 _mesa_printf("HIT %d %d\n", slot, i);
+	 assert(draw->vcache.idx[slot].out < draw->vs.queue_nr);
+	 return draw->vs.queue[draw->vcache.idx[slot].out].vertex;
       }
 
+      /* Otherwise a collision
+       */
+      slot = VCACHE_SIZE + draw->vcache.overflow++;
+//      _mesa_printf("XXX %d --> %d\n", i, slot);
+   }
+
+   /* Deal with the cache miss: 
+    */
+   {
+      unsigned out;
+      
       assert(slot < Elements(draw->vcache.idx));
 
-      draw->vcache.idx[slot] = i;
+//      _mesa_printf("NEW %d %d\n", slot, i);
+      draw->vcache.idx[slot].in = i;
+      draw->vcache.idx[slot].out = out = draw->vs.queue_nr++;
+      draw->vcache.referenced |= (1 << slot);
+
 
       /* Add to vertex shader queue:
        */
       assert(draw->vs.queue_nr < VS_QUEUE_LENGTH);
-      draw->vs.queue[draw->vs.queue_nr].dest = draw->vcache.vertex[slot];
-      draw->vs.queue[draw->vs.queue_nr].elt = i;
-      draw->vs.queue_nr++;
+
+      draw->vs.queue[out].elt = i;
+      draw->vs.queue[out].vertex->clipmask = 0;
+      draw->vs.queue[out].vertex->edgeflag = 1; /*XXX use user's edge flag! */
+      draw->vs.queue[out].vertex->pad = 0;
+      draw->vs.queue[out].vertex->vertex_id = UNDEFINED_VERTEX_ID;
 
       /* Need to set the vertex's edge flag here.  If we're being called
        * by do_ef_triangle(), that function needs edge flag info!
        */
-      draw->vcache.vertex[slot]->clipmask = 0;
-      draw->vcache.vertex[slot]->edgeflag = 1; /*XXX use user's edge flag! */
-      draw->vcache.vertex[slot]->pad = 0;
-      draw->vcache.vertex[slot]->vertex_id = UNDEFINED_VERTEX_ID;
-   }
 
-
-   /* primitive flushing may have cleared the bitfield but did not
-    * clear the idx[] array values.  Set the bit now.  This fixes a
-    * bug found when drawing long triangle fans.
-    */
-   draw->vcache.referenced |= (1 << slot);
-   return draw->vcache.vertex[slot];
+      return draw->vs.queue[draw->vcache.idx[slot].out].vertex;
+   }
 }
 
 
@@ -130,8 +138,8 @@ void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw )
 {
    unsigned i;
 
-   for (i = 0; i < Elements(draw->vcache.vertex); i++)
-      draw->vcache.vertex[i]->vertex_id = UNDEFINED_VERTEX_ID;
+   for (i = 0; i < draw->vs.post_nr; i++)
+      draw->vs.queue[i].vertex->vertex_id = UNDEFINED_VERTEX_ID;
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c
index f68f6e3244..5d2f5c9c43 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_shader.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c
@@ -55,7 +55,7 @@ draw_vertex_shader_queue_flush(struct draw_context *draw)
     */
    shader->prepare( shader, draw );
 
-//   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
+//   fprintf(stderr, "%s %d\n", __FUNCTION__, draw->vs.queue_nr );
 
    /* run vertex shader on vertex cache entries, four per invokation */
    for (i = 0; i < draw->vs.queue_nr; i += 4) {
@@ -65,12 +65,12 @@ draw_vertex_shader_queue_flush(struct draw_context *draw)
 
       for (j = 0; j < n; j++) {
          elts[j] = draw->vs.queue[i + j].elt;
-         dests[j] = draw->vs.queue[i + j].dest;
+         dests[j] = draw->vs.queue[i + j].vertex;
       }
 
       for ( ; j < 4; j++) {
 	 elts[j] = elts[0];
-	 dests[j] = dests[0];
+         dests[j] = draw->vs.queue[i + j].vertex;
       }
 
       assert(n > 0);
@@ -79,6 +79,7 @@ draw_vertex_shader_queue_flush(struct draw_context *draw)
       shader->run(shader, draw, elts, n, dests);
    }
 
+   draw->vs.post_nr = draw->vs.queue_nr;
    draw->vs.queue_nr = 0;
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_vf.c b/src/gallium/auxiliary/draw/draw_vf.c
index dc3a5ecd21..901ff20a7e 100644
--- a/src/gallium/auxiliary/draw/draw_vf.c
+++ b/src/gallium/auxiliary/draw/draw_vf.c
@@ -30,6 +30,7 @@
 
 #include "pipe/p_compiler.h"
 #include "pipe/p_util.h"
+#include "rtasm/rtasm_execmem.h"
 
 #include "draw_vf.h"
 
@@ -37,11 +38,6 @@
 #define DRAW_VF_DBG 0
 
 
-/* TODO: remove this */
-extern void 
-_mesa_exec_free( void *addr );
-
-
 static boolean match_fastpath( struct draw_vertex_fetch *vf,
 				 const struct draw_vf_fastpath *fp)
 {
@@ -414,12 +410,12 @@ void draw_vf_destroy( struct draw_vertex_fetch *vf )
       FREE(fp->attr);
 
       /* KW: At the moment, fp->func is constrained to be allocated by
-       * _mesa_exec_alloc(), as the hardwired fastpaths in
+       * rtasm_exec_alloc(), as the hardwired fastpaths in
        * t_vertex_generic.c are handled specially.  It would be nice
        * to unify them, but this probably won't change until this
        * module gets another overhaul.
        */
-      //_mesa_exec_free((void *) fp->func);
+      //rtasm_exec_free((void *) fp->func);
       FREE(fp);
    }
    
diff --git a/src/gallium/auxiliary/draw/draw_vf_sse.c b/src/gallium/auxiliary/draw/draw_vf_sse.c
index 1ad2ae756d..aff4ffd985 100644
--- a/src/gallium/auxiliary/draw/draw_vf_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vf_sse.c
@@ -26,17 +26,16 @@
  */
 
 
-#include "simple_list.h"
-
 #include "pipe/p_compiler.h"
+#include "util/u_simple_list.h"
 
 #include "draw_vf.h"
 
 
 #if defined(USE_SSE_ASM)
 
-#include "x86/rtasm/x86sse.h"
-#include "x86/common_x86_asm.h"
+#include "rtasm/rtasm_cpu.h"
+#include "rtasm/rtasm_x86sse.h"
 
 
 #define X    0
@@ -576,7 +575,7 @@ void draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf )
 {
    struct x86_program p;   
 
-   if (!cpu_has_xmm) {
+   if (!rtasm_cpu_has_sse()) {
       vf->codegen_emit = NULL;
       return;
    }
@@ -586,7 +585,7 @@ void draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf )
    p.vf = vf;
    p.inputs_safe = 0;		/* for now */
    p.outputs_safe = 1;		/* for now */
-   p.have_sse2 = cpu_has_xmm2;
+   p.have_sse2 = rtasm_cpu_has_sse2();
    p.identity = x86_make_reg(file_XMM, 6);
    p.chan0 = x86_make_reg(file_XMM, 7);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 44022b6e07..0fd557d667 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -40,7 +40,7 @@
 
 #ifdef MESA_LLVM
 
-#include "llvm/gallivm.h"
+#include "gallivm/gallivm.h"
 
 struct draw_llvm_vertex_shader {
    struct draw_vertex_shader base;
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 27bc66812c..11ef0c503d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -41,7 +41,7 @@
 #include "draw_private.h"
 #include "draw_context.h"
 
-#include "x86/rtasm/x86sse.h"
+#include "rtasm/rtasm_x86sse.h"
 #include "tgsi/exec/tgsi_sse2.h"
 
 
diff --git a/src/gallium/auxiliary/draw/draw_wide_prims.c b/src/gallium/auxiliary/draw/draw_wide_prims.c
index 655774b155..1f8069bdca 100644
--- a/src/gallium/auxiliary/draw/draw_wide_prims.c
+++ b/src/gallium/auxiliary/draw/draw_wide_prims.c
@@ -162,70 +162,6 @@ static void wide_line( struct draw_stage *stage,
 
 
 /**
- * Draw a wide line by drawing a quad, using geometry which will
- * fullfill GL's antialiased line requirements.
- */
-static void wide_line_aa(struct draw_stage *stage,
-                         struct prim_header *header)
-{
-   const struct wide_stage *wide = wide_stage(stage);
-   const float half_width = wide->half_line_width;
-   struct prim_header tri;
-   struct vertex_header *v[4];
-   float *pos;
-   float dx = header->v[1]->data[0][0] - header->v[0]->data[0][0];
-   float dy = header->v[1]->data[0][1] - header->v[0]->data[0][1];
-   const float len = (float) sqrt(dx * dx + dy * dy);
-   uint i;
-
-   dx = dx * half_width / len;
-   dy = dy * half_width / len;
-
-   /* allocate/dup new verts */
-   for (i = 0; i < 4; i++) {
-      v[i] = dup_vert(stage, header->v[i/2], i);
-   }
-
-   /*
-    * Quad for line from v0 to v1:
-    *
-    *  1                         3
-    *  +-------------------------+
-    *  |                         |
-    *  *v0                     v1*
-    *  |                         |
-    *  +-------------------------+
-    *  0                         2
-    */
-
-   pos = v[0]->data[0];
-   pos[0] += dy;
-   pos[1] -= dx;
-
-   pos = v[1]->data[0];
-   pos[0] -= dy;
-   pos[1] += dx;
-
-   pos = v[2]->data[0];
-   pos[0] += dy;
-   pos[1] -= dx;
-
-   pos = v[3]->data[0];
-   pos[0] -= dy;
-   pos[1] += dx;
-
-   tri.det = header->det;  /* only the sign matters */
-
-   tri.v[0] = v[2];  tri.v[1] = v[1];  tri.v[2] = v[0];
-   stage->next->tri( stage->next, &tri );
-
-   tri.v[0] = v[3];  tri.v[1] = v[1];  tri.v[2] = v[2];
-   stage->next->tri( stage->next, &tri );
-
-}
-
-
-/**
  * Set the vertex texcoords for sprite mode.
  * Coords may be left untouched or set to a right-side-up or upside-down
  * orientation.
@@ -379,10 +315,7 @@ static void wide_first_line( struct draw_stage *stage,
    wide->half_line_width = 0.5f * draw->rasterizer->line_width;
 
    if (draw->rasterizer->line_width != 1.0) {
-      if (draw->rasterizer->line_smooth)
-         wide->stage.line = wide_line_aa;
-      else
-         wide->stage.line = wide_line;
+      wide->stage.line = wide_line;
    }
    else {
       wide->stage.line = passthrough_line;
diff --git a/src/gallium/auxiliary/llvm/Makefile b/src/gallium/auxiliary/gallivm/Makefile
index 39fac6ea4a..39fac6ea4a 100644
--- a/src/gallium/auxiliary/llvm/Makefile
+++ b/src/gallium/auxiliary/gallivm/Makefile
diff --git a/src/gallium/auxiliary/gallivm/SConscript b/src/gallium/auxiliary/gallivm/SConscript
new file mode 100644
index 0000000000..c0aa51b90a
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/SConscript
@@ -0,0 +1,16 @@
+Import('*')
+
+gallivm = env.ConvenienceLibrary(
+	target = 'gallivm',
+	source = [
+        'gallivm.cpp',
+        'gallivm_cpu.cpp',
+        'instructions.cpp',
+        'loweringpass.cpp',
+        'tgsitollvm.cpp',
+        'storage.cpp',
+        'storagesoa.cpp',
+        'instructionssoa.cpp',
+	])
+
+auxiliaries.insert(0, gallivm)
diff --git a/src/gallium/auxiliary/llvm/gallivm.cpp b/src/gallium/auxiliary/gallivm/gallivm.cpp
index d14bb3b99a..d14bb3b99a 100644
--- a/src/gallium/auxiliary/llvm/gallivm.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm.cpp
diff --git a/src/gallium/auxiliary/llvm/gallivm.h b/src/gallium/auxiliary/gallivm/gallivm.h
index 92da4bca7f..92da4bca7f 100644
--- a/src/gallium/auxiliary/llvm/gallivm.h
+++ b/src/gallium/auxiliary/gallivm/gallivm.h
diff --git a/src/gallium/auxiliary/llvm/gallivm_builtins.cpp b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
index 1796f0a177..1796f0a177 100644
--- a/src/gallium/auxiliary/llvm/gallivm_builtins.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
diff --git a/src/gallium/auxiliary/llvm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
index 8f9830d0b1..8f9830d0b1 100644
--- a/src/gallium/auxiliary/llvm/gallivm_cpu.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
diff --git a/src/gallium/auxiliary/llvm/gallivm_p.h b/src/gallium/auxiliary/gallivm/gallivm_p.h
index cfe7b1901b..cfe7b1901b 100644
--- a/src/gallium/auxiliary/llvm/gallivm_p.h
+++ b/src/gallium/auxiliary/gallivm/gallivm_p.h
diff --git a/src/gallium/auxiliary/llvm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp
index 55d39fa5f1..55d39fa5f1 100644
--- a/src/gallium/auxiliary/llvm/instructions.cpp
+++ b/src/gallium/auxiliary/gallivm/instructions.cpp
diff --git a/src/gallium/auxiliary/llvm/instructions.h b/src/gallium/auxiliary/gallivm/instructions.h
index 9ebc17dd8e..9ebc17dd8e 100644
--- a/src/gallium/auxiliary/llvm/instructions.h
+++ b/src/gallium/auxiliary/gallivm/instructions.h
diff --git a/src/gallium/auxiliary/llvm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
index a4d5046637..a4d5046637 100644
--- a/src/gallium/auxiliary/llvm/instructionssoa.cpp
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
diff --git a/src/gallium/auxiliary/llvm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h
index 4169dcbb2e..4169dcbb2e 100644
--- a/src/gallium/auxiliary/llvm/instructionssoa.h
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.h
diff --git a/src/gallium/auxiliary/llvm/llvm_builtins.c b/src/gallium/auxiliary/gallivm/llvm_builtins.c
index 4f98d754ba..4f98d754ba 100644
--- a/src/gallium/auxiliary/llvm/llvm_builtins.c
+++ b/src/gallium/auxiliary/gallivm/llvm_builtins.c
diff --git a/src/gallium/auxiliary/llvm/loweringpass.cpp b/src/gallium/auxiliary/gallivm/loweringpass.cpp
index 556dbec366..556dbec366 100644
--- a/src/gallium/auxiliary/llvm/loweringpass.cpp
+++ b/src/gallium/auxiliary/gallivm/loweringpass.cpp
diff --git a/src/gallium/auxiliary/llvm/loweringpass.h b/src/gallium/auxiliary/gallivm/loweringpass.h
index f62dcf6ba7..f62dcf6ba7 100644
--- a/src/gallium/auxiliary/llvm/loweringpass.h
+++ b/src/gallium/auxiliary/gallivm/loweringpass.h
diff --git a/src/gallium/auxiliary/llvm/storage.cpp b/src/gallium/auxiliary/gallivm/storage.cpp
index c4326de8c5..c4326de8c5 100644
--- a/src/gallium/auxiliary/llvm/storage.cpp
+++ b/src/gallium/auxiliary/gallivm/storage.cpp
diff --git a/src/gallium/auxiliary/llvm/storage.h b/src/gallium/auxiliary/gallivm/storage.h
index 8574f7554e..8574f7554e 100644
--- a/src/gallium/auxiliary/llvm/storage.h
+++ b/src/gallium/auxiliary/gallivm/storage.h
diff --git a/src/gallium/auxiliary/llvm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp
index ed0674a96f..ed0674a96f 100644
--- a/src/gallium/auxiliary/llvm/storagesoa.cpp
+++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp
diff --git a/src/gallium/auxiliary/llvm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h
index 6443351f27..6443351f27 100644
--- a/src/gallium/auxiliary/llvm/storagesoa.h
+++ b/src/gallium/auxiliary/gallivm/storagesoa.h
diff --git a/src/gallium/auxiliary/llvm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
index 2cb4acce32..2cb4acce32 100644
--- a/src/gallium/auxiliary/llvm/tgsitollvm.cpp
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
diff --git a/src/gallium/auxiliary/llvm/tgsitollvm.h b/src/gallium/auxiliary/gallivm/tgsitollvm.h
index 7ada04d629..7ada04d629 100644
--- a/src/gallium/auxiliary/llvm/tgsitollvm.h
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.h
diff --git a/src/gallium/auxiliary/pipebuffer/Makefile b/src/gallium/auxiliary/pipebuffer/Makefile
index 588629e870..a9fa518c67 100644
--- a/src/gallium/auxiliary/pipebuffer/Makefile
+++ b/src/gallium/auxiliary/pipebuffer/Makefile
@@ -1,10 +1,9 @@
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
 LIBNAME = pipebuffer
 
-DRIVER_SOURCES = \
+C_SOURCES = \
 	pb_buffer_fenced.c \
 	pb_buffer_malloc.c \
 	pb_bufmgr_fenced.c \
@@ -12,11 +11,6 @@ DRIVER_SOURCES = \
 	pb_bufmgr_pool.c \
 	pb_winsys.c
 
-C_SOURCES = \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES = 
-
 include ../../Makefile.template
 
 symlinks:
diff --git a/src/gallium/auxiliary/pipebuffer/SConscript b/src/gallium/auxiliary/pipebuffer/SConscript
new file mode 100644
index 0000000000..3d41fd84a6
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/SConscript
@@ -0,0 +1,14 @@
+Import('*')
+
+pipebuffer = env.ConvenienceLibrary(
+	target = 'pipebuffer',
+	source = [
+		'pb_buffer_fenced.c',
+		'pb_buffer_malloc.c',
+		'pb_bufmgr_fenced.c',
+		'pb_bufmgr_mm.c',
+		'pb_bufmgr_pool.c',
+		'pb_winsys.c',
+	])
+
+auxiliaries.insert(0, pipebuffer)
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index f4fc3f6d71..e2ee72ed1f 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -34,13 +34,12 @@
  */
 
 
-#include "linked_list.h"
-
-#include "p_compiler.h"
-#include "p_debug.h"
-#include "p_winsys.h"
-#include "p_thread.h"
-#include "p_util.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_winsys.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_util.h"
+#include "util/u_double_list.h"
 
 #include "pb_buffer.h"
 #include "pb_buffer_fenced.h"
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
index c535d3276c..bffca5b244 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
@@ -34,8 +34,8 @@
  */
 
 
-#include "p_debug.h"
-#include "p_util.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_util.h"
 
 #include "pb_buffer.h"
 #include "pb_buffer_fenced.h"
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
index 8b1b51c0e2..ff4fd123f3 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
@@ -1,7 +1,6 @@
 /**************************************************************************
  *
  * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
- * Copyright 1999 Wittawat Yamwong
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -34,12 +33,12 @@
  */
 
 
-#include "linked_list.h"
-
-#include "p_defines.h"
-#include "p_debug.h"
-#include "p_thread.h"
-#include "p_util.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_util.h"
+#include "util/u_double_list.h"
+#include "util/u_mm.h"
 #include "pb_buffer.h"
 #include "pb_bufmgr.h"
 
@@ -50,306 +49,6 @@
 #define SUPER(__derived) (&(__derived)->base)
 
 
-struct mem_block 
-{
-   struct mem_block *next, *prev;
-   struct mem_block *next_free, *prev_free;
-   struct mem_block *heap;
-   int ofs, size;
-   unsigned int free:1;
-   unsigned int reserved:1;
-};
-
-
-#ifdef DEBUG
-/**
- * For debugging purposes.
- */
-static void
-mmDumpMemInfo(const struct mem_block *heap)
-{
-   debug_printf("Memory heap %p:\n", (void *)heap);
-   if (heap == 0) {
-      debug_printf("  heap == 0\n");
-   } else {
-      const struct mem_block *p;
-
-      for(p = heap->next; p != heap; p = p->next) {
-	 debug_printf("  Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
-		 p->free ? 'F':'.',
-		 p->reserved ? 'R':'.');
-      }
-
-      debug_printf("\nFree list:\n");
-
-      for(p = heap->next_free; p != heap; p = p->next_free) {
-	 debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
-		 p->free ? 'F':'.',
-		 p->reserved ? 'R':'.');
-      }
-
-   }
-   debug_printf("End of memory blocks\n");
-}
-#endif
-
-
-/** 
- * input: total size in bytes
- * return: a heap pointer if OK, NULL if error
- */
-static struct mem_block *
-mmInit(int ofs, int size)
-{
-   struct mem_block *heap, *block;
-  
-   if (size <= 0) 
-      return NULL;
-
-   heap = CALLOC_STRUCT(mem_block);
-   if (!heap) 
-      return NULL;
-   
-   block = CALLOC_STRUCT(mem_block);
-   if (!block) {
-      FREE(heap);
-      return NULL;
-   }
-
-   heap->next = block;
-   heap->prev = block;
-   heap->next_free = block;
-   heap->prev_free = block;
-
-   block->heap = heap;
-   block->next = heap;
-   block->prev = heap;
-   block->next_free = heap;
-   block->prev_free = heap;
-
-   block->ofs = ofs;
-   block->size = size;
-   block->free = 1;
-
-   return heap;
-}
-
-
-static struct mem_block *
-SliceBlock(struct mem_block *p, 
-           int startofs, int size, 
-           int reserved, int alignment)
-{
-   struct mem_block *newblock;
-
-   /* break left  [p, newblock, p->next], then p = newblock */
-   if (startofs > p->ofs) {
-      newblock = CALLOC_STRUCT(mem_block);
-      if (!newblock)
-	 return NULL;
-      newblock->ofs = startofs;
-      newblock->size = p->size - (startofs - p->ofs);
-      newblock->free = 1;
-      newblock->heap = p->heap;
-
-      newblock->next = p->next;
-      newblock->prev = p;
-      p->next->prev = newblock;
-      p->next = newblock;
-
-      newblock->next_free = p->next_free;
-      newblock->prev_free = p;
-      p->next_free->prev_free = newblock;
-      p->next_free = newblock;
-
-      p->size -= newblock->size;
-      p = newblock;
-   }
-
-   /* break right, also [p, newblock, p->next] */
-   if (size < p->size) {
-      newblock = CALLOC_STRUCT(mem_block);
-      if (!newblock)
-	 return NULL;
-      newblock->ofs = startofs + size;
-      newblock->size = p->size - size;
-      newblock->free = 1;
-      newblock->heap = p->heap;
-
-      newblock->next = p->next;
-      newblock->prev = p;
-      p->next->prev = newblock;
-      p->next = newblock;
-
-      newblock->next_free = p->next_free;
-      newblock->prev_free = p;
-      p->next_free->prev_free = newblock;
-      p->next_free = newblock;
-	 
-      p->size = size;
-   }
-
-   /* p = middle block */
-   p->free = 0;
-
-   /* Remove p from the free list: 
-    */
-   p->next_free->prev_free = p->prev_free;
-   p->prev_free->next_free = p->next_free;
-
-   p->next_free = 0;
-   p->prev_free = 0;
-
-   p->reserved = reserved;
-   return p;
-}
-
-
-/**
- * Allocate 'size' bytes with 2^align2 bytes alignment,
- * restrict the search to free memory after 'startSearch'
- * depth and back buffers should be in different 4mb banks
- * to get better page hits if possible
- * input:	size = size of block
- *       	align2 = 2^align2 bytes alignment
- *		startSearch = linear offset from start of heap to begin search
- * return: pointer to the allocated block, 0 if error
- */
-static struct mem_block *
-mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch)
-{
-   struct mem_block *p;
-   const int mask = (1 << align2)-1;
-   int startofs = 0;
-   int endofs;
-
-   if (!heap || align2 < 0 || size <= 0)
-      return NULL;
-
-   for (p = heap->next_free; p != heap; p = p->next_free) {
-      assert(p->free);
-
-      startofs = (p->ofs + mask) & ~mask;
-      if ( startofs < startSearch ) {
-	 startofs = startSearch;
-      }
-      endofs = startofs+size;
-      if (endofs <= (p->ofs+p->size))
-	 break;
-   }
-
-   if (p == heap) 
-      return NULL;
-
-   assert(p->free);
-   p = SliceBlock(p,startofs,size,0,mask+1);
-
-   return p;
-}
-
-
-#if 0
-/**
- * Free block starts at offset
- * input: pointer to a heap, start offset
- * return: pointer to a block
- */
-static struct mem_block *
-mmFindBlock(struct mem_block *heap, int start)
-{
-   struct mem_block *p;
-
-   for (p = heap->next; p != heap; p = p->next) {
-      if (p->ofs == start) 
-	 return p;
-   }
-
-   return NULL;
-}
-#endif
-
-
-static INLINE int
-Join2Blocks(struct mem_block *p)
-{
-   /* XXX there should be some assertions here */
-
-   /* NOTE: heap->free == 0 */
-
-   if (p->free && p->next->free) {
-      struct mem_block *q = p->next;
-
-      assert(p->ofs + p->size == q->ofs);
-      p->size += q->size;
-
-      p->next = q->next;
-      q->next->prev = p;
-
-      q->next_free->prev_free = q->prev_free; 
-      q->prev_free->next_free = q->next_free;
-     
-      FREE(q);
-      return 1;
-   }
-   return 0;
-}
-
-
-/**
- * Free block starts at offset
- * input: pointer to a block
- * return: 0 if OK, -1 if error
- */
-static int
-mmFreeMem(struct mem_block *b)
-{
-   if (!b)
-      return 0;
-
-   if (b->free) {
-      debug_printf("block already free\n");
-      return -1;
-   }
-   if (b->reserved) {
-      debug_printf("block is reserved\n");
-      return -1;
-   }
-
-   b->free = 1;
-   b->next_free = b->heap->next_free;
-   b->prev_free = b->heap;
-   b->next_free->prev_free = b;
-   b->prev_free->next_free = b;
-
-   Join2Blocks(b);
-   if (b->prev != b->heap)
-      Join2Blocks(b->prev);
-
-   return 0;
-}
-
-
-/**
- * destroy MM
- */
-static void
-mmDestroy(struct mem_block *heap)
-{
-   struct mem_block *p;
-
-   if (!heap)
-      return;
-
-   for (p = heap->next; p != heap; ) {
-      struct mem_block *next = p->next;
-      FREE(p);
-      p = next;
-   }
-
-   FREE(heap);
-}
-
-
 struct mm_pb_manager
 {
    struct pb_manager base;
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
index 04477a865a..528e9528f6 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
@@ -35,13 +35,12 @@
  */
 
 
-#include "linked_list.h"
-
-#include "p_compiler.h"
-#include "p_debug.h"
-#include "p_thread.h"
-#include "p_defines.h"
-#include "p_util.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_util.h"
+#include "util/u_double_list.h"
 
 #include "pb_buffer.h"
 #include "pb_bufmgr.h"
diff --git a/src/gallium/auxiliary/rtasm/Makefile b/src/gallium/auxiliary/rtasm/Makefile
new file mode 100644
index 0000000000..39b8a4dbd7
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/Makefile
@@ -0,0 +1,15 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = rtasm
+
+C_SOURCES = \
+	rtasm_cpu.c \
+	rtasm_execmem.c \
+	rtasm_x86sse.c \
+	rtasm_ppc_spe.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/rtasm/SConscript b/src/gallium/auxiliary/rtasm/SConscript
new file mode 100644
index 0000000000..8ea25922aa
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/SConscript
@@ -0,0 +1,12 @@
+Import('*')
+
+rtasm = env.ConvenienceLibrary(
+	target = 'rtasm',
+	source = [
+		'rtasm_cpu.c',
+		'rtasm_execmem.c',
+		'rtasm_x86sse.c',
+		'rtasm_ppc_spe.c',
+	])
+
+auxiliaries.insert(0, rtasm)
diff --git a/src/gallium/drivers/cell/ppu/cell_state_sampler.c b/src/gallium/auxiliary/rtasm/rtasm_cpu.c
index a33421a4ad..eb3359750b 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_sampler.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_cpu.c
@@ -1,6 +1,6 @@
 /**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -10,11 +10,11 @@
  * 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.
@@ -22,63 +22,29 @@
  * 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:
- *  Brian Paul
- */
-
-#include "pipe/p_util.h"
-#include "draw/draw_context.h"
-#include "cell_context.h"
-#include "cell_state.h"
-#include "cell_texture.h"
-
-
-void *
-cell_create_sampler_state(struct pipe_context *pipe,
-                          const struct pipe_sampler_state *sampler)
-{
-   return mem_dup(sampler, sizeof(*sampler));
-}
-
-void
-cell_bind_sampler_state(struct pipe_context *pipe,
-                            unsigned unit, void *sampler)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   draw_flush(cell->draw);
 
-   assert(unit < PIPE_MAX_SAMPLERS);
-   cell->sampler[unit] = (struct pipe_sampler_state *)sampler;
+#include "rtasm_cpu.h"
 
-   cell->dirty |= CELL_NEW_SAMPLER;
-}
 
-
-void
-cell_delete_sampler_state(struct pipe_context *pipe,
-                              void *sampler)
+int rtasm_cpu_has_sse(void)
 {
-   FREE( sampler );
+   /* FIXME: actually detect this at run-time */
+#if defined(__i386__) || defined(__386__)
+   return 1;
+#else
+   return 0;
+#endif
 }
 
-
-
-void
-cell_set_sampler_texture(struct pipe_context *pipe,
-                         unsigned sampler,
-                         struct pipe_texture *texture)
+int rtasm_cpu_has_sse2(void) 
 {
-   struct cell_context *cell = cell_context(pipe);
-
-   draw_flush(cell->draw);
-
-   cell->texture[sampler] = texture;
-
-   cell_update_texture_mapping(cell);
-
-   cell->dirty |= CELL_NEW_TEXTURE;
+   /* FIXME: actually detect this at run-time */
+#if defined(__i386__) || defined(__386__)
+   return 1;
+#else
+   return 0;
+#endif
 }
diff --git a/src/gallium/auxiliary/rtasm/rtasm_cpu.h b/src/gallium/auxiliary/rtasm/rtasm_cpu.h
new file mode 100644
index 0000000000..ebc71634fd
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_cpu.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ *
+ * 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Runtime detection of CPU capabilities.
+ */
+
+#ifndef _RTASM_CPU_H_
+#define _RTASM_CPU_H_
+
+
+int rtasm_cpu_has_sse(void);
+
+int rtasm_cpu_has_sse2(void);
+
+
+#endif /* _RTASM_CPU_H_ */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_execmem.c b/src/gallium/auxiliary/rtasm/rtasm_execmem.c
new file mode 100644
index 0000000000..300c1c2d9d
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_execmem.c
@@ -0,0 +1,135 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_util.h"
+
+#include "rtasm_execmem.h"
+
+
+#if defined(__linux__)
+
+/*
+ * Allocate a large block of memory which can hold code then dole it out
+ * in pieces by means of the generic memory manager code.
+*/
+
+#include <unistd.h>
+#include <sys/mman.h>
+#include "util/u_mm.h"
+
+#define EXEC_HEAP_SIZE (10*1024*1024)
+
+_glthread_DECLARE_STATIC_MUTEX(exec_mutex);
+
+static struct mem_block *exec_heap = NULL;
+static unsigned char *exec_mem = NULL;
+
+
+static void
+init_heap(void)
+{
+   if (!exec_heap)
+      exec_heap = mmInit( 0, EXEC_HEAP_SIZE );
+   
+   if (!exec_mem)
+      exec_mem = (unsigned char *) mmap(0, EXEC_HEAP_SIZE, 
+					PROT_EXEC | PROT_READ | PROT_WRITE, 
+					MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+}
+
+
+void *
+rtasm_exec_malloc(size_t size)
+{
+   struct mem_block *block = NULL;
+   void *addr = NULL;
+
+   _glthread_LOCK_MUTEX(exec_mutex);
+
+   init_heap();
+
+   if (exec_heap) {
+      size = (size + 31) & ~31;
+      block = mmAllocMem( exec_heap, size, 32, 0 );
+   }
+
+   if (block)
+      addr = exec_mem + block->ofs;
+   else 
+      debug_printf("rtasm_exec_malloc failed\n");
+   
+   _glthread_UNLOCK_MUTEX(exec_mutex);
+   
+   return addr;
+}
+
+ 
+void 
+rtasm_exec_free(void *addr)
+{
+   _glthread_LOCK_MUTEX(exec_mutex);
+
+   if (exec_heap) {
+      struct mem_block *block = mmFindBlock(exec_heap, (unsigned char *)addr - exec_mem);
+   
+      if (block)
+	 mmFreeMem(block);
+   }
+
+   _glthread_UNLOCK_MUTEX(exec_mutex);
+}
+
+
+#else
+
+/*
+ * Just use regular memory.
+ */
+
+void *
+rtasm_exec_malloc(size_t size)
+{
+   return MALLOC( size );
+}
+
+ 
+void 
+rtasm_exec_free(void *addr)
+{
+   FREE(addr);
+}
+
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/rtasm_execmem.h b/src/gallium/auxiliary/rtasm/rtasm_execmem.h
new file mode 100644
index 0000000000..155c6d34e0
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_execmem.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+#ifndef _RTASM_EXECMEM_H_
+#define _RTASM_EXECMEM_H_
+
+#include "pipe/p_compiler.h"
+
+
+extern void *
+rtasm_exec_malloc( size_t size );
+
+
+extern void 
+rtasm_exec_free( void *addr );
+
+
+#endif
diff --git a/src/mesa/ppc/rtasm/spe_asm.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index 1037637250..95a2d6fcbb 100644
--- a/src/mesa/ppc/rtasm/spe_asm.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -23,16 +23,17 @@
  */
 
 /**
- * \file spe_asm.c
+ * \file
  * Real-time assembly generation interface for Cell B.E. SPEs.
  *
  * \author Ian Romanick <idr@us.ibm.com>
  */
-#ifdef GALLIUM_CELL
-#include <inttypes.h>
-#include <imports.h>
-#include "spe_asm.h"
 
+#include "pipe/p_compiler.h"
+#include "pipe/p_util.h"
+#include "rtasm_ppc_spe.h"
+
+#ifdef GALLIUM_CELL
 /**
  * SPE instruction types
  *
@@ -296,21 +297,21 @@ void _name (struct spe_function *p, int imm) \
     emit_RI16(p, _op, 0, imm); \
 }
 
-#include "spe_asm.h"
+#include "rtasm_ppc_spe.h"
 
 
 /*
  */
 void spe_init_func(struct spe_function *p, unsigned code_size)
 {
-    p->store = _mesa_align_malloc(code_size, 16);
+    p->store = align_malloc(code_size, 16);
     p->csr = p->store;
 }
 
 
 void spe_release_func(struct spe_function *p)
 {
-    _mesa_align_free(p->store);
+    align_free(p->store);
     p->store = NULL;
     p->csr = NULL;
 }
diff --git a/src/mesa/ppc/rtasm/spe_asm.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
index 6d69ae655d..10ce44b3a0 100644
--- a/src/mesa/ppc/rtasm/spe_asm.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
@@ -23,14 +23,14 @@
  */
 
 /**
- * \file spe_asm.h
+ * \file
  * Real-time assembly generation interface for Cell B.E. SPEs.
  *
  * \author Ian Romanick <idr@us.ibm.com>
  */
 
-#ifndef SPE_ASM_H
-#define SPE_ASM_H
+#ifndef RTASM_PPC_SPE_H
+#define RTASM_PPC_SPE_H
 
 struct spe_function {
     /**
@@ -44,7 +44,7 @@ struct spe_function {
 extern void spe_init_func(struct spe_function *p, unsigned code_size);
 extern void spe_release_func(struct spe_function *p);
 
-#endif /* SPE_ASM_H */
+#endif /* RTASM_PPC_SPE_H */
 
 #ifndef EMIT_
 #define EMIT_(name, _op) \
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
new file mode 100644
index 0000000000..dcbf76f600
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -0,0 +1,1219 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#if defined(__i386__) || defined(__386__) || defined(i386)
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "rtasm_execmem.h"
+#include "rtasm_x86sse.h"
+
+#define DISASSEM 0
+#define X86_TWOB 0x0f
+
+static unsigned char *cptr( void (*label)() )
+{
+   return (unsigned char *)(unsigned long)label;
+}
+
+
+static void do_realloc( struct x86_function *p )
+{
+   if (p->size == 0) {
+      p->size = 1024;
+      p->store = rtasm_exec_malloc(p->size);
+      p->csr = p->store;
+   }
+   else {
+      unsigned used = p->csr - p->store;
+      unsigned char *tmp = p->store;
+      p->size *= 2;
+      p->store = rtasm_exec_malloc(p->size);
+      memcpy(p->store, tmp, used);
+      p->csr = p->store + used;
+      rtasm_exec_free(tmp);
+   }
+}
+
+/* Emit bytes to the instruction stream:
+ */
+static unsigned char *reserve( struct x86_function *p, int bytes )
+{
+   if (p->csr + bytes - p->store > p->size)
+      do_realloc(p);
+
+   {
+      unsigned char *csr = p->csr;
+      p->csr += bytes;
+      return csr;
+   }
+}
+
+
+
+static void emit_1b( struct x86_function *p, char b0 )
+{
+   char *csr = (char *)reserve(p, 1);
+   *csr = b0;
+}
+
+static void emit_1i( struct x86_function *p, int i0 )
+{
+   int *icsr = (int *)reserve(p, sizeof(i0));
+   *icsr = i0;
+}
+
+static void emit_1ub( struct x86_function *p, unsigned char b0 )
+{
+   unsigned char *csr = reserve(p, 1);
+   *csr++ = b0;
+}
+
+static void emit_2ub( struct x86_function *p, unsigned char b0, unsigned char b1 )
+{
+   unsigned char *csr = reserve(p, 2);
+   *csr++ = b0;
+   *csr++ = b1;
+}
+
+static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1, unsigned char b2 )
+{
+   unsigned char *csr = reserve(p, 3);
+   *csr++ = b0;
+   *csr++ = b1;
+   *csr++ = b2;
+}
+
+
+/* Build a modRM byte + possible displacement.  No treatment of SIB
+ * indexing.  BZZT - no way to encode an absolute address.
+ */
+static void emit_modrm( struct x86_function *p, 
+			struct x86_reg reg, 
+			struct x86_reg regmem )
+{
+   unsigned char val = 0;
+   
+   assert(reg.mod == mod_REG);
+   
+   val |= regmem.mod << 6;     	/* mod field */
+   val |= reg.idx << 3;		/* reg field */
+   val |= regmem.idx;		/* r/m field */
+   
+   emit_1ub(p, val);
+
+   /* Oh-oh we've stumbled into the SIB thing.
+    */
+   if (regmem.file == file_REG32 &&
+       regmem.idx == reg_SP) {
+      emit_1ub(p, 0x24);		/* simplistic! */
+   }
+
+   switch (regmem.mod) {
+   case mod_REG:
+   case mod_INDIRECT:
+      break;
+   case mod_DISP8:
+      emit_1b(p, regmem.disp);
+      break;
+   case mod_DISP32:
+      emit_1i(p, regmem.disp);
+      break;
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+static void emit_modrm_noreg( struct x86_function *p,
+			      unsigned op,
+			      struct x86_reg regmem )
+{
+   struct x86_reg dummy = x86_make_reg(file_REG32, op);
+   emit_modrm(p, dummy, regmem);
+}
+
+/* Many x86 instructions have two opcodes to cope with the situations
+ * where the destination is a register or memory reference
+ * respectively.  This function selects the correct opcode based on
+ * the arguments presented.
+ */
+static void emit_op_modrm( struct x86_function *p,
+			   unsigned char op_dst_is_reg, 
+			   unsigned char op_dst_is_mem,
+			   struct x86_reg dst,
+			   struct x86_reg src )
+{  
+   switch (dst.mod) {
+   case mod_REG:
+      emit_1ub(p, op_dst_is_reg);
+      emit_modrm(p, dst, src);
+      break;
+   case mod_INDIRECT:
+   case mod_DISP32:
+   case mod_DISP8:
+      assert(src.mod == mod_REG);
+      emit_1ub(p, op_dst_is_mem);
+      emit_modrm(p, src, dst);
+      break;
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+
+
+
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+			     enum x86_reg_name idx )
+{
+   struct x86_reg reg;
+
+   reg.file = file;
+   reg.idx = idx;
+   reg.mod = mod_REG;
+   reg.disp = 0;
+
+   return reg;
+}
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+			      int disp )
+{
+   assert(reg.file == file_REG32);
+
+   if (reg.mod == mod_REG)
+      reg.disp = disp;
+   else
+      reg.disp += disp;
+
+   if (reg.disp == 0)
+      reg.mod = mod_INDIRECT;
+   else if (reg.disp <= 127 && reg.disp >= -128)
+      reg.mod = mod_DISP8;
+   else
+      reg.mod = mod_DISP32;
+
+   return reg;
+}
+
+struct x86_reg x86_deref( struct x86_reg reg )
+{
+   return x86_make_disp(reg, 0);
+}
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg )
+{
+   return x86_make_reg( reg.file, reg.idx );
+}
+
+unsigned char *x86_get_label( struct x86_function *p )
+{
+   return p->csr;
+}
+
+
+
+/***********************************************************************
+ * x86 instructions
+ */
+
+
+void x86_jcc( struct x86_function *p,
+	      enum x86_cc cc,
+	      unsigned char *label )
+{
+   int offset = label - (x86_get_label(p) + 2);
+   
+   if (offset <= 127 && offset >= -128) {
+      emit_1ub(p, 0x70 + cc);
+      emit_1b(p, (char) offset);
+   }
+   else {
+      offset = label - (x86_get_label(p) + 6);
+      emit_2ub(p, 0x0f, 0x80 + cc);
+      emit_1i(p, offset);
+   }
+}
+
+/* Always use a 32bit offset for forward jumps:
+ */
+unsigned char *x86_jcc_forward( struct x86_function *p,
+			  enum x86_cc cc )
+{
+   emit_2ub(p, 0x0f, 0x80 + cc);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+unsigned char *x86_jmp_forward( struct x86_function *p)
+{
+   emit_1ub(p, 0xe9);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+unsigned char *x86_call_forward( struct x86_function *p)
+{
+   emit_1ub(p, 0xe8);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+/* Fixup offset from forward jump:
+ */
+void x86_fixup_fwd_jump( struct x86_function *p,
+			 unsigned char *fixup )
+{
+   *(int *)(fixup - 4) = x86_get_label(p) - fixup;
+}
+
+void x86_jmp( struct x86_function *p, unsigned char *label)
+{
+   emit_1ub(p, 0xe9);
+   emit_1i(p, label - x86_get_label(p) - 4);
+}
+
+#if 0
+/* This doesn't work once we start reallocating & copying the
+ * generated code on buffer fills, because the call is relative to the
+ * current pc.
+ */
+void x86_call( struct x86_function *p, void (*label)())
+{
+   emit_1ub(p, 0xe8);
+   emit_1i(p, cptr(label) - x86_get_label(p) - 4);
+}
+#else
+void x86_call( struct x86_function *p, struct x86_reg reg)
+{
+   emit_1ub(p, 0xff);
+   emit_modrm(p, reg, reg);
+}
+#endif
+
+
+/* michal:
+ * Temporary. As I need immediate operands, and dont want to mess with the codegen,
+ * I load the immediate into general purpose register and use it.
+ */
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   assert(dst.mod == mod_REG);
+   emit_1ub(p, 0xb8 + dst.idx);
+   emit_1i(p, imm);
+}
+
+void x86_push( struct x86_function *p,
+	       struct x86_reg reg )
+{
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x50 + reg.idx);
+   p->stack_offset += 4;
+}
+
+void x86_pop( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x58 + reg.idx);
+   p->stack_offset -= 4;
+}
+
+void x86_inc( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x40 + reg.idx);
+}
+
+void x86_dec( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x48 + reg.idx);
+}
+
+void x86_ret( struct x86_function *p )
+{
+   emit_1ub(p, 0xc3);
+}
+
+void x86_sahf( struct x86_function *p )
+{
+   emit_1ub(p, 0x9e);
+}
+
+void x86_mov( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   emit_op_modrm( p, 0x8b, 0x89, dst, src );
+}
+
+void x86_xor( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   emit_op_modrm( p, 0x33, 0x31, dst, src );
+}
+
+void x86_cmp( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   emit_op_modrm( p, 0x3b, 0x39, dst, src );
+}
+
+void x86_lea( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   emit_1ub(p, 0x8d);
+   emit_modrm( p, dst, src );
+}
+
+void x86_test( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   emit_1ub(p, 0x85);
+   emit_modrm( p, dst, src );
+}
+
+void x86_add( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   emit_op_modrm(p, 0x03, 0x01, dst, src );
+}
+
+void x86_mul( struct x86_function *p,
+	       struct x86_reg src )
+{
+   assert (src.file == file_REG32 && src.mod == mod_REG);
+   emit_op_modrm(p, 0xf7, 0, x86_make_reg (file_REG32, reg_SP), src );
+}
+
+void x86_sub( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   emit_op_modrm(p, 0x2b, 0x29, dst, src );
+}
+
+void x86_or( struct x86_function *p,
+             struct x86_reg dst,
+             struct x86_reg src )
+{
+   emit_op_modrm( p, 0x0b, 0x09, dst, src );
+}
+
+void x86_and( struct x86_function *p,
+              struct x86_reg dst,
+              struct x86_reg src )
+{
+   emit_op_modrm( p, 0x23, 0x21, dst, src );
+}
+
+
+
+/***********************************************************************
+ * SSE instructions
+ */
+
+
+void sse_movss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, 0xF3, X86_TWOB);
+   emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movaps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x28, 0x29, dst, src );
+}
+
+void sse_movups( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movhps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   assert(dst.mod != mod_REG || src.mod != mod_REG);
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x16, 0x17, dst, src ); /* cf movlhps */
+}
+
+void sse_movlps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   assert(dst.mod != mod_REG || src.mod != mod_REG);
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x12, 0x13, dst, src ); /* cf movhlps */
+}
+
+void sse_maxps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x5F);
+   emit_modrm( p, dst, src );
+}
+
+void sse_maxss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x5F);
+   emit_modrm( p, dst, src );
+}
+
+void sse_divss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x5E);
+   emit_modrm( p, dst, src );
+}
+
+void sse_minps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x5D);
+   emit_modrm( p, dst, src );
+}
+
+void sse_subps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x5C);
+   emit_modrm( p, dst, src );
+}
+
+void sse_mulps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x59);
+   emit_modrm( p, dst, src );
+}
+
+void sse_mulss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x59);
+   emit_modrm( p, dst, src );
+}
+
+void sse_addps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x58);
+   emit_modrm( p, dst, src );
+}
+
+void sse_addss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x58);
+   emit_modrm( p, dst, src );
+}
+
+void sse_andnps( struct x86_function *p,
+                 struct x86_reg dst,
+                 struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x55);
+   emit_modrm( p, dst, src );
+}
+
+void sse_andps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x54);
+   emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtps( struct x86_function *p,
+                  struct x86_reg dst,
+                  struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x52);
+   emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtss( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x52);
+   emit_modrm( p, dst, src );
+
+}
+
+void sse_movhlps( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   assert(dst.mod == mod_REG && src.mod == mod_REG);
+   emit_2ub(p, X86_TWOB, 0x12);
+   emit_modrm( p, dst, src );
+}
+
+void sse_movlhps( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   assert(dst.mod == mod_REG && src.mod == mod_REG);
+   emit_2ub(p, X86_TWOB, 0x16);
+   emit_modrm( p, dst, src );
+}
+
+void sse_orps( struct x86_function *p,
+               struct x86_reg dst,
+               struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x56);
+   emit_modrm( p, dst, src );
+}
+
+void sse_xorps( struct x86_function *p,
+                struct x86_reg dst,
+                struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x57);
+   emit_modrm( p, dst, src );
+}
+
+void sse_cvtps2pi( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   assert(dst.file == file_MMX && 
+	  (src.file == file_XMM || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x2d);
+   emit_modrm( p, dst, src );
+}
+
+
+/* Shufps can also be used to implement a reduced swizzle when dest ==
+ * arg0.
+ */
+void sse_shufps( struct x86_function *p,
+		 struct x86_reg dest,
+		 struct x86_reg arg0,
+		 unsigned char shuf) 
+{
+   emit_2ub(p, X86_TWOB, 0xC6);
+   emit_modrm(p, dest, arg0);
+   emit_1ub(p, shuf); 
+}
+
+void sse_cmpps( struct x86_function *p,
+		struct x86_reg dest,
+		struct x86_reg arg0,
+		unsigned char cc) 
+{
+   emit_2ub(p, X86_TWOB, 0xC2);
+   emit_modrm(p, dest, arg0);
+   emit_1ub(p, cc); 
+}
+
+void sse_pmovmskb( struct x86_function *p,
+                   struct x86_reg dest,
+                   struct x86_reg src)
+{
+    emit_3ub(p, 0x66, X86_TWOB, 0xD7);
+    emit_modrm(p, dest, src);
+}
+
+/***********************************************************************
+ * SSE2 instructions
+ */
+
+/**
+ * Perform a reduced swizzle:
+ */
+void sse2_pshufd( struct x86_function *p,
+		  struct x86_reg dest,
+		  struct x86_reg arg0,
+		  unsigned char shuf) 
+{
+   emit_3ub(p, 0x66, X86_TWOB, 0x70);
+   emit_modrm(p, dest, arg0);
+   emit_1ub(p, shuf); 
+}
+
+void sse2_cvttps2dq( struct x86_function *p,
+                     struct x86_reg dst,
+                     struct x86_reg src )
+{
+   emit_3ub( p, 0xF3, X86_TWOB, 0x5B );
+   emit_modrm( p, dst, src );
+}
+
+void sse2_cvtps2dq( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   emit_3ub(p, 0x66, X86_TWOB, 0x5B);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packssdw( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   emit_3ub(p, 0x66, X86_TWOB, 0x6B);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packsswb( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   emit_3ub(p, 0x66, X86_TWOB, 0x63);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packuswb( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   emit_3ub(p, 0x66, X86_TWOB, 0x67);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_rcpps( struct x86_function *p,
+                 struct x86_reg dst,
+                 struct x86_reg src )
+{
+   emit_2ub(p, X86_TWOB, 0x53);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_rcpss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_3ub(p, 0xF3, X86_TWOB, 0x53);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_movd( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   emit_2ub(p, 0x66, X86_TWOB);
+   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+
+
+
+/***********************************************************************
+ * x87 instructions
+ */
+void x87_fist( struct x86_function *p, struct x86_reg dst )
+{
+   emit_1ub(p, 0xdb);
+   emit_modrm_noreg(p, 2, dst);
+}
+
+void x87_fistp( struct x86_function *p, struct x86_reg dst )
+{
+   emit_1ub(p, 0xdb);
+   emit_modrm_noreg(p, 3, dst);
+}
+
+void x87_fild( struct x86_function *p, struct x86_reg arg )
+{
+   emit_1ub(p, 0xdf);
+   emit_modrm_noreg(p, 0, arg);
+}
+
+void x87_fldz( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xee);
+}
+
+
+void x87_fldcw( struct x86_function *p, struct x86_reg arg )
+{
+   assert(arg.file == file_REG32);
+   assert(arg.mod != mod_REG);
+   emit_1ub(p, 0xd9);
+   emit_modrm_noreg(p, 5, arg);
+}
+
+void x87_fld1( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xe8);
+}
+
+void x87_fldl2e( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xea);
+}
+
+void x87_fldln2( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xed);
+}
+
+void x87_fwait( struct x86_function *p )
+{
+   emit_1ub(p, 0x9b);
+}
+
+void x87_fnclex( struct x86_function *p )
+{
+   emit_2ub(p, 0xdb, 0xe2);
+}
+
+void x87_fclex( struct x86_function *p )
+{
+   x87_fwait(p);
+   x87_fnclex(p);
+}
+
+
+static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86_reg arg,
+			  unsigned char dst0ub0,
+			  unsigned char dst0ub1,
+			  unsigned char arg0ub0,
+			  unsigned char arg0ub1,
+			  unsigned char argmem_noreg)
+{
+   assert(dst.file == file_x87);
+
+   if (arg.file == file_x87) {
+      if (dst.idx == 0) 
+	 emit_2ub(p, dst0ub0, dst0ub1+arg.idx);
+      else if (arg.idx == 0) 
+	 emit_2ub(p, arg0ub0, arg0ub1+arg.idx);
+      else
+	 assert(0);
+   }
+   else if (dst.idx == 0) {
+      assert(arg.file == file_REG32);
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, argmem_noreg, arg);
+   }
+   else
+      assert(0);
+}
+
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xc8,
+		0xdc, 0xc8,
+		4);
+}
+
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xe0,
+		0xdc, 0xe8,
+		4);
+}
+
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xe8,
+		0xdc, 0xe0,
+		5);
+}
+
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xc0,
+		0xdc, 0xc0,
+		0);
+}
+
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xf0,
+		0xdc, 0xf8,
+		6);
+}
+
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg )
+{
+   x87_arith_op(p, dst, arg, 
+		0xd8, 0xf8,
+		0xdc, 0xf0,
+		7);
+}
+
+void x87_fmulp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xc8+dst.idx);
+}
+
+void x87_fsubp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xe8+dst.idx);
+}
+
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xe0+dst.idx);
+}
+
+void x87_faddp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xc0+dst.idx);
+}
+
+void x87_fdivp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xf8+dst.idx);
+}
+
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xf0+dst.idx);
+}
+
+void x87_fucom( struct x86_function *p, struct x86_reg arg )
+{
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdd, 0xe0+arg.idx);
+}
+
+void x87_fucomp( struct x86_function *p, struct x86_reg arg )
+{
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdd, 0xe8+arg.idx);
+}
+
+void x87_fucompp( struct x86_function *p )
+{
+   emit_2ub(p, 0xda, 0xe9);
+}
+
+void x87_fxch( struct x86_function *p, struct x86_reg arg )
+{
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xd9, 0xc8+arg.idx);
+}
+
+void x87_fabs( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xe1);
+}
+
+void x87_fchs( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xe0);
+}
+
+void x87_fcos( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xff);
+}
+
+
+void x87_fprndint( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xfc);
+}
+
+void x87_fscale( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xfd);
+}
+
+void x87_fsin( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xfe);
+}
+
+void x87_fsincos( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xfb);
+}
+
+void x87_fsqrt( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xfa);
+}
+
+void x87_fxtract( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xf4);
+}
+
+/* st0 = (2^st0)-1
+ *
+ * Restrictions: -1.0 <= st0 <= 1.0
+ */
+void x87_f2xm1( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xf0);
+}
+
+/* st1 = st1 * log2(st0);
+ * pop_stack;
+ */
+void x87_fyl2x( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xf1);
+}
+
+/* st1 = st1 * log2(st0 + 1.0);
+ * pop_stack;
+ *
+ * A fast operation, with restrictions: -.29 < st0 < .29 
+ */
+void x87_fyl2xp1( struct x86_function *p )
+{
+   emit_2ub(p, 0xd9, 0xf9);
+}
+
+
+void x87_fld( struct x86_function *p, struct x86_reg arg )
+{
+   if (arg.file == file_x87) 
+      emit_2ub(p, 0xd9, 0xc0 + arg.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 0, arg);
+   }
+}
+
+void x87_fst( struct x86_function *p, struct x86_reg dst )
+{
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xdd, 0xd0 + dst.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 2, dst);
+   }
+}
+
+void x87_fstp( struct x86_function *p, struct x86_reg dst )
+{
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xdd, 0xd8 + dst.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 3, dst);
+   }
+}
+
+void x87_fcom( struct x86_function *p, struct x86_reg dst )
+{
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xd8, 0xd0 + dst.idx);
+   else {
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, 2, dst);
+   }
+}
+
+void x87_fcomp( struct x86_function *p, struct x86_reg dst )
+{
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xd8, 0xd8 + dst.idx);
+   else {
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, 3, dst);
+   }
+}
+
+
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst )
+{
+   assert(dst.file == file_REG32);
+
+   if (dst.idx == reg_AX &&
+       dst.mod == mod_REG) 
+      emit_2ub(p, 0xdf, 0xe0);
+   else {
+      emit_1ub(p, 0xdd);
+      emit_modrm_noreg(p, 7, dst);
+   }
+}
+
+
+
+
+/***********************************************************************
+ * MMX instructions
+ */
+
+void mmx_emms( struct x86_function *p )
+{
+   assert(p->need_emms);
+   emit_2ub(p, 0x0f, 0x77);
+   p->need_emms = 0;
+}
+
+void mmx_packssdw( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   assert(dst.file == file_MMX && 
+	  (src.file == file_MMX || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x6b);
+   emit_modrm( p, dst, src );
+}
+
+void mmx_packuswb( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   assert(dst.file == file_MMX && 
+	  (src.file == file_MMX || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x67);
+   emit_modrm( p, dst, src );
+}
+
+void mmx_movd( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   p->need_emms = 1;
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+void mmx_movq( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   p->need_emms = 1;
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x6f, 0x7f, dst, src );
+}
+
+
+/***********************************************************************
+ * Helper functions
+ */
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity:
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p,
+			   unsigned arg )
+{
+   return x86_make_disp(x86_make_reg(file_REG32, reg_SP), 
+			p->stack_offset + arg * 4);	/* ??? */
+}
+
+
+void x86_init_func( struct x86_function *p )
+{
+   p->size = 0;
+   p->store = NULL;
+   p->csr = p->store;
+}
+
+void x86_init_func_size( struct x86_function *p, unsigned code_size )
+{
+   p->size = code_size;
+   p->store = rtasm_exec_malloc(code_size);
+   p->csr = p->store;
+}
+
+void x86_release_func( struct x86_function *p )
+{
+   rtasm_exec_free(p->store);
+   p->store = NULL;
+   p->csr = NULL;
+   p->size = 0;
+}
+
+
+void (*x86_get_func( struct x86_function *p ))(void)
+{
+   if (DISASSEM && p->store)
+      debug_printf("disassemble %p %p\n", p->store, p->csr);
+   return (void (*)(void)) (unsigned long) p->store;
+}
+
+#else
+
+void x86sse_dummy( void )
+{
+}
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
new file mode 100644
index 0000000000..606b41eb35
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -0,0 +1,278 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef _RTASM_X86SSE_H_
+#define _RTASM_X86SSE_H_
+
+#if defined(__i386__) || defined(__386__) || defined(i386)
+
+/* It is up to the caller to ensure that instructions issued are
+ * suitable for the host cpu.  There are no checks made in this module
+ * for mmx/sse/sse2 support on the cpu.
+ */
+struct x86_reg {
+   unsigned file:3;
+   unsigned idx:3;
+   unsigned mod:2;		/* mod_REG if this is just a register */
+   int      disp:24;		/* only +/- 23bits of offset - should be enough... */
+};
+
+struct x86_function {
+   unsigned size;
+   unsigned char *store;
+   unsigned char *csr;
+   unsigned stack_offset;
+   int need_emms;
+   const char *fn;
+};
+
+enum x86_reg_file {
+   file_REG32,
+   file_MMX,
+   file_XMM,
+   file_x87
+};
+
+/* Values for mod field of modr/m byte
+ */
+enum x86_reg_mod {
+   mod_INDIRECT,
+   mod_DISP8,
+   mod_DISP32,
+   mod_REG
+};
+
+enum x86_reg_name {
+   reg_AX,
+   reg_CX,
+   reg_DX,
+   reg_BX,
+   reg_SP,
+   reg_BP,
+   reg_SI,
+   reg_DI
+};
+
+
+enum x86_cc {
+   cc_O,			/* overflow */
+   cc_NO,			/* not overflow */
+   cc_NAE,			/* not above or equal / carry */
+   cc_AE,			/* above or equal / not carry */
+   cc_E,			/* equal / zero */
+   cc_NE			/* not equal / not zero */
+};
+
+enum sse_cc {
+   cc_Equal,
+   cc_LessThan,
+   cc_LessThanEqual,
+   cc_Unordered,
+   cc_NotEqual,
+   cc_NotLessThan,
+   cc_NotLessThanEqual,
+   cc_Ordered
+};
+
+#define cc_Z  cc_E
+#define cc_NZ cc_NE
+
+/* Begin/end/retreive function creation:
+ */
+
+
+void x86_init_func( struct x86_function *p );
+void x86_init_func_size( struct x86_function *p, unsigned code_size );
+void x86_release_func( struct x86_function *p );
+void (*x86_get_func( struct x86_function *p ))( void );
+
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+			     enum x86_reg_name idx );
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+			      int disp );
+
+struct x86_reg x86_deref( struct x86_reg reg );
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg );
+
+
+/* Labels, jumps and fixup:
+ */
+unsigned char *x86_get_label( struct x86_function *p );
+
+void x86_jcc( struct x86_function *p,
+	      enum x86_cc cc,
+	      unsigned char *label );
+
+unsigned char *x86_jcc_forward( struct x86_function *p,
+			  enum x86_cc cc );
+
+unsigned char *x86_jmp_forward( struct x86_function *p);
+
+unsigned char *x86_call_forward( struct x86_function *p);
+
+void x86_fixup_fwd_jump( struct x86_function *p,
+			 unsigned char *fixup );
+
+void x86_jmp( struct x86_function *p, unsigned char *label );
+
+/* void x86_call( struct x86_function *p, void (*label)() ); */
+void x86_call( struct x86_function *p, struct x86_reg reg);
+
+/* michal:
+ * Temporary. As I need immediate operands, and dont want to mess with the codegen,
+ * I load the immediate into general purpose register and use it.
+ */
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
+
+
+/* Macro for sse_shufps() and sse2_pshufd():
+ */
+#define SHUF(_x,_y,_z,_w)       (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6))
+#define SHUF_NOOP               RSW(0,1,2,3)
+#define GET_SHUF(swz, idx)      (((swz) >> ((idx)*2)) & 0x3)
+
+void mmx_emms( struct x86_function *p );
+void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                  unsigned char shuf );
+void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
+                unsigned char cc );
+void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                 unsigned char shuf );
+void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
+
+void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_dec( struct x86_function *p, struct x86_reg reg );
+void x86_inc( struct x86_function *p, struct x86_reg reg );
+void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mul( struct x86_function *p, struct x86_reg src );
+void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_pop( struct x86_function *p, struct x86_reg reg );
+void x86_push( struct x86_function *p, struct x86_reg reg );
+void x86_ret( struct x86_function *p );
+void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_sahf( struct x86_function *p );
+
+void x87_f2xm1( struct x86_function *p );
+void x87_fabs( struct x86_function *p );
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_faddp( struct x86_function *p, struct x86_reg dst );
+void x87_fchs( struct x86_function *p );
+void x87_fclex( struct x86_function *p );
+void x87_fcom( struct x86_function *p, struct x86_reg dst );
+void x87_fcomp( struct x86_function *p, struct x86_reg dst );
+void x87_fcos( struct x86_function *p );
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivp( struct x86_function *p, struct x86_reg dst );
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst );
+void x87_fild( struct x86_function *p, struct x86_reg arg );
+void x87_fist( struct x86_function *p, struct x86_reg dst );
+void x87_fistp( struct x86_function *p, struct x86_reg dst );
+void x87_fld( struct x86_function *p, struct x86_reg arg );
+void x87_fld1( struct x86_function *p );
+void x87_fldcw( struct x86_function *p, struct x86_reg arg );
+void x87_fldl2e( struct x86_function *p );
+void x87_fldln2( struct x86_function *p );
+void x87_fldz( struct x86_function *p );
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fmulp( struct x86_function *p, struct x86_reg dst );
+void x87_fnclex( struct x86_function *p );
+void x87_fprndint( struct x86_function *p );
+void x87_fscale( struct x86_function *p );
+void x87_fsin( struct x86_function *p );
+void x87_fsincos( struct x86_function *p );
+void x87_fsqrt( struct x86_function *p );
+void x87_fst( struct x86_function *p, struct x86_reg dst );
+void x87_fstp( struct x86_function *p, struct x86_reg dst );
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubp( struct x86_function *p, struct x86_reg dst );
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst );
+void x87_fxch( struct x86_function *p, struct x86_reg dst );
+void x87_fxtract( struct x86_function *p );
+void x87_fyl2x( struct x86_function *p );
+void x87_fyl2xp1( struct x86_function *p );
+void x87_fwait( struct x86_function *p );
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst );
+void x87_fucompp( struct x86_function *p );
+void x87_fucomp( struct x86_function *p, struct x86_reg arg );
+void x87_fucom( struct x86_function *p, struct x86_reg arg );
+
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity.  Note - doesn't track explict
+ * manipulation of ESP by other instructions.
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
+
+#endif
+#endif
diff --git a/src/gallium/auxiliary/tgsi/Makefile b/src/gallium/auxiliary/tgsi/Makefile
index 12a8bd0409..71f64b747c 100644
--- a/src/gallium/auxiliary/tgsi/Makefile
+++ b/src/gallium/auxiliary/tgsi/Makefile
@@ -1,3 +1,18 @@
-default:
-	cd ../.. ; make
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = tgsi
+
+C_SOURCES = \
+	exec/tgsi_exec.c \
+	exec/tgsi_sse2.c \
+	util/tgsi_build.c \
+	util/tgsi_dump.c \
+	util/tgsi_parse.c \
+	util/tgsi_transform.c \
+	util/tgsi_util.c
+
+include ../../Makefile.template
+
+symlinks:
 
diff --git a/src/gallium/auxiliary/tgsi/SConscript b/src/gallium/auxiliary/tgsi/SConscript
new file mode 100644
index 0000000000..8464bfe944
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/SConscript
@@ -0,0 +1,14 @@
+Import('*')
+
+tgsi = env.ConvenienceLibrary(
+	target = 'tgsi',
+	source = [
+		'exec/tgsi_exec.c',
+		'exec/tgsi_sse2.c',
+		'util/tgsi_build.c',
+		'util/tgsi_dump.c',
+		'util/tgsi_parse.c',
+		'util/tgsi_util.c',
+	])
+
+auxiliaries.insert(0, tgsi)
diff --git a/src/gallium/auxiliary/tgsi/exec/Makefile b/src/gallium/auxiliary/tgsi/exec/Makefile
deleted file mode 100644
index eb8b14e0e8..0000000000
--- a/src/gallium/auxiliary/tgsi/exec/Makefile
+++ /dev/null
@@ -1,3 +0,0 @@
-default:
-	cd ../../.. ; make
-
diff --git a/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
index 62c6a69c63..779b901f2b 100755
--- a/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
@@ -32,7 +32,7 @@
 #include "tgsi_exec.h"
 #include "tgsi_sse2.h"
 
-#include "x86/rtasm/x86sse.h"
+#include "rtasm/rtasm_x86sse.h"
 
 #if defined(__i386__) || defined(__386__)
 
@@ -2356,11 +2356,17 @@ tgsi_emit_sse2_fs(
          ok = emit_instruction(
             func,
             &parse.FullToken.FullInstruction );
+
+	 if (!ok) {
+	    debug_printf("failed to translate tgsi opcode %d\n", 
+			 parse.FullToken.FullInstruction.Instruction.Opcode );
+	 }
          break;
 
       case TGSI_TOKEN_TYPE_IMMEDIATE:
          /* XXX implement this */
 	 ok = 0;
+	 debug_printf("failed to emit immediate value\n");
          break;
 
       default:
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
new file mode 100644
index 0000000000..906a46d6b4
--- /dev/null
+++ b/src/gallium/auxiliary/util/Makefile
@@ -0,0 +1,15 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = util
+
+C_SOURCES = \
+	p_debug.c \
+	p_tile.c \
+	p_util.c \
+	u_mm.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
new file mode 100644
index 0000000000..4717941434
--- /dev/null
+++ b/src/gallium/auxiliary/util/SConscript
@@ -0,0 +1,12 @@
+Import('*')
+
+util = env.ConvenienceLibrary(
+	target = 'util',
+	source = [
+		'p_debug.c',
+		'p_tile.c',
+		'p_util.c',
+		'u_mm.c',
+	])
+
+auxiliaries.insert(0, util)
diff --git a/src/gallium/auxiliary/pipebuffer/linked_list.h b/src/gallium/auxiliary/util/u_double_list.h
index e99817fb13..8cb10c9820 100644
--- a/src/gallium/auxiliary/pipebuffer/linked_list.h
+++ b/src/gallium/auxiliary/util/u_double_list.h
@@ -34,8 +34,8 @@
  * be protected using an external mutex.
  */
 
-#ifndef LINKED_LIST_H_
-#define LINKED_LIST_H_
+#ifndef _U_DOUBLE_LIST_H_
+#define _U_DOUBLE_LIST_H_
 
 
 #include <stddef.h>
@@ -88,4 +88,4 @@ struct list_head
     ((__type *)(((char *)(__item)) - offsetof(__type, __field)))
 
 
-#endif /*LINKED_LIST_H_*/
+#endif /*_U_DOUBLE_LIST_H_*/
diff --git a/src/gallium/auxiliary/util/u_mm.c b/src/gallium/auxiliary/util/u_mm.c
new file mode 100644
index 0000000000..b49ae074e0
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_mm.c
@@ -0,0 +1,283 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * 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
+ * WITTAWAT YAMWONG, OR ANY OTHER CONTRIBUTORS 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 "pipe/p_compiler.h"
+#include "pipe/p_util.h"
+#include "pipe/p_debug.h"
+
+#include "util/u_mm.h"
+
+
+void
+mmDumpMemInfo(const struct mem_block *heap)
+{
+   debug_printf("Memory heap %p:\n", (void *)heap);
+   if (heap == 0) {
+      debug_printf("  heap == 0\n");
+   } else {
+      const struct mem_block *p;
+
+      for(p = heap->next; p != heap; p = p->next) {
+	 debug_printf("  Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+      debug_printf("\nFree list:\n");
+
+      for(p = heap->next_free; p != heap; p = p->next_free) {
+	 debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+   }
+   debug_printf("End of memory blocks\n");
+}
+
+struct mem_block *
+mmInit(int ofs, int size)
+{
+   struct mem_block *heap, *block;
+  
+   if (size <= 0) 
+      return NULL;
+
+   heap = CALLOC_STRUCT(mem_block);
+   if (!heap) 
+      return NULL;
+   
+   block = CALLOC_STRUCT(mem_block);
+   if (!block) {
+      FREE(heap);
+      return NULL;
+   }
+
+   heap->next = block;
+   heap->prev = block;
+   heap->next_free = block;
+   heap->prev_free = block;
+
+   block->heap = heap;
+   block->next = heap;
+   block->prev = heap;
+   block->next_free = heap;
+   block->prev_free = heap;
+
+   block->ofs = ofs;
+   block->size = size;
+   block->free = 1;
+
+   return heap;
+}
+
+
+static struct mem_block *
+SliceBlock(struct mem_block *p, 
+           int startofs, int size, 
+           int reserved, int alignment)
+{
+   struct mem_block *newblock;
+
+   /* break left  [p, newblock, p->next], then p = newblock */
+   if (startofs > p->ofs) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs;
+      newblock->size = p->size - (startofs - p->ofs);
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+
+      p->size -= newblock->size;
+      p = newblock;
+   }
+
+   /* break right, also [p, newblock, p->next] */
+   if (size < p->size) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs + size;
+      newblock->size = p->size - size;
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+	 
+      p->size = size;
+   }
+
+   /* p = middle block */
+   p->free = 0;
+
+   /* Remove p from the free list: 
+    */
+   p->next_free->prev_free = p->prev_free;
+   p->prev_free->next_free = p->next_free;
+
+   p->next_free = 0;
+   p->prev_free = 0;
+
+   p->reserved = reserved;
+   return p;
+}
+
+
+struct mem_block *
+mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch)
+{
+   struct mem_block *p;
+   const int mask = (1 << align2)-1;
+   int startofs = 0;
+   int endofs;
+
+   if (!heap || align2 < 0 || size <= 0)
+      return NULL;
+
+   for (p = heap->next_free; p != heap; p = p->next_free) {
+      assert(p->free);
+
+      startofs = (p->ofs + mask) & ~mask;
+      if ( startofs < startSearch ) {
+	 startofs = startSearch;
+      }
+      endofs = startofs+size;
+      if (endofs <= (p->ofs+p->size))
+	 break;
+   }
+
+   if (p == heap) 
+      return NULL;
+
+   assert(p->free);
+   p = SliceBlock(p,startofs,size,0,mask+1);
+
+   return p;
+}
+
+
+struct mem_block *
+mmFindBlock(struct mem_block *heap, int start)
+{
+   struct mem_block *p;
+
+   for (p = heap->next; p != heap; p = p->next) {
+      if (p->ofs == start) 
+	 return p;
+   }
+
+   return NULL;
+}
+
+
+static INLINE int
+Join2Blocks(struct mem_block *p)
+{
+   /* XXX there should be some assertions here */
+
+   /* NOTE: heap->free == 0 */
+
+   if (p->free && p->next->free) {
+      struct mem_block *q = p->next;
+
+      assert(p->ofs + p->size == q->ofs);
+      p->size += q->size;
+
+      p->next = q->next;
+      q->next->prev = p;
+
+      q->next_free->prev_free = q->prev_free; 
+      q->prev_free->next_free = q->next_free;
+     
+      FREE(q);
+      return 1;
+   }
+   return 0;
+}
+
+int
+mmFreeMem(struct mem_block *b)
+{
+   if (!b)
+      return 0;
+
+   if (b->free) {
+      debug_printf("block already free\n");
+      return -1;
+   }
+   if (b->reserved) {
+      debug_printf("block is reserved\n");
+      return -1;
+   }
+
+   b->free = 1;
+   b->next_free = b->heap->next_free;
+   b->prev_free = b->heap;
+   b->next_free->prev_free = b;
+   b->prev_free->next_free = b;
+
+   Join2Blocks(b);
+   if (b->prev != b->heap)
+      Join2Blocks(b->prev);
+
+   return 0;
+}
+
+
+void
+mmDestroy(struct mem_block *heap)
+{
+   struct mem_block *p;
+
+   if (!heap)
+      return;
+
+   for (p = heap->next; p != heap; ) {
+      struct mem_block *next = p->next;
+      FREE(p);
+      p = next;
+   }
+
+   FREE(heap);
+}
diff --git a/src/gallium/auxiliary/util/u_mm.h b/src/gallium/auxiliary/util/u_mm.h
new file mode 100644
index 0000000000..b226b101cb
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_mm.h
@@ -0,0 +1,91 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * 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
+ * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS 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
+ * Memory manager code.  Primarily used by device drivers to manage texture
+ * heaps, etc.
+ */
+
+
+#ifndef _U_MM_H_
+#define _U_MM_H_
+
+
+struct mem_block {
+   struct mem_block *next, *prev;
+   struct mem_block *next_free, *prev_free;
+   struct mem_block *heap;
+   int ofs,size;
+   unsigned int free:1;
+   unsigned int reserved:1;
+};
+
+
+
+/** 
+ * input: total size in bytes
+ * return: a heap pointer if OK, NULL if error
+ */
+extern struct mem_block *mmInit(int ofs, int size);
+
+/**
+ * Allocate 'size' bytes with 2^align2 bytes alignment,
+ * restrict the search to free memory after 'startSearch'
+ * depth and back buffers should be in different 4mb banks
+ * to get better page hits if possible
+ * input:	size = size of block
+ *       	align2 = 2^align2 bytes alignment
+ *		startSearch = linear offset from start of heap to begin search
+ * return: pointer to the allocated block, 0 if error
+ */
+extern struct mem_block *mmAllocMem(struct mem_block *heap, int size, int align2, 
+                            int startSearch);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a block
+ * return: 0 if OK, -1 if error
+ */
+extern int mmFreeMem(struct mem_block *b);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a heap, start offset
+ * return: pointer to a block
+ */
+extern struct mem_block *mmFindBlock(struct mem_block *heap, int start);
+
+/**
+ * destroy MM
+ */
+extern void mmDestroy(struct mem_block *mmInit);
+
+/**
+ * For debuging purpose.
+ */
+extern void mmDumpMemInfo(const struct mem_block *mmInit);
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_simple_list.h b/src/gallium/auxiliary/util/u_simple_list.h
new file mode 100644
index 0000000000..f5f43b0faa
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_list.h
@@ -0,0 +1,197 @@
+/**
+ * \file simple_list.h
+ * Simple macros for type-safe, intrusive lists.
+ *
+ *  Intended to work with a list sentinal which is created as an empty
+ *  list.  Insert & delete are O(1).
+ *  
+ * \author
+ *  (C) 1997, Keith Whitwell
+ */
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef _U_SIMPLE_LIST_H_
+#define _U_SIMPLE_LIST_H_
+
+/**
+ * Remove an element from list.
+ *
+ * \param elem element to remove.
+ */
+#define remove_from_list(elem)			\
+do {						\
+   (elem)->next->prev = (elem)->prev;		\
+   (elem)->prev->next = (elem)->next;		\
+} while (0)
+
+/**
+ * Insert an element to the list head.
+ *
+ * \param list list.
+ * \param elem element to insert.
+ */
+#define insert_at_head(list, elem)		\
+do {						\
+   (elem)->prev = list;				\
+   (elem)->next = (list)->next;			\
+   (list)->next->prev = elem;			\
+   (list)->next = elem;				\
+} while(0)
+
+/**
+ * Insert an element to the list tail.
+ *
+ * \param list list.
+ * \param elem element to insert.
+ */
+#define insert_at_tail(list, elem)		\
+do {						\
+   (elem)->next = list;				\
+   (elem)->prev = (list)->prev;			\
+   (list)->prev->next = elem;			\
+   (list)->prev = elem;				\
+} while(0)
+
+/**
+ * Move an element to the list head.
+ *
+ * \param list list.
+ * \param elem element to move.
+ */
+#define move_to_head(list, elem)		\
+do {						\
+   remove_from_list(elem);			\
+   insert_at_head(list, elem);			\
+} while (0)
+
+/**
+ * Move an element to the list tail.
+ *
+ * \param list list.
+ * \param elem element to move.
+ */
+#define move_to_tail(list, elem)		\
+do {						\
+   remove_from_list(elem);			\
+   insert_at_tail(list, elem);			\
+} while (0)
+
+/**
+ * Make a empty list empty.
+ *
+ * \param sentinal list (sentinal element).
+ */
+#define make_empty_list(sentinal)		\
+do {						\
+   (sentinal)->next = sentinal;			\
+   (sentinal)->prev = sentinal;			\
+} while (0)
+
+/**
+ * Get list first element.
+ *
+ * \param list list.
+ *
+ * \return pointer to first element.
+ */
+#define first_elem(list)       ((list)->next)
+
+/**
+ * Get list last element.
+ *
+ * \param list list.
+ *
+ * \return pointer to last element.
+ */
+#define last_elem(list)        ((list)->prev)
+
+/**
+ * Get next element.
+ *
+ * \param elem element.
+ *
+ * \return pointer to next element.
+ */
+#define next_elem(elem)        ((elem)->next)
+
+/**
+ * Get previous element.
+ *
+ * \param elem element.
+ *
+ * \return pointer to previous element.
+ */
+#define prev_elem(elem)        ((elem)->prev)
+
+/**
+ * Test whether element is at end of the list.
+ * 
+ * \param list list.
+ * \param elem element.
+ * 
+ * \return non-zero if element is at end of list, or zero otherwise.
+ */
+#define at_end(list, elem)     ((elem) == (list))
+
+/**
+ * Test if a list is empty.
+ * 
+ * \param list list.
+ * 
+ * \return non-zero if list empty, or zero otherwise.
+ */
+#define is_empty_list(list)    ((list)->next == (list))
+
+/**
+ * Walk through the elements of a list.
+ *
+ * \param ptr pointer to the current element.
+ * \param list list.
+ *
+ * \note It should be followed by a { } block or a single statement, as in a \c
+ * for loop.
+ */
+#define foreach(ptr, list)     \
+        for( ptr=(list)->next ;  ptr!=list ;  ptr=(ptr)->next )
+
+/**
+ * Walk through the elements of a list.
+ *
+ * Same as #foreach but lets you unlink the current value during a list
+ * traversal.  Useful for freeing a list, element by element.
+ * 
+ * \param ptr pointer to the current element.
+ * \param t temporary pointer.
+ * \param list list.
+ *
+ * \note It should be followed by a { } block or a single statement, as in a \c
+ * for loop.
+ */
+#define foreach_s(ptr, t, list)   \
+        for(ptr=(list)->next,t=(ptr)->next; list != ptr; ptr=t, t=(t)->next)
+
+#endif /* _U_SIMPLE_LIST_H_ */
diff --git a/src/gallium/drivers/Makefile b/src/gallium/drivers/Makefile
index 58df6c5009..6161cb6ff8 100644
--- a/src/gallium/drivers/Makefile
+++ b/src/gallium/drivers/Makefile
@@ -2,12 +2,7 @@ TOP = ../../..
 include $(TOP)/configs/current
 
 
-ifeq ($(CONFIG_NAME), linux-cell)
-CELL_DIR = cell
-endif
-
-SUBDIRS = softpipe i915simple i965simple nv30 nv40 nv50 \
-	  failover pipebuffer $(CELL_DIR)
+SUBDIRS = $(GALLIUM_DRIVER_DIRS)
 
 
 default: subdirs
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index 74b131fbef..9a4004535e 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -87,10 +87,13 @@
 #define CELL_CMD_STATE_TEXTURE       13
 #define CELL_CMD_STATE_VERTEX_INFO   14
 #define CELL_CMD_STATE_VIEWPORT      15
-#define CELL_CMD_STATE_VS_ARRAY_INFO 16
-#define CELL_CMD_STATE_BLEND         17
-#define CELL_CMD_VS_EXECUTE          18
-#define CELL_CMD_STATE_ATTRIB_FETCH  19
+#define CELL_CMD_STATE_UNIFORMS      16
+#define CELL_CMD_STATE_VS_ARRAY_INFO 17
+#define CELL_CMD_STATE_BIND_VS       18
+#define CELL_CMD_STATE_BLEND         19
+#define CELL_CMD_STATE_ATTRIB_FETCH  20
+#define CELL_CMD_VS_EXECUTE          21
+#define CELL_CMD_FLUSH_BUFFER_RANGE  22
 
 
 #define CELL_NUM_BUFFERS 4
@@ -134,7 +137,7 @@ struct cell_array_info
    uint pitch;         /**< Byte pitch from one entry to the next. */
    uint size;
    uint function_offset;
-} ALIGN16_ATTRIB;
+};
 
 
 struct cell_attribute_fetch_code {
@@ -142,32 +145,37 @@ struct cell_attribute_fetch_code {
    uint size;
 };
 
+
+struct cell_buffer_range {
+   uint64_t base;
+   unsigned size;
+};
+
+
 struct cell_shader_info
 {
-   unsigned num_outputs;
-
    uint64_t declarations;
-   unsigned num_declarations;
    uint64_t instructions;
-   unsigned num_instructions;
-   uint64_t uniforms;
    uint64_t  immediates;
+
+   unsigned num_outputs;
+   unsigned num_declarations;
+   unsigned num_instructions;
    unsigned num_immediates;
-} ALIGN16_ATTRIB;
+};
 
 
 #define SPU_VERTS_PER_BATCH 64
 struct cell_command_vs
 {
    uint64_t opcode;       /**< CELL_CMD_VS_EXECUTE */
-   struct cell_shader_info   shader;
+   uint64_t vOut[SPU_VERTS_PER_BATCH];
    unsigned num_elts;
    unsigned elts[SPU_VERTS_PER_BATCH];
-   uint64_t vOut[SPU_VERTS_PER_BATCH];
    float plane[12][4];
    unsigned nr_planes;
    unsigned nr_attrs;
-} ALIGN16_ATTRIB;
+};
 
 
 struct cell_command_render
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index 196ab777f5..164dde762c 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -1,6 +1,6 @@
 # Gallium3D Cell driver: PPU code
 
-# This makefile builds the g3dcell.a library which gets pulled into
+# This makefile builds the libcell.a library which gets pulled into
 # the main libGL.so library
 
 
@@ -8,10 +8,14 @@ TOP = ../../../../..
 include $(TOP)/configs/linux-cell
 
 
-#PROG = gl4
+# This is the "top-level" cell PPU driver code, will get pulled into libGL.so
+# by the winsys Makefile.
+CELL_LIB = ../libcell.a
 
-CELL_LIB = libcell.a
 
+# This is the SPU code.  We'd like to be able to put this into the libcell.a
+# archive with the PPU code, but nesting .a libs doesn't seem to work.
+# So, it's pulled into libGL.so in gallium/winsys/xlib/Makefile
 SPU_CODE_MODULE = ../spu/g3d_spu.a
 
 
@@ -21,14 +25,10 @@ SOURCES = \
 	cell_context.c \
 	cell_draw_arrays.c \
 	cell_flush.c \
-	cell_state_blend.c \
-	cell_state_clip.c \
 	cell_state_derived.c \
 	cell_state_emit.c \
-	cell_state_fs.c \
-	cell_state_rasterizer.c \
-	cell_state_sampler.c \
-	cell_state_surface.c \
+	cell_state_shader.c \
+	cell_pipe_state.c \
 	cell_state_vertex.c \
 	cell_spu.c \
 	cell_surface.c \
@@ -56,7 +56,7 @@ default: $(CELL_LIB)
 
 
 $(CELL_LIB): $(OBJECTS) $(SPU_CODE_MODULE)
-#	ar -ru $(CELL_LIB) $(OBJECTS) $(SPU_CODE_MODULE)
+#	ar -ru $(CELL_LIB) $(OBJECTS) $(SPU_CODE_MODULE) # doesn't work
 	ar -ru $(CELL_LIB) $(OBJECTS)
 
 #$(PROG): $(PPU_OBJECTS)
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
index e588a30d5b..3ffe09add6 100644
--- a/src/gallium/drivers/cell/ppu/cell_clear.c
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -41,6 +41,7 @@
 #include "cell_batch.h"
 #include "cell_flush.h"
 #include "cell_spu.h"
+#include "cell_state.h"
 
 
 void
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
index e1eb22f468..98c314f45c 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.c
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -48,6 +48,7 @@
 #include "cell_state.h"
 #include "cell_surface.h"
 #include "cell_spu.h"
+#include "cell_pipe_state.h"
 #include "cell_texture.h"
 #include "cell_vbuf.h"
 
@@ -198,40 +199,6 @@ cell_create_context(struct pipe_winsys *winsys, struct cell_winsys *cws)
 
 
    /* state setters */
-   cell->pipe.create_blend_state = cell_create_blend_state;
-   cell->pipe.bind_blend_state   = cell_bind_blend_state;
-   cell->pipe.delete_blend_state = cell_delete_blend_state;
-
-   cell->pipe.create_sampler_state = cell_create_sampler_state;
-   cell->pipe.bind_sampler_state   = cell_bind_sampler_state;
-   cell->pipe.delete_sampler_state = cell_delete_sampler_state;
-
-   cell->pipe.create_depth_stencil_alpha_state = cell_create_depth_stencil_alpha_state;
-   cell->pipe.bind_depth_stencil_alpha_state   = cell_bind_depth_stencil_alpha_state;
-   cell->pipe.delete_depth_stencil_alpha_state = cell_delete_depth_stencil_alpha_state;
-
-   cell->pipe.create_rasterizer_state = cell_create_rasterizer_state;
-   cell->pipe.bind_rasterizer_state   = cell_bind_rasterizer_state;
-   cell->pipe.delete_rasterizer_state = cell_delete_rasterizer_state;
-
-   cell->pipe.create_fs_state = cell_create_fs_state;
-   cell->pipe.bind_fs_state   = cell_bind_fs_state;
-   cell->pipe.delete_fs_state = cell_delete_fs_state;
-
-   cell->pipe.create_vs_state = cell_create_vs_state;
-   cell->pipe.bind_vs_state   = cell_bind_vs_state;
-   cell->pipe.delete_vs_state = cell_delete_vs_state;
-
-   cell->pipe.set_blend_color = cell_set_blend_color;
-   cell->pipe.set_clip_state = cell_set_clip_state;
-   cell->pipe.set_constant_buffer = cell_set_constant_buffer;
-
-   cell->pipe.set_framebuffer_state = cell_set_framebuffer_state;
-
-   cell->pipe.set_polygon_stipple = cell_set_polygon_stipple;
-   cell->pipe.set_scissor_state = cell_set_scissor_state;
-   cell->pipe.set_viewport_state = cell_set_viewport_state;
-
    cell->pipe.set_vertex_buffer = cell_set_vertex_buffer;
    cell->pipe.set_vertex_element = cell_set_vertex_element;
 
@@ -241,20 +208,16 @@ cell_create_context(struct pipe_winsys *winsys, struct cell_winsys *cws)
    cell->pipe.clear = cell_clear_surface;
    cell->pipe.flush = cell_flush;
 
-   /* textures */
-   cell->pipe.texture_create = cell_texture_create;
-   cell->pipe.texture_release = cell_texture_release;
-   cell->pipe.get_tex_surface = cell_get_tex_surface;
-
-   cell->pipe.set_sampler_texture = cell_set_sampler_texture;
-
 #if 0
    cell->pipe.begin_query = cell_begin_query;
    cell->pipe.end_query = cell_end_query;
    cell->pipe.wait_query = cell_wait_query;
 #endif
 
+   cell_init_state_functions(cell);
+   cell_init_shader_functions(cell);
    cell_init_surface_functions(cell);
+   cell_init_texture_functions(cell);
 
    cell->draw = cell_draw_create(cell);
 
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 91f8e542a2..3b687bb868 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -36,7 +36,7 @@
 #include "draw/draw_vbuf.h"
 #include "cell_winsys.h"
 #include "cell/common.h"
-#include "ppc/rtasm/spe_asm.h"
+#include "rtasm/rtasm_ppc_spe.h"
 
 
 struct cell_vbuf_render;
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index f12613649b..cbd387f014 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -49,9 +49,12 @@ 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].size)
+      if (sp->constants[i].size) {
          sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer,
                                                   PIPE_BUFFER_USAGE_CPU_READ);
+         cell_flush_buffer_range(sp, sp->mapped_constants[i], 
+                                 sp->constants[i].buffer->size);
+      }
    }
 
    draw_set_mapped_constant_buffer(sp->draw,
@@ -124,6 +127,7 @@ cell_draw_elements(struct pipe_context *pipe,
          void *buf = pipe->winsys->buffer_map(pipe->winsys,
                                               sp->vertex_buffer[i].buffer,
                                               PIPE_BUFFER_USAGE_CPU_READ);
+	 cell_flush_buffer_range(sp, buf, sp->vertex_buffer[i].buffer->size);
          draw_set_mapped_vertex_buffer(draw, i, buf);
       }
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.c b/src/gallium/drivers/cell/ppu/cell_flush.c
index 20f27531fc..66a5627d84 100644
--- a/src/gallium/drivers/cell/ppu/cell_flush.c
+++ b/src/gallium/drivers/cell/ppu/cell_flush.c
@@ -82,3 +82,17 @@ cell_flush_int(struct pipe_context *pipe, unsigned flags)
 
    flushing = FALSE;
 }
+
+
+void
+cell_flush_buffer_range(struct cell_context *cell, void *ptr,
+			unsigned size)
+{
+   uint64_t batch[1 + (ROUNDUP8(sizeof(struct cell_buffer_range)) / 8)];
+   struct cell_buffer_range *br = (struct cell_buffer_range *) & batch[1];
+
+   batch[0] = CELL_CMD_FLUSH_BUFFER_RANGE;
+   br->base = (uintptr_t) ptr;
+   br->size = size;
+   cell_batch_append(cell, batch, sizeof(batch));
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
new file mode 100644
index 0000000000..95bfc29fbe
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -0,0 +1,326 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/* Authors:
+ *  Keith Whitwell <keith@tungstengraphics.com>
+ *  Brian Paul
+ */
+
+#include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
+#include "draw/draw_context.h"
+#include "cell_context.h"
+#include "cell_state.h"
+#include "cell_texture.h"
+
+
+
+static void *
+cell_create_blend_state(struct pipe_context *pipe,
+                        const struct pipe_blend_state *blend)
+{
+   return mem_dup(blend, sizeof(*blend));
+}
+
+
+static void
+cell_bind_blend_state(struct pipe_context *pipe, void *blend)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->blend = (const struct pipe_blend_state *)blend;
+
+   cell->dirty |= CELL_NEW_BLEND;
+}
+
+
+static void
+cell_delete_blend_state(struct pipe_context *pipe, void *blend)
+{
+   FREE(blend);
+}
+
+
+static void
+cell_set_blend_color(struct pipe_context *pipe,
+                     const struct pipe_blend_color *blend_color)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->blend_color = *blend_color;
+
+   cell->dirty |= CELL_NEW_BLEND;
+}
+
+
+
+
+static void *
+cell_create_depth_stencil_alpha_state(struct pipe_context *pipe,
+                 const struct pipe_depth_stencil_alpha_state *depth_stencil)
+{
+   return mem_dup(depth_stencil, sizeof(*depth_stencil));
+}
+
+
+static void
+cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe,
+                                    void *depth_stencil)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->depth_stencil
+      = (const struct pipe_depth_stencil_alpha_state *) depth_stencil;
+
+   cell->dirty |= CELL_NEW_DEPTH_STENCIL;
+}
+
+
+static void
+cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth)
+{
+   FREE(depth);
+}
+
+
+static void cell_set_clip_state( struct pipe_context *pipe,
+			     const struct pipe_clip_state *clip )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   /* pass the clip state to the draw module */
+   draw_set_clip_state(cell->draw, clip);
+}
+
+
+
+/* Called when driver state tracker notices changes to the viewport
+ * matrix:
+ */
+static void
+cell_set_viewport_state( struct pipe_context *pipe,
+                         const struct pipe_viewport_state *viewport )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   cell->viewport = *viewport; /* struct copy */
+   cell->dirty |= CELL_NEW_VIEWPORT;
+
+   /* pass the viewport info to the draw module */
+   draw_set_viewport_state(cell->draw, viewport);
+
+   /* Using tnl/ and vf/ modules is temporary while getting started.
+    * Full pipe will have vertex shader, vertex fetch of its own.
+    */
+}
+
+
+static void
+cell_set_scissor_state( struct pipe_context *pipe,
+                        const struct pipe_scissor_state *scissor )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   memcpy( &cell->scissor, scissor, sizeof(*scissor) );
+   cell->dirty |= CELL_NEW_SCISSOR;
+}
+
+
+static void
+cell_set_polygon_stipple( struct pipe_context *pipe,
+                          const struct pipe_poly_stipple *stipple )
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   memcpy( &cell->poly_stipple, stipple, sizeof(*stipple) );
+   cell->dirty |= CELL_NEW_STIPPLE;
+}
+
+
+
+static void *
+cell_create_rasterizer_state(struct pipe_context *pipe,
+                             const struct pipe_rasterizer_state *setup)
+{
+   struct pipe_rasterizer_state *state
+      = MALLOC(sizeof(struct pipe_rasterizer_state));
+   memcpy(state, setup, sizeof(struct pipe_rasterizer_state));
+   return state;
+}
+
+
+static void
+cell_bind_rasterizer_state(struct pipe_context *pipe, void *setup)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   /* pass-through to draw module */
+   draw_set_rasterizer_state(cell->draw, setup);
+
+   cell->rasterizer = (struct pipe_rasterizer_state *)setup;
+
+   cell->dirty |= CELL_NEW_RASTERIZER;
+}
+
+
+static void
+cell_delete_rasterizer_state(struct pipe_context *pipe, void *rasterizer)
+{
+   FREE(rasterizer);
+}
+
+
+
+static void *
+cell_create_sampler_state(struct pipe_context *pipe,
+                          const struct pipe_sampler_state *sampler)
+{
+   return mem_dup(sampler, sizeof(*sampler));
+}
+
+
+static void
+cell_bind_sampler_state(struct pipe_context *pipe,
+                            unsigned unit, void *sampler)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+   cell->sampler[unit] = (struct pipe_sampler_state *)sampler;
+
+   cell->dirty |= CELL_NEW_SAMPLER;
+}
+
+
+static void
+cell_delete_sampler_state(struct pipe_context *pipe,
+                              void *sampler)
+{
+   FREE( sampler );
+}
+
+
+
+static void
+cell_set_sampler_texture(struct pipe_context *pipe,
+                         unsigned sampler,
+                         struct pipe_texture *texture)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   pipe_texture_reference(pipe,
+                          (struct pipe_texture **) &cell->texture[sampler],
+                          texture);
+
+   cell_update_texture_mapping(cell);
+
+   cell->dirty |= CELL_NEW_TEXTURE;
+}
+
+
+
+static void
+cell_set_framebuffer_state(struct pipe_context *pipe,
+                           const struct pipe_framebuffer_state *fb)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   if (1 /*memcmp(&cell->framebuffer, fb, sizeof(*fb))*/) {
+      struct pipe_surface *csurf = fb->cbufs[0];
+      struct pipe_surface *zsurf = fb->zsbuf;
+      uint i;
+
+      /* unmap old surfaces */
+      for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+         if (cell->framebuffer.cbufs[i] && cell->cbuf_map[i]) {
+            pipe_surface_unmap(cell->framebuffer.cbufs[i]);
+            cell->cbuf_map[i] = NULL;
+         }
+      }
+
+      if (cell->framebuffer.zsbuf && cell->zsbuf_map) {
+         pipe_surface_unmap(cell->framebuffer.zsbuf);
+         cell->zsbuf_map = NULL;
+      }
+
+      /* update my state */
+      cell->framebuffer = *fb;
+
+      /* map new surfaces */
+      if (csurf)
+         cell->cbuf_map[0] = pipe_surface_map(csurf);
+
+      if (zsurf)
+         cell->zsbuf_map = pipe_surface_map(zsurf);
+
+      cell->dirty |= CELL_NEW_FRAMEBUFFER;
+   }
+}
+
+
+
+void
+cell_init_state_functions(struct cell_context *cell)
+{
+   cell->pipe.create_blend_state = cell_create_blend_state;
+   cell->pipe.bind_blend_state   = cell_bind_blend_state;
+   cell->pipe.delete_blend_state = cell_delete_blend_state;
+
+   cell->pipe.create_sampler_state = cell_create_sampler_state;
+   cell->pipe.bind_sampler_state   = cell_bind_sampler_state;
+   cell->pipe.delete_sampler_state = cell_delete_sampler_state;
+
+   cell->pipe.set_sampler_texture = cell_set_sampler_texture;
+
+   cell->pipe.create_depth_stencil_alpha_state = cell_create_depth_stencil_alpha_state;
+   cell->pipe.bind_depth_stencil_alpha_state   = cell_bind_depth_stencil_alpha_state;
+   cell->pipe.delete_depth_stencil_alpha_state = cell_delete_depth_stencil_alpha_state;
+
+   cell->pipe.create_rasterizer_state = cell_create_rasterizer_state;
+   cell->pipe.bind_rasterizer_state   = cell_bind_rasterizer_state;
+   cell->pipe.delete_rasterizer_state = cell_delete_rasterizer_state;
+
+   cell->pipe.set_blend_color = cell_set_blend_color;
+   cell->pipe.set_clip_state = cell_set_clip_state;
+
+   cell->pipe.set_framebuffer_state = cell_set_framebuffer_state;
+
+   cell->pipe.set_polygon_stipple = cell_set_polygon_stipple;
+   cell->pipe.set_scissor_state = cell_set_scissor_state;
+   cell->pipe.set_viewport_state = cell_set_viewport_state;
+}
+
diff --git a/src/gallium/drivers/cell/ppu/cell_state_surface.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.h
index 287610b76b..1889bd52ff 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_surface.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.h
@@ -1,8 +1,8 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 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
@@ -26,46 +26,14 @@
  **************************************************************************/
 
 
-#include "pipe/p_inlines.h"
-#include "cell_context.h"
-#include "cell_state.h"
-
-
-void
-cell_set_framebuffer_state(struct pipe_context *pipe,
-                           const struct pipe_framebuffer_state *fb)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   if (1 /*memcmp(&cell->framebuffer, fb, sizeof(*fb))*/) {
-      struct pipe_surface *csurf = fb->cbufs[0];
-      struct pipe_surface *zsurf = fb->zsbuf;
-      uint i;
-
-      /* unmap old surfaces */
-      for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
-         if (cell->framebuffer.cbufs[i] && cell->cbuf_map[i]) {
-            pipe_surface_unmap(cell->framebuffer.cbufs[i]);
-            cell->cbuf_map[i] = NULL;
-         }
-      }
-
-      if (cell->framebuffer.zsbuf && cell->zsbuf_map) {
-         pipe_surface_unmap(cell->framebuffer.zsbuf);
-         cell->zsbuf_map = NULL;
-      }
+#ifndef CELL_PIPE_STATE_H
+#define CELL_PIPE_STATE_H
 
-      /* update my state */
-      cell->framebuffer = *fb;
 
-      /* map new surfaces */
-      if (csurf)
-         cell->cbuf_map[0] = pipe_surface_map(csurf);
+struct cell_context;
 
-      if (zsurf)
-         cell->zsbuf_map = pipe_surface_map(zsurf);
+extern void
+cell_init_state_functions(struct cell_context *cell);
 
-      cell->dirty |= CELL_NEW_FRAMEBUFFER;
-   }
-}
 
+#endif /* CELL_PIPE_STATE_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_state.h b/src/gallium/drivers/cell/ppu/cell_state.h
index 3a71ba14fa..31ce505e21 100644
--- a/src/gallium/drivers/cell/ppu/cell_state.h
+++ b/src/gallium/drivers/cell/ppu/cell_state.h
@@ -1,3 +1,29 @@
+/**************************************************************************
+ * 
+ * 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_STATE_H
@@ -22,82 +48,6 @@
 #define CELL_NEW_VERTEX_INFO   0x8000
 
 
-
-extern void
-cell_set_framebuffer_state( struct pipe_context *,
-                            const struct pipe_framebuffer_state * );
-
-
-
-extern void *
-cell_create_blend_state(struct pipe_context *, const struct pipe_blend_state *);
-extern void cell_bind_blend_state(struct pipe_context *, void *);
-extern void cell_delete_blend_state(struct pipe_context *, void *);
-
-extern void cell_set_blend_color( struct pipe_context *pipe,
-                                  const struct pipe_blend_color *blend_color );
-
-
-void *
-cell_create_sampler_state(struct pipe_context *,
-                          const struct pipe_sampler_state *);
-
-extern void
-cell_bind_sampler_state(struct pipe_context *, unsigned, void *);
-
-extern void
-cell_delete_sampler_state(struct pipe_context *, void *);
-
-
-extern void *
-cell_create_depth_stencil_alpha_state(struct pipe_context *,
-                                const struct pipe_depth_stencil_alpha_state *);
-
-extern void
-cell_bind_depth_stencil_alpha_state(struct pipe_context *, void *);
-
-extern void
-cell_delete_depth_stencil_alpha_state(struct pipe_context *, void *);
-
-
-void *cell_create_fs_state(struct pipe_context *,
-                               const struct pipe_shader_state *);
-void cell_bind_fs_state(struct pipe_context *, void *);
-void cell_delete_fs_state(struct pipe_context *, void *);
-void *cell_create_vs_state(struct pipe_context *,
-                               const struct pipe_shader_state *);
-void cell_bind_vs_state(struct pipe_context *, void *);
-void cell_delete_vs_state(struct pipe_context *, void *);
-
-
-void *
-cell_create_rasterizer_state(struct pipe_context *,
-                             const struct pipe_rasterizer_state *);
-void cell_bind_rasterizer_state(struct pipe_context *, void *);
-void cell_delete_rasterizer_state(struct pipe_context *, void *);
-
-
-void cell_set_clip_state( struct pipe_context *,
-                          const struct pipe_clip_state * );
-
-void cell_set_constant_buffer(struct pipe_context *pipe,
-                              uint shader, uint index,
-                              const struct pipe_constant_buffer *buf);
-
-void cell_set_polygon_stipple( struct pipe_context *,
-                               const struct pipe_poly_stipple * );
-
-void
-cell_set_sampler_texture(struct pipe_context *pipe,
-                         unsigned sampler,
-                         struct pipe_texture *texture);
-
-void cell_set_scissor_state( struct pipe_context *,
-                             const struct pipe_scissor_state * );
-
-void cell_set_texture_state( struct pipe_context *,
-                             unsigned unit, struct pipe_texture * );
-
 void cell_set_vertex_element(struct pipe_context *,
                              unsigned index,
                              const struct pipe_vertex_element *);
@@ -106,10 +56,11 @@ void cell_set_vertex_buffer(struct pipe_context *,
                             unsigned index,
                             const struct pipe_vertex_buffer *);
 
-void cell_set_viewport_state( struct pipe_context *,
-                              const struct pipe_viewport_state * );
+void cell_update_derived( struct cell_context *softpipe );
 
 
-void cell_update_derived( struct cell_context *softpipe );
+void
+cell_init_shader_functions(struct cell_context *cell);
+
+#endif /* CELL_STATE_H */
 
-#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_state_blend.c b/src/gallium/drivers/cell/ppu/cell_state_blend.c
deleted file mode 100644
index b6d6d71f0c..0000000000
--- a/src/gallium/drivers/cell/ppu/cell_state_blend.c
+++ /dev/null
@@ -1,109 +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.
- * 
- **************************************************************************/
-
-/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
- */
-
-#include "pipe/p_util.h"
-#include "draw/draw_context.h"
-#include "cell_context.h"
-#include "cell_state.h"
-
-
-
-void *
-cell_create_blend_state(struct pipe_context *pipe,
-                        const struct pipe_blend_state *blend)
-{
-   return mem_dup(blend, sizeof(*blend));
-}
-
-
-void
-cell_bind_blend_state(struct pipe_context *pipe, void *blend)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   draw_flush(cell->draw);
-
-   cell->blend = (const struct pipe_blend_state *)blend;
-
-   cell->dirty |= CELL_NEW_BLEND;
-}
-
-
-void
-cell_delete_blend_state(struct pipe_context *pipe, void *blend)
-{
-   FREE(blend);
-}
-
-
-void
-cell_set_blend_color(struct pipe_context *pipe,
-                     const struct pipe_blend_color *blend_color)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   draw_flush(cell->draw);
-
-   cell->blend_color = *blend_color;
-
-   cell->dirty |= CELL_NEW_BLEND;
-}
-
-
-
-
-void *
-cell_create_depth_stencil_alpha_state(struct pipe_context *pipe,
-                 const struct pipe_depth_stencil_alpha_state *depth_stencil)
-{
-   return mem_dup(depth_stencil, sizeof(*depth_stencil));
-}
-
-
-void
-cell_bind_depth_stencil_alpha_state(struct pipe_context *pipe,
-                                    void *depth_stencil)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   draw_flush(cell->draw);
-
-   cell->depth_stencil
-      = (const struct pipe_depth_stencil_alpha_state *) depth_stencil;
-
-   cell->dirty |= CELL_NEW_DEPTH_STENCIL;
-}
-
-
-void
-cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *depth)
-{
-   FREE(depth);
-}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_clip.c b/src/gallium/drivers/cell/ppu/cell_state_clip.c
deleted file mode 100644
index 0482f87e88..0000000000
--- a/src/gallium/drivers/cell/ppu/cell_state_clip.c
+++ /dev/null
@@ -1,84 +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.
- * 
- **************************************************************************/
-
-/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
- */
-
-#include "cell_context.h"
-#include "cell_state.h"
-#include "draw/draw_context.h"
-
-
-void cell_set_clip_state( struct pipe_context *pipe,
-			     const struct pipe_clip_state *clip )
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   /* pass the clip state to the draw module */
-   draw_set_clip_state(cell->draw, clip);
-}
-
-
-
-/* Called when driver state tracker notices changes to the viewport
- * matrix:
- */
-void cell_set_viewport_state( struct pipe_context *pipe,
-                              const struct pipe_viewport_state *viewport )
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   cell->viewport = *viewport; /* struct copy */
-   cell->dirty |= CELL_NEW_VIEWPORT;
-
-   /* pass the viewport info to the draw module */
-   draw_set_viewport_state(cell->draw, viewport);
-
-   /* Using tnl/ and vf/ modules is temporary while getting started.
-    * Full pipe will have vertex shader, vertex fetch of its own.
-    */
-}
-
-
-void cell_set_scissor_state( struct pipe_context *pipe,
-                             const struct pipe_scissor_state *scissor )
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   memcpy( &cell->scissor, scissor, sizeof(*scissor) );
-   cell->dirty |= CELL_NEW_SCISSOR;
-}
-
-
-void cell_set_polygon_stipple( struct pipe_context *pipe,
-                                   const struct pipe_poly_stipple *stipple )
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   memcpy( &cell->poly_stipple, stipple, sizeof(*stipple) );
-   cell->dirty |= CELL_NEW_STIPPLE;
-}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index 5d2a786449..49c0d130c5 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -31,6 +31,8 @@
 #include "cell_state_emit.h"
 #include "cell_batch.h"
 #include "cell_texture.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
 
 
 static void
@@ -100,4 +102,20 @@ cell_emit_state(struct cell_context *cell)
       emit_state_cmd(cell, CELL_CMD_STATE_VERTEX_INFO,
                      &cell->vertex_info, sizeof(struct vertex_info));
    }
+   
+   if (cell->dirty & CELL_NEW_VS) {
+      const struct draw_context *const draw = cell->draw;
+      struct cell_shader_info info;
+
+      info.num_outputs = draw->num_vs_outputs;
+      info.declarations = (uintptr_t) draw->machine.Declarations;
+      info.num_declarations = draw->machine.NumDeclarations;
+      info.instructions = (uintptr_t) draw->machine.Instructions;
+      info.num_instructions = draw->machine.NumInstructions;
+      info.immediates = (uintptr_t) draw->machine.Imms;
+      info.num_immediates = draw->machine.ImmLimit / 4;
+
+      emit_state_cmd(cell, CELL_CMD_STATE_BIND_VS,
+		     & info, sizeof(info));
+   }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c b/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c
deleted file mode 100644
index 7eca5b5765..0000000000
--- a/src/gallium/drivers/cell/ppu/cell_state_rasterizer.c
+++ /dev/null
@@ -1,106 +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.
- * 
- **************************************************************************/
-
-#include "pipe/p_defines.h"
-#include "pipe/p_util.h"
-#include "draw/draw_context.h"
-#include "cell_context.h"
-#include "cell_state.h"
-
-
-
-struct spu_rasterizer_state
-{
-   unsigned flatshade:1;
-#if 0
-   unsigned light_twoside:1;
-   unsigned front_winding:2;  /**< PIPE_WINDING_x */
-   unsigned cull_mode:2;      /**< PIPE_WINDING_x */
-   unsigned fill_cw:2;        /**< PIPE_POLYGON_MODE_x */
-   unsigned fill_ccw:2;       /**< PIPE_POLYGON_MODE_x */
-   unsigned offset_cw:1;
-   unsigned offset_ccw:1;
-#endif
-   unsigned scissor:1;
-   unsigned poly_smooth:1;
-   unsigned poly_stipple_enable:1;
-   unsigned point_smooth:1;
-#if 0
-   unsigned point_sprite:1;
-   unsigned point_size_per_vertex:1; /**< size computed in vertex shader */
-#endif
-   unsigned multisample:1;         /* XXX maybe more ms state in future */
-   unsigned line_smooth:1;
-   unsigned line_stipple_enable:1;
-   unsigned line_stipple_factor:8;  /**< [1..256] actually */
-   unsigned line_stipple_pattern:16;
-#if 0
-   unsigned bypass_clipping:1;
-#endif
-   unsigned origin_lower_left:1;  /**< Is (0,0) the lower-left corner? */
-
-   float line_width;
-   float point_size;           /**< used when no per-vertex size */
-#if 0
-   float offset_units;
-   float offset_scale;
-   ubyte sprite_coord_mode[PIPE_MAX_SHADER_OUTPUTS]; /**< PIPE_SPRITE_COORD_ */
-#endif
-};
-
-
-
-void *
-cell_create_rasterizer_state(struct pipe_context *pipe,
-                             const struct pipe_rasterizer_state *setup)
-{
-   struct pipe_rasterizer_state *state
-      = MALLOC(sizeof(struct pipe_rasterizer_state));
-   memcpy(state, setup, sizeof(struct pipe_rasterizer_state));
-   return state;
-}
-
-
-void
-cell_bind_rasterizer_state(struct pipe_context *pipe, void *setup)
-{
-   struct cell_context *cell = cell_context(pipe);
-
-   /* pass-through to draw module */
-   draw_set_rasterizer_state(cell->draw, setup);
-
-   cell->rasterizer = (struct pipe_rasterizer_state *)setup;
-
-   cell->dirty |= CELL_NEW_RASTERIZER;
-}
-
-
-void
-cell_delete_rasterizer_state(struct pipe_context *pipe, void *rasterizer)
-{
-   FREE(rasterizer);
-}
diff --git a/src/gallium/drivers/cell/ppu/cell_state_fs.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index b2ed699a5b..935501441b 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_fs.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -32,7 +32,7 @@
 #include "draw/draw_context.h"
 #if 0
 #include "pipe/p_shader_tokens.h"
-#include "llvm/gallivm.h"
+#include "gallivm/gallivm.h"
 #include "tgsi/util/tgsi_dump.h"
 #include "tgsi/exec/tgsi_sse2.h"
 #endif
@@ -41,7 +41,7 @@
 #include "cell_state.h"
 
 
-void *
+static void *
 cell_create_fs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
 {
@@ -80,7 +80,7 @@ cell_create_fs_state(struct pipe_context *pipe,
 }
 
 
-void
+static void
 cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 {
    struct cell_context *cell = cell_context(pipe);
@@ -91,7 +91,7 @@ cell_bind_fs_state(struct pipe_context *pipe, void *fs)
 }
 
 
-void
+static void
 cell_delete_fs_state(struct pipe_context *pipe, void *fs)
 {
    struct cell_fragment_shader_state *state =
@@ -101,7 +101,7 @@ cell_delete_fs_state(struct pipe_context *pipe, void *fs)
 }
 
 
-void *
+static void *
 cell_create_vs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
 {
@@ -124,7 +124,7 @@ cell_create_vs_state(struct pipe_context *pipe,
 }
 
 
-void
+static void
 cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct cell_context *cell = cell_context(pipe);
@@ -137,7 +137,7 @@ cell_bind_vs_state(struct pipe_context *pipe, void *vs)
 }
 
 
-void
+static void
 cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct cell_context *cell = cell_context(pipe);
@@ -150,7 +150,7 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 }
 
 
-void
+static void
 cell_set_constant_buffer(struct pipe_context *pipe,
                          uint shader, uint index,
                          const struct pipe_constant_buffer *buf)
@@ -169,3 +169,18 @@ cell_set_constant_buffer(struct pipe_context *pipe,
 
    cell->dirty |= CELL_NEW_CONSTANTS;
 }
+
+
+void
+cell_init_shader_functions(struct cell_context *cell)
+{
+   cell->pipe.create_fs_state = cell_create_fs_state;
+   cell->pipe.bind_fs_state   = cell_bind_fs_state;
+   cell->pipe.delete_fs_state = cell_delete_fs_state;
+
+   cell->pipe.create_vs_state = cell_create_vs_state;
+   cell->pipe.bind_vs_state   = cell_bind_vs_state;
+   cell->pipe.delete_vs_state = cell_delete_vs_state;
+
+   cell->pipe.set_constant_buffer = cell_set_constant_buffer;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index c8ef36002f..0edefa5f05 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -79,14 +79,16 @@ cell_texture_layout(struct cell_texture * spt)
 }
 
 
-struct pipe_texture *
-cell_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat)
+static struct pipe_texture *
+cell_texture_create(struct pipe_context *pipe,
+                    const struct pipe_texture *templat)
 {
    struct cell_texture *spt = CALLOC_STRUCT(cell_texture);
    if (!spt)
       return NULL;
 
    spt->base = *templat;
+   spt->base.refcount = 1;
 
    cell_texture_layout(spt);
 
@@ -103,7 +105,7 @@ cell_texture_create(struct pipe_context *pipe, const struct pipe_texture *templa
 }
 
 
-void
+static void
 cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
 {
    if (!*pt)
@@ -128,10 +130,18 @@ cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
 }
 
 
+static void
+cell_texture_update(struct pipe_context *pipe, struct pipe_texture *texture)
+{
+   /* XXX TO DO:  re-tile the texture data ... */
+
+}
+
+
 /**
  * Called via pipe->get_tex_surface()
  */
-struct pipe_surface *
+static struct pipe_surface *
 cell_get_tex_surface(struct pipe_context *pipe,
                          struct pipe_texture *pt,
                          unsigned face, unsigned level, unsigned zslice)
@@ -250,3 +260,13 @@ cell_update_texture_mapping(struct cell_context *cell)
    cell->tex_map = pipe_surface_map(cell->tex_surf);
 #endif
 }
+
+
+void
+cell_init_texture_functions(struct cell_context *cell)
+{
+   cell->pipe.texture_create = cell_texture_create;
+   cell->pipe.texture_release = cell_texture_release;
+   cell->pipe.texture_update = cell_texture_update;
+   cell->pipe.get_tex_surface = cell_get_tex_surface;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
index 0264fed88e..824fb3e20f 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -29,7 +29,7 @@
 #define CELL_TEXTURE_H
 
 
-struct pipe_context;
+struct cell_context;
 struct pipe_texture;
 
 
@@ -60,21 +60,12 @@ cell_texture(struct pipe_texture *pt)
 
 
 
-extern struct pipe_texture *
-cell_texture_create(struct pipe_context *pipe,
-                    const struct pipe_texture *templat);
-
 extern void
-cell_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
-
-extern struct pipe_surface *
-cell_get_tex_surface(struct pipe_context *pipe,
-                     struct pipe_texture *pt,
-                     unsigned face, unsigned level, unsigned zslice);
+cell_update_texture_mapping(struct cell_context *cell);
 
 
 extern void
-cell_update_texture_mapping(struct cell_context *cell);
+cell_init_texture_functions(struct cell_context *cell);
 
 
-#endif /* CELL_TEXTURE */
+#endif /* CELL_TEXTURE_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
index f10689a959..9cf74bab47 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c
@@ -31,7 +31,7 @@
 #include "../auxiliary/draw/draw_private.h"
 
 #include "cell_context.h"
-#include "ppc/rtasm/spe_asm.h"
+#include "rtasm/rtasm_ppc_spe.h"
 
 typedef uint64_t register_mask;
 
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
index 6a1d3bc20a..f7ef72e5a2 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -35,6 +35,7 @@
 
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
+#include "cell_flush.h"
 #include "cell_spu.h"
 #include "cell_batch.h"
 
@@ -54,6 +55,7 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
    struct cell_command_vs *const vs = &cell_global.command[0].vs;
    uint64_t *batch;
    struct cell_array_info *array_info;
+   struct cell_shader_info *shader_info;
    unsigned i, j;
    struct cell_attribute_fetch_code *cf;
 
@@ -100,17 +102,17 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
    (void) memcpy(&batch[1], &draw->viewport,
                  sizeof(struct pipe_viewport_state));
 
+   {
+      uint64_t uniforms = (uintptr_t) draw->user.constants;
+
+      batch = cell_batch_alloc(cell, 2 *sizeof(batch[0]));
+      batch[0] = CELL_CMD_STATE_UNIFORMS;
+      batch[1] = uniforms;
+   }
+
    cell_batch_flush(cell);
 
    vs->opcode = CELL_CMD_VS_EXECUTE;
-   vs->shader.num_outputs = draw->num_vs_outputs;
-   vs->shader.declarations = (uintptr_t) draw->machine.Declarations;
-   vs->shader.num_declarations = draw->machine.NumDeclarations;
-   vs->shader.instructions = (uintptr_t) draw->machine.Instructions;
-   vs->shader.num_instructions = draw->machine.NumInstructions;
-   vs->shader.uniforms = (uintptr_t) draw->user.constants;
-   vs->shader.immediates = (uintptr_t) draw->machine.Imms;
-   vs->shader.num_immediates = draw->machine.ImmLimit / 4;
    vs->nr_attrs = draw->vertex_fetch.nr_attrs;
 
    (void) memcpy(vs->plane, draw->plane, sizeof(draw->plane));
diff --git a/src/gallium/drivers/cell/spu/Makefile b/src/gallium/drivers/cell/spu/Makefile
index 30ef2450ec..c071de1900 100644
--- a/src/gallium/drivers/cell/spu/Makefile
+++ b/src/gallium/drivers/cell/spu/Makefile
@@ -18,6 +18,7 @@ PROG_SPU_EMBED_O = $(PROG)_spu-embed.o
 SOURCES = \
 	spu_main.c \
 	spu_blend.c \
+	spu_dcache.c \
 	spu_render.c \
 	spu_texture.c \
 	spu_tile.c \
diff --git a/src/gallium/drivers/cell/spu/spu_dcache.c b/src/gallium/drivers/cell/spu/spu_dcache.c
new file mode 100644
index 0000000000..698a5790bb
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_dcache.c
@@ -0,0 +1,111 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, 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 "cell/common.h"
+#include "spu_main.h"
+#include "spu_dcache.h"
+
+#define CACHELINE_LOG2SIZE    7
+#define LINE_SIZE             (1U << 7)
+#define ALIGN_MASK            (~(LINE_SIZE - 1))
+
+#define CACHE_NAME            data
+#define CACHED_TYPE           qword
+#define CACHE_TYPE            CACHE_TYPE_RO
+#define CACHE_SET_TAGID(set)  TAG_VERTEX_BUFFER
+#define CACHE_LOG2NNWAY       2
+#define CACHE_LOG2NSETS       6
+#include <cache-api.h>
+
+/* Yes folks, this is ugly.
+ */
+#undef CACHE_NWAY
+#undef CACHE_NSETS
+#define CACHE_NAME            data
+#define CACHE_NWAY            4
+#define CACHE_NSETS           (1U << 6)
+
+
+/**
+ * Fetch between arbitrary number of bytes from an unaligned address
+ */
+void
+spu_dcache_fetch_unaligned(qword *dst, unsigned ea, unsigned size)
+{
+   const int shift = ea & 0x0f;
+   const unsigned aligned_start_ea = ea & ~0x0f;
+   const unsigned aligned_end_ea = ROUNDUP16(ea + size);
+   const unsigned num_entries = (aligned_end_ea - aligned_start_ea) / 16;
+   unsigned i;
+
+
+   if (shift == 0) {
+      /* Data is already aligned.  Fetch directly into the destination buffer.
+       */
+      for (i = 0; i < num_entries; i++) {
+         dst[i] = cache_rd(data, ea + (i * 16));
+      }
+   } else {
+      qword tmp[2] ALIGN16_ATTRIB;
+
+
+      tmp[0] = cache_rd(data, (ea & ~0x0f));
+      for (i = 0; i < (num_entries & ~1); i++) {
+         const unsigned curr = i & 1;
+         const unsigned next = curr ^ 1;
+
+         tmp[next] = cache_rd(data, (ea & ~0x0f) + (next * 16));
+
+         dst[i] = si_or((qword) spu_slqwbyte(tmp[curr], shift),
+                        (qword) spu_rlmaskqwbyte(tmp[next], shift - 16));
+      }
+
+      if (i < num_entries) {
+         dst[i] = si_or((qword) spu_slqwbyte(tmp[(i & 1)], shift),
+                        si_il(0));
+      }
+   }
+}
+
+
+/**
+ * Notify the cache that a range of main memory may have been modified
+ */
+void
+spu_dcache_mark_dirty(unsigned ea, unsigned size)
+{
+   unsigned i;
+   const unsigned aligned_start = (ea & ALIGN_MASK);
+   const unsigned aligned_end = (ea + size + (LINE_SIZE - 1)) 
+       & ALIGN_MASK;
+
+
+   for (i = 0; i < (CACHE_NWAY * CACHE_NSETS); i++) {
+      const unsigned entry = __cache_dir[i];
+      const unsigned addr = entry & ~0x0f;
+
+      __cache_dir[i] = ((addr >= aligned_start) && (addr < aligned_end))
+          ? (entry & ~CACHELINE_VALID) : entry;
+   }
+}
diff --git a/src/gallium/drivers/cell/spu/spu_dcache.h b/src/gallium/drivers/cell/spu/spu_dcache.h
new file mode 100644
index 0000000000..7a06b8c25a
--- /dev/null
+++ b/src/gallium/drivers/cell/spu/spu_dcache.h
@@ -0,0 +1,34 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, 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.
+ */
+
+#ifndef SPU_DCACHE_H
+#define SPU_DCACHE_H
+
+extern void
+spu_dcache_fetch_unaligned(qword *dst, unsigned ea, unsigned size);
+
+extern void
+spu_dcache_mark_dirty(unsigned ea, unsigned size);
+
+#endif /* SPU_DCACHE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c
index 109540b1f7..cf81bee8fd 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.c
+++ b/src/gallium/drivers/cell/spu/spu_exec.c
@@ -72,6 +72,8 @@
 #include "spu_exec.h"
 #include "spu_main.h"
 #include "spu_vertex_shader.h"
+#include "spu_dcache.h"
+#include "cell/common.h"
 
 #define TILE_TOP_LEFT     0
 #define TILE_TOP_RIGHT    1
@@ -146,17 +148,13 @@ spu_exec_machine_init(struct spu_exec_machine *mach,
                       struct spu_sampler *samplers,
                       unsigned processor)
 {
-   qword zero;
-   qword not_zero;
-   uint i;
+   const qword zero = si_il(0);
+   const qword not_zero = si_il(~0);
 
    mach->Samplers = samplers;
    mach->Processor = processor;
    mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS];
 
-   zero = si_xor(zero, zero);
-   not_zero = si_xori(zero, 0xff);
-
    /* Setup constants. */
    mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].q = zero;
    mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].q = not_zero;
@@ -356,19 +354,17 @@ fetch_src_file_channel(
    case TGSI_EXTSWIZZLE_W:
       switch( file ) {
       case TGSI_FILE_CONSTANT: {
-         unsigned char buffer[32] ALIGN16_ATTRIB;
          unsigned i;
 
          for (i = 0; i < 4; i++) {
             const float *ptr = mach->Consts[index->i[i]];
-            const uint64_t addr = (uint64_t)(uintptr_t) ptr;
-            const unsigned size = ((addr & 0x0f) == 0) ? 16 : 32;
+            float tmp[4];
 
-            mfc_get(buffer, addr & ~0x0f, size, TAG_VERTEX_BUFFER, 0, 0);
-            wait_on_mask(1 << TAG_VERTEX_BUFFER);
+            spu_dcache_fetch_unaligned((qword *) tmp,
+                                       (uintptr_t)(ptr + swizzle),
+                                       sizeof(float));
 
-            (void) memcpy(& chan->f[i], &buffer[(addr & 0x0f) 
-                + (sizeof(float) * swizzle)], sizeof(float));
+            chan->f[i] = tmp[0];
          }
          break;
       }
@@ -1903,32 +1899,28 @@ spu_exec_machine_run( struct spu_exec_machine *mach )
    /* execute declarations (interpolants) */
    if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
       for (i = 0; i < mach->NumDeclarations; i++) {
-	 uint8_t buffer[sizeof(struct tgsi_full_declaration) + 32] ALIGN16_ATTRIB;
-	 struct tgsi_full_declaration decl;
-	 unsigned long decl_addr = (unsigned long) (mach->Declarations+i);
-	 unsigned size = ((sizeof(decl) + (decl_addr & 0x0f) + 0x0f) & ~0x0f);
+         union {
+            struct tgsi_full_declaration decl;
+            qword buffer[ROUNDUP16(sizeof(struct tgsi_full_declaration)) / 16];
+         } d ALIGN16_ATTRIB;
+         unsigned ea = (unsigned) (mach->Declarations + pc);
 
-	 mfc_get(buffer, decl_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0);
-	 wait_on_mask(1 << TAG_INSTRUCTION_FETCH);
+         spu_dcache_fetch_unaligned(d.buffer, ea, sizeof(d.decl));
 
-	 memcpy(& decl, buffer + (decl_addr & 0x0f), sizeof(decl));
-	 exec_declaration( mach, &decl );
+         exec_declaration( mach, &d.decl );
       }
    }
 
    /* execute instructions, until pc is set to -1 */
    while (pc != -1) {
-      uint8_t buffer[sizeof(struct tgsi_full_instruction) + 32] ALIGN16_ATTRIB;
-      struct tgsi_full_instruction inst;
-      unsigned long inst_addr = (unsigned long) (mach->Instructions + pc);
-      unsigned size = ((sizeof(inst) + (inst_addr & 0x0f) + 0x0f) & ~0x0f);
-
-      assert(pc < mach->NumInstructions);
-      mfc_get(buffer, inst_addr & ~0x0f, size, TAG_INSTRUCTION_FETCH, 0, 0);
-      wait_on_mask(1 << TAG_INSTRUCTION_FETCH);
-
-      memcpy(& inst, buffer + (inst_addr & 0x0f), sizeof(inst));
-      exec_instruction( mach, & inst, &pc );
+      union {
+         struct tgsi_full_instruction inst;
+         qword buffer[ROUNDUP16(sizeof(struct tgsi_full_instruction)) / 16];
+      } i ALIGN16_ATTRIB;
+      unsigned ea = (unsigned) (mach->Instructions + pc);
+
+      spu_dcache_fetch_unaligned(i.buffer, ea, sizeof(i.inst));
+      exec_instruction( mach, & i.inst, &pc );
    }
 
 #if 0
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index fcbf0f841e..1136dba62d 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -433,10 +433,19 @@ cmd_batch(uint opcode)
                        sizeof(struct pipe_viewport_state));
          pos += (1 + ROUNDUP8(sizeof(struct pipe_viewport_state)) / 8);
          break;
+      case CELL_CMD_STATE_UNIFORMS:
+         draw.constants = (float (*)[4]) (uintptr_t) buffer[pos + 1];
+         pos += 2;
+         break;
       case CELL_CMD_STATE_VS_ARRAY_INFO:
          cmd_state_vs_array_info((struct cell_array_info *) &buffer[pos+1]);
          pos += (1 + ROUNDUP8(sizeof(struct cell_array_info)) / 8);
          break;
+      case CELL_CMD_STATE_BIND_VS:
+         spu_bind_vertex_shader(&draw,
+                                (struct cell_shader_info *) &buffer[pos+1]);
+         pos += (1 + ROUNDUP8(sizeof(struct cell_shader_info)) / 8);
+         break;
       case CELL_CMD_STATE_ATTRIB_FETCH: {
          struct cell_attribute_fetch_code *code =
              (struct cell_attribute_fetch_code *) &buffer[pos+1];
@@ -453,6 +462,14 @@ cmd_batch(uint opcode)
          pos += (1 + ROUNDUP8(sizeof(struct cell_attribute_fetch_code)) / 8);
          break;
       }
+      case CELL_CMD_FLUSH_BUFFER_RANGE: {
+	 struct cell_buffer_range *br = (struct cell_buffer_range *)
+	     &buffer[pos+1];
+
+	 spu_dcache_mark_dirty((unsigned) br->base, br->size);
+         pos += (1 + ROUNDUP8(sizeof(struct cell_buffer_range)) / 8);
+	 break;
+      }
       default:
          printf("SPU %u: bad opcode: 0x%llx\n", spu.init.id, buffer[pos]);
          ASSERT(0);
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
index 55c6c28717..f7e4e653e3 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
@@ -40,31 +40,13 @@
 #include "spu_exec.h"
 #include "spu_vertex_shader.h"
 #include "spu_main.h"
-
-#define CACHE_NAME            attribute
-#define CACHED_TYPE           qword
-#define CACHE_TYPE            CACHE_TYPE_RO
-#define CACHE_SET_TAGID(set)  TAG_VERTEX_BUFFER
-#define CACHE_LOG2NNWAY       2
-#define CACHE_LOG2NSETS       6
-#include <cache-api.h>
-
-/* Yes folks, this is ugly.
- */
-#undef CACHE_NWAY
-#undef CACHE_NSETS
-#define CACHE_NAME            attribute
-#define CACHE_NWAY            4
-#define CACHE_NSETS           (1U << 6)
-
-
-#define DRAW_DBG 0
+#include "spu_dcache.h"
 
 typedef void (*spu_fetch_func)(qword *out, const qword *in,
 			       const qword *shuffle_data);
 
 
-static const qword fetch_shuffle_data[] = {
+static const qword fetch_shuffle_data[5] ALIGN16_ATTRIB = {
    /* Shuffle used by CVT_64_FLOAT
     */
    {
@@ -103,44 +85,6 @@ static const qword fetch_shuffle_data[] = {
 
 
 /**
- * Fetch between 1 and 32 bytes from an unaligned address
- */
-static INLINE void
-fetch_unaligned(qword *dst, unsigned ea, unsigned size)
-{
-   qword tmp[4];
-   const int shift = ea & 0x0f;
-   const unsigned aligned_start_ea = ea & ~0x0f;
-   const unsigned aligned_end_ea = (ea + size) & ~0x0f;
-   const unsigned num_entries = ((aligned_end_ea - aligned_start_ea) / 16) + 1;
-   unsigned i;
-
-
-   if (shift == 0) {
-      /* Data is already aligned.  Fetch directly into the destination buffer.
-       */
-      for (i = 0; i < num_entries; i++) {
-	 dst[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16));
-      }
-   } else {
-      /* Fetch data from the cache to the local buffer.
-       */
-      for (i = 0; i < num_entries; i++) {
-	 tmp[i] = cache_rd(attribute, (ea & ~0x0f) + (i * 16));
-      }
-
-
-      /* Fix the alignment of the data and write to the destination buffer.
-       */
-      for (i = 0; i < ((size + 15) / 16); i++) {
-	 dst[i] = si_or((qword) spu_slqwbyte(tmp[i], shift),
-			(qword) spu_rlmaskqwbyte(tmp[i + 1], shift - 16));
-      }
-   }
-}
-
-
-/**
  * Fetch vertex attributes for 'count' vertices.
  */
 static void generic_vertex_fetch(struct spu_vs_context *draw,
@@ -169,7 +113,7 @@ static void generic_vertex_fetch(struct spu_vs_context *draw,
       unsigned idx;
       const unsigned bytes_per_entry = draw->vertex_fetch.size[attr];
       const unsigned quads_per_entry = (bytes_per_entry + 15) / 16;
-      qword in[2 * 4];
+      qword in[2 * 4] ALIGN16_ATTRIB;
 
 
       /* Fetch four attributes for four vertices.  
@@ -182,7 +126,7 @@ static void generic_vertex_fetch(struct spu_vs_context *draw,
          printf("SPU: fetching = 0x%llx\n", addr);
 #endif
 
-         fetch_unaligned(& in[idx], addr, bytes_per_entry);
+         spu_dcache_fetch_unaligned(& in[idx], addr, bytes_per_entry);
          idx += quads_per_entry;
       }
 
@@ -200,15 +144,5 @@ static void generic_vertex_fetch(struct spu_vs_context *draw,
 
 void spu_update_vertex_fetch( struct spu_vs_context *draw )
 {
-   unsigned i;
-
-   
-   /* Invalidate the vertex cache.
-    */
-   for (i = 0; i < (CACHE_NWAY * CACHE_NSETS); i++) {
-      CACHELINE_CLEARVALID(i);
-   }
-
-
    draw->vertex_fetch.fetch_func = generic_vertex_fetch;
 }
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
index 3f5bf41aa2..8363efeeb6 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_shader.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c
@@ -165,63 +165,55 @@ run_vertex_program(struct spu_vs_context *draw,
 }
 
 
-static void
-spu_bind_vertex_shader(struct spu_vs_context *draw,
-		       void *uniforms,
-		       void *planes,
-		       unsigned nr_planes,
-		       unsigned num_outputs
-		       )
-{
-   draw->constants = (float (*)[4]) uniforms;
-
-   (void) memcpy(draw->plane, planes, sizeof(float) * 4 * nr_planes);
-   draw->nr_planes = nr_planes;
-   draw->num_vs_outputs = num_outputs;
-
-   /* specify the shader to interpret/execute */
-   spu_exec_machine_init(&draw->machine,
-			 PIPE_MAX_SAMPLERS,
-			 NULL /*samplers*/,
-			 PIPE_SHADER_VERTEX);
-}
-
-
 unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]
     ALIGN16_ATTRIB;
 
+
 void
-spu_execute_vertex_shader(struct spu_vs_context *draw,
-                          const struct cell_command_vs *vs)
+spu_bind_vertex_shader(struct spu_vs_context *draw,
+		       struct cell_shader_info *vs)
 {
-   unsigned i;
-
-   const uint64_t immediate_addr = vs->shader.immediates;
+   const unsigned immediate_addr = vs->immediates;
    const unsigned immediate_size = 
-       ROUNDUP16((sizeof(float) * 4 * vs->shader.num_immediates)
-                 + (immediate_addr & 0x0f));
+       ROUNDUP16((sizeof(float) * 4 * vs->num_immediates)
+		 + (immediate_addr & 0x0f));
+ 
 
    mfc_get(immediates, immediate_addr & ~0x0f, immediate_size,
            TAG_VERTEX_BUFFER, 0, 0);
 
    draw->machine.Instructions = (struct tgsi_full_instruction *)
-       vs->shader.instructions;
-   draw->machine.NumInstructions = vs->shader.num_instructions;
+       vs->instructions;
+   draw->machine.NumInstructions = vs->num_instructions;
 
    draw->machine.Declarations = (struct tgsi_full_declaration *)
-       vs->shader.declarations;
-   draw->machine.NumDeclarations = vs->shader.num_declarations;
+       vs->declarations;
+   draw->machine.NumDeclarations = vs->num_declarations;
 
-   draw->vertex_fetch.nr_attrs = vs->nr_attrs;
+   draw->num_vs_outputs = vs->num_outputs;
+
+   /* specify the shader to interpret/execute */
+   spu_exec_machine_init(&draw->machine,
+			 PIPE_MAX_SAMPLERS,
+			 NULL /*samplers*/,
+			 PIPE_SHADER_VERTEX);
 
    wait_on_mask(1 << TAG_VERTEX_BUFFER);
 
    (void) memcpy(& draw->machine.Imms, &immediates[immediate_addr & 0x0f],
-                 sizeof(float) * 4 * vs->shader.num_immediates);
+                 sizeof(float) * 4 * vs->num_immediates);
+}
 
-   spu_bind_vertex_shader(draw, vs->shader.uniforms,
-                          vs->plane, vs->nr_planes,
-                          vs->shader.num_outputs);
+
+void
+spu_execute_vertex_shader(struct spu_vs_context *draw,
+                          const struct cell_command_vs *vs)
+{
+   unsigned i;
+
+   (void) memcpy(draw->plane, vs->plane, sizeof(float) * 4 * vs->nr_planes);
+   draw->nr_planes = vs->nr_planes;
+   draw->vertex_fetch.nr_attrs = vs->nr_attrs;
 
    for (i = 0; i < vs->num_elts; i += 4) {
       const unsigned batch_size = MIN2(vs->num_elts - i, 4);
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.h b/src/gallium/drivers/cell/spu/spu_vertex_shader.h
index 0fb0bc28d0..54a4b8d9b9 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_shader.h
+++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.h
@@ -1,6 +1,7 @@
 #ifndef SPU_VERTEX_SHADER_H
 #define SPU_VERTEX_SHADER_H
 
+#include "cell/common.h"
 #include "pipe/p_format.h"
 #include "spu_exec.h"
 
@@ -55,6 +56,10 @@ static INLINE void spu_vertex_fetch(struct spu_vs_context *draw,
 struct cell_command_vs;
 
 extern void
+spu_bind_vertex_shader(struct spu_vs_context *draw,
+		       struct cell_shader_info *vs);
+
+extern void
 spu_execute_vertex_shader(struct spu_vs_context *draw,
 			  const struct cell_command_vs *vs);
 
diff --git a/src/gallium/drivers/failover/Makefile b/src/gallium/drivers/failover/Makefile
index 14389bd055..f08b8df07a 100644
--- a/src/gallium/drivers/failover/Makefile
+++ b/src/gallium/drivers/failover/Makefile
@@ -1,20 +1,13 @@
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
 LIBNAME = failover
 
-DRIVER_SOURCES = \
+C_SOURCES = \
 	fo_state.c \
 	fo_state_emit.c \
 	fo_context.c 
 
-C_SOURCES = \
-	$(COMMON_SOURCES) \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES = 
-
 include ../../Makefile.template
 
 symlinks:
diff --git a/src/gallium/drivers/failover/SConscript b/src/gallium/drivers/failover/SConscript
new file mode 100644
index 0000000000..f8e9b1b491
--- /dev/null
+++ b/src/gallium/drivers/failover/SConscript
@@ -0,0 +1,13 @@
+Import('*')
+
+env = env.Clone()
+
+failover = env.ConvenienceLibrary(
+	target = 'failover',
+	source = [
+		'fo_state.c',
+		'fo_state_emit.c',
+		'fo_context.c',
+	])
+
+Export('failover')
diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c
index 7ce4a7df17..156f7399b0 100644
--- a/src/gallium/drivers/failover/fo_context.c
+++ b/src/gallium/drivers/failover/fo_context.c
@@ -137,15 +137,14 @@ struct pipe_context *failover_create( struct pipe_context *hw,
 
    failover_init_state_functions( failover );
 
-#if 0
-   failover->pipe.surface_alloc = hw->surface_alloc;
-#endif
-   failover->pipe.get_tex_surface = hw->get_tex_surface;
-
    failover->pipe.surface_copy = hw->surface_copy;
    failover->pipe.surface_fill = hw->surface_fill;
+
    failover->pipe.texture_create = hw->texture_create;
    failover->pipe.texture_release = hw->texture_release;
+   failover->pipe.texture_update = hw->texture_update;
+   failover->pipe.get_tex_surface = hw->get_tex_surface;
+
    failover->pipe.flush = hw->flush;
 
    failover->dirty = 0;
diff --git a/src/gallium/drivers/i915simple/Makefile b/src/gallium/drivers/i915simple/Makefile
index ee22ba86f9..2a75f5d57c 100644
--- a/src/gallium/drivers/i915simple/Makefile
+++ b/src/gallium/drivers/i915simple/Makefile
@@ -1,10 +1,9 @@
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
 LIBNAME = i915simple
 
-DRIVER_SOURCES = \
+C_SOURCES = \
 	i915_blit.c \
 	i915_clear.c \
 	i915_flush.c \
@@ -26,12 +25,6 @@ DRIVER_SOURCES = \
 	i915_fpc_translate.c \
 	i915_surface.c 
 
-C_SOURCES = \
-	$(COMMON_SOURCES) \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES = 
-
 include ../../Makefile.template
 
 symlinks:
diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c
index 7f71f8fd4f..acfa349439 100644
--- a/src/gallium/drivers/i915simple/i915_context.c
+++ b/src/gallium/drivers/i915simple/i915_context.c
@@ -296,12 +296,11 @@ struct pipe_context *i915_create( struct pipe_winsys *pipe_winsys,
    i915_init_state_functions(i915);
    i915_init_flush_functions(i915);
    i915_init_string_functions(i915);
+   i915_init_texture_functions(i915);
 
    i915->pci_id = pci_id;
    i915->flags.is_i945 = is_i945;
 
-   i915->pipe.texture_create = i915_texture_create;
-   i915->pipe.texture_release = i915_texture_release;
 
    i915->dirty = ~0;
    i915->hardware_dirty = ~0;
diff --git a/src/gallium/drivers/i915simple/i915_debug.c b/src/gallium/drivers/i915simple/i915_debug.c
index 94db44e1aa..78102dbac2 100644
--- a/src/gallium/drivers/i915simple/i915_debug.c
+++ b/src/gallium/drivers/i915simple/i915_debug.c
@@ -25,8 +25,6 @@
  * 
  **************************************************************************/
 
-//#include "imports.h"
-
 #include "i915_reg.h"
 #include "i915_context.h"
 #include "i915_winsys.h"
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
index 294e6fad03..e055eed7e0 100644
--- a/src/gallium/drivers/i915simple/i915_state.c
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -32,6 +32,7 @@
 #include "draw/draw_context.h"
 #include "pipe/p_winsys.h"
 #include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
@@ -505,7 +506,9 @@ static void i915_set_sampler_texture(struct pipe_context *pipe,
 {
    struct i915_context *i915 = i915_context(pipe);
 
-   i915->texture[sampler] = (struct i915_texture*)texture;  /* ptr, not struct */
+   pipe_texture_reference(pipe,
+                          (struct pipe_texture **) &i915->texture[sampler],
+                          texture);
 
    i915->dirty |= I915_NEW_TEXTURE;
 }
diff --git a/src/gallium/drivers/i915simple/i915_state_immediate.c b/src/gallium/drivers/i915simple/i915_state_immediate.c
index 07031fc6c5..dfbbcab624 100644
--- a/src/gallium/drivers/i915simple/i915_state_immediate.c
+++ b/src/gallium/drivers/i915simple/i915_state_immediate.c
@@ -33,7 +33,7 @@
 #include "i915_context.h"
 #include "i915_state.h"
 #include "i915_reg.h"
-#include "p_util.h"
+#include "pipe/p_util.h"
 
 
 /* All state expressable with the LOAD_STATE_IMMEDIATE_1 packet.
diff --git a/src/gallium/drivers/i915simple/i915_state_inlines.h b/src/gallium/drivers/i915simple/i915_state_inlines.h
index 0934ac79a4..378de8f9c4 100644
--- a/src/gallium/drivers/i915simple/i915_state_inlines.h
+++ b/src/gallium/drivers/i915simple/i915_state_inlines.h
@@ -28,8 +28,8 @@
 #ifndef I915_STATE_INLINES_H
 #define I915_STATE_INLINES_H
 
-#include "p_compiler.h"
-#include "p_defines.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_defines.h"
 #include "i915_reg.h"
 
 
diff --git a/src/gallium/drivers/i915simple/i915_surface.c b/src/gallium/drivers/i915simple/i915_surface.c
index 17fd27895a..f4fbedbe9b 100644
--- a/src/gallium/drivers/i915simple/i915_surface.c
+++ b/src/gallium/drivers/i915simple/i915_surface.c
@@ -36,48 +36,6 @@
 #include "util/p_tile.h"
 
 
-/*
- * XXX note: same as code in sp_surface.c
- */
-static struct pipe_surface *
-i915_get_tex_surface(struct pipe_context *pipe,
-                     struct pipe_texture *pt,
-                     unsigned face, unsigned level, unsigned zslice)
-{
-   struct i915_texture *tex = (struct i915_texture *)pt;
-   struct pipe_surface *ps;
-   unsigned offset;  /* in bytes */
-
-   offset = tex->level_offset[level];
-
-   if (pt->target == PIPE_TEXTURE_CUBE) {
-      offset += tex->image_offset[level][face] * pt->cpp;
-   }
-   else if (pt->target == PIPE_TEXTURE_3D) {
-      offset += tex->image_offset[level][zslice] * pt->cpp;
-   }
-   else {
-      assert(face == 0);
-      assert(zslice == 0);
-   }
-
-   ps = pipe->winsys->surface_alloc(pipe->winsys);
-   if (ps) {
-      assert(ps->refcount);
-      assert(ps->winsys);
-      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
-      ps->format = pt->format;
-      ps->cpp = pt->cpp;
-      ps->width = pt->width[level];
-      ps->height = pt->height[level];
-      ps->pitch = tex->pitch;
-      ps->offset = offset;
-   }
-   return ps;
-}
-
-
-
 /* Assumes all values are within bounds -- no checking at this level -
  * do it higher up if required.
  */
@@ -115,6 +73,7 @@ i915_surface_copy(struct pipe_context *pipe,
    }
 }
 
+
 /* Fill a rectangular sub-region.  Need better logic about when to
  * push buffers into AGP - will currently do so whenever possible.
  */
@@ -184,8 +143,6 @@ i915_surface_fill(struct pipe_context *pipe,
 void
 i915_init_surface_functions(struct i915_context *i915)
 {
-   i915->pipe.get_tex_surface = i915_get_tex_surface;
-
    i915->pipe.surface_copy = i915_surface_copy;
    i915->pipe.surface_fill = i915_surface_fill;
 }
diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c
index 6d37ae3d74..1b415a94d4 100644
--- a/src/gallium/drivers/i915simple/i915_texture.c
+++ b/src/gallium/drivers/i915simple/i915_texture.c
@@ -478,7 +478,7 @@ i945_miptree_layout(struct pipe_context *pipe, struct i915_texture * tex)
 }
 
 
-struct pipe_texture *
+static struct pipe_texture *
 i915_texture_create(struct pipe_context *pipe,
                     const struct pipe_texture *templat)
 {
@@ -488,6 +488,7 @@ i915_texture_create(struct pipe_context *pipe,
       struct i915_context *i915 = i915_context(pipe);
 
       tex->base = *templat;
+      tex->base.refcount = 1;
 
       if (i915->flags.is_i945 ? i945_miptree_layout(pipe, tex) :
 	  i915_miptree_layout(pipe, tex))
@@ -506,7 +507,7 @@ i915_texture_create(struct pipe_context *pipe,
 }
 
 
-void
+static void
 i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
 {
    if (!*pt)
@@ -534,3 +535,62 @@ i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
    }
    *pt = NULL;
 }
+
+
+static void
+i915_texture_update(struct pipe_context *pipe, struct pipe_texture *texture)
+{
+   /* no-op? */
+}
+
+
+/*
+ * XXX note: same as code in sp_surface.c
+ */
+static struct pipe_surface *
+i915_get_tex_surface(struct pipe_context *pipe,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice)
+{
+   struct i915_texture *tex = (struct i915_texture *)pt;
+   struct pipe_surface *ps;
+   unsigned offset;  /* in bytes */
+
+   offset = tex->level_offset[level];
+
+   if (pt->target == PIPE_TEXTURE_CUBE) {
+      offset += tex->image_offset[level][face] * pt->cpp;
+   }
+   else if (pt->target == PIPE_TEXTURE_3D) {
+      offset += tex->image_offset[level][zslice] * pt->cpp;
+   }
+   else {
+      assert(face == 0);
+      assert(zslice == 0);
+   }
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->refcount);
+      assert(ps->winsys);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = tex->pitch;
+      ps->offset = offset;
+   }
+   return ps;
+}
+
+
+
+void
+i915_init_texture_functions(struct i915_context *i915)
+{
+   i915->pipe.texture_create = i915_texture_create;
+   i915->pipe.texture_release = i915_texture_release;
+   i915->pipe.texture_update = i915_texture_update;
+   i915->pipe.get_tex_surface = i915_get_tex_surface;
+}
diff --git a/src/gallium/drivers/i915simple/i915_texture.h b/src/gallium/drivers/i915simple/i915_texture.h
index 330d111dc7..6d8d41178f 100644
--- a/src/gallium/drivers/i915simple/i915_texture.h
+++ b/src/gallium/drivers/i915simple/i915_texture.h
@@ -1,17 +1,38 @@
+/**************************************************************************
+ * 
+ * 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 I915_TEXTURE_H
 #define I915_TEXTURE_H
 
 struct pipe_context;
-struct pipe_texture;
 
 
-struct pipe_texture *
-i915_texture_create(struct pipe_context *pipe,
-                    const struct pipe_texture *templat);
-
 extern void
-i915_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+i915_init_texture_functions(struct i915_context *i915);
 
 
 #endif /* I915_TEXTURE_H */
diff --git a/src/gallium/drivers/i965simple/Makefile b/src/gallium/drivers/i965simple/Makefile
index 1dec1f9749..cc8580836c 100644
--- a/src/gallium/drivers/i965simple/Makefile
+++ b/src/gallium/drivers/i965simple/Makefile
@@ -1,14 +1,13 @@
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
 LIBNAME = i965simple
 
-DRIVER_SOURCES = \
-        brw_blit.c \
-        brw_flush.c \
-        brw_strings.c \
-        brw_surface.c \
+C_SOURCES = \
+	brw_blit.c \
+	brw_flush.c \
+	brw_strings.c \
+	brw_surface.c \
 	brw_cc.c \
 	brw_clip.c \
 	brw_clip_line.c \
@@ -31,8 +30,8 @@ DRIVER_SOURCES = \
 	brw_sf.c \
 	brw_sf_emit.c \
 	brw_sf_state.c \
-        brw_shader_info.c \
-        brw_state.c \
+	brw_shader_info.c \
+	brw_state.c \
 	brw_state_batch.c \
 	brw_state_cache.c \
 	brw_state_pool.c \
@@ -51,16 +50,6 @@ DRIVER_SOURCES = \
 	brw_wm_state.c \
 	brw_wm_surface_state.c
 
-C_SOURCES = \
-	$(COMMON_SOURCES) \
-	$(COMMON_BM_SOURCES) \
-	$(MINIGLX_SOURCES) \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES =
-
-DRIVER_DEFINES = -I.
-
 include ../../Makefile.template
 
 symlinks:
diff --git a/src/gallium/drivers/i965simple/brw_context.c b/src/gallium/drivers/i965simple/brw_context.c
index 5e58701e91..6fb840708e 100644
--- a/src/gallium/drivers/i965simple/brw_context.c
+++ b/src/gallium/drivers/i965simple/brw_context.c
@@ -222,10 +222,9 @@ struct pipe_context *brw_create(struct pipe_winsys *pipe_winsys,
    brw->pipe.get_param = brw_get_param;
    brw->pipe.get_paramf = brw_get_paramf;
    brw->pipe.clear = brw_clear;
-   brw->pipe.texture_create  = brw_texture_create;
-   brw->pipe.texture_release = brw_texture_release;
 
    brw_init_surface_functions(brw);
+   brw_init_texture_functions(brw);
    brw_init_state_functions(brw);
    brw_init_flush_functions(brw);
    brw_init_string_functions(brw);
diff --git a/src/gallium/drivers/i965simple/brw_state.c b/src/gallium/drivers/i965simple/brw_state.c
index f746d1cc57..2fc048bde0 100644
--- a/src/gallium/drivers/i965simple/brw_state.c
+++ b/src/gallium/drivers/i965simple/brw_state.c
@@ -32,6 +32,7 @@
 
 #include "pipe/p_winsys.h"
 #include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/util/tgsi_dump.h"
 
@@ -327,7 +328,9 @@ static void brw_set_sampler_texture(struct pipe_context *pipe,
 {
    struct brw_context *brw = brw_context(pipe);
 
-   brw->attribs.Texture[unit] = (struct brw_texture*)texture;  /* ptr, not struct */
+   pipe_texture_reference(pipe,
+                          (struct pipe_texture **) &brw->attribs.Texture[unit],
+                          texture);
 
    brw->state.dirty.brw |= BRW_NEW_TEXTURE;
 }
diff --git a/src/gallium/drivers/i965simple/brw_surface.c b/src/gallium/drivers/i965simple/brw_surface.c
index 376a42b1a6..dc4846d39f 100644
--- a/src/gallium/drivers/i965simple/brw_surface.c
+++ b/src/gallium/drivers/i965simple/brw_surface.c
@@ -35,47 +35,6 @@
 #include "util/p_tile.h"
 
 
-/*
- * XXX note: same as code in sp_surface.c
- */
-static struct pipe_surface *
-brw_get_tex_surface(struct pipe_context *pipe,
-                     struct pipe_texture *pt,
-                     unsigned face, unsigned level, unsigned zslice)
-{
-   struct brw_texture *tex = (struct brw_texture *)pt;
-   struct pipe_surface *ps;
-   unsigned offset;  /* in bytes */
-
-   offset = tex->level_offset[level];
-
-   if (pt->target == PIPE_TEXTURE_CUBE) {
-      offset += tex->image_offset[level][face] * pt->cpp;
-   }
-   else if (pt->target == PIPE_TEXTURE_3D) {
-      offset += tex->image_offset[level][zslice] * pt->cpp;
-   }
-   else {
-      assert(face == 0);
-      assert(zslice == 0);
-   }
-
-   ps = pipe->winsys->surface_alloc(pipe->winsys);
-   if (ps) {
-      assert(ps->format);
-      assert(ps->refcount);
-      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
-      ps->format = pt->format;
-      ps->cpp = pt->cpp;
-      ps->width = pt->width[level];
-      ps->height = pt->height[level];
-      ps->pitch = tex->pitch;
-      ps->offset = offset;
-   }
-   return ps;
-}
-
-
 /* Upload data to a rectangular sub-region.  Lots of choices how to do this:
  *
  * - memcpy by span to current destination
@@ -201,10 +160,10 @@ brw_surface_fill(struct pipe_context *pipe,
    }
 }
 
+
 void
 brw_init_surface_functions(struct brw_context *brw)
 {
-   brw->pipe.get_tex_surface = brw_get_tex_surface;
    brw->pipe.surface_copy  = brw_surface_copy;
    brw->pipe.surface_fill  = brw_surface_fill;
 }
diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.c b/src/gallium/drivers/i965simple/brw_tex_layout.c
index 90561f1307..86ce3d0cc3 100644
--- a/src/gallium/drivers/i965simple/brw_tex_layout.c
+++ b/src/gallium/drivers/i965simple/brw_tex_layout.c
@@ -300,13 +300,15 @@ static boolean brw_miptree_layout(struct pipe_context *pipe, struct brw_texture
 }
 
 
-struct pipe_texture *
-brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat)
+static struct pipe_texture *
+brw_texture_create(struct pipe_context *pipe,
+                   const struct pipe_texture *templat)
 {
    struct brw_texture *tex = CALLOC_STRUCT(brw_texture);
 
    if (tex) {
       tex->base = *templat;
+      tex->base.refcount = 1;
 
       if (brw_miptree_layout(pipe, tex))
 	 tex->buffer = pipe->winsys->buffer_create(pipe->winsys, 64,
@@ -323,7 +325,8 @@ brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat
    return &tex->base;
 }
 
-void
+
+static void
 brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
 {
    if (!*pt)
@@ -351,3 +354,61 @@ brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
    }
    *pt = NULL;
 }
+
+
+static void
+brw_texture_update(struct pipe_context *pipe, struct pipe_texture *texture)
+{
+   /* no-op? */
+}
+
+
+/*
+ * XXX note: same as code in sp_surface.c
+ */
+static struct pipe_surface *
+brw_get_tex_surface(struct pipe_context *pipe,
+                     struct pipe_texture *pt,
+                     unsigned face, unsigned level, unsigned zslice)
+{
+   struct brw_texture *tex = (struct brw_texture *)pt;
+   struct pipe_surface *ps;
+   unsigned offset;  /* in bytes */
+
+   offset = tex->level_offset[level];
+
+   if (pt->target == PIPE_TEXTURE_CUBE) {
+      offset += tex->image_offset[level][face] * pt->cpp;
+   }
+   else if (pt->target == PIPE_TEXTURE_3D) {
+      offset += tex->image_offset[level][zslice] * pt->cpp;
+   }
+   else {
+      assert(face == 0);
+      assert(zslice == 0);
+   }
+
+   ps = pipe->winsys->surface_alloc(pipe->winsys);
+   if (ps) {
+      assert(ps->format);
+      assert(ps->refcount);
+      pipe_buffer_reference(pipe->winsys, &ps->buffer, tex->buffer);
+      ps->format = pt->format;
+      ps->cpp = pt->cpp;
+      ps->width = pt->width[level];
+      ps->height = pt->height[level];
+      ps->pitch = tex->pitch;
+      ps->offset = offset;
+   }
+   return ps;
+}
+
+
+void
+brw_init_texture_functions(struct brw_context *brw)
+{
+   brw->pipe.texture_create  = brw_texture_create;
+   brw->pipe.texture_release = brw_texture_release;
+   brw->pipe.texture_update = brw_texture_update;
+   brw->pipe.get_tex_surface = brw_get_tex_surface;
+}
diff --git a/src/gallium/drivers/i965simple/brw_tex_layout.h b/src/gallium/drivers/i965simple/brw_tex_layout.h
index cfd6b1ef3a..ed49baeef8 100644
--- a/src/gallium/drivers/i965simple/brw_tex_layout.h
+++ b/src/gallium/drivers/i965simple/brw_tex_layout.h
@@ -1,15 +1,12 @@
 #ifndef BRW_TEX_LAYOUT_H
 #define BRW_TEX_LAYOUT_H
 
-#include "pipe/p_compiler.h"
 
-struct pipe_context;
-struct pipe_texture;
+struct brw_context;
 
-extern struct pipe_texture *
-brw_texture_create(struct pipe_context *pipe, const struct pipe_texture *templat);
 
 extern void
-brw_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
+brw_init_texture_functions(struct brw_context *brw);
+
 
 #endif
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index 5479daf8ea..539ffb77f5 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -1,10 +1,9 @@
-
 TOP = ../../../..
 include $(TOP)/configs/current
 
 LIBNAME = softpipe
 
-DRIVER_SOURCES = \
+C_SOURCES = \
 	sp_fs_exec.c \
 	sp_fs_sse.c \
 	sp_fs_llvm.c \
@@ -41,12 +40,6 @@ DRIVER_SOURCES = \
 	sp_tile_cache.c \
 	sp_surface.c 
 
-C_SOURCES = \
-	$(COMMON_SOURCES) \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES = 
-
 include ../../Makefile.template
 
 symlinks:
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 5e98f190bb..2cdf3c75bf 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -283,6 +283,7 @@ struct pipe_context *softpipe_create( struct pipe_winsys *pipe_winsys,
    /* textures */
    softpipe->pipe.texture_create = softpipe_texture_create;
    softpipe->pipe.texture_release = softpipe_texture_release;
+   softpipe->pipe.texture_update = softpipe_texture_update;
    softpipe->pipe.get_tex_surface = softpipe_get_tex_surface;
 
    /*
@@ -327,6 +328,18 @@ struct pipe_context *softpipe_create( struct pipe_winsys *pipe_winsys,
       draw_set_rasterize_stage(softpipe->draw, softpipe->setup);
    }
 
+   /* plug in AA line/point stages */
+   draw_install_aaline_stage(softpipe->draw, &softpipe->pipe);
+   draw_install_aapoint_stage(softpipe->draw, &softpipe->pipe);
+
+#if USE_DRAW_STAGE_PSTIPPLE
+   /* Do polygon stipple w/ texture map + frag prog? */
+   draw_install_pstipple_stage(softpipe->draw, &softpipe->pipe);
+#endif
+
+   /* sp_prim_setup can do wide points (don't convert to quads) */
+   draw_convert_wide_points(softpipe->draw, FALSE);
+
    sp_init_surface_functions(softpipe);
 
    return &softpipe->pipe;
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index b70d4fea85..feeafc7084 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -39,6 +39,13 @@
 #include "sp_quad.h"
 
 
+/**
+ * This is a temporary variable for testing draw-stage polygon stipple.
+ * If zero, do stipple in sp_quad_stipple.c
+ */
+#define USE_DRAW_STAGE_PSTIPPLE 1
+
+
 struct softpipe_winsys;
 struct softpipe_vbuf_render;
 struct draw_context;
@@ -68,7 +75,7 @@ struct softpipe_context {
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
-   struct softpipe_texture *texture[PIPE_MAX_SAMPLERS];
+   struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index 9ad30a7681..8cb0534342 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -44,6 +44,41 @@ struct sp_exec_fragment_shader {
 
 
 
+/**
+ * Compute quad X,Y,Z,W for the four fragments in a quad.
+ *
+ * This should really be part of the compiled shader.
+ */
+void
+sp_setup_pos_vector(const struct tgsi_interp_coef *coef,
+		    float x, float y,
+		    struct tgsi_exec_vector *quadpos)
+{
+   uint chan;
+   /* do X */
+   quadpos->xyzw[0].f[0] = x;
+   quadpos->xyzw[0].f[1] = x + 1;
+   quadpos->xyzw[0].f[2] = x;
+   quadpos->xyzw[0].f[3] = x + 1;
+
+   /* do Y */
+   quadpos->xyzw[1].f[0] = y;
+   quadpos->xyzw[1].f[1] = y;
+   quadpos->xyzw[1].f[2] = y + 1;
+   quadpos->xyzw[1].f[3] = y + 1;
+
+   /* do Z and W for all fragments in the quad */
+   for (chan = 2; chan < 4; chan++) {
+      const float dadx = coef->dadx[chan];
+      const float dady = coef->dady[chan];
+      const float a0 = coef->a0[chan] + dadx * x + dady * y;
+      quadpos->xyzw[chan].f[0] = a0;
+      quadpos->xyzw[chan].f[1] = a0 + dadx;
+      quadpos->xyzw[chan].f[2] = a0 + dady;
+      quadpos->xyzw[chan].f[3] = a0 + dadx + dady;
+   }
+}
+
 
 static void
 exec_prepare( struct sp_fragment_shader *base,
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index d90066e025..8095d662ee 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -42,7 +42,7 @@
 
 #if defined(__i386__) || defined(__386__)
 
-#include "x86/rtasm/x86sse.h"
+#include "rtasm/rtasm_x86sse.h"
 
 /* Surely this should be defined somewhere in a tgsi header:
  */
@@ -63,41 +63,6 @@ struct sp_sse_fragment_shader {
 };
 
 
-/**
- * Compute quad X,Y,Z,W for the four fragments in a quad.
- *
- * This should really be part of the compiled shader.
- */
-void
-sp_setup_pos_vector(const struct tgsi_interp_coef *coef,
-		    float x, float y,
-		    struct tgsi_exec_vector *quadpos)
-{
-   uint chan;
-   /* do X */
-   quadpos->xyzw[0].f[0] = x;
-   quadpos->xyzw[0].f[1] = x + 1;
-   quadpos->xyzw[0].f[2] = x;
-   quadpos->xyzw[0].f[3] = x + 1;
-
-   /* do Y */
-   quadpos->xyzw[1].f[0] = y;
-   quadpos->xyzw[1].f[1] = y;
-   quadpos->xyzw[1].f[2] = y + 1;
-   quadpos->xyzw[1].f[3] = y + 1;
-
-   /* do Z and W for all fragments in the quad */
-   for (chan = 2; chan < 4; chan++) {
-      const float dadx = coef->dadx[chan];
-      const float dady = coef->dady[chan];
-      const float a0 = coef->a0[chan] + dadx * x + dady * y;
-      quadpos->xyzw[chan].f[0] = a0;
-      quadpos->xyzw[chan].f[1] = a0 + dadx;
-      quadpos->xyzw[chan].f[2] = a0 + dady;
-      quadpos->xyzw[chan].f[3] = a0 + dadx + dady;
-   }
-}
-
 
 static void
 fs_sse_prepare( struct sp_fragment_shader *base,
@@ -124,6 +89,9 @@ fs_sse_run( struct sp_fragment_shader *base,
 		       (float)quad->x0, (float)quad->y0, 
 		       machine->Temps);
 
+   /* init kill mask */
+   machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] = 0x0;
+
    shader->func( machine->Inputs,
 		 machine->Outputs,
 		 machine->Consts,
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
index d73521ccbe..7b1e131ee1 100644
--- a/src/gallium/drivers/softpipe/sp_prim_setup.c
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.c
@@ -476,33 +476,33 @@ static void tri_persp_coeff( struct setup_stage *setup,
  * We could do a bit less work if we'd examine gl_FragCoord's swizzle mask.
  */
 static void
-setup_fragcoord_coeff(struct setup_stage *setup)
+setup_fragcoord_coeff(struct setup_stage *setup, uint slot)
 {
    /*X*/
-   setup->coef[0].a0[0] = 0;
-   setup->coef[0].dadx[0] = 1.0;
-   setup->coef[0].dady[0] = 0.0;
+   setup->coef[slot].a0[0] = 0;
+   setup->coef[slot].dadx[0] = 1.0;
+   setup->coef[slot].dady[0] = 0.0;
    /*Y*/
    if (setup->softpipe->rasterizer->origin_lower_left) {
       /* y=0=bottom */
       const int winHeight = setup->softpipe->framebuffer.cbufs[0]->height;
-      setup->coef[0].a0[1] = (float) (winHeight - 1);
-      setup->coef[0].dady[1] = -1.0;
+      setup->coef[slot].a0[1] = (float) (winHeight - 1);
+      setup->coef[slot].dady[1] = -1.0;
    }
    else {
       /* y=0=top */
-      setup->coef[0].a0[1] = 0.0;
-      setup->coef[0].dady[1] = 1.0;
+      setup->coef[slot].a0[1] = 0.0;
+      setup->coef[slot].dady[1] = 1.0;
    }
-   setup->coef[0].dadx[1] = 0.0;
+   setup->coef[slot].dadx[1] = 0.0;
    /*Z*/
-   setup->coef[0].a0[2] = setup->posCoef.a0[2];
-   setup->coef[0].dadx[2] = setup->posCoef.dadx[2];
-   setup->coef[0].dady[2] = setup->posCoef.dady[2];
+   setup->coef[slot].a0[2] = setup->posCoef.a0[2];
+   setup->coef[slot].dadx[2] = setup->posCoef.dadx[2];
+   setup->coef[slot].dady[2] = setup->posCoef.dady[2];
    /*W*/
-   setup->coef[0].a0[3] = setup->posCoef.a0[3];
-   setup->coef[0].dadx[3] = setup->posCoef.dadx[3];
-   setup->coef[0].dady[3] = setup->posCoef.dady[3];
+   setup->coef[slot].a0[3] = setup->posCoef.a0[3];
+   setup->coef[slot].dadx[3] = setup->posCoef.dadx[3];
+   setup->coef[slot].dady[3] = setup->posCoef.dady[3];
 }
 
 
@@ -543,8 +543,7 @@ static void setup_tri_coefficients( struct setup_stage *setup )
             tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
          break;
       case INTERP_POS:
-         assert(fragSlot == 0);
-         setup_fragcoord_coeff(setup);
+         setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
@@ -798,9 +797,7 @@ setup_line_coefficients(struct setup_stage *setup, struct prim_header *prim)
             line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
          break;
       case INTERP_POS:
-         assert(fragSlot == 0);
-         assert(0); /* XXX fix this: */
-         setup_fragcoord_coeff(setup);
+         setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
@@ -1022,9 +1019,7 @@ setup_point(struct draw_stage *stage, struct prim_header *prim)
                               &setup->coef[fragSlot], vertSlot, j);
          break;
       case INTERP_POS:
-         assert(fragSlot == 0);
-         assert(0); /* XXX fix this: */
-         setup_fragcoord_coeff(setup);
+         setup_fragcoord_coeff(setup, fragSlot);
          break;
       default:
          assert(0);
diff --git a/src/gallium/drivers/softpipe/sp_quad.c b/src/gallium/drivers/softpipe/sp_quad.c
index 6bd468a51c..15b5594547 100644
--- a/src/gallium/drivers/softpipe/sp_quad.c
+++ b/src/gallium/drivers/softpipe/sp_quad.c
@@ -112,7 +112,9 @@ sp_build_quad_pipeline(struct softpipe_context *sp)
       sp_push_quad_first( sp, sp->quad.earlyz );
    }
 
+#if !USE_DRAW_STAGE_PSTIPPLE
    if (sp->rasterizer->poly_stipple_enable) {
       sp_push_quad_first( sp, sp->quad.polygon_stipple );
    }
+#endif
 }
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index cf1b1eff75..2f40e09d5c 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -142,7 +142,7 @@ static void shade_begin(struct quad_stage *qs)
    /* set TGSI sampler state that varies */
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
       qss->samplers[i].state = softpipe->sampler[i];
-      qss->samplers[i].texture = &softpipe->texture[i]->base;
+      qss->samplers[i].texture = softpipe->texture[i];
    }
 
    /* find output slots for depth, color */
diff --git a/src/gallium/drivers/softpipe/sp_state_clip.c b/src/gallium/drivers/softpipe/sp_state_clip.c
index c797c0dd3b..4946c776e3 100644
--- a/src/gallium/drivers/softpipe/sp_state_clip.c
+++ b/src/gallium/drivers/softpipe/sp_state_clip.c
@@ -42,24 +42,16 @@ void softpipe_set_clip_state( struct pipe_context *pipe,
 }
 
 
-
-/* Called when driver state tracker notices changes to the viewport
- * matrix:
- */
 void softpipe_set_viewport_state( struct pipe_context *pipe,
                                   const struct pipe_viewport_state *viewport )
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->viewport = *viewport; /* struct copy */
-   softpipe->dirty |= SP_NEW_VIEWPORT;
-
    /* pass the viewport info to the draw module */
    draw_set_viewport_state(softpipe->draw, viewport);
 
-   /* Using tnl/ and vf/ modules is temporary while getting started.
-    * Full pipe will have vertex shader, vertex fetch of its own.
-    */
+   softpipe->viewport = *viewport; /* struct copy */
+   softpipe->dirty |= SP_NEW_VIEWPORT;
 }
 
 
@@ -68,7 +60,9 @@ void softpipe_set_scissor_state( struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   memcpy( &softpipe->scissor, scissor, sizeof(*scissor) );
+   draw_flush(softpipe->draw);
+
+   softpipe->scissor = *scissor; /* struct copy */
    softpipe->dirty |= SP_NEW_SCISSOR;
 }
 
@@ -78,6 +72,8 @@ void softpipe_set_polygon_stipple( struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   memcpy( &softpipe->poly_stipple, stipple, sizeof(*stipple) );
+   draw_flush(softpipe->draw);
+
+   softpipe->poly_stipple = *stipple; /* struct copy */
    softpipe->dirty |= SP_NEW_STIPPLE;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index 9d8fd8b750..f9f2c5eaa8 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -44,7 +44,8 @@
  * condition that users shouldn't hit anyway.
  */
 static int
-find_vs_output(const struct pipe_shader_state *vs,
+find_vs_output(struct softpipe_context *sp,
+               const struct pipe_shader_state *vs,
                uint semantic_name,
                uint semantic_index)
 {
@@ -54,7 +55,9 @@ find_vs_output(const struct pipe_shader_state *vs,
           vs->output_semantic_index[i] == semantic_index)
          return i;
    }
-   return 0;
+
+   /* See if the draw module is introducing a new attribute... */
+   return draw_find_vs_output(sp->draw, semantic_name, semantic_index);
 }
 
 
@@ -111,24 +114,24 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
          int src;
          switch (fs->input_semantic_name[i]) {
          case TGSI_SEMANTIC_POSITION:
-            src = find_vs_output(vs, TGSI_SEMANTIC_POSITION, 0);
+            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_POSITION, 0);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src);
             break;
 
          case TGSI_SEMANTIC_COLOR:
-            src = find_vs_output(vs, TGSI_SEMANTIC_COLOR, 
+            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_COLOR, 
                                  fs->input_semantic_index[i]);
             draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
             break;
 
          case TGSI_SEMANTIC_FOG:
-            src = find_vs_output(vs, TGSI_SEMANTIC_FOG, 0);
+            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_FOG, 0);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
             break;
 
          case TGSI_SEMANTIC_GENERIC:
             /* this includes texcoords and varying vars */
-            src = find_vs_output(vs, TGSI_SEMANTIC_GENERIC,
+            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_GENERIC,
                                  fs->input_semantic_index[i]);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
             break;
@@ -138,7 +141,7 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
          }
       }
 
-      softpipe->psize_slot = find_vs_output(vs, TGSI_SEMANTIC_PSIZE, 0);
+      softpipe->psize_slot = find_vs_output(softpipe, vs, TGSI_SEMANTIC_PSIZE, 0);
       if (softpipe->psize_slot > 0) {
          draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_CONSTANT,
                                softpipe->psize_slot);
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index 460adccec4..18669a1c6e 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -30,6 +30,7 @@
  */
 
 #include "pipe/p_util.h"
+#include "pipe/p_inlines.h"
 
 #include "draw/draw_context.h"
 
@@ -82,9 +83,9 @@ softpipe_set_sampler_texture(struct pipe_context *pipe,
    draw_flush(softpipe->draw);
 
    assert(unit < PIPE_MAX_SAMPLERS);
-   softpipe->texture[unit] = softpipe_texture(texture);  /* ptr, not struct */
+   pipe_texture_reference(pipe, &softpipe->texture[unit], texture);
 
-   sp_tile_cache_set_texture(softpipe->tex_cache[unit], texture);
+   sp_tile_cache_set_texture(pipe, softpipe->tex_cache[unit], texture);
 
    softpipe->dirty |= SP_NEW_TEXTURE;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c
index e2c6893e9f..124b18b708 100644
--- a/src/gallium/drivers/softpipe/sp_state_surface.c
+++ b/src/gallium/drivers/softpipe/sp_state_surface.c
@@ -27,7 +27,7 @@
 
 /* Authors:  Keith Whitwell <keith@tungstengraphics.com>
  */
-#include "p_inlines.h"
+#include "pipe/p_inlines.h"
 
 #include "sp_context.h"
 #include "sp_state.h"
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index c54e9d385c..43d5085895 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -332,6 +332,61 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
 }
 
 
+/**
+ * For RECT textures / unnormalized texcoords
+ * Only a subset of wrap modes supported.
+ */
+static INLINE int
+nearest_texcoord_unnorm(unsigned wrapMode, float s, unsigned size)
+{
+   int i;
+   switch (wrapMode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      i = ifloor(s);
+      return CLAMP(i, 0, size-1);
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      /* fall-through */
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+/**
+ * For RECT textures / unnormalized texcoords.
+ * Only a subset of wrap modes supported.
+ */
+static INLINE void
+linear_texcoord_unnorm(unsigned wrapMode, float s, unsigned size,
+                       int *i0, int *i1, float *a)
+{
+   switch (wrapMode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      /* Not exactly what the spec says, but it matches NVIDIA output */
+      s = CLAMP(s - 0.5F, 0.0, (float) size - 1.0);
+      *i0 = ifloor(s);
+      *i1 = *i0 + 1;
+      break;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      /* fall-through */
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      s = CLAMP(s, 0.5F, (float) size - 0.5F);
+      s -= 0.5F;
+      *i0 = ifloor(s);
+      *i1 = *i0 + 1;
+      if (*i1 > size - 1)
+         *i1 = size - 1;
+      break;
+   default:
+      assert(0);
+   }
+   *a = FRAC(s);
+}
+
+
 static unsigned
 choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
 {
@@ -415,15 +470,15 @@ compute_lambda(struct tgsi_sampler *sampler,
 {
    float rho, lambda;
 
+   assert(sampler->state->normalized_coords);
+
    assert(s);
    {
       float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT];
       float dsdy = s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT];
       dsdx = FABSF(dsdx);
       dsdy = FABSF(dsdy);
-      rho = MAX2(dsdx, dsdy);
-      if (sampler->state->normalized_coords)
-         rho *= sampler->texture->width[0];
+      rho = MAX2(dsdx, dsdy) * sampler->texture->width[0];
    }
    if (t) {
       float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
@@ -431,9 +486,7 @@ compute_lambda(struct tgsi_sampler *sampler,
       float max;
       dtdx = FABSF(dtdx);
       dtdy = FABSF(dtdy);
-      max = MAX2(dtdx, dtdy);
-      if (sampler->state->normalized_coords)
-         max *= sampler->texture->height[0];
+      max = MAX2(dtdx, dtdy) * sampler->texture->height[0];
       rho = MAX2(rho, max);
    }
    if (p) {
@@ -442,9 +495,7 @@ compute_lambda(struct tgsi_sampler *sampler,
       float max;
       dpdx = FABSF(dpdx);
       dpdy = FABSF(dpdy);
-      max = MAX2(dpdx, dpdy);
-      if (sampler->state->normalized_coords)
-         max *= sampler->texture->depth[0];
+      max = MAX2(dpdx, dpdy) * sampler->texture->depth[0];
       rho = MAX2(rho, max);
    }
 
@@ -628,13 +679,10 @@ sp_get_samples_2d_common(struct tgsi_sampler *sampler,
    choose_mipmap_levels(sampler, s, t, p, lodbias,
                         &level0, &level1, &levelBlend, &imgFilter);
 
-   if (sampler->state->normalized_coords) {
-      width = sampler->texture->width[level0];
-      height = sampler->texture->height[level0];
-   }
-   else {
-      width = height = 1;
-   }
+   assert(sampler->state->normalized_coords);
+
+   width = sampler->texture->width[level0];
+   height = sampler->texture->height[level0];
 
    assert(width > 0);
 
@@ -765,14 +813,11 @@ sp_get_samples_3d(struct tgsi_sampler *sampler,
    choose_mipmap_levels(sampler, s, t, p, lodbias,
                         &level0, &level1, &levelBlend, &imgFilter);
 
-   if (sampler->state->normalized_coords) {
-      width = sampler->texture->width[level0];
-      height = sampler->texture->height[level0];
-      depth = sampler->texture->depth[level0];
-   }
-   else {
-      width = height = depth = 1;
-   }
+   assert(sampler->state->normalized_coords);
+
+   width = sampler->texture->width[level0];
+   height = sampler->texture->height[level0];
+   depth = sampler->texture->depth[level0];
 
    assert(width > 0);
    assert(height > 0);
@@ -889,6 +934,73 @@ sp_get_samples_cube(struct tgsi_sampler *sampler,
 }
 
 
+static void
+sp_get_samples_rect(struct tgsi_sampler *sampler,
+                    const float s[QUAD_SIZE],
+                    const float t[QUAD_SIZE],
+                    const float p[QUAD_SIZE],
+                    float lodbias,
+                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   //sp_get_samples_2d_common(sampler, s, t, p, lodbias, rgba, faces);
+   static const uint face = 0;
+   const uint compare_func = sampler->state->compare_func;
+   unsigned level0, level1, j, imgFilter;
+   int width, height;
+   float levelBlend;
+
+   choose_mipmap_levels(sampler, s, t, p, lodbias,
+                        &level0, &level1, &levelBlend, &imgFilter);
+
+   /* texture RECTS cannot be mipmapped */
+   assert(level0 == level1);
+
+   width = sampler->texture->width[level0];
+   height = sampler->texture->height[level0];
+
+   assert(width > 0);
+
+   switch (imgFilter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = nearest_texcoord_unnorm(sampler->state->wrap_s, s[j], width);
+         int y = nearest_texcoord_unnorm(sampler->state->wrap_t, t[j], height);
+         get_texel(sampler, face, level0, x, y, 0, rgba, j);
+         if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+            shadow_compare(compare_func, rgba, p, j);
+         }
+      }
+      break;
+   case PIPE_TEX_FILTER_LINEAR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         float tx[4][4], a, b;
+         int x0, y0, x1, y1, c;
+         linear_texcoord_unnorm(sampler->state->wrap_s, s[j], width,  &x0, &x1, &a);
+         linear_texcoord_unnorm(sampler->state->wrap_t, t[j], height, &y0, &y1, &b);
+         get_texel(sampler, face, level0, x0, y0, 0, tx, 0);
+         get_texel(sampler, face, level0, x1, y0, 0, tx, 1);
+         get_texel(sampler, face, level0, x0, y1, 0, tx, 2);
+         get_texel(sampler, face, level0, x1, y1, 0, tx, 3);
+         if (sampler->state->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+            shadow_compare(compare_func, tx, p, 0);
+            shadow_compare(compare_func, tx, p, 1);
+            shadow_compare(compare_func, tx, p, 2);
+            shadow_compare(compare_func, tx, p, 3);
+         }
+
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp_2d(a, b, tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+
+
 /**
  * Called via tgsi_sampler::get_samples()
  * Use the sampler's state setting to get a filtered RGBA value
@@ -914,15 +1026,21 @@ sp_get_samples(struct tgsi_sampler *sampler,
 
    switch (sampler->texture->target) {
    case PIPE_TEXTURE_1D:
+      assert(sampler->state->normalized_coords);
       sp_get_samples_1d(sampler, s, t, p, lodbias, rgba);
       break;
    case PIPE_TEXTURE_2D:
-      sp_get_samples_2d(sampler, s, t, p, lodbias, rgba);
+      if (sampler->state->normalized_coords)
+         sp_get_samples_2d(sampler, s, t, p, lodbias, rgba);
+      else
+         sp_get_samples_rect(sampler, s, t, p, lodbias, rgba);
       break;
    case PIPE_TEXTURE_3D:
+      assert(sampler->state->normalized_coords);
       sp_get_samples_3d(sampler, s, t, p, lodbias, rgba);
       break;
    case PIPE_TEXTURE_CUBE:
+      assert(sampler->state->normalized_coords);
       sp_get_samples_cube(sampler, s, t, p, lodbias, rgba);
       break;
    default:
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 6de7a9b543..6ba0f09e0a 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -39,6 +39,7 @@
 #include "sp_context.h"
 #include "sp_state.h"
 #include "sp_texture.h"
+#include "sp_tile_cache.h"
 
 
 /* Simple, maximally packed layout.
@@ -88,6 +89,7 @@ softpipe_texture_create(struct pipe_context *pipe,
       return NULL;
 
    spt->base = *templat;
+   spt->base.refcount = 1;
 
    softpipe_texture_layout(spt);
 
@@ -99,6 +101,8 @@ softpipe_texture_create(struct pipe_context *pipe,
       return NULL;
    }
 
+   assert(spt->base.refcount == 1);
+
    return &spt->base;
 }
 
@@ -128,6 +132,20 @@ softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt)
 }
 
 
+void
+softpipe_texture_update(struct pipe_context *pipe,
+                        struct pipe_texture *texture)
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+   uint unit;
+   for (unit = 0; unit < PIPE_MAX_SAMPLERS; unit++) {
+      if (softpipe->texture[unit] == texture) {
+         sp_flush_tile_cache(softpipe, softpipe->tex_cache[unit]);
+      }
+   }
+}
+
+
 /**
  * Called via pipe->get_tex_surface()
  */
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index fa646c0de9..50fc100427 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -62,6 +62,10 @@ softpipe_texture_create(struct pipe_context *pipe,
 extern void
 softpipe_texture_release(struct pipe_context *pipe, struct pipe_texture **pt);
 
+extern void
+softpipe_texture_update(struct pipe_context *pipe,
+                        struct pipe_texture *texture);
+
 extern struct pipe_surface *
 softpipe_get_tex_surface(struct pipe_context *pipe,
                          struct pipe_texture *pt,
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index dde3fabc81..da30dd6c48 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -212,14 +212,15 @@ sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc)
  * Specify the texture to cache.
  */
 void
-sp_tile_cache_set_texture(struct softpipe_tile_cache *tc,
+sp_tile_cache_set_texture(struct pipe_context *pipe,
+                          struct softpipe_tile_cache *tc,
                           struct pipe_texture *texture)
 {
    uint i;
 
    assert(!tc->surface);
 
-   tc->texture = texture;
+   pipe_texture_reference(pipe, &tc->texture, texture);
 
    if (tc->tex_surf_map) {
       pipe_surface_unmap(tc->tex_surf);
@@ -358,30 +359,37 @@ sp_flush_tile_cache(struct softpipe_context *softpipe,
    struct pipe_surface *ps = tc->surface;
    int inuse = 0, pos;
 
-   if (!ps || !ps->buffer)
-      return;
-
-   for (pos = 0; pos < NUM_ENTRIES; pos++) {
-      struct softpipe_cached_tile *tile = tc->entries + pos;
-      if (tile->x >= 0) {
-         if (tc->depth_stencil) {
-            pipe_put_tile_raw(pipe, ps,
-                           tile->x, tile->y, TILE_SIZE, TILE_SIZE,
-                           tile->data.depth32, 0/*STRIDE*/);
-         }
-         else {
-            pipe_put_tile_rgba(pipe, ps,
-                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
-                               (float *) tile->data.color);
+   if (ps && ps->buffer) {
+      /* caching a drawing surface */
+      for (pos = 0; pos < NUM_ENTRIES; pos++) {
+         struct softpipe_cached_tile *tile = tc->entries + pos;
+         if (tile->x >= 0) {
+            if (tc->depth_stencil) {
+               pipe_put_tile_raw(pipe, ps,
+                              tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                              tile->data.depth32, 0/*STRIDE*/);
+            }
+            else {
+               pipe_put_tile_rgba(pipe, ps,
+                                  tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                                  (float *) tile->data.color);
+            }
+            tile->x = tile->y = -1;  /* mark as empty */
+            inuse++;
          }
-         tile->x = tile->y = -1;  /* mark as empty */
-         inuse++;
       }
-   }
 
 #if TILE_CLEAR_OPTIMIZATION
-   sp_tile_cache_flush_clear(&softpipe->pipe, tc);
+      sp_tile_cache_flush_clear(&softpipe->pipe, tc);
 #endif
+   }
+   else if (tc->texture) {
+      /* caching a texture, mark all entries as embpy */
+      for (pos = 0; pos < NUM_ENTRIES; pos++) {
+         tc->entries[pos].x = -1;
+      }
+      tc->tex_face = -1;
+   }
 
 #if 0
    debug_printf("flushed tiles in use: %d\n", inuse);
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h
index 7fd1081286..2631e29a3a 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.h
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.h
@@ -80,7 +80,8 @@ extern void
 sp_tile_cache_unmap_surfaces(struct softpipe_tile_cache *tc);
 
 extern void
-sp_tile_cache_set_texture(struct softpipe_tile_cache *tc,
+sp_tile_cache_set_texture(struct pipe_context *pipe,
+                          struct softpipe_tile_cache *tc,
                           struct pipe_texture *texture);
 
 extern void
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index 39f95695fb..036c4c8964 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -206,6 +206,14 @@ struct pipe_context {
    void (*texture_release)(struct pipe_context *pipe,
 			   struct pipe_texture **pt);
 
+   /**
+    * Called when texture data is changed.
+    * Note: we could pass some hints about which mip levels or cube faces
+    * have changed...
+    */
+   void (*texture_update)(struct pipe_context *pipe,
+                          struct pipe_texture *texture);
+
    /** Get a surface which is a "view" into a texture */
    struct pipe_surface *(*get_tex_surface)(struct pipe_context *pipe,
                                            struct pipe_texture *texture,
diff --git a/src/gallium/include/pipe/p_shader_tokens.h b/src/gallium/include/pipe/p_shader_tokens.h
index 3ce35310f6..10c47e0ef0 100644
--- a/src/gallium/include/pipe/p_shader_tokens.h
+++ b/src/gallium/include/pipe/p_shader_tokens.h
@@ -5,6 +5,8 @@
 extern "C" {
 #endif // defined __cplusplus
 
+#include "p_compiler.h"
+
 struct tgsi_version
 {
    unsigned MajorVersion  : 8;
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 1082343e2f..47fa78c31d 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -66,7 +66,8 @@ struct pipe_winsys;
  * The driver will certainly subclass this to include actual memory
  * management information.
  */
-struct pipe_buffer {
+struct pipe_buffer
+{
    unsigned alignment;
    unsigned usage;
    unsigned size;
@@ -76,8 +77,6 @@ struct pipe_buffer {
 };
 
 
-
-
 /**
  * Primitive (point/line/tri) rasterization info
  */
@@ -113,18 +112,21 @@ struct pipe_rasterizer_state
 };
 
 
-struct pipe_poly_stipple {
+struct pipe_poly_stipple
+{
    unsigned stipple[32];
 };
 
 
-struct pipe_viewport_state {
+struct pipe_viewport_state
+{
    float scale[4];
    float translate[4];
 };
 
 
-struct pipe_scissor_state {
+struct pipe_scissor_state
+{
    unsigned minx:16;
    unsigned miny:16;
    unsigned maxx:16;
@@ -132,7 +134,8 @@ struct pipe_scissor_state {
 };
 
 
-struct pipe_clip_state {
+struct pipe_clip_state
+{
    float ucp[PIPE_MAX_CLIP_PLANES][4];
    unsigned nr;
 };
@@ -141,13 +144,15 @@ struct pipe_clip_state {
 /**
  * Constants for vertex/fragment shaders
  */
-struct pipe_constant_buffer {
+struct pipe_constant_buffer
+{
    struct pipe_buffer *buffer;
    unsigned size;    /** in bytes */
 };
 
 
-struct pipe_shader_state {
+struct pipe_shader_state
+{
    const struct tgsi_token *tokens;
    ubyte num_inputs;
    ubyte num_outputs;
@@ -185,7 +190,8 @@ struct pipe_depth_stencil_alpha_state
 };
 
 
-struct pipe_blend_state {
+struct pipe_blend_state
+{
    unsigned blend_enable:1;
 
    unsigned rgb_func:3;          /**< PIPE_BLEND_x */
@@ -204,7 +210,8 @@ struct pipe_blend_state {
 };
 
 
-struct pipe_blend_color {
+struct pipe_blend_color
+{
    float color[4];
 };
 
@@ -224,19 +231,19 @@ struct pipe_framebuffer_state
  */
 struct pipe_sampler_state
 {
-   unsigned wrap_s:3;        /**< PIPE_TEX_WRAP_x */
-   unsigned wrap_t:3;        /**< PIPE_TEX_WRAP_x */
-   unsigned wrap_r:3;        /**< PIPE_TEX_WRAP_x */
+   unsigned wrap_s:3;            /**< PIPE_TEX_WRAP_x */
+   unsigned wrap_t:3;            /**< PIPE_TEX_WRAP_x */
+   unsigned wrap_r:3;            /**< PIPE_TEX_WRAP_x */
    unsigned min_img_filter:2;    /**< PIPE_TEX_FILTER_x */
    unsigned min_mip_filter:2;    /**< PIPE_TEX_MIPFILTER_x */
    unsigned mag_img_filter:2;    /**< PIPE_TEX_FILTER_x */
-   unsigned compare:1;       /**< shadow/depth compare enabled? */
-   unsigned compare_mode:1;  /**< PIPE_TEX_COMPARE_x */
-   unsigned compare_func:3;  /**< PIPE_FUNC_x */
-   unsigned normalized_coords:1;  /**< Are coords normalized to [0,1]? */
-   float shadow_ambient;          /**< shadow test fail color/intensity */
-   float lod_bias;                /**< LOD/lambda bias */
-   float min_lod, max_lod;        /**< LOD clamp range, after bias */
+   unsigned compare:1;           /**< shadow/depth compare enabled? */
+   unsigned compare_mode:1;      /**< PIPE_TEX_COMPARE_x */
+   unsigned compare_func:3;      /**< PIPE_FUNC_x */
+   unsigned normalized_coords:1; /**< Are coords normalized to [0,1]? */
+   float shadow_ambient;         /**< shadow test fail color/intensity */
+   float lod_bias;               /**< LOD/lambda bias */
+   float min_lod, max_lod;       /**< LOD clamp range, after bias */
    float border_color[4];
    float max_anisotropy;
 };
@@ -248,10 +255,10 @@ struct pipe_sampler_state
  */
 struct pipe_surface
 {
-   struct pipe_buffer *buffer; /**< driver private buffer handle */
+   struct pipe_buffer *buffer;   /**< surface's buffer/memory */
    enum pipe_format format;      /**< PIPE_FORMAT_x */
    unsigned status;              /**< PIPE_SURFACE_STATUS_x */
-   unsigned clear_value;         /**< may be temporary */
+   unsigned clear_value;         /**< XXX may be temporary */
    unsigned cpp;                 /**< bytes per pixel */
    unsigned width;
    unsigned height;
diff --git a/src/gallium/include/pipe/p_thread.h b/src/gallium/include/pipe/p_thread.h
index cd432c547c..4325abc951 100644
--- a/src/gallium/include/pipe/p_thread.h
+++ b/src/gallium/include/pipe/p_thread.h
@@ -1,54 +1,317 @@
 /**************************************************************************
  * 
+ * Copyright 1999-2006 Brian Paul
  * 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.
- * 
+ * 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 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.
+ * 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 P_THREAD_H
-#define P_THREAD_H
+/**
+ * @file
+ * Thread
+ *
+ * Initial version by John Stone (j.stone@acm.org) (johns@cs.umr.edu)
+ *                and Christoph Poliwoda (poliwoda@volumegraphics.com)
+ * Revised by Keith Whitwell
+ * Adapted for new gl dispatcher by Brian Paul
+ *
+ *
+ *
+ * DOCUMENTATION
+ *
+ * This thread module exports the following types:
+ *   _glthread_TSD     Thread-specific data area
+ *   _glthread_Thread  Thread datatype
+ *   _glthread_Mutex   Mutual exclusion lock
+ *
+ * Macros:
+ *   _glthread_DECLARE_STATIC_MUTEX(name)   Declare a non-local mutex
+ *   _glthread_INIT_MUTEX(name)             Initialize a mutex
+ *   _glthread_LOCK_MUTEX(name)             Lock a mutex
+ *   _glthread_UNLOCK_MUTEX(name)           Unlock a mutex
+ *
+ * Functions:
+ *   _glthread_GetID(v)      Get integer thread ID
+ *   _glthread_InitTSD()     Initialize thread-specific data
+ *   _glthread_GetTSD()      Get thread-specific data
+ *   _glthread_SetTSD()      Set thread-specific data
+ *
+ * If this file is accidentally included by a non-threaded build,
+ * it should not cause the build to fail, or otherwise cause problems.
+ * In general, it should only be included when needed however.
+ */
+
+#ifndef _P_THREAD_H_
+#define _P_THREAD_H_
+
+
+#if defined(USE_MGL_NAMESPACE)
+#define _glapi_Dispatch _mglapi_Dispatch
+#endif
+
+
+
+#if (defined(PTHREADS) || defined(SOLARIS_THREADS) ||\
+     defined(WIN32_THREADS) || defined(USE_XTHREADS) || defined(BEOS_THREADS)) \
+    && !defined(THREADS)
+# define THREADS
+#endif
+
+#ifdef VMS
+#include <GL/vms_x_fix.h>
+#endif
+
+/*
+ * POSIX threads. This should be your choice in the Unix world
+ * whenever possible.  When building with POSIX threads, be sure
+ * to enable any compiler flags which will cause the MT-safe
+ * libc (if one exists) to be used when linking, as well as any
+ * header macros for MT-safe errno, etc.  For Solaris, this is the -mt
+ * compiler flag.  On Solaris with gcc, use -D_REENTRANT to enable
+ * proper compiling for MT-safe libc etc.
+ */
+#if defined(PTHREADS)
+#include <pthread.h> /* POSIX threads headers */
+
+typedef struct {
+   pthread_key_t  key;
+   int initMagic;
+} _glthread_TSD;
+
+typedef pthread_t _glthread_Thread;
+
+typedef pthread_mutex_t _glthread_Mutex;
+
+#define _glthread_DECLARE_STATIC_MUTEX(name) \
+   static _glthread_Mutex name = PTHREAD_MUTEX_INITIALIZER
+
+#define _glthread_INIT_MUTEX(name) \
+   pthread_mutex_init(&(name), NULL)
+
+#define _glthread_DESTROY_MUTEX(name) \
+   pthread_mutex_destroy(&(name))
+
+#define _glthread_LOCK_MUTEX(name) \
+   (void) pthread_mutex_lock(&(name))
+
+#define _glthread_UNLOCK_MUTEX(name) \
+   (void) pthread_mutex_unlock(&(name))
+
+#endif /* PTHREADS */
+
+
+
+
+/*
+ * Solaris threads. Use only up to Solaris 2.4.
+ * Solaris 2.5 and higher provide POSIX threads.
+ * Be sure to compile with -mt on the Solaris compilers, or
+ * use -D_REENTRANT if using gcc.
+ */
+#ifdef SOLARIS_THREADS
+#include <thread.h>
+
+typedef struct {
+   thread_key_t key;
+   mutex_t      keylock;
+   int          initMagic;
+} _glthread_TSD;
+
+typedef thread_t _glthread_Thread;
+
+typedef mutex_t _glthread_Mutex;
+
+/* XXX need to really implement mutex-related macros */
+#define _glthread_DECLARE_STATIC_MUTEX(name)  static _glthread_Mutex name = 0
+#define _glthread_INIT_MUTEX(name)  (void) name
+#define _glthread_DESTROY_MUTEX(name) (void) name
+#define _glthread_LOCK_MUTEX(name)  (void) name
+#define _glthread_UNLOCK_MUTEX(name)  (void) name
+
+#endif /* SOLARIS_THREADS */
+
+
+
+
+/*
+ * Windows threads. Should work with Windows NT and 95.
+ * IMPORTANT: Link with multithreaded runtime library when THREADS are
+ * used!
+ */
+#ifdef WIN32_THREADS
+#include <windows.h>
+
+typedef struct {
+   DWORD key;
+   int   initMagic;
+} _glthread_TSD;
+
+typedef HANDLE _glthread_Thread;
+
+typedef CRITICAL_SECTION _glthread_Mutex;
+
+#define _glthread_DECLARE_STATIC_MUTEX(name)  /*static*/ _glthread_Mutex name = {0,0,0,0,0,0}
+#define _glthread_INIT_MUTEX(name)  InitializeCriticalSection(&name)
+#define _glthread_DESTROY_MUTEX(name)  DeleteCriticalSection(&name)
+#define _glthread_LOCK_MUTEX(name)  EnterCriticalSection(&name)
+#define _glthread_UNLOCK_MUTEX(name)  LeaveCriticalSection(&name)
+
+#endif /* WIN32_THREADS */
+
+
 
-#include "p_compiler.h"
 
 /*
- * XXX: We should come up with a system-independent thread definitions.
- * XXX: Patching glthread defs for now.
+ * XFree86 has its own thread wrapper, Xthreads.h
+ * We wrap it again for GL.
  */
+#ifdef USE_XTHREADS
+#include <X11/Xthreads.h>
+
+typedef struct {
+   xthread_key_t key;
+   int initMagic;
+} _glthread_TSD;
+
+typedef xthread_t _glthread_Thread;
+
+typedef xmutex_rec _glthread_Mutex;
+
+#ifdef XMUTEX_INITIALIZER
+#define _glthread_DECLARE_STATIC_MUTEX(name) \
+   static _glthread_Mutex name = XMUTEX_INITIALIZER
+#else
+#define _glthread_DECLARE_STATIC_MUTEX(name) \
+   static _glthread_Mutex name
+#endif
+
+#define _glthread_INIT_MUTEX(name) \
+   xmutex_init(&(name))
+
+#define _glthread_DESTROY_MUTEX(name) \
+   xmutex_clear(&(name))
+
+#define _glthread_LOCK_MUTEX(name) \
+   (void) xmutex_lock(&(name))
+
+#define _glthread_UNLOCK_MUTEX(name) \
+   (void) xmutex_unlock(&(name))
+
+#endif /* USE_XTHREADS */
+
+
+
+/*
+ * BeOS threads. R5.x required.
+ */
+#ifdef BEOS_THREADS
+
+#include <kernel/OS.h>
+#include <support/TLS.h>
+
+typedef struct {
+   int32        key;
+   int          initMagic;
+} _glthread_TSD;
+
+typedef thread_id _glthread_Thread;
+
+/* Use Benaphore, aka speeder semaphore */
+typedef struct {
+    int32   lock;
+    sem_id  sem;
+} benaphore;
+typedef benaphore _glthread_Mutex;
+
+#define _glthread_DECLARE_STATIC_MUTEX(name)  static _glthread_Mutex name = { 0, 0 }
+#define _glthread_INIT_MUTEX(name)    	name.sem = create_sem(0, #name"_benaphore"), name.lock = 0
+#define _glthread_DESTROY_MUTEX(name) 	delete_sem(name.sem), name.lock = 0
+#define _glthread_LOCK_MUTEX(name)    	if (name.sem == 0) _glthread_INIT_MUTEX(name); \
+									  	if (atomic_add(&(name.lock), 1) >= 1) acquire_sem(name.sem)
+#define _glthread_UNLOCK_MUTEX(name)  	if (atomic_add(&(name.lock), -1) > 1) release_sem(name.sem)
+
+#endif /* BEOS_THREADS */
+
+
+
+#ifndef THREADS
+
+/*
+ * THREADS not defined
+ */
+
+typedef unsigned _glthread_TSD;
+
+typedef unsigned _glthread_Thread;
+
+typedef unsigned _glthread_Mutex;
+
+#define _glthread_DECLARE_STATIC_MUTEX(name)  static _glthread_Mutex name = 0
+
+#define _glthread_INIT_MUTEX(name)  (void) name
+
+#define _glthread_DESTROY_MUTEX(name)  (void) name
+
+#define _glthread_LOCK_MUTEX(name)  (void) name
+
+#define _glthread_UNLOCK_MUTEX(name)  (void) name
+
+#endif /* THREADS */
+
+
+
+/*
+ * Platform independent thread specific data API.
+ */
+
+extern unsigned long
+_glthread_GetID(void);
+
+
+extern void
+_glthread_InitTSD(_glthread_TSD *);
+
+
+extern void *
+_glthread_GetTSD(_glthread_TSD *);
 
-#ifndef __MSC__
 
-#include "glapi/glthread.h"
+extern void
+_glthread_SetTSD(_glthread_TSD *, void *);
 
-#else /* __MSC__ */
+#if defined(GLX_USE_TLS)
 
-typedef int _glthread_Mutex;
+extern __thread struct _glapi_table * _glapi_tls_Dispatch
+    __attribute__((tls_model("initial-exec")));
 
-#define _glthread_INIT_MUTEX( M )   ((void) (M))
-#define _glthread_LOCK_MUTEX( M )   ((void) (M))
-#define _glthread_UNLOCK_MUTEX( M ) ((void) (M))
+#define GET_DISPATCH() _glapi_tls_Dispatch
 
-#define sched_yield() ((void) 0)
+#elif !defined(GL_CALL)
+# if defined(THREADS)
+#  define GET_DISPATCH() \
+   ((__builtin_expect( _glapi_Dispatch != NULL, 1 )) \
+       ? _glapi_Dispatch : _glapi_get_dispatch())
+# else
+#  define GET_DISPATCH() _glapi_Dispatch
+# endif /* defined(THREADS) */
+#endif  /* ndef GL_CALL */
 
-#endif /* __MSC__ */
 
-#endif /* P_THREAD_H */
+#endif /* _P_THREAD_H_ */
diff --git a/src/gallium/winsys/Makefile b/src/gallium/winsys/Makefile
new file mode 100644
index 0000000000..3dc5ea5744
--- /dev/null
+++ b/src/gallium/winsys/Makefile
@@ -0,0 +1,20 @@
+TOP = ../../..
+include $(TOP)/configs/current
+
+
+SUBDIRS = $(GALLIUM_WINSYS_DIRS)
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`
diff --git a/src/gallium/winsys/SConscript b/src/gallium/winsys/SConscript
new file mode 100644
index 0000000000..32215d8d58
--- /dev/null
+++ b/src/gallium/winsys/SConscript
@@ -0,0 +1,10 @@
+Import('*')
+
+if dri:
+	SConscript([
+		'dri/SConscript',
+	])
+else:
+	SConscript([
+		'xlib/SConscript',
+	])
diff --git a/src/gallium/winsys/dri/Makefile.template b/src/gallium/winsys/dri/Makefile.template
index 2a261ed669..3bc1fdd4d4 100644
--- a/src/gallium/winsys/dri/Makefile.template
+++ b/src/gallium/winsys/dri/Makefile.template
@@ -1,7 +1,9 @@
 # -*-makefile-*-
 
-MESA_MODULES = $(TOP)/src/mesa/libmesa.a
-
+MESA_MODULES = \
+	$(TOP)/src/mesa/libmesa.a \
+	$(GALLIUM_AUXILIARIES)
+	
 COMMON_GALLIUM_SOURCES = \
         $(TOP)/src/mesa/drivers/dri/common/utils.c \
         $(TOP)/src/mesa/drivers/dri/common/vblank.c \
@@ -23,8 +25,9 @@ WINOBJ=
 WINLIB=
 INCLUDES = $(SHARED_INCLUDES) $(EXPAT_INCLUDES)
 
-OBJECTS = $(C_SOURCES:.c=.o) \
-	  $(ASM_SOURCES:.S=.o) 
+OBJECTS = \
+	$(C_SOURCES:.c=.o) \
+	$(ASM_SOURCES:.S=.o) 
 
 else
 # miniglx
diff --git a/src/mesa/drivers/dri/SConscript b/src/gallium/winsys/dri/SConscript
index d32bd08669..8c56ce917c 100644
--- a/src/mesa/drivers/dri/SConscript
+++ b/src/gallium/winsys/dri/SConscript
@@ -6,6 +6,9 @@ drienv.Replace(CPPPATH = [
 	'#src/mesa/drivers/dri/common',
 	'#include',
 	'#include/GL/internal',
+	'#src/gallium/include',
+	'#src/gallium/auxiliary',
+	'#src/gallium/drivers',
 	'#src/mesa',
 	'#src/mesa/main',
 	'#src/mesa/glapi',
@@ -21,15 +24,15 @@ drienv.Replace(CPPPATH = [
 drienv.ParseConfig('pkg-config --cflags --libs libdrm')
 
 COMMON_GALLIUM_SOURCES = [
-	'../common/utils.c',
-	'../common/vblank.c',
-	'../common/dri_util.c',
-	'../common/xmlconfig.c',
+	'#src/mesa/drivers/dri/common/utils.c',
+	'#src/mesa/drivers/dri/common/vblank.c',
+	'#src/mesa/drivers/dri/common/dri_util.c',
+	'#src/mesa/drivers/dri/common/xmlconfig.c',
 ]
 
 COMMON_BM_SOURCES = [
-	'../common/dri_bufmgr.c',
-	'../common/dri_drmpool.c',
+	'#src/mesa/drivers/dri/common/dri_bufmgr.c',
+	'#src/mesa/drivers/dri/common/dri_drmpool.c',
 ]
 
 Export([
@@ -44,5 +47,5 @@ Export([
 #	$(INSTALL) -m 755 $(LIBNAME) $(DRI_DRIVER_INSTALL_DIR)
 
 SConscript([
-	'intel_winsys/SConscript',
+	'intel/SConscript',
 ])
diff --git a/src/gallium/winsys/dri/intel/SConscript b/src/gallium/winsys/dri/intel/SConscript
index a7cc10450e..0ad19d42a8 100644
--- a/src/gallium/winsys/dri/intel/SConscript
+++ b/src/gallium/winsys/dri/intel/SConscript
@@ -9,11 +9,6 @@ env.Append(CPPPATH = [
 
 #MINIGLX_SOURCES = server/intel_dri.c
 
-pipe_drivers = [
-	softpipe,
-	i915simple
-]
-
 DRIVER_SOURCES = [
 	'intel_winsys_pipe.c',
 	'intel_winsys_softpipe.c',
@@ -31,11 +26,14 @@ sources = \
 	COMMON_BM_SOURCES + \
 	DRIVER_SOURCES
 
-# DRIVER_DEFINES = -I../intel $(shell pkg-config libdrm --atleast-version=2.3.1 \
-#				&& echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
+drivers = [
+	softpipe,
+	i915simple
+]
 
+# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions
 env.SharedLibrary(
 	target ='i915tex_dri.so',
 	source = sources,
-	LIBS = pipe_drivers + env['LIBS'],
+	LIBS = drivers + mesa + auxiliaries + env['LIBS'],
 )
\ No newline at end of file
diff --git a/src/gallium/winsys/dri/intel/intel_batchpool.c b/src/gallium/winsys/dri/intel/intel_batchpool.c
index 33b56817f6..ce154c7b88 100644
--- a/src/gallium/winsys/dri/intel/intel_batchpool.c
+++ b/src/gallium/winsys/dri/intel/intel_batchpool.c
@@ -36,9 +36,12 @@
 
 #include <xf86drm.h>
 #include <stdlib.h>
+#include <stdio.h>
 #include <errno.h>
-#include "imports.h"
-#include "glthread.h"
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_thread.h"
+
 #include "dri_bufpool.h"
 #include "dri_bufmgr.h"
 #include "intel_batchpool.h"
@@ -196,7 +199,7 @@ pool_create(struct _DriBufferPool *pool,
    _glthread_LOCK_MUTEX(p->mutex);
 
    if (p->numFree == 0)
-      pool_checkFree(p, GL_TRUE);
+      pool_checkFree(p, TRUE);
 
    if (p->numFree == 0) {
       fprintf(stderr, "Out of fixed size buffer objects\n");
@@ -278,7 +281,7 @@ pool_map(struct _DriBufferPool *pool, void *private, unsigned flags,
       return -EBUSY;
    }
 
-   buf->mapped = GL_TRUE;
+   buf->mapped = TRUE;
    *virtual = (unsigned char *) p->virtual + buf->start;
    _glthread_UNLOCK_MUTEX(p->mutex);
    return 0;
@@ -361,7 +364,7 @@ pool_validate(struct _DriBufferPool *pool, void *private)
    BBuf *buf = (BBuf *) private;
    BPool *p = buf->parent;
    _glthread_LOCK_MUTEX(p->mutex);
-   buf->unfenced = GL_TRUE;
+   buf->unfenced = TRUE;
    _glthread_UNLOCK_MUTEX(p->mutex);
    return 0;
 }
@@ -379,7 +382,7 @@ pool_takedown(struct _DriBufferPool *pool)
    while ((p->numFree < p->numTot) && p->numDelayed) {
       _glthread_UNLOCK_MUTEX(p->mutex);
       sched_yield();
-      pool_checkFree(p, GL_TRUE);
+      pool_checkFree(p, TRUE);
       _glthread_LOCK_MUTEX(p->mutex);
    }
 
diff --git a/src/gallium/winsys/xlib/Makefile b/src/gallium/winsys/xlib/Makefile
new file mode 100644
index 0000000000..c443330942
--- /dev/null
+++ b/src/gallium/winsys/xlib/Makefile
@@ -0,0 +1,93 @@
+# src/gallium/winsys/xlib/Makefile
+
+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/auxiliary
+
+XLIB_WINSYS_SOURCES = \
+	glxapi.c	\
+	fakeglx.c	\
+	xfonts.c	\
+	xm_api.c	\
+	xm_winsys.c	\
+	xm_winsys_aub.c	\
+	brw_aub.c
+
+XLIB_WINSYS_OBJECTS = $(XLIB_WINSYS_SOURCES:.c=.o)
+
+
+ifeq ($(CONFIG_NAME), linux-cell)
+# The SPU code is in a separate .a file, unfortunately
+CELL_SPU_LIB = $(TOP)/src/gallium/drivers/cell/spu/g3d_spu.a
+endif
+
+LIBS = \
+	$(GALLIUM_DRIVERS) \
+	$(TOP)/src/mesa/libglapi.a \
+	$(TOP)/src/mesa/libmesa.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)/$(GL_LIB_NAME)
+
+
+# Make the libGL.so library
+$(TOP)/$(LIB_DIR)/$(GL_LIB_NAME): $(XLIB_WINSYS_OBJECTS) $(LIBS)
+	$(TOP)/bin/mklib -o $(GL_LIB) \
+		-linker "$(CC)" \
+		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
+		-install $(TOP)/$(LIB_DIR) \
+		$(MKLIB_OPTIONS) $(XLIB_WINSYS_OBJECTS) \
+		--start-group $(LIBS) --end-group $(GL_LIB_DEPS)
+
+
+depend: $(ALL_SOURCES)
+	@ echo "running $(MKDEP)"
+	@ rm -f depend  # workaround oops on gutsy?!?
+	@ touch depend
+	@ $(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDE_DIRS) $(ALL_SOURCES) \
+		> /dev/null 2>/dev/null
+
+
+install: default
+	$(INSTALL) -d $(INSTALL_DIR)/include/GL
+	$(INSTALL) -d $(INSTALL_DIR)/$(LIB_DIR)
+	$(INSTALL) -m 644 $(TOP)/include/GL/*.h $(INSTALL_DIR)/include/GL
+	@if [ -e $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME) ]; then \
+		$(INSTALL) $(TOP)/$(LIB_DIR)/libGL* $(INSTALL_DIR)/$(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/winsys/xlib/SConscript b/src/gallium/winsys/xlib/SConscript
new file mode 100644
index 0000000000..c38b5be52c
--- /dev/null
+++ b/src/gallium/winsys/xlib/SConscript
@@ -0,0 +1,34 @@
+#######################################################################
+# SConscript for xlib winsys
+
+Import('*')
+
+env = env.Clone()
+
+env.Append(CPPPATH = [
+	'#/src/mesa',
+	'#/src/mesa/main',
+])
+
+sources = [
+	'glxapi.c',
+	'fakeglx.c',
+	'xfonts.c',
+	'xm_api.c',
+	'xm_winsys.c',
+	'xm_winsys_aub.c',
+	'brw_aub.c',
+]
+	
+drivers = [
+	softpipe,
+	i915simple,
+	i965simple,
+]
+
+# TODO: write a wrapper function http://www.scons.org/wiki/WrapperFunctions
+env.SharedLibrary(
+	target ='GL',
+	source = sources,
+	LIBS = glapi + mesa + drivers + auxiliaries + env['LIBS'],
+)
diff --git a/src/gallium/winsys/xlib/brw_aub.c b/src/gallium/winsys/xlib/brw_aub.c
index 541d50c6e4..10eedd8402 100644
--- a/src/gallium/winsys/xlib/brw_aub.c
+++ b/src/gallium/winsys/xlib/brw_aub.c
@@ -29,11 +29,13 @@
   *   Keith Whitwell <keith@tungstengraphics.com>
   */
 
+#include <stdio.h>
+#include <stdlib.h>
 #include "brw_aub.h"
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "imports.h"
-//#include "intel_winsys.h"
+#include "pipe/p_util.h"
+#include "pipe/p_debug.h"
 
 
 struct brw_aubfile {
@@ -350,9 +352,9 @@ struct brw_aubfile *brw_aubfile_create( void )
 
    i++;
 
-   if (_mesa_getenv("INTEL_AUBFILE")) {
-      val = snprintf(filename, sizeof(filename), "%s%d.aub", _mesa_getenv("INTEL_AUBFILE"), i%4);
-      _mesa_printf("--> Aub file: %s\n", filename);
+   if (getenv("INTEL_AUBFILE")) {
+      val = snprintf(filename, sizeof(filename), "%s%d.aub", getenv("INTEL_AUBFILE"), i%4);
+      debug_printf("--> Aub file: %s\n", filename);
       aubfile->file = fopen(filename, "w");
    }
    else {
@@ -360,12 +362,12 @@ struct brw_aubfile *brw_aubfile_create( void )
       if (val < 0 || val > sizeof(filename)) 
 	 strcpy(filename, "default.aub");   
    
-      _mesa_printf("--> Aub file: %s\n", filename);
+      debug_printf("--> Aub file: %s\n", filename);
       aubfile->file = fopen(filename, "w");
    }
 
    if (!aubfile->file) {
-      _mesa_printf("couldn't open aubfile\n");
+      debug_printf("couldn't open aubfile\n");
       exit(1);
    }
 
diff --git a/src/glx/x11/Makefile b/src/glx/x11/Makefile
index 5f74fcff06..b404727f08 100644
--- a/src/glx/x11/Makefile
+++ b/src/glx/x11/Makefile
@@ -35,7 +35,7 @@ SOURCES = \
 
 include $(TOP)/src/mesa/sources
 
-MESA_ASM_API = $(addprefix $(TOP)/src/mesa/, $(ASM_API))
+MESA_GLAPI_ASM_SOURCES = $(addprefix $(TOP)/src/mesa/, $(GLAPI_ASM_SOURCES))
 MESA_GLAPI_SOURCES = $(addprefix $(TOP)/src/mesa/, $(GLAPI_SOURCES))
 MESA_GLAPI_OBJECTS = $(addprefix $(TOP)/src/mesa/, $(GLAPI_OBJECTS))
 
@@ -70,11 +70,11 @@ $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME):  $(OBJECTS) Makefile
 		-install $(TOP)/$(LIB_DIR) $(GL_LIB_DEPS) $(OBJECTS)
 
 
-depend: $(SOURCES) $(MESA_GLAPI_SOURCES) $(MESA_ASM_API) Makefile
+depend: $(SOURCES) $(MESA_GLAPI_SOURCES) $(MESA_GLAPI_ASM_SOURCES) Makefile
 	rm -f depend
 	touch depend
 	$(MKDEP) $(MKDEP_OPTIONS) $(INCLUDES) $(SOURCES) \
-		$(MESA_GLAPI_SOURCES) $(MESA_ASM_API) 
+		$(MESA_GLAPI_SOURCES) $(MESA_GLAPI_ASM_SOURCES) 
 
 
 # Emacs tags
diff --git a/src/mesa/Makefile b/src/mesa/Makefile
index c8cb2b592f..2403223db2 100644
--- a/src/mesa/Makefile
+++ b/src/mesa/Makefile
@@ -11,19 +11,6 @@ GL_MINOR = 5
 GL_TINY = 0$(MESA_MAJOR)0$(MESA_MINOR)0$(MESA_TINY)
 
 
-PIPE_LIB = \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/i965simple/libi965simple.a
-
-ifeq ($(CONFIG_NAME), linux-cell)
-CELL_LIB = $(TOP)/src/gallium/drivers/cell/ppu/libcell.a
-CELL_LIB_SPU = $(TOP)/src/gallium/drivers/cell/spu/g3d_spu.a
-endif
-
-ifeq ($(CONFIG_NAME), linux-llvm)
-LLVM_LIB = $(TOP)/src/gallium/auxiliary/llvm/libgallivm.a
-endif
-
 .SUFFIXES : .cpp
 
 .c.o:
@@ -36,33 +23,14 @@ endif
 	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
 
 
-# Figure out what to make here
-default:
-	@if [ "${DRIVER_DIRS}" = "dri" ] ; then \
-		$(MAKE) linux-solo ; \
-	elif [ "${DRIVER_DIRS}" = "osmesa" ] ; then \
-		$(MAKE) osmesa-only ; \
-	elif [ "$(DRIVER_DIRS)" = "beos" ]; then \
-		$(MAKE) beos ; \
-	elif [ "$(DRIVER_DIRS)" = "directfb" ]; then \
-		$(MAKE) directfb ; \
-	elif [ "$(DRIVER_DIRS)" = "fbdev osmesa" ]; then \
-		$(MAKE) fbdev ; $(MAKE) osmesa-only ; \
-	else \
-		$(MAKE) stand-alone ; \
-	fi
-
+default: depend subdirs libmesa.a
 
-######################################################################
-# BeOS driver target
-
-beos: depend subdirs libmesa.a
-	cd drivers/beos; $(MAKE)
+ifneq ($(DRIVER_DIRS),dri)
+default: libglapi.a
+endif
 
 
 ######################################################################
-# Linux DRI drivers
-
 # Make archive of core object files
 libmesa.a: $(SOLO_OBJECTS)
 	@ $(TOP)/bin/mklib -o mesa -static $(SOLO_OBJECTS);
@@ -70,32 +38,8 @@ libmesa.a: $(SOLO_OBJECTS)
 		mimeset -f "$@" ; \
 	fi
 
-linux-solo: depend subdirs libmesa.a
-	cd $(TOP)/src/gallium/winsys/dri ; $(MAKE)
-
-
-#####################################################################
-# Stand-alone Mesa libGL, no built-in drivers (DirectFB)
-
-libgl-core: $(CORE_OBJECTS)
-	@ $(TOP)/bin/mklib -o $(GL_LIB) \
-		-major $(MESA_MAJOR) -minor $(MESA_MINOR) -patch $(MESA_TINY) \
-		-install $(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) $(CORE_OBJECTS) \
-		$(GL_LIB_DEPS)
-
-directfb: depend subdirs libgl-core
-	cd drivers/directfb ; $(MAKE)
-
-
-#####################################################################
-# fbdev Mesa driver (libGL.so)
-
-fbdev: $(CORE_OBJECTS) $(FBDEV_DRIVER_OBJECTS) $(COMMON_DRIVER_OBJECTS)
-	@ $(TOP)/bin/mklib -o $(GL_LIB) \
-		-major $(MESA_MAJOR) -minor $(MESA_MINOR) -patch $(MESA_TINY) \
-		-install $(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
-		$(CORE_OBJECTS) $(FBDEV_DRIVER_OBJECTS) \
-		$(COMMON_DRIVER_OBJECTS) $(GL_LIB_DEPS)
+libglapi.a: $(GLAPI_OBJECTS)
+	@ $(TOP)/bin/mklib -o glapi -static $(GLAPI_OBJECTS)
 
 
 ######################################################################
@@ -123,13 +67,13 @@ stand-alone: depend subdirs $(TOP)/$(LIB_DIR)/$(GL_LIB_NAME) $(TOP)/$(LIB_DIR)/$
 osmesa-only: depend subdirs $(TOP)/$(LIB_DIR)/$(OSMESA_LIB_NAME)
 
 # Make the GL library
-$(TOP)/$(LIB_DIR)/$(GL_LIB_NAME): $(STAND_ALONE_OBJECTS) $(PIPE_LIB) $(CELL_LIB) $(CELL_LIB_SPU) $(LLVM_LIB)
+$(TOP)/$(LIB_DIR)/$(GL_LIB_NAME): $(STAND_ALONE_OBJECTS) $(PIPE_LIB) $(CELL_LIB) $(CELL_LIB_SPU)
 	@ $(TOP)/bin/mklib -o $(GL_LIB) \
 		-linker "$(CC)" \
 		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
 		-install $(TOP)/$(LIB_DIR) \
 		$(MKLIB_OPTIONS) $(STAND_ALONE_OBJECTS) \
-		$(PIPE_LIB) $(CELL_LIB) $(CELL_LIB_SPU) $(LLVM_LIB) $(GL_LIB_DEPS)
+		$(PIPE_LIB) $(CELL_LIB) $(CELL_LIB_SPU) $(GL_LIB_DEPS)
 
 # Make the OSMesa library
 $(TOP)/$(LIB_DIR)/$(OSMESA_LIB_NAME): $(OSMESA_DRIVER_OBJECTS) \
@@ -176,9 +120,6 @@ install: default
 	@if [ -e $(TOP)/$(LIB_DIR)/$(OSMESA_LIB_NAME) ]; then \
 		$(INSTALL) $(TOP)/$(LIB_DIR)/libOSMesa* $(INSTALL_DIR)/$(LIB_DIR); \
 	fi
-	@if [ "${DRIVER_DIRS}" = "dri" ] ; then \
-		cd $(TOP)/gallium/winsys/dri ; $(MAKE) install ; \
-	fi
 
 ## NOT INSTALLED YET:
 ## $(INSTALL) -d $(INSTALL_DIR)/include/GLES
@@ -192,9 +133,8 @@ tags:
 clean:
 	-rm -f */*.o
 	-rm -f */*/*.o
-	-rm -f depend depend.bak libmesa.a
+	-rm -f depend depend.bak libmesa.a libglapi.a
 	-rm -f drivers/*/*.o
-	(cd drivers/dri && $(MAKE) clean)
 	(cd x86 && $(MAKE) clean)
 	(cd x86-64 && $(MAKE) clean)
 
diff --git a/src/mesa/SConscript b/src/mesa/SConscript
index faf8c84872..db18c61fac 100644
--- a/src/mesa/SConscript
+++ b/src/mesa/SConscript
@@ -1,11 +1,16 @@
 #######################################################################
-# SConscript for mesa
-#
-# TODO: Split this into per-module SConscripts 
+# SConscript for Mesa
 
 
 Import('*')
 
+env = env.Clone()
+
+# Includes
+env.Append(CPPPATH = [
+	'#/src/mesa',
+	'#/src/mesa/main',
+])
 
 #######################################################################
 # Core sources
@@ -116,53 +121,6 @@ VF_SOURCES = [
 	'vf/vf_sse.c',
 ]
 
-DRAW_SOURCES = [
-	'pipe/draw/draw_clip.c',
-	'pipe/draw/draw_context.c',
-	'pipe/draw/draw_cull.c',
-	'pipe/draw/draw_debug.c',
-	'pipe/draw/draw_flatshade.c',
-	'pipe/draw/draw_offset.c',
-	'pipe/draw/draw_prim.c',
-	'pipe/draw/draw_stipple.c',
-	'pipe/draw/draw_twoside.c',
-	'pipe/draw/draw_unfilled.c',
-	'pipe/draw/draw_validate.c',
-	'pipe/draw/draw_vbuf.c',
-	'pipe/draw/draw_vertex.c',
-	'pipe/draw/draw_vertex_cache.c',
-	'pipe/draw/draw_vertex_fetch.c',
-	'pipe/draw/draw_vertex_shader.c',
-	'pipe/draw/draw_vertex_shader_llvm.c',
-	'pipe/draw/draw_vf.c',
-	'pipe/draw/draw_vf_generic.c',
-	'pipe/draw/draw_vf_sse.c',
-	'pipe/draw/draw_wide_prims.c',
-]
-
-TGSIEXEC_SOURCES = [
-	'pipe/tgsi/exec/tgsi_exec.c',
-	'pipe/tgsi/exec/tgsi_sse2.c',
-]
-
-TGSIUTIL_SOURCES = [
-	'pipe/tgsi/util/tgsi_build.c',
-	'pipe/tgsi/util/tgsi_dump.c',
-	'pipe/tgsi/util/tgsi_parse.c',
-	'pipe/tgsi/util/tgsi_util.c',
-]
-
-STATECACHE_SOURCES = [
-	'pipe/cso_cache/cso_hash.c',
-	'pipe/cso_cache/cso_cache.c',
-]
-
-PIPEUTIL_SOURCES = [
-	'pipe/util/p_debug.c',
-	'pipe/util/p_tile.c',
-	'pipe/util/p_util.c',
-]
-
 STATETRACKER_SOURCES = [
 	'state_tracker/st_atom.c',
 	'state_tracker/st_atom_blend.c',
@@ -311,126 +269,25 @@ else:
 	ASM_SOURCES = []
 	API_SOURCES = []
 
-
-#######################################################################
-# Driver sources
-
-
-X11_DRIVER_SOURCES = [
-	'pipe/xlib/glxapi.c',
-	'pipe/xlib/fakeglx.c',
-	'pipe/xlib/xfonts.c',
-	'pipe/xlib/xm_api.c',
-	'pipe/xlib/xm_winsys.c',
-	'pipe/xlib/xm_winsys_aub.c',
-	'pipe/xlib/brw_aub.c',
-]
-
-OSMESA_DRIVER_SOURCES = [
-	'drivers/osmesa/osmesa.c',
-]
-
-GLIDE_DRIVER_SOURCES = [
-	'drivers/glide/fxapi.c',
-	'drivers/glide/fxdd.c',
-	'drivers/glide/fxddspan.c',
-	'drivers/glide/fxddtex.c',
-	'drivers/glide/fxsetup.c',
-	'drivers/glide/fxtexman.c',
-	'drivers/glide/fxtris.c',
-	'drivers/glide/fxvb.c',
-	'drivers/glide/fxglidew.c',
-	'drivers/glide/fxg.c',
-]
-
-SVGA_DRIVER_SOURCES = [
-	'drivers/svga/svgamesa.c',
-	'drivers/svga/svgamesa8.c',
-	'drivers/svga/svgamesa15.c',
-	'drivers/svga/svgamesa16.c',
-	'drivers/svga/svgamesa24.c',
-	'drivers/svga/svgamesa32.c',
-]
-
-FBDEV_DRIVER_SOURCES = [
-	'drivers/fbdev/glfbdev.c',
-]
-
-
-### All the core C sources
-
 SOLO_SOURCES = \
 	MAIN_SOURCES + \
 	MATH_SOURCES + \
 	VBO_SOURCES + \
 	VF_SOURCES + \
-	DRAW_SOURCES + \
-	TGSIEXEC_SOURCES + \
-	TGSIUTIL_SOURCES + \
-	PIPEUTIL_SOURCES + \
-	STATECACHE_SOURCES + \
 	STATETRACKER_SOURCES + \
 	SHADER_SOURCES + \
 	ASM_SOURCES + \
 	SLANG_SOURCES
 
-CORE_SOURCES = \
-	GLAPI_SOURCES + API_SOURCES + \
-	SOLO_SOURCES
-
-ALL_SOURCES = \
-	GLAPI_SOURCES + API_SOURCES + \
-	SOLO_SOURCES + \
-	ASM_SOURCES + \
-	X11_DRIVER_SOURCES + \
-	FBDEV_DRIVER_SOURCES + \
-	OSMESA_DRIVER_SOURCES
-
-
-######################################################################
-# Gallium sources
-
-SConscript([
-	'pipe/SConscript',
-])
-
-
-######################################################################
-# libGL
+mesa = env.ConvenienceLibrary(
+	target = 'mesa',
+	source = SOLO_SOURCES,
+)
+Export('mesa')
 
 if not dri:
-	STAND_ALONE_DRIVER_SOURCES = \
-		CORE_SOURCES + \
-		X11_DRIVER_SOURCES
-	
-	Import(
-		'softpipe', 
-		'i915simple',
-		'i965simple'
-	)
-	
-	pipe_drivers = [
-		softpipe,
-		i965simple
-	]
-	
-	env.SharedLibrary(
-		target ='GL',
-		source = STAND_ALONE_DRIVER_SOURCES,
-		LIBS = [softpipe, i965simple] + env['LIBS'],
+	glapi = env.ConvenienceLibrary(
+		target = 'glapi',
+		source = GLAPI_SOURCES + API_SOURCES,
 	)
-
-
-######################################################################
-# Driver sources
-
-if dri:
-	mesa = env.ConvenienceLibrary(
-		target = 'mesa',
-		source = SOLO_SOURCES,
-	)
-	env.Prepend(LIBS = [mesa])
-
-	SConscript([
-		'drivers/dri/SConscript',
-	])
+	Export('glapi')
diff --git a/src/mesa/sources b/src/mesa/sources
index f83d247a1e..f0bf7b31fb 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -156,51 +156,6 @@ VF_SOURCES = \
 	vf/vf_generic.c \
 	vf/vf_sse.c
 
-
-DRAW_SOURCES = \
-	$(TOP)/src/gallium/auxiliary/draw/draw_clip.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vs_exec.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vs_sse.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vs_llvm.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_context.c\
-	$(TOP)/src/gallium/auxiliary/draw/draw_cull.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_debug.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_flatshade.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_offset.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_prim.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_stipple.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_twoside.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_unfilled.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_validate.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vbuf.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vertex.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vertex_cache.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vertex_fetch.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vertex_shader.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vf.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vf_generic.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_vf_sse.c \
-	$(TOP)/src/gallium/auxiliary/draw/draw_wide_prims.c
-
-TGSIEXEC_SOURCES = \
-	$(TOP)/src/gallium/auxiliary/tgsi/exec/tgsi_exec.c \
-	$(TOP)/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
-
-TGSIUTIL_SOURCES = \
-	$(TOP)/src/gallium/auxiliary/tgsi/util/tgsi_build.c \
-	$(TOP)/src/gallium/auxiliary/tgsi/util/tgsi_dump.c \
-	$(TOP)/src/gallium/auxiliary/tgsi/util/tgsi_parse.c \
-	$(TOP)/src/gallium/auxiliary/tgsi/util/tgsi_util.c
-
-STATECACHE_SOURCES = \
-	$(TOP)/src/gallium/auxiliary/cso_cache/cso_hash.c \
-	$(TOP)/src/gallium/auxiliary/cso_cache/cso_cache.c
-
-PIPEUTIL_SOURCES = \
-	$(TOP)/src/gallium/auxiliary/util/p_debug.c \
-	$(TOP)/src/gallium/auxiliary/util/p_tile.c \
-	$(TOP)/src/gallium/auxiliary/util/p_util.c
-
 STATETRACKER_SOURCES = \
 	state_tracker/st_atom.c \
 	state_tracker/st_atom_blend.c \
@@ -229,7 +184,7 @@ STATETRACKER_SOURCES = \
 	state_tracker/st_cb_readpixels.c \
 	state_tracker/st_cb_strings.c \
 	state_tracker/st_cb_texture.c \
-        state_tracker/st_cache.c \
+	state_tracker/st_cache.c \
 	state_tracker/st_context.c \
 	state_tracker/st_debug.c \
 	state_tracker/st_draw.c \
@@ -291,7 +246,6 @@ ASM_C_SOURCES =	\
 	x86/rtasm/x86sse.c \
 	sparc/sparc.c \
 	ppc/common_ppc.c \
-	ppc/rtasm/spe_asm.c \
 	x86-64/x86-64.c
 
 X86_SOURCES =			\
@@ -333,15 +287,6 @@ SPARC_API =			\
 __COMMON_DRIVER_SOURCES =			\
 	drivers/common/driverfuncs.c
 
-X11_DRIVER_SOURCES =		\
-	$(TOP)/src/gallium/winsys/xlib/glxapi.c	\
-	$(TOP)/src/gallium/winsys/xlib/fakeglx.c	\
-	$(TOP)/src/gallium/winsys/xlib/xfonts.c	\
-	$(TOP)/src/gallium/winsys/xlib/xm_api.c	\
-	$(TOP)/src/gallium/winsys/xlib/xm_winsys.c	\
-	$(TOP)/src/gallium/winsys/xlib/xm_winsys_aub.c	\
-	$(TOP)/src/gallium/winsys/xlib/brw_aub.c
-
 OSMESA_DRIVER_SOURCES = \
 	drivers/osmesa/osmesa.c
 
@@ -374,7 +319,7 @@ FBDEV_DRIVER_SOURCES =			\
 ALL_SOURCES = \
 	$(GLAPI_SOURCES)	\
 	$(SOLO_SOURCES)		\
-	$(ASM_SOURCES)		\
+	$(MESA_ASM_SOURCES)		\
 	$(COMMON_DRIVER_SOURCES)\
 	$(X11_DRIVER_SOURCES)	\
 	$(FBDEV_DRIVER_SOURCES) \
@@ -407,11 +352,11 @@ CORE_SOURCES = \
 
 SOLO_OBJECTS = \
 	$(SOLO_SOURCES:.c=.o) \
-	$(ASM_SOURCES:.S=.o)
+	$(MESA_ASM_SOURCES:.S=.o)
 
 GLAPI_OBJECTS = \
 	$(GLAPI_SOURCES:.c=.o) \
-	$(ASM_API:.S=.o)
+	$(GLAPI_ASM_SOURCES:.S=.o)
 
 CORE_OBJECTS = $(SOLO_OBJECTS) $(GLAPI_OBJECTS)
 
diff --git a/src/mesa/state_tracker/st_atom_texture.c b/src/mesa/state_tracker/st_atom_texture.c
index 2a836d630b..a4ac726816 100644
--- a/src/mesa/state_tracker/st_atom_texture.c
+++ b/src/mesa/state_tracker/st_atom_texture.c
@@ -34,6 +34,7 @@
 
 #include "st_context.h"
 #include "st_atom.h"
+#include "st_texture.h"
 #include "st_cb_texture.h"
 #include "pipe/p_context.h"
 
@@ -53,28 +54,33 @@ update_textures(struct st_context *st)
    for (unit = 0; unit < st->ctx->Const.MaxTextureCoordUnits; unit++) {
       const GLuint su = fprog->Base.SamplerUnits[unit];
       struct gl_texture_object *texObj = st->ctx->Texture.Unit[su]._Current;
-      struct pipe_texture *pt;
+      struct st_texture_object *stObj = st_texture_object(texObj);
 
       if (texObj) {
          GLboolean flush, retval;
 
          retval = st_finalize_texture(st->ctx, st->pipe, texObj, &flush);
          /* XXX retval indicates whether there's a texture border */
-
-         pt = st_get_texobj_texture(texObj);
-      }
-      else {
-         pt = NULL;
       }
 
       /* XXX: need to ensure that textures are unbound/removed from
        * this table before being deleted, otherwise the pointer
        * comparison below could fail.
        */
-      if (st->state.sampler_texture[unit] != pt) {
-	 st->state.sampler_texture[unit] = pt;
-	 st->pipe->set_sampler_texture(st->pipe, unit, pt);
+      if (st->state.sampler_texture[unit] != stObj) {
+         struct pipe_texture *pt = st_get_stobj_texture(stObj);
+         st->state.sampler_texture[unit] = stObj;
+         st->pipe->set_sampler_texture(st->pipe, unit, pt);
+      }
+
+      stObj = st->state.sampler_texture[unit];
+
+      if (stObj && stObj->dirtyData) {
+         struct pipe_texture *pt = st_get_stobj_texture(stObj);
+         st->pipe->texture_update(st->pipe, pt);
+         stObj->dirtyData = GL_FALSE;
       }
+
    }
 }
 
diff --git a/src/mesa/state_tracker/st_cb_drawpixels.c b/src/mesa/state_tracker/st_cb_drawpixels.c
index e2d4e06da1..585cae3743 100644
--- a/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -726,7 +726,8 @@ draw_textured_quad(GLcontext *ctx, GLint x, GLint y, GLfloat z,
    pipe->bind_rasterizer_state(pipe, ctx->st->state.rasterizer->data);
    pipe->bind_fs_state(pipe, ctx->st->state.fs->data);
    pipe->bind_vs_state(pipe, ctx->st->state.vs->cso->data);
-   pipe->set_sampler_texture(pipe, unit, ctx->st->state.sampler_texture[unit]);
+   pipe->set_sampler_texture(pipe, unit,
+                st_get_stobj_texture(ctx->st->state.sampler_texture[unit]));
    pipe->bind_sampler_state(pipe, unit, ctx->st->state.sampler[unit]->data);
    pipe->set_viewport_state(pipe, &ctx->st->state.viewport);
 }
diff --git a/src/mesa/state_tracker/st_cb_fbo.c b/src/mesa/state_tracker/st_cb_fbo.c
index 4341623267..781425b546 100644
--- a/src/mesa/state_tracker/st_cb_fbo.c
+++ b/src/mesa/state_tracker/st_cb_fbo.c
@@ -48,6 +48,7 @@
 #include "st_cb_texture.h"
 #include "st_format.h"
 #include "st_public.h"
+#include "st_texture.h"
 
 
 
diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c
index 03dbb30b0f..778fb536bc 100644
--- a/src/mesa/state_tracker/st_cb_texture.c
+++ b/src/mesa/state_tracker/st_cb_texture.c
@@ -53,33 +53,6 @@
 #define DBG if (0) printf
 
 
-struct st_texture_object
-{
-   struct gl_texture_object base;       /* The "parent" object */
-
-   /* The texture must include at levels [0..lastLevel] once validated:
-    */
-   GLuint lastLevel;
-
-   /* On validation any active images held in main memory or in other
-    * textures will be copied to this texture and the old storage freed.
-    */
-   struct pipe_texture *pt;
-
-   GLboolean imageOverride;
-   GLint depthOverride;
-   GLuint pitchOverride;
-};
-
-
-
-static INLINE struct st_texture_object *
-st_texture_object(struct gl_texture_object *obj)
-{
-   return (struct st_texture_object *) obj;
-}
-
-
 static INLINE struct st_texture_image *
 st_texture_image(struct gl_texture_image *img)
 {
@@ -87,14 +60,6 @@ st_texture_image(struct gl_texture_image *img)
 }
 
 
-struct pipe_texture *
-st_get_texobj_texture(struct gl_texture_object *texObj)
-{
-   struct st_texture_object *stObj = st_texture_object(texObj);
-   return stObj->pt;
-}
-
-
 static enum pipe_texture_target
 gl_target_to_pipe(GLenum target)
 {
@@ -725,11 +690,12 @@ st_TexImage(GLcontext * ctx,
       texImage->Data = NULL;
    }
 
-#if 01
+   /* flag data as dirty */
+   stObj->dirtyData = GL_TRUE;
+
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
       ctx->Driver.GenerateMipmap(ctx, target, texObj);
    }
-#endif
 }
 
 
@@ -900,6 +866,7 @@ st_TexSubimage(GLcontext * ctx,
                  struct gl_texture_object *texObj,
                  struct gl_texture_image *texImage)
 {
+   struct st_texture_object *stObj = st_texture_object(texObj);
    struct st_texture_image *stImage = st_texture_image(texImage);
    GLuint dstRowStride;
    GLuint srcImageStride = _mesa_image_image_stride(packing, width, height,
@@ -946,14 +913,9 @@ st_TexSubimage(GLcontext * ctx,
       }
    }
 
-#if 0
-   /* GL_SGIS_generate_mipmap */
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-      _mesa_generate_mipmap(ctx, target,
-                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
-                            texObj);
+      ctx->Driver.GenerateMipmap(ctx, target, texObj);
    }
-#endif
 
    _mesa_unmap_teximage_pbo(ctx, packing);
 
@@ -961,6 +923,9 @@ st_TexSubimage(GLcontext * ctx,
       st_texture_image_unmap(stImage);
       texImage->Data = NULL;
    }
+
+   /* flag data as dirty */
+   stObj->dirtyData = GL_TRUE;
 }
 
 
@@ -1126,6 +1091,7 @@ do_copy_texsubimage(GLcontext *ctx,
    struct gl_texture_image *texImage =
       _mesa_select_tex_image(ctx, texObj, target, level);
    struct st_texture_image *stImage = st_texture_image(texImage);
+   struct st_texture_object *stObj = st_texture_object(texObj);
    GLenum baseFormat = texImage->InternalFormat;
    struct gl_framebuffer *fb = ctx->ReadBuffer;
    struct st_renderbuffer *strb;
@@ -1212,17 +1178,12 @@ do_copy_texsubimage(GLcontext *ctx,
 
    pipe_surface_reference(&dest_surface, NULL);
 
-#if 0
-   /* GL_SGIS_generate_mipmap -- this can be accelerated now.
-    * XXX Add a ctx->Driver.GenerateMipmaps() function?
-    */
+   /* flag data as dirty */
+   stObj->dirtyData = GL_TRUE;
+
    if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
-      intel_generate_mipmap(ctx, target,
-                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
-                            texObj);
+      ctx->Driver.GenerateMipmap(ctx, target, texObj);
    }
-#endif
-
 }
 
 
diff --git a/src/mesa/state_tracker/st_cb_texture.h b/src/mesa/state_tracker/st_cb_texture.h
index 878256ec26..843745fcd6 100644
--- a/src/mesa/state_tracker/st_cb_texture.h
+++ b/src/mesa/state_tracker/st_cb_texture.h
@@ -1,9 +1,33 @@
-#ifndef ST_CB_TEXTURE_H
-#define ST_CB_TEXTURE_H
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
 
 
-extern struct pipe_texture *
-st_get_texobj_texture(struct gl_texture_object *texObj);
+#ifndef ST_CB_TEXTURE_H
+#define ST_CB_TEXTURE_H
 
 
 extern GLboolean
diff --git a/src/mesa/state_tracker/st_context.h b/src/mesa/state_tracker/st_context.h
index 59d1590f05..5be4769be4 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -59,26 +59,6 @@ struct st_tracked_state {
 
 
 
-struct st_texture_image
-{
-   struct gl_texture_image base;
-
-   /* These aren't stored in gl_texture_image 
-    */
-   GLuint level;
-   GLuint face;
-
-   /* If stImage->pt != NULL, image data is stored here.
-    * Else if stImage->base.Data != NULL, image is stored there.
-    * Else there is no image data.
-    */
-   struct pipe_texture *pt;
-
-   struct pipe_surface *surface;
-};
-
-
-
 struct st_context
 {
    GLcontext *ctx;
@@ -106,7 +86,7 @@ struct st_context
       struct pipe_clip_state clip;
       struct pipe_constant_buffer constants[2];
       struct pipe_framebuffer_state framebuffer;
-      struct pipe_texture *sampler_texture[PIPE_MAX_SAMPLERS];
+      struct st_texture_object *sampler_texture[PIPE_MAX_SAMPLERS];
       struct pipe_poly_stipple poly_stipple;
       struct pipe_scissor_state scissor;
       struct pipe_viewport_state viewport;
diff --git a/src/mesa/state_tracker/st_gen_mipmap.c b/src/mesa/state_tracker/st_gen_mipmap.c
index 6c09b86033..c9765b2003 100644
--- a/src/mesa/state_tracker/st_gen_mipmap.c
+++ b/src/mesa/state_tracker/st_gen_mipmap.c
@@ -43,6 +43,7 @@
 #include "st_draw.h"
 #include "st_gen_mipmap.h"
 #include "st_program.h"
+#include "st_texture.h"
 #include "st_cb_drawpixels.h"
 #include "st_cb_texture.h"
 
@@ -110,14 +111,22 @@ st_init_generate_mipmap(struct st_context *st)
    struct pipe_rasterizer_state rasterizer;
    struct pipe_depth_stencil_alpha_state depthstencil;
 
+   /* we don't use blending, but need to set valid values */
    memset(&blend, 0, sizeof(blend));
+   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
    blend.colormask = PIPE_MASK_RGBA;
    st->gen_mipmap.blend_cso = pipe->create_blend_state(pipe, &blend);
 
    memset(&depthstencil, 0, sizeof(depthstencil));
    st->gen_mipmap.depthstencil_cso = pipe->create_depth_stencil_alpha_state(pipe, &depthstencil);
 
+   /* Note: we're assuming zero is valid for all non-specified fields */
    memset(&rasterizer, 0, sizeof(rasterizer));
+   rasterizer.front_winding = PIPE_WINDING_CW;
+   rasterizer.cull_mode = PIPE_WINDING_NONE;
    st->gen_mipmap.rasterizer_cso = pipe->create_rasterizer_state(pipe, &rasterizer);
 
    st->gen_mipmap.stfp = make_tex_fragment_program(st->ctx);
@@ -302,7 +311,8 @@ st_render_mipmap(struct st_context *st,
       pipe->bind_vs_state(pipe, st->state.vs->cso->data);
    if (st->state.sampler[0])
       pipe->bind_sampler_state(pipe, 0, st->state.sampler[0]->data);
-   pipe->set_sampler_texture(pipe, 0, st->state.sampler_texture[0]);
+   pipe->set_sampler_texture(pipe, 0,
+                        st_get_stobj_texture(st->state.sampler_texture[0]));
    pipe->set_viewport_state(pipe, &st->state.viewport);
 
    return TRUE;
diff --git a/src/mesa/state_tracker/st_program.h b/src/mesa/state_tracker/st_program.h
index ea1dde4a7a..25cf3e94a8 100644
--- a/src/mesa/state_tracker/st_program.h
+++ b/src/mesa/state_tracker/st_program.h
@@ -36,7 +36,6 @@
 
 #include "mtypes.h"
 #include "pipe/p_shader_tokens.h"
-#include "x86/rtasm/x86sse.h"
 
 
 #define ST_MAX_SHADER_TOKENS 1024
diff --git a/src/mesa/state_tracker/st_texture.c b/src/mesa/state_tracker/st_texture.c
index b86f416c9b..ad284170e4 100644
--- a/src/mesa/state_tracker/st_texture.c
+++ b/src/mesa/state_tracker/st_texture.c
@@ -76,7 +76,7 @@ st_texture_create(struct st_context *st,
 		  GLuint depth0,
 		  GLuint compress_byte)
 {
-   struct pipe_texture pt;
+   struct pipe_texture pt, *newtex;
 
    assert(target <= PIPE_TEXTURE_CUBE);
 
@@ -95,9 +95,12 @@ st_texture_create(struct st_context *st,
    pt.depth[0] = depth0;
    pt.compressed = compress_byte ? 1 : 0;
    pt.cpp = pt.compressed ? compress_byte : st_sizeof_format(format);
-   pt.refcount = 1; 
 
-   return st->pipe->texture_create(st->pipe, &pt);
+   newtex = st->pipe->texture_create(st->pipe, &pt);
+
+   assert(!newtex || newtex->refcount == 1);
+
+   return newtex;
 }
 
 
diff --git a/src/mesa/state_tracker/st_texture.h b/src/mesa/state_tracker/st_texture.h
index 72324cd9ab..78f5f451ed 100644
--- a/src/mesa/state_tracker/st_texture.h
+++ b/src/mesa/state_tracker/st_texture.h
@@ -35,6 +35,70 @@ struct pipe_context;
 struct pipe_texture;
 
 
+struct st_texture_image
+{
+   struct gl_texture_image base;
+
+   /* These aren't stored in gl_texture_image 
+    */
+   GLuint level;
+   GLuint face;
+
+   /* If stImage->pt != NULL, image data is stored here.
+    * Else if stImage->base.Data != NULL, image is stored there.
+    * Else there is no image data.
+    */
+   struct pipe_texture *pt;
+
+   struct pipe_surface *surface;
+};
+
+
+
+struct st_texture_object
+{
+   struct gl_texture_object base;       /* The "parent" object */
+
+   /* The texture must include at levels [0..lastLevel] once validated:
+    */
+   GLuint lastLevel;
+
+   /* On validation any active images held in main memory or in other
+    * textures will be copied to this texture and the old storage freed.
+    */
+   struct pipe_texture *pt;
+
+   GLboolean imageOverride;
+   GLint depthOverride;
+   GLuint pitchOverride;
+
+   GLboolean dirtyData;
+};
+
+
+static INLINE struct st_texture_object *
+st_texture_object(struct gl_texture_object *obj)
+{
+   return (struct st_texture_object *) obj;
+}
+
+
+static INLINE struct pipe_texture *
+st_get_texobj_texture(struct gl_texture_object *texObj)
+{
+   struct st_texture_object *stObj = st_texture_object(texObj);
+   return stObj ? stObj->pt : NULL;
+}
+
+
+static INLINE struct pipe_texture *
+st_get_stobj_texture(struct st_texture_object *stObj)
+{
+   return stObj ? stObj->pt : NULL;
+}
+
+
+
 extern struct pipe_texture *
 st_texture_create(struct st_context *st,
                   enum pipe_texture_target target,
diff --git a/src/mesa/x86/common_x86_asm.S b/src/mesa/x86/common_x86_asm.S
index ef3cc9eb59..09c86b05ba 100644
--- a/src/mesa/x86/common_x86_asm.S
+++ b/src/mesa/x86/common_x86_asm.S
@@ -39,7 +39,7 @@
  * in there will break the build on some platforms.
  */
 
-#include "matypes.h"
+#include "assyntax.h"
 #include "common_x86_features.h"
 
 	SEG_TEXT