Merge branch 'master' into r300-compiler

214 files changed, 13988 insertions, 10981 deletions

diff --git a/src/gallium/SConscript b/src/gallium/SConscript
index 89c69d7205..8be84cddbe 100644
--- a/src/gallium/SConscript
+++ b/src/gallium/SConscript
@@ -23,6 +23,7 @@ SConscript([
 	'auxiliary/pipebuffer/SConscript',
 	'auxiliary/indices/SConscript',
 	'auxiliary/rbug/SConscript',
+	'auxiliary/vl/SConscript',
 ])
 
 for driver in env['drivers']:
diff --git a/src/gallium/auxiliary/draw/draw_pt_post_vs.c b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
index 00d7197b13..e25f16c354 100644
--- a/src/gallium/auxiliary/draw/draw_pt_post_vs.c
+++ b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
@@ -104,7 +104,7 @@ static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs,
    unsigned clipped = 0;
    unsigned j;
 
-   if (0) debug_printf("%s\n");
+   if (0) debug_printf("%s\n", __FUNCTION__);
 
    for (j = 0; j < count; j++) {
       float *position = out->data[pos];
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index 109ac7c9d6..d01f866622 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -542,7 +542,7 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
       debug_printf("%10p %7u %7u\n",
                    fenced_buf,
                    fenced_buf->base.base.size,
-                   fenced_buf->base.base.reference.count);
+                   p_atomic_read(&fenced_buf->base.base.reference.count));
       curr = next; 
       next = curr->next;
    }
@@ -556,7 +556,7 @@ fenced_buffer_list_dump(struct fenced_buffer_list *fenced_list)
       debug_printf("%10p %7u %7u %10p %s\n",
                    fenced_buf,
                    fenced_buf->base.base.size,
-                   fenced_buf->base.base.reference.count,
+                   p_atomic_read(&fenced_buf->base.base.reference.count),
                    fenced_buf->fence,
                    signaled == 0 ? "y" : "n");
       curr = next; 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
index 1b4df28c70..6e3214ca9c 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
@@ -350,7 +350,7 @@ pb_debug_manager_dump(struct pb_debug_manager *mgr)
       buf = LIST_ENTRY(struct pb_debug_buffer, curr, head);
 
       debug_printf("buffer = %p\n", buf);
-      debug_printf("    .size = %p\n", buf->base.base.size);
+      debug_printf("    .size = 0x%x\n", buf->base.base.size);
       debug_backtrace_dump(buf->create_backtrace, PB_DEBUG_CREATE_BACKTRACE);
       
       curr = next; 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 501fc05e72..5f6b83b236 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -1445,11 +1445,11 @@ fetch_texel( struct tgsi_sampler **sampler,
    {
       float rgba[NUM_CHANNELS][QUAD_SIZE];
       (*sampler)->get_samples(*sampler, 
-                              &store[0], 
-                              &store[4], 
-                              &store[8], 
-                              0.0f, /*store[12],  lodbias */
-                              rgba);
+                              &store[0],  /* s */
+                              &store[4],  /* t */
+                              &store[8],  /* r */
+                              store[12],  /* lodbias */
+                              rgba);      /* results */
 
       memcpy( store, rgba, 16 * sizeof(float));
    }
@@ -1855,7 +1855,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_RCP:
-   /* TGSI_OPCODE_RECIP */
       FETCH( func, *inst, 0, 0, CHAN_X );
       emit_rcp( func, 0, 0 );
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
@@ -1864,7 +1863,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_RSQ:
-   /* TGSI_OPCODE_RECIPSQRT */
       FETCH( func, *inst, 0, 0, CHAN_X );
       emit_abs( func, 0 );
       emit_rsqrt( func, 1, 0 );
@@ -1962,7 +1960,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_DP3:
-   /* TGSI_OPCODE_DOT3 */
       FETCH( func, *inst, 0, 0, CHAN_X );
       FETCH( func, *inst, 1, 1, CHAN_X );
       emit_mul( func, 0, 1 );
@@ -1980,7 +1977,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_DP4:
-   /* TGSI_OPCODE_DOT4 */
       FETCH( func, *inst, 0, 0, CHAN_X );
       FETCH( func, *inst, 1, 1, CHAN_X );
       emit_mul( func, 0, 1 );
@@ -2051,17 +2047,14 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SLT:
-   /* TGSI_OPCODE_SETLT */
       emit_setcc( func, inst, cc_LessThan );
       break;
 
    case TGSI_OPCODE_SGE:
-   /* TGSI_OPCODE_SETGE */
       emit_setcc( func, inst, cc_NotLessThan );
       break;
 
    case TGSI_OPCODE_MAD:
-   /* TGSI_OPCODE_MADD */
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
          FETCH( func, *inst, 0, 0, chan_index );
          FETCH( func, *inst, 1, 1, chan_index );
@@ -2291,7 +2284,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SEQ:
-      return 0;
+      emit_setcc( func, inst, cc_Equal );
       break;
 
    case TGSI_OPCODE_SFL:
@@ -2299,7 +2292,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SGT:
-      return 0;
+      emit_setcc( func, inst, cc_NotLessThanEqual );
       break;
 
    case TGSI_OPCODE_SIN:
@@ -2311,11 +2304,11 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SLE:
-      return 0;
+      emit_setcc( func, inst, cc_LessThanEqual );
       break;
 
    case TGSI_OPCODE_SNE:
-      return 0;
+      emit_setcc( func, inst, cc_NotEqual );
       break;
 
    case TGSI_OPCODE_STR:
@@ -2379,7 +2372,6 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_SSG:
-   /* TGSI_OPCODE_SGN */
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
          FETCH( func, *inst, 0, 0, chan_index );
          emit_sgn( func, 0, 0 );
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
index ae8d330a78..1d8bb55bbd 100644
--- a/src/gallium/auxiliary/util/Makefile
+++ b/src/gallium/auxiliary/util/Makefile
@@ -10,6 +10,7 @@ C_SOURCES = \
 	u_debug_stack.c \
 	u_blit.c \
 	u_cache.c \
+	u_cpu_detect.c \
 	u_draw_quad.c \
 	u_format.c \
 	u_format_access.c \
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
index 28a5ab4256..2187935fa4 100644
--- a/src/gallium/auxiliary/util/SConscript
+++ b/src/gallium/auxiliary/util/SConscript
@@ -24,6 +24,7 @@ util = env.ConvenienceLibrary(
 		'u_bitmask.c',
 		'u_blit.c',
 		'u_cache.c',
+		'u_cpu_detect.c',
 		'u_debug.c',
 		'u_debug_dump.c',
 		'u_debug_memory.c',
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.c b/src/gallium/auxiliary/util/u_cpu_detect.c
index d9f2f8fc28..ecfb96138d 100644
--- a/src/gallium/auxiliary/util/u_cpu_detect.c
+++ b/src/gallium/auxiliary/util/u_cpu_detect.c
@@ -24,23 +24,21 @@
  * 
  **************************************************************************/
 
-/*
- * Based on the work of Eric Anholt <anholt@FreeBSD.org>
+/**
+ * @file
+ * CPU feature detection.
+ *
+ * @author Dennis Smit
+ * @author Based on the work of Eric Anholt <anholt@FreeBSD.org>
  */
 
-/* FIXME: clean this entire file up */
+#include "pipe/p_config.h"
 
+#include "u_debug.h"
 #include "u_cpu_detect.h"
 
-#ifdef __linux__
-#define OS_LINUX
-#endif
-#ifdef WIN32
-#define OS_WIN32
-#endif
-
-#if defined(ARCH_POWERPC)
-#if defined(OS_DARWIN)
+#if defined(PIPE_ARCH_PPC)
+#if defined(PIPE_OS_DARWIN)
 #include <sys/sysctl.h>
 #else
 #include <signal.h>
@@ -48,137 +46,140 @@
 #endif
 #endif
 
-#if defined(OS_NETBSD) || defined(OS_OPENBSD)
+#if defined(PIPE_OS_NETBSD) || defined(PIPE_OS_OPENBSD)
 #include <sys/param.h>
 #include <sys/sysctl.h>
 #include <machine/cpu.h>
 #endif
 
-#if defined(OS_FREEBSD)
+#if defined(PIPE_OS_FREEBSD)
 #include <sys/types.h>
 #include <sys/sysctl.h>
 #endif
 
-#if defined(OS_LINUX)
+#if defined(PIPE_OS_LINUX)
 #include <signal.h>
 #endif
 
-#if defined(OS_WIN32)
-#include <windows.h>
+#ifdef PIPE_OS_UNIX
+#include <unistd.h>
 #endif
 
-#include <stdio.h>
-#include <stdlib.h>
-#include <unistd.h>
-#include <string.h>
+#if defined(PIPE_OS_WINDOWS)
+#include <windows.h>
+#endif
 
 
-static struct cpu_detect_caps __cpu_detect_caps;
-static int __cpu_detect_initialized = 0;
+struct util_cpu_caps util_cpu_caps;
 
 static int has_cpuid(void);
 static int cpuid(unsigned int ax, unsigned int *p);
 
+#if defined(PIPE_ARCH_X86)
+
 /* The sigill handlers */
-#if defined(ARCH_X86) /* x86 (linux katmai handler check thing) */
-#if defined(OS_LINUX) && defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
-static void sigill_handler_sse(int signal, struct sigcontext sc)
+#if defined(PIPE_OS_LINUX) //&& defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
+static void
+sigill_handler_sse(int signal, struct sigcontext sc)
 {
-	/* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
-	 * instructions are 3 bytes long.  We must increment the instruction
-	 * pointer manually to avoid repeated execution of the offending
-	 * instruction.
-	 *
-	 * If the SIGILL is caused by a divide-by-zero when unmasked
-	 * exceptions aren't supported, the SIMD FPU status and control
-	 * word will be restored at the end of the test, so we don't need
-	 * to worry about doing it here.  Besides, we may not be able to...
-	 */
-	sc.eip += 3;
-
-	__cpu_detect_caps.hasSSE=0;
+   /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
+    * instructions are 3 bytes long.  We must increment the instruction
+    * pointer manually to avoid repeated execution of the offending
+    * instruction.
+    *
+    * If the SIGILL is caused by a divide-by-zero when unmasked
+    * exceptions aren't supported, the SIMD FPU status and control
+    * word will be restored at the end of the test, so we don't need
+    * to worry about doing it here.  Besides, we may not be able to...
+    */
+   sc.eip += 3;
+
+   util_cpu_caps.has_sse=0;
 }
 
-static void sigfpe_handler_sse(int signal, struct sigcontext sc)
+static void
+sigfpe_handler_sse(int signal, struct sigcontext sc)
 {
-	if (sc.fpstate->magic != 0xffff) {
-		/* Our signal context has the extended FPU state, so reset the
-		 * divide-by-zero exception mask and clear the divide-by-zero
-		 * exception bit.
-		 */
-		sc.fpstate->mxcsr |= 0x00000200;
-		sc.fpstate->mxcsr &= 0xfffffffb;
-	} else {
-		/* If we ever get here, we're completely hosed.
-		*/
-	}
+   if (sc.fpstate->magic != 0xffff) {
+      /* Our signal context has the extended FPU state, so reset the
+       * divide-by-zero exception mask and clear the divide-by-zero
+       * exception bit.
+       */
+      sc.fpstate->mxcsr |= 0x00000200;
+      sc.fpstate->mxcsr &= 0xfffffffb;
+   } else {
+      /* If we ever get here, we're completely hosed.
+      */
+   }
 }
-#endif
-#endif /* OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
+#endif /* PIPE_OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
 
-#if defined(OS_WIN32)
-LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
+#if defined(PIPE_OS_WINDOWS)
+static LONG CALLBACK
+win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
 {
-	if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
-		ep->ContextRecord->Eip +=3;
-		__cpu_detect_caps.hasSSE=0;
-		return EXCEPTION_CONTINUE_EXECUTION;
-	}
-	return EXCEPTION_CONTINUE_SEARCH;
+   if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
+      ep->ContextRecord->Eip +=3;
+      util_cpu_caps.has_sse=0;
+      return EXCEPTION_CONTINUE_EXECUTION;
+   }
+   return EXCEPTION_CONTINUE_SEARCH;
 }
-#endif /* OS_WIN32 */
+#endif /* PIPE_OS_WINDOWS */
+
+#endif /* PIPE_ARCH_X86 */
 
 
-#if defined(ARCH_POWERPC) && !defined(OS_DARWIN)
+#if defined(PIPE_ARCH_PPC) && !defined(PIPE_OS_DARWIN)
 static sigjmp_buf __lv_powerpc_jmpbuf;
 static volatile sig_atomic_t __lv_powerpc_canjump = 0;
 
-static void sigill_handler (int sig);
-
-static void sigill_handler (int sig)
+static void
+sigill_handler(int sig)
 {
-	if (!__lv_powerpc_canjump) {
-		signal (sig, SIG_DFL);
-		raise (sig);
-	}
+   if (!__lv_powerpc_canjump) {
+      signal (sig, SIG_DFL);
+      raise (sig);
+   }
 
-	__lv_powerpc_canjump = 0;
-	siglongjmp(__lv_powerpc_jmpbuf, 1);
+   __lv_powerpc_canjump = 0;
+   siglongjmp(__lv_powerpc_jmpbuf, 1);
 }
 
-static void check_os_altivec_support(void)
+static void
+check_os_altivec_support(void)
 {
-#if defined(OS_DARWIN)
-	int sels[2] = {CTL_HW, HW_VECTORUNIT};
-	int has_vu = 0;
-	int len = sizeof (has_vu);
-	int err;
-
-	err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
-
-	if (err == 0) {
-		if (has_vu != 0) {
-			__cpu_detect_caps.hasAltiVec = 1;
-		}
-	}
-#else /* !OS_DARWIN */
-	/* no Darwin, do it the brute-force way */
-	/* this is borrowed from the libmpeg2 library */
-	signal(SIGILL, sigill_handler);
-	if (sigsetjmp(__lv_powerpc_jmpbuf, 1)) {
-		signal(SIGILL, SIG_DFL);
-	} else {
-		__lv_powerpc_canjump = 1;
-
-		__asm __volatile
-			("mtspr 256, %0\n\t"
-			 "vand %%v0, %%v0, %%v0"
-			 :
-			 : "r" (-1));
-
-		signal(SIGILL, SIG_DFL);
-		__cpu_detect_caps.hasAltiVec = 1;
-	}
+#if defined(PIPE_OS_DARWIN)
+   int sels[2] = {CTL_HW, HW_VECTORUNIT};
+   int has_vu = 0;
+   int len = sizeof (has_vu);
+   int err;
+
+   err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
+
+   if (err == 0) {
+      if (has_vu != 0) {
+         util_cpu_caps.has_altivec = 1;
+      }
+   }
+#else /* !PIPE_OS_DARWIN */
+   /* no Darwin, do it the brute-force way */
+   /* this is borrowed from the libmpeg2 library */
+   signal(SIGILL, sigill_handler);
+   if (sigsetjmp(__lv_powerpc_jmpbuf, 1)) {
+      signal(SIGILL, SIG_DFL);
+   } else {
+      __lv_powerpc_canjump = 1;
+
+      __asm __volatile
+         ("mtspr 256, %0\n\t"
+          "vand %%v0, %%v0, %%v0"
+          :
+          : "r" (-1));
+
+      signal(SIGILL, SIG_DFL);
+      util_cpu_caps.has_altivec = 1;
+   }
 #endif
 }
 #endif
@@ -189,318 +190,312 @@ static void check_os_altivec_support(void)
  * and RedHat patched 2.2 kernels that have broken exception handling
  * support for user space apps that do SSE.
  */
-static void check_os_katmai_support(void)
+static void
+check_os_katmai_support(void)
 {
-#if defined(ARCH_X86)
-#if defined(OS_FREEBSD)
-	int has_sse=0, ret;
-	int len = sizeof (has_sse);
-
-	ret = sysctlbyname("hw.instruction_sse", &has_sse, &len, NULL, 0);
-	if (ret || !has_sse)
-		__cpu_detect_caps.hasSSE=0;
-
-#elif defined(OS_NETBSD) || defined(OS_OPENBSD)
-	int has_sse, has_sse2, ret, mib[2];
-	int varlen;
-
-	mib[0] = CTL_MACHDEP;
-	mib[1] = CPU_SSE;
-	varlen = sizeof (has_sse);
-
-	ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
-	if (ret < 0 || !has_sse) {
-		__cpu_detect_caps.hasSSE = 0;
-	} else {
-		__cpu_detect_caps.hasSSE = 1;
-	}
-
-	mib[1] = CPU_SSE2;
-	varlen = sizeof (has_sse2);
-	ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
-	if (ret < 0 || !has_sse2) {
-		__cpu_detect_caps.hasSSE2 = 0;
-	} else {
-		__cpu_detect_caps.hasSSE2 = 1;
-	}
-	__cpu_detect_caps.hasSSE = 0; /* FIXME ?!?!? */
-
-#elif defined(OS_WIN32)
-	LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
-	if (__cpu_detect_caps.hasSSE) {
-		exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
-		__asm __volatile ("xorps %xmm0, %xmm0");
-		SetUnhandledExceptionFilter(exc_fil);
-	}
-#elif defined(OS_LINUX)
-	struct sigaction saved_sigill;
-	struct sigaction saved_sigfpe;
-
-	/* Save the original signal handlers.
-	*/
-	sigaction(SIGILL, NULL, &saved_sigill);
-	sigaction(SIGFPE, NULL, &saved_sigfpe);
-
-	signal(SIGILL, (void (*)(int))sigill_handler_sse);
-	signal(SIGFPE, (void (*)(int))sigfpe_handler_sse);
-
-	/* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
-	 * supports the extended FPU save and restore required for SSE.  If
-	 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
-	 * doesn't support Streaming SIMD Exceptions, even if the processor
-	 * does.
-	 */
-	if (__cpu_detect_caps.hasSSE) {
-		__asm __volatile ("xorps %xmm1, %xmm0");
-	}
-
-	/* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
-	 * it supports unmasked SIMD FPU exceptions.  If we unmask the
-	 * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
-	 * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
-	 * as expected, we're okay but we need to clean up after it.
-	 *
-	 * Are we being too stringent in our requirement that the OS support
-	 * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
-	 * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
-	 * doesn't even support them.  We at least know the user-space SSE
-	 * support is good in kernels that do support unmasked exceptions,
-	 * and therefore to be safe I'm going to leave this test in here.
-	 */
-	if (__cpu_detect_caps.hasSSE) {
-		//      test_os_katmai_exception_support();
-	}
-
-	/* Restore the original signal handlers.
-	*/
-	sigaction(SIGILL, &saved_sigill, NULL);
-	sigaction(SIGFPE, &saved_sigfpe, NULL);
+#if defined(PIPE_ARCH_X86)
+#if defined(PIPE_OS_FREEBSD)
+   int has_sse=0, ret;
+   int len = sizeof (has_sse);
+
+   ret = sysctlbyname("hw.instruction_sse", &has_sse, &len, NULL, 0);
+   if (ret || !has_sse)
+      util_cpu_caps.has_sse=0;
+
+#elif defined(PIPE_OS_NETBSD) || defined(PIPE_OS_OPENBSD)
+   int has_sse, has_sse2, ret, mib[2];
+   int varlen;
+
+   mib[0] = CTL_MACHDEP;
+   mib[1] = CPU_SSE;
+   varlen = sizeof (has_sse);
+
+   ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
+   if (ret < 0 || !has_sse) {
+      util_cpu_caps.has_sse = 0;
+   } else {
+      util_cpu_caps.has_sse = 1;
+   }
+
+   mib[1] = CPU_SSE2;
+   varlen = sizeof (has_sse2);
+   ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
+   if (ret < 0 || !has_sse2) {
+      util_cpu_caps.has_sse2 = 0;
+   } else {
+      util_cpu_caps.has_sse2 = 1;
+   }
+   util_cpu_caps.has_sse = 0; /* FIXME ?!?!? */
+
+#elif defined(PIPE_OS_WINDOWS)
+   LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
+   if (util_cpu_caps.has_sse) {
+      exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
+#if defined(PIPE_CC_GCC)
+      __asm __volatile ("xorps %xmm0, %xmm0");
+#elif defined(PIPE_CC_MSVC)
+      __asm {
+          xorps xmm0, xmm0        // executing SSE instruction
+      }
+#else
+#error Unsupported compiler
+#endif
+      SetUnhandledExceptionFilter(exc_fil);
+   }
+#elif defined(PIPE_OS_LINUX)
+   struct sigaction saved_sigill;
+   struct sigaction saved_sigfpe;
+
+   /* Save the original signal handlers.
+   */
+   sigaction(SIGILL, NULL, &saved_sigill);
+   sigaction(SIGFPE, NULL, &saved_sigfpe);
+
+   signal(SIGILL, (void (*)(int))sigill_handler_sse);
+   signal(SIGFPE, (void (*)(int))sigfpe_handler_sse);
+
+   /* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
+    * supports the extended FPU save and restore required for SSE.  If
+    * we execute an SSE instruction on a PIII and get a SIGILL, the OS
+    * doesn't support Streaming SIMD Exceptions, even if the processor
+    * does.
+    */
+   if (util_cpu_caps.has_sse) {
+      __asm __volatile ("xorps %xmm1, %xmm0");
+   }
+
+   /* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
+    * it supports unmasked SIMD FPU exceptions.  If we unmask the
+    * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
+    * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
+    * as expected, we're okay but we need to clean up after it.
+    *
+    * Are we being too stringent in our requirement that the OS support
+    * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
+    * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
+    * doesn't even support them.  We at least know the user-space SSE
+    * support is good in kernels that do support unmasked exceptions,
+    * and therefore to be safe I'm going to leave this test in here.
+    */
+   if (util_cpu_caps.has_sse) {
+      //      test_os_katmai_exception_support();
+   }
+
+   /* Restore the original signal handlers.
+   */
+   sigaction(SIGILL, &saved_sigill, NULL);
+   sigaction(SIGFPE, &saved_sigfpe, NULL);
 
 #else
-	/* We can't use POSIX signal handling to test the availability of
-	 * SSE, so we disable it by default.
-	 */
-	__cpu_detect_caps.hasSSE = 0;
+   /* We can't use POSIX signal handling to test the availability of
+    * SSE, so we disable it by default.
+    */
+   util_cpu_caps.has_sse = 0;
 #endif /* __linux__ */
 #endif
+
+#if defined(PIPE_ARCH_X86_64)
+   util_cpu_caps.has_sse = 1;
+#endif
 }
 
 
 static int has_cpuid(void)
 {
-#if defined(ARCH_X86)
-	int a, c;
-
-	__asm __volatile
-		("pushf\n"
-		 "popl %0\n"
-		 "movl %0, %1\n"
-		 "xorl $0x200000, %0\n"
-		 "push %0\n"
-		 "popf\n"
-		 "pushf\n"
-		 "popl %0\n"
-		 : "=a" (a), "=c" (c)
-		 :
-		 : "cc");
-
-	return a != c;
+#if defined(PIPE_ARCH_X86)
+#if defined(PIPE_OS_GCC)
+   int a, c;
+
+   __asm __volatile
+      ("pushf\n"
+       "popl %0\n"
+       "movl %0, %1\n"
+       "xorl $0x200000, %0\n"
+       "push %0\n"
+       "popf\n"
+       "pushf\n"
+       "popl %0\n"
+       : "=a" (a), "=c" (c)
+       :
+       : "cc");
+
+   return a != c;
+#else
+   /* FIXME */
+   return 1;
+#endif
+#elif defined(PIPE_ARCH_X86_64)
+   return 1;
 #else
-	return 0;
+   return 0;
 #endif
 }
 
-static int cpuid(unsigned int ax, unsigned int *p)
+static INLINE int
+cpuid(unsigned int ax, unsigned int *p)
 {
-#if defined(ARCH_X86)
-	unsigned int flags;
-
-	__asm __volatile
-		("movl %%ebx, %%esi\n\t"
-		 "cpuid\n\t"
-		 "xchgl %%ebx, %%esi"
-		 : "=a" (p[0]), "=S" (p[1]),
-		 "=c" (p[2]), "=d" (p[3])
-		 : "0" (ax));
-
-	return 0;
-#else
-	return -1;
+   int ret = -1;
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+#if defined(PIPE_CC_GCC)
+   __asm __volatile
+      ("movl %%ebx, %%esi\n\t"
+       "cpuid\n\t"
+       "xchgl %%ebx, %%esi"
+       : "=a" (p[0]), "=S" (p[1]),
+       "=c" (p[2]), "=d" (p[3])
+       : "0" (ax));
+
+   ret = 0;
+#elif defined(PIPE_CC_MSVC)
+   __cpuid(ax, p);
+
+   ret = 0;
+#endif
 #endif
+
+   return ret;
 }
 
-void cpu_detect_initialize()
+void
+util_cpu_detect(void)
 {
-	unsigned int regs[4];
-	unsigned int regs2[4];
-
-	int mib[2], ncpu;
-	int len;
-
-	memset(&__cpu_detect_caps, 0, sizeof (struct cpu_detect_caps));
-
-	/* Check for arch type */
-#if defined(ARCH_MIPS)
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_MIPS;
-#elif defined(ARCH_ALPHA)
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_ALPHA;
-#elif defined(ARCH_SPARC)
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_SPARC;
-#elif defined(ARCH_X86)
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_X86;
-#elif defined(ARCH_POWERPC)
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_POWERPC;
+   static boolean util_cpu_detect_initialized = FALSE;
+
+   if(util_cpu_detect_initialized)
+      return;
+
+   memset(&util_cpu_caps, 0, sizeof util_cpu_caps);
+
+   /* Check for arch type */
+#if defined(PIPE_ARCH_MIPS)
+   util_cpu_caps.arch = UTIL_CPU_ARCH_MIPS;
+#elif defined(PIPE_ARCH_ALPHA)
+   util_cpu_caps.arch = UTIL_CPU_ARCH_ALPHA;
+#elif defined(PIPE_ARCH_SPARC)
+   util_cpu_caps.arch = UTIL_CPU_ARCH_SPARC;
+#elif defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   util_cpu_caps.arch = UTIL_CPU_ARCH_X86;
+#elif defined(PIPE_ARCH_PPC)
+   util_cpu_caps.arch = UTIL_CPU_ARCH_POWERPC;
 #else
-	__cpu_detect_caps.type = CPU_DETECT_TYPE_OTHER;
+   util_cpu_caps.arch = UTIL_CPU_ARCH_UNKNOWN;
 #endif
 
-	/* Count the number of CPUs in system */
-#if !defined(OS_WIN32) && !defined(OS_UNKNOWN) && defined(_SC_NPROCESSORS_ONLN)
-	__cpu_detect_caps.nrcpu = sysconf(_SC_NPROCESSORS_ONLN);
-	if (__cpu_detect_caps.nrcpu == -1)
-		__cpu_detect_caps.nrcpu = 1;
-
-#elif defined(OS_NETBSD) || defined(OS_FREEBSD) || defined(OS_OPENBSD)
+   /* Count the number of CPUs in system */
+#if !defined(PIPE_OS_WINDOWS) && !defined(PIPE_OS_UNKNOWN) && defined(_SC_NPROCESSORS_ONLN)
+   util_cpu_caps.nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
+   if (util_cpu_caps.nr_cpus == -1)
+      util_cpu_caps.nr_cpus = 1;
 
-	mib[0] = CTL_HW;
-	mib[1] = HW_NCPU;
+#elif defined(PIPE_OS_NETBSD) || defined(PIPE_OS_FREEBSD) || defined(PIPE_OS_OPENBSD)
+   {
+      int mib[2], ncpu;
+      int len;
 
-	len = sizeof (ncpu);
-	sysctl(mib, 2, &ncpu, &len, NULL, 0);
-	__cpu_detect_caps.nrcpu = ncpu;
+      mib[0] = CTL_HW;
+      mib[1] = HW_NCPU;
 
+      len = sizeof (ncpu);
+      sysctl(mib, 2, &ncpu, &len, NULL, 0);
+      util_cpu_caps.nr_cpus = ncpu;
+   }
 #else
-	__cpu_detect_caps.nrcpu = 1;
+   util_cpu_caps.nr_cpus = 1;
 #endif
 
-#if defined(ARCH_X86)
-	/* No cpuid, old 486 or lower */
-	if (has_cpuid() == 0)
-		return;
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   if (has_cpuid()) {
+      unsigned int regs[4];
+      unsigned int regs2[4];
 
-	__cpu_detect_caps.cacheline = 32;
+      util_cpu_caps.cacheline = 32;
 
-	/* Get max cpuid level */
-	cpuid(0x00000000, regs);
+      /* Get max cpuid level */
+      cpuid(0x00000000, regs);
 
-	if (regs[0] >= 0x00000001) {
-		unsigned int cacheline;
+      if (regs[0] >= 0x00000001) {
+         unsigned int cacheline;
 
-		cpuid (0x00000001, regs2);
+         cpuid (0x00000001, regs2);
 
-		__cpu_detect_caps.x86cpuType = (regs2[0] >> 8) & 0xf;
-		if (__cpu_detect_caps.x86cpuType == 0xf)
-		    __cpu_detect_caps.x86cpuType = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
+         util_cpu_caps.x86_cpu_type = (regs2[0] >> 8) & 0xf;
+         if (util_cpu_caps.x86_cpu_type == 0xf)
+             util_cpu_caps.x86_cpu_type = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
 
-		/* general feature flags */
-		__cpu_detect_caps.hasTSC  = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
-		__cpu_detect_caps.hasMMX  = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
-		__cpu_detect_caps.hasSSE  = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
-		__cpu_detect_caps.hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
-		__cpu_detect_caps.hasSSE3 = (regs2[2] & (1));	       /* 0x0000001 */
-		__cpu_detect_caps.hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
-		__cpu_detect_caps.hasMMX2 = __cpu_detect_caps.hasSSE; /* SSE cpus supports mmxext too */
+         /* general feature flags */
+         util_cpu_caps.has_tsc    = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
+         util_cpu_caps.has_mmx    = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+         util_cpu_caps.has_sse    = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
+         util_cpu_caps.has_sse2   = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
+         util_cpu_caps.has_sse3   = (regs2[2] & (1));          /* 0x0000001 */
+         util_cpu_caps.has_ssse3  = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
+         util_cpu_caps.has_sse4_1 = (regs2[2] & (1 << 19)) >> 19;
+         util_cpu_caps.has_mmx2   = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
 
-		cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
-		if (cacheline > 0)
-			__cpu_detect_caps.cacheline = cacheline;
-	}
+         cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
+         if (cacheline > 0)
+            util_cpu_caps.cacheline = cacheline;
+      }
 
-	cpuid(0x80000000, regs);
+      cpuid(0x80000000, regs);
 
-	if (regs[0] >= 0x80000001) {
+      if (regs[0] >= 0x80000001) {
 
-		cpuid(0x80000001, regs2);
+         cpuid(0x80000001, regs2);
 
-		__cpu_detect_caps.hasMMX  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
-		__cpu_detect_caps.hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
-		__cpu_detect_caps.has3DNow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
-		__cpu_detect_caps.has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
-	}
+         util_cpu_caps.has_mmx  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+         util_cpu_caps.has_mmx2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
+         util_cpu_caps.has_3dnow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
+         util_cpu_caps.has_3dnow_ext = (regs2[3] & (1 << 30 )) >> 30;
+      }
 
-	if (regs[0] >= 0x80000006) {
-		cpuid(0x80000006, regs2);
-		__cpu_detect_caps.cacheline = regs2[2] & 0xFF;
-	}
+      if (regs[0] >= 0x80000006) {
+         cpuid(0x80000006, regs2);
+         util_cpu_caps.cacheline = regs2[2] & 0xFF;
+      }
 
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_FREEBSD) || defined(PIPE_OS_NETBSD) || defined(PIPE_OS_CYGWIN) || defined(PIPE_OS_OPENBSD)
+      if (util_cpu_caps.has_sse)
+         check_os_katmai_support();
 
-#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_CYGWIN) || defined(OS_OPENBSD)
-	if (__cpu_detect_caps.hasSSE)
-		check_os_katmai_support();
-
-	if (!__cpu_detect_caps.hasSSE) {
-		__cpu_detect_caps.hasSSE2 = 0;
-		__cpu_detect_caps.hasSSE3 = 0;
-		__cpu_detect_caps.hasSSSE3 = 0;
-	}
+      if (!util_cpu_caps.has_sse) {
+         util_cpu_caps.has_sse2 = 0;
+         util_cpu_caps.has_sse3 = 0;
+         util_cpu_caps.has_ssse3 = 0;
+      }
 #else
-	__cpu_detect_caps.hasSSE = 0;
-	__cpu_detect_caps.hasSSE2 = 0;
-	__cpu_detect_caps.hasSSE3 = 0;
-	__cpu_detect_caps.hasSSSE3 = 0;
+      util_cpu_caps.has_sse = 0;
+      util_cpu_caps.has_sse2 = 0;
+      util_cpu_caps.has_sse3 = 0;
+      util_cpu_caps.has_ssse3 = 0;
+#endif
+   }
+#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
+
+#if defined(PIPE_ARCH_PPC)
+   check_os_altivec_support();
+#endif /* PIPE_ARCH_PPC */
+
+#ifdef DEBUG
+   debug_printf("util_cpu_caps.arch = %i\n", util_cpu_caps.arch);
+   debug_printf("util_cpu_caps.nr_cpus = %u\n", util_cpu_caps.nr_cpus);
+
+   debug_printf("util_cpu_caps.x86_cpu_type = %u\n", util_cpu_caps.x86_cpu_type);
+   debug_printf("util_cpu_caps.cacheline = %u\n", util_cpu_caps.cacheline);
+
+   debug_printf("util_cpu_caps.has_tsc = %u\n", util_cpu_caps.has_tsc);
+   debug_printf("util_cpu_caps.has_mmx = %u\n", util_cpu_caps.has_mmx);
+   debug_printf("util_cpu_caps.has_mmx2 = %u\n", util_cpu_caps.has_mmx2);
+   debug_printf("util_cpu_caps.has_sse = %u\n", util_cpu_caps.has_sse);
+   debug_printf("util_cpu_caps.has_sse2 = %u\n", util_cpu_caps.has_sse2);
+   debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps.has_sse3);
+   debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps.has_ssse3);
+   debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps.has_sse4_1);
+   debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps.has_3dnow);
+   debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps.has_3dnow_ext);
+   debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps.has_altivec);
 #endif
-#endif /* ARCH_X86 */
-
-#if defined(ARCH_POWERPC)
-	check_os_altivec_support();
-#endif /* ARCH_POWERPC */
-
-	__cpu_detect_initialized = 1;
-}
-
-struct cpu_detect_caps *cpu_detect_get_caps()
-{
-	return &__cpu_detect_caps;
-}
-
-/* The getters and setters for feature flags */
-int cpu_detect_get_tsc()
-{
-	return __cpu_detect_caps.hasTSC;
-}
-
-int cpu_detect_get_mmx()
-{
-	return __cpu_detect_caps.hasMMX;
-}
-
-int cpu_detect_get_mmx2()
-{
-	return __cpu_detect_caps.hasMMX2;
-}
-
-int cpu_detect_get_sse()
-{
-	return __cpu_detect_caps.hasSSE;
-}
-
-int cpu_detect_get_sse2()
-{
-	return __cpu_detect_caps.hasSSE2;
-}
-
-int cpu_detect_get_sse3()
-{
-	return __cpu_detect_caps.hasSSE3;
-}
-
-int cpu_detect_get_ssse3()
-{
-	return __cpu_detect_caps.hasSSSE3;
-}
-
-int cpu_detect_get_3dnow()
-{
-	return __cpu_detect_caps.has3DNow;
-}
-
-int cpu_detect_get_3dnow2()
-{
-	return __cpu_detect_caps.has3DNowExt;
-}
 
-int cpu_detect_get_altivec()
-{
-	return __cpu_detect_caps.hasAltiVec;
+   util_cpu_detect_initialized = TRUE;
 }
-
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.h b/src/gallium/auxiliary/util/u_cpu_detect.h
index 1612d49286..7ea0121c07 100644
--- a/src/gallium/auxiliary/util/u_cpu_detect.h
+++ b/src/gallium/auxiliary/util/u_cpu_detect.h
@@ -24,55 +24,53 @@
  *
  ***************************************************************************/
 
-/*
- * Based on the work of Eric Anholt <anholt@FreeBSD.org>
+/**
+ * @file
+ * CPU feature detection.
+ *
+ * @author Dennis Smit
+ * @author Based on the work of Eric Anholt <anholt@FreeBSD.org>
  */
 
-#ifndef _CPU_DETECT_H
-#define _CPU_DETECT_H
+#ifndef _UTIL_CPU_DETECT_H
+#define _UTIL_CPU_DETECT_H
+
+#include "pipe/p_compiler.h"
 
-typedef enum {
-	CPU_DETECT_TYPE_MIPS,
-	CPU_DETECT_TYPE_ALPHA,
-	CPU_DETECT_TYPE_SPARC,
-	CPU_DETECT_TYPE_X86,
-	CPU_DETECT_TYPE_POWERPC,
-	CPU_DETECT_TYPE_OTHER
-} cpu_detect_type;
+enum util_cpu_arch {
+   UTIL_CPU_ARCH_UNKNOWN = 0,
+   UTIL_CPU_ARCH_MIPS,
+   UTIL_CPU_ARCH_ALPHA,
+   UTIL_CPU_ARCH_SPARC,
+   UTIL_CPU_ARCH_X86,
+   UTIL_CPU_ARCH_POWERPC
+};
 
-struct cpu_detect_caps {
-	cpu_detect_type	type;
-	int		nrcpu;
+struct util_cpu_caps {
+   enum util_cpu_arch arch;
+   unsigned nr_cpus;
 
-	/* Feature flags */
-	int		x86cpuType;
-	int		cacheline;
+   /* Feature flags */
+   int x86_cpu_type;
+   unsigned cacheline;
 
-	int		hasTSC;
-	int		hasMMX;
-	int		hasMMX2;
-	int		hasSSE;
-	int		hasSSE2;
-	int		hasSSE3;
-	int		hasSSSE3;
-	int		has3DNow;
-	int		has3DNowExt;
-	int		hasAltiVec;
+   unsigned has_tsc:1;
+   unsigned has_mmx:1;
+   unsigned has_mmx2:1;
+   unsigned has_sse:1;
+   unsigned has_sse2:1;
+   unsigned has_sse3:1;
+   unsigned has_ssse3:1;
+   unsigned has_sse4_1:1;
+   unsigned has_3dnow:1;
+   unsigned has_3dnow_ext:1;
+   unsigned has_altivec:1;
 };
 
-/* prototypes */
-void cpu_detect_initialize(void);
-struct cpu_detect_caps *cpu_detect_get_caps(void);
+extern struct util_cpu_caps
+util_cpu_caps;
+
+void util_cpu_detect(void);
 
-int cpu_detect_get_tsc(void);
-int cpu_detect_get_mmx(void);
-int cpu_detect_get_mmx2(void);
-int cpu_detect_get_sse(void);
-int cpu_detect_get_sse2(void);
-int cpu_detect_get_sse3(void);
-int cpu_detect_get_ssse3(void);
-int cpu_detect_get_3dnow(void);
-int cpu_detect_get_3dnow2(void);
-int cpu_detect_get_altivec(void);
 
-#endif /* _CPU_DETECT_H */
+#endif /* _UTIL_CPU_DETECT_H */
diff --git a/src/gallium/auxiliary/util/u_debug.h b/src/gallium/auxiliary/util/u_debug.h
index 1380d98d7e..b82e7cb4d4 100644
--- a/src/gallium/auxiliary/util/u_debug.h
+++ b/src/gallium/auxiliary/util/u_debug.h
@@ -65,6 +65,11 @@ extern "C" {
 #define __FUNCTION__ "???"
 #endif
 
+#if defined(__GNUC__)
+#define _util_printf_format(fmt, list) __attribute__ ((format (printf, fmt, list)))
+#else
+#define _util_printf_format(fmt, list)
+#endif
 
 void _debug_vprintf(const char *format, va_list ap);
    
@@ -82,14 +87,17 @@ _debug_printf(const char *format, ...)
 /**
  * Print debug messages.
  *
- * The actual channel used to output debug message is platform specific. To 
- * avoid misformating or truncation, follow these rules of thumb:   
+ * The actual channel used to output debug message is platform specific. To
+ * avoid misformating or truncation, follow these rules of thumb:
  * - output whole lines
- * - avoid outputing large strings (512 bytes is the current maximum length 
+ * - avoid outputing large strings (512 bytes is the current maximum length
  * that is guaranteed to be printed in all platforms)
  */
 #if !defined(PIPE_OS_HAIKU)
 static INLINE void
+debug_printf(const char *format, ...) _util_printf_format(1,2);
+
+static INLINE void
 debug_printf(const char *format, ...)
 {
 #ifdef DEBUG
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
index f06c0e463d..4e3d35f40e 100644
--- a/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -1515,6 +1515,17 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    uint zslice = 0;
    uint offset;
 
+   /* The texture object should have room for the levels which we're
+    * about to generate.
+    */
+   assert(lastLevel <= pt->last_level);
+
+   /* If this fails, why are we here? */
+   assert(lastLevel > baseLevel);
+
+   assert(filter == PIPE_TEX_FILTER_LINEAR ||
+          filter == PIPE_TEX_FILTER_NEAREST);
+
    /* check if we can render in the texture's format */
    if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D,
                                     PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) {
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index b428dc544c..75b075f160 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -283,6 +283,14 @@ util_fast_pow(float x, float y)
    return util_fast_exp2(util_fast_log2(x) * y);
 }
 
+/* Note that this counts zero as a power of two.
+ */
+static INLINE boolean
+util_is_power_of_two( unsigned v )
+{
+   return (v & (v-1)) == 0;
+}
+
 
 /**
  * Floor(x), returned as int.
@@ -463,6 +471,26 @@ util_logbase2(unsigned n)
 
 
 /**
+ * Returns the smallest power of two >= x
+ */
+static INLINE unsigned
+util_next_power_of_two(unsigned x)
+{
+   unsigned i;
+
+   if (x == 0)
+      return 1;
+
+   --x;
+
+   for (i = 1; i < sizeof(unsigned) * 8; i <<= 1)
+      x |= x >> i;
+
+   return x + 1;
+}
+
+
+/**
  * Clamp X to [MIN, MAX].
  * This is a macro to allow float, int, uint, etc. types.
  */
diff --git a/src/gallium/auxiliary/util/u_network.c b/src/gallium/auxiliary/util/u_network.c
index bc4b758406..6269c72e12 100644
--- a/src/gallium/auxiliary/util/u_network.c
+++ b/src/gallium/auxiliary/util/u_network.c
@@ -6,7 +6,7 @@
 #if defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 #  include <winsock2.h>
 #  include <windows.h>
-#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
+#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) || defined(PIPE_OS_BSD)
 #  include <sys/socket.h>
 #  include <netinet/in.h>
 #  include <unistd.h>
@@ -54,7 +54,7 @@ u_socket_close(int s)
    if (s < 0)
       return;
 
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) || defined(PIPE_OS_BSD)
    shutdown(s, SHUT_RDWR);
    close(s);
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
@@ -169,7 +169,7 @@ u_socket_listen_on_port(uint16_t portnum)
 void
 u_socket_block(int s, boolean block)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) || defined(PIPE_OS_BSD)
    int old = fcntl(s, F_GETFL, 0);
    if (old == -1)
       return;
diff --git a/src/gallium/auxiliary/util/u_network.h b/src/gallium/auxiliary/util/u_network.h
index 8c778f492c..0aa898b967 100644
--- a/src/gallium/auxiliary/util/u_network.h
+++ b/src/gallium/auxiliary/util/u_network.h
@@ -6,7 +6,7 @@
 
 #if defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 #  define PIPE_HAVE_SOCKETS
-#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
+#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU) || defined(PIPE_OS_BSD)
 #  define PIPE_HAVE_SOCKETS
 #endif
 
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
index 0d6489c26e..8a22f584be 100644
--- a/src/gallium/auxiliary/util/u_tile.c
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -1452,7 +1452,7 @@ pipe_put_tile_z(struct pipe_transfer *pt,
    case PIPE_FORMAT_S8Z24_UNORM:
       {
          uint *pDest = (uint *) (map + y * pt->stride + x*4);
-         assert(pt->usage == PIPE_TRANSFER_READ_WRITE);
+         assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);
          for (i = 0; i < h; i++) {
             for (j = 0; j < w; j++) {
                /* convert 32-bit Z to 24-bit Z, preserve stencil */
@@ -1479,7 +1479,7 @@ pipe_put_tile_z(struct pipe_transfer *pt,
    case PIPE_FORMAT_Z24S8_UNORM:
       {
          uint *pDest = (uint *) (map + y * pt->stride + x*4);
-         assert(pt->usage == PIPE_TRANSFER_READ_WRITE);
+         assert((pt->usage & PIPE_TRANSFER_READ_WRITE) == PIPE_TRANSFER_READ_WRITE);
          for (i = 0; i < h; i++) {
             for (j = 0; j < w; j++) {
                /* convert 32-bit Z to 24-bit Z, preserve stencil */
diff --git a/src/gallium/auxiliary/vl/Makefile b/src/gallium/auxiliary/vl/Makefile
new file mode 100644
index 0000000000..4314c1e8d6
--- /dev/null
+++ b/src/gallium/auxiliary/vl/Makefile
@@ -0,0 +1,13 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = vl
+
+C_SOURCES = \
+	vl_bitstream_parser.c \
+	vl_mpeg12_mc_renderer.c \
+	vl_compositor.c \
+        vl_csc.c \
+	vl_shader_build.c
+
+include ../../Makefile.template
diff --git a/src/gallium/auxiliary/vl/SConscript b/src/gallium/auxiliary/vl/SConscript
new file mode 100644
index 0000000000..aed69f5efe
--- /dev/null
+++ b/src/gallium/auxiliary/vl/SConscript
@@ -0,0 +1,13 @@
+Import('*')
+
+vl = env.ConvenienceLibrary(
+	target = 'vl',
+	source = [
+		'vl_bitstream_parser.c',
+		'vl_mpeg12_mc_renderer.c',
+		'vl_compositor.c',
+                'vl_csc.c',
+		'vl_shader_build.c',
+	])
+
+auxiliaries.insert(0, vl)
diff --git a/src/gallium/auxiliary/vl/vl_bitstream_parser.c b/src/gallium/auxiliary/vl/vl_bitstream_parser.c
new file mode 100644
index 0000000000..3193ea5f41
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_bitstream_parser.c
@@ -0,0 +1,167 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "vl_bitstream_parser.h"
+#include <assert.h>
+#include <limits.h>
+#include <util/u_memory.h>
+
+static unsigned
+grab_bits(unsigned cursor, unsigned how_many_bits, unsigned bitstream_elt)
+{
+   unsigned excess_bits = sizeof(unsigned) * CHAR_BIT - how_many_bits - cursor;
+	
+   assert(cursor < sizeof(unsigned) * CHAR_BIT);
+   assert(how_many_bits > 0 && how_many_bits <= sizeof(unsigned) * CHAR_BIT);
+   assert(cursor + how_many_bits <= sizeof(unsigned) * CHAR_BIT);
+
+   return (bitstream_elt << excess_bits) >> (excess_bits + cursor);
+}
+
+static unsigned
+show_bits(unsigned cursor, unsigned how_many_bits, const unsigned *bitstream)
+{	
+   unsigned cur_int = cursor / (sizeof(unsigned) * CHAR_BIT);
+   unsigned cur_bit = cursor % (sizeof(unsigned) * CHAR_BIT);
+	
+   assert(bitstream);
+	
+   if (cur_bit + how_many_bits > sizeof(unsigned) * CHAR_BIT) {
+      unsigned lower = grab_bits(cur_bit, sizeof(unsigned) * CHAR_BIT - cur_bit,
+                                 bitstream[cur_int]);
+      unsigned upper = grab_bits(0, cur_bit + how_many_bits - sizeof(unsigned) * CHAR_BIT,
+                                 bitstream[cur_int + 1]);
+      return lower | upper << (sizeof(unsigned) * CHAR_BIT - cur_bit);
+   }
+   else
+      return grab_bits(cur_bit, how_many_bits, bitstream[cur_int]);
+}
+
+bool vl_bitstream_parser_init(struct vl_bitstream_parser *parser,
+                              unsigned num_bitstreams,
+                              const void **bitstreams,
+                              const unsigned *sizes)
+{
+   assert(parser);
+   assert(num_bitstreams);
+   assert(bitstreams);
+   assert(sizes);
+
+   parser->num_bitstreams = num_bitstreams;
+   parser->bitstreams = (const unsigned**)bitstreams;
+   parser->sizes = sizes;
+   parser->cur_bitstream = 0;
+   parser->cursor = 0;
+
+   return true;
+}
+
+void vl_bitstream_parser_cleanup(struct vl_bitstream_parser *parser)
+{
+   assert(parser);
+}
+
+unsigned
+vl_bitstream_parser_get_bits(struct vl_bitstream_parser *parser,
+                             unsigned how_many_bits)
+{
+   unsigned bits;
+
+   assert(parser);
+
+   bits = vl_bitstream_parser_show_bits(parser, how_many_bits);
+
+   vl_bitstream_parser_forward(parser, how_many_bits);
+
+   return bits;
+}
+
+unsigned
+vl_bitstream_parser_show_bits(struct vl_bitstream_parser *parser,
+                              unsigned how_many_bits)
+{	
+   unsigned bits = 0;
+   unsigned shift = 0;
+   unsigned cursor;
+   unsigned cur_bitstream;
+
+   assert(parser);
+
+   cursor = parser->cursor;
+   cur_bitstream = parser->cur_bitstream;
+
+   while (1) {
+      unsigned bits_left = parser->sizes[cur_bitstream] * CHAR_BIT - cursor;
+      unsigned bits_to_show = how_many_bits > bits_left ? bits_left : how_many_bits;
+
+      bits |= show_bits(cursor, bits_to_show,
+                        parser->bitstreams[cur_bitstream]) << shift;
+		
+      if (how_many_bits > bits_to_show) {
+         how_many_bits -= bits_to_show;
+         cursor = 0;
+         ++cur_bitstream;
+         shift += bits_to_show;
+      }
+      else
+         break;
+   }
+
+   return bits;
+}
+
+void vl_bitstream_parser_forward(struct vl_bitstream_parser *parser,
+                                 unsigned how_many_bits)
+{
+   assert(parser);
+   assert(how_many_bits);
+
+   parser->cursor += how_many_bits;
+
+   while (parser->cursor > parser->sizes[parser->cur_bitstream] * CHAR_BIT) {
+      parser->cursor -= parser->sizes[parser->cur_bitstream++] * CHAR_BIT;
+      assert(parser->cur_bitstream < parser->num_bitstreams);
+   }
+}
+
+void vl_bitstream_parser_rewind(struct vl_bitstream_parser *parser,
+                                unsigned how_many_bits)
+{
+   signed c;
+	
+   assert(parser);
+   assert(how_many_bits);
+	
+   c = parser->cursor - how_many_bits;
+
+   while (c < 0) {
+      c += parser->sizes[parser->cur_bitstream--] * CHAR_BIT;
+      assert(parser->cur_bitstream < parser->num_bitstreams);
+   }
+
+   parser->cursor = (unsigned)c;
+}
diff --git a/src/gallium/state_trackers/wgl/shared/stw_public.h b/src/gallium/auxiliary/vl/vl_bitstream_parser.h
index 7fe9cfb356..30ec743fa7 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_public.h
+++ b/src/gallium/auxiliary/vl/vl_bitstream_parser.h
@@ -1,8 +1,8 @@
 /**************************************************************************
- *
- * Copyright 2009 VMware, Inc.
+ * 
+ * Copyright 2009 Younes Manton.
  * 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
@@ -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.
@@ -22,52 +22,42 @@
  * 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 STW_PUBLIC_H
-#define STW_PUBLIC_H
+#ifndef vl_bitstream_parser_h
+#define vl_bitstream_parser_h
 
-#include <windows.h>
+#include "pipe/p_compiler.h"
 
-BOOL stw_copy_context( UINT_PTR hglrcSrc,
-                       UINT_PTR hglrcDst,
-                       UINT mask );
+struct vl_bitstream_parser
+{
+   unsigned num_bitstreams;
+   const unsigned **bitstreams;
+   const unsigned *sizes;
+   unsigned cur_bitstream;
+   unsigned cursor;
+};
 
-UINT_PTR stw_create_layer_context( HDC hdc, 
-                                   int iLayerPlane );
+bool vl_bitstream_parser_init(struct vl_bitstream_parser *parser,
+                              unsigned num_bitstreams,
+                              const void **bitstreams,
+                              const unsigned *sizes);
 
-BOOL stw_share_lists( UINT_PTR hglrc1, UINT_PTR hglrc2 );
+void vl_bitstream_parser_cleanup(struct vl_bitstream_parser *parser);
 
-BOOL stw_delete_context( UINT_PTR hglrc );
+unsigned
+vl_bitstream_parser_get_bits(struct vl_bitstream_parser *parser,
+                             unsigned how_many_bits);
 
-BOOL
-stw_release_context( UINT_PTR dhglrc );
+unsigned
+vl_bitstream_parser_show_bits(struct vl_bitstream_parser *parser,
+                              unsigned how_many_bits);
 
-UINT_PTR stw_get_current_context( void );
+void vl_bitstream_parser_forward(struct vl_bitstream_parser *parser,
+                                 unsigned how_many_bits);
 
-HDC stw_get_current_dc( void );
+void vl_bitstream_parser_rewind(struct vl_bitstream_parser *parser,
+                                unsigned how_many_bits);
 
-BOOL stw_make_current( HDC hdc, UINT_PTR hglrc );
-
-BOOL stw_swap_buffers( HDC hdc );
-
-BOOL
-stw_swap_layer_buffers( HDC hdc, UINT fuPlanes );
-
-PROC stw_get_proc_address( LPCSTR lpszProc );
-
-int stw_pixelformat_describe( HDC hdc,
-                              int iPixelFormat,
-                              UINT nBytes,
-                              LPPIXELFORMATDESCRIPTOR ppfd );
-
-int stw_pixelformat_get( HDC hdc );
-
-BOOL stw_pixelformat_set( HDC hdc,
-                          int iPixelFormat );
-
-int stw_pixelformat_choose( HDC hdc,
-                            CONST PIXELFORMATDESCRIPTOR *ppfd );
-
-#endif
+#endif /* vl_bitstream_parser_h */
diff --git a/src/gallium/auxiliary/vl/vl_compositor.c b/src/gallium/auxiliary/vl/vl_compositor.c
new file mode 100644
index 0000000000..b36dbeb208
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_compositor.c
@@ -0,0 +1,532 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "vl_compositor.h"
+#include <assert.h>
+#include <pipe/p_context.h>
+#include <pipe/p_inlines.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+#include <util/u_memory.h>
+#include "vl_csc.h"
+#include "vl_shader_build.h"
+
+struct vertex2f
+{
+   float x, y;
+};
+
+struct vertex4f
+{
+   float x, y, z, w;
+};
+
+struct vertex_shader_consts
+{
+   struct vertex4f dst_scale;
+   struct vertex4f dst_trans;
+   struct vertex4f src_scale;
+   struct vertex4f src_trans;
+};
+
+struct fragment_shader_consts
+{
+   float matrix[16];
+};
+
+/*
+ * Represents 2 triangles in a strip in normalized coords.
+ * Used to render the surface onto the frame buffer.
+ */
+static const struct vertex2f surface_verts[4] =
+{
+   {0.0f, 0.0f},
+   {0.0f, 1.0f},
+   {1.0f, 0.0f},
+   {1.0f, 1.0f}
+};
+
+/*
+ * Represents texcoords for the above. We can use the position values directly.
+ * TODO: Duplicate these in the shader, no need to create a buffer.
+ */
+static const struct vertex2f *surface_texcoords = surface_verts;
+
+static void
+create_vert_shader(struct vl_compositor *c)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(c);
+
+   tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version*)&tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header*)&tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0             ; Vertex pos
+    * decl i1             ; Vertex texcoords
+    */
+   for (i = 0; i < 2; i++) {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl c0             ; Scaling vector to scale vertex pos rect to destination size
+    * decl c1             ; Translation vector to move vertex pos rect into position
+    * decl c2             ; Scaling vector to scale texcoord rect to source size
+    * decl c3             ; Translation vector to move texcoord rect into position
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 3);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl o0             ; Vertex pos
+    * decl o1             ; Vertex texcoords
+    */
+   for (i = 0; i < 2; i++) {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * mad o0, i0, c0, c1  ; Scale and translate unit output rect to destination size and pos
+    * mad o1, i1, c2, c3  ; Scale and translate unit texcoord rect to source size and pos
+    */
+   for (i = 0; i < 2; ++i) {
+      inst = vl_inst4(TGSI_OPCODE_MAD, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i, TGSI_FILE_CONSTANT, i * 2, TGSI_FILE_CONSTANT, i * 2 + 1);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   c->vertex_shader = c->pipe->create_vs_state(c->pipe, &vs);
+   FREE(tokens);
+}
+
+static void
+create_frag_shader(struct vl_compositor *c)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(c);
+
+   tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version*)&tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header*)&tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /* decl i0             ; Texcoords for s0 */
+   decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, 1, 0, 0, TGSI_INTERPOLATE_LINEAR);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl c0-c3          ; CSC matrix c0-c3
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 3);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0             ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0 */
+   decl = vl_decl_temps(0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl s0             ; Sampler for tex containing picture to display */
+   decl = vl_decl_samplers(0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* tex2d t0, i0, s0    ; Read src pixel */
+   inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_SAMPLER, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * dp4 o0.x, t0, c0    ; Multiply pixel by the color conversion matrix
+    * dp4 o0.y, t0, c1
+    * dp4 o0.z, t0, c2
+    * dp4 o0.w, t0, c3
+    */
+   for (i = 0; i < 4; ++i) {
+      inst = vl_inst3(TGSI_OPCODE_DP4, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i);
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+	
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   c->fragment_shader = c->pipe->create_fs_state(c->pipe, &fs);
+   FREE(tokens);
+}
+
+static bool
+init_pipe_state(struct vl_compositor *c)
+{
+   struct pipe_sampler_state sampler;
+
+   assert(c);
+
+   c->fb_state.nr_cbufs = 1;
+   c->fb_state.zsbuf = NULL;
+
+   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_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
+   sampler.compare_func = PIPE_FUNC_ALWAYS;
+   sampler.normalized_coords = 1;
+   /*sampler.prefilter = ;*/
+   /*sampler.lod_bias = ;*/
+   /*sampler.min_lod = ;*/
+   /*sampler.max_lod = ;*/
+   /*sampler.border_color[i] = ;*/
+   /*sampler.max_anisotropy = ;*/
+   c->sampler = c->pipe->create_sampler_state(c->pipe, &sampler);
+	
+   return true;
+}
+
+static void cleanup_pipe_state(struct vl_compositor *c)
+{
+   assert(c);
+	
+   c->pipe->delete_sampler_state(c->pipe, c->sampler);
+}
+
+static bool
+init_shaders(struct vl_compositor *c)
+{
+   assert(c);
+
+   create_vert_shader(c);
+   create_frag_shader(c);
+
+   return true;
+}
+
+static void cleanup_shaders(struct vl_compositor *c)
+{
+   assert(c);
+	
+   c->pipe->delete_vs_state(c->pipe, c->vertex_shader);
+   c->pipe->delete_fs_state(c->pipe, c->fragment_shader);
+}
+
+static bool
+init_buffers(struct vl_compositor *c)
+{
+   struct fragment_shader_consts fsc;
+
+   assert(c);
+	
+   /*
+    * Create our vertex buffer and vertex buffer element
+    * VB contains 4 vertices that render a quad covering the entire window
+    * to display a rendered surface
+    * Quad is rendered as a tri strip
+    */
+   c->vertex_bufs[0].stride = sizeof(struct vertex2f);
+   c->vertex_bufs[0].max_index = 3;
+   c->vertex_bufs[0].buffer_offset = 0;
+   c->vertex_bufs[0].buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_VERTEX,
+      sizeof(struct vertex2f) * 4
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(c->pipe->screen, c->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      surface_verts,
+      sizeof(struct vertex2f) * 4
+   );
+
+   pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[0].buffer);
+
+   c->vertex_elems[0].src_offset = 0;
+   c->vertex_elems[0].vertex_buffer_index = 0;
+   c->vertex_elems[0].nr_components = 2;
+   c->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /*
+    * Create our texcoord buffer and texcoord buffer element
+    * Texcoord buffer contains the TCs for mapping the rendered surface to the 4 vertices
+    */
+   c->vertex_bufs[1].stride = sizeof(struct vertex2f);
+   c->vertex_bufs[1].max_index = 3;
+   c->vertex_bufs[1].buffer_offset = 0;
+   c->vertex_bufs[1].buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_VERTEX,
+      sizeof(struct vertex2f) * 4
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(c->pipe->screen, c->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      surface_texcoords,
+      sizeof(struct vertex2f) * 4
+   );
+
+   pipe_buffer_unmap(c->pipe->screen, c->vertex_bufs[1].buffer);
+
+   c->vertex_elems[1].src_offset = 0;
+   c->vertex_elems[1].vertex_buffer_index = 1;
+   c->vertex_elems[1].nr_components = 2;
+   c->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /*
+    * Create our vertex shader's constant buffer
+    * Const buffer contains scaling and translation vectors
+    */
+   c->vs_const_buf.buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex_shader_consts)
+   );
+
+   /*
+    * Create our fragment shader's constant buffer
+    * Const buffer contains the color conversion matrix and bias vectors
+    */
+   c->fs_const_buf.buffer = pipe_buffer_create
+   (
+      c->pipe->screen,
+      1,
+      PIPE_BUFFER_USAGE_CONSTANT,
+      sizeof(struct fragment_shader_consts)
+   );
+
+   vl_csc_get_matrix(VL_CSC_COLOR_STANDARD_IDENTITY, NULL, true, fsc.matrix);
+
+   vl_compositor_set_csc_matrix(c, fsc.matrix);
+
+   return true;
+}
+
+static void
+cleanup_buffers(struct vl_compositor *c)
+{
+   unsigned i;
+
+   assert(c);
+	
+   for (i = 0; i < 2; ++i)
+      pipe_buffer_reference(&c->vertex_bufs[i].buffer, NULL);
+
+   pipe_buffer_reference(&c->vs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&c->fs_const_buf.buffer, NULL);
+}
+
+bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe)
+{
+   assert(compositor);
+
+   memset(compositor, 0, sizeof(struct vl_compositor));
+
+   compositor->pipe = pipe;
+
+   if (!init_pipe_state(compositor))
+      return false;
+   if (!init_shaders(compositor)) {
+      cleanup_pipe_state(compositor);
+      return false;
+   }
+   if (!init_buffers(compositor)) {
+      cleanup_shaders(compositor);
+      cleanup_pipe_state(compositor);
+      return false;
+   }
+
+   return true;
+}
+
+void vl_compositor_cleanup(struct vl_compositor *compositor)
+{
+   assert(compositor);
+	
+   cleanup_buffers(compositor);
+   cleanup_shaders(compositor);
+   cleanup_pipe_state(compositor);
+}
+
+void vl_compositor_render(struct vl_compositor          *compositor,
+                          /*struct pipe_texture         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_texture           *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_texture           *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_texture           *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_texture           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas*/
+                          struct pipe_fence_handle      **fence)
+{
+   struct vertex_shader_consts *vs_consts;
+
+   assert(compositor);
+   assert(src_surface);
+   assert(src_area);
+   assert(dst_surface);
+   assert(dst_area);
+   assert(picture_type == PIPE_MPEG12_PICTURE_TYPE_FRAME);
+
+   compositor->fb_state.width = dst_surface->width[0];
+   compositor->fb_state.height = dst_surface->height[0];
+   compositor->fb_state.cbufs[0] = compositor->pipe->screen->get_tex_surface
+   (
+      compositor->pipe->screen,
+      dst_surface,
+      0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
+   );
+
+   compositor->viewport.scale[0] = compositor->fb_state.width;
+   compositor->viewport.scale[1] = compositor->fb_state.height;
+   compositor->viewport.scale[2] = 1;
+   compositor->viewport.scale[3] = 1;
+   compositor->viewport.translate[0] = 0;
+   compositor->viewport.translate[1] = 0;
+   compositor->viewport.translate[2] = 0;
+   compositor->viewport.translate[3] = 0;
+
+   compositor->pipe->set_framebuffer_state(compositor->pipe, &compositor->fb_state);
+   compositor->pipe->set_viewport_state(compositor->pipe, &compositor->viewport);
+   compositor->pipe->bind_sampler_states(compositor->pipe, 1, &compositor->sampler);
+   compositor->pipe->set_sampler_textures(compositor->pipe, 1, &src_surface);
+   compositor->pipe->bind_vs_state(compositor->pipe, compositor->vertex_shader);
+   compositor->pipe->bind_fs_state(compositor->pipe, compositor->fragment_shader);
+   compositor->pipe->set_vertex_buffers(compositor->pipe, 2, compositor->vertex_bufs);
+   compositor->pipe->set_vertex_elements(compositor->pipe, 2, compositor->vertex_elems);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_VERTEX, 0, &compositor->vs_const_buf);
+   compositor->pipe->set_constant_buffer(compositor->pipe, PIPE_SHADER_FRAGMENT, 0, &compositor->fs_const_buf);
+
+   vs_consts = pipe_buffer_map
+   (
+      compositor->pipe->screen,
+      compositor->vs_const_buf.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   vs_consts->dst_scale.x = dst_area->w / (float)compositor->fb_state.cbufs[0]->width;
+   vs_consts->dst_scale.y = dst_area->h / (float)compositor->fb_state.cbufs[0]->height;
+   vs_consts->dst_scale.z = 1;
+   vs_consts->dst_scale.w = 1;
+   vs_consts->dst_trans.x = dst_area->x / (float)compositor->fb_state.cbufs[0]->width;
+   vs_consts->dst_trans.y = dst_area->y / (float)compositor->fb_state.cbufs[0]->height;
+   vs_consts->dst_trans.z = 0;
+   vs_consts->dst_trans.w = 0;
+
+   vs_consts->src_scale.x = src_area->w / (float)src_surface->width[0];
+   vs_consts->src_scale.y = src_area->h / (float)src_surface->height[0];
+   vs_consts->src_scale.z = 1;
+   vs_consts->src_scale.w = 1;
+   vs_consts->src_trans.x = src_area->x / (float)src_surface->width[0];
+   vs_consts->src_trans.y = src_area->y / (float)src_surface->height[0];
+   vs_consts->src_trans.z = 0;
+   vs_consts->src_trans.w = 0;
+
+   pipe_buffer_unmap(compositor->pipe->screen, compositor->vs_const_buf.buffer);
+
+   compositor->pipe->draw_arrays(compositor->pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
+   compositor->pipe->flush(compositor->pipe, PIPE_FLUSH_RENDER_CACHE, fence);
+
+   pipe_surface_reference(&compositor->fb_state.cbufs[0], NULL);
+}
+
+void vl_compositor_set_csc_matrix(struct vl_compositor *compositor, const float *mat)
+{
+   assert(compositor);
+
+   memcpy
+   (
+      pipe_buffer_map(compositor->pipe->screen, compositor->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      mat,
+      sizeof(struct fragment_shader_consts)
+   );
+
+   pipe_buffer_unmap(compositor->pipe->screen, compositor->fs_const_buf.buffer);
+}
diff --git a/src/gallium/auxiliary/vl/vl_compositor.h b/src/gallium/auxiliary/vl/vl_compositor.h
new file mode 100644
index 0000000000..17e2afd353
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_compositor.h
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 vl_compositor_h
+#define vl_compositor_h
+
+#include <pipe/p_compiler.h>
+#include <pipe/p_state.h>
+#include <pipe/p_video_state.h>
+
+struct pipe_context;
+struct pipe_texture;
+
+struct vl_compositor
+{
+   struct pipe_context *pipe;
+
+   struct pipe_framebuffer_state fb_state;
+   void *sampler;
+   void *vertex_shader;
+   void *fragment_shader;
+   struct pipe_viewport_state viewport;
+   struct pipe_vertex_buffer vertex_bufs[2];
+   struct pipe_vertex_element vertex_elems[2];
+   struct pipe_constant_buffer vs_const_buf, fs_const_buf;
+};
+
+bool vl_compositor_init(struct vl_compositor *compositor, struct pipe_context *pipe);
+
+void vl_compositor_cleanup(struct vl_compositor *compositor);
+
+void vl_compositor_render(struct vl_compositor          *compositor,
+                          /*struct pipe_texture         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_texture           *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_texture           *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_texture           *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_texture           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas,*/
+                          struct pipe_fence_handle      **fence);
+
+void vl_compositor_set_csc_matrix(struct vl_compositor *compositor, const float *mat);
+
+#endif /* vl_compositor_h */
diff --git a/src/gallium/auxiliary/vl/vl_csc.c b/src/gallium/auxiliary/vl/vl_csc.c
new file mode 100644
index 0000000000..5ecc43a5fa
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_csc.c
@@ -0,0 +1,206 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "vl_csc.h"
+#include <util/u_math.h>
+#include <util/u_debug.h>
+
+/*
+ * Color space conversion formulas
+ *
+ * To convert YCbCr to RGB,
+ *    vec4  ycbcr, rgb
+ *    mat44 csc
+ *    rgb = csc * ycbcr
+ *
+ * To calculate the color space conversion matrix csc with ProcAmp adjustments,
+ *    mat44 csc, cstd, procamp, bias
+ *    csc = cstd * (procamp * bias)
+ *
+ * Where cstd is a matrix corresponding to one of the color standards (BT.601, BT.709, etc)
+ * adjusted for the kind of YCbCr -> RGB mapping wanted (1:1, full),
+ * bias is a matrix corresponding to the kind of YCbCr -> RGB mapping wanted (1:1, full)
+ *
+ * To calculate procamp,
+ *    mat44 procamp, hue, saturation, brightness, contrast
+ *    procamp = brightness * (saturation * (contrast * hue))
+ * Alternatively,
+ *    procamp = saturation * (brightness * (contrast * hue))
+ *
+ * contrast
+ * [ c, 0, 0, 0]
+ * [ 0, c, 0, 0]
+ * [ 0, 0, c, 0]
+ * [ 0, 0, 0, 1]
+ *
+ * brightness
+ * [ 1, 0, 0, b]
+ * [ 0, 1, 0, 0]
+ * [ 0, 0, 1, 0]
+ * [ 0, 0, 0, 1]
+ *
+ * saturation
+ * [ 1, 0, 0, 0]
+ * [ 0, s, 0, 0]
+ * [ 0, 0, s, 0]
+ * [ 0, 0, 0, 1]
+ *
+ * hue
+ * [ 1,       0,      0, 0]
+ * [ 0,  cos(h), sin(h), 0]
+ * [ 0, -sin(h), cos(h), 0]
+ * [ 0,       0,      0, 1]
+ *
+ * procamp
+ * [ c,           0,          0, b]
+ * [ 0,  c*s*cos(h), c*s*sin(h), 0]
+ * [ 0, -c*s*sin(h), c*s*cos(h), 0]
+ * [ 0,           0,          0, 1]
+ *
+ * bias
+ * [ 1, 0, 0,  ybias]
+ * [ 0, 1, 0, cbbias]
+ * [ 0, 0, 1, crbias]
+ * [ 0, 0, 0,      1]
+ *
+ * csc
+ * [ c*cstd[ 0], c*cstd[ 1]*s*cos(h) - c*cstd[ 2]*s*sin(h), c*cstd[ 2]*s*cos(h) + c*cstd[ 1]*s*sin(h), cstd[ 3] + cstd[ 0]*(b + c*ybias) + cstd[ 1]*(c*cbbias*s*cos(h) + c*crbias*s*sin(h)) + cstd[ 2]*(c*crbias*s*cos(h) - c*cbbias*s*sin(h))]
+ * [ c*cstd[ 4], c*cstd[ 5]*s*cos(h) - c*cstd[ 6]*s*sin(h), c*cstd[ 6]*s*cos(h) + c*cstd[ 5]*s*sin(h), cstd[ 7] + cstd[ 4]*(b + c*ybias) + cstd[ 5]*(c*cbbias*s*cos(h) + c*crbias*s*sin(h)) + cstd[ 6]*(c*crbias*s*cos(h) - c*cbbias*s*sin(h))]
+ * [ c*cstd[ 8], c*cstd[ 9]*s*cos(h) - c*cstd[10]*s*sin(h), c*cstd[10]*s*cos(h) + c*cstd[ 9]*s*sin(h), cstd[11] + cstd[ 8]*(b + c*ybias) + cstd[ 9]*(c*cbbias*s*cos(h) + c*crbias*s*sin(h)) + cstd[10]*(c*crbias*s*cos(h) - c*cbbias*s*sin(h))]
+ * [ c*cstd[12], c*cstd[13]*s*cos(h) - c*cstd[14]*s*sin(h), c*cstd[14]*s*cos(h) + c*cstd[13]*s*sin(h), cstd[15] + cstd[12]*(b + c*ybias) + cstd[13]*(c*cbbias*s*cos(h) + c*crbias*s*sin(h)) + cstd[14]*(c*crbias*s*cos(h) - c*cbbias*s*sin(h))]
+ */
+
+/*
+ * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [16,235]
+ */
+static const float bt_601[16] =
+{
+   1.0f,  0.0f,    1.371f, 0.0f,
+   1.0f, -0.336f, -0.698f, 0.0f,
+   1.0f,  1.732f,  0.0f,   0.0f,
+   0.0f,  0.0f,    0.0f,   1.0f
+};
+
+/*
+ * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [0,255]
+ */
+static const float bt_601_full[16] =
+{
+   1.164f,  0.0f,    1.596f, 0.0f,
+   1.164f, -0.391f, -0.813f, 0.0f,
+   1.164f,  2.018f,  0.0f,   0.0f,
+   0.0f,    0.0f,    0.0f,   1.0f
+};
+
+/*
+ * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [16,235]
+ */
+static const float bt_709[16] =
+{
+   1.0f,  0.0f,    1.540f, 0.0f,
+   1.0f, -0.183f, -0.459f, 0.0f,
+   1.0f,  1.816f,  0.0f,   0.0f,
+   0.0f,  0.0f,    0.0f,   1.0f
+};
+
+/*
+ * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
+ * Y is in [16,235], Cb and Cr are in [16,240]
+ * R, G, and B are in [0,255]
+ */
+static const float bt_709_full[16] =
+{
+   1.164f,  0.0f,    1.793f, 0.0f,
+   1.164f, -0.213f, -0.534f, 0.0f,
+   1.164f,  2.115f,  0.0f,   0.0f,
+   0.0f,    0.0f,    0.0f,   1.0f
+};
+
+static const float identity[16] =
+{
+   1.0f, 0.0f, 0.0f, 0.0f,
+   0.0f, 1.0f, 0.0f, 0.0f,
+   0.0f, 0.0f, 1.0f, 0.0f,
+   0.0f, 0.0f, 0.0f, 1.0f
+};
+
+void vl_csc_get_matrix(enum VL_CSC_COLOR_STANDARD cs,
+                       struct vl_procamp *procamp,
+                       bool full_range,
+                       float *matrix)
+{
+   float ybias = full_range ? -16.0f/255.0f : 0.0f;
+   float cbbias = -128.0f/255.0f;
+   float crbias = -128.0f/255.0f;
+   float c = procamp ? procamp->contrast : 1.0f;
+   float s = procamp ? procamp->saturation : 1.0f;
+   float b = procamp ? procamp->brightness : 0.0f;
+   float h = procamp ? procamp->hue : 0.0f;
+   const float *cstd;
+
+   assert(matrix);
+
+   switch (cs) {
+      case VL_CSC_COLOR_STANDARD_BT_601:
+         cstd = full_range ? &bt_601_full[0] : &bt_601[0];
+         break;
+      case VL_CSC_COLOR_STANDARD_BT_709:
+         cstd = full_range ? &bt_709_full[0] : &bt_709[0];
+         break;
+      case VL_CSC_COLOR_STANDARD_IDENTITY:
+      default:
+         assert(cs == VL_CSC_COLOR_STANDARD_IDENTITY);
+         memcpy(matrix, &identity[0], sizeof(float) * 16);
+         return;
+   }
+
+   matrix[ 0] = c*cstd[ 0];
+   matrix[ 1] = c*cstd[ 1]*s*cosf(h) - c*cstd[ 2]*s*sinf(h);
+   matrix[ 2] = c*cstd[ 2]*s*cosf(h) + c*cstd[ 1]*s*sinf(h);
+   matrix[ 3] = cstd[ 3] + cstd[ 0]*(b + c*ybias) + cstd[ 1]*(c*cbbias*s*cosf(h) + c*crbias*s*sinf(h)) + cstd[ 2]*(c*crbias*s*cosf(h) - c*cbbias*s*sinf(h));
+
+   matrix[ 4] = c*cstd[ 4];
+   matrix[ 5] = c*cstd[ 5]*s*cosf(h) - c*cstd[ 6]*s*sinf(h);
+   matrix[ 6] = c*cstd[ 6]*s*cosf(h) + c*cstd[ 5]*s*sinf(h);
+   matrix[ 7] = cstd[ 7] + cstd[ 4]*(b + c*ybias) + cstd[ 5]*(c*cbbias*s*cosf(h) + c*crbias*s*sinf(h)) + cstd[ 6]*(c*crbias*s*cosf(h) - c*cbbias*s*sinf(h));
+
+   matrix[ 8] = c*cstd[ 8];
+   matrix[ 9] = c*cstd[ 9]*s*cosf(h) - c*cstd[10]*s*sinf(h);
+   matrix[10] = c*cstd[10]*s*cosf(h) + c*cstd[ 9]*s*sinf(h);
+   matrix[11] = cstd[11] + cstd[ 8]*(b + c*ybias) + cstd[ 9]*(c*cbbias*s*cosf(h) + c*crbias*s*sinf(h)) + cstd[10]*(c*crbias*s*cosf(h) - c*cbbias*s*sinf(h));
+
+   matrix[12] = c*cstd[12];
+   matrix[13] = c*cstd[13]*s*cos(h) - c*cstd[14]*s*sin(h);
+   matrix[14] = c*cstd[14]*s*cos(h) + c*cstd[13]*s*sin(h);
+   matrix[15] = cstd[15] + cstd[12]*(b + c*ybias) + cstd[13]*(c*cbbias*s*cos(h) + c*crbias*s*sin(h)) + cstd[14]*(c*crbias*s*cos(h) - c*cbbias*s*sin(h));
+}
diff --git a/src/gallium/state_trackers/wgl/shared/stw_winsys.h b/src/gallium/auxiliary/vl/vl_csc.h
index c0bf82c9ed..722ca35f33 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_winsys.h
+++ b/src/gallium/auxiliary/vl/vl_csc.h
@@ -1,8 +1,8 @@
 /**************************************************************************
- *
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * 
+ * Copyright 2009 Younes Manton.
  * 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
@@ -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.
@@ -22,44 +22,32 @@
  * 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 STW_WINSYS_H
-#define STW_WINSYS_H
+#ifndef vl_csc_h
+#define vl_csc_h
 
-#include <windows.h> /* for HDC */
+#include <pipe/p_compiler.h>
 
-#include "pipe/p_compiler.h"
-
-struct pipe_screen;
-struct pipe_context;
-struct pipe_surface;
-
-struct stw_winsys
+struct vl_procamp
 {
-   struct pipe_screen *
-   (*create_screen)( void );
-
-   struct pipe_context *
-   (*create_context)( struct pipe_screen *screen );
-
-   void
-   (*flush_frontbuffer)( struct pipe_screen *screen,
-                         struct pipe_surface *surf,
-                         HDC hDC );
+   float brightness;
+   float contrast;
+   float saturation;
+   float hue;
 };
 
-boolean
-stw_init(const struct stw_winsys *stw_winsys);
-
-boolean
-stw_init_thread(void);
-
-void
-stw_cleanup_thread(void);
+enum VL_CSC_COLOR_STANDARD
+{
+   VL_CSC_COLOR_STANDARD_IDENTITY,
+   VL_CSC_COLOR_STANDARD_BT_601,
+   VL_CSC_COLOR_STANDARD_BT_709
+};
 
-void
-stw_cleanup(void);
+void vl_csc_get_matrix(enum VL_CSC_COLOR_STANDARD cs,
+                       struct vl_procamp *procamp,
+                       bool full_range,
+                       float *matrix);
 
-#endif /* STW_WINSYS_H */
+#endif /* vl_csc_h */
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
new file mode 100644
index 0000000000..6b3614821c
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c
@@ -0,0 +1,1654 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "vl_mpeg12_mc_renderer.h"
+#include <assert.h>
+#include <pipe/p_context.h>
+#include <pipe/p_inlines.h>
+#include <util/u_math.h>
+#include <util/u_memory.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+#include "vl_shader_build.h"
+
+#define DEFAULT_BUF_ALIGNMENT 1
+#define MACROBLOCK_WIDTH 16
+#define MACROBLOCK_HEIGHT 16
+#define BLOCK_WIDTH 8
+#define BLOCK_HEIGHT 8
+#define ZERO_BLOCK_NIL -1.0f
+#define ZERO_BLOCK_IS_NIL(zb) ((zb).x < 0.0f)
+
+struct vertex2f
+{
+   float x, y;
+};
+
+struct vertex4f
+{
+   float x, y, z, w;
+};
+
+struct vertex_shader_consts
+{
+   struct vertex4f denorm;
+};
+
+struct fragment_shader_consts
+{
+   struct vertex4f multiplier;
+   struct vertex4f div;
+};
+
+/*
+ * Muliplier renormalizes block samples from 16 bits to 12 bits.
+ * Divider is used when calculating Y % 2 for choosing top or bottom
+ * field for P or B macroblocks.
+ * TODO: Use immediates.
+ */
+static const struct fragment_shader_consts fs_consts = {
+   {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
+   {0.5f, 2.0f, 0.0f, 0.0f}
+};
+
+struct vert_stream_0
+{
+   struct vertex2f pos;
+   struct vertex2f luma_tc;
+   struct vertex2f cb_tc;
+   struct vertex2f cr_tc;
+};
+
+enum MACROBLOCK_TYPE
+{
+   MACROBLOCK_TYPE_INTRA,
+   MACROBLOCK_TYPE_FWD_FRAME_PRED,
+   MACROBLOCK_TYPE_FWD_FIELD_PRED,
+   MACROBLOCK_TYPE_BKWD_FRAME_PRED,
+   MACROBLOCK_TYPE_BKWD_FIELD_PRED,
+   MACROBLOCK_TYPE_BI_FRAME_PRED,
+   MACROBLOCK_TYPE_BI_FIELD_PRED,
+
+   NUM_MACROBLOCK_TYPES
+};
+
+static void
+create_intra_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 50;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    */
+   for (i = 0; i < 4; i++) {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    */
+   for (i = 0; i < 4; i++) {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (i = 0; i < 4; ++i) {
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->i_vs = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_intra_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    */
+   for (i = 0; i < 3; ++i) {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    */
+   for (i = 0; i < 3; ++i) {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (i = 0; i < 3; ++i) {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul o0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->i_fs = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_frame_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    * decl i4              ; Ref surface top field texcoords
+    * decl i5              ; Ref surface bottom field texcoords (unused, packed in the same stream)
+    */
+   for (i = 0; i < 6; i++) {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    * decl o4              ; Ref macroblock texcoords
+    */
+   for (i = 0; i < 5; i++) {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (i = 0; i < 4; ++i) {
+        inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+        ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* add o4, i0, i4       ; Translate vertex pos by motion vec to form ref macroblock texcoords */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, 4);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->p_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_field_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    * decl i3                      ; Ref macroblock texcoords
+    */
+   for (i = 0; i < 4; ++i) {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0, t1 */
+   decl = vl_decl_temps(0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    * decl s3                      ; Sampler for ref surface texture
+    */
+   for (i = 0; i < 4; ++i) {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (i = 0; i < 3; ++i) {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul t0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* tex2d t1, i3, s3             ; Read texel from ref macroblock */
+   inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 3, TGSI_FILE_SAMPLER, 3);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* add o0, t0, t1               ; Add ref and differential to form final output */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->p_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_field_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state vs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
+
+   ti = 3;
+
+   /*
+    * decl i0              ; Vertex pos
+    * decl i1              ; Luma texcoords
+    * decl i2              ; Chroma Cb texcoords
+    * decl i3              ; Chroma Cr texcoords
+    * decl i4              ; First ref macroblock top field texcoords
+    * decl i5              ; First ref macroblock bottom field texcoords (unused, packed in the same stream)
+    * decl i6              ; Second ref macroblock top field texcoords
+    * decl i7              ; Second ref macroblock bottom field texcoords (unused, packed in the same stream)
+    */
+   for (i = 0; i < 8; i++) {
+      decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl o0              ; Vertex pos
+    * decl o1              ; Luma texcoords
+    * decl o2              ; Chroma Cb texcoords
+    * decl o3              ; Chroma Cr texcoords
+    * decl o4              ; First ref macroblock texcoords
+    * decl o5              ; Second ref macroblock texcoords
+    */
+   for (i = 0; i < 6; i++) {
+      decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * mov o0, i0           ; Move input vertex pos to output
+    * mov o1, i1           ; Move input luma texcoords to output
+    * mov o2, i2           ; Move input chroma Cb texcoords to output
+    * mov o3, i3           ; Move input chroma Cr texcoords to output
+    */
+   for (i = 0; i < 4; ++i) {
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * add o4, i0, i4       ; Translate vertex pos by motion vec to form first ref macroblock texcoords
+    * add o5, i0, i6       ; Translate vertex pos by motion vec to form second ref macroblock texcoords
+    */
+   for (i = 0; i < 2; ++i) {
+      inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, (i + 2) * 2);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   vs.tokens = tokens;
+   r->b_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
+   free(tokens);
+}
+
+static void
+create_field_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+create_frame_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   const unsigned max_tokens = 100;
+
+   struct pipe_shader_state fs;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+
+   unsigned ti;
+
+   unsigned i;
+
+   assert(r);
+
+   tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+   *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
+
+   ti = 3;
+
+   /*
+    * decl i0                      ; Luma texcoords
+    * decl i1                      ; Chroma Cb texcoords
+    * decl i2                      ; Chroma Cr texcoords
+    * decl i3                      ; First ref macroblock texcoords
+    * decl i4                      ; Second ref macroblock texcoords
+    */
+   for (i = 0; i < 5; ++i) {
+      decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * decl c0                      ; Scaling factor, rescales 16-bit snorm to 9-bit snorm
+    * decl c1                      ; Constant 1/2 in .x channel to use as weight to blend past and future texels
+    */
+   decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl o0                      ; Fragment color */
+   decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /* decl t0-t2 */
+   decl = vl_decl_temps(0, 2);
+   ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * decl s0                      ; Sampler for luma texture
+    * decl s1                      ; Sampler for chroma Cb texture
+    * decl s2                      ; Sampler for chroma Cr texture
+    * decl s3                      ; Sampler for first ref surface texture
+    * decl s4                      ; Sampler for second ref surface texture
+    */
+   for (i = 0; i < 5; ++i) {
+      decl = vl_decl_samplers(i, i);
+      ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /*
+    * tex2d t1, i0, s0             ; Read texel from luma texture
+    * mov t0.x, t1.x               ; Move luma sample into .x component
+    * tex2d t1, i1, s1             ; Read texel from chroma Cb texture
+    * mov t0.y, t1.x               ; Move Cb sample into .y component
+    * tex2d t1, i2, s2             ; Read texel from chroma Cr texture
+    * mov t0.z, t1.x               ; Move Cr sample into .z component
+    */
+   for (i = 0; i < 3; ++i) {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+      inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+      inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* mul t0, t0, c0               ; Rescale texel to correct range */
+   inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /*
+    * tex2d t1, i3, s3             ; Read texel from first ref macroblock
+    * tex2d t2, i4, s4             ; Read texel from second ref macroblock
+    */
+   for (i = 0; i < 2; ++i) {
+      inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, i + 3);
+      ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+   }
+
+   /* lerp t1, c1.x, t1, t2        ; Blend past and future texels */
+   inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
+   inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* add o0, t0, t1               ; Add past/future ref and differential to form final output */
+   inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   /* end */
+   inst = vl_end();
+   ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
+
+   assert(ti <= max_tokens);
+
+   fs.tokens = tokens;
+   r->b_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
+   free(tokens);
+}
+
+static void
+create_field_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(false);
+}
+
+static void
+xfer_buffers_map(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned i;
+
+   assert(r);
+
+   for (i = 0; i < 3; ++i) {
+      r->tex_transfer[i] = r->pipe->screen->get_tex_transfer
+      (
+         r->pipe->screen, r->textures.all[i],
+         0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0,
+         r->textures.all[i]->width[0], r->textures.all[i]->height[0]
+      );
+
+      r->texels[i] = r->pipe->screen->transfer_map(r->pipe->screen, r->tex_transfer[i]);
+   }
+}
+
+static void
+xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned i;
+
+   assert(r);
+
+   for (i = 0; i < 3; ++i) {
+      r->pipe->screen->transfer_unmap(r->pipe->screen, r->tex_transfer[i]);
+      r->pipe->screen->tex_transfer_destroy(r->tex_transfer[i]);
+   }
+}
+
+static bool
+init_pipe_state(struct vl_mpeg12_mc_renderer *r)
+{
+   struct pipe_sampler_state sampler;
+   unsigned filters[5];
+   unsigned i;
+
+   assert(r);
+
+   r->viewport.scale[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   r->viewport.scale[1] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   r->viewport.scale[2] = 1;
+   r->viewport.scale[3] = 1;
+   r->viewport.translate[0] = 0;
+   r->viewport.translate[1] = 0;
+   r->viewport.translate[2] = 0;
+   r->viewport.translate[3] = 0;
+
+   r->fb_state.width = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   r->fb_state.height = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   r->fb_state.nr_cbufs = 1;
+   r->fb_state.zsbuf = NULL;
+
+   /* Luma filter */
+   filters[0] = PIPE_TEX_FILTER_NEAREST;
+   /* Chroma filters */
+   if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_444 ||
+       r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
+      filters[1] = PIPE_TEX_FILTER_NEAREST;
+      filters[2] = PIPE_TEX_FILTER_NEAREST;
+   }
+   else {
+      filters[1] = PIPE_TEX_FILTER_LINEAR;
+      filters[2] = PIPE_TEX_FILTER_LINEAR;
+   }
+   /* Fwd, bkwd ref filters */
+   filters[3] = PIPE_TEX_FILTER_LINEAR;
+   filters[4] = PIPE_TEX_FILTER_LINEAR;
+
+   for (i = 0; i < 5; ++i) {
+      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_img_filter = filters[i];
+      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+      sampler.mag_img_filter = filters[i];
+      sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
+      sampler.compare_func = PIPE_FUNC_ALWAYS;
+      sampler.normalized_coords = 1;
+      /*sampler.prefilter = ; */
+      /*sampler.shadow_ambient = ; */
+      /*sampler.lod_bias = ; */
+      sampler.min_lod = 0;
+      /*sampler.max_lod = ; */
+      /*sampler.border_color[i] = ; */
+      /*sampler.max_anisotropy = ; */
+      r->samplers.all[i] = r->pipe->create_sampler_state(r->pipe, &sampler);
+   }
+
+   return true;
+}
+
+static void
+cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned i;
+
+   assert(r);
+
+   for (i = 0; i < 5; ++i)
+      r->pipe->delete_sampler_state(r->pipe, r->samplers.all[i]);
+}
+
+static bool
+init_shaders(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   create_intra_vert_shader(r);
+   create_intra_frag_shader(r);
+   create_frame_pred_vert_shader(r);
+   create_frame_pred_frag_shader(r);
+   create_frame_bi_pred_vert_shader(r);
+   create_frame_bi_pred_frag_shader(r);
+
+   return true;
+}
+
+static void
+cleanup_shaders(struct vl_mpeg12_mc_renderer *r)
+{
+   assert(r);
+
+   r->pipe->delete_vs_state(r->pipe, r->i_vs);
+   r->pipe->delete_fs_state(r->pipe, r->i_fs);
+   r->pipe->delete_vs_state(r->pipe, r->p_vs[0]);
+   r->pipe->delete_fs_state(r->pipe, r->p_fs[0]);
+   r->pipe->delete_vs_state(r->pipe, r->b_vs[0]);
+   r->pipe->delete_fs_state(r->pipe, r->b_fs[0]);
+}
+
+static bool
+init_buffers(struct vl_mpeg12_mc_renderer *r)
+{
+   struct pipe_texture template;
+
+   const unsigned mbw =
+      align(r->picture_width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
+   const unsigned mbh =
+      align(r->picture_height, MACROBLOCK_HEIGHT) / MACROBLOCK_HEIGHT;
+
+   unsigned i;
+
+   assert(r);
+
+   r->macroblocks_per_batch =
+      mbw * (r->bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE ? mbh : 1);
+   r->num_macroblocks = 0;
+   r->macroblock_buf = MALLOC(r->macroblocks_per_batch * sizeof(struct pipe_mpeg12_macroblock));
+
+   memset(&template, 0, sizeof(struct pipe_texture));
+   template.target = PIPE_TEXTURE_2D;
+   /* TODO: Accomodate HW that can't do this and also for cases when this isn't precise enough */
+   template.format = PIPE_FORMAT_R16_SNORM;
+   template.last_level = 0;
+   template.width[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_width) : r->picture_width;
+   template.height[0] = r->pot_buffers ?
+      util_next_power_of_two(r->picture_height) : r->picture_height;
+   template.depth[0] = 1;
+   pf_get_block(template.format, &template.block);
+   template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC;
+
+   r->textures.individual.y = r->pipe->screen->texture_create(r->pipe->screen, &template);
+
+   if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
+      template.width[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_width / 2) :
+         r->picture_width / 2;
+      template.height[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_height / 2) :
+         r->picture_height / 2;
+   }
+   else if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422)
+      template.height[0] = r->pot_buffers ?
+         util_next_power_of_two(r->picture_height / 2) :
+         r->picture_height / 2;
+
+   r->textures.individual.cb =
+      r->pipe->screen->texture_create(r->pipe->screen, &template);
+   r->textures.individual.cr =
+      r->pipe->screen->texture_create(r->pipe->screen, &template);
+
+   r->vertex_bufs.individual.ycbcr.stride = sizeof(struct vertex2f) * 4;
+   r->vertex_bufs.individual.ycbcr.max_index = 24 * r->macroblocks_per_batch - 1;
+   r->vertex_bufs.individual.ycbcr.buffer_offset = 0;
+   r->vertex_bufs.individual.ycbcr.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex2f) * 4 * 24 * r->macroblocks_per_batch
+   );
+
+   for (i = 1; i < 3; ++i) {
+      r->vertex_bufs.all[i].stride = sizeof(struct vertex2f) * 2;
+      r->vertex_bufs.all[i].max_index = 24 * r->macroblocks_per_batch - 1;
+      r->vertex_bufs.all[i].buffer_offset = 0;
+      r->vertex_bufs.all[i].buffer = pipe_buffer_create
+      (
+         r->pipe->screen,
+         DEFAULT_BUF_ALIGNMENT,
+         PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
+         sizeof(struct vertex2f) * 2 * 24 * r->macroblocks_per_batch
+      );
+   }
+
+   /* Position element */
+   r->vertex_elems[0].src_offset = 0;
+   r->vertex_elems[0].vertex_buffer_index = 0;
+   r->vertex_elems[0].nr_components = 2;
+   r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Luma, texcoord element */
+   r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[1].vertex_buffer_index = 0;
+   r->vertex_elems[1].nr_components = 2;
+   r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cr texcoord element */
+   r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
+   r->vertex_elems[2].vertex_buffer_index = 0;
+   r->vertex_elems[2].nr_components = 2;
+   r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Chroma Cb texcoord element */
+   r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
+   r->vertex_elems[3].vertex_buffer_index = 0;
+   r->vertex_elems[3].nr_components = 2;
+   r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface top field texcoord element */
+   r->vertex_elems[4].src_offset = 0;
+   r->vertex_elems[4].vertex_buffer_index = 1;
+   r->vertex_elems[4].nr_components = 2;
+   r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* First ref surface bottom field texcoord element */
+   r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[5].vertex_buffer_index = 1;
+   r->vertex_elems[5].nr_components = 2;
+   r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface top field texcoord element */
+   r->vertex_elems[6].src_offset = 0;
+   r->vertex_elems[6].vertex_buffer_index = 2;
+   r->vertex_elems[6].nr_components = 2;
+   r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   /* Second ref surface bottom field texcoord element */
+   r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
+   r->vertex_elems[7].vertex_buffer_index = 2;
+   r->vertex_elems[7].nr_components = 2;
+   r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
+
+   r->vs_const_buf.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
+      sizeof(struct vertex_shader_consts)
+   );
+
+   r->fs_const_buf.buffer = pipe_buffer_create
+   (
+      r->pipe->screen,
+      DEFAULT_BUF_ALIGNMENT,
+      PIPE_BUFFER_USAGE_CONSTANT, sizeof(struct fragment_shader_consts)
+   );
+
+   memcpy
+   (
+      pipe_buffer_map(r->pipe->screen, r->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
+      &fs_consts, sizeof(struct fragment_shader_consts)
+   );
+
+   pipe_buffer_unmap(r->pipe->screen, r->fs_const_buf.buffer);
+
+   return true;
+}
+
+static void
+cleanup_buffers(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned i;
+
+   assert(r);
+
+   pipe_buffer_reference(&r->vs_const_buf.buffer, NULL);
+   pipe_buffer_reference(&r->fs_const_buf.buffer, NULL);
+
+   for (i = 0; i < 3; ++i)
+      pipe_buffer_reference(&r->vertex_bufs.all[i].buffer, NULL);
+
+   for (i = 0; i < 3; ++i)
+      pipe_texture_reference(&r->textures.all[i], NULL);
+
+   FREE(r->macroblock_buf);
+}
+
+static enum MACROBLOCK_TYPE
+get_macroblock_type(struct pipe_mpeg12_macroblock *mb)
+{
+   assert(mb);
+
+   switch (mb->mb_type) {
+      case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
+         return MACROBLOCK_TYPE_INTRA;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_FWD_FRAME_PRED : MACROBLOCK_TYPE_FWD_FIELD_PRED;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_BKWD_FRAME_PRED : MACROBLOCK_TYPE_BKWD_FIELD_PRED;
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
+         return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
+            MACROBLOCK_TYPE_BI_FRAME_PRED : MACROBLOCK_TYPE_BI_FIELD_PRED;
+      default:
+         assert(0);
+   }
+
+   /* Unreachable */
+   return -1;
+}
+
+/* XXX: One of these days this will have to be killed with fire */
+#define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, use_zb, zb)				\
+	do {															\
+	(vb)[0].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[0].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[1].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[1].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+	(vb)[2].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[3].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].pos.y = (mby) * (unity) + (ofsy);			\
+	(vb)[4].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[4].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+	(vb)[5].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
+																\
+	if (!use_zb || (cbp) & (lm))												\
+	{															\
+		(vb)[0].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].luma_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].luma_tc.x = (zb)[0].x;		(vb)[0].luma_tc.y = (zb)[0].y;						\
+		(vb)[1].luma_tc.x = (zb)[0].x;		(vb)[1].luma_tc.y = (zb)[0].y + (hy);					\
+		(vb)[2].luma_tc.x = (zb)[0].x + (hx);	(vb)[2].luma_tc.y = (zb)[0].y;						\
+		(vb)[3].luma_tc.x = (zb)[0].x + (hx);	(vb)[3].luma_tc.y = (zb)[0].y;						\
+		(vb)[4].luma_tc.x = (zb)[0].x;		(vb)[4].luma_tc.y = (zb)[0].y + (hy);					\
+		(vb)[5].luma_tc.x = (zb)[0].x + (hx);	(vb)[5].luma_tc.y = (zb)[0].y + (hy);					\
+	}															\
+																\
+	if (!use_zb || (cbp) & (cbm))												\
+	{															\
+		(vb)[0].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cb_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].cb_tc.x = (zb)[1].x;		(vb)[0].cb_tc.y = (zb)[1].y;						\
+		(vb)[1].cb_tc.x = (zb)[1].x;		(vb)[1].cb_tc.y = (zb)[1].y + (hy);					\
+		(vb)[2].cb_tc.x = (zb)[1].x + (hx);	(vb)[2].cb_tc.y = (zb)[1].y;						\
+		(vb)[3].cb_tc.x = (zb)[1].x + (hx);	(vb)[3].cb_tc.y = (zb)[1].y;						\
+		(vb)[4].cb_tc.x = (zb)[1].x;		(vb)[4].cb_tc.y = (zb)[1].y + (hy);					\
+		(vb)[5].cb_tc.x = (zb)[1].x + (hx);	(vb)[5].cb_tc.y = (zb)[1].y + (hy);					\
+	}															\
+																\
+	if (!use_zb || (cbp) & (crm))												\
+	{															\
+		(vb)[0].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[1].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[2].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[3].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cr_tc.y = (mby) * (unity) + (ofsy);		\
+		(vb)[4].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+		(vb)[5].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
+	}															\
+	else															\
+	{															\
+		(vb)[0].cr_tc.x = (zb)[2].x;		(vb)[0].cr_tc.y = (zb)[2].y;						\
+		(vb)[1].cr_tc.x = (zb)[2].x;		(vb)[1].cr_tc.y = (zb)[2].y + (hy);					\
+		(vb)[2].cr_tc.x = (zb)[2].x + (hx);	(vb)[2].cr_tc.y = (zb)[2].y;						\
+		(vb)[3].cr_tc.x = (zb)[2].x + (hx);	(vb)[3].cr_tc.y = (zb)[2].y;						\
+		(vb)[4].cr_tc.x = (zb)[2].x;		(vb)[4].cr_tc.y = (zb)[2].y + (hy);					\
+		(vb)[5].cr_tc.x = (zb)[2].x + (hx);	(vb)[5].cr_tc.y = (zb)[2].y + (hy);					\
+	}															\
+	} while (0)
+
+static void
+gen_macroblock_verts(struct vl_mpeg12_mc_renderer *r,
+                     struct pipe_mpeg12_macroblock *mb, unsigned pos,
+                     struct vert_stream_0 *ycbcr_vb, struct vertex2f **ref_vb)
+{
+   struct vertex2f mo_vec[2];
+
+   unsigned i;
+
+   assert(r);
+   assert(mb);
+   assert(ycbcr_vb);
+   assert(pos < r->macroblocks_per_batch);
+
+   switch (mb->mb_type) {
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
+      {
+         struct vertex2f *vb;
+
+         assert(ref_vb && ref_vb[1]);
+
+         vb = ref_vb[1] + pos * 2 * 24;
+
+         mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
+         mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+         if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME) {
+            for (i = 0; i < 24 * 2; i += 2) {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+            }
+         }
+         else {
+            mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
+            mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+            for (i = 0; i < 24 * 2; i += 2) {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+               vb[i + 1].x = mo_vec[1].x;
+               vb[i + 1].y = mo_vec[1].y;
+            }
+         }
+
+         /* fall-through */
+      }
+      case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
+      case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
+      {
+         struct vertex2f *vb;
+
+         assert(ref_vb && ref_vb[0]);
+
+         vb = ref_vb[0] + pos * 2 * 24;
+
+         if (mb->mb_type == PIPE_MPEG12_MACROBLOCK_TYPE_BKWD) {
+             mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
+             mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
+
+             if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD) {
+                mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
+                mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
+             }
+         }
+         else {
+            mo_vec[0].x = mb->pmv[0][0][0] * 0.5f * r->surface_tex_inv_size.x;
+            mo_vec[0].y = mb->pmv[0][0][1] * 0.5f * r->surface_tex_inv_size.y;
+
+            if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD) {
+               mo_vec[1].x = mb->pmv[1][0][0] * 0.5f * r->surface_tex_inv_size.x;
+               mo_vec[1].y = mb->pmv[1][0][1] * 0.5f * r->surface_tex_inv_size.y;
+            }
+         }
+
+         if (mb->mb_type == PIPE_MPEG12_MOTION_TYPE_FRAME) {
+            for (i = 0; i < 24 * 2; i += 2) {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+            }
+         }
+         else {
+            for (i = 0; i < 24 * 2; i += 2) {
+               vb[i].x = mo_vec[0].x;
+               vb[i].y = mo_vec[0].y;
+               vb[i + 1].x = mo_vec[1].x;
+               vb[i + 1].y = mo_vec[1].y;
+            }
+         }
+
+         /* fall-through */
+      }
+      case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
+      {
+         const struct vertex2f unit =
+         {
+            r->surface_tex_inv_size.x * MACROBLOCK_WIDTH,
+            r->surface_tex_inv_size.y * MACROBLOCK_HEIGHT
+         };
+         const struct vertex2f half =
+         {
+            r->surface_tex_inv_size.x * (MACROBLOCK_WIDTH / 2),
+            r->surface_tex_inv_size.y * (MACROBLOCK_HEIGHT / 2)
+         };
+         const bool use_zb = r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE;
+
+         struct vert_stream_0 *vb = ycbcr_vb + pos * 24;
+
+         SET_BLOCK(vb, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, 0, 0, half.x, half.y,
+                   32, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 6, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, half.x, 0, half.x, half.y,
+                   16, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 12, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, 0, half.y, half.x, half.y,
+                   8, 2, 1, use_zb, r->zero_block);
+
+         SET_BLOCK(vb + 18, mb->cbp, mb->mbx, mb->mby,
+                   unit.x, unit.y, half.x, half.y, half.x, half.y,
+                   4, 2, 1, use_zb, r->zero_block);
+
+         break;
+      }
+      default:
+         assert(0);
+   }
+}
+
+static void
+gen_macroblock_stream(struct vl_mpeg12_mc_renderer *r,
+                      unsigned *num_macroblocks)
+{
+   unsigned offset[NUM_MACROBLOCK_TYPES];
+   struct vert_stream_0 *ycbcr_vb;
+   struct vertex2f *ref_vb[2];
+   unsigned i;
+
+   assert(r);
+   assert(num_macroblocks);
+
+   for (i = 0; i < r->num_macroblocks; ++i) {
+      enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
+      ++num_macroblocks[mb_type];
+   }
+
+   offset[0] = 0;
+
+   for (i = 1; i < NUM_MACROBLOCK_TYPES; ++i)
+      offset[i] = offset[i - 1] + num_macroblocks[i - 1];
+
+   ycbcr_vb = (struct vert_stream_0 *)pipe_buffer_map
+   (
+      r->pipe->screen,
+      r->vertex_bufs.individual.ycbcr.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   for (i = 0; i < 2; ++i)
+      ref_vb[i] = (struct vertex2f *)pipe_buffer_map
+      (
+         r->pipe->screen,
+         r->vertex_bufs.individual.ref[i].buffer,
+         PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+      );
+
+   for (i = 0; i < r->num_macroblocks; ++i) {
+      enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
+
+      gen_macroblock_verts(r, &r->macroblock_buf[i], offset[mb_type],
+                           ycbcr_vb, ref_vb);
+
+      ++offset[mb_type];
+   }
+
+   pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ycbcr.buffer);
+   for (i = 0; i < 2; ++i)
+      pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ref[i].buffer);
+}
+
+static void
+flush(struct vl_mpeg12_mc_renderer *r)
+{
+   unsigned num_macroblocks[NUM_MACROBLOCK_TYPES] = { 0 };
+   unsigned vb_start = 0;
+   struct vertex_shader_consts *vs_consts;
+   unsigned i;
+
+   assert(r);
+   assert(r->num_macroblocks == r->macroblocks_per_batch);
+
+   gen_macroblock_stream(r, num_macroblocks);
+
+   r->fb_state.cbufs[0] = r->pipe->screen->get_tex_surface
+   (
+      r->pipe->screen, r->surface,
+      0, 0, 0, PIPE_BUFFER_USAGE_GPU_WRITE
+   );
+
+   r->pipe->set_framebuffer_state(r->pipe, &r->fb_state);
+   r->pipe->set_viewport_state(r->pipe, &r->viewport);
+
+   vs_consts = pipe_buffer_map
+   (
+      r->pipe->screen, r->vs_const_buf.buffer,
+      PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
+   );
+
+   vs_consts->denorm.x = r->surface->width[0];
+   vs_consts->denorm.y = r->surface->height[0];
+
+   pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf.buffer);
+
+   r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_VERTEX, 0,
+                                &r->vs_const_buf);
+   r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_FRAGMENT, 0,
+                                &r->fs_const_buf);
+
+   if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
+      r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
+      r->pipe->set_sampler_textures(r->pipe, 3, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 3, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->i_vs);
+      r->pipe->bind_fs_state(r->pipe, r->i_fs);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0) {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ ) {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0) {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ ) {
+      r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
+      r->textures.individual.ref[0] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24;
+   }
+
+   if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0) {
+      r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->textures.individual.ref[1] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->b_vs[0]);
+      r->pipe->bind_fs_state(r->pipe, r->b_fs[0]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24;
+   }
+
+   if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ ) {
+      r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
+      r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
+      r->textures.individual.ref[0] = r->past;
+      r->textures.individual.ref[1] = r->future;
+      r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
+      r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
+      r->pipe->bind_vs_state(r->pipe, r->b_vs[1]);
+      r->pipe->bind_fs_state(r->pipe, r->b_fs[1]);
+
+      r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
+                           num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24);
+      vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24;
+   }
+
+   r->pipe->flush(r->pipe, PIPE_FLUSH_RENDER_CACHE, r->fence);
+   pipe_surface_reference(&r->fb_state.cbufs[0], NULL);
+
+   if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
+      for (i = 0; i < 3; ++i)
+         r->zero_block[i].x = ZERO_BLOCK_NIL;
+
+   r->num_macroblocks = 0;
+}
+
+static void
+grab_frame_coded_block(short *src, short *dst, unsigned dst_pitch)
+{
+   unsigned y;
+
+   assert(src);
+   assert(dst);
+
+   for (y = 0; y < BLOCK_HEIGHT; ++y)
+      memcpy(dst + y * dst_pitch, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
+}
+
+static void
+grab_field_coded_block(short *src, short *dst, unsigned dst_pitch)
+{
+   unsigned y;
+
+   assert(src);
+   assert(dst);
+
+   for (y = 0; y < BLOCK_HEIGHT; ++y)
+      memcpy(dst + y * dst_pitch * 2, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
+}
+
+static void
+fill_zero_block(short *dst, unsigned dst_pitch)
+{
+   unsigned y;
+
+   assert(dst);
+
+   for (y = 0; y < BLOCK_HEIGHT; ++y)
+      memset(dst + y * dst_pitch, 0, BLOCK_WIDTH * 2);
+}
+
+static void
+grab_blocks(struct vl_mpeg12_mc_renderer *r, unsigned mbx, unsigned mby,
+            enum pipe_mpeg12_dct_type dct_type, unsigned cbp, short *blocks)
+{
+   unsigned tex_pitch;
+   short *texels;
+   unsigned tb = 0, sb = 0;
+   unsigned mbpx = mbx * MACROBLOCK_WIDTH, mbpy = mby * MACROBLOCK_HEIGHT;
+   unsigned x, y;
+
+   assert(r);
+   assert(blocks);
+
+   tex_pitch = r->tex_transfer[0]->stride / r->tex_transfer[0]->block.size;
+   texels = r->texels[0] + mbpy * tex_pitch + mbpx;
+
+   for (y = 0; y < 2; ++y) {
+      for (x = 0; x < 2; ++x, ++tb) {
+         if ((cbp >> (5 - tb)) & 1) {
+            if (dct_type == PIPE_MPEG12_DCT_TYPE_FRAME) {
+               grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
+                                      texels + y * tex_pitch * BLOCK_WIDTH +
+                                      x * BLOCK_WIDTH, tex_pitch);
+            }
+            else {
+               grab_field_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
+                                      texels + y * tex_pitch + x * BLOCK_WIDTH,
+                                      tex_pitch);
+            }
+
+            ++sb;
+         }
+         else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
+            if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
+                ZERO_BLOCK_IS_NIL(r->zero_block[0])) {
+               fill_zero_block(texels + y * tex_pitch * BLOCK_WIDTH + x * BLOCK_WIDTH, tex_pitch);
+               if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
+                  r->zero_block[0].x = (mbpx + x * 8) * r->surface_tex_inv_size.x;
+                  r->zero_block[0].y = (mbpy + y * 8) * r->surface_tex_inv_size.y;
+               }
+            }
+         }
+      }
+   }
+
+   /* TODO: Implement 422, 444 */
+   assert(r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
+
+   mbpx /= 2;
+   mbpy /= 2;
+
+   for (tb = 0; tb < 2; ++tb) {
+      tex_pitch = r->tex_transfer[tb + 1]->stride / r->tex_transfer[tb + 1]->block.size;
+      texels = r->texels[tb + 1] + mbpy * tex_pitch + mbpx;
+
+      if ((cbp >> (1 - tb)) & 1) {
+         grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT, texels, tex_pitch);
+         ++sb;
+      }
+      else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
+         if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
+             ZERO_BLOCK_IS_NIL(r->zero_block[tb + 1])) {
+            fill_zero_block(texels, tex_pitch);
+            if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
+               r->zero_block[tb + 1].x = (mbpx << 1) * r->surface_tex_inv_size.x;
+               r->zero_block[tb + 1].y = (mbpy << 1) * r->surface_tex_inv_size.y;
+            }
+         }
+      }
+   }
+}
+
+static void
+grab_macroblock(struct vl_mpeg12_mc_renderer *r,
+                struct pipe_mpeg12_macroblock *mb)
+{
+   assert(r);
+   assert(mb);
+   assert(r->num_macroblocks < r->macroblocks_per_batch);
+
+   memcpy(&r->macroblock_buf[r->num_macroblocks], mb,
+          sizeof(struct pipe_mpeg12_macroblock));
+
+   grab_blocks(r, mb->mbx, mb->mby, mb->dct_type, mb->cbp, mb->blocks);
+
+   ++r->num_macroblocks;
+}
+
+bool
+vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
+                           struct pipe_context *pipe,
+                           unsigned picture_width,
+                           unsigned picture_height,
+                           enum pipe_video_chroma_format chroma_format,
+                           enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
+                           enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
+                           bool pot_buffers)
+{
+   unsigned i;
+
+   assert(renderer);
+   assert(pipe);
+   /* TODO: Implement other policies */
+   assert(bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE);
+   /* TODO: Implement this */
+   /* XXX: XFER_ALL sampling issue at block edges when using bilinear filtering */
+   assert(eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE);
+   /* TODO: Non-pot buffers untested, probably doesn't work without changes to texcoord generation, vert shader, etc */
+   assert(pot_buffers);
+
+   memset(renderer, 0, sizeof(struct vl_mpeg12_mc_renderer));
+
+   renderer->pipe = pipe;
+   renderer->picture_width = picture_width;
+   renderer->picture_height = picture_height;
+   renderer->chroma_format = chroma_format;
+   renderer->bufmode = bufmode;
+   renderer->eb_handling = eb_handling;
+   renderer->pot_buffers = pot_buffers;
+
+   if (!init_pipe_state(renderer))
+      return false;
+   if (!init_shaders(renderer)) {
+      cleanup_pipe_state(renderer);
+      return false;
+   }
+   if (!init_buffers(renderer)) {
+      cleanup_shaders(renderer);
+      cleanup_pipe_state(renderer);
+      return false;
+   }
+
+   renderer->surface = NULL;
+   renderer->past = NULL;
+   renderer->future = NULL;
+   for (i = 0; i < 3; ++i)
+      renderer->zero_block[i].x = ZERO_BLOCK_NIL;
+   renderer->num_macroblocks = 0;
+
+   xfer_buffers_map(renderer);
+
+   return true;
+}
+
+void
+vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
+{
+   assert(renderer);
+
+   xfer_buffers_unmap(renderer);
+
+   cleanup_pipe_state(renderer);
+   cleanup_shaders(renderer);
+   cleanup_buffers(renderer);
+}
+
+void
+vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer
+                                         *renderer,
+                                         struct pipe_texture *surface,
+                                         struct pipe_texture *past,
+                                         struct pipe_texture *future,
+                                         unsigned num_macroblocks,
+                                         struct pipe_mpeg12_macroblock
+                                         *mpeg12_macroblocks,
+                                         struct pipe_fence_handle **fence)
+{
+   bool new_surface = false;
+
+   assert(renderer);
+   assert(surface);
+   assert(num_macroblocks);
+   assert(mpeg12_macroblocks);
+
+   if (renderer->surface) {
+      if (surface != renderer->surface) {
+         if (renderer->num_macroblocks > 0) {
+            xfer_buffers_unmap(renderer);
+            flush(renderer);
+         }
+         
+         new_surface = true;
+      }
+
+      /* If the surface we're rendering hasn't changed the ref frames shouldn't change. */
+      assert(surface != renderer->surface || renderer->past == past);
+      assert(surface != renderer->surface || renderer->future == future);
+   }
+   else
+      new_surface = true;
+
+   if (new_surface) {
+      renderer->surface = surface;
+      renderer->past = past;
+      renderer->future = future;
+      renderer->fence = fence;
+      renderer->surface_tex_inv_size.x = 1.0f / surface->width[0];
+      renderer->surface_tex_inv_size.y = 1.0f / surface->height[0];
+   }
+
+   while (num_macroblocks) {
+      unsigned left_in_batch = renderer->macroblocks_per_batch - renderer->num_macroblocks;
+      unsigned num_to_submit = MIN2(num_macroblocks, left_in_batch);
+      unsigned i;
+
+      for (i = 0; i < num_to_submit; ++i) {
+         assert(mpeg12_macroblocks[i].base.codec == PIPE_VIDEO_CODEC_MPEG12);
+         grab_macroblock(renderer, &mpeg12_macroblocks[i]);
+      }
+
+      num_macroblocks -= num_to_submit;
+
+      if (renderer->num_macroblocks == renderer->macroblocks_per_batch) {
+         xfer_buffers_unmap(renderer);
+         flush(renderer);
+         xfer_buffers_map(renderer);
+         /* Next time we get this surface it may have new ref frames */
+         renderer->surface = NULL;
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
new file mode 100644
index 0000000000..5d2c1273ee
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.h
@@ -0,0 +1,120 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 vl_mpeg12_mc_renderer_h
+#define vl_mpeg12_mc_renderer_h
+
+#include <pipe/p_compiler.h>
+#include <pipe/p_state.h>
+#include <pipe/p_video_state.h>
+
+struct pipe_context;
+struct pipe_video_surface;
+struct pipe_macroblock;
+
+/* A slice is video-width (rounded up to a multiple of macroblock width) x macroblock height */
+enum VL_MPEG12_MC_RENDERER_BUFFER_MODE
+{
+   VL_MPEG12_MC_RENDERER_BUFFER_SLICE,  /* Saves memory at the cost of smaller batches */
+   VL_MPEG12_MC_RENDERER_BUFFER_PICTURE /* Larger batches, more memory */
+};
+
+enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK
+{
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL, /* Waste of memory bandwidth */
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE, /* Can only do point-filtering when interpolating subsampled chroma channels */
+   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE /* Needs conditional texel fetch! */
+};
+
+struct vl_mpeg12_mc_renderer
+{
+   struct pipe_context *pipe;
+   unsigned picture_width;
+   unsigned picture_height;
+   enum pipe_video_chroma_format chroma_format;
+   enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode;
+   enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling;
+   bool pot_buffers;
+   unsigned macroblocks_per_batch;
+
+   struct pipe_viewport_state viewport;
+   struct pipe_constant_buffer vs_const_buf;
+   struct pipe_constant_buffer fs_const_buf;
+   struct pipe_framebuffer_state fb_state;
+   struct pipe_vertex_element vertex_elems[8];
+	
+   union
+   {
+      void *all[5];
+      struct { void *y, *cb, *cr, *ref[2]; } individual;
+   } samplers;
+	
+   void *i_vs, *p_vs[2], *b_vs[2];
+   void *i_fs, *p_fs[2], *b_fs[2];
+	
+   union
+   {
+      struct pipe_texture *all[5];
+      struct { struct pipe_texture *y, *cb, *cr, *ref[2]; } individual;
+   } textures;
+
+   union
+   {
+      struct pipe_vertex_buffer all[3];
+      struct { struct pipe_vertex_buffer ycbcr, ref[2]; } individual;
+   } vertex_bufs;
+	
+   struct pipe_texture *surface, *past, *future;
+   struct pipe_fence_handle **fence;
+   unsigned num_macroblocks;
+   struct pipe_mpeg12_macroblock *macroblock_buf;
+   struct pipe_transfer *tex_transfer[3];
+   short *texels[3];
+   struct { float x, y; } surface_tex_inv_size;
+   struct { float x, y; } zero_block[3];
+};
+
+bool vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
+                                struct pipe_context *pipe,
+                                unsigned picture_width,
+                                unsigned picture_height,
+                                enum pipe_video_chroma_format chroma_format,
+                                enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
+                                enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
+                                bool pot_buffers);
+
+void vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer);
+
+void vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer *renderer,
+                                              struct pipe_texture *surface,
+                                              struct pipe_texture *past,
+                                              struct pipe_texture *future,
+                                              unsigned num_macroblocks,
+                                              struct pipe_mpeg12_macroblock *mpeg12_macroblocks,
+                                              struct pipe_fence_handle **fence);
+
+#endif /* vl_mpeg12_mc_renderer_h */
diff --git a/src/gallium/auxiliary/vl/vl_shader_build.c b/src/gallium/auxiliary/vl/vl_shader_build.c
new file mode 100644
index 0000000000..faa20a903c
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_shader_build.c
@@ -0,0 +1,242 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "vl_shader_build.h"
+#include <assert.h>
+#include <tgsi/tgsi_parse.h>
+#include <tgsi/tgsi_build.h>
+
+struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_interpolated_input
+(
+   unsigned int name,
+   unsigned int index,
+   unsigned int first,
+   unsigned int last,
+   int interpolation
+)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   assert
+   (
+      interpolation == TGSI_INTERPOLATE_CONSTANT ||
+      interpolation == TGSI_INTERPOLATE_LINEAR ||
+      interpolation == TGSI_INTERPOLATE_PERSPECTIVE
+   );
+
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.Declaration.Interpolate = interpolation;;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_CONSTANT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl.Declaration.File = TGSI_FILE_OUTPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = name;
+   decl.Semantic.SemanticIndex = index;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_TEMPORARY;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last)
+{
+   struct tgsi_full_declaration decl = tgsi_default_full_declaration();
+
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_SAMPLER;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   return decl;
+}
+
+struct tgsi_full_instruction vl_inst2
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src_file,
+   unsigned int src_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 1;
+   inst.FullSrcRegisters[0].SrcRegister.File = src_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_inst3
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_tex
+(
+   int tex,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = TGSI_OPCODE_TEX;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.InstructionExtTexture.Texture = tex;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_inst4
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index,
+   enum tgsi_file_type src3_file,
+   unsigned int src3_index
+)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = opcode;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = dst_file;
+   inst.FullDstRegisters[0].DstRegister.Index = dst_index;
+   inst.Instruction.NumSrcRegs = 3;
+   inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
+   inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
+   inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
+   inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
+   inst.FullSrcRegisters[2].SrcRegister.File = src3_file;
+   inst.FullSrcRegisters[2].SrcRegister.Index = src3_index;
+
+   return inst;
+}
+
+struct tgsi_full_instruction vl_end(void)
+{
+   struct tgsi_full_instruction inst = tgsi_default_full_instruction();
+
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+
+   return inst;
+}
diff --git a/src/gallium/auxiliary/vl/vl_shader_build.h b/src/gallium/auxiliary/vl/vl_shader_build.h
new file mode 100644
index 0000000000..5da71f8e13
--- /dev/null
+++ b/src/gallium/auxiliary/vl/vl_shader_build.h
@@ -0,0 +1,88 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 vl_shader_build_h
+#define vl_shader_build_h
+
+#include <pipe/p_shader_tokens.h>
+
+struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_interpolated_input
+(
+   unsigned int name,
+   unsigned int index,
+   unsigned int first,
+   unsigned int last,
+   int interpolation
+);
+struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last);
+struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last);
+struct tgsi_full_instruction vl_inst2
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src_file,
+   unsigned int src_index
+);
+struct tgsi_full_instruction vl_inst3
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+);
+struct tgsi_full_instruction vl_tex
+(
+   int tex,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index
+);
+struct tgsi_full_instruction vl_inst4
+(
+   int opcode,
+   enum tgsi_file_type dst_file,
+   unsigned int dst_index,
+   enum tgsi_file_type src1_file,
+   unsigned int src1_index,
+   enum tgsi_file_type src2_file,
+   unsigned int src2_index,
+   enum tgsi_file_type src3_file,
+   unsigned int src3_index
+);
+struct tgsi_full_instruction vl_end(void);
+
+#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index 9161747fdb..d185c6b849 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -41,7 +41,7 @@
 static const char *
 cell_get_vendor(struct pipe_screen *screen)
 {
-   return "Tungsten Graphics, Inc.";
+   return "VMware, Inc.";
 }
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 6a63a0e6ce..ae4c61efb3 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -389,22 +389,14 @@ cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
    const uint texWidth = pt->width[level];
    const uint texHeight = pt->height[level];
    const uint stride = ct->stride[level];
-   unsigned flags = 0x0;
    unsigned size;
 
    assert(transfer->texture);
 
-   if (transfer->usage != PIPE_TRANSFER_READ) {
-      flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
-   }
-
-   if (transfer->usage != PIPE_TRANSFER_WRITE) {
-      flags |= PIPE_BUFFER_USAGE_CPU_READ;
-   }
-
    if (!ct->mapped) {
       /* map now */
-      ct->mapped = pipe_buffer_map(screen, ct->buffer, flags);
+      ct->mapped = pipe_buffer_map(screen, ct->buffer,
+                                   pipe_transfer_buffer_flags(transfer));
    }
 
    /*
@@ -417,8 +409,7 @@ cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
    if (!ctrans->map)
       return NULL; /* out of memory */
 
-   if (transfer->usage == PIPE_TRANSFER_READ ||
-       transfer->usage == PIPE_TRANSFER_READ_WRITE) {
+   if (transfer->usage & PIPE_TRANSFER_READ) {
       /* need to untwiddle the texture to make a linear version */
       const uint bpp = pf_get_size(ct->base.format);
       if (bpp == 4) {
@@ -459,8 +450,7 @@ cell_transfer_unmap(struct pipe_screen *screen,
                                    PIPE_BUFFER_USAGE_CPU_READ);
    }
 
-   if (transfer->usage == PIPE_TRANSFER_WRITE ||
-       transfer->usage == PIPE_TRANSFER_READ_WRITE) {
+   if (transfer->usage & PIPE_TRANSFER_WRITE) {
       /* The user wrote new texture data into the mapped buffer.
        * We need to convert the new linear data into the twiddled/tiled format.
        */
diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c
index b43f735245..e745f3342d 100644
--- a/src/gallium/drivers/i915simple/i915_context.c
+++ b/src/gallium/drivers/i915simple/i915_context.c
@@ -175,12 +175,19 @@ i915_is_buffer_referenced(struct pipe_context *pipe,
 static void i915_destroy(struct pipe_context *pipe)
 {
    struct i915_context *i915 = i915_context(pipe);
+   int i;
 
    draw_destroy(i915->draw);
    
    if(i915->batch)
       i915->iws->batchbuffer_destroy(i915->batch);
 
+   /* unbind framebuffer */
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&i915->framebuffer.cbufs[i], NULL);
+   }
+   pipe_surface_reference(&i915->framebuffer.zsbuf, NULL);
+
    FREE(i915);
 }
 
diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
index b3a7774fd6..8a3e466c84 100644
--- a/src/gallium/drivers/i915simple/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
@@ -198,7 +198,7 @@ i915_vbuf_render_map_vertices(struct vbuf_render *render)
    struct intel_winsys *iws = i915->iws;
 
    if (i915->vbo_flushed)
-      debug_printf("%s bad vbo flush occured stalling on hw\n");
+      debug_printf("%s bad vbo flush occured stalling on hw\n", __FUNCTION__);
 
    i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE);
 
@@ -389,14 +389,43 @@ i915_vbuf_render_draw_arrays(struct vbuf_render *render,
                              uint nr)
 {
    struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
 
    if (i915_render->fallback) {
       draw_arrays_fallback(render, start, nr);
       return;
    }
 
-   /* JB: TODO submit direct cmds */
-   draw_arrays_fallback(render, start, nr);
+   if (i915->dirty)
+      i915_update_derived(i915);
+
+   if (i915->hardware_dirty)
+      i915_emit_hardware_state(i915);
+
+   if (!BEGIN_BATCH(2, 0)) {
+      FLUSH_BATCH(NULL);
+
+      /* Make sure state is re-emitted after a flush:
+       */
+      i915_update_derived(i915);
+      i915_emit_hardware_state(i915);
+      i915->vbo_flushed = 1;
+
+      if (!BEGIN_BATCH(2, 0)) {
+         assert(0);
+         goto out;
+      }
+   }
+
+   OUT_BATCH(_3DPRIMITIVE |
+             PRIM_INDIRECT |
+             PRIM_INDIRECT_SEQUENTIAL |
+             i915_render->hwprim |
+             nr);
+   OUT_BATCH(start); /* Beginning vertex index */
+
+out:
+   return;
 }
 
 /**
diff --git a/src/gallium/drivers/i915simple/i915_screen.c b/src/gallium/drivers/i915simple/i915_screen.c
index a1dd43c1bc..c66558c320 100644
--- a/src/gallium/drivers/i915simple/i915_screen.c
+++ b/src/gallium/drivers/i915simple/i915_screen.c
@@ -46,7 +46,7 @@
 static const char *
 i915_get_vendor(struct pipe_screen *screen)
 {
-   return "Tungsten Graphics, Inc.";
+   return "VMware, Inc.";
 }
 
 static const char *
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
index 0087dfa410..7d48e6e84d 100644
--- a/src/gallium/drivers/i915simple/i915_state.c
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -588,9 +588,17 @@ static void i915_set_framebuffer_state(struct pipe_context *pipe,
 				       const struct pipe_framebuffer_state *fb)
 {
    struct i915_context *i915 = i915_context(pipe);
+   int i;
+
    draw_flush(i915->draw);
 
-   i915->framebuffer = *fb; /* struct copy */
+   i915->framebuffer.width = fb->width;
+   i915->framebuffer.height = fb->height;
+   i915->framebuffer.nr_cbufs = fb->nr_cbufs;
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&i915->framebuffer.cbufs[i], fb->cbufs[i]);
+   }
+   pipe_surface_reference(&i915->framebuffer.zsbuf, fb->zsbuf);
 
    i915->dirty |= I915_NEW_FRAMEBUFFER;
 }
diff --git a/src/gallium/drivers/i915simple/i915_texture.c b/src/gallium/drivers/i915simple/i915_texture.c
index 6a6c654271..286c9ace8e 100644
--- a/src/gallium/drivers/i915simple/i915_texture.c
+++ b/src/gallium/drivers/i915simple/i915_texture.c
@@ -165,7 +165,7 @@ i915_scanout_layout(struct i915_texture *tex)
    struct pipe_texture *pt = &tex->base;
 
    if (pt->last_level > 0 || pt->block.size != 4)
-      return 0;
+      return FALSE;
 
    i915_miptree_set_level_info(tex, 0, 1,
                                tex->base.width[0],
@@ -191,6 +191,38 @@ i915_scanout_layout(struct i915_texture *tex)
    return TRUE;
 }
 
+/**
+ * Special case to deal with shared textures.
+ */
+static boolean
+i915_display_target_layout(struct i915_texture *tex)
+{
+   struct pipe_texture *pt = &tex->base;
+
+   if (pt->last_level > 0 || pt->block.size != 4)
+      return FALSE;
+
+   /* fallback to normal textures for small textures */
+   if (tex->base.width[0] < 240)
+      return FALSE;
+
+   i915_miptree_set_level_info(tex, 0, 1,
+                               tex->base.width[0],
+                               tex->base.height[0],
+                               1);
+   i915_miptree_set_image_offset(tex, 0, 0, 0, 0);
+
+   tex->stride = power_of_two(tex->base.nblocksx[0] * pt->block.size);
+   tex->total_nblocksy = round_up(tex->base.nblocksy[0], 8);
+   tex->hw_tiled = INTEL_TILE_X;
+
+   debug_printf("%s size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
+      tex->base.width[0], tex->base.height[0], pt->block.size,
+      tex->stride, tex->total_nblocksy, tex->stride * tex->total_nblocksy);
+
+   return TRUE;
+}
+
 static void
 i915_miptree_layout_2d(struct i915_texture *tex)
 {
@@ -201,6 +233,16 @@ i915_miptree_layout_2d(struct i915_texture *tex)
    unsigned nblocksx = pt->nblocksx[0];
    unsigned nblocksy = pt->nblocksy[0];
 
+   /* used for scanouts that need special layouts */
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+      if (i915_scanout_layout(tex))
+         return;
+
+   /* for shared buffers we use some very like scanout */
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+      if (i915_display_target_layout(tex))
+         return;
+
    tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4);
    tex->total_nblocksy = 0;
 
@@ -351,6 +393,11 @@ i945_miptree_layout_2d(struct i915_texture *tex)
       if (i915_scanout_layout(tex))
          return;
 
+   /* for shared buffers we use some very like scanout */
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+      if (i915_display_target_layout(tex))
+         return;
+
    tex->stride = round_up(pt->nblocksx[0] * pt->block.size, 4);
 
    /* May need to adjust pitch to accomodate the placement of
@@ -812,7 +859,7 @@ i915_transfer_map(struct pipe_screen *screen,
    char *map;
    boolean write = FALSE;
 
-   if (transfer->usage != PIPE_TRANSFER_READ)
+   if (transfer->usage & PIPE_TRANSFER_WRITE)
       write = TRUE;
 
    map = iws->buffer_map(iws, tex->buffer, write);
diff --git a/src/gallium/drivers/i965simple/brw_screen.c b/src/gallium/drivers/i965simple/brw_screen.c
index fb68fd624b..4a84c4db23 100644
--- a/src/gallium/drivers/i965simple/brw_screen.c
+++ b/src/gallium/drivers/i965simple/brw_screen.c
@@ -39,7 +39,7 @@
 static const char *
 brw_get_vendor( struct pipe_screen *screen )
 {
-   return "Tungsten Graphics, Inc.";
+   return "VMware, Inc.";
 }
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
index 0b115fc9b0..d27ef0de04 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
@@ -48,11 +48,13 @@
 #include "util/u_memory.h"
 #include "util/u_debug.h"
 #include "util/u_string.h"
+#include "util/u_cpu_detect.h"
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
 #include "lp_bld_intr.h"
 #include "lp_bld_logic.h"
+#include "lp_bld_debug.h"
 #include "lp_bld_arit.h"
 
 
@@ -71,30 +73,28 @@ lp_build_min_simple(struct lp_build_context *bld,
 
    /* TODO: optimize the constant case */
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
    if(type.width * type.length == 128) {
       if(type.floating) {
-         if(type.width == 32)
+         if(type.width == 32 && util_cpu_caps.has_sse)
             intrinsic = "llvm.x86.sse.min.ps";
-         if(type.width == 64)
+         if(type.width == 64 && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.min.pd";
       }
       else {
-         if(type.width == 8 && !type.sign)
+         if(type.width == 8 && !type.sign && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.pminu.b";
-         if(type.width == 8 && type.sign)
+         if(type.width == 8 && type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pminsb";
-         if(type.width == 16 && !type.sign)
+         if(type.width == 16 && !type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pminuw";
-         if(type.width == 16 && type.sign)
+         if(type.width == 16 && type.sign && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.pmins.w";
-         if(type.width == 32 && !type.sign)
+         if(type.width == 32 && !type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pminud";
-         if(type.width == 32 && type.sign)
+         if(type.width == 32 && type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pminsd";
       }
    }
-#endif
 
    if(intrinsic)
       return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
@@ -119,30 +119,28 @@ lp_build_max_simple(struct lp_build_context *bld,
 
    /* TODO: optimize the constant case */
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
    if(type.width * type.length == 128) {
       if(type.floating) {
-         if(type.width == 32)
+         if(type.width == 32 && util_cpu_caps.has_sse)
             intrinsic = "llvm.x86.sse.max.ps";
-         if(type.width == 64)
+         if(type.width == 64 && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.max.pd";
       }
       else {
-         if(type.width == 8 && !type.sign)
+         if(type.width == 8 && !type.sign && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.pmaxu.b";
-         if(type.width == 8 && type.sign)
+         if(type.width == 8 && type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pmaxsb";
-         if(type.width == 16 && !type.sign)
+         if(type.width == 16 && !type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pmaxuw";
-         if(type.width == 16 && type.sign)
+         if(type.width == 16 && type.sign && util_cpu_caps.has_sse2)
             intrinsic = "llvm.x86.sse2.pmaxs.w";
-         if(type.width == 32 && !type.sign)
+         if(type.width == 32 && !type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pmaxud";
-         if(type.width == 32 && type.sign)
+         if(type.width == 32 && type.sign && util_cpu_caps.has_sse4_1)
             intrinsic = "llvm.x86.sse41.pmaxsd";
       }
    }
-#endif
 
    if(intrinsic)
       return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
@@ -204,15 +202,14 @@ lp_build_add(struct lp_build_context *bld,
       if(a == bld->one || b == bld->one)
         return bld->one;
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-      if(type.width * type.length == 128 &&
+      if(util_cpu_caps.has_sse2 &&
+         type.width * type.length == 128 &&
          !type.floating && !type.fixed) {
          if(type.width == 8)
             intrinsic = type.sign ? "llvm.x86.sse2.padds.b" : "llvm.x86.sse2.paddus.b";
          if(type.width == 16)
             intrinsic = type.sign ? "llvm.x86.sse2.padds.w" : "llvm.x86.sse2.paddus.w";
       }
-#endif
    
       if(intrinsic)
          return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
@@ -257,15 +254,14 @@ lp_build_sub(struct lp_build_context *bld,
       if(b == bld->one)
         return bld->zero;
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-      if(type.width * type.length == 128 &&
+      if(util_cpu_caps.has_sse2 &&
+         type.width * type.length == 128 &&
          !type.floating && !type.fixed) {
          if(type.width == 8)
             intrinsic = type.sign ? "llvm.x86.sse2.psubs.b" : "llvm.x86.sse2.psubus.b";
          if(type.width == 16)
             intrinsic = type.sign ? "llvm.x86.sse2.psubs.w" : "llvm.x86.sse2.psubus.w";
       }
-#endif
    
       if(intrinsic)
          return lp_build_intrinsic_binary(bld->builder, intrinsic, lp_build_vec_type(bld->type), a, b);
@@ -419,8 +415,7 @@ lp_build_mul(struct lp_build_context *bld,
       return bld->undef;
 
    if(!type.floating && !type.fixed && type.norm) {
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-      if(type.width == 8 && type.length == 16) {
+      if(util_cpu_caps.has_sse2 && type.width == 8 && type.length == 16) {
          LLVMTypeRef i16x8 = LLVMVectorType(LLVMInt16Type(), 8);
          LLVMTypeRef i8x16 = LLVMVectorType(LLVMInt8Type(), 16);
          static LLVMValueRef ml = NULL;
@@ -456,7 +451,6 @@ lp_build_mul(struct lp_build_context *bld,
          
          return ab;
       }
-#endif
 
       /* FIXME */
       assert(0);
@@ -493,10 +487,8 @@ lp_build_div(struct lp_build_context *bld,
    if(LLVMIsConstant(a) && LLVMIsConstant(b))
       return LLVMConstFDiv(a, b);
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(type.width == 32 && type.length == 4)
+   if(util_cpu_caps.has_sse && type.width == 32 && type.length == 4)
       return lp_build_mul(bld, a, lp_build_rcp(bld, b));
-#endif
 
    return LLVMBuildFDiv(bld->builder, a, b, "");
 }
@@ -606,8 +598,7 @@ lp_build_abs(struct lp_build_context *bld,
       return a;
    }
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(type.width*type.length == 128) {
+   if(type.width*type.length == 128 && util_cpu_caps.has_ssse3) {
       switch(type.width) {
       case 8:
          return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a);
@@ -617,7 +608,6 @@ lp_build_abs(struct lp_build_context *bld,
          return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a);
       }
    }
-#endif
 
    return lp_build_max(bld, a, LLVMBuildNeg(bld->builder, a, ""));
 }
@@ -684,6 +674,8 @@ lp_build_round_sse41(struct lp_build_context *bld,
 
    assert(type.floating);
    assert(type.width*type.length == 128);
+   assert(lp_check_value(type, a));
+   assert(util_cpu_caps.has_sse4_1);
 
    switch(type.width) {
    case 32:
@@ -703,20 +695,45 @@ lp_build_round_sse41(struct lp_build_context *bld,
 
 
 LLVMValueRef
-lp_build_round(struct lp_build_context *bld,
+lp_build_trunc(struct lp_build_context *bld,
                LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
 
    assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   if(util_cpu_caps.has_sse4_1)
+      return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
+   else {
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMValueRef res;
+      res = LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      return res;
+   }
+}
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
-#endif
 
-   /* FIXME */
-   assert(0);
-   return bld->undef;
+LLVMValueRef
+lp_build_round(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const struct lp_type type = bld->type;
+
+   assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   if(util_cpu_caps.has_sse4_1)
+      return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
+   else {
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMValueRef res;
+      res = lp_build_iround(bld, a);
+      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      return res;
+   }
 }
 
 
@@ -728,13 +745,15 @@ lp_build_floor(struct lp_build_context *bld,
 
    assert(type.floating);
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
-#endif
-
-   /* FIXME */
-   assert(0);
-   return bld->undef;
+   if(util_cpu_caps.has_sse4_1)
+      return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
+   else {
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMValueRef res;
+      res = lp_build_ifloor(bld, a);
+      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      return res;
+   }
 }
 
 
@@ -745,59 +764,143 @@ lp_build_ceil(struct lp_build_context *bld,
    const struct lp_type type = bld->type;
 
    assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   if(util_cpu_caps.has_sse4_1)
+      return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
+   else {
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMValueRef res;
+      res = lp_build_iceil(bld, a);
+      res = LLVMBuildSIToFP(bld->builder, res, vec_type, "");
+      return res;
+   }
+}
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
-#endif
 
-   /* FIXME */
-   assert(0);
-   return bld->undef;
+/**
+ * Convert to integer, through whichever rounding method that's fastest,
+ * typically truncating to zero.
+ */
+LLVMValueRef
+lp_build_itrunc(struct lp_build_context *bld,
+                LLVMValueRef a)
+{
+   const struct lp_type type = bld->type;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+
+   assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
 }
 
 
 LLVMValueRef
-lp_build_trunc(struct lp_build_context *bld,
-               LLVMValueRef a)
+lp_build_iround(struct lp_build_context *bld,
+                LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMValueRef res;
 
    assert(type.floating);
+   assert(lp_check_value(type, a));
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
-#endif
+   if(util_cpu_caps.has_sse4_1) {
+      res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
+   }
+   else {
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+      LLVMValueRef sign;
+      LLVMValueRef half;
+
+      /* get sign bit */
+      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+
+      /* sign * 0.5 */
+      half = lp_build_const_scalar(type, 0.5);
+      half = LLVMBuildBitCast(bld->builder, half, int_vec_type, "");
+      half = LLVMBuildOr(bld->builder, sign, half, "");
+      half = LLVMBuildBitCast(bld->builder, half, vec_type, "");
 
-   /* FIXME */
-   assert(0);
-   return bld->undef;
+      res = LLVMBuildAdd(bld->builder, a, half, "");
+   }
+
+   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "");
+
+   return res;
 }
 
 
-/**
- * Convert to integer, through whichever rounding method that's fastest,
- * typically truncating to zero.
- */
 LLVMValueRef
-lp_build_int(struct lp_build_context *bld,
-             LLVMValueRef a)
+lp_build_ifloor(struct lp_build_context *bld,
+                LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
    LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMValueRef res;
 
    assert(type.floating);
+   assert(lp_check_value(type, a));
 
-   return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+   if(util_cpu_caps.has_sse4_1) {
+      res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
+   }
+   else {
+      /* Take the sign bit and add it to 1 constant */
+      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      unsigned mantissa = lp_mantissa(type);
+      LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+      LLVMValueRef sign;
+      LLVMValueRef offset;
+
+      /* sign = a < 0 ? ~0 : 0 */
+      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+      sign = LLVMBuildAShr(bld->builder, sign, lp_build_int_const_scalar(type, type.width - 1), "");
+
+      /* offset = -0.99999(9)f */
+      offset = lp_build_const_scalar(type, -(double)(((unsigned long long)1 << mantissa) - 1)/((unsigned long long)1 << mantissa));
+      offset = LLVMConstBitCast(offset, int_vec_type);
+
+      /* offset = a < 0 ? -0.99999(9)f : 0.0f */
+      offset = LLVMBuildAnd(bld->builder, offset, sign, "");
+      offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "");
+
+      res = LLVMBuildAdd(bld->builder, a, offset, "");
+   }
+
+   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "");
+
+   return res;
 }
 
 
 LLVMValueRef
-lp_build_ifloor(struct lp_build_context *bld,
-                LLVMValueRef a)
+lp_build_iceil(struct lp_build_context *bld,
+               LLVMValueRef a)
 {
-   a = lp_build_floor(bld, a);
-   a = lp_build_int(bld, a);
-   return a;
+   const struct lp_type type = bld->type;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMValueRef res;
+
+   assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   if(util_cpu_caps.has_sse4_1) {
+      res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
+   }
+   else {
+      assert(0);
+      res = bld->undef;
+   }
+
+   res = LLVMBuildFPToSI(bld->builder, res, int_vec_type, "");
+
+   return res;
 }
 
 
@@ -837,11 +940,9 @@ lp_build_rcp(struct lp_build_context *bld,
    if(LLVMIsConstant(a))
       return LLVMConstFDiv(bld->one, a);
 
-   /* XXX: is this really necessary? */
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(type.width == 32 && type.length == 4)
+   if(util_cpu_caps.has_sse && type.width == 32 && type.length == 4)
+      /* FIXME: improve precision */
       return lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rcp.ps", lp_build_vec_type(type), a);
-#endif
 
    return LLVMBuildFDiv(bld->builder, bld->one, a, "");
 }
@@ -858,11 +959,8 @@ lp_build_rsqrt(struct lp_build_context *bld,
 
    assert(type.floating);
 
-   /* XXX: is this really necessary? */
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(type.width == 32 && type.length == 4)
+   if(util_cpu_caps.has_sse && type.width == 32 && type.length == 4)
       return lp_build_intrinsic_unary(bld->builder, "llvm.x86.sse.rsqrt.ps", lp_build_vec_type(type), a);
-#endif
 
    return lp_build_rcp(bld, lp_build_sqrt(bld, a));
 }
@@ -918,7 +1016,8 @@ lp_build_pow(struct lp_build_context *bld,
 {
    /* TODO: optimize the constant case */
    if(LLVMIsConstant(x) && LLVMIsConstant(y))
-      debug_printf("%s: inefficient/imprecise constant arithmetic\n");
+      debug_printf("%s: inefficient/imprecise constant arithmetic\n",
+                   __FUNCTION__);
 
    return lp_build_exp2(bld, lp_build_mul(bld, lp_build_log2(bld, x), y));
 }
@@ -972,7 +1071,8 @@ lp_build_polynomial(struct lp_build_context *bld,
 
    /* TODO: optimize the constant case */
    if(LLVMIsConstant(x))
-      debug_printf("%s: inefficient/imprecise constant arithmetic\n");
+      debug_printf("%s: inefficient/imprecise constant arithmetic\n",
+                   __FUNCTION__);
 
    for (i = num_coeffs; i--; ) {
       LLVMValueRef coeff = lp_build_const_scalar(type, coeffs[i]);
@@ -1026,7 +1126,8 @@ lp_build_exp2_approx(struct lp_build_context *bld,
    if(p_exp2_int_part || p_frac_part || p_exp2) {
       /* TODO: optimize the constant case */
       if(LLVMIsConstant(x))
-         debug_printf("%s: inefficient/imprecise constant arithmetic\n");
+         debug_printf("%s: inefficient/imprecise constant arithmetic\n",
+                      __FUNCTION__);
 
       assert(type.floating && type.width == 32);
 
@@ -1125,7 +1226,8 @@ lp_build_log2_approx(struct lp_build_context *bld,
    if(p_exp || p_floor_log2 || p_log2) {
       /* TODO: optimize the constant case */
       if(LLVMIsConstant(x))
-         debug_printf("%s: inefficient/imprecise constant arithmetic\n");
+         debug_printf("%s: inefficient/imprecise constant arithmetic\n",
+                      __FUNCTION__);
 
       assert(type.floating && type.width == 32);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.h b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
index d68a97c4b8..095a8e1cab 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
@@ -126,11 +126,18 @@ lp_build_trunc(struct lp_build_context *bld,
                LLVMValueRef a);
 
 LLVMValueRef
-lp_build_int(struct lp_build_context *bld,
-             LLVMValueRef a);
+lp_build_ifloor(struct lp_build_context *bld,
+                LLVMValueRef a);
+LLVMValueRef
+lp_build_iceil(struct lp_build_context *bld,
+               LLVMValueRef a);
 
 LLVMValueRef
-lp_build_ifloor(struct lp_build_context *bld,
+lp_build_iround(struct lp_build_context *bld,
+                LLVMValueRef a);
+
+LLVMValueRef
+lp_build_itrunc(struct lp_build_context *bld,
                 LLVMValueRef a);
 
 LLVMValueRef
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.c b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
index 186cac70f6..20c8710214 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
@@ -63,6 +63,7 @@
 
 #include "util/u_debug.h"
 #include "util/u_math.h"
+#include "util/u_cpu_detect.h"
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
@@ -334,8 +335,7 @@ lp_build_pack2(LLVMBuilderRef builder,
    assert(!src_type.floating);
    assert(!dst_type.floating);
 
-#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(src_type.width * src_type.length == 128) {
+   if(util_cpu_caps.has_sse2 && src_type.width * src_type.length == 128) {
       /* All X86 non-interleaved pack instructions all take signed inputs and
        * saturate them, so saturate beforehand. */
       if(!src_type.sign && !clamped) {
@@ -349,7 +349,7 @@ lp_build_pack2(LLVMBuilderRef builder,
 
       switch(src_type.width) {
       case 32:
-         if(dst_type.sign)
+         if(dst_type.sign || !util_cpu_caps.has_sse4_1)
             res = lp_build_intrinsic_binary(builder, "llvm.x86.sse2.packssdw.128", src_vec_type, lo, hi);
          else
             /* PACKUSDW is the only instrinsic with a consistent signature */
@@ -372,7 +372,6 @@ lp_build_pack2(LLVMBuilderRef builder,
       res = LLVMBuildBitCast(builder, res, dst_vec_type, "");
       return res;
    }
-#endif
 
    lo = LLVMBuildBitCast(builder, lo, dst_vec_type, "");
    hi = LLVMBuildBitCast(builder, hi, dst_vec_type, "");
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
index 21c665c4d4..98ec1cb1b9 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
@@ -210,7 +210,4 @@ lp_build_depth_test(LLVMBuilderRef builder,
       dst = lp_build_select(&bld, z_bitmask, src, dst);
       LLVMBuildStore(builder, dst, dst_ptr);
    }
-
-   /* FIXME */
-   assert(!state->occlusion_count);
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.c b/src/gallium/drivers/llvmpipe/lp_bld_logic.c
index 6b6f820769..db22a8028a 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_logic.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.c
@@ -33,6 +33,8 @@
  */
 
 
+#include "util/u_cpu_detect.h"
+
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
 #include "lp_bld_intr.h"
@@ -65,7 +67,7 @@ lp_build_cmp(struct lp_build_context *bld,
 
 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
    if(type.width * type.length == 128) {
-      if(type.floating) {
+      if(type.floating && util_cpu_caps.has_sse) {
          LLVMValueRef args[3];
          unsigned cc;
          boolean swap;
@@ -114,7 +116,7 @@ lp_build_cmp(struct lp_build_context *bld,
          res = LLVMBuildBitCast(bld->builder, res, int_vec_type, "");
          return res;
       }
-      else {
+      else if(util_cpu_caps.has_sse2) {
          static const struct {
             unsigned swap:1;
             unsigned eq:1;
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c
index 8ca1be6f1b..1a47ca32d2 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c
@@ -274,8 +274,8 @@ lp_build_sample_2d_linear_soa(struct lp_build_sample_context *bld,
    s_fpart = lp_build_sub(&bld->coord_bld, s, s_ipart);
    t_fpart = lp_build_sub(&bld->coord_bld, t, t_ipart);
 
-   x0 = lp_build_int(&bld->coord_bld, s_ipart);
-   y0 = lp_build_int(&bld->coord_bld, t_ipart);
+   x0 = lp_build_itrunc(&bld->coord_bld, s_ipart);
+   y0 = lp_build_itrunc(&bld->coord_bld, t_ipart);
 
    x0 = lp_build_sample_wrap(bld, x0, width,  bld->static_state->pot_width,  bld->static_state->wrap_s);
    y0 = lp_build_sample_wrap(bld, y0, height, bld->static_state->pot_height, bld->static_state->wrap_t);
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index a4b2bd8c2a..202cb8ef43 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -107,11 +107,16 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
    if (llvmpipe->draw)
       draw_destroy( llvmpipe->draw );
 
-   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++)
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       lp_destroy_tile_cache(llvmpipe->cbuf_cache[i]);
+      pipe_surface_reference(&llvmpipe->framebuffer.cbufs[i], NULL);
+   }
+   pipe_surface_reference(&llvmpipe->framebuffer.zsbuf, NULL);
 
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
       lp_destroy_tex_tile_cache(llvmpipe->tex_cache[i]);
+      pipe_texture_reference(&llvmpipe->texture[i], NULL);
+   }
 
    for (i = 0; i < Elements(llvmpipe->constants); i++) {
       if (llvmpipe->constants[i].buffer) {
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index b4a22ff4a9..1126bf90b9 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -36,6 +36,7 @@
 #include <llvm-c/Transforms/Scalar.h>
 
 #include "util/u_memory.h"
+#include "util/u_cpu_detect.h"
 #include "lp_screen.h"
 #include "lp_bld_intr.h"
 #include "lp_jit.h"
@@ -147,6 +148,19 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
 {
    char *error = NULL;
 
+   util_cpu_detect();
+
+#if 0
+   /* For simulating less capable machines */
+   util_cpu_caps.has_sse3 = 0;
+   util_cpu_caps.has_sse4_1 = 0;
+#endif
+
+#ifdef LLVM_NATIVE_ARCH
+   LLVMLinkInJIT();
+   LLVMInitializeNativeTarget();
+#endif
+
    screen->module = LLVMModuleCreateWithName("llvmpipe");
 
    screen->provider = LLVMCreateModuleProviderForExistingModule(screen->module);
@@ -163,8 +177,15 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
    LLVMAddTargetData(screen->target, screen->pass);
    /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
     * but there are more on SVN. */
+   /* TODO: Add more passes */
    LLVMAddConstantPropagationPass(screen->pass);
-   LLVMAddInstructionCombiningPass(screen->pass);
+   if(util_cpu_caps.has_sse4_1) {
+      /* FIXME: There is a bug in this pass, whereby the combination of fptosi
+       * and sitofp (necessary for trunc/floor/ceil/round implementation)
+       * somehow becomes invalid code.
+       */
+      LLVMAddInstructionCombiningPass(screen->pass);
+   }
    LLVMAddPromoteMemoryToRegisterPass(screen->pass);
    LLVMAddGVNPass(screen->pass);
    LLVMAddCFGSimplificationPass(screen->pass);
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index ff7ef8658a..0518927458 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -82,7 +82,7 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
       return 13; /* max 4Kx4K */
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return 9;  /* max 256x256x256 */
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
       return 13; /* max 4Kx4K */
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 9faed5a0b1..b00be0cc32 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -401,7 +401,6 @@ generate_fragment(struct llvmpipe_context *lp,
    if(key->depth.enabled) {
       debug_printf("depth.func = %s\n", debug_dump_func(key->depth.func, TRUE));
       debug_printf("depth.writemask = %u\n", key->depth.writemask);
-      debug_printf("depth.occlusion_count = %u\n", key->depth.occlusion_count);
    }
    if(key->alpha.enabled) {
       debug_printf("alpha.func = %s\n", debug_dump_func(key->alpha.func, TRUE));
@@ -582,6 +581,11 @@ generate_fragment(struct llvmpipe_context *lp,
     * Translate the LLVM IR into machine code.
     */
 
+   if(LLVMVerifyFunction(variant->function, LLVMPrintMessageAction)) {
+      LLVMDumpValue(variant->function);
+      abort();
+   }
+
    LLVMRunFunctionPassManager(screen->pass, variant->function);
 
 #ifdef DEBUG
@@ -589,11 +593,6 @@ generate_fragment(struct llvmpipe_context *lp,
    debug_printf("\n");
 #endif
 
-   if(LLVMVerifyFunction(variant->function, LLVMPrintMessageAction)) {
-      LLVMDumpValue(variant->function);
-      abort();
-   }
-
    variant->jit_function = (lp_jit_frag_func)LLVMGetPointerToGlobal(screen->engine, variant->function);
 
 #ifdef DEBUG
diff --git a/src/gallium/drivers/llvmpipe/lp_state_surface.c b/src/gallium/drivers/llvmpipe/lp_state_surface.c
index 177a26b7b1..2c29144c03 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_surface.c
@@ -56,7 +56,7 @@ llvmpipe_set_framebuffer_state(struct pipe_context *pipe,
          lp_flush_tile_cache(lp->cbuf_cache[i]);
 
          /* assign new */
-         lp->framebuffer.cbufs[i] = fb->cbufs[i];
+         pipe_surface_reference(&lp->framebuffer.cbufs[i], fb->cbufs[i]);
 
          /* update cache */
          lp_tile_cache_set_surface(lp->cbuf_cache[i], fb->cbufs[i]);
@@ -81,7 +81,7 @@ llvmpipe_set_framebuffer_state(struct pipe_context *pipe,
       }
 
       /* assign new */
-      lp->framebuffer.zsbuf = fb->zsbuf;
+      pipe_surface_reference(&lp->framebuffer.zsbuf, fb->zsbuf);
 
       /* Tell draw module how deep the Z/depth buffer is */
       if (lp->framebuffer.zsbuf) {
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index d8455e5649..7d83f899e6 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -264,6 +264,11 @@ int main(int argc, char **argv)
    unsigned i;
    int ret;
 
+#ifdef LLVM_NATIVE_ARCH
+   LLVMLinkInJIT();
+   LLVMInitializeNativeTarget();
+#endif
+
    for (i = 0; i < sizeof(test_cases)/sizeof(test_cases[0]); ++i)
       if(!test_format(&test_cases[i]))
         ret = 1;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c
index 4592dc0b2d..f07fa256f1 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_main.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_main.c
@@ -365,6 +365,11 @@ int main(int argc, char **argv)
          n = atoi(argv[i]);
    }
 
+#ifdef LLVM_NATIVE_ARCH
+   LLVMLinkInJIT();
+   LLVMInitializeNativeTarget();
+#endif
+
    if(fp) {
       /* Warm up the caches */
       test_some(0, NULL, 100);
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index 724d437833..08f0950d47 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -353,17 +353,9 @@ llvmpipe_transfer_map( struct pipe_screen *_screen,
 
    if(lpt->dt) {
       struct llvmpipe_winsys *winsys = screen->winsys;
-      unsigned flags = 0;
 
-      if (transfer->usage != PIPE_TRANSFER_READ) {
-         flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
-      }
-
-      if (transfer->usage != PIPE_TRANSFER_WRITE) {
-         flags |= PIPE_BUFFER_USAGE_CPU_READ;
-      }
-
-      map = winsys->displaytarget_map(winsys, lpt->dt, flags);
+      map = winsys->displaytarget_map(winsys, lpt->dt,
+                                      pipe_transfer_buffer_flags(transfer));
       if (map == NULL)
          return NULL;
    }
@@ -373,7 +365,7 @@ llvmpipe_transfer_map( struct pipe_screen *_screen,
    /* May want to different things here depending on read/write nature
     * of the map:
     */
-   if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) 
+   if (transfer->texture && (transfer->usage & PIPE_TRANSFER_WRITE))
    {
       /* Do something to notify sharing contexts of a texture change.
        * In llvmpipe, that would mean flushing the texture cache.
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_cache.c b/src/gallium/drivers/llvmpipe/lp_tile_cache.c
index 2e576e6039..0c06b659a1 100644
--- a/src/gallium/drivers/llvmpipe/lp_tile_cache.c
+++ b/src/gallium/drivers/llvmpipe/lp_tile_cache.c
@@ -44,10 +44,53 @@
 #include "lp_tile_cache.h"
 
 
+#define MAX_WIDTH 4096
+#define MAX_HEIGHT 4096
+
+
+enum llvmpipe_tile_status
+{
+   LP_TILE_STATUS_UNDEFINED = 0,
+   LP_TILE_STATUS_CLEAR = 1,
+   LP_TILE_STATUS_DEFINED = 2
+};
+
+
+struct llvmpipe_cached_tile
+{
+   enum llvmpipe_tile_status status;
+
+   /** color in SOA format */
+   uint8_t *color;
+};
+
+
+struct llvmpipe_tile_cache
+{
+   struct pipe_screen *screen;
+   struct pipe_surface *surface;  /**< the surface we're caching */
+   struct pipe_transfer *transfer;
+   void *transfer_map;
+
+   struct llvmpipe_cached_tile entries[MAX_WIDTH/TILE_SIZE][MAX_HEIGHT/TILE_SIZE];
+
+   uint8_t clear_color[4];  /**< for color bufs */
+   uint clear_val;        /**< for z+stencil, or packed color clear value */
+
+   struct llvmpipe_cached_tile *last_tile;  /**< most recently retrieved tile */
+};
+
+
 struct llvmpipe_tile_cache *
 lp_create_tile_cache( struct pipe_screen *screen )
 {
    struct llvmpipe_tile_cache *tc;
+   int maxLevels, maxTexSize;
+
+   /* sanity checking: max sure MAX_WIDTH/HEIGHT >= largest texture image */
+   maxLevels = screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS);
+   maxTexSize = 1 << (maxLevels - 1);
+   assert(MAX_WIDTH >= maxTexSize);
 
    tc = CALLOC_STRUCT( llvmpipe_tile_cache );
    if(!tc)
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_cache.h b/src/gallium/drivers/llvmpipe/lp_tile_cache.h
index 6d8ba5ece7..161bab3799 100644
--- a/src/gallium/drivers/llvmpipe/lp_tile_cache.h
+++ b/src/gallium/drivers/llvmpipe/lp_tile_cache.h
@@ -33,42 +33,7 @@
 #include "lp_tile_soa.h"
 
 
-enum llvmpipe_tile_status
-{
-   LP_TILE_STATUS_UNDEFINED = 0,
-   LP_TILE_STATUS_CLEAR = 1,
-   LP_TILE_STATUS_DEFINED = 2
-};
-
-
-struct llvmpipe_cached_tile
-{
-   enum llvmpipe_tile_status status;
-
-   /** color in SOA format */
-   uint8_t *color;
-};
-
-
-/** XXX move these */
-#define MAX_WIDTH 2048
-#define MAX_HEIGHT 2048
-
-
-struct llvmpipe_tile_cache
-{
-   struct pipe_screen *screen;
-   struct pipe_surface *surface;  /**< the surface we're caching */
-   struct pipe_transfer *transfer;
-   void *transfer_map;
-
-   struct llvmpipe_cached_tile entries[MAX_WIDTH/TILE_SIZE][MAX_HEIGHT/TILE_SIZE];
-
-   uint8_t clear_color[4];  /**< for color bufs */
-   uint clear_val;        /**< for z+stencil, or packed color clear value */
-
-   struct llvmpipe_cached_tile *last_tile;  /**< most recently retrieved tile */
-};
+struct llvmpipe_tile_cache;  /* opaque */
 
 
 extern struct llvmpipe_tile_cache *
diff --git a/src/gallium/drivers/nv04/nv04_transfer.c b/src/gallium/drivers/nv04/nv04_transfer.c
index 854b855d64..6618660743 100644
--- a/src/gallium/drivers/nv04/nv04_transfer.c
+++ b/src/gallium/drivers/nv04/nv04_transfer.c
@@ -13,22 +13,6 @@ struct nv04_transfer {
 	bool direct;
 };
 
-static unsigned nv04_usage_tx_to_buf(unsigned tx_usage)
-{
-	switch (tx_usage) {
-		case PIPE_TRANSFER_READ:
-			return PIPE_BUFFER_USAGE_CPU_READ;
-		case PIPE_TRANSFER_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_WRITE;
-		case PIPE_TRANSFER_READ_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-		default:
-			assert(0);
-	}
-
-	return -1;
-}
-
 static void
 nv04_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
                              struct pipe_texture *template)
@@ -86,7 +70,7 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		tx->direct = true;
 		tx->surface = pscreen->get_tex_surface(pscreen, pt,
 	                                               0, 0, 0,
-	                                               nv04_usage_tx_to_buf(usage));
+	                                               pipe_transfer_buffer_flags(&tx->base));
 		return &tx->base;
 	}
 
@@ -103,7 +87,7 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       face, level, zslice,
-	                                       nv04_usage_tx_to_buf(usage));
+	                                       pipe_transfer_buffer_flags(&tx->base));
 
 	pipe_texture_reference(&tx_tex, NULL);
 
@@ -114,7 +98,7 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		struct nv04_screen *nvscreen = nv04_screen(pscreen);
 		struct pipe_surface *src;
 
@@ -140,7 +124,7 @@ nv04_transfer_del(struct pipe_transfer *ptx)
 {
 	struct nv04_transfer *tx = (struct nv04_transfer *)ptx;
 
-	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+	if (!tx->direct && (ptx->usage & PIPE_TRANSFER_WRITE)) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		struct nv04_screen *nvscreen = nv04_screen(pscreen);
 		struct pipe_surface *dst;
@@ -170,7 +154,7 @@ nv04_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
 	struct nv04_miptree *mt = (struct nv04_miptree *)tx->surface->texture;
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
-	                            nv04_usage_tx_to_buf(ptx->usage));
+	                            pipe_transfer_buffer_flags(ptx));
 
 	return map + ns->base.offset +
 	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
diff --git a/src/gallium/drivers/nv10/nv10_transfer.c b/src/gallium/drivers/nv10/nv10_transfer.c
index c06b8d34c7..8feb85e4bd 100644
--- a/src/gallium/drivers/nv10/nv10_transfer.c
+++ b/src/gallium/drivers/nv10/nv10_transfer.c
@@ -13,22 +13,6 @@ struct nv10_transfer {
 	bool direct;
 };
 
-static unsigned nv10_usage_tx_to_buf(unsigned tx_usage)
-{
-	switch (tx_usage) {
-		case PIPE_TRANSFER_READ:
-			return PIPE_BUFFER_USAGE_CPU_READ;
-		case PIPE_TRANSFER_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_WRITE;
-		case PIPE_TRANSFER_READ_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-		default:
-			assert(0);
-	}
-
-	return -1;
-}
-
 static void
 nv10_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
                              struct pipe_texture *template)
@@ -86,7 +70,7 @@ nv10_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		tx->direct = true;
 		tx->surface = pscreen->get_tex_surface(pscreen, pt,
 	                                               0, 0, 0,
-	                                               nv10_usage_tx_to_buf(usage));
+	                                               pipe_transfer_buffer_flags(&tx->base));
 		return &tx->base;
 	}
 
@@ -103,7 +87,7 @@ nv10_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       face, level, zslice,
-	                                       nv10_usage_tx_to_buf(usage));
+	                                       pipe_transfer_buffer_flags(&tx->base));
 
 	pipe_texture_reference(&tx_tex, NULL);
 
@@ -114,7 +98,7 @@ nv10_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		struct nv10_screen *nvscreen = nv10_screen(pscreen);
 		struct pipe_surface *src;
 
@@ -140,7 +124,7 @@ nv10_transfer_del(struct pipe_transfer *ptx)
 {
 	struct nv10_transfer *tx = (struct nv10_transfer *)ptx;
 
-	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+	if (!tx->direct && (ptx->usage & PIPE_TRANSFER_WRITE)) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		struct nv10_screen *nvscreen = nv10_screen(pscreen);
 		struct pipe_surface *dst;
@@ -170,7 +154,7 @@ nv10_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
 	struct nv10_miptree *mt = (struct nv10_miptree *)tx->surface->texture;
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
-	                            nv10_usage_tx_to_buf(ptx->usage));
+	                            pipe_transfer_buffer_flags(ptx));
 
 	return map + ns->base.offset +
 	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
diff --git a/src/gallium/drivers/nv20/nv20_transfer.c b/src/gallium/drivers/nv20/nv20_transfer.c
index 5018995596..81b4f1a917 100644
--- a/src/gallium/drivers/nv20/nv20_transfer.c
+++ b/src/gallium/drivers/nv20/nv20_transfer.c
@@ -13,22 +13,6 @@ struct nv20_transfer {
 	bool direct;
 };
 
-static unsigned nv20_usage_tx_to_buf(unsigned tx_usage)
-{
-	switch (tx_usage) {
-		case PIPE_TRANSFER_READ:
-			return PIPE_BUFFER_USAGE_CPU_READ;
-		case PIPE_TRANSFER_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_WRITE;
-		case PIPE_TRANSFER_READ_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-		default:
-			assert(0);
-	}
-
-	return -1;
-}
-
 static void
 nv20_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
                              struct pipe_texture *template)
@@ -86,7 +70,7 @@ nv20_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		tx->direct = true;
 		tx->surface = pscreen->get_tex_surface(pscreen, pt,
 	                                               0, 0, 0,
-	                                               nv20_usage_tx_to_buf(usage));
+	                                               pipe_transfer_buffer_flags(&tx->base));
 		return &tx->base;
 	}
 
@@ -103,7 +87,7 @@ nv20_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       face, level, zslice,
-	                                       nv20_usage_tx_to_buf(usage));
+	                                       pipe_transfer_buffer_flags(&tx->base));
 
 	pipe_texture_reference(&tx_tex, NULL);
 
@@ -114,7 +98,7 @@ nv20_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		struct nv20_screen *nvscreen = nv20_screen(pscreen);
 		struct pipe_surface *src;
 
@@ -140,7 +124,7 @@ nv20_transfer_del(struct pipe_transfer *ptx)
 {
 	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
 
-	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+	if (!tx->direct && (ptx->usage = PIPE_TRANSFER_WRITE)) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		struct nv20_screen *nvscreen = nv20_screen(pscreen);
 		struct pipe_surface *dst;
@@ -170,7 +154,7 @@ nv20_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
 	struct nv20_miptree *mt = (struct nv20_miptree *)tx->surface->texture;
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
-	                            nv20_usage_tx_to_buf(ptx->usage));
+	                            pipe_transfer_buffer_flags(ptx));
 
 	return map + ns->base.offset +
 	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
diff --git a/src/gallium/drivers/nv30/nv30_transfer.c b/src/gallium/drivers/nv30/nv30_transfer.c
index 2367571878..98011decf7 100644
--- a/src/gallium/drivers/nv30/nv30_transfer.c
+++ b/src/gallium/drivers/nv30/nv30_transfer.c
@@ -13,22 +13,6 @@ struct nv30_transfer {
 	bool direct;
 };
 
-static unsigned nv30_usage_tx_to_buf(unsigned tx_usage)
-{
-	switch (tx_usage) {
-		case PIPE_TRANSFER_READ:
-			return PIPE_BUFFER_USAGE_CPU_READ;
-		case PIPE_TRANSFER_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_WRITE;
-		case PIPE_TRANSFER_READ_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-		default:
-			assert(0);
-	}
-
-	return -1;
-}
-
 static void
 nv30_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
                              struct pipe_texture *template)
@@ -86,7 +70,7 @@ nv30_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		tx->direct = true;
 		tx->surface = pscreen->get_tex_surface(pscreen, pt,
 	                                               face, level, zslice,
-	                                               nv30_usage_tx_to_buf(usage));
+	                                               pipe_transfer_buffer_flags(&tx->base));
 		return &tx->base;
 	}
 
@@ -103,7 +87,7 @@ nv30_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       0, 0, 0,
-	                                       nv30_usage_tx_to_buf(usage));
+	                                       pipe_transfer_buffer_flags(&tx->base));
 
 	pipe_texture_reference(&tx_tex, NULL);
 
@@ -114,7 +98,7 @@ nv30_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		struct nv30_screen *nvscreen = nv30_screen(pscreen);
 		struct pipe_surface *src;
 
@@ -140,7 +124,7 @@ nv30_transfer_del(struct pipe_transfer *ptx)
 {
 	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
 
-	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+	if (!tx->direct && (ptx->usage & PIPE_TRANSFER_WRITE)) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		struct nv30_screen *nvscreen = nv30_screen(pscreen);
 		struct pipe_surface *dst;
@@ -170,7 +154,7 @@ nv30_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
 	struct nv30_miptree *mt = (struct nv30_miptree *)tx->surface->texture;
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
-	                            nv30_usage_tx_to_buf(ptx->usage));
+	                            pipe_transfer_buffer_flags(ptx));
 
 	return map + ns->base.offset +
 	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
diff --git a/src/gallium/drivers/nv40/nv40_transfer.c b/src/gallium/drivers/nv40/nv40_transfer.c
index 6d92ac3db9..92caee6f38 100644
--- a/src/gallium/drivers/nv40/nv40_transfer.c
+++ b/src/gallium/drivers/nv40/nv40_transfer.c
@@ -13,22 +13,6 @@ struct nv40_transfer {
 	bool direct;
 };
 
-static unsigned nv40_usage_tx_to_buf(unsigned tx_usage)
-{
-	switch (tx_usage) {
-		case PIPE_TRANSFER_READ:
-			return PIPE_BUFFER_USAGE_CPU_READ;
-		case PIPE_TRANSFER_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_WRITE;
-		case PIPE_TRANSFER_READ_WRITE:
-			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
-		default:
-			assert(0);
-	}
-
-	return -1;
-}
-
 static void
 nv40_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
                              struct pipe_texture *template)
@@ -86,7 +70,7 @@ nv40_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		tx->direct = true;
 		tx->surface = pscreen->get_tex_surface(pscreen, pt,
 	                                               face, level, zslice,
-	                                               nv40_usage_tx_to_buf(usage));
+	                                               pipe_transfer_buffer_flags(&tx->base));
 		return &tx->base;
 	}
 
@@ -103,7 +87,7 @@ nv40_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       0, 0, 0,
-	                                       nv40_usage_tx_to_buf(usage));
+	                                       pipe_transfer_buffer_flags(&tx->base));
 
 	pipe_texture_reference(&tx_tex, NULL);
 
@@ -114,7 +98,7 @@ nv40_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		struct nv40_screen *nvscreen = nv40_screen(pscreen);
 		struct pipe_surface *src;
 
@@ -140,7 +124,7 @@ nv40_transfer_del(struct pipe_transfer *ptx)
 {
 	struct nv40_transfer *tx = (struct nv40_transfer *)ptx;
 
-	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+	if (!tx->direct && (ptx->usage & PIPE_TRANSFER_WRITE)) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		struct nv40_screen *nvscreen = nv40_screen(pscreen);
 		struct pipe_surface *dst;
@@ -170,7 +154,7 @@ nv40_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
 	struct nv40_miptree *mt = (struct nv40_miptree *)tx->surface->texture;
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
-	                            nv40_usage_tx_to_buf(ptx->usage));
+	                            pipe_transfer_buffer_flags(ptx));
 
 	return map + ns->base.offset +
 	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index eb90d5e66f..576d075318 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -90,6 +90,10 @@ struct nv50_reg {
 	int acc; /* instruction where this reg is last read (first insn == 1) */
 };
 
+/* arbitrary limits */
+#define MAX_IF_DEPTH 4
+#define MAX_LOOP_DEPTH 4
+
 struct nv50_pc {
 	struct nv50_program *p;
 
@@ -121,6 +125,13 @@ struct nv50_pc {
 	struct nv50_reg *iv_p;
 	struct nv50_reg *iv_c;
 
+	struct nv50_program_exec *if_cond;
+	struct nv50_program_exec *if_insn[MAX_IF_DEPTH];
+	struct nv50_program_exec *br_join[MAX_IF_DEPTH];
+	struct nv50_program_exec *br_loop[MAX_LOOP_DEPTH]; /* for BRK branch */
+	int if_lvl, loop_lvl;
+	unsigned loop_pos[MAX_LOOP_DEPTH];
+
 	/* current instruction and total number of insns */
 	unsigned insn_cur;
 	unsigned insn_nr;
@@ -196,6 +207,10 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg)
 	assert(0);
 }
 
+/* XXX: For shaders that aren't executed linearly (e.g. shaders that
+ * contain loops), we need to assign all hw regs to TGSI TEMPs early,
+ * lest we risk temp_temps overwriting regs alloc'd "later".
+ */
 static struct nv50_reg *
 alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst)
 {
@@ -559,6 +574,22 @@ check_swap_src_0_1(struct nv50_pc *pc,
 }
 
 static void
+set_src_0_restricted(struct nv50_pc *pc, struct nv50_reg *src,
+		     struct nv50_program_exec *e)
+{
+	struct nv50_reg *temp;
+
+	if (src->type != P_TEMP) {
+		temp = temp_temp(pc);
+		emit_mov(pc, temp, src);
+		src = temp;
+	}
+
+	alloc_reg(pc, src);
+	e->inst[0] |= (src->hw << 9);
+}
+
+static void
 set_src_0(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e)
 {
 	if (src->type == P_ATTR) {
@@ -760,7 +791,11 @@ emit_flop(struct nv50_pc *pc, unsigned sub,
 	}
 
 	set_dst(pc, dst, e);
-	set_src_0(pc, src, e);
+
+	if (sub == 0 || sub == 2)
+		set_src_0_restricted(pc, src, e);
+	else
+		set_src_0(pc, src, e);
 
 	emit(pc, e);
 }
@@ -810,7 +845,8 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 #define CVT_F32_U32 0x64
 #define CVT_S32_F32 0x8c
 #define CVT_S32_S32 0x0c
-#define CVT_F32_F32_ROP 0xcc
+#define CVT_NEG     0x20
+#define CVT_RI      0x08
 
 static void
 emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src,
@@ -890,6 +926,7 @@ emit_set(struct nv50_pc *pc, unsigned ccode, struct nv50_reg *dst, int wp,
 	set_src_1(pc, src1, e);
 
 	emit(pc, e);
+	pc->if_cond = pc->p->exec_tail; /* record for OPCODE_IF */
 
 	/* cvt.f32.u32/s32 (?) if we didn't only write the predicate */
 	if (rdst)
@@ -917,7 +954,7 @@ map_tgsi_setop_cc(unsigned op)
 static INLINE void
 emit_flr(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src)
 {
-	emit_cvt(pc, dst, src, -1, CVTOP_FLOOR, CVT_F32_F32_ROP);
+	emit_cvt(pc, dst, src, -1, CVTOP_FLOOR, CVT_F32_F32 | CVT_RI);
 }
 
 static void
@@ -1129,10 +1166,11 @@ emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
 	emit(pc, e);
 
 #if 1
-	if (mask & 1) emit_mov(pc, dst[0], t[0]);
-	if (mask & 2) emit_mov(pc, dst[1], t[1]);
-	if (mask & 4) emit_mov(pc, dst[2], t[2]);
-	if (mask & 8) emit_mov(pc, dst[3], t[3]);
+	c = 0;
+	if (mask & 1) emit_mov(pc, dst[0], t[c++]);
+	if (mask & 2) emit_mov(pc, dst[1], t[c++]);
+	if (mask & 4) emit_mov(pc, dst[2], t[c++]);
+	if (mask & 8) emit_mov(pc, dst[3], t[c]);
 
 	free_temp4(pc, t);
 #else
@@ -1149,6 +1187,38 @@ emit_tex(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask,
 }
 
 static void
+emit_branch(struct nv50_pc *pc, int pred, unsigned cc,
+	    struct nv50_program_exec **join)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	if (join) {
+		set_long(pc, e);
+		e->inst[0] |= 0xa0000002;
+		emit(pc, e);
+		*join = e;
+		e = exec(pc);
+	}
+
+	set_long(pc, e);
+	e->inst[0] |= 0x10000002;
+	if (pred >= 0)
+		set_pred(pc, cc, pred, e);
+	emit(pc, e);
+}
+
+static void
+emit_nop(struct nv50_pc *pc)
+{
+	struct nv50_program_exec *e = exec(pc);
+
+	e->inst[0] = 0xf0000000;
+	set_long(pc, e);
+	e->inst[1] = 0xe0000000;
+	emit(pc, e);
+}
+
+static void
 convert_to_long(struct nv50_pc *pc, struct nv50_program_exec *e)
 {
 	unsigned q = 0, m = ~0;
@@ -1444,6 +1514,55 @@ nv50_tgsi_dst_revdep(unsigned op, int s, int c)
 	}
 }
 
+static INLINE boolean
+has_pred(struct nv50_program_exec *e, unsigned cc)
+{
+	if (!is_long(e) || is_immd(e))
+		return FALSE;
+	return ((e->inst[1] & 0x780) == (cc << 7));
+}
+
+/* on ENDIF see if we can do "@p0.neu single_op" instead of:
+ *        join_at ENDIF
+ *        @p0.eq bra ENDIF
+ *        single_op
+ * ENDIF: nop.join
+ */
+static boolean
+nv50_kill_branch(struct nv50_pc *pc)
+{
+	int lvl = pc->if_lvl;
+
+	if (pc->if_insn[lvl]->next != pc->p->exec_tail)
+		return FALSE;
+
+	/* if ccode == 'true', the BRA is from an ELSE and the predicate
+	 * reg may no longer be valid, since we currently always use $p0
+	 */
+	if (has_pred(pc->if_insn[lvl], 0xf))
+		return FALSE;
+	assert(pc->if_insn[lvl] && pc->br_join[lvl]);
+
+	/* We'll use the exec allocated for JOIN_AT (as we can't easily
+	 * update prev's next); if exec_tail is BRK, update the pointer.
+	 */
+	if (pc->loop_lvl && pc->br_loop[pc->loop_lvl - 1] == pc->p->exec_tail)
+		pc->br_loop[pc->loop_lvl - 1] = pc->br_join[lvl];
+
+	pc->p->exec_size -= 4; /* remove JOIN_AT and BRA */
+
+	*pc->br_join[lvl] = *pc->p->exec_tail;
+
+	FREE(pc->if_insn[lvl]);
+	FREE(pc->p->exec_tail);
+
+	pc->p->exec_tail = pc->br_join[lvl];
+	pc->p->exec_tail->next = NULL;
+	set_pred(pc, 0xd, 0, pc->p->exec_tail);
+
+	return TRUE;
+}
+
 static boolean
 nv50_program_tx_insn(struct nv50_pc *pc,
 		     const struct tgsi_full_instruction *inst)
@@ -1513,12 +1632,20 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_add(pc, dst[c], src[0][c], src[1][c]);
 		}
 		break;
+	case TGSI_OPCODE_BGNLOOP:
+		pc->loop_pos[pc->loop_lvl++] = pc->p->exec_size;
+		break;
+	case TGSI_OPCODE_BRK:
+		emit_branch(pc, -1, 0, NULL);
+		assert(pc->loop_lvl > 0);
+		pc->br_loop[pc->loop_lvl - 1] = pc->p->exec_tail;
+		break;
 	case TGSI_OPCODE_CEIL:
 		for (c = 0; c < 4; c++) {
 			if (!(mask & (1 << c)))
 				continue;
 			emit_cvt(pc, dst[c], src[0][c], -1,
-				 CVTOP_CEIL, CVT_F32_F32);
+				 CVTOP_CEIL, CVT_F32_F32 | CVT_RI);
 		}
 		break;
 	case TGSI_OPCODE_COS:
@@ -1560,6 +1687,33 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 		if (mask & (1 << 0))
 			emit_mov_immdval(pc, dst[0], 1.0f);
 		break;
+	case TGSI_OPCODE_ELSE:
+		emit_branch(pc, -1, 0, NULL);
+		pc->if_insn[--pc->if_lvl]->param.index = pc->p->exec_size;
+		pc->if_insn[pc->if_lvl++] = pc->p->exec_tail;
+		break;
+	case TGSI_OPCODE_ENDIF:
+		pc->if_insn[--pc->if_lvl]->param.index = pc->p->exec_size;
+
+		/* try to replace branch over 1 insn with a predicated insn */
+		if (nv50_kill_branch(pc) == TRUE)
+			break;
+
+		if (pc->br_join[pc->if_lvl]) {
+			pc->br_join[pc->if_lvl]->param.index = pc->p->exec_size;
+			pc->br_join[pc->if_lvl] = NULL;
+		}
+		/* emit a NOP as join point, we could set it on the next
+		 * one, but would have to make sure it is long and !immd
+		 */
+		emit_nop(pc);
+		pc->p->exec_tail->inst[1] |= 2;
+		break;
+	case TGSI_OPCODE_ENDLOOP:
+		emit_branch(pc, -1, 0, NULL);
+		pc->p->exec_tail->param.index = pc->loop_pos[--pc->loop_lvl];
+		pc->br_loop[pc->loop_lvl]->param.index = pc->p->exec_size;
+		break;
 	case TGSI_OPCODE_EX2:
 		emit_preex2(pc, temp, src[0][0]);
 		emit_flop(pc, 6, brdc, temp);
@@ -1580,6 +1734,13 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			emit_sub(pc, dst[c], src[0][c], temp);
 		}
 		break;
+	case TGSI_OPCODE_IF:
+		/* emitting a join_at may not be necessary */
+		assert(pc->if_lvl < MAX_IF_DEPTH);
+		set_pred_wr(pc, 1, 0, pc->if_cond);
+		emit_branch(pc, 0, 2, &pc->br_join[pc->if_lvl]);
+		pc->if_insn[pc->if_lvl++] = pc->p->exec_tail;
+		break;
 	case TGSI_OPCODE_KIL:
 		emit_kil(pc, src[0][0]);
 		emit_kil(pc, src[0][1]);
@@ -1704,7 +1865,7 @@ nv50_program_tx_insn(struct nv50_pc *pc,
 			if (!(mask & (1 << c)))
 				continue;
 			emit_cvt(pc, dst[c], src[0][c], -1,
-				 CVTOP_TRUNC, CVT_F32_F32);
+				 CVTOP_TRUNC, CVT_F32_F32 | CVT_RI);
 		}
 		break;
 	case TGSI_OPCODE_XPD:
@@ -2237,6 +2398,8 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 			pc->result[i].rhw = rid++;
 		if (p->info.writes_z)
 			pc->result[2].rhw = rid;
+
+		p->cfg.high_result = rid;
 	}
 
 	if (pc->immd_nr) {
@@ -2362,12 +2525,75 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p)
 	return TRUE;
 }
 
+static void
+nv50_fp_move_results(struct nv50_pc *pc)
+{
+	struct nv50_reg reg;
+	unsigned i;
+
+	ctor_reg(&reg, P_TEMP, -1, -1);
+
+	for (i = 0; i < pc->result_nr * 4; ++i) {
+		if (pc->result[i].rhw < 0 || pc->result[i].hw < 0)
+			continue;
+		if (pc->result[i].rhw != pc->result[i].hw) {
+			reg.hw = pc->result[i].rhw;
+			emit_mov(pc, &reg, &pc->result[i]);
+		}
+	}
+}
+
+static void
+nv50_program_fixup_insns(struct nv50_pc *pc)
+{
+	struct nv50_program_exec *e, *prev = NULL, **bra_list;
+	unsigned i, n, pos;
+
+	bra_list = CALLOC(pc->p->exec_size, sizeof(struct nv50_program_exec *));
+
+	/* Collect branch instructions, we need to adjust their offsets
+	 * when converting 32 bit instructions to 64 bit ones
+	 */
+	for (n = 0, e = pc->p->exec_head; e; e = e->next)
+		if (e->param.index >= 0 && !e->param.mask)
+			bra_list[n++] = e;
+
+	/* Make sure we don't have any single 32 bit instructions. */
+	for (e = pc->p->exec_head, pos = 0; e; e = e->next) {
+		pos += is_long(e) ? 2 : 1;
+
+		if ((pos & 1) && (!e->next || is_long(e->next))) {
+			for (i = 0; i < n; ++i)
+				if (bra_list[i]->param.index >= pos)
+					bra_list[i]->param.index += 1;
+			convert_to_long(pc, e);
+			++pos;
+		}
+		if (e->next)
+			prev = e;
+	}
+
+	assert(!is_immd(pc->p->exec_head));
+	assert(!is_immd(pc->p->exec_tail));
+
+	/* last instruction must be long so it can have the end bit set */
+	if (!is_long(pc->p->exec_tail)) {
+		convert_to_long(pc, pc->p->exec_tail);
+		if (prev)
+			convert_to_long(pc, prev);
+	}
+	assert(!(pc->p->exec_tail->inst[1] & 2));
+	/* set the end-bit */
+	pc->p->exec_tail->inst[1] |= 1;
+
+	FREE(bra_list);
+}
+
 static boolean
 nv50_program_tx(struct nv50_program *p)
 {
 	struct tgsi_parse_context parse;
 	struct nv50_pc *pc;
-	unsigned k;
 	boolean ret;
 
 	pc = CALLOC_STRUCT(nv50_pc);
@@ -2405,48 +2631,10 @@ nv50_program_tx(struct nv50_program *p)
 		}
 	}
 
-	if (p->type == PIPE_SHADER_FRAGMENT) {
-		struct nv50_reg out;
-		ctor_reg(&out, P_TEMP, -1, -1);
-
-		for (k = 0; k < pc->result_nr * 4; k++) {
-			if (pc->result[k].rhw == -1)
-				continue;
-			if (pc->result[k].hw != pc->result[k].rhw) {
-				out.hw = pc->result[k].rhw;
-				emit_mov(pc, &out, &pc->result[k]);
-			}
-			if (pc->p->cfg.high_result < (pc->result[k].rhw + 1))
-				pc->p->cfg.high_result = pc->result[k].rhw + 1;
-		}
-	}
-
-	/* look for single half instructions and make them long */
-	struct nv50_program_exec *e, *e_prev;
+	if (pc->p->type == PIPE_SHADER_FRAGMENT)
+		nv50_fp_move_results(pc);
 
-	for (k = 0, e = pc->p->exec_head, e_prev = NULL; e; e = e->next) {
-		if (!is_long(e))
-			k++;
-
-		if (!e->next || is_long(e->next)) {
-			if (k & 1)
-				convert_to_long(pc, e);
-			k = 0;
-		}
-
-		if (e->next)
-			e_prev = e;
-	}
-
-	if (!is_long(pc->p->exec_tail)) {
-		/* this may occur if moving FP results */
-		assert(e_prev && !is_long(e_prev));
-		convert_to_long(pc, e_prev);
-		convert_to_long(pc, pc->p->exec_tail);
-	}
-
-	assert(is_long(pc->p->exec_tail) && !is_immd(pc->p->exec_head));
-	pc->p->exec_tail->inst[1] |= 0x00000001;
+	nv50_program_fixup_insns(pc);
 
 	p->param_nr = pc->param_nr * 4;
 	p->immd_nr = pc->immd_nr * 4;
@@ -2558,6 +2746,17 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 
 		if (e->param.index < 0)
 			continue;
+
+		if (e->param.mask == 0) {
+			assert(!(e->param.index & 1));
+			/* seem to be 8 byte steps */
+			ei = (e->param.index >> 1) + 0 /* START_ID */;
+
+			e->inst[0] &= 0xf0000fff;
+			e->inst[0] |= ei << 12;
+			continue;
+		}
+
 		bs = (e->inst[1] >> 22) & 0x07;
 		assert(bs < 2);
 		ei = e->param.shift >> 5;
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 5a3559ed18..4ed76973c4 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -312,7 +312,7 @@ scissor_uptodate:
 			goto viewport_uptodate;
 		nv50->state.viewport_bypass = bypass;
 
-		so = so_new(12, 0);
+		so = so_new(14, 0);
 		if (!bypass) {
 			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE(0), 3);
 			so_data  (so, fui(nv50->viewport.translate[0]));
@@ -325,12 +325,21 @@ scissor_uptodate:
 
 			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
 			so_data  (so, 1);
+			/* 0x0000 = remove whole primitive only (xyz)
+			 * 0x1018 = remove whole primitive only (xy), clamp z
+			 * 0x1080 = clip primitive (xyz)
+			 * 0x1098 = clip primitive (xy), clamp z
+			 */
+			so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
+			so_data  (so, 0x1080);
 			/* no idea what 0f90 does */
 			so_method(so, tesla, 0x0f90, 1);
 			so_data  (so, 0);
 		} else {
 			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
 			so_data  (so, 0);
+			so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
+			so_data  (so, 0x0000);
 			so_method(so, tesla, 0x0f90, 1);
 			so_data  (so, 1);
 		}
diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c
index bb7731855c..9c289026bb 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.c
+++ b/src/gallium/drivers/nv50/nv50_transfer.c
@@ -161,7 +161,7 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
-	if (usage != PIPE_TRANSFER_WRITE) {
+	if (usage & PIPE_TRANSFER_READ) {
 		nv50_transfer_rect_m2mf(pscreen, mt->base.bo, tx->level_offset,
 					tx->level_pitch, tx->level_tiling,
 					x, y,
@@ -183,7 +183,7 @@ nv50_transfer_del(struct pipe_transfer *ptx)
 	struct nv50_transfer *tx = (struct nv50_transfer *)ptx;
 	struct nv50_miptree *mt = nv50_miptree(ptx->texture);
 
-	if (ptx->usage != PIPE_TRANSFER_READ) {
+	if (ptx->usage & PIPE_TRANSFER_WRITE) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
 		nv50_transfer_rect_m2mf(pscreen, tx->bo, 0, tx->base.stride,
 					tx->bo->tile_mode, 0, 0,
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index 0a7e470363..883f0a02dc 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -68,11 +68,17 @@
 } while (0)
 
 #define OUT_CS(value) do { \
+    if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \
+        DBG(cs_context_copy, DBG_CS, "r300: writing %08x\n", value); \
+    } \
     cs_winsys->write_cs_dword(cs_winsys, (value)); \
     cs_count--; \
 } while (0)
 
 #define OUT_CS_32F(value) do { \
+    if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \
+        DBG(cs_context_copy, DBG_CS, "r300: writing %f\n", value); \
+    } \
     cs_winsys->write_cs_dword(cs_winsys, fui(value)); \
     cs_count--; \
 } while (0)
@@ -82,8 +88,9 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing 0x%08X to register 0x%04X\n", \
             value, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, 0)); \
-    OUT_CS(value); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0(register, 0)); \
+    cs_winsys->write_cs_dword(cs_winsys, value); \
+    cs_count -= 2; \
 } while (0)
 
 /* Note: This expects count to be the number of registers,
@@ -93,7 +100,8 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing register sequence of %d to 0x%04X\n", \
             count, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, ((count) - 1))); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1))); \
+    cs_count--; \
 } while (0)
 
 #define OUT_CS_RELOC(bo, offset, rd, wd, flags) do { \
@@ -101,9 +109,9 @@
             "domains (%d, %d, %d)\n", \
         bo, offset, rd, wd, flags); \
     assert(bo); \
-    OUT_CS(offset); \
+    cs_winsys->write_cs_dword(cs_winsys, offset); \
     cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \
-    cs_count -= 2; \
+    cs_count -= 3; \
 } while (0)
 
 #define END_CS do { \
@@ -131,24 +139,26 @@
         DBG(cs_context_copy, DBG_CS, "r300: writing data sequence of %d to 0x%04X\n", \
             count, register); \
     assert(register); \
-    OUT_CS(CP_PACKET0(register, ((count) - 1)) | RADEON_ONE_REG_WR); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1)) | RADEON_ONE_REG_WR); \
+    cs_count--; \
 } while (0)
 
 #define CP_PACKET3(op, count) \
     (RADEON_CP_PACKET3 | (op) | ((count) << 16))
 
 #define OUT_CS_PKT3(op, count) do { \
-    OUT_CS(CP_PACKET3(op, count)); \
+    cs_winsys->write_cs_dword(cs_winsys, CP_PACKET3(op, count)); \
+    cs_count--; \
 } while (0)
 
 #define OUT_CS_INDEX_RELOC(bo, offset, count, rd, wd, flags) do { \
     DBG(cs_context_copy, DBG_CS, "r300: writing relocation for index buffer %p," \
             "offset %d\n", bo, offset); \
     assert(bo); \
-    OUT_CS(offset); \
-    OUT_CS(count); \
+    cs_winsys->write_cs_dword(cs_winsys, offset); \
+    cs_winsys->write_cs_dword(cs_winsys, count); \
     cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \
-    cs_count -= 2; \
+    cs_count -= 4; \
 } while (0)
 
 #endif /* R300_CS_H */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index a1b36ba2ed..77ce431cdc 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -381,6 +381,7 @@ void r300_emit_query_end(struct r300_context* r300,
             OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
             OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0),
                     0, RADEON_GEM_DOMAIN_GTT, 0);
+	    break;
         default:
             debug_printf("r300: Implementation error: Chipset reports %d"
                     " pixel pipes!\n", caps->num_frag_pipes);
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index a0e848a59a..546ad545a5 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -126,9 +126,14 @@ void r300_translate_fragment_shader(struct r300_context* r300,
     /* Invoke the compiler */
     r3xx_compile_fragment_program(&compiler);
     if (compiler.Base.Error) {
-        /* Todo: Fail gracefully */
-        fprintf(stderr, "r300 FP: Compiler error\n");
-        abort();
+        /* Todo: Fallback to software rendering gracefully? */
+        fprintf(stderr, "r300 FP: Compiler error: %s\n", compiler.Base.ErrorMsg);
+
+        if (compiler.is_r500) {
+            memcpy(compiler.code, &r5xx_passthrough_fragment_shader, sizeof(r5xx_passthrough_fragment_shader));
+        } else {
+            memcpy(compiler.code, &r3xx_passthrough_fragment_shader, sizeof(r3xx_passthrough_fragment_shader));
+        }
     }
 
     /* And, finally... */
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index 9fab789402..967e9f697e 100644
--- a/src/gallium/drivers/r300/r300_fs.h
+++ b/src/gallium/drivers/r300/r300_fs.h
@@ -48,4 +48,4 @@ struct r300_fragment_shader {
 void r300_translate_fragment_shader(struct r300_context* r300,
                                     struct r300_fragment_shader* fs);
 
-    #endif /* R300_FS_H */
+#endif /* R300_FS_H */
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 03cd219cde..3abff5db62 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -1478,6 +1478,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_TX_PITCH_EN                  (1 << 31)
 #       define R300_TX_WIDTH(x)                  ((x) << 0)
 #       define R300_TX_HEIGHT(x)                 ((x) << 11)
+#       define R300_TX_DEPTH(x)                  ((x) << 22)
 #       define R300_TX_NUM_LEVELS(x)             ((x) << 26)
 
 #define R300_TX_FORMAT1_0                   0x44C0
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 737396d8d9..ca44e0f661 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -79,12 +79,14 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
     struct pipe_screen* screen = r300->context.screen;
     size_t size = (size_t)vertex_size * (size_t)count;
 
-    if (size + r300render->vbo_offset > r300render->vbo_size) 
+    if (size + r300render->vbo_offset > r300render->vbo_size)
     {
+        pipe_buffer_reference(&r300->vbo, NULL);
         r300render->vbo = pipe_buffer_create(screen,
                                              64,
                                              PIPE_BUFFER_USAGE_VERTEX,
                                              R300_MAX_VBO_SIZE);
+        r300render->vbo_offset = 0;
         r300render->vbo_size = R300_MAX_VBO_SIZE;
     }
 
@@ -117,7 +119,7 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
     END_CS;
 
-    r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 
+    r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
                                     r300render->vertex_size * (max + 1));
     pipe_buffer_unmap(screen, r300render->vbo);
 }
@@ -129,7 +131,6 @@ static void r300_render_release_vertices(struct vbuf_render* render)
 
     r300render->vbo_offset += r300render->vbo_max_used;
     r300render->vbo_max_used = 0;
-    r300->vbo = NULL;
 }
 
 static boolean r300_render_set_primitive(struct vbuf_render* render,
@@ -222,13 +223,6 @@ static void r300_render_draw(struct vbuf_render* render,
 
     r300_prepare_render(r300render, count);
 
-    /* Send our indices into an index buffer. */
-    index_buffer = pipe_buffer_create(screen, 64, PIPE_BUFFER_USAGE_VERTEX,
-                                      count * 2);
-    if (!index_buffer) {
-        return;
-    }
-
     BEGIN_CS(2 + (count+1)/2);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 3b5b1bbd37..f2659ca61f 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -101,11 +101,9 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
         case PIPE_CAP_MAX_RENDER_TARGETS:
             return 4;
         case PIPE_CAP_OCCLUSION_QUERY:
-            /* IN THEORY */
-            return 0;
+            return 1;
         case PIPE_CAP_TEXTURE_SHADOW_MAP:
-            /* IN THEORY */
-            return 0;
+            return 1;
         case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
             if (r300screen->caps->is_r500) {
                 /* 13 == 4096x4096 */
@@ -347,16 +345,9 @@ static void* r300_transfer_map(struct pipe_screen* screen,
 {
     struct r300_texture* tex = (struct r300_texture*)transfer->texture;
     char* map;
-    unsigned flags = 0;
-
-    if (transfer->usage != PIPE_TRANSFER_WRITE) {
-        flags |= PIPE_BUFFER_USAGE_CPU_READ;
-    }
-    if (transfer->usage != PIPE_TRANSFER_READ) {
-        flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
-    }
 
-    map = pipe_buffer_map(screen, tex->buffer, flags);
+    map = pipe_buffer_map(screen, tex->buffer,
+                          pipe_transfer_buffer_flags(transfer));
 
     if (!map) {
         return NULL;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 5f6b225d34..02b7ab9107 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -53,7 +53,7 @@ static void r300_vs_tab_routes(struct r300_context* r300,
     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
     {
         for (i = 0; i < info->num_inputs; i++) {
-            switch (info->input_semantic_name[i]) {
+            switch (r300->vs->code.inputs[i]) {
                 case TGSI_SEMANTIC_POSITION:
                     pos = TRUE;
                     tab[i] = 0;
@@ -63,10 +63,12 @@ static void r300_vs_tab_routes(struct r300_context* r300,
                     cols++;
                     break;
                 case TGSI_SEMANTIC_PSIZE:
+                    assert(psize == FALSE);
                     psize = TRUE;
                     tab[i] = 15;
                     break;
                 case TGSI_SEMANTIC_FOG:
+                    assert(fog == FALSE);
                     fog = TRUE;
                     /* Fall through */
                 case TGSI_SEMANTIC_GENERIC:
@@ -125,7 +127,9 @@ static void r300_vs_tab_routes(struct r300_context* r300,
 
     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 
-    if (!pos) {
+    /* We need to add vertex position attribute only for SW TCL case,
+     * for HW TCL case it could be generated by vertex shader */
+    if (!pos && !r300screen->caps->has_tcl) {
         debug_printf("r300: Forcing vertex position attribute emit...\n");
         /* Make room for the position attribute
          * at the beginning of the tab. */
@@ -164,7 +168,7 @@ static void r300_vs_tab_routes(struct r300_context* r300,
         vinfo->hwfmt[3] |= (4 << (3 * i));
     }
 
-    for (i; i < texs; i++) {
+    for (; i < texs; i++) {
         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
@@ -334,48 +338,37 @@ static void r300_update_rs_block(struct r300_context* r300)
     struct r300_rs_block* rs = r300->rs_block;
     struct tgsi_shader_info* info = &r300->fs->info;
     int* tab = r300->vertex_info.fs_tab;
-    int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0;
-
+    int col_count = 0, fp_offset = 0, i, tex_count = 0;
+    int rs_tex_comp = 0;
     memset(rs, 0, sizeof(struct r300_rs_block));
 
     if (r300_screen(r300->context.screen)->caps->is_r500) {
         for (i = 0; i < info->num_inputs; i++) {
             assert(tab[i] != -1);
-            memory_pos = tab[i] * 4;
             switch (info->input_semantic_name[i]) {
                 case TGSI_SEMANTIC_COLOR:
                     rs->ip[col_count] |=
-                        R500_RS_COL_PTR(memory_pos) |
+                        R500_RS_COL_PTR(col_count) |
                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
                     col_count++;
                     break;
                 case TGSI_SEMANTIC_GENERIC:
                     rs->ip[tex_count] |=
-                        R500_RS_SEL_S(memory_pos) |
-                        R500_RS_SEL_T(memory_pos + 1) |
-                        R500_RS_SEL_R(memory_pos + 2) |
-                        R500_RS_SEL_Q(memory_pos + 3);
+                        R500_RS_SEL_S(rs_tex_comp) |
+                        R500_RS_SEL_T(rs_tex_comp + 1) |
+                        R500_RS_SEL_R(rs_tex_comp + 2) |
+                        R500_RS_SEL_Q(rs_tex_comp + 3);
                     tex_count++;
+                    rs_tex_comp += 4;
                     break;
                 default:
                     break;
             }
         }
 
-        if (col_count == 0) {
-            rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
-        }
-
-        if (tex_count == 0) {
-            rs->ip[0] |=
-                R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
-                R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
-        }
-
         /* Rasterize at least one color, or bad things happen. */
         if ((col_count == 0) && (tex_count == 0)) {
+            rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
             col_count++;
         }
 
@@ -393,22 +386,22 @@ static void r300_update_rs_block(struct r300_context* r300)
     } else {
         for (i = 0; i < info->num_inputs; i++) {
             assert(tab[i] != -1);
-            memory_pos = tab[i] * 4;
             switch (info->input_semantic_name[i]) {
                 case TGSI_SEMANTIC_COLOR:
                     rs->ip[col_count] |=
-                        R300_RS_COL_PTR(memory_pos) |
+                        R300_RS_COL_PTR(col_count) |
                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
                     col_count++;
                     break;
                 case TGSI_SEMANTIC_GENERIC:
                     rs->ip[tex_count] |=
-                        R300_RS_TEX_PTR(memory_pos) |
+                        R300_RS_TEX_PTR(rs_tex_comp) |
                         R300_RS_SEL_S(R300_RS_SEL_C0) |
                         R300_RS_SEL_T(R300_RS_SEL_C1) |
                         R300_RS_SEL_R(R300_RS_SEL_C2) |
                         R300_RS_SEL_Q(R300_RS_SEL_C3);
                     tex_count++;
+                    rs_tex_comp+=4;
                     break;
                 default:
                     break;
@@ -445,7 +438,7 @@ static void r300_update_rs_block(struct r300_context* r300)
         }
     }
 
-    rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
+    rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
         R300_HIRES_EN;
 
     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 6e8c368320..ce60ded7ca 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -22,34 +22,39 @@
 
 #include "r300_texture.h"
 
-static void r300_setup_texture_state(struct r300_texture* tex,
-                                     unsigned width,
-                                     unsigned height,
-                                     unsigned levels)
+static void r300_setup_texture_state(struct r300_texture* tex)
 {
     struct r300_texture_state* state = &tex->state;
+    struct pipe_texture *pt = &tex->tex;
 
-    state->format0 = R300_TX_WIDTH((width - 1) & 0x7ff) |
-        R300_TX_HEIGHT((height - 1) & 0x7ff) |
-        R300_TX_NUM_LEVELS(levels) |
+    state->format0 = R300_TX_WIDTH((pt->width[0] - 1) & 0x7ff) |
+        R300_TX_HEIGHT((pt->height[0] - 1) & 0x7ff) |
+        R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf) |
+        R300_TX_NUM_LEVELS(pt->last_level) |
         R300_TX_PITCH_EN;
 
     /* XXX */
-    state->format1 = r300_translate_texformat(tex->tex.format);
+    state->format1 = r300_translate_texformat(pt->format);
+    if (pt->target == PIPE_TEXTURE_CUBE) {
+	state->format1 |= R300_TX_FORMAT_CUBIC_MAP;
+    }
+    if (pt->target == PIPE_TEXTURE_3D) {
+	state->format1 |= R300_TX_FORMAT_3D;
+    }
 
-    state->format2 = r300_texture_get_stride(tex, 0);
+    state->format2 = (r300_texture_get_stride(tex, 0) / pt->block.size) - 1;
 
     /* Assume (somewhat foolishly) that oversized textures will
      * not be permitted by the state tracker. */
-    if (width > 2048) {
+    if (pt->width[0] > 2048) {
         state->format2 |= R500_TXWIDTH_BIT11;
     }
-    if (height > 2048) {
+    if (pt->height[0] > 2048) {
         state->format2 |= R500_TXHEIGHT_BIT11;
     }
 
-    debug_printf("r300: Set texture state (%dx%d, pitch %d, %d levels)\n",
-            width, height, levels);
+    debug_printf("r300: Set texture state (%dx%d, %d levels)\n",
+		 pt->width[0], pt->height[0], pt->last_level);
 }
 
 /**
@@ -62,7 +67,7 @@ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
         return tex->stride_override;
 
     if (level > tex->tex.last_level) {
-        debug_printf("%s: level (%u) > last_level (%u)\n", level, tex->tex.last_level);
+        debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__, level, tex->tex.last_level);
         return 0;
     }
 
@@ -120,8 +125,7 @@ static struct pipe_texture*
 
     r300_setup_miptree(tex);
 
-    r300_setup_texture_state(tex, template->width[0], template->height[0],
-                             template->last_level);
+    r300_setup_texture_state(tex);
 
     tex->buffer = screen->buffer_create(screen, 1024,
                                         PIPE_BUFFER_USAGE_PIXEL,
@@ -204,7 +208,7 @@ static struct pipe_texture*
     tex->stride_override = *stride;
 
     /* XXX */
-    r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0], 0);
+    r300_setup_texture_state(tex);
 
     pipe_buffer_reference(&tex->buffer, buffer);
 
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 3109af5bac..78ee0f1611 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -72,6 +72,11 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
         /* W24_FP */
         case PIPE_FORMAT_Z24S8_UNORM:
             return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP);
+	/* Z5_Y6_X5 */
+        case PIPE_FORMAT_R16_SNORM:
+            return R300_EASY_TX_FORMAT(X, X, X, X, Z5Y6X5);
+        case PIPE_FORMAT_Z16_UNORM:
+	    return R300_EASY_TX_FORMAT(X, X, X, X, X16);
         default:
             debug_printf("r300: Implementation error: "
                 "Got unsupported texture format %s in %s\n",
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index 516e3992fd..bcb887a0b2 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -6,26 +6,17 @@ LIBNAME = softpipe
 C_SOURCES = \
 	sp_fs_exec.c \
 	sp_fs_sse.c \
-	sp_fs_llvm.c \
 	sp_clear.c \
 	sp_flush.c \
 	sp_query.c \
 	sp_context.c \
 	sp_draw_arrays.c \
-	sp_prim_setup.c \
 	sp_prim_vbuf.c \
 	sp_quad_pipe.c \
-	sp_quad_alpha_test.c \
-	sp_quad_blend.c \
-	sp_quad_colormask.c \
-	sp_quad_coverage.c \
+	sp_quad_stipple.c \
 	sp_quad_depth_test.c \
-	sp_quad_earlyz.c \
 	sp_quad_fs.c \
-	sp_quad_occlusion.c \
-	sp_quad_output.c \
-	sp_quad_stencil.c \
-	sp_quad_stipple.c \
+	sp_quad_blend.c \
 	sp_screen.c \
         sp_setup.c \
 	sp_state_blend.c \
@@ -38,7 +29,9 @@ C_SOURCES = \
 	sp_state_vertex.c \
 	sp_texture.c \
 	sp_tex_sample.c \
+	sp_tex_tile_cache.c \
 	sp_tile_cache.c \
-	sp_surface.c 
+	sp_surface.c \
+	sp_video_context.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index f8720638a7..aac9edf44e 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -7,25 +7,16 @@ softpipe = env.ConvenienceLibrary(
 	source = [
 		'sp_fs_exec.c',
 		'sp_fs_sse.c',
-		'sp_fs_llvm.c',
 		'sp_clear.c',
 		'sp_context.c',
 		'sp_draw_arrays.c',
 		'sp_flush.c',
-		'sp_prim_setup.c',
 		'sp_prim_vbuf.c',
 		'sp_setup.c',
-		'sp_quad_alpha_test.c',
 		'sp_quad_blend.c',
 		'sp_quad_pipe.c',
-		'sp_quad_colormask.c',
-		'sp_quad_coverage.c',
 		'sp_quad_depth_test.c',
-		'sp_quad_earlyz.c',
 		'sp_quad_fs.c',
-		'sp_quad_occlusion.c',
-		'sp_quad_output.c',
-		'sp_quad_stencil.c',
 		'sp_quad_stipple.c',
 		'sp_query.c',
 		'sp_screen.c',
@@ -39,8 +30,10 @@ softpipe = env.ConvenienceLibrary(
 		'sp_state_vertex.c',
 		'sp_surface.c',
 		'sp_tex_sample.c',
+		'sp_tex_tile_cache.c',
 		'sp_texture.c',
 		'sp_tile_cache.c',
+		'sp_video_context.c',
 	])
 
-Export('softpipe')
\ No newline at end of file
+Export('softpipe')
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index c699c433e9..94d000a5ac 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -31,17 +31,18 @@
  */
 
 #include "draw/draw_context.h"
+#include "draw/draw_vbuf.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 #include "sp_clear.h"
 #include "sp_context.h"
 #include "sp_flush.h"
-#include "sp_prim_setup.h"
 #include "sp_prim_vbuf.h"
 #include "sp_state.h"
 #include "sp_surface.h"
 #include "sp_tile_cache.h"
+#include "sp_tex_tile_cache.h"
 #include "sp_texture.h"
 #include "sp_winsys.h"
 #include "sp_query.h"
@@ -72,13 +73,10 @@ softpipe_unmap_transfers(struct softpipe_context *sp)
 {
    uint i;
 
-   for (i = 0; i < sp->framebuffer.nr_cbufs; i++)
-      sp_flush_tile_cache(sp, sp->cbuf_cache[i]);
-   sp_flush_tile_cache(sp, sp->zsbuf_cache);
-
    for (i = 0; i < sp->framebuffer.nr_cbufs; i++) {
       sp_tile_cache_unmap_transfers(sp->cbuf_cache[i]);
    }
+
    sp_tile_cache_unmap_transfers(sp->zsbuf_cache);
 }
 
@@ -92,26 +90,21 @@ softpipe_destroy( struct pipe_context *pipe )
    if (softpipe->draw)
       draw_destroy( softpipe->draw );
 
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      softpipe->quad[i].polygon_stipple->destroy( softpipe->quad[i].polygon_stipple );
-      softpipe->quad[i].earlyz->destroy( softpipe->quad[i].earlyz );
-      softpipe->quad[i].shade->destroy( softpipe->quad[i].shade );
-      softpipe->quad[i].alpha_test->destroy( softpipe->quad[i].alpha_test );
-      softpipe->quad[i].depth_test->destroy( softpipe->quad[i].depth_test );
-      softpipe->quad[i].stencil_test->destroy( softpipe->quad[i].stencil_test );
-      softpipe->quad[i].occlusion->destroy( softpipe->quad[i].occlusion );
-      softpipe->quad[i].coverage->destroy( softpipe->quad[i].coverage );
-      softpipe->quad[i].blend->destroy( softpipe->quad[i].blend );
-      softpipe->quad[i].colormask->destroy( softpipe->quad[i].colormask );
-      softpipe->quad[i].output->destroy( softpipe->quad[i].output );
-   }
+      softpipe->quad.shade->destroy( softpipe->quad.shade );
+      softpipe->quad.depth_test->destroy( softpipe->quad.depth_test );
+      softpipe->quad.blend->destroy( softpipe->quad.blend );
 
-   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++)
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       sp_destroy_tile_cache(softpipe->cbuf_cache[i]);
+      pipe_surface_reference(&softpipe->framebuffer.cbufs[i], NULL);
+   }
    sp_destroy_tile_cache(softpipe->zsbuf_cache);
+   pipe_surface_reference(&softpipe->framebuffer.zsbuf, NULL);
 
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
-      sp_destroy_tile_cache(softpipe->tex_cache[i]);
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      sp_destroy_tex_tile_cache(softpipe->tex_cache[i]);
+      pipe_texture_reference(&softpipe->texture[i], NULL);
+   }
 
    for (i = 0; i < Elements(softpipe->constants); i++) {
       if (softpipe->constants[i].buffer) {
@@ -246,41 +239,14 @@ softpipe_create( struct pipe_screen *screen )
    softpipe->zsbuf_cache = sp_create_tile_cache( screen );
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++)
-      softpipe->tex_cache[i] = sp_create_tile_cache( screen );
+      softpipe->tex_cache[i] = sp_create_tex_tile_cache( screen );
 
 
    /* setup quad rendering stages */
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      softpipe->quad[i].polygon_stipple = sp_quad_polygon_stipple_stage(softpipe);
-      softpipe->quad[i].earlyz = sp_quad_earlyz_stage(softpipe);
-      softpipe->quad[i].shade = sp_quad_shade_stage(softpipe);
-      softpipe->quad[i].alpha_test = sp_quad_alpha_test_stage(softpipe);
-      softpipe->quad[i].depth_test = sp_quad_depth_test_stage(softpipe);
-      softpipe->quad[i].stencil_test = sp_quad_stencil_test_stage(softpipe);
-      softpipe->quad[i].occlusion = sp_quad_occlusion_stage(softpipe);
-      softpipe->quad[i].coverage = sp_quad_coverage_stage(softpipe);
-      softpipe->quad[i].blend = sp_quad_blend_stage(softpipe);
-      softpipe->quad[i].colormask = sp_quad_colormask_stage(softpipe);
-      softpipe->quad[i].output = sp_quad_output_stage(softpipe);
-   }
+      softpipe->quad.shade = sp_quad_shade_stage(softpipe);
+      softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
+      softpipe->quad.blend = sp_quad_blend_stage(softpipe);
 
-   /* vertex shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples_vertex;
-      softpipe->tgsi.vert_samplers[i].unit = i;
-      softpipe->tgsi.vert_samplers[i].sp = softpipe;
-      softpipe->tgsi.vert_samplers[i].cache = softpipe->tex_cache[i];
-      softpipe->tgsi.vert_samplers_list[i] = &softpipe->tgsi.vert_samplers[i];
-   }
-
-   /* fragment shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples_fragment;
-      softpipe->tgsi.frag_samplers[i].unit = i;
-      softpipe->tgsi.frag_samplers[i].sp = softpipe;
-      softpipe->tgsi.frag_samplers[i].cache = softpipe->tex_cache[i];
-      softpipe->tgsi.frag_samplers_list[i] = &softpipe->tgsi.frag_samplers[i];
-   }
 
    /*
     * Create drawing context and plug our rendering stage into it.
@@ -294,30 +260,28 @@ softpipe_create( struct pipe_screen *screen )
                          (struct tgsi_sampler **)
                             softpipe->tgsi.vert_samplers_list);
 
-   softpipe->setup = sp_draw_render_stage(softpipe);
-   if (!softpipe->setup)
-      goto fail;
-
    if (debug_get_bool_option( "SP_NO_RAST", FALSE ))
       softpipe->no_rast = TRUE;
 
-   if (debug_get_bool_option( "SP_NO_VBUF", FALSE )) {
-      /* Deprecated path -- vbuf is the intended interface to the draw module:
-       */
-      draw_set_rasterize_stage(softpipe->draw, softpipe->setup);
-   }
-   else {
-      sp_init_vbuf(softpipe);
-   }
+   softpipe->vbuf_backend = sp_create_vbuf_backend(softpipe);
+   if (!softpipe->vbuf_backend)
+      goto fail;
+
+   softpipe->vbuf = draw_vbuf_stage(softpipe->draw, softpipe->vbuf_backend);
+   if (!softpipe->vbuf)
+      goto fail;
+
+   draw_set_rasterize_stage(softpipe->draw, softpipe->vbuf);
+   draw_set_render(softpipe->draw, softpipe->vbuf_backend);
+
+
 
    /* 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_init_surface_functions(softpipe);
 
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 7888c2f644..43a195c8ef 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -36,24 +36,13 @@
 #include "draw/draw_vertex.h"
 
 #include "sp_quad_pipe.h"
-#include "sp_tex_sample.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
-
-/* Number of threads working on individual quads.
- * Setting to 1 disables this feature.
- */
-#define SP_NUM_QUAD_THREADS 1
-
 struct softpipe_vbuf_render;
 struct draw_context;
 struct draw_stage;
 struct softpipe_tile_cache;
+struct softpipe_tex_tile_cache;
 struct sp_fragment_shader;
 struct sp_vertex_shader;
 
@@ -62,12 +51,12 @@ struct softpipe_context {
    struct pipe_context pipe;  /**< base class */
 
    /** Constant state objects */
-   const struct pipe_blend_state *blend;
-   const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
-   const struct pipe_depth_stencil_alpha_state *depth_stencil;
-   const struct pipe_rasterizer_state *rasterizer;
-   const struct sp_fragment_shader *fs;
-   const struct sp_vertex_shader *vs;
+   struct pipe_blend_state *blend;
+   struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   struct pipe_depth_stencil_alpha_state *depth_stencil;
+   struct pipe_rasterizer_state *rasterizer;
+   struct sp_fragment_shader *fs;
+   struct sp_vertex_shader *vs;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -107,7 +96,16 @@ struct softpipe_context {
    /** Which vertex shader output slot contains point size */
    int psize_slot;
 
-   unsigned reduced_api_prim;  /**< PIPE_PRIM_POINTS, _LINES or _TRIANGLES */
+   /* The reduced version of the primitive supplied by the state
+    * tracker.
+    */
+   unsigned reduced_api_prim;
+
+   /* The reduced primitive after unfilled triangles, wide-line
+    * decomposition, etc, are taken into account.  This is the
+    * primitive actually rasterized.
+    */
+   unsigned reduced_prim;
 
    /** Derived from scissor and surface bounds: */
    struct pipe_scissor_state cliprect;
@@ -116,41 +114,33 @@ struct softpipe_context {
 
    /** Software quad rendering pipeline */
    struct {
-      struct quad_stage *polygon_stipple;
-      struct quad_stage *earlyz;
       struct quad_stage *shade;
-      struct quad_stage *alpha_test;
-      struct quad_stage *stencil_test;
       struct quad_stage *depth_test;
-      struct quad_stage *occlusion;
-      struct quad_stage *coverage;
       struct quad_stage *blend;
-      struct quad_stage *colormask;
-      struct quad_stage *output;
 
       struct quad_stage *first; /**< points to one of the above stages */
-   } quad[SP_NUM_QUAD_THREADS];
+   } quad;
 
    /** TGSI exec things */
    struct {
-      struct sp_shader_sampler vert_samplers[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler *vert_samplers_list[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler frag_samplers[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler *frag_samplers_list[PIPE_MAX_SAMPLERS];
+      struct sp_sampler_varient *vert_samplers_list[PIPE_MAX_SAMPLERS];
+      struct sp_sampler_varient *frag_samplers_list[PIPE_MAX_SAMPLERS];
    } tgsi;
 
    /** The primitive drawing context */
    struct draw_context *draw;
-   struct draw_stage *setup;
+
+   /** Draw module backend */
+   struct vbuf_render *vbuf_backend;
    struct draw_stage *vbuf;
-   struct softpipe_vbuf_render *vbuf_render;
 
    boolean dirty_render_cache;
    
    struct softpipe_tile_cache *cbuf_cache[PIPE_MAX_COLOR_BUFS];
    struct softpipe_tile_cache *zsbuf_cache;
-
-   struct softpipe_tile_cache *tex_cache[PIPE_MAX_SAMPLERS];
+   
+   unsigned tex_timestamp;
+   struct softpipe_tex_tile_cache *tex_cache[PIPE_MAX_SAMPLERS];
 
    unsigned use_sse : 1;
    unsigned dump_fs : 1;
@@ -164,5 +154,9 @@ softpipe_context( struct pipe_context *pipe )
    return (struct softpipe_context *)pipe;
 }
 
+void
+softpipe_reset_sampler_varients(struct softpipe_context *softpipe);
+
+
 #endif /* SP_CONTEXT_H */
 
diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c
index 4a14d49686..e38b767cf2 100644
--- a/src/gallium/drivers/softpipe/sp_flush.c
+++ b/src/gallium/drivers/softpipe/sp_flush.c
@@ -37,6 +37,7 @@
 #include "sp_surface.h"
 #include "sp_state.h"
 #include "sp_tile_cache.h"
+#include "sp_tex_tile_cache.h"
 #include "sp_winsys.h"
 
 
@@ -52,17 +53,19 @@ softpipe_flush( struct pipe_context *pipe,
 
    if (flags & PIPE_FLUSH_TEXTURE_CACHE) {
       for (i = 0; i < softpipe->num_textures; i++) {
-         sp_flush_tile_cache(softpipe, softpipe->tex_cache[i]);
+         sp_flush_tex_tile_cache(softpipe->tex_cache[i]);
       }
    }
 
-   if (flags & PIPE_FLUSH_RENDER_CACHE) {
+   if (flags & PIPE_FLUSH_SWAPBUFFERS) {
+      /* If this is a swapbuffers, just flush color buffers.
+       *
+       * The zbuffer changes are not discarded, but held in the cache
+       * in the hope that a later clear will wipe them out.
+       */
       for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++)
          if (softpipe->cbuf_cache[i])
-            sp_flush_tile_cache(softpipe, softpipe->cbuf_cache[i]);
-
-      if (softpipe->zsbuf_cache)
-         sp_flush_tile_cache(softpipe, softpipe->zsbuf_cache);
+            sp_flush_tile_cache(softpipe->cbuf_cache[i]);
 
       /* Need this call for hardware buffers before swapbuffers.
        *
@@ -71,7 +74,15 @@ softpipe_flush( struct pipe_context *pipe,
        * to unmap surfaces when flushing.
        */
       softpipe_unmap_transfers(softpipe);
-      
+   }
+   else if (flags & PIPE_FLUSH_RENDER_CACHE) {
+      for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++)
+         if (softpipe->cbuf_cache[i])
+            sp_flush_tile_cache(softpipe->cbuf_cache[i]);
+
+      if (softpipe->zsbuf_cache)
+         sp_flush_tile_cache(softpipe->zsbuf_cache);
+     
       softpipe->dirty_render_cache = FALSE;
    }
 
diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index 9ee86fe787..4076114d39 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -59,15 +59,34 @@ sp_exec_fragment_shader(const struct sp_fragment_shader *base)
 }
 
 
+static void
+exec_prepare( const struct sp_fragment_shader *base,
+	      struct tgsi_exec_machine *machine,
+	      struct tgsi_sampler **samplers )
+{
+   /*
+    * Bind tokens/shader to the interpreter's machine state.
+    * Avoid redundant binding.
+    */
+   if (machine->Tokens != base->shader.tokens) {
+      tgsi_exec_machine_bind_shader( machine,
+                                     base->shader.tokens,
+                                     PIPE_MAX_SAMPLERS,
+                                     samplers );
+   }
+}
+
+
+
 /**
  * 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)
+static void
+setup_pos_vector(const struct tgsi_interp_coef *coef,
+                 float x, float y,
+                 struct tgsi_exec_vector *quadpos)
 {
    uint chan;
    /* do X */
@@ -95,24 +114,6 @@ sp_setup_pos_vector(const struct tgsi_interp_coef *coef,
 }
 
 
-static void
-exec_prepare( const struct sp_fragment_shader *base,
-	      struct tgsi_exec_machine *machine,
-	      struct tgsi_sampler **samplers )
-{
-   /*
-    * Bind tokens/shader to the interpreter's machine state.
-    * Avoid redundant binding.
-    */
-   if (machine->Tokens != base->shader.tokens) {
-      tgsi_exec_machine_bind_shader( machine,
-                                     base->shader.tokens,
-                                     PIPE_MAX_SAMPLERS,
-                                     samplers );
-   }
-}
-
-
 /* TODO: hide the machine struct in here somewhere, remove from this
  * interface:
  */
@@ -122,11 +123,43 @@ exec_run( const struct sp_fragment_shader *base,
 	  struct quad_header *quad )
 {
    /* Compute X, Y, Z, W vals for this quad */
-   sp_setup_pos_vector(quad->posCoef, 
-		       (float)quad->input.x0, (float)quad->input.y0, 
-		       &machine->QuadPos);
+   setup_pos_vector(quad->posCoef, 
+                    (float)quad->input.x0, (float)quad->input.y0, 
+                    &machine->QuadPos);
    
-   return tgsi_exec_machine_run( machine );
+   quad->inout.mask &= tgsi_exec_machine_run( machine );
+   if (quad->inout.mask == 0)
+      return FALSE;
+
+   /* store outputs */
+   {
+      const ubyte *sem_name = base->info.output_semantic_name;
+      const ubyte *sem_index = base->info.output_semantic_index;
+      const uint n = base->info.num_outputs;
+      uint i;
+      for (i = 0; i < n; i++) {
+         switch (sem_name[i]) {
+         case TGSI_SEMANTIC_COLOR:
+            {
+               uint cbuf = sem_index[i];
+               memcpy(quad->output.color[cbuf],
+                      &machine->Outputs[i].xyzw[0].f[0],
+                      sizeof(quad->output.color[0]) );
+            }
+            break;
+         case TGSI_SEMANTIC_POSITION:
+            {
+               uint j;
+               for (j = 0; j < 4; j++) {
+                  quad->output.depth[j] = machine->Outputs[i].xyzw[2].f[j];
+               }
+            }
+            break;
+         }
+      }
+   }
+
+   return TRUE;
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_fs_llvm.c b/src/gallium/drivers/softpipe/sp_fs_llvm.c
deleted file mode 100644
index 95c0d982d1..0000000000
--- a/src/gallium/drivers/softpipe/sp_fs_llvm.c
+++ /dev/null
@@ -1,205 +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.
- * 
- **************************************************************************/
-
-/**
- * Execute fragment shader using LLVM code generation.
- * Authors:
- *   Zack Rusin
- */
-
-#include "sp_context.h"
-#include "sp_state.h"
-#include "sp_fs.h"
-
-#include "pipe/p_state.h"
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-#include "tgsi/tgsi_sse2.h"
-
-#if 0
-
-/**
- * Subclass of sp_fragment_shader
- */
-struct sp_llvm_fragment_shader
-{
-   struct sp_fragment_shader base;
-   struct gallivm_prog *llvm_prog;
-};
-
-
-static void
-shade_quad_llvm(struct quad_stage *qs,
-                struct quad_header *quad)
-{
-   struct quad_shade_stage *qss = quad_shade_stage(qs);
-   struct softpipe_context *softpipe = qs->softpipe;
-   float dests[4][16][4] ALIGN16_ATTRIB;
-   float inputs[4][16][4] ALIGN16_ATTRIB;
-   const float fx = (float) quad->x0;
-   const float fy = (float) quad->y0;
-   struct gallivm_prog *llvm = qss->llvm_prog;
-
-   inputs[0][0][0] = fx;
-   inputs[1][0][0] = fx + 1.0f;
-   inputs[2][0][0] = fx;
-   inputs[3][0][0] = fx + 1.0f;
-
-   inputs[0][0][1] = fy;
-   inputs[1][0][1] = fy;
-   inputs[2][0][1] = fy + 1.0f;
-   inputs[3][0][1] = fy + 1.0f;
-
-
-   gallivm_prog_inputs_interpolate(llvm, inputs, quad->coef);
-
-#if DLLVM
-   debug_printf("MASK = %d\n", quad->mask);
-   for (int i = 0; i < 4; ++i) {
-      for (int j = 0; j < 2; ++j) {
-         debug_printf("IN(%d,%d) [%f %f %f %f]\n", i, j, 
-                inputs[i][j][0], inputs[i][j][1], inputs[i][j][2], inputs[i][j][3]);
-      }
-   }
-#endif
-
-   quad->mask &=
-      gallivm_fragment_shader_exec(llvm, fx, fy, dests, inputs,
-                                   softpipe->mapped_constants[PIPE_SHADER_FRAGMENT],
-                                   qss->samplers);
-#if DLLVM
-   debug_printf("OUT LLVM = 1[%f %f %f %f], 2[%f %f %f %f]\n",
-          dests[0][0][0], dests[0][0][1], dests[0][0][2], dests[0][0][3], 
-          dests[0][1][0], dests[0][1][1], dests[0][1][2], dests[0][1][3]);
-#endif
-
-   /* store result color */
-   if (qss->colorOutSlot >= 0) {
-      unsigned i;
-      /* XXX need to handle multiple color outputs someday */
-      allvmrt(qss->stage.softpipe->fs->info.output_semantic_name[qss->colorOutSlot]
-             == TGSI_SEMANTIC_COLOR);
-      for (i = 0; i < QUAD_SIZE; ++i) {
-         quad->outputs.color[0][0][i] = dests[i][qss->colorOutSlot][0];
-         quad->outputs.color[0][1][i] = dests[i][qss->colorOutSlot][1];
-         quad->outputs.color[0][2][i] = dests[i][qss->colorOutSlot][2];
-         quad->outputs.color[0][3][i] = dests[i][qss->colorOutSlot][3];
-      }
-   }
-#if DLLVM
-   for (int i = 0; i < QUAD_SIZE; ++i) {
-      debug_printf("QLLVM%d(%d) [%f, %f, %f, %f]\n", i, qss->colorOutSlot,
-             quad->outputs.color[0][0][i],
-             quad->outputs.color[0][1][i],
-             quad->outputs.color[0][2][i],
-             quad->outputs.color[0][3][i]);
-   }
-#endif
-
-   /* store result Z */
-   if (qss->depthOutSlot >= 0) {
-      /* output[slot] is new Z */
-      uint i;
-      for (i = 0; i < 4; i++) {
-         quad->outputs.depth[i] = dests[i][0][2];
-      }
-   }
-   else {
-      /* copy input Z (which was interpolated by the executor) to output Z */
-      uint i;
-      for (i = 0; i < 4; i++) {
-         quad->outputs.depth[i] = inputs[i][0][2];
-      }
-   }
-#if DLLVM
-   debug_printf("D [%f, %f, %f, %f] mask = %d\n",
-             quad->outputs.depth[0],
-             quad->outputs.depth[1],
-             quad->outputs.depth[2],
-             quad->outputs.depth[3], quad->mask);
-#endif
-
-   /* shader may cull fragments */
-   if( quad->mask ) {
-      qs->next->run( qs->next, quad );
-   }
-}
-
-
-unsigned 
-run_llvm_fs( const struct sp_fragment_shader *base,
-	     struct foo *machine )
-{
-}
-
-
-void 
-delete_llvm_fs( struct sp_fragment_shader *base )
-{
-   FREE(base);
-}
-
-
-struct sp_fragment_shader *
-softpipe_create_fs_llvm(struct softpipe_context *softpipe,
-                        const struct pipe_shader_state *templ)
-{
-   struct sp_llvm_fragment_shader *shader = NULL;
-
-   /* LLVM fragment shaders currently disabled:
-    */
-   state = CALLOC_STRUCT(sp_llvm_shader_state);
-   if (!state)
-      return NULL;
-
-   state->llvm_prog = 0;
-
-   if (!gallivm_global_cpu_engine()) {
-      gallivm_cpu_engine_create(state->llvm_prog);
-   }
-   else
-      gallivm_cpu_jit_compile(gallivm_global_cpu_engine(), state->llvm_prog);
-   
-   if (shader) {
-      shader->base.run = run_llvm_fs;
-      shader->base.delete = delete_llvm_fs;
-   }
-
-   return shader;
-}
-
-
-#else
-
-struct sp_fragment_shader *
-softpipe_create_fs_llvm(struct softpipe_context *softpipe,
-		       const struct pipe_shader_state *templ)
-{
-   return NULL;
-}
-
-#endif
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index f4fa0905d7..9d3e4670ee 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -76,6 +76,43 @@ fs_sse_prepare( const struct sp_fragment_shader *base,
 }
 
 
+
+/**
+ * Compute quad X,Y,Z,W for the four fragments in a quad.
+ *
+ * This should really be part of the compiled shader.
+ */
+static void
+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;
+   }
+}
+
+
 /* TODO: codegenerate the whole run function, skip this wrapper.
  * TODO: break dependency on tgsi_exec_machine struct
  * TODO: push Position calculation into the generated shader
@@ -89,9 +126,9 @@ fs_sse_run( const struct sp_fragment_shader *base,
    struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base);
 
    /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */
-   sp_setup_pos_vector(quad->posCoef, 
-		       (float)quad->input.x0, (float)quad->input.y0, 
-		       machine->Temps);
+   setup_pos_vector(quad->posCoef, 
+                    (float)quad->input.x0, (float)quad->input.y0, 
+                    machine->Temps);
 
    /* init kill mask */
    tgsi_set_kill_mask(machine, 0x0);
@@ -104,7 +141,39 @@ fs_sse_run( const struct sp_fragment_shader *base,
 		 //	 , &machine->QuadPos
       );
 
-   return ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
+   quad->inout.mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]);
+   if (quad->inout.mask == 0)
+      return FALSE;
+
+   /* store outputs */
+   {
+      const ubyte *sem_name = base->info.output_semantic_name;
+      const ubyte *sem_index = base->info.output_semantic_index;
+      const uint n = base->info.num_outputs;
+      uint i;
+      for (i = 0; i < n; i++) {
+         switch (sem_name[i]) {
+         case TGSI_SEMANTIC_COLOR:
+            {
+               uint cbuf = sem_index[i];
+               memcpy(quad->output.color[cbuf],
+                      &machine->Outputs[i].xyzw[0].f[0],
+                      sizeof(quad->output.color[0]) );
+            }
+            break;
+         case TGSI_SEMANTIC_POSITION:
+            {
+               uint j;
+               for (j = 0; j < 4; j++) {
+                  quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
+               }
+            }
+            break;
+         }
+      }
+   }
+
+   return TRUE;
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
deleted file mode 100644
index 038ff04d4f..0000000000
--- a/src/gallium/drivers/softpipe/sp_prim_setup.c
+++ /dev/null
@@ -1,190 +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.
- * 
- **************************************************************************/
-
-/**
- * \brief A draw stage that drives our triangle setup routines from
- * within the draw pipeline.  One of two ways to drive setup, the
- * other being in sp_prim_vbuf.c.
- *
- * \author  Keith Whitwell <keith@tungstengraphics.com>
- * \author  Brian Paul
- */
-
-
-#include "sp_context.h"
-#include "sp_setup.h"
-#include "sp_state.h"
-#include "sp_prim_setup.h"
-#include "draw/draw_pipe.h"
-#include "draw/draw_vertex.h"
-#include "util/u_memory.h"
-
-/**
- * Triangle setup info (derived from draw_stage).
- * Also used for line drawing (taking some liberties).
- */
-struct setup_stage {
-   struct draw_stage stage; /**< This must be first (base class) */
-
-   struct setup_context *setup;
-};
-
-
-
-/**
- * Basically a cast wrapper.
- */
-static INLINE struct setup_stage *setup_stage( struct draw_stage *stage )
-{
-   return (struct setup_stage *)stage;
-}
-
-
-typedef const float (*cptrf4)[4];
-
-static void
-do_tri(struct draw_stage *stage, struct prim_header *prim)
-{
-   struct setup_stage *setup = setup_stage( stage );
-   
-   setup_tri( setup->setup,
-              (cptrf4)prim->v[0]->data,
-              (cptrf4)prim->v[1]->data,
-              (cptrf4)prim->v[2]->data );
-}
-
-static void
-do_line(struct draw_stage *stage, struct prim_header *prim)
-{
-   struct setup_stage *setup = setup_stage( stage );
-
-   setup_line( setup->setup,
-               (cptrf4)prim->v[0]->data,
-               (cptrf4)prim->v[1]->data );
-}
-
-static void
-do_point(struct draw_stage *stage, struct prim_header *prim)
-{
-   struct setup_stage *setup = setup_stage( stage );
-
-   setup_point( setup->setup,
-                (cptrf4)prim->v[0]->data );
-}
-
-
-
-
-static void setup_begin( struct draw_stage *stage )
-{
-   struct setup_stage *setup = setup_stage(stage);
-
-   setup_prepare( setup->setup );
-
-   stage->point = do_point;
-   stage->line = do_line;
-   stage->tri = do_tri;
-}
-
-
-static void setup_first_point( struct draw_stage *stage,
-			       struct prim_header *header )
-{
-   setup_begin(stage);
-   stage->point( stage, header );
-}
-
-static void setup_first_line( struct draw_stage *stage,
-			       struct prim_header *header )
-{
-   setup_begin(stage);
-   stage->line( stage, header );
-}
-
-
-static void setup_first_tri( struct draw_stage *stage,
-			       struct prim_header *header )
-{
-   setup_begin(stage);
-   stage->tri( stage, header );
-}
-
-
-
-static void setup_flush( struct draw_stage *stage,
-			 unsigned flags )
-{
-   stage->point = setup_first_point;
-   stage->line = setup_first_line;
-   stage->tri = setup_first_tri;
-}
-
-
-static void reset_stipple_counter( struct draw_stage *stage )
-{
-}
-
-
-static void render_destroy( struct draw_stage *stage )
-{
-   struct setup_stage *ssetup = setup_stage(stage);
-   setup_destroy_context(ssetup->setup);
-   FREE( stage );
-}
-
-
-/**
- * Create a new primitive setup/render stage.
- */
-struct draw_stage *sp_draw_render_stage( struct softpipe_context *softpipe )
-{
-   struct setup_stage *sstage = CALLOC_STRUCT(setup_stage);
-
-   sstage->setup = setup_create_context(softpipe);
-   sstage->stage.draw = softpipe->draw;
-   sstage->stage.point = setup_first_point;
-   sstage->stage.line = setup_first_line;
-   sstage->stage.tri = setup_first_tri;
-   sstage->stage.flush = setup_flush;
-   sstage->stage.reset_stipple_counter = reset_stipple_counter;
-   sstage->stage.destroy = render_destroy;
-
-   return (struct draw_stage *)sstage;
-}
-
-struct setup_context *
-sp_draw_setup_context( struct draw_stage *stage )
-{
-   struct setup_stage *ssetup = setup_stage(stage);
-   return ssetup->setup;
-}
-
-void
-sp_draw_flush( struct draw_stage *stage )
-{
-   stage->flush( stage, 0 );
-}
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.h b/src/gallium/drivers/softpipe/sp_prim_setup.h
deleted file mode 100644
index 49bdd98ed8..0000000000
--- a/src/gallium/drivers/softpipe/sp_prim_setup.h
+++ /dev/null
@@ -1,85 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-
-#ifndef SP_PRIM_SETUP_H
-#define SP_PRIM_SETUP_H
-
-
-/**
- * vbuf is a special stage to gather the stream of triangles, lines, points
- * together and reconstruct vertex buffers for hardware upload.
- *
- * First attempt, work in progress.
- * 
- * TODO:
- *    - separate out vertex buffer building and primitive emit, ie >1 draw per vb.
- *    - tell vbuf stage how to build hw vertices directly
- *    - pass vbuf stage a buffer pointer for direct emit to agp/vram.
- *
- *
- *
- * Vertices are just an array of floats, with all the attributes
- * packed.  We currently assume a layout like:
- *
- * attr[0][0..3] - window position
- * attr[1..n][0..3] - remaining attributes.
- *
- * Attributes are assumed to be 4 floats wide but are packed so that
- * all the enabled attributes run contiguously.
- */
-
-
-struct draw_stage;
-struct softpipe_context;
-
-
-typedef void (*vbuf_draw_func)( struct pipe_context *pipe,
-                                unsigned prim,
-                                const ushort *elements,
-                                unsigned nr_elements,
-                                const void *vertex_buffer,
-                                unsigned nr_vertices );
-
-
-extern struct draw_stage *
-sp_draw_render_stage( struct softpipe_context *softpipe );
-
-extern struct setup_context *
-sp_draw_setup_context( struct draw_stage * );
-
-extern void
-sp_draw_flush( struct draw_stage * );
-
-
-extern struct draw_stage *
-sp_draw_vbuf_stage( struct draw_context *draw_context,
-                    struct pipe_context *pipe,
-                    vbuf_draw_func draw );
-
-
-#endif /* SP_PRIM_SETUP_H */
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index 42021789ea..e603c20fc4 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -37,13 +37,13 @@
 
 
 #include "sp_context.h"
+#include "sp_setup.h"
 #include "sp_state.h"
 #include "sp_prim_vbuf.h"
-#include "sp_prim_setup.h"
-#include "sp_setup.h"
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
 #include "util/u_memory.h"
+#include "util/u_prim.h"
 
 
 #define SP_MAX_VBUF_INDEXES 1024
@@ -58,6 +58,8 @@ struct softpipe_vbuf_render
 {
    struct vbuf_render base;
    struct softpipe_context *softpipe;
+   struct setup_context *setup;
+
    uint prim;
    uint vertex_size;
    uint nr_vertices;
@@ -74,6 +76,11 @@ softpipe_vbuf_render(struct vbuf_render *vbr)
 }
 
 
+
+
+
+
+
 static const struct vertex_info *
 sp_vbuf_get_vertex_info(struct vbuf_render *vbr)
 {
@@ -104,36 +111,6 @@ sp_vbuf_allocate_vertices(struct vbuf_render *vbr,
 static void
 sp_vbuf_release_vertices(struct vbuf_render *vbr)
 {
-#if 0
-   {
-      struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
-      const struct vertex_info *info = 
-         softpipe_get_vbuf_vertex_info(cvbr->softpipe);
-      const float *vtx = (const float *) cvbr->vertex_buffer;
-      uint i, j;
-      debug_printf("%s (vtx_size = %u,  vtx_used = %u)\n",
-             __FUNCTION__, cvbr->vertex_size, cvbr->nr_vertices);
-      for (i = 0; i < cvbr->nr_vertices; i++) {
-         for (j = 0; j < info->num_attribs; j++) {
-            uint k;
-            switch (info->attrib[j].emit) {
-            case EMIT_4F:  k = 4;   break;
-            case EMIT_3F:  k = 3;   break;
-            case EMIT_2F:  k = 2;   break;
-            case EMIT_1F:  k = 1;   break;
-            default: assert(0);
-            }
-            debug_printf("Vert %u attr %u: ", i, j);
-            while (k-- > 0) {
-               debug_printf("%g ", vtx[0]);
-               vtx++;
-            }
-            debug_printf("\n");
-         }
-      }
-   }
-#endif
-
    /* keep the old allocation for next time */
 }
 
@@ -159,14 +136,11 @@ static boolean
 sp_vbuf_set_primitive(struct vbuf_render *vbr, unsigned prim)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
-
-   /* XXX: break this dependency - make setup_context live under
-    * softpipe, rename the old "setup" draw stage to something else.
-    */
-   struct setup_context *setup_ctx = sp_draw_setup_context(cvbr->softpipe->setup);
+   struct setup_context *setup_ctx = cvbr->setup;
    
    setup_prepare( setup_ctx );
 
+   cvbr->softpipe->reduced_prim = u_reduced_prim(prim);
    cvbr->prim = prim;
    return TRUE;
 
@@ -191,14 +165,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
    struct softpipe_context *softpipe = cvbr->softpipe;
    const unsigned stride = softpipe->vertex_info_vbuf.size * sizeof(float);
    const void *vertex_buffer = cvbr->vertex_buffer;
+   struct setup_context *setup_ctx = cvbr->setup;
    unsigned i;
 
-   /* XXX: break this dependency - make setup_context live under
-    * softpipe, rename the old "setup" draw stage to something else.
-    */
-   struct draw_stage *setup = softpipe->setup;
-   struct setup_context *setup_ctx = sp_draw_setup_context(setup);
-
    switch (cvbr->prim) {
    case PIPE_PRIM_POINTS:
       for (i = 0; i < nr; i++) {
@@ -237,14 +206,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLES:
-      for (i = 2; i < nr; i += 3) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i += 3) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-1], stride),
                        get_vert(vertex_buffer, indices[i-0], stride),
                        get_vert(vertex_buffer, indices[i-2], stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i += 3) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
@@ -254,14 +225,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLE_STRIP:
-      for (i = 2; i < nr; i += 1) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i+(i&1)-1], stride),
                        get_vert(vertex_buffer, indices[i-(i&1)], stride),
                        get_vert(vertex_buffer, indices[i-2], stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i+(i&1)-2], stride),
                        get_vert(vertex_buffer, indices[i-(i&1)-1], stride),
@@ -271,14 +244,16 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLE_FAN:
-      for (i = 2; i < nr; i += 1) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-0], stride),
                        get_vert(vertex_buffer, indices[0], stride),
                        get_vert(vertex_buffer, indices[i-1], stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[0], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
@@ -288,8 +263,8 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      for (i = 3; i < nr; i += 4) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 3; i < nr; i += 4) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
@@ -299,7 +274,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
                        get_vert(vertex_buffer, indices[i-0], stride),
                        get_vert(vertex_buffer, indices[i-3], stride) );
          }
-         else {
+      }
+      else {
+         for (i = 3; i < nr; i += 4) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
@@ -314,8 +291,8 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      for (i = 3; i < nr; i += 2) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 3; i < nr; i += 2) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-0], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
@@ -325,7 +302,9 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
                        get_vert(vertex_buffer, indices[i-0], stride),
                        get_vert(vertex_buffer, indices[i-3], stride) );
          }
-         else {
+      }
+      else {
+         for (i = 3; i < nr; i += 2) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
@@ -355,11 +334,6 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
    default:
       assert(0);
    }
-
-   /* XXX: why are we calling this???  If we had to call something, it
-    * would be a function in sp_setup.c:
-    */
-   sp_draw_flush( setup );
 }
 
 
@@ -372,17 +346,12 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
    struct softpipe_context *softpipe = cvbr->softpipe;
+   struct setup_context *setup_ctx = cvbr->setup;
    const unsigned stride = softpipe->vertex_info_vbuf.size * sizeof(float);
    const void *vertex_buffer =
       (void *) get_vert(cvbr->vertex_buffer, start, stride);
    unsigned i;
 
-   /* XXX: break this dependency - make setup_context live under
-    * softpipe, rename the old "setup" draw stage to something else.
-    */
-   struct draw_stage *setup = softpipe->setup;
-   struct setup_context *setup_ctx = sp_draw_setup_context(setup);
-
    switch (cvbr->prim) {
    case PIPE_PRIM_POINTS:
       for (i = 0; i < nr; i++) {
@@ -421,14 +390,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLES:
-      for (i = 2; i < nr; i += 3) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i += 3) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-1, stride),
                        get_vert(vertex_buffer, i-0, stride),
                        get_vert(vertex_buffer, i-2, stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i += 3) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-1, stride),
@@ -438,14 +409,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLE_STRIP:
-      for (i = 2; i < nr; i++) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i++) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i+(i&1)-1, stride),
                        get_vert(vertex_buffer, i-(i&1), stride),
                        get_vert(vertex_buffer, i-2, stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i++) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i+(i&1)-2, stride),
                        get_vert(vertex_buffer, i-(i&1)-1, stride),
@@ -455,14 +428,16 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_TRIANGLE_FAN:
-      for (i = 2; i < nr; i += 1) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-0, stride),
                        get_vert(vertex_buffer, 0, stride),
                        get_vert(vertex_buffer, i-1, stride) );
          }
-         else {
+      }
+      else {
+         for (i = 2; i < nr; i += 1) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, 0, stride),
                        get_vert(vertex_buffer, i-1, stride),
@@ -472,8 +447,8 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      for (i = 3; i < nr; i += 4) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 3; i < nr; i += 4) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-1, stride),
@@ -483,7 +458,9 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
                        get_vert(vertex_buffer, i-0, stride),
                        get_vert(vertex_buffer, i-3, stride) );
          }
-         else {
+      }
+      else {
+         for (i = 3; i < nr; i += 4) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
@@ -497,8 +474,8 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      for (i = 3; i < nr; i += 2) {
-         if (softpipe->rasterizer->flatshade_first) {
+      if (softpipe->rasterizer->flatshade_first) {
+         for (i = 3; i < nr; i += 2) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-0, stride),
                        get_vert(vertex_buffer, i-1, stride),
@@ -508,7 +485,9 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
                        get_vert(vertex_buffer, i-0, stride),
                        get_vert(vertex_buffer, i-3, stride) );
          }
-         else {
+      }
+      else {
+         for (i = 3; i < nr; i += 2) {
             setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
@@ -546,40 +525,38 @@ static void
 sp_vbuf_destroy(struct vbuf_render *vbr)
 {
    struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
-   cvbr->softpipe->vbuf_render = NULL;
+   setup_destroy_context(cvbr->setup);
    FREE(cvbr);
 }
 
 
 /**
- * Initialize the post-transform vertex buffer information for the given
- * context.
+ * Create the post-transform vertex handler for the given context.
  */
-void
-sp_init_vbuf(struct softpipe_context *sp)
+struct vbuf_render *
+sp_create_vbuf_backend(struct softpipe_context *sp)
 {
-   assert(sp->draw);
+   struct softpipe_vbuf_render *cvbr = CALLOC_STRUCT(softpipe_vbuf_render);
 
-   sp->vbuf_render = CALLOC_STRUCT(softpipe_vbuf_render);
+   assert(sp->draw);
 
-   sp->vbuf_render->base.max_indices = SP_MAX_VBUF_INDEXES;
-   sp->vbuf_render->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE;
 
-   sp->vbuf_render->base.get_vertex_info = sp_vbuf_get_vertex_info;
-   sp->vbuf_render->base.allocate_vertices = sp_vbuf_allocate_vertices;
-   sp->vbuf_render->base.map_vertices = sp_vbuf_map_vertices;
-   sp->vbuf_render->base.unmap_vertices = sp_vbuf_unmap_vertices;
-   sp->vbuf_render->base.set_primitive = sp_vbuf_set_primitive;
-   sp->vbuf_render->base.draw = sp_vbuf_draw;
-   sp->vbuf_render->base.draw_arrays = sp_vbuf_draw_arrays;
-   sp->vbuf_render->base.release_vertices = sp_vbuf_release_vertices;
-   sp->vbuf_render->base.destroy = sp_vbuf_destroy;
+   cvbr->base.max_indices = SP_MAX_VBUF_INDEXES;
+   cvbr->base.max_vertex_buffer_bytes = SP_MAX_VBUF_SIZE;
 
-   sp->vbuf_render->softpipe = sp;
+   cvbr->base.get_vertex_info = sp_vbuf_get_vertex_info;
+   cvbr->base.allocate_vertices = sp_vbuf_allocate_vertices;
+   cvbr->base.map_vertices = sp_vbuf_map_vertices;
+   cvbr->base.unmap_vertices = sp_vbuf_unmap_vertices;
+   cvbr->base.set_primitive = sp_vbuf_set_primitive;
+   cvbr->base.draw = sp_vbuf_draw;
+   cvbr->base.draw_arrays = sp_vbuf_draw_arrays;
+   cvbr->base.release_vertices = sp_vbuf_release_vertices;
+   cvbr->base.destroy = sp_vbuf_destroy;
 
-   sp->vbuf = draw_vbuf_stage(sp->draw, &sp->vbuf_render->base);
+   cvbr->softpipe = sp;
 
-   draw_set_rasterize_stage(sp->draw, sp->vbuf);
+   cvbr->setup = setup_create_context(cvbr->softpipe);
 
-   draw_set_render(sp->draw, &sp->vbuf_render->base);
+   return &cvbr->base;
 }
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.h b/src/gallium/drivers/softpipe/sp_prim_vbuf.h
index 1de9cc2a89..ad01cc2f28 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.h
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.h
@@ -31,8 +31,8 @@
 
 struct softpipe_context;
 
-extern void
-sp_init_vbuf(struct softpipe_context *softpipe);
+extern struct vbuf_render *
+sp_create_vbuf_backend(struct softpipe_context *softpipe);
 
 
 #endif /* SP_VBUF_H */
diff --git a/src/gallium/drivers/softpipe/sp_quad.h b/src/gallium/drivers/softpipe/sp_quad.h
index bd6c6cb912..a3236bd116 100644
--- a/src/gallium/drivers/softpipe/sp_quad.h
+++ b/src/gallium/drivers/softpipe/sp_quad.h
@@ -97,10 +97,10 @@ struct quad_header {
    struct quad_header_inout inout;
    struct quad_header_output output;
 
-   const struct tgsi_interp_coef *coef;
+   /* Redundant/duplicated:
+    */
    const struct tgsi_interp_coef *posCoef;
-
-   unsigned nr_attrs;
+   const struct tgsi_interp_coef *coef;
 };
 
 #endif /* SP_QUAD_H */
diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
deleted file mode 100644
index 0845bae0e6..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
+++ /dev/null
@@ -1,108 +0,0 @@
-
-/**
- * quad alpha test
- */
-
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_quad_pipe.h"
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-
-
-static void
-alpha_test_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-   const float ref = softpipe->depth_stencil->alpha.ref_value;
-   unsigned passMask = 0x0, j;
-   const uint cbuf = 0; /* only output[0].alpha is tested */
-   const float *aaaa = quad->output.color[cbuf][3];
-
-   switch (softpipe->depth_stencil->alpha.func) {
-   case PIPE_FUNC_NEVER:
-      break;
-   case PIPE_FUNC_LESS:
-      /*
-       * If mask were an array [4] we could do this SIMD-style:
-       * passMask = (quad->outputs.color[0][3] <= vec4(ref));
-       */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] < ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_EQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] == ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_LEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] <= ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_GREATER:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] > ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_NOTEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] != ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_GEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (aaaa[j] >= ref) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_ALWAYS:
-      passMask = MASK_ALL;
-      break;
-   default:
-      assert(0);
-   }
-
-   quad->inout.mask &= passMask;
-
-   if (quad->inout.mask)
-      qs->next->run(qs->next, quad);
-}
-
-
-static void alpha_test_begin(struct quad_stage *qs)
-{
-   qs->next->begin(qs->next);
-}
-
-
-static void alpha_test_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-struct quad_stage *
-sp_quad_alpha_test_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = alpha_test_begin;
-   stage->run = alpha_test_quad;
-   stage->destroy = alpha_test_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
index b1e18805c7..e243c63fa2 100644
--- a/src/gallium/drivers/softpipe/sp_quad_blend.c
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -117,644 +117,865 @@ do { \
 
 
 static void
-logicop_quad(struct quad_stage *qs, struct quad_header *quad)
+logicop_quad(struct quad_stage *qs, 
+             float (*quadColor)[4],
+             float (*dest)[4])
 {
    struct softpipe_context *softpipe = qs->softpipe;
-   uint cbuf;
+   ubyte src[4][4], dst[4][4], res[4][4];
+   uint *src4 = (uint *) src;
+   uint *dst4 = (uint *) dst;
+   uint *res4 = (uint *) res;
+   uint j;
+
+
+   /* convert to ubyte */
+   for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
+      dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */
+      dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */
+      dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */
+      dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */
+
+      src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */
+      src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */
+      src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */
+      src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
+   }
 
-   /* loop over colorbuffer outputs */
-   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
-      float dest[4][QUAD_SIZE];
-      ubyte src[4][4], dst[4][4], res[4][4];
-      uint *src4 = (uint *) src;
-      uint *dst4 = (uint *) dst;
-      uint *res4 = (uint *) res;
-      struct softpipe_cached_tile *
-         tile = sp_get_cached_tile(softpipe,
-                                   softpipe->cbuf_cache[cbuf],
-                                   quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
-      uint i, j;
+   switch (softpipe->blend->logicop_func) {
+   case PIPE_LOGICOP_CLEAR:
+      for (j = 0; j < 4; j++)
+         res4[j] = 0;
+      break;
+   case PIPE_LOGICOP_NOR:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] | dst4[j]);
+      break;
+   case PIPE_LOGICOP_AND_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j] & dst4[j];
+      break;
+   case PIPE_LOGICOP_COPY_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j];
+      break;
+   case PIPE_LOGICOP_AND_REVERSE:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] & ~dst4[j];
+      break;
+   case PIPE_LOGICOP_INVERT:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~dst4[j];
+      break;
+   case PIPE_LOGICOP_XOR:
+      for (j = 0; j < 4; j++)
+         res4[j] = dst4[j] ^ src4[j];
+      break;
+   case PIPE_LOGICOP_NAND:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] & dst4[j]);
+      break;
+   case PIPE_LOGICOP_AND:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] & dst4[j];
+      break;
+   case PIPE_LOGICOP_EQUIV:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~(src4[j] ^ dst4[j]);
+      break;
+   case PIPE_LOGICOP_NOOP:
+      for (j = 0; j < 4; j++)
+         res4[j] = dst4[j];
+      break;
+   case PIPE_LOGICOP_OR_INVERTED:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~src4[j] | dst4[j];
+      break;
+   case PIPE_LOGICOP_COPY:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j];
+      break;
+   case PIPE_LOGICOP_OR_REVERSE:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] | ~dst4[j];
+      break;
+   case PIPE_LOGICOP_OR:
+      for (j = 0; j < 4; j++)
+         res4[j] = src4[j] | dst4[j];
+      break;
+   case PIPE_LOGICOP_SET:
+      for (j = 0; j < 4; j++)
+         res4[j] = ~0;
+      break;
+   default:
+      assert(0);
+   }
 
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
-         }
-      }
+   for (j = 0; j < 4; j++) {
+      quadColor[j][0] = ubyte_to_float(res[j][0]);
+      quadColor[j][1] = ubyte_to_float(res[j][1]);
+      quadColor[j][2] = ubyte_to_float(res[j][2]);
+      quadColor[j][3] = ubyte_to_float(res[j][3]);
+   }
+}
 
-      /* convert to ubyte */
-      for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
-         dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */
-         dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */
-         dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */
-         dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */
-
-         src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */
-         src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */
-         src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */
-         src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
-      }
 
-      switch (softpipe->blend->logicop_func) {
-      case PIPE_LOGICOP_CLEAR:
-         for (j = 0; j < 4; j++)
-            res4[j] = 0;
-         break;
-      case PIPE_LOGICOP_NOR:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] | dst4[j]);
-         break;
-      case PIPE_LOGICOP_AND_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j] & dst4[j];
-         break;
-      case PIPE_LOGICOP_COPY_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j];
-         break;
-      case PIPE_LOGICOP_AND_REVERSE:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] & ~dst4[j];
-         break;
-      case PIPE_LOGICOP_INVERT:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~dst4[j];
-         break;
-      case PIPE_LOGICOP_XOR:
-         for (j = 0; j < 4; j++)
-            res4[j] = dst4[j] ^ src4[j];
-         break;
-      case PIPE_LOGICOP_NAND:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] & dst4[j]);
-         break;
-      case PIPE_LOGICOP_AND:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] & dst4[j];
-         break;
-      case PIPE_LOGICOP_EQUIV:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] ^ dst4[j]);
-         break;
-      case PIPE_LOGICOP_NOOP:
-         for (j = 0; j < 4; j++)
-            res4[j] = dst4[j];
-         break;
-      case PIPE_LOGICOP_OR_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j] | dst4[j];
-         break;
-      case PIPE_LOGICOP_COPY:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j];
-         break;
-      case PIPE_LOGICOP_OR_REVERSE:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] | ~dst4[j];
-         break;
-      case PIPE_LOGICOP_OR:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] | dst4[j];
-         break;
-      case PIPE_LOGICOP_SET:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~0;
-         break;
-      default:
-         assert(0);
-      }
 
-      for (j = 0; j < 4; j++) {
-         quadColor[j][0] = ubyte_to_float(res[j][0]);
-         quadColor[j][1] = ubyte_to_float(res[j][1]);
-         quadColor[j][2] = ubyte_to_float(res[j][2]);
-         quadColor[j][3] = ubyte_to_float(res[j][3]);
-      }
+static void
+blend_quad(struct quad_stage *qs, 
+           float (*quadColor)[4],
+           float (*dest)[4])
+{
+   static const float zero[4] = { 0, 0, 0, 0 };
+   static const float one[4] = { 1, 1, 1, 1 };
+   struct softpipe_context *softpipe = qs->softpipe;
+   float source[4][QUAD_SIZE];
+
+   /*
+    * Compute src/first term RGB
+    */
+   switch (softpipe->blend->rgb_src_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      VEC4_COPY(source[0], quadColor[0]); /* R */
+      VEC4_COPY(source[1], quadColor[1]); /* G */
+      VEC4_COPY(source[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */
+      VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */
+      VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+   {
+      const float *alpha = quadColor[3];
+      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */
+      VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */
+      VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+   {
+      const float *alpha = dest[3];
+      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+   {
+      const float *alpha = quadColor[3];
+      float diff[4], temp[4];
+      VEC4_SUB(diff, one, dest[3]);
+      VEC4_MIN(temp, alpha, diff);
+      VEC4_MUL(source[0], quadColor[0], temp); /* R */
+      VEC4_MUL(source[1], quadColor[1], temp); /* G */
+      VEC4_MUL(source[2], quadColor[2], temp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+   {
+      float comp[4];
+      VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
+      VEC4_MUL(source[0], quadColor[0], comp); /* R */
+      VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
+      VEC4_MUL(source[1], quadColor[1], comp); /* G */
+      VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
+      VEC4_MUL(source[2], quadColor[2], comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+   {
+      float alpha[4];
+      VEC4_SCALAR(alpha, softpipe->blend_color.color[3]);
+      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_SRC1_COLOR:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(source[0], zero); /* R */
+      VEC4_COPY(source[1], zero); /* G */
+      VEC4_COPY(source[2], zero); /* B */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
+      VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
+      VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
+      VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
+      VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
+      VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+   {
+      float inv_alpha[4];
+      VEC4_SUB(inv_alpha, one, quadColor[3]);
+      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+   {
+      float inv_alpha[4];
+      VEC4_SUB(inv_alpha, one, dest[3]);
+      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, dest[0]); /* R */
+      VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
+      VEC4_SUB(inv_comp, one, dest[1]); /* G */
+      VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
+      VEC4_SUB(inv_comp, one, dest[2]); /* B */
+      VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+   {
+      float inv_comp[4];
+      /* R */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
+      VEC4_MUL(source[0], quadColor[0], inv_comp);
+      /* G */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
+      VEC4_MUL(source[1], quadColor[1], inv_comp);
+      /* B */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
+      VEC4_MUL(source[2], quadColor[2], inv_comp);
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+   {
+      float inv_alpha[4];
+      VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]);
+      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+      assert(0); /* to do */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+      assert(0); /* to do */
+      break;
+   default:
+      assert(0);
+   }
+
+   /*
+    * Compute src/first term A
+    */
+   switch (softpipe->blend->alpha_src_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      VEC4_COPY(source[3], quadColor[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+   {
+      const float *alpha = quadColor[3];
+      VEC4_MUL(source[3], quadColor[3], alpha); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      /* multiply alpha by 1.0 */
+      VEC4_COPY(source[3], quadColor[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+   {
+      float comp[4];
+      VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+      VEC4_MUL(source[3], quadColor[3], comp); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(source[3], zero); /* A */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+   {
+      float inv_alpha[4];
+      VEC4_SUB(inv_alpha, one, quadColor[3]);
+      VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+   {
+      float inv_alpha[4];
+      VEC4_SUB(inv_alpha, one, dest[3]);
+      VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+   {
+      float inv_comp[4];
+      /* A */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+      VEC4_MUL(source[3], quadColor[3], inv_comp);
+   }
+   break;
+   default:
+      assert(0);
    }
 
-   /* pass quad to next stage */
-   qs->next->run(qs->next, quad);
-}
 
+   /*
+    * Compute dest/second term RGB
+    */
+   switch (softpipe->blend->rgb_dst_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      /* dest = dest * 1   NO-OP, leave dest as-is */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */
+      VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */
+      VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */
+      VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */
+      VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */
+      VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */
+      VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      VEC4_MUL(dest[0], dest[0], dest[0]); /* R */
+      VEC4_MUL(dest[1], dest[1], dest[1]); /* G */
+      VEC4_MUL(dest[2], dest[2], dest[2]); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      assert(0); /* illegal */
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+   {
+      float comp[4];
+      VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
+      VEC4_MUL(dest[0], dest[0], comp); /* R */
+      VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
+      VEC4_MUL(dest[1], dest[1], comp); /* G */
+      VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
+      VEC4_MUL(dest[2], dest[2], comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+   {
+      float comp[4];
+      VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+      VEC4_MUL(dest[0], dest[0], comp); /* R */
+      VEC4_MUL(dest[1], dest[1], comp); /* G */
+      VEC4_MUL(dest[2], dest[2], comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(dest[0], zero); /* R */
+      VEC4_COPY(dest[1], zero); /* G */
+      VEC4_COPY(dest[2], zero); /* B */
+      break;
+   case PIPE_BLENDFACTOR_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_SRC1_ALPHA:
+      /* XXX what are these? */
+      assert(0);
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
+      VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
+      VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
+      VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
+      VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
+      VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+   {
+      float one_minus_alpha[QUAD_SIZE];
+      VEC4_SUB(one_minus_alpha, one, quadColor[3]);
+      VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
+      VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
+      VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, dest[3]); /* A */
+      VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
+      VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
+      VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, dest[0]); /* R */
+      VEC4_MUL(dest[0], dest[0], inv_comp); /* R */
+      VEC4_SUB(inv_comp, one, dest[1]); /* G */
+      VEC4_MUL(dest[1], dest[1], inv_comp); /* G */
+      VEC4_SUB(inv_comp, one, dest[2]); /* B */
+      VEC4_MUL(dest[2], dest[2], inv_comp); /* B */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+   {
+      float inv_comp[4];
+      /* R */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
+      VEC4_MUL(dest[0], dest[0], inv_comp);
+      /* G */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
+      VEC4_MUL(dest[1], dest[1], inv_comp);
+      /* B */
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
+      VEC4_MUL(dest[2], dest[2], inv_comp);
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+   {
+      float inv_comp[4];
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+      VEC4_MUL(dest[0], dest[0], inv_comp);
+      VEC4_MUL(dest[1], dest[1], inv_comp);
+      VEC4_MUL(dest[2], dest[2], inv_comp);
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+      /* XXX what are these? */
+      assert(0);
+      break;
+   default:
+      assert(0);
+   }
 
+   /*
+    * Compute dest/second term A
+    */
+   switch (softpipe->blend->alpha_dst_factor) {
+   case PIPE_BLENDFACTOR_ONE:
+      /* dest = dest * 1   NO-OP, leave dest as-is */
+      break;
+   case PIPE_BLENDFACTOR_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_SRC_ALPHA:
+      VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */
+      break;
+   case PIPE_BLENDFACTOR_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_DST_ALPHA:
+      VEC4_MUL(dest[3], dest[3], dest[3]); /* A */
+      break;
+   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+      assert(0); /* illegal */
+      break;
+   case PIPE_BLENDFACTOR_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_CONST_ALPHA:
+   {
+      float comp[4];
+      VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
+      VEC4_MUL(dest[3], dest[3], comp); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_ZERO:
+      VEC4_COPY(dest[3], zero); /* A */
+      break;
+   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+   {
+      float one_minus_alpha[QUAD_SIZE];
+      VEC4_SUB(one_minus_alpha, one, quadColor[3]);
+      VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_DST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+   {
+      float inv_comp[4];
+      VEC4_SUB(inv_comp, one, dest[3]); /* A */
+      VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */
+   }
+   break;
+   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+      /* fall-through */
+   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+   {
+      float inv_comp[4];
+      VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
+      VEC4_MUL(dest[3], dest[3], inv_comp);
+   }
+   break;
+   default:
+      assert(0);
+   }
 
+   /*
+    * Combine RGB terms
+    */
+   switch (softpipe->blend->rgb_func) {
+   case PIPE_BLEND_ADD:
+      VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_SUBTRACT:
+      VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */
+      VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */
+      VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */
+      break;
+   case PIPE_BLEND_MIN:
+      VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   case PIPE_BLEND_MAX:
+      VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */
+      break;
+   default:
+      assert(0);
+   }
+
+   /*
+    * Combine A terms
+    */
+   switch (softpipe->blend->alpha_func) {
+   case PIPE_BLEND_ADD:
+      VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_SUBTRACT:
+      VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_REVERSE_SUBTRACT:
+      VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */
+      break;
+   case PIPE_BLEND_MIN:
+      VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   case PIPE_BLEND_MAX:
+      VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */
+      break;
+   default:
+      assert(0);
+   }
+}
 
 static void
-blend_quad(struct quad_stage *qs, struct quad_header *quad)
+colormask_quad(struct quad_stage *qs,
+               float (*quadColor)[4],
+               float (*dest)[4])
 {
-   static const float zero[4] = { 0, 0, 0, 0 };
-   static const float one[4] = { 1, 1, 1, 1 };
+   struct softpipe_context *softpipe = qs->softpipe;
 
+   /* R */
+   if (!(softpipe->blend->colormask & PIPE_MASK_R))
+      COPY_4V(quadColor[0], dest[0]);
+
+   /* G */
+   if (!(softpipe->blend->colormask & PIPE_MASK_G))
+      COPY_4V(quadColor[1], dest[1]);
+
+   /* B */
+   if (!(softpipe->blend->colormask & PIPE_MASK_B))
+      COPY_4V(quadColor[2], dest[2]);
+
+   /* A */
+   if (!(softpipe->blend->colormask & PIPE_MASK_A))
+      COPY_4V(quadColor[3], dest[3]);
+}
+
+
+static void
+blend_fallback(struct quad_stage *qs, 
+               struct quad_header *quads[],
+               unsigned nr)
+{
    struct softpipe_context *softpipe = qs->softpipe;
-   uint cbuf;
+   const struct pipe_blend_state *blend = softpipe->blend;
+   unsigned cbuf;
+
+   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) 
+   {
+      float dest[4][QUAD_SIZE];
+      struct softpipe_cached_tile *tile
+         = sp_get_cached_tile(softpipe->cbuf_cache[cbuf],
+                              quads[0]->input.x0, 
+                              quads[0]->input.y0);
+      uint q, i, j;
+
+      for (q = 0; q < nr; q++) {
+         struct quad_header *quad = quads[q];
+         float (*quadColor)[4] = quad->output.color[cbuf];
+         const int itx = (quad->input.x0 & (TILE_SIZE-1));
+         const int ity = (quad->input.y0 & (TILE_SIZE-1));
+
+         /* get/swizzle dest colors 
+          */
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = itx + (j & 1);
+            int y = ity + (j >> 1);
+            for (i = 0; i < 4; i++) {
+               dest[i][j] = tile->data.color[y][x][i];
+            }
+         }
 
-   if (softpipe->blend->logicop_enable) {
-      logicop_quad(qs, quad);
-      return;
+
+         if (blend->logicop_enable) {
+            logicop_quad( qs, quadColor, dest );
+         }
+         else if (blend->blend_enable) {
+            blend_quad( qs, quadColor, dest );
+         }
+
+         if (blend->colormask != 0xf)
+            colormask_quad( qs, quadColor, dest );
+   
+         /* Output color values
+          */
+         for (j = 0; j < QUAD_SIZE; j++) {
+            if (quad->inout.mask & (1 << j)) {
+               int x = itx + (j & 1);
+               int y = ity + (j >> 1);
+               for (i = 0; i < 4; i++) { /* loop over color chans */
+                  tile->data.color[y][x][i] = quadColor[i][j];
+               }
+            }
+         }
+      }
    }
+}
 
-   /* loop over colorbuffer outputs */
-   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
-      float source[4][QUAD_SIZE], dest[4][QUAD_SIZE];
-      struct softpipe_cached_tile *tile
-         = sp_get_cached_tile(softpipe,
-                              softpipe->cbuf_cache[cbuf],
-                              quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
-      uint i, j;
 
+static void
+blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, 
+                                         struct quad_header *quads[],
+                                         unsigned nr)
+{
+   static const float one[4] = { 1, 1, 1, 1 };
+   float one_minus_alpha[QUAD_SIZE];
+   float dest[4][QUAD_SIZE];
+   float source[4][QUAD_SIZE];
+   uint i, j, q;
+
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(qs->softpipe->cbuf_cache[0],
+                           quads[0]->input.x0, 
+                           quads[0]->input.y0);
+
+   for (q = 0; q < nr; q++) {
+      struct quad_header *quad = quads[q];
+      float (*quadColor)[4] = quad->output.color[0];
+      const float *alpha = quadColor[3];
+      const int itx = (quad->input.x0 & (TILE_SIZE-1));
+      const int ity = (quad->input.y0 & (TILE_SIZE-1));
+      
       /* get/swizzle dest colors */
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
+         int x = itx + (j & 1);
+         int y = ity + (j >> 1);
          for (i = 0; i < 4; i++) {
             dest[i][j] = tile->data.color[y][x][i];
          }
       }
 
-      /*
-       * Compute src/first term RGB
-       */
-      switch (softpipe->blend->rgb_src_factor) {
-      case PIPE_BLENDFACTOR_ONE:
-         VEC4_COPY(source[0], quadColor[0]); /* R */
-         VEC4_COPY(source[1], quadColor[1]); /* G */
-         VEC4_COPY(source[2], quadColor[2]); /* B */
-         break;
-      case PIPE_BLENDFACTOR_SRC_COLOR:
-         VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */
-         VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */
-         VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA:
-         {
-            const float *alpha = quadColor[3];
-            VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_DST_COLOR:
-         VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */
-         VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */
-         VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */
-         break;
-      case PIPE_BLENDFACTOR_DST_ALPHA:
-         {
-            const float *alpha = dest[3];
-            VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-         {
-            const float *alpha = quadColor[3];
-            float diff[4], temp[4];
-            VEC4_SUB(diff, one, dest[3]);
-            VEC4_MIN(temp, alpha, diff);
-            VEC4_MUL(source[0], quadColor[0], temp); /* R */
-            VEC4_MUL(source[1], quadColor[1], temp); /* G */
-            VEC4_MUL(source[2], quadColor[2], temp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_CONST_COLOR:
-         {
-            float comp[4];
-            VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
-            VEC4_MUL(source[0], quadColor[0], comp); /* R */
-            VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
-            VEC4_MUL(source[1], quadColor[1], comp); /* G */
-            VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
-            VEC4_MUL(source[2], quadColor[2], comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
-         {
-            float alpha[4];
-            VEC4_SCALAR(alpha, softpipe->blend_color.color[3]);
-            VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_SRC1_COLOR:
-         assert(0); /* to do */
-         break;
-      case PIPE_BLENDFACTOR_SRC1_ALPHA:
-         assert(0); /* to do */
-         break;
-      case PIPE_BLENDFACTOR_ZERO:
-         VEC4_COPY(source[0], zero); /* R */
-         VEC4_COPY(source[1], zero); /* G */
-         VEC4_COPY(source[2], zero); /* B */
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
-            VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
-            VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
-            VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
-            VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
-            VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-         {
-            float inv_alpha[4];
-            VEC4_SUB(inv_alpha, one, quadColor[3]);
-            VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-         {
-            float inv_alpha[4];
-            VEC4_SUB(inv_alpha, one, dest[3]);
-            VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, dest[0]); /* R */
-            VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
-            VEC4_SUB(inv_comp, one, dest[1]); /* G */
-            VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
-            VEC4_SUB(inv_comp, one, dest[2]); /* B */
-            VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-         {
-            float inv_comp[4];
-            /* R */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
-            VEC4_MUL(source[0], quadColor[0], inv_comp);
-            /* G */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
-            VEC4_MUL(source[1], quadColor[1], inv_comp);
-            /* B */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
-            VEC4_MUL(source[2], quadColor[2], inv_comp);
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-         {
-            float inv_alpha[4];
-            VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]);
-            VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-            VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-            VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
+      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
+      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
+      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
+      VEC4_MUL(source[3], quadColor[3], alpha); /* A */
+
+      VEC4_SUB(one_minus_alpha, one, alpha);
+      VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
+      VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
+      VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
+      VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* B */
+
+      VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
+      VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
+      VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */
+      VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
+
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (quad->inout.mask & (1 << j)) {
+            int x = itx + (j & 1);
+            int y = ity + (j >> 1);
+            for (i = 0; i < 4; i++) { /* loop over color chans */
+               tile->data.color[y][x][i] = quadColor[i][j];
+            }
          }
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-         assert(0); /* to do */
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
-         assert(0); /* to do */
-         break;
-      default:
-         assert(0);
       }
+   }
+}
 
-      /*
-       * Compute src/first term A
-       */
-      switch (softpipe->blend->alpha_src_factor) {
-      case PIPE_BLENDFACTOR_ONE:
-         VEC4_COPY(source[3], quadColor[3]); /* A */
-         break;
-      case PIPE_BLENDFACTOR_SRC_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_SRC_ALPHA:
-         {
-            const float *alpha = quadColor[3];
-            VEC4_MUL(source[3], quadColor[3], alpha); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_DST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_DST_ALPHA:
-         VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-         /* multiply alpha by 1.0 */
-         VEC4_COPY(source[3], quadColor[3]); /* A */
-         break;
-      case PIPE_BLENDFACTOR_CONST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
-         {
-            float comp[4];
-            VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
-            VEC4_MUL(source[3], quadColor[3], comp); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_ZERO:
-         VEC4_COPY(source[3], zero); /* A */
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-         {
-            float inv_alpha[4];
-            VEC4_SUB(inv_alpha, one, quadColor[3]);
-            VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-         {
-            float inv_alpha[4];
-            VEC4_SUB(inv_alpha, one, dest[3]);
-            VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-         {
-            float inv_comp[4];
-            /* A */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
-            VEC4_MUL(source[3], quadColor[3], inv_comp);
+static void
+blend_single_add_one_one(struct quad_stage *qs, 
+                         struct quad_header *quads[],
+                         unsigned nr)
+{
+   float dest[4][QUAD_SIZE];
+   uint i, j, q;
+
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(qs->softpipe->cbuf_cache[0],
+                           quads[0]->input.x0, 
+                           quads[0]->input.y0);
+
+   for (q = 0; q < nr; q++) {
+      struct quad_header *quad = quads[q];
+      float (*quadColor)[4] = quad->output.color[0];
+      const int itx = (quad->input.x0 & (TILE_SIZE-1));
+      const int ity = (quad->input.y0 & (TILE_SIZE-1));
+      
+      /* get/swizzle dest colors */
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = itx + (j & 1);
+         int y = ity + (j >> 1);
+         for (i = 0; i < 4; i++) {
+            dest[i][j] = tile->data.color[y][x][i];
          }
-         break;
-      default:
-         assert(0);
       }
+     
+      VEC4_ADD_SAT(quadColor[0], quadColor[0], dest[0]); /* R */
+      VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */
+      VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */
+      VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */
 
-
-      /*
-       * Compute dest/second term RGB
-       */
-      switch (softpipe->blend->rgb_dst_factor) {
-      case PIPE_BLENDFACTOR_ONE:
-         /* dest = dest * 1   NO-OP, leave dest as-is */
-         break;
-      case PIPE_BLENDFACTOR_SRC_COLOR:
-         VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */
-         VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */
-         VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA:
-         VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */
-         VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */
-         VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */
-         break;
-      case PIPE_BLENDFACTOR_DST_ALPHA:
-         VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */
-         VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */
-         VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */
-         break;
-      case PIPE_BLENDFACTOR_DST_COLOR:
-         VEC4_MUL(dest[0], dest[0], dest[0]); /* R */
-         VEC4_MUL(dest[1], dest[1], dest[1]); /* G */
-         VEC4_MUL(dest[2], dest[2], dest[2]); /* B */
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-         assert(0); /* illegal */
-         break;
-      case PIPE_BLENDFACTOR_CONST_COLOR:
-         {
-            float comp[4];
-            VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
-            VEC4_MUL(dest[0], dest[0], comp); /* R */
-            VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */
-            VEC4_MUL(dest[1], dest[1], comp); /* G */
-            VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */
-            VEC4_MUL(dest[2], dest[2], comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
-         {
-            float comp[4];
-            VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
-            VEC4_MUL(dest[0], dest[0], comp); /* R */
-            VEC4_MUL(dest[1], dest[1], comp); /* G */
-            VEC4_MUL(dest[2], dest[2], comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_ZERO:
-         VEC4_COPY(dest[0], zero); /* R */
-         VEC4_COPY(dest[1], zero); /* G */
-         VEC4_COPY(dest[2], zero); /* B */
-         break;
-      case PIPE_BLENDFACTOR_SRC1_COLOR:
-      case PIPE_BLENDFACTOR_SRC1_ALPHA:
-         /* XXX what are these? */
-         assert(0);
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
-            VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
-            VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
-            VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
-            VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
-            VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-         {
-            float one_minus_alpha[QUAD_SIZE];
-            VEC4_SUB(one_minus_alpha, one, quadColor[3]);
-            VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
-            VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
-            VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, dest[3]); /* A */
-            VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
-            VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
-            VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, dest[0]); /* R */
-            VEC4_MUL(dest[0], dest[0], inv_comp); /* R */
-            VEC4_SUB(inv_comp, one, dest[1]); /* G */
-            VEC4_MUL(dest[1], dest[1], inv_comp); /* G */
-            VEC4_SUB(inv_comp, one, dest[2]); /* B */
-            VEC4_MUL(dest[2], dest[2], inv_comp); /* B */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-         {
-            float inv_comp[4];
-            /* R */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[0]);
-            VEC4_MUL(dest[0], dest[0], inv_comp);
-            /* G */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[1]);
-            VEC4_MUL(dest[1], dest[1], inv_comp);
-            /* B */
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[2]);
-            VEC4_MUL(dest[2], dest[2], inv_comp);
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-         {
-            float inv_comp[4];
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
-            VEC4_MUL(dest[0], dest[0], inv_comp);
-            VEC4_MUL(dest[1], dest[1], inv_comp);
-            VEC4_MUL(dest[2], dest[2], inv_comp);
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (quad->inout.mask & (1 << j)) {
+            int x = itx + (j & 1);
+            int y = ity + (j >> 1);
+            for (i = 0; i < 4; i++) { /* loop over color chans */
+               tile->data.color[y][x][i] = quadColor[i][j];
+            }
          }
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-      case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
-         /* XXX what are these? */
-         assert(0);
-         break;
-      default:
-         assert(0);
       }
+   }
+}
 
-      /*
-       * Compute dest/second term A
-       */
-      switch (softpipe->blend->alpha_dst_factor) {
-      case PIPE_BLENDFACTOR_ONE:
-         /* dest = dest * 1   NO-OP, leave dest as-is */
-         break;
-      case PIPE_BLENDFACTOR_SRC_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_SRC_ALPHA:
-         VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */
-         break;
-      case PIPE_BLENDFACTOR_DST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_DST_ALPHA:
-         VEC4_MUL(dest[3], dest[3], dest[3]); /* A */
-         break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-         assert(0); /* illegal */
-         break;
-      case PIPE_BLENDFACTOR_CONST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
-         {
-            float comp[4];
-            VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
-            VEC4_MUL(dest[3], dest[3], comp); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_ZERO:
-         VEC4_COPY(dest[3], zero); /* A */
-         break;
-      case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-         {
-            float one_minus_alpha[QUAD_SIZE];
-            VEC4_SUB(one_minus_alpha, one, quadColor[3]);
-            VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-         {
-            float inv_comp[4];
-            VEC4_SUB(inv_comp, one, dest[3]); /* A */
-            VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */
-         }
-         break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-         /* fall-through */
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-         {
-            float inv_comp[4];
-            VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
-            VEC4_MUL(dest[3], dest[3], inv_comp);
+
+static void
+single_output_color(struct quad_stage *qs, 
+                    struct quad_header *quads[],
+                    unsigned nr)
+{
+   uint i, j, q;
+
+   struct softpipe_cached_tile *tile
+      = sp_get_cached_tile(qs->softpipe->cbuf_cache[0],
+                           quads[0]->input.x0, 
+                           quads[0]->input.y0);
+
+   for (q = 0; q < nr; q++) {
+      struct quad_header *quad = quads[q];
+      float (*quadColor)[4] = quad->output.color[0];
+      const int itx = (quad->input.x0 & (TILE_SIZE-1));
+      const int ity = (quad->input.y0 & (TILE_SIZE-1));
+      
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (quad->inout.mask & (1 << j)) {
+            int x = itx + (j & 1);
+            int y = ity + (j >> 1);
+            for (i = 0; i < 4; i++) { /* loop over color chans */
+               tile->data.color[y][x][i] = quadColor[i][j];
+            }
          }
-         break;
-      default:
-         assert(0);
       }
+   }
+}
+
+static void
+blend_noop(struct quad_stage *qs, 
+           struct quad_header *quads[],
+           unsigned nr)
+{
+}
 
-      /*
-       * Combine RGB terms
-       */
-      switch (softpipe->blend->rgb_func) {
-      case PIPE_BLEND_ADD:
-         VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
-         VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
-         VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */
-         break;
-      case PIPE_BLEND_SUBTRACT:
-         VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */
-         VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */
-         VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */
-         break;
-      case PIPE_BLEND_REVERSE_SUBTRACT:
-         VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */
-         VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */
-         VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */
-         break;
-      case PIPE_BLEND_MIN:
-         VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */
-         VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */
-         VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */
-         break;
-      case PIPE_BLEND_MAX:
-         VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */
-         VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */
-         VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */
-         break;
-      default:
-         assert(0);
-      }
 
-      /*
-       * Combine A terms
-       */
-      switch (softpipe->blend->alpha_func) {
-      case PIPE_BLEND_ADD:
-         VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
-         break;
-      case PIPE_BLEND_SUBTRACT:
-         VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */
-         break;
-      case PIPE_BLEND_REVERSE_SUBTRACT:
-         VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */
-         break;
-      case PIPE_BLEND_MIN:
-         VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */
-         break;
-      case PIPE_BLEND_MAX:
-         VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */
-         break;
-      default:
-         assert(0);
+static void
+choose_blend_quad(struct quad_stage *qs, 
+                  struct quad_header *quads[],
+                  unsigned nr)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   const struct pipe_blend_state *blend = softpipe->blend;
+
+   qs->run = blend_fallback;
+   
+   if (softpipe->framebuffer.nr_cbufs == 0) {
+      qs->run = blend_noop;
+   }
+   else if (!softpipe->blend->logicop_enable &&
+            softpipe->blend->colormask == 0xf) 
+   {
+      if (!blend->blend_enable) {
+         qs->run = single_output_color;
       }
+      else if (blend->rgb_src_factor == blend->alpha_src_factor &&
+               blend->rgb_dst_factor == blend->alpha_dst_factor &&
+               blend->rgb_func == blend->alpha_func &&
+               softpipe->framebuffer.nr_cbufs == 1)
+      {
+         if (blend->alpha_func == PIPE_BLEND_ADD) {
+            if (blend->rgb_src_factor == PIPE_BLENDFACTOR_ONE &&
+                blend->rgb_dst_factor == PIPE_BLENDFACTOR_ONE) {
+               qs->run = blend_single_add_one_one;
+            }
+            else if (blend->rgb_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA &&
+                blend->rgb_dst_factor == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
+               qs->run = blend_single_add_src_alpha_inv_src_alpha;
 
-   } /* cbuf loop */
+         }
+      }
+   }
 
-   /* pass blended quad to next stage */
-   qs->next->run(qs->next, quad);
+   qs->run(qs, quads, nr);
 }
 
 
 static void blend_begin(struct quad_stage *qs)
 {
-   qs->next->begin(qs->next);
+   qs->run = choose_blend_quad;
 }
 
 
@@ -770,7 +991,7 @@ struct quad_stage *sp_quad_blend_stage( struct softpipe_context *softpipe )
 
    stage->softpipe = softpipe;
    stage->begin = blend_begin;
-   stage->run = blend_quad;
+   stage->run = choose_blend_quad;
    stage->destroy = blend_destroy;
 
    return stage;
diff --git a/src/gallium/drivers/softpipe/sp_quad_bufloop.c b/src/gallium/drivers/softpipe/sp_quad_bufloop.c
deleted file mode 100644
index 953d8516b9..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_bufloop.c
+++ /dev/null
@@ -1,74 +0,0 @@
-
-#include "util/u_memory.h"
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_quad_pipe.h"
-
-
-/**
- * Loop over colorbuffers, passing quad to next stage each time.
- */
-static void
-cbuf_loop_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-   float tmp[PIPE_MAX_COLOR_BUFS][4][QUAD_SIZE];
-   unsigned i;
-
-   assert(sizeof(quad->outputs.color) == sizeof(tmp));
-   assert(softpipe->framebuffer.nr_cbufs <= PIPE_MAX_COLOR_BUFS);
-
-   /* make copy of original colors since they can get modified
-    * by blending and masking.
-    * XXX we won't have to do this if the fragment program actually emits
-    * N separate colors and we're drawing to N color buffers (MRT).
-    * But if we emitted one color and glDrawBuffer(GL_FRONT_AND_BACK) is
-    * in effect, we need to save/restore colors like this.
-    */
-   memcpy(tmp, quad->outputs.color, sizeof(tmp));
-
-   for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) {
-      /* set current cbuffer */
-#if 0 /* obsolete & going away */
-      softpipe->current_cbuf = i;
-#endif
-
-      /* pass blended quad to next stage */
-      qs->next->run(qs->next, quad);
-
-      /* restore quad's colors for next buffer */
-      memcpy(quad->outputs.color, tmp, sizeof(tmp));
-   }
-}
-
-
-static void cbuf_loop_begin(struct quad_stage *qs)
-{
-   qs->next->begin(qs->next);
-}
-
-
-static void cbuf_loop_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-/**
- * Create the colorbuffer loop stage.
- * This is used to implement multiple render targets and GL_FRONT_AND_BACK
- * rendering.
- */
-struct quad_stage *sp_quad_bufloop_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = cbuf_loop_begin;
-   stage->run = cbuf_loop_quad;
-   stage->destroy = cbuf_loop_destroy;
-
-   return stage;
-}
-
diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c
deleted file mode 100644
index dc90e5d5e9..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_colormask.c
+++ /dev/null
@@ -1,116 +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.
- * 
- **************************************************************************/
-
-/**
- * \brief  quad colormask stage
- * \author Brian Paul
- */
-
-#include "pipe/p_defines.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_quad_pipe.h"
-#include "sp_tile_cache.h"
-
-
-
-/**
- * XXX colormask could be rolled into blending...
- */
-static void
-colormask_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-   uint cbuf;
-
-   /* loop over colorbuffer outputs */
-   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
-      float dest[4][QUAD_SIZE];
-      struct softpipe_cached_tile *tile
-         = sp_get_cached_tile(softpipe,
-                              softpipe->cbuf_cache[cbuf],
-                              quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
-      uint i, j;
-
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
-         }
-      }
-
-      /* R */
-      if (!(softpipe->blend->colormask & PIPE_MASK_R))
-          COPY_4V(quadColor[0], dest[0]);
-
-      /* G */
-      if (!(softpipe->blend->colormask & PIPE_MASK_G))
-          COPY_4V(quadColor[1], dest[1]);
-
-      /* B */
-      if (!(softpipe->blend->colormask & PIPE_MASK_B))
-          COPY_4V(quadColor[2], dest[2]);
-
-      /* A */
-      if (!(softpipe->blend->colormask & PIPE_MASK_A))
-          COPY_4V(quadColor[3], dest[3]);
-   }
-
-   /* pass quad to next stage */
-   qs->next->run(qs->next, quad);
-}
-
-
-static void colormask_begin(struct quad_stage *qs)
-{
-   qs->next->begin(qs->next);
-}
-
-
-static void colormask_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-struct quad_stage *sp_quad_colormask_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = colormask_begin;
-   stage->run = colormask_quad;
-   stage->destroy = colormask_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c
deleted file mode 100644
index 4aeee85870..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_coverage.c
+++ /dev/null
@@ -1,94 +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.
- * 
- **************************************************************************/
-
-
-/**
- * \brief  Apply AA coverage to quad alpha valus
- * \author  Brian Paul
- */
-
-
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_quad_pipe.h"
-
-
-/**
- * Multiply quad's alpha values by the fragment coverage.
- */
-static void
-coverage_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-   const uint prim = quad->input.prim;
-
-   if ((softpipe->rasterizer->poly_smooth && prim == QUAD_PRIM_TRI) ||
-       (softpipe->rasterizer->line_smooth && prim == QUAD_PRIM_LINE) ||
-       (softpipe->rasterizer->point_smooth && prim == QUAD_PRIM_POINT)) {
-      uint cbuf;
-
-      /* loop over colorbuffer outputs */
-      for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
-         float (*quadColor)[4] = quad->output.color[cbuf];
-         unsigned j;
-         for (j = 0; j < QUAD_SIZE; j++) {
-            assert(quad->input.coverage[j] >= 0.0);
-            assert(quad->input.coverage[j] <= 1.0);
-         quadColor[3][j] *= quad->input.coverage[j];
-         }
-      }
-   }
-
-   qs->next->run(qs->next, quad);
-}
-
-
-static void coverage_begin(struct quad_stage *qs)
-{
-   qs->next->begin(qs->next);
-}
-
-
-static void coverage_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-struct quad_stage *sp_quad_coverage_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = coverage_begin;
-   stage->run = coverage_quad;
-   stage->destroy = coverage_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index d463930bae..0ca86c4e1c 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -31,61 +31,109 @@
 
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
+#include "tgsi/tgsi_scan.h"
 #include "sp_context.h"
 #include "sp_quad.h"
 #include "sp_surface.h"
 #include "sp_quad_pipe.h"
 #include "sp_tile_cache.h"
+#include "sp_state.h"           /* for sp_fragment_shader */
 
 
-/**
- * Do depth testing for a quad.
- * Not static since it's used by the stencil code.
- */
+struct depth_data {
+   struct pipe_surface *ps;
+   enum pipe_format format;
+   unsigned bzzzz[QUAD_SIZE];  /**< Z values fetched from depth buffer */
+   unsigned qzzzz[QUAD_SIZE];  /**< Z values from the quad */
+   ubyte stencilVals[QUAD_SIZE];
+   struct softpipe_cached_tile *tile;
+};
 
-/*
- * To increase efficiency, we should probably have multiple versions
- * of this function that are specifically for Z16, Z32 and FP Z buffers.
- * Try to effectively do that with codegen...
- */
 
-void
-sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
+
+static void
+get_depth_stencil_values( struct depth_data *data,
+                          const struct quad_header *quad )
 {
-   struct softpipe_context *softpipe = qs->softpipe;
-   struct pipe_surface *ps = softpipe->framebuffer.zsbuf;
-   const enum pipe_format format = ps->format;
-   unsigned bzzzz[QUAD_SIZE];  /**< Z values fetched from depth buffer */
-   unsigned qzzzz[QUAD_SIZE];  /**< Z values from the quad */
-   unsigned zmask = 0;
    unsigned j;
-   struct softpipe_cached_tile *tile
-      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0);
+   const struct softpipe_cached_tile *tile = data->tile;
+
+   switch (data->format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         data->bzzzz[j] = tile->data.depth16[y][x];
+      }
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         data->bzzzz[j] = tile->data.depth32[y][x];
+      }
+      break;
+   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         data->bzzzz[j] = tile->data.depth32[y][x] & 0xffffff;
+         data->stencilVals[j] = tile->data.depth32[y][x] >> 24;
+      }
+   break;
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         data->bzzzz[j] = tile->data.depth32[y][x] >> 8;
+         data->stencilVals[j] = tile->data.depth32[y][x] & 0xff;
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
 
-   assert(ps); /* shouldn't get here if there's no zbuffer */
+/* If the shader has not been run, interpolate the depth values
+ * ourselves.
+ */
+static void
+interpolate_quad_depth( struct quad_header *quad )
+{
+   const float fx = (float) quad->input.x0;
+   const float fy = (float) quad->input.y0;
+   const float dzdx = quad->posCoef->dadx[2];
+   const float dzdy = quad->posCoef->dady[2];
+   const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
 
-   /*
-    * Convert quad's float depth values to int depth values (qzzzz).
+   quad->output.depth[0] = z0;
+   quad->output.depth[1] = z0 + dzdx;
+   quad->output.depth[2] = z0 + dzdy;
+   quad->output.depth[3] = z0 + dzdx + dzdy;
+}
+
+
+static void
+convert_quad_depth( struct depth_data *data, 
+                    const struct quad_header *quad )
+{
+   unsigned j;
+
+   /* Convert quad's float depth values to int depth values (qzzzz).
     * If the Z buffer stores integer values, we _have_ to do the depth
     * compares with integers (not floats).  Otherwise, the float->int->float
     * conversion of Z values (which isn't an identity function) will cause
     * Z-fighting errors.
-    *
-    * Also, get the zbuffer values (bzzzz) from the cached tile.
     */
-   switch (format) {
+   switch (data->format) {
    case PIPE_FORMAT_Z16_UNORM:
       {
          float scale = 65535.0;
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
-         }
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            bzzzz[j] = tile->data.depth16[y][x];
+            data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
       }
       break;
@@ -94,47 +142,247 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          double scale = (double) (uint) ~0UL;
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
-         }
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            bzzzz[j] = tile->data.depth32[y][x];
+            data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
       }
       break;
    case PIPE_FORMAT_X8Z24_UNORM:
-      /* fall-through */
    case PIPE_FORMAT_S8Z24_UNORM:
       {
          float scale = (float) ((1 << 24) - 1);
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
-         }
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            bzzzz[j] = tile->data.depth32[y][x] & 0xffffff;
+            data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
       }
       break;
    case PIPE_FORMAT_Z24X8_UNORM:
-      /* fall-through */
    case PIPE_FORMAT_Z24S8_UNORM:
       {
          float scale = (float) ((1 << 24) - 1);
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
+            data->qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
 
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            bzzzz[j] = tile->data.depth32[y][x] >> 8;
+
+
+static void
+write_depth_stencil_values( struct depth_data *data,
+                            struct quad_header *quad )
+{
+   struct softpipe_cached_tile *tile = data->tile;
+   unsigned j;
+
+   /* put updated Z values back into cached tile */
+   switch (data->format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         tile->data.depth16[y][x] = (ushort) data->bzzzz[j];
+      }
+      break;
+   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_Z32_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         tile->data.depth32[y][x] = data->bzzzz[j];
+      }
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         tile->data.depth32[y][x] = (data->stencilVals[j] << 24) | data->bzzzz[j];
+      }
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         tile->data.depth32[y][x] = (data->bzzzz[j] << 8) | data->stencilVals[j];
+      }
+      break;
+   case PIPE_FORMAT_Z24X8_UNORM:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
+         tile->data.depth32[y][x] = data->bzzzz[j] << 8;
+      }
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+
+
+/** Only 8-bit stencil supported */
+#define STENCIL_MAX 0xff
+
+
+/**
+ * Do the basic stencil test (compare stencil buffer values against the
+ * reference value.
+ *
+ * \param data->stencilVals  the stencil values from the stencil buffer
+ * \param func  the stencil func (PIPE_FUNC_x)
+ * \param ref  the stencil reference value
+ * \param valMask  the stencil value mask indicating which bits of the stencil
+ *                 values and ref value are to be used.
+ * \return mask indicating which pixels passed the stencil test
+ */
+static unsigned
+do_stencil_test(struct depth_data *data,
+                unsigned func,
+                unsigned ref, unsigned valMask)
+{
+   unsigned passMask = 0x0;
+   unsigned j;
+
+   ref &= valMask;
+
+   switch (func) {
+   case PIPE_FUNC_NEVER:
+      /* passMask = 0x0 */
+      break;
+   case PIPE_FUNC_LESS:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref < (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_EQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref == (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_LEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref <= (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GREATER:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref > (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref != (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_GEQUAL:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (ref >= (data->stencilVals[j] & valMask)) {
+            passMask |= (1 << j);
+         }
+      }
+      break;
+   case PIPE_FUNC_ALWAYS:
+      passMask = MASK_ALL;
+      break;
+   default:
+      assert(0);
+   }
+
+   return passMask;
+}
+
+
+/**
+ * Apply the stencil operator to stencil values.
+ *
+ * \param data->stencilVals  the stencil buffer values (read and written)
+ * \param mask  indicates which pixels to update
+ * \param op  the stencil operator (PIPE_STENCIL_OP_x)
+ * \param ref  the stencil reference value
+ * \param wrtMask  writemask controlling which bits are changed in the
+ *                 stencil values
+ */
+static void
+apply_stencil_op(struct depth_data *data,
+                 unsigned mask, unsigned op, ubyte ref, ubyte wrtMask)
+{
+   unsigned j;
+   ubyte newstencil[QUAD_SIZE];
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      newstencil[j] = data->stencilVals[j];
+   }
+
+   switch (op) {
+   case PIPE_STENCIL_OP_KEEP:
+      /* no-op */
+      break;
+   case PIPE_STENCIL_OP_ZERO:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = 0;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_REPLACE:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = ref;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INCR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            if (data->stencilVals[j] < STENCIL_MAX) {
+               newstencil[j] = data->stencilVals[j] + 1;
+            }
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_DECR:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            if (data->stencilVals[j] > 0) {
+               newstencil[j] = data->stencilVals[j] - 1;
+            }
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INCR_WRAP:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = data->stencilVals[j] + 1;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_DECR_WRAP:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = data->stencilVals[j] - 1;
+         }
+      }
+      break;
+   case PIPE_STENCIL_OP_INVERT:
+      for (j = 0; j < QUAD_SIZE; j++) {
+         if (mask & (1 << j)) {
+            newstencil[j] = ~data->stencilVals[j];
          }
       }
       break;
@@ -142,6 +390,39 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
       assert(0);
    }
 
+   /*
+    * update the stencil values
+    */
+   if (wrtMask != STENCIL_MAX) {
+      /* apply bit-wise stencil buffer writemask */
+      for (j = 0; j < QUAD_SIZE; j++) {
+         data->stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & data->stencilVals[j]);
+      }
+   }
+   else {
+      for (j = 0; j < QUAD_SIZE; j++) {
+         data->stencilVals[j] = newstencil[j];
+      }
+   }
+}
+
+   
+
+/*
+ * To increase efficiency, we should probably have multiple versions
+ * of this function that are specifically for Z16, Z32 and FP Z buffers.
+ * Try to effectively do that with codegen...
+ */
+
+static boolean
+depth_test_quad(struct quad_stage *qs, 
+                struct depth_data *data,
+                struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   unsigned zmask = 0;
+   unsigned j;
+
    switch (softpipe->depth_stencil->depth.func) {
    case PIPE_FUNC_NEVER:
       /* zmask = 0 */
@@ -151,37 +432,37 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
        * Like this:  quad->mask &= (quad->outputs.depth < zzzz);
        */
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] < bzzzz[j]) 
+	 if (data->qzzzz[j] < data->bzzzz[j]) 
 	    zmask |= 1 << j;
       }
       break;
    case PIPE_FUNC_EQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] == bzzzz[j]) 
+	 if (data->qzzzz[j] == data->bzzzz[j]) 
 	    zmask |= 1 << j;
       }
       break;
    case PIPE_FUNC_LEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] <= bzzzz[j]) 
+	 if (data->qzzzz[j] <= data->bzzzz[j]) 
 	    zmask |= (1 << j);
       }
       break;
    case PIPE_FUNC_GREATER:
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] > bzzzz[j]) 
+	 if (data->qzzzz[j] > data->bzzzz[j]) 
 	    zmask |= (1 << j);
       }
       break;
    case PIPE_FUNC_NOTEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] != bzzzz[j]) 
+	 if (data->qzzzz[j] != data->bzzzz[j]) 
 	    zmask |= (1 << j);
       }
       break;
    case PIPE_FUNC_GEQUAL:
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (qzzzz[j] >= bzzzz[j]) 
+	 if (data->qzzzz[j] >= data->bzzzz[j]) 
 	    zmask |= (1 << j);
       }
       break;
@@ -193,80 +474,480 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
    }
 
    quad->inout.mask &= zmask;
+   if (quad->inout.mask == 0)
+      return FALSE;
 
+   /* Update our internal copy only if writemask set.  Even if
+    * depth.writemask is FALSE, may still need to write out buffer
+    * data due to stencil changes.
+    */
    if (softpipe->depth_stencil->depth.writemask) {
-      
-      /* This is also efficient with sse / spe instructions: 
-       */
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (quad->inout.mask & (1 << j)) {
-	    bzzzz[j] = qzzzz[j];
-	 }
+         if (quad->inout.mask & (1 << j)) {
+            data->bzzzz[j] = data->qzzzz[j];
+         }
       }
+   }
 
-      /* put updated Z values back into cached tile */
-      switch (format) {
-      case PIPE_FORMAT_Z16_UNORM:
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            tile->data.depth16[y][x] = (ushort) bzzzz[j];
+   return TRUE;
+}
+
+
+
+/**
+ * Do stencil (and depth) testing.  Stenciling depends on the outcome of
+ * depth testing.
+ */
+static void
+depth_stencil_test_quad(struct quad_stage *qs, 
+                        struct depth_data *data,
+                        struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   unsigned func, zFailOp, zPassOp, failOp;
+   ubyte ref, wrtMask, valMask;
+   uint face = quad->input.facing;
+
+   if (!softpipe->depth_stencil->stencil[1].enabled) {
+      /* single-sided stencil test, use front (face=0) state */
+      face = 0;
+   }
+
+   /* choose front or back face function, operator, etc */
+   /* XXX we could do these initializations once per primitive */
+   func    = softpipe->depth_stencil->stencil[face].func;
+   failOp  = softpipe->depth_stencil->stencil[face].fail_op;
+   zFailOp = softpipe->depth_stencil->stencil[face].zfail_op;
+   zPassOp = softpipe->depth_stencil->stencil[face].zpass_op;
+   ref     = softpipe->depth_stencil->stencil[face].ref_value;
+   wrtMask = softpipe->depth_stencil->stencil[face].writemask;
+   valMask = softpipe->depth_stencil->stencil[face].valuemask;
+
+
+   /* do the stencil test first */
+   {
+      unsigned passMask, failMask;
+      passMask = do_stencil_test(data, func, ref, valMask);
+      failMask = quad->inout.mask & ~passMask;
+      quad->inout.mask &= passMask;
+
+      if (failOp != PIPE_STENCIL_OP_KEEP) {
+         apply_stencil_op(data, failMask, failOp, ref, wrtMask);
+      }
+   }
+
+   if (quad->inout.mask) {
+      /* now the pixels that passed the stencil test are depth tested */
+      if (softpipe->depth_stencil->depth.enabled) {
+         const unsigned origMask = quad->inout.mask;
+
+         depth_test_quad(qs, data, quad);  /* quad->mask is updated */
+
+         /* update stencil buffer values according to z pass/fail result */
+         if (zFailOp != PIPE_STENCIL_OP_KEEP) {
+            const unsigned zFailMask = origMask & ~quad->inout.mask;
+            apply_stencil_op(data, zFailMask, zFailOp, ref, wrtMask);
          }
-         break;
-      case PIPE_FORMAT_X8Z24_UNORM:
-         /* fall-through */
-         /* (yes, this falls through to a different case than above) */
-      case PIPE_FORMAT_Z32_UNORM:
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            tile->data.depth32[y][x] = bzzzz[j];
+
+         if (zPassOp != PIPE_STENCIL_OP_KEEP) {
+            const unsigned zPassMask = origMask & quad->inout.mask;
+            apply_stencil_op(data, zPassMask, zPassOp, ref, wrtMask);
          }
-         break;
-      case PIPE_FORMAT_S8Z24_UNORM:
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            uint s8z24 = tile->data.depth32[y][x];
-            s8z24 = (s8z24 & 0xff000000) | bzzzz[j];
-            tile->data.depth32[y][x] = s8z24;
+      }
+      else {
+         /* no depth test, apply Zpass operator to stencil buffer values */
+         apply_stencil_op(data, quad->inout.mask, zPassOp, ref, wrtMask);
+      }
+   }
+}
+
+
+#define ALPHATEST( FUNC, COMP )                                         \
+   static int                                                          \
+   alpha_test_quads_##FUNC( struct quad_stage *qs,                      \
+                           struct quad_header *quads[],                 \
+                           unsigned nr )                                \
+   {                                                                    \
+      const float ref = qs->softpipe->depth_stencil->alpha.ref_value;   \
+      const uint cbuf = 0; /* only output[0].alpha is tested */         \
+      unsigned pass_nr = 0;                                             \
+      unsigned i;                                                       \
+                                                                        \
+      for (i = 0; i < nr; i++) {                                        \
+         const float *aaaa = quads[i]->output.color[cbuf][3];           \
+         unsigned passMask = 0;                                         \
+                                                                        \
+         if (aaaa[0] COMP ref) passMask |= (1 << 0);                    \
+         if (aaaa[1] COMP ref) passMask |= (1 << 1);                    \
+         if (aaaa[2] COMP ref) passMask |= (1 << 2);                    \
+         if (aaaa[3] COMP ref) passMask |= (1 << 3);                    \
+                                                                        \
+         quads[i]->inout.mask &= passMask;                              \
+                                                                        \
+         if (quads[i]->inout.mask)                                      \
+            quads[pass_nr++] = quads[i];                                \
+      }                                                                 \
+                                                                        \
+      return pass_nr;                                                   \
+   }
+
+
+ALPHATEST( LESS,     < )
+ALPHATEST( EQUAL,    == )
+ALPHATEST( LEQUAL,   <= )
+ALPHATEST( GREATER,  > )
+ALPHATEST( NOTEQUAL, != )
+ALPHATEST( GEQUAL,   >= )
+
+
+/* XXX: Incorporate into shader using KILP.
+ */
+static int
+alpha_test_quads(struct quad_stage *qs, 
+                 struct quad_header *quads[], 
+                 unsigned nr)
+{
+   switch (qs->softpipe->depth_stencil->alpha.func) {
+   case PIPE_FUNC_LESS:
+      return alpha_test_quads_LESS( qs, quads, nr );
+   case PIPE_FUNC_EQUAL:
+      return alpha_test_quads_EQUAL( qs, quads, nr );
+      break;
+   case PIPE_FUNC_LEQUAL:
+      return alpha_test_quads_LEQUAL( qs, quads, nr );
+   case PIPE_FUNC_GREATER:
+      return alpha_test_quads_GREATER( qs, quads, nr );
+   case PIPE_FUNC_NOTEQUAL:
+      return alpha_test_quads_NOTEQUAL( qs, quads, nr );
+   case PIPE_FUNC_GEQUAL:
+      return alpha_test_quads_GEQUAL( qs, quads, nr );
+   case PIPE_FUNC_ALWAYS:
+      return nr;
+   case PIPE_FUNC_NEVER:
+   default:
+      return 0;
+   }
+}
+
+static unsigned mask_count[16] = 
+{
+   0,                           /* 0x0 */
+   1,                           /* 0x1 */
+   1,                           /* 0x2 */
+   2,                           /* 0x3 */
+   1,                           /* 0x4 */
+   2,                           /* 0x5 */
+   2,                           /* 0x6 */
+   3,                           /* 0x7 */
+   1,                           /* 0x8 */
+   2,                           /* 0x9 */
+   2,                           /* 0xa */
+   3,                           /* 0xb */
+   2,                           /* 0xc */
+   3,                           /* 0xd */
+   3,                           /* 0xe */
+   4,                           /* 0xf */
+};
+
+
+
+static void
+depth_test_quads_fallback(struct quad_stage *qs, 
+                          struct quad_header *quads[],
+                          unsigned nr)
+{
+   unsigned i, pass = 0;
+   const struct sp_fragment_shader *fs = qs->softpipe->fs;
+   boolean interp_depth = !fs->info.writes_z;
+   struct depth_data data;
+
+
+   if (qs->softpipe->depth_stencil->alpha.enabled) {
+      nr = alpha_test_quads(qs, quads, nr);
+   }
+
+   if (qs->softpipe->framebuffer.zsbuf && 
+       (qs->softpipe->depth_stencil->depth.enabled ||
+        qs->softpipe->depth_stencil->stencil[0].enabled)) {
+
+      data.ps = qs->softpipe->framebuffer.zsbuf;
+      data.format = data.ps->format;
+      data.tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, 
+                                     quads[0]->input.x0, 
+                                     quads[0]->input.y0);
+
+      for (i = 0; i < nr; i++) {
+         get_depth_stencil_values(&data, quads[i]);
+
+         if (qs->softpipe->depth_stencil->depth.enabled) {
+            if (interp_depth)
+               interpolate_quad_depth(quads[i]);
+
+            convert_quad_depth(&data, quads[i]);
          }
-         break;
-      case PIPE_FORMAT_Z24S8_UNORM:
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            uint z24s8 = tile->data.depth32[y][x];
-            z24s8 = (z24s8 & 0xff) | (bzzzz[j] << 8);
-            tile->data.depth32[y][x] = z24s8;
+
+         if (qs->softpipe->depth_stencil->stencil[0].enabled) {
+            depth_stencil_test_quad(qs, &data, quads[i]);
+            write_depth_stencil_values(&data, quads[i]);
          }
-         break;
-      case PIPE_FORMAT_Z24X8_UNORM:
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->input.x0 % TILE_SIZE + (j & 1);
-            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-            tile->data.depth32[y][x] = bzzzz[j] << 8;
+         else {
+            if (!depth_test_quad(qs, &data, quads[i]))
+               continue;
+
+            if (qs->softpipe->depth_stencil->depth.writemask)
+               write_depth_stencil_values(&data, quads[i]);
          }
-         break;
-      default:
-         assert(0);
+
+
+         quads[pass++] = quads[i];
+      }
+
+      nr = pass;
+   }
+
+   if (qs->softpipe->active_query_count) {
+      for (i = 0; i < nr; i++) 
+         qs->softpipe->occlusion_count += mask_count[quads[i]->inout.mask];
+   }
+
+   if (nr)
+      qs->next->run(qs->next, quads, nr);
+}
+
+/* XXX: this function assumes setup function actually emits linear
+ * spans of quads.  It seems a lot more natural to do (early)
+ * depth-testing on spans rather than quads.
+ */
+static void
+depth_interp_z16_less_write(struct quad_stage *qs, 
+                            struct quad_header *quads[],
+                            unsigned nr)
+{
+   unsigned i, pass = 0;
+   const unsigned ix = quads[0]->input.x0;
+   const unsigned iy = quads[0]->input.y0;
+   const float fx = (float) ix;
+   const float fy = (float) iy;
+   const float dzdx = quads[0]->posCoef->dadx[2];
+   const float dzdy = quads[0]->posCoef->dady[2];
+   const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
+   struct softpipe_cached_tile *tile;
+   ushort (*depth16)[TILE_SIZE];
+   ushort idepth[4], depth_step;
+   const float scale = 65535.0;
+
+   idepth[0] = (ushort)((z0) * scale);
+   idepth[1] = (ushort)((z0 + dzdx) * scale);
+   idepth[2] = (ushort)((z0 + dzdy) * scale);
+   idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+
+   depth_step = (ushort)(dzdx * 2 * scale);
+
+   tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
+
+   depth16 = (ushort (*)[TILE_SIZE])
+      &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
+
+   for (i = 0; i < nr; i++) {
+      unsigned outmask = quads[i]->inout.mask;
+      unsigned mask = 0;
+      
+      if ((outmask & 1) && idepth[0] < depth16[0][0]) {
+         depth16[0][0] = idepth[0];
+         mask |= (1 << 0);
+      }
+
+      if ((outmask & 2) && idepth[1] < depth16[0][1]) {
+         depth16[0][1] = idepth[1];
+         mask |= (1 << 1);
+      }
+
+      if ((outmask & 4) && idepth[2] < depth16[1][0]) {
+         depth16[1][0] = idepth[2];
+         mask |= (1 << 2);
+      }
+
+      if ((outmask & 8) && idepth[3] < depth16[1][1]) {
+         depth16[1][1] = idepth[3];
+         mask |= (1 << 3);
       }
+
+      idepth[0] += depth_step;
+      idepth[1] += depth_step;
+      idepth[2] += depth_step;
+      idepth[3] += depth_step;
+
+      depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
+
+      quads[i]->inout.mask = mask;
+      if (quads[i]->inout.mask)
+         quads[pass++] = quads[i];
    }
+
+   if (pass)
+      qs->next->run(qs->next, quads, pass);
+
 }
 
 
 static void
-depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
+depth_interp_z16_lequal_write(struct quad_stage *qs, 
+                            struct quad_header *quads[],
+                            unsigned nr)
 {
-   sp_depth_test_quad(qs, quad);
+   unsigned i, pass = 0;
+   const unsigned ix = quads[0]->input.x0;
+   const unsigned iy = quads[0]->input.y0;
+   const float fx = (float) ix;
+   const float fy = (float) iy;
+   const float dzdx = quads[0]->posCoef->dadx[2];
+   const float dzdy = quads[0]->posCoef->dady[2];
+   const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
+   struct softpipe_cached_tile *tile;
+   ushort (*depth16)[TILE_SIZE];
+   ushort idepth[4], depth_step;
+   const float scale = 65535.0;
+
+   idepth[0] = (ushort)((z0) * scale);
+   idepth[1] = (ushort)((z0 + dzdx) * scale);
+   idepth[2] = (ushort)((z0 + dzdy) * scale);
+   idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+
+   depth_step = (ushort)(dzdx * 2 * scale);
+
+   tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
+
+   depth16 = (ushort (*)[TILE_SIZE])
+      &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
+
+   for (i = 0; i < nr; i++) {
+      unsigned outmask = quads[i]->inout.mask;
+      unsigned mask = 0;
+      
+      if ((outmask & 1) && idepth[0] <= depth16[0][0]) {
+         depth16[0][0] = idepth[0];
+         mask |= (1 << 0);
+      }
+
+      if ((outmask & 2) && idepth[1] <= depth16[0][1]) {
+         depth16[0][1] = idepth[1];
+         mask |= (1 << 1);
+      }
+
+      if ((outmask & 4) && idepth[2] <= depth16[1][0]) {
+         depth16[1][0] = idepth[2];
+         mask |= (1 << 2);
+      }
+
+      if ((outmask & 8) && idepth[3] <= depth16[1][1]) {
+         depth16[1][1] = idepth[3];
+         mask |= (1 << 3);
+      }
+
+      idepth[0] += depth_step;
+      idepth[1] += depth_step;
+      idepth[2] += depth_step;
+      idepth[3] += depth_step;
+
+      depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
+
+      quads[i]->inout.mask = mask;
+      if (quads[i]->inout.mask)
+         quads[pass++] = quads[i];
+   }
+
+   if (pass)
+      qs->next->run(qs->next, quads, pass);
 
-   if (quad->inout.mask)
-      qs->next->run(qs->next, quad);
 }
 
 
+
+
+
+static void
+depth_noop(struct quad_stage *qs, 
+           struct quad_header *quads[],
+           unsigned nr)
+{
+   qs->next->run(qs->next, quads, nr);
+}
+
+
+
+static void
+choose_depth_test(struct quad_stage *qs, 
+                  struct quad_header *quads[],
+                  unsigned nr)
+{
+   boolean interp_depth = !qs->softpipe->fs->info.writes_z;
+
+   boolean alpha = qs->softpipe->depth_stencil->alpha.enabled;
+
+   boolean depth = (qs->softpipe->framebuffer.zsbuf && 
+                    qs->softpipe->depth_stencil->depth.enabled);
+
+   unsigned depthfunc = qs->softpipe->depth_stencil->depth.func;
+
+   boolean stencil = qs->softpipe->depth_stencil->stencil[0].enabled;
+
+   boolean depthwrite = qs->softpipe->depth_stencil->depth.writemask;
+
+   boolean occlusion = qs->softpipe->active_query_count;
+
+
+   if (!alpha &&
+       !depth &&
+       !stencil) {
+      qs->run = depth_noop;
+   }
+   else if (!alpha && 
+            interp_depth && 
+            depth && 
+            depthwrite && 
+            !occlusion &&
+            !stencil) 
+   {
+      switch (depthfunc) {
+      case PIPE_FUNC_LESS:
+         switch (qs->softpipe->framebuffer.zsbuf->format) {
+         case PIPE_FORMAT_Z16_UNORM:
+            qs->run = depth_interp_z16_less_write;
+            break;
+         default:
+            qs->run = depth_test_quads_fallback;
+            break;
+         }
+         break;
+      case PIPE_FUNC_LEQUAL:
+         switch (qs->softpipe->framebuffer.zsbuf->format) {
+         case PIPE_FORMAT_Z16_UNORM:
+            qs->run = depth_interp_z16_lequal_write;
+            break;
+         default:
+            qs->run = depth_test_quads_fallback;
+            break;
+         }
+         break;
+      default:
+         qs->run = depth_test_quads_fallback;
+      }
+   }
+   else {
+      qs->run = depth_test_quads_fallback;
+   }
+
+
+   qs->run( qs, quads, nr );
+}
+
+
+
+
+
 static void depth_test_begin(struct quad_stage *qs)
 {
+   qs->run = choose_depth_test;
    qs->next->begin(qs->next);
 }
 
@@ -283,7 +964,7 @@ struct quad_stage *sp_quad_depth_test_stage( struct softpipe_context *softpipe )
 
    stage->softpipe = softpipe;
    stage->begin = depth_test_begin;
-   stage->run = depth_test_quad;
+   stage->run = choose_depth_test;
    stage->destroy = depth_test_destroy;
 
    return stage;
diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c
deleted file mode 100644
index 496fd39ed1..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_earlyz.c
+++ /dev/null
@@ -1,88 +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.
- * 
- **************************************************************************/
-
-/**
- * \brief  Quad early-z testing
- */
-
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-#include "sp_quad.h"
-#include "sp_quad_pipe.h"
-
-
-/**
- * All this stage does is compute the quad's Z values (which is normally
- * done by the shading stage).
- * The next stage will do the actual depth test.
- */
-static void
-earlyz_quad(
-   struct quad_stage    *qs,
-   struct quad_header   *quad )
-{
-   const float fx = (float) quad->input.x0;
-   const float fy = (float) quad->input.y0;
-   const float dzdx = quad->posCoef->dadx[2];
-   const float dzdy = quad->posCoef->dady[2];
-   const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
-
-   quad->output.depth[0] = z0;
-   quad->output.depth[1] = z0 + dzdx;
-   quad->output.depth[2] = z0 + dzdy;
-   quad->output.depth[3] = z0 + dzdx + dzdy;
-
-   qs->next->run( qs->next, quad );
-}
-
-static void
-earlyz_begin(
-   struct quad_stage *qs )
-{
-   qs->next->begin( qs->next );
-}
-
-static void
-earlyz_destroy(
-   struct quad_stage *qs )
-{
-   FREE( qs );
-}
-
-struct quad_stage *
-sp_quad_earlyz_stage(
-   struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT( quad_stage );
-
-   stage->softpipe = softpipe;
-   stage->begin = earlyz_begin;
-   stage->run = earlyz_quad;
-   stage->destroy = earlyz_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index 28f8d1a60e..1e7533d0f9 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -68,72 +68,69 @@ quad_shade_stage(struct quad_stage *qs)
 /**
  * Execute fragment shader for the four fragments in the quad.
  */
-static void
+static INLINE boolean
 shade_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    struct quad_shade_stage *qss = quad_shade_stage( qs );
    struct softpipe_context *softpipe = qs->softpipe;
    struct tgsi_exec_machine *machine = qss->machine;
-   boolean z_written;
-   
-   /* Consts do not require 16 byte alignment. */
-   machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT];
-
-   machine->InterpCoefs = quad->coef;
 
    /* run shader */
-   quad->inout.mask &= softpipe->fs->run( softpipe->fs, machine, quad );
-
-   /* store outputs */
-   z_written = FALSE;
-   {
-      const ubyte *sem_name = softpipe->fs->info.output_semantic_name;
-      const ubyte *sem_index = softpipe->fs->info.output_semantic_index;
-      const uint n = qss->stage.softpipe->fs->info.num_outputs;
-      uint i;
-      for (i = 0; i < n; i++) {
-         switch (sem_name[i]) {
-         case TGSI_SEMANTIC_COLOR:
-            {
-               uint cbuf = sem_index[i];
-               memcpy(quad->output.color[cbuf],
-                      &machine->Outputs[i].xyzw[0].f[0],
-                      sizeof(quad->output.color[0]) );
-            }
-            break;
-         case TGSI_SEMANTIC_POSITION:
-            {
-               uint j;
-               for (j = 0; j < 4; j++) {
-                  quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
-               }
-               z_written = TRUE;
-            }
-            break;
-         }
+   return softpipe->fs->run( softpipe->fs, machine, quad );
+}
+
+
+
+static void
+coverage_quad(struct quad_stage *qs, struct quad_header *quad)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+   uint cbuf;
+
+   /* loop over colorbuffer outputs */
+   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
+      float (*quadColor)[4] = quad->output.color[cbuf];
+      unsigned j;
+      for (j = 0; j < QUAD_SIZE; j++) {
+         assert(quad->input.coverage[j] >= 0.0);
+         assert(quad->input.coverage[j] <= 1.0);
+         quadColor[3][j] *= quad->input.coverage[j];
       }
    }
+}
+
 
-   if (!z_written) {
-      /* compute Z values now, as in the quad earlyz stage */
-      /* XXX we should really only do this if the earlyz stage is not used */
-      const float fx = (float) quad->input.x0;
-      const float fy = (float) quad->input.y0;
-      const float dzdx = quad->posCoef->dadx[2];
-      const float dzdy = quad->posCoef->dady[2];
-      const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
-
-      quad->output.depth[0] = z0;
-      quad->output.depth[1] = z0 + dzdx;
-      quad->output.depth[2] = z0 + dzdy;
-      quad->output.depth[3] = z0 + dzdx + dzdy;
-   }
 
-   /* shader may cull fragments */
-   if (quad->inout.mask) {
-      qs->next->run( qs->next, quad );
+static void
+shade_quads(struct quad_stage *qs, 
+                 struct quad_header *quads[],
+                 unsigned nr)
+{
+   struct quad_shade_stage *qss = quad_shade_stage( qs );
+   struct softpipe_context *softpipe = qs->softpipe;
+   struct tgsi_exec_machine *machine = qss->machine;
+
+   unsigned i, pass = 0;
+   
+   machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT];
+   machine->InterpCoefs = quads[0]->coef;
+
+   for (i = 0; i < nr; i++) {
+      if (!shade_quad(qs, quads[i]))
+         continue;
+
+      if (/*do_coverage*/ 0)
+         coverage_quad( qs, quads[i] );
+
+      quads[pass++] = quads[i];
    }
+   
+   if (pass)
+      qs->next->run(qs->next, quads, pass);
 }
+   
+
+
 
 
 /**
@@ -174,7 +171,7 @@ sp_quad_shade_stage( struct softpipe_context *softpipe )
 
    qss->stage.softpipe = softpipe;
    qss->stage.begin = shade_begin;
-   qss->stage.run = shade_quad;
+   qss->stage.run = shade_quads;
    qss->stage.destroy = shade_destroy;
 
    qss->machine = tgsi_exec_machine_create();
diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c
deleted file mode 100644
index 92d5f9f3c1..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_output.c
+++ /dev/null
@@ -1,103 +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 "util/u_memory.h"
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_quad_pipe.h"
-#include "sp_tile_cache.h"
-
-
-/**
- * Last step of quad processing: write quad colors to the framebuffer,
- * taking mask into account.
- */
-static void
-output_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   /* in-tile pos: */
-   const int itx = quad->input.x0 % TILE_SIZE;
-   const int ity = quad->input.y0 % TILE_SIZE;
-
-   struct softpipe_context *softpipe = qs->softpipe;
-   uint cbuf;
-
-   /* loop over colorbuffer outputs */
-   for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
-      struct softpipe_cached_tile *tile
-         = sp_get_cached_tile(softpipe,
-                              softpipe->cbuf_cache[cbuf],
-                              quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
-      int i, j;
-
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (quad->inout.mask & (1 << j)) {
-            int x = itx + (j & 1);
-            int y = ity + (j >> 1);
-            for (i = 0; i < 4; i++) { /* loop over color chans */
-               tile->data.color[y][x][i] = quadColor[i][j];
-            }
-            if (0) {
-               debug_printf("sp write pixel %d,%d: %g, %g, %g\n",
-                            quad->input.x0 + x,
-                            quad->input.y0 + y,
-                            quadColor[0][j],
-                            quadColor[1][j],
-                            quadColor[2][j]);
-            }
-         }
-      }
-   }
-}
-
-
-static void output_begin(struct quad_stage *qs)
-{
-   assert(qs->next == NULL);
-}
-
-
-static void output_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = output_begin;
-   stage->run = output_quad;
-   stage->destroy = output_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.c b/src/gallium/drivers/softpipe/sp_quad_pipe.c
index b5f69b7426..1b5bab4eca 100644
--- a/src/gallium/drivers/softpipe/sp_quad_pipe.c
+++ b/src/gallium/drivers/softpipe/sp_quad_pipe.c
@@ -31,88 +31,33 @@
 #include "pipe/p_shader_tokens.h"
 
 static void
-sp_push_quad_first(
-   struct softpipe_context *sp,
-   struct quad_stage *quad,
-   uint i )
+sp_push_quad_first( struct softpipe_context *sp,
+                    struct quad_stage *quad )
 {
-   quad->next = sp->quad[i].first;
-   sp->quad[i].first = quad;
+   quad->next = sp->quad.first;
+   sp->quad.first = quad;
 }
 
-static void
-sp_build_depth_stencil(
-   struct softpipe_context *sp,
-   uint i )
-{
-   if (sp->depth_stencil->stencil[0].enabled ||
-       sp->depth_stencil->stencil[1].enabled) {
-      sp_push_quad_first( sp, sp->quad[i].stencil_test, i );
-   }
-   else if (sp->depth_stencil->depth.enabled &&
-            sp->framebuffer.zsbuf) {
-      sp_push_quad_first( sp, sp->quad[i].depth_test, i );
-   }
-}
 
 void
 sp_build_quad_pipeline(struct softpipe_context *sp)
 {
-   uint i;
-
    boolean early_depth_test =
-               sp->depth_stencil->depth.enabled &&
-               sp->framebuffer.zsbuf &&
-               !sp->depth_stencil->alpha.enabled &&
-               !sp->fs->info.uses_kill &&
-               !sp->fs->info.writes_z;
-
-   /* build up the pipeline in reverse order... */
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      sp->quad[i].first = sp->quad[i].output;
-
-      if (sp->blend->colormask != 0xf) {
-         sp_push_quad_first( sp, sp->quad[i].colormask, i );
-      }
+      sp->depth_stencil->depth.enabled &&
+      sp->framebuffer.zsbuf &&
+      !sp->depth_stencil->alpha.enabled &&
+      !sp->fs->info.uses_kill &&
+      !sp->fs->info.writes_z;
 
-      if (sp->blend->blend_enable ||
-          sp->blend->logicop_enable) {
-         sp_push_quad_first( sp, sp->quad[i].blend, i );
-      }
+   sp->quad.first = sp->quad.blend;
 
-      if (sp->active_query_count) {
-         sp_push_quad_first( sp, sp->quad[i].occlusion, i );
-      }
-
-      if (sp->rasterizer->poly_smooth ||
-          sp->rasterizer->line_smooth ||
-          sp->rasterizer->point_smooth) {
-         sp_push_quad_first( sp, sp->quad[i].coverage, i );
-      }
-
-      if (!early_depth_test) {
-         sp_build_depth_stencil( sp, i );
-      }
-
-      if (sp->depth_stencil->alpha.enabled) {
-         sp_push_quad_first( sp, sp->quad[i].alpha_test, i );
-      }
-
-      /* XXX always enable shader? */
-      if (1) {
-         sp_push_quad_first( sp, sp->quad[i].shade, i );
-      }
-
-      if (early_depth_test) {
-         sp_build_depth_stencil( sp, i );
-         sp_push_quad_first( sp, sp->quad[i].earlyz, i );
-      }
-
-#if !USE_DRAW_STAGE_PSTIPPLE
-      if (sp->rasterizer->poly_stipple_enable) {
-         sp_push_quad_first( sp, sp->quad[i].polygon_stipple, i );
-      }
-#endif
+   if (early_depth_test) {
+      sp_push_quad_first( sp, sp->quad.shade );
+      sp_push_quad_first( sp, sp->quad.depth_test );
+   }
+   else {
+      sp_push_quad_first( sp, sp->quad.depth_test );
+      sp_push_quad_first( sp, sp->quad.shade );
    }
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_pipe.h b/src/gallium/drivers/softpipe/sp_quad_pipe.h
index 0e40586ffc..c0aa134831 100644
--- a/src/gallium/drivers/softpipe/sp_quad_pipe.h
+++ b/src/gallium/drivers/softpipe/sp_quad_pipe.h
@@ -49,7 +49,7 @@ struct quad_stage {
    void (*begin)(struct quad_stage *qs);
 
    /** the stage action */
-   void (*run)(struct quad_stage *qs, struct quad_header *quad);
+   void (*run)(struct quad_stage *qs, struct quad_header *quad[], unsigned nr);
 
    void (*destroy)(struct quad_stage *qs);
 };
@@ -69,6 +69,4 @@ struct quad_stage *sp_quad_output_stage( struct softpipe_context *softpipe );
 
 void sp_build_quad_pipeline(struct softpipe_context *sp);
 
-void sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad);
-
 #endif /* SP_QUAD_PIPE_H */
diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c
deleted file mode 100644
index 5e9d447737..0000000000
--- a/src/gallium/drivers/softpipe/sp_quad_stencil.c
+++ /dev/null
@@ -1,352 +0,0 @@
-
-/**
- * \brief Quad stencil testing
- */
-
-
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_tile_cache.h"
-#include "sp_quad_pipe.h"
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-
-
-/** Only 8-bit stencil supported */
-#define STENCIL_MAX 0xff
-
-
-/**
- * Do the basic stencil test (compare stencil buffer values against the
- * reference value.
- *
- * \param stencilVals  the stencil values from the stencil buffer
- * \param func  the stencil func (PIPE_FUNC_x)
- * \param ref  the stencil reference value
- * \param valMask  the stencil value mask indicating which bits of the stencil
- *                 values and ref value are to be used.
- * \return mask indicating which pixels passed the stencil test
- */
-static unsigned
-do_stencil_test(const ubyte stencilVals[QUAD_SIZE], unsigned func,
-                unsigned ref, unsigned valMask)
-{
-   unsigned passMask = 0x0;
-   unsigned j;
-
-   ref &= valMask;
-
-   switch (func) {
-   case PIPE_FUNC_NEVER:
-      /* passMask = 0x0 */
-      break;
-   case PIPE_FUNC_LESS:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref < (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_EQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref == (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_LEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref <= (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_GREATER:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref > (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_NOTEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref != (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_GEQUAL:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (ref >= (stencilVals[j] & valMask)) {
-            passMask |= (1 << j);
-         }
-      }
-      break;
-   case PIPE_FUNC_ALWAYS:
-      passMask = MASK_ALL;
-      break;
-   default:
-      assert(0);
-   }
-
-   return passMask;
-}
-
-
-/**
- * Apply the stencil operator to stencil values.
- *
- * \param stencilVals  the stencil buffer values (read and written)
- * \param mask  indicates which pixels to update
- * \param op  the stencil operator (PIPE_STENCIL_OP_x)
- * \param ref  the stencil reference value
- * \param wrtMask  writemask controlling which bits are changed in the
- *                 stencil values
- */
-static void
-apply_stencil_op(ubyte stencilVals[QUAD_SIZE],
-                 unsigned mask, unsigned op, ubyte ref, ubyte wrtMask)
-{
-   unsigned j;
-   ubyte newstencil[QUAD_SIZE];
-
-   for (j = 0; j < QUAD_SIZE; j++) {
-      newstencil[j] = stencilVals[j];
-   }
-
-   switch (op) {
-   case PIPE_STENCIL_OP_KEEP:
-      /* no-op */
-      break;
-   case PIPE_STENCIL_OP_ZERO:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            newstencil[j] = 0;
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_REPLACE:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            newstencil[j] = ref;
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_INCR:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            if (stencilVals[j] < STENCIL_MAX) {
-               newstencil[j] = stencilVals[j] + 1;
-            }
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_DECR:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            if (stencilVals[j] > 0) {
-               newstencil[j] = stencilVals[j] - 1;
-            }
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_INCR_WRAP:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            newstencil[j] = stencilVals[j] + 1;
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_DECR_WRAP:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            newstencil[j] = stencilVals[j] - 1;
-         }
-      }
-      break;
-   case PIPE_STENCIL_OP_INVERT:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (mask & (1 << j)) {
-            newstencil[j] = ~stencilVals[j];
-         }
-      }
-      break;
-   default:
-      assert(0);
-   }
-
-   /*
-    * update the stencil values
-    */
-   if (wrtMask != STENCIL_MAX) {
-      /* apply bit-wise stencil buffer writemask */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         stencilVals[j] = (wrtMask & newstencil[j]) | (~wrtMask & stencilVals[j]);
-      }
-   }
-   else {
-      for (j = 0; j < QUAD_SIZE; j++) {
-         stencilVals[j] = newstencil[j];
-      }
-   }
-}
-
-
-/**
- * Do stencil (and depth) testing.  Stenciling depends on the outcome of
- * depth testing.
- */
-static void
-stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-   struct pipe_surface *ps = softpipe->framebuffer.zsbuf;
-   unsigned func, zFailOp, zPassOp, failOp;
-   ubyte ref, wrtMask, valMask;
-   ubyte stencilVals[QUAD_SIZE];
-   struct softpipe_cached_tile *tile
-      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0);
-   uint j;
-   uint face = quad->input.facing;
-
-   if (!softpipe->depth_stencil->stencil[1].enabled) {
-      /* single-sided stencil test, use front (face=0) state */
-      face = 0;
-   }
-
-   /* choose front or back face function, operator, etc */
-   /* XXX we could do these initializations once per primitive */
-   func    = softpipe->depth_stencil->stencil[face].func;
-   failOp  = softpipe->depth_stencil->stencil[face].fail_op;
-   zFailOp = softpipe->depth_stencil->stencil[face].zfail_op;
-   zPassOp = softpipe->depth_stencil->stencil[face].zpass_op;
-   ref     = softpipe->depth_stencil->stencil[face].ref_value;
-   wrtMask = softpipe->depth_stencil->stencil[face].writemask;
-   valMask = softpipe->depth_stencil->stencil[face].valuemask;
-
-   assert(ps); /* shouldn't get here if there's no stencil buffer */
-
-   /* get stencil values from cached tile */
-   switch (ps->format) {
-   case PIPE_FORMAT_S8Z24_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         stencilVals[j] = tile->data.depth32[y][x] >> 24;
-      }
-      break;
-   case PIPE_FORMAT_Z24S8_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         stencilVals[j] = tile->data.depth32[y][x] & 0xff;
-      }
-      break;
-   case PIPE_FORMAT_S8_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         stencilVals[j] = tile->data.stencil8[y][x];
-      }
-      break;
-   default:
-      assert(0);
-   }
-
-   /* do the stencil test first */
-   {
-      unsigned passMask, failMask;
-      passMask = do_stencil_test(stencilVals, func, ref, valMask);
-      failMask = quad->inout.mask & ~passMask;
-      quad->inout.mask &= passMask;
-
-      if (failOp != PIPE_STENCIL_OP_KEEP) {
-         apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask);
-      }
-   }
-
-   if (quad->inout.mask) {
-      /* now the pixels that passed the stencil test are depth tested */
-      if (softpipe->depth_stencil->depth.enabled) {
-         const unsigned origMask = quad->inout.mask;
-
-         sp_depth_test_quad(qs, quad);  /* quad->mask is updated */
-
-         /* update stencil buffer values according to z pass/fail result */
-         if (zFailOp != PIPE_STENCIL_OP_KEEP) {
-            const unsigned failMask = origMask & ~quad->inout.mask;
-            apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask);
-         }
-
-         if (zPassOp != PIPE_STENCIL_OP_KEEP) {
-            const unsigned passMask = origMask & quad->inout.mask;
-            apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask);
-         }
-      }
-      else {
-         /* no depth test, apply Zpass operator to stencil buffer values */
-         apply_stencil_op(stencilVals, quad->inout.mask, zPassOp, ref, wrtMask);
-      }
-
-   }
-
-   /* put new stencil values into cached tile */
-   switch (ps->format) {
-   case PIPE_FORMAT_S8Z24_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         uint s8z24 = tile->data.depth32[y][x];
-         s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff);
-         tile->data.depth32[y][x] = s8z24;
-      }
-      break;
-   case PIPE_FORMAT_Z24S8_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         uint z24s8 = tile->data.depth32[y][x];
-         z24s8 = (z24s8 & 0xffffff00) | stencilVals[j];
-         tile->data.depth32[y][x] = z24s8;
-      }
-      break;
-   case PIPE_FORMAT_S8_UNORM:
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->input.x0 % TILE_SIZE + (j & 1);
-         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
-         tile->data.stencil8[y][x] = stencilVals[j];
-      }
-      break;
-   default:
-      assert(0);
-   }
-
-   if (quad->inout.mask)
-      qs->next->run(qs->next, quad);
-}
-
-
-static void stencil_begin(struct quad_stage *qs)
-{
-   qs->next->begin(qs->next);
-}
-
-
-static void stencil_destroy(struct quad_stage *qs)
-{
-   FREE( qs );
-}
-
-
-struct quad_stage *sp_quad_stencil_test_stage( struct softpipe_context *softpipe )
-{
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = stencil_begin;
-   stage->run = stencil_test_quad;
-   stage->destroy = stencil_destroy;
-
-   return stage;
-}
diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c
index 07162db7b6..a0527a596a 100644
--- a/src/gallium/drivers/softpipe/sp_quad_stipple.c
+++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c
@@ -14,14 +14,20 @@
  * Apply polygon stipple to quads produced by triangle rasterization
  */
 static void
-stipple_quad(struct quad_stage *qs, struct quad_header *quad)
+stipple_quad(struct quad_stage *qs, struct quad_header *quads[], unsigned nr)
 {
    static const uint bit31 = 1 << 31;
    static const uint bit30 = 1 << 30;
+   unsigned pass = nr;
+
+   struct softpipe_context *softpipe = qs->softpipe;
+   unsigned q;
+
+   pass = 0;
+
+   for (q = 0; q < nr; q++)  {
+      struct quad_header *quad = quads[q];
 
-   if (quad->input.prim == QUAD_PRIM_TRI) {
-      struct softpipe_context *softpipe = qs->softpipe;
-      /* need to invert Y to index into OpenGL's stipple pattern */
       const int col0 = quad->input.x0 % 32;
       const int y0 = quad->input.y0;
       const int y1 = y0 + 1;
@@ -41,13 +47,11 @@ stipple_quad(struct quad_stage *qs, struct quad_header *quad)
       if ((stipple1 & (bit30 >> col0)) == 0)
          quad->inout.mask &= ~MASK_BOTTOM_RIGHT;
 
-      if (!quad->inout.mask) {
-         /* all fragments failed stipple test, end of quad pipeline */
-         return;
-      }
+      if (quad->inout.mask)
+         quads[pass++] = quad;
    }
 
-   qs->next->run(qs->next, quad);
+   qs->next->run(qs->next, quads, pass);
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index ce77018415..81fb7aa20c 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -40,7 +40,7 @@
 static const char *
 softpipe_get_vendor(struct pipe_screen *screen)
 {
-   return "Tungsten Graphics, Inc.";
+   return "VMware, Inc.";
 }
 
 
@@ -82,7 +82,7 @@ softpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
       return 13; /* max 4Kx4K */
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return 9;  /* max 256x256x256 */
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
       return 13; /* max 4Kx4K */
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
@@ -135,6 +135,9 @@ softpipe_is_format_supported( struct pipe_screen *screen,
           target == PIPE_TEXTURE_CUBE);
 
    switch(format) {
+   case PIPE_FORMAT_L16_UNORM:
+   case PIPE_FORMAT_YCBCR_REV:
+   case PIPE_FORMAT_YCBCR:
    case PIPE_FORMAT_DXT1_RGB:
    case PIPE_FORMAT_DXT1_RGBA:
    case PIPE_FORMAT_DXT3_RGBA:
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index de3ae3c369..ade125662a 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -33,7 +33,6 @@
  */
 
 #include "sp_context.h"
-#include "sp_prim_setup.h"
 #include "sp_quad.h"
 #include "sp_quad_pipe.h"
 #include "sp_setup.h"
@@ -61,87 +60,9 @@ struct edge {
    int lines;		/**< number of lines on this edge */
 };
 
-#if SP_NUM_QUAD_THREADS > 1
 
-/* Set to 1 if you want other threads to be instantly
- * notified of pending jobs.
- */
-#define INSTANT_NOTEMPTY_NOTIFY 0
-
-struct thread_info
-{
-   struct setup_context *setup;
-   uint id;
-   pipe_thread handle;
-};
-
-struct quad_job;
-
-typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job );
-
-struct quad_job
-{
-   struct quad_header_input input;
-   struct quad_header_inout inout;
-   quad_job_routine routine;
-};
-
-#define NUM_QUAD_JOBS 64
-
-struct quad_job_que
-{
-   struct quad_job jobs[NUM_QUAD_JOBS];
-   uint first;
-   uint last;
-   pipe_mutex que_mutex;
-   pipe_condvar que_notfull_condvar;
-   pipe_condvar que_notempty_condvar;
-   uint jobs_added;
-   uint jobs_done;
-   pipe_condvar que_done_condvar;
-};
-
-static void
-add_quad_job( struct quad_job_que *que, struct quad_header *quad, quad_job_routine routine )
-{
-#if INSTANT_NOTEMPTY_NOTIFY
-   boolean empty;
-#endif
-
-   /* Wait for empty slot, see if the que is empty.
-    */
-   pipe_mutex_lock( que->que_mutex );
-   while ((que->last + 1) % NUM_QUAD_JOBS == que->first) {
-#if !INSTANT_NOTEMPTY_NOTIFY
-      pipe_condvar_broadcast( que->que_notempty_condvar );
-#endif
-      pipe_condvar_wait( que->que_notfull_condvar, que->que_mutex );
-   }
-#if INSTANT_NOTEMPTY_NOTIFY
-   empty = que->last == que->first;
-#endif
-   que->jobs_added++;
-   pipe_mutex_unlock( que->que_mutex );
+#define MAX_QUADS 16
 
-   /* Submit new job.
-    */
-   que->jobs[que->last].input = quad->input;
-   que->jobs[que->last].inout = quad->inout;
-   que->jobs[que->last].routine = routine;
-   que->last = (que->last + 1) % NUM_QUAD_JOBS;
-
-#if INSTANT_NOTEMPTY_NOTIFY
-   /* If the que was empty, notify consumers there's a job to be done.
-    */
-   if (empty) {
-      pipe_mutex_lock( que->que_mutex );
-      pipe_condvar_broadcast( que->que_notempty_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-   }
-#endif
-}
-
-#endif
 
 /**
  * Triangle setup info (derived from draw_stage).
@@ -164,22 +85,19 @@ struct setup_context {
    struct edge emaj;
 
    float oneoverarea;
+   int facing;
+
+   struct quad_header quad[MAX_QUADS];
+   struct quad_header *quad_ptrs[MAX_QUADS];
+   unsigned count;
 
    struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS];
    struct tgsi_interp_coef posCoef;  /* For Z, W */
-   struct quad_header quad;
-
-#if SP_NUM_QUAD_THREADS > 1
-   struct quad_job_que que;
-   struct thread_info threads[SP_NUM_QUAD_THREADS];
-#endif
 
    struct {
       int left[2];   /**< [0] = row0, [1] = row1 */
       int right[2];
       int y;
-      unsigned y_flags;
-      unsigned mask;     /**< mask of MASK_BOTTOM/TOP_LEFT/RIGHT bits */
    } span;
 
 #if DEBUG_FRAGS
@@ -190,67 +108,6 @@ struct setup_context {
    unsigned winding;		/* which winding to cull */
 };
 
-#if SP_NUM_QUAD_THREADS > 1
-
-static PIPE_THREAD_ROUTINE( quad_thread, param )
-{
-   struct thread_info *info = (struct thread_info *) param;
-   struct quad_job_que *que = &info->setup->que;
-
-   for (;;) {
-      struct quad_job job;
-      boolean full;
-
-      /* Wait for an available job.
-       */
-      pipe_mutex_lock( que->que_mutex );
-      while (que->last == que->first)
-         pipe_condvar_wait( que->que_notempty_condvar, que->que_mutex );
-
-      /* See if the que is full.
-       */
-      full = (que->last + 1) % NUM_QUAD_JOBS == que->first;
-
-      /* Take a job and remove it from que.
-       */
-      job = que->jobs[que->first];
-      que->first = (que->first + 1) % NUM_QUAD_JOBS;
-
-      /* Notify the producer if the que is not full.
-       */
-      if (full)
-         pipe_condvar_signal( que->que_notfull_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-
-      job.routine( info->setup, info->id, &job );
-
-      /* Notify the producer if that's the last finished job.
-       */
-      pipe_mutex_lock( que->que_mutex );
-      que->jobs_done++;
-      if (que->jobs_added == que->jobs_done)
-         pipe_condvar_signal( que->que_done_condvar );
-      pipe_mutex_unlock( que->que_mutex );
-   }
-
-   return NULL;
-}
-
-#define WAIT_FOR_COMPLETION(setup) \
-   do {\
-      pipe_mutex_lock( setup->que.que_mutex );\
-      if (!INSTANT_NOTEMPTY_NOTIFY)\
-         pipe_condvar_broadcast( setup->que.que_notempty_condvar );\
-      while (setup->que.jobs_added != setup->que.jobs_done)\
-         pipe_condvar_wait( setup->que.que_done_condvar, setup->que.que_mutex );\
-      pipe_mutex_unlock( setup->que.que_mutex );\
-   } while (0)
-
-#else
-
-#define WAIT_FOR_COMPLETION(setup) ((void) 0)
-
-#endif
 
 
 
@@ -313,98 +170,18 @@ quad_clip( struct setup_context *setup, struct quad_header *quad )
  * Emit a quad (pass to next stage) with clipping.
  */
 static INLINE void
-clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
+clip_emit_quad( struct setup_context *setup, struct quad_header *quad )
 {
    quad_clip( setup, quad );
+
    if (quad->inout.mask) {
       struct softpipe_context *sp = setup->softpipe;
 
-      sp->quad[thread].first->run( sp->quad[thread].first, quad );
+      sp->quad.first->run( sp->quad.first, &quad, 1 );
    }
 }
 
-#if SP_NUM_QUAD_THREADS > 1
-
-static void
-clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
-{
-   struct quad_header quad;
-
-   quad.input = job->input;
-   quad.inout = job->inout;
-   quad.coef = setup->quad.coef;
-   quad.posCoef = setup->quad.posCoef;
-   quad.nr_attrs = setup->quad.nr_attrs;
-   clip_emit_quad( setup, &quad, thread );
-}
-
-#define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, &setup->quad, clip_emit_quad_job )
-
-#else
-
-#define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 )
-
-#endif
-
-/**
- * Emit a quad (pass to next stage).  No clipping is done.
- */
-static INLINE void
-emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
-{
-   struct softpipe_context *sp = setup->softpipe;
-#if DEBUG_FRAGS
-   uint mask = quad->inout.mask;
-#endif
-
-#if DEBUG_FRAGS
-   if (mask & 1) setup->numFragsEmitted++;
-   if (mask & 2) setup->numFragsEmitted++;
-   if (mask & 4) setup->numFragsEmitted++;
-   if (mask & 8) setup->numFragsEmitted++;
-#endif
-   sp->quad[thread].first->run( sp->quad[thread].first, quad );
-#if DEBUG_FRAGS
-   mask = quad->inout.mask;
-   if (mask & 1) setup->numFragsWritten++;
-   if (mask & 2) setup->numFragsWritten++;
-   if (mask & 4) setup->numFragsWritten++;
-   if (mask & 8) setup->numFragsWritten++;
-#endif
-}
-
-#if SP_NUM_QUAD_THREADS > 1
 
-static void
-emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
-{
-   struct quad_header quad;
-
-   quad.input = job->input;
-   quad.inout = job->inout;
-   quad.coef = setup->quad.coef;
-   quad.posCoef = setup->quad.posCoef;
-   quad.nr_attrs = setup->quad.nr_attrs;
-   emit_quad( setup, &quad, thread );
-}
-
-#define EMIT_QUAD(setup,x,y,mask) do {\
-      setup->quad.input.x0 = x;\
-      setup->quad.input.y0 = y;\
-      setup->quad.inout.mask = mask;\
-      add_quad_job( &setup->que, &setup->quad, emit_quad_job );\
-   } while (0)
-
-#else
-
-#define EMIT_QUAD(setup,x,y,mask) do {\
-      setup->quad.input.x0 = x;\
-      setup->quad.input.y0 = y;\
-      setup->quad.inout.mask = mask;\
-      emit_quad( setup, &setup->quad, 0 );\
-   } while (0)
-
-#endif
 
 /**
  * Given an X or Y coordinate, return the block/quad coordinate that it
@@ -412,7 +189,12 @@ emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
  */
 static INLINE int block( int x )
 {
-   return x & ~1;
+   return x & ~(2-1);
+}
+
+static INLINE int block_x( int x )
+{
+   return x & ~(16-1);
 }
 
 
@@ -421,72 +203,63 @@ static INLINE int block( int x )
  */
 static void flush_spans( struct setup_context *setup )
 {
+   const int step = 16;
    const int xleft0 = setup->span.left[0];
    const int xleft1 = setup->span.left[1];
    const int xright0 = setup->span.right[0];
    const int xright1 = setup->span.right[1];
-   int minleft, maxright;
+   struct quad_stage *pipe = setup->softpipe->quad.first;
+
+
+   int minleft = block_x(MIN2(xleft0, xleft1));
+   int maxright = MAX2(xright0, xright1);
    int x;
 
-   switch (setup->span.y_flags) {
-   case 0x3:
-      /* both odd and even lines written (both quad rows) */
-      minleft = block(MIN2(xleft0, xleft1));
-      maxright = block(MAX2(xright0, xright1));
-      for (x = minleft; x <= maxright; x += 2) {
-         /* determine which of the four pixels is inside the span bounds */
-         uint mask = 0x0;
-         if (x >= xleft0 && x < xright0)
-            mask |= MASK_TOP_LEFT;
-         if (x >= xleft1 && x < xright1)
-            mask |= MASK_BOTTOM_LEFT;
-         if (x+1 >= xleft0 && x+1 < xright0)
-            mask |= MASK_TOP_RIGHT;
-         if (x+1 >= xleft1 && x+1 < xright1)
-            mask |= MASK_BOTTOM_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
-
-   case 0x1:
-      /* only even line written (quad top row) */
-      minleft = block(xleft0);
-      maxright = block(xright0);
-      for (x = minleft; x <= maxright; x += 2) {
-         uint mask = 0x0;
-         if (x >= xleft0 && x < xright0)
-            mask |= MASK_TOP_LEFT;
-         if (x+1 >= xleft0 && x+1 < xright0)
-            mask |= MASK_TOP_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
-
-   case 0x2:
-      /* only odd line written (quad bottom row) */
-      minleft = block(xleft1);
-      maxright = block(xright1);
-      for (x = minleft; x <= maxright; x += 2) {
-         uint mask = 0x0;
-         if (x >= xleft1 && x < xright1)
-            mask |= MASK_BOTTOM_LEFT;
-         if (x+1 >= xleft1 && x+1 < xright1)
-            mask |= MASK_BOTTOM_RIGHT;
-         if (mask)
-            EMIT_QUAD( setup, x, setup->span.y, mask );
-      }
-      break;
+   for (x = minleft; x < maxright; x += step) {
+      unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
+      unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
+      unsigned skip_right0 = CLAMP(x + step - xright0, 0, step);
+      unsigned skip_right1 = CLAMP(x + step - xright1, 0, step);
+      unsigned lx = x;
+      unsigned q = 0;
 
-   default:
-      return;
+      unsigned skipmask_left0 = (1U << skip_left0) - 1U;
+      unsigned skipmask_left1 = (1U << skip_left1) - 1U;
+
+      /* These calculations fail when step == 32 and skip_right == 0.
+       */
+      unsigned skipmask_right0 = ~0U << (unsigned)(step - skip_right0);
+      unsigned skipmask_right1 = ~0U << (unsigned)(step - skip_right1);
+
+      unsigned mask0 = ~skipmask_left0 & ~skipmask_right0;
+      unsigned mask1 = ~skipmask_left1 & ~skipmask_right1;
+
+      if (mask0 | mask1) {
+         do {
+            unsigned quadmask = (mask0 & 3) | ((mask1 & 3) << 2);
+            if (quadmask) {
+               setup->quad[q].input.x0 = lx;
+               setup->quad[q].input.y0 = setup->span.y;
+               setup->quad[q].input.facing = setup->facing;
+               setup->quad[q].inout.mask = quadmask;
+               setup->quad_ptrs[q] = &setup->quad[q];
+               q++;
+            }
+            mask0 >>= 2;
+            mask1 >>= 2;
+            lx += 2;
+         } while (mask0 | mask1);
+
+         pipe->run( pipe, setup->quad_ptrs, q );
+      }
    }
 
+
    setup->span.y = 0;
-   setup->span.y_flags = 0;
    setup->span.right[0] = 0;
    setup->span.right[1] = 0;
+   setup->span.left[0] = 1000000;     /* greater than right[0] */
+   setup->span.left[1] = 1000000;     /* greater than right[1] */
 }
 
 
@@ -496,7 +269,7 @@ static void print_vertex(const struct setup_context *setup,
 {
    int i;
    debug_printf("   Vertex: (%p)\n", v);
-   for (i = 0; i < setup->quad.nr_attrs; i++) {
+   for (i = 0; i < setup->quad[0].nr_attrs; i++) {
       debug_printf("     %d: %f %f %f %f\n",  i,
               v[i][0], v[i][1], v[i][2], v[i][3]);
       if (util_is_inf_or_nan(v[i][0])) {
@@ -601,7 +374,9 @@ static boolean setup_sort_vertices( struct setup_context *setup,
     *  - the GLSL gl_FrontFacing fragment attribute (bool)
     *  - two-sided stencil test
     */
-   setup->quad.input.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+   setup->facing = 
+      ((det > 0.0) ^ 
+       (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW));
 
    return TRUE;
 }
@@ -788,7 +563,7 @@ static void setup_tri_coefficients( struct setup_context *setup )
       }
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+         setup->coef[fragSlot].a0[0] = 1.0f - setup->facing;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
@@ -844,11 +619,10 @@ static void subtriangle( struct setup_context *setup,
 
    /* clip top/bottom */
    start_y = sy;
-   finish_y = sy + lines;
-
    if (start_y < miny)
       start_y = miny;
 
+   finish_y = sy + lines;
    if (finish_y > maxy)
       finish_y = maxy;
 
@@ -885,7 +659,6 @@ static void subtriangle( struct setup_context *setup,
 
          setup->span.left[_y&1] = left;
          setup->span.right[_y&1] = right;
-         setup->span.y_flags |= 1<<(_y&1);
       }
    }
 
@@ -958,10 +731,9 @@ void setup_tri( struct setup_context *setup,
    setup_tri_coefficients( setup );
    setup_tri_edges( setup );
 
-   setup->quad.input.prim = QUAD_PRIM_TRI;
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_TRIANGLES);
 
    setup->span.y = 0;
-   setup->span.y_flags = 0;
    setup->span.right[0] = 0;
    setup->span.right[1] = 0;
    /*   setup->span.z_mode = tri_z_mode( setup->ctx ); */
@@ -983,8 +755,6 @@ void setup_tri( struct setup_context *setup,
 
    flush_spans( setup );
 
-   WAIT_FOR_COMPLETION(setup);
-
 #if DEBUG_FRAGS
    printf("Tri: %u frags emitted, %u written\n",
           setup->numFragsEmitted,
@@ -1101,7 +871,7 @@ setup_line_coefficients(struct setup_context *setup,
       }
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+         setup->coef[fragSlot].a0[0] = 1.0f - setup->facing;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
@@ -1122,20 +892,20 @@ plot(struct setup_context *setup, int x, int y)
    const int quadY = y - iy;
    const int mask = (1 << ix) << (2 * iy);
 
-   if (quadX != setup->quad.input.x0 ||
-       quadY != setup->quad.input.y0)
+   if (quadX != setup->quad[0].input.x0 ||
+       quadY != setup->quad[0].input.y0)
    {
       /* flush prev quad, start new quad */
 
-      if (setup->quad.input.x0 != -1)
-         CLIP_EMIT_QUAD(setup);
+      if (setup->quad[0].input.x0 != -1)
+         clip_emit_quad( setup, &setup->quad[0] );
 
-      setup->quad.input.x0 = quadX;
-      setup->quad.input.y0 = quadY;
-      setup->quad.inout.mask = 0x0;
+      setup->quad[0].input.x0 = quadX;
+      setup->quad[0].input.y0 = quadY;
+      setup->quad[0].inout.mask = 0x0;
    }
 
-   setup->quad.inout.mask |= mask;
+   setup->quad[0].inout.mask |= mask;
 }
 
 
@@ -1195,17 +965,18 @@ setup_line(struct setup_context *setup,
 
    assert(dx >= 0);
    assert(dy >= 0);
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_LINES);
+
+   setup->quad[0].input.x0 = setup->quad[0].input.y0 = -1;
+   setup->quad[0].inout.mask = 0x0;
 
-   setup->quad.input.x0 = setup->quad.input.y0 = -1;
-   setup->quad.inout.mask = 0x0;
-   setup->quad.input.prim = QUAD_PRIM_LINE;
    /* XXX temporary: set coverage to 1.0 so the line appears
     * if AA mode happens to be enabled.
     */
-   setup->quad.input.coverage[0] =
-   setup->quad.input.coverage[1] =
-   setup->quad.input.coverage[2] =
-   setup->quad.input.coverage[3] = 1.0;
+   setup->quad[0].input.coverage[0] =
+   setup->quad[0].input.coverage[1] =
+   setup->quad[0].input.coverage[2] =
+   setup->quad[0].input.coverage[3] = 1.0;
 
    if (dx > dy) {
       /*** X-major line ***/
@@ -1249,11 +1020,9 @@ setup_line(struct setup_context *setup,
    }
 
    /* draw final quad */
-   if (setup->quad.inout.mask) {
-      CLIP_EMIT_QUAD(setup);
+   if (setup->quad[0].inout.mask) {
+      clip_emit_quad( setup, &setup->quad[0] );
    }
-
-   WAIT_FOR_COMPLETION(setup);
 }
 
 
@@ -1300,6 +1069,8 @@ setup_point( struct setup_context *setup,
    if (softpipe->no_rast)
       return;
 
+   assert(setup->softpipe->reduced_prim == PIPE_PRIM_POINTS);
+
    /* For points, all interpolants are constant-valued.
     * However, for point sprites, we'll need to setup texcoords appropriately.
     * XXX: which coefficients are the texcoords???
@@ -1346,22 +1117,21 @@ setup_point( struct setup_context *setup,
       }
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
-         setup->coef[fragSlot].a0[0] = 1.0f - setup->quad.input.facing;
+         setup->coef[fragSlot].a0[0] = 1.0f - setup->facing;
          setup->coef[fragSlot].dadx[0] = 0.0;
          setup->coef[fragSlot].dady[0] = 0.0;
       }
    }
 
-   setup->quad.input.prim = QUAD_PRIM_POINT;
 
    if (halfSize <= 0.5 && !round) {
       /* special case for 1-pixel points */
       const int ix = ((int) x) & 1;
       const int iy = ((int) y) & 1;
-      setup->quad.input.x0 = (int) x - ix;
-      setup->quad.input.y0 = (int) y - iy;
-      setup->quad.inout.mask = (1 << ix) << (2 * iy);
-      CLIP_EMIT_QUAD(setup);
+      setup->quad[0].input.x0 = (int) x - ix;
+      setup->quad[0].input.y0 = (int) y - iy;
+      setup->quad[0].inout.mask = (1 << ix) << (2 * iy);
+      clip_emit_quad( setup, &setup->quad[0] );
    }
    else {
       if (round) {
@@ -1381,15 +1151,15 @@ setup_point( struct setup_context *setup,
             for (ix = ixmin; ix <= ixmax; ix += 2) {
                float dx, dy, dist2, cover;
 
-               setup->quad.inout.mask = 0x0;
+               setup->quad[0].inout.mask = 0x0;
 
                dx = (ix + 0.5f) - x;
                dy = (iy + 0.5f) - y;
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_TOP_LEFT;
+                  setup->quad[0].input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_TOP_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1397,8 +1167,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_TOP_RIGHT;
+                  setup->quad[0].input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_TOP_RIGHT;
                }
 
                dx = (ix + 0.5f) - x;
@@ -1406,8 +1176,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_BOTTOM_LEFT;
+                  setup->quad[0].input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_BOTTOM_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1415,14 +1185,14 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.inout.mask |= MASK_BOTTOM_RIGHT;
+                  setup->quad[0].input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad[0].inout.mask |= MASK_BOTTOM_RIGHT;
                }
 
-               if (setup->quad.inout.mask) {
-                  setup->quad.input.x0 = ix;
-                  setup->quad.input.y0 = iy;
-                  CLIP_EMIT_QUAD(setup);
+               if (setup->quad[0].inout.mask) {
+                  setup->quad[0].input.x0 = ix;
+                  setup->quad[0].input.y0 = iy;
+                  clip_emit_quad( setup, &setup->quad[0] );
                }
             }
          }
@@ -1466,33 +1236,25 @@ setup_point( struct setup_context *setup,
                   mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
                }
 
-               setup->quad.inout.mask = mask;
-               setup->quad.input.x0 = ix;
-               setup->quad.input.y0 = iy;
-               CLIP_EMIT_QUAD(setup);
+               setup->quad[0].inout.mask = mask;
+               setup->quad[0].input.x0 = ix;
+               setup->quad[0].input.y0 = iy;
+               clip_emit_quad( setup, &setup->quad[0] );
             }
          }
       }
    }
-
-   WAIT_FOR_COMPLETION(setup);
 }
 
 void setup_prepare( struct setup_context *setup )
 {
    struct softpipe_context *sp = setup->softpipe;
-   unsigned i;
 
    if (sp->dirty) {
       softpipe_update_derived(sp);
    }
 
-   /* Note: nr_attrs is only used for debugging (vertex printing) */
-   setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw);
-
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      sp->quad[i].first->begin( sp->quad[i].first );
-   }
+   sp->quad.first->begin( sp->quad.first );
 
    if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES &&
        sp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL &&
@@ -1520,30 +1282,17 @@ void setup_destroy_context( struct setup_context *setup )
 struct setup_context *setup_create_context( struct softpipe_context *softpipe )
 {
    struct setup_context *setup = CALLOC_STRUCT(setup_context);
-#if SP_NUM_QUAD_THREADS > 1
-   uint i;
-#endif
+   unsigned i;
 
    setup->softpipe = softpipe;
 
-   setup->quad.coef = setup->coef;
-   setup->quad.posCoef = &setup->posCoef;
-
-#if SP_NUM_QUAD_THREADS > 1
-   setup->que.first = 0;
-   setup->que.last = 0;
-   pipe_mutex_init( setup->que.que_mutex );
-   pipe_condvar_init( setup->que.que_notfull_condvar );
-   pipe_condvar_init( setup->que.que_notempty_condvar );
-   setup->que.jobs_added = 0;
-   setup->que.jobs_done = 0;
-   pipe_condvar_init( setup->que.que_done_condvar );
-   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
-      setup->threads[i].setup = setup;
-      setup->threads[i].id = i;
-      setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] );
+   for (i = 0; i < MAX_QUADS; i++) {
+      setup->quad[i].coef = setup->coef;
+      setup->quad[i].posCoef = &setup->posCoef;
    }
-#endif
+
+   setup->span.left[0] = 1000000;     /* greater than right[0] */
+   setup->span.left[1] = 1000000;     /* greater than right[1] */
 
    return setup;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 9776e978e3..77ee3c1136 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -87,6 +87,7 @@ struct sp_fragment_shader {
 struct sp_vertex_shader {
    struct pipe_shader_state shader;
    struct draw_vertex_shader *draw_data;
+   int max_sampler;             /* -1 if no samplers */
 };
 
 
diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c
index 384fe559af..efed082f82 100644
--- a/src/gallium/drivers/softpipe/sp_state_blend.c
+++ b/src/gallium/drivers/softpipe/sp_state_blend.c
@@ -45,7 +45,7 @@ void softpipe_bind_blend_state( struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->blend = (const struct pipe_blend_state *)blend;
+   softpipe->blend = (struct pipe_blend_state *)blend;
 
    softpipe->dirty |= SP_NEW_BLEND;
 }
@@ -86,7 +86,7 @@ softpipe_bind_depth_stencil_state(struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil;
+   softpipe->depth_stencil = (struct pipe_depth_stencil_alpha_state *)depth_stencil;
 
    softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index 9258f64109..1faeca1c2a 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -32,7 +32,10 @@
 #include "draw/draw_vertex.h"
 #include "draw/draw_private.h"
 #include "sp_context.h"
+#include "sp_screen.h"
 #include "sp_state.h"
+#include "sp_texture.h"
+#include "sp_tex_tile_cache.h"
 
 
 /**
@@ -65,24 +68,19 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
       const struct sp_fragment_shader *spfs = softpipe->fs;
       const enum interp_mode colorInterp
          = softpipe->rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
+      struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf;
+      const uint num = draw_num_vs_outputs(softpipe->draw);
       uint i;
 
-      if (softpipe->vbuf) {
-         /* if using the post-transform vertex buffer, tell draw_vbuf to
-          * simply emit the whole post-xform vertex as-is:
-          */
-         struct vertex_info *vinfo_vbuf = &softpipe->vertex_info_vbuf;
-         const uint num = draw_num_vs_outputs(softpipe->draw);
-         uint i;
-
-         /* No longer any need to try and emit draw vertex_header info.
-          */
-         vinfo_vbuf->num_attribs = 0;
-         for (i = 0; i < num; i++) {
-            draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i);
-         }
-         draw_compute_vertex_size(vinfo_vbuf);
+      /* Tell draw_vbuf to simply emit the whole post-xform vertex
+       * as-is.  No longer any need to try and emit draw vertex_header
+       * info.
+       */
+      vinfo_vbuf->num_attribs = 0;
+      for (i = 0; i < num; i++) {
+	 draw_emit_vertex_attr(vinfo_vbuf, EMIT_4F, INTERP_PERSPECTIVE, i);
       }
+      draw_compute_vertex_size(vinfo_vbuf);
 
       /*
        * Loop over fragment shader inputs, searching for the matching output
@@ -181,11 +179,19 @@ softpipe_get_vbuf_vertex_info(struct softpipe_context *softpipe)
 static void
 compute_cliprect(struct softpipe_context *sp)
 {
+   /* SP_NEW_FRAMEBUFFER
+    */
    uint surfWidth = sp->framebuffer.width;
    uint surfHeight = sp->framebuffer.height;
 
+   /* SP_NEW_RASTERIZER
+    */
    if (sp->rasterizer->scissor) {
-      /* clip to scissor rect */
+
+      /* SP_NEW_SCISSOR
+       *
+       * clip to scissor rect:
+       */
       sp->cliprect.minx = MAX2(sp->scissor.minx, 0);
       sp->cliprect.miny = MAX2(sp->scissor.miny, 0);
       sp->cliprect.maxx = MIN2(sp->scissor.maxx, surfWidth);
@@ -201,27 +207,63 @@ compute_cliprect(struct softpipe_context *sp)
 }
 
 
+static void
+update_tgsi_samplers( struct softpipe_context *softpipe )
+{
+   unsigned i;
+
+   softpipe_reset_sampler_varients( softpipe );
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      struct softpipe_tex_tile_cache *tc = softpipe->tex_cache[i];
+      if (tc->texture) {
+         struct softpipe_texture *spt = softpipe_texture(tc->texture);
+         if (spt->timestamp != tc->timestamp) {
+	    sp_tex_tile_cache_validate_texture( tc );
+            /*
+            _debug_printf("INV %d %d\n", tc->timestamp, spt->timestamp);
+            */
+            tc->timestamp = spt->timestamp;
+         }
+      }
+   }
+}
+
+
 /* Hopefully this will remain quite simple, otherwise need to pull in
  * something like the state tracker mechanism.
  */
 void softpipe_update_derived( struct softpipe_context *softpipe )
 {
+   struct softpipe_screen *sp_screen = softpipe_screen(softpipe->pipe.screen);
+
+   /* Check for updated textures.
+    */
+   if (softpipe->tex_timestamp != sp_screen->timestamp) {
+      softpipe->tex_timestamp = sp_screen->timestamp;
+      softpipe->dirty |= SP_NEW_TEXTURE;
+   }
+      
+   if (softpipe->dirty & (SP_NEW_SAMPLER |
+                          SP_NEW_TEXTURE |
+                          SP_NEW_FS | 
+                          SP_NEW_VS))
+      update_tgsi_samplers( softpipe );
+
    if (softpipe->dirty & (SP_NEW_RASTERIZER |
                           SP_NEW_FS |
                           SP_NEW_VS))
       invalidate_vertex_layout( softpipe );
 
    if (softpipe->dirty & (SP_NEW_SCISSOR |
-                          SP_NEW_DEPTH_STENCIL_ALPHA |
+                          SP_NEW_RASTERIZER |
                           SP_NEW_FRAMEBUFFER))
       compute_cliprect(softpipe);
 
    if (softpipe->dirty & (SP_NEW_BLEND |
                           SP_NEW_DEPTH_STENCIL_ALPHA |
                           SP_NEW_FRAMEBUFFER |
-                          SP_NEW_RASTERIZER |
-                          SP_NEW_FS | 
-			  SP_NEW_QUERY))
+                          SP_NEW_FS))
       sp_build_quad_pipeline(softpipe);
 
    softpipe->dirty = 0;
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index 4330c20393..256faa94b8 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -31,9 +31,8 @@
 
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
-#include "pipe/internal/p_winsys_screen.h"
-#include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
+#include "draw/draw_vs.h"
 #include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_scan.h"
 #include "tgsi/tgsi_parse.h"
@@ -51,12 +50,9 @@ softpipe_create_fs_state(struct pipe_context *pipe,
       tgsi_dump(templ->tokens, 0);
 
    /* codegen */
-   state = softpipe_create_fs_llvm( softpipe, templ );
+   state = softpipe_create_fs_sse( softpipe, templ );
    if (!state) {
-      state = softpipe_create_fs_sse( softpipe, templ );
-      if (!state) {
-         state = softpipe_create_fs_exec( softpipe, templ );
-      }
+      state = softpipe_create_fs_exec( softpipe, templ );
    }
 
    assert(state);
@@ -111,6 +107,8 @@ softpipe_create_vs_state(struct pipe_context *pipe,
    if (state->draw_data == NULL) 
       goto fail;
 
+   state->max_sampler = state->draw_data->info.file_max[TGSI_FILE_SAMPLER];
+
    return state;
 
 fail:
@@ -128,7 +126,7 @@ softpipe_bind_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->vs = (const struct sp_vertex_shader *)vs;
+   softpipe->vs = (struct sp_vertex_shader *) vs;
 
    draw_bind_vertex_shader(softpipe->draw,
                            (softpipe->vs ? softpipe->vs->draw_data : NULL));
@@ -142,8 +140,7 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   struct sp_vertex_shader *state =
-      (struct sp_vertex_shader *)vs;
+   struct sp_vertex_shader *state = (struct sp_vertex_shader *) vs;
 
    draw_delete_vertex_shader(softpipe->draw, state->draw_data);
    FREE( state );
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index cb517b02e4..db0b8ab76b 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -32,21 +32,37 @@
 #include "util/u_memory.h"
 
 #include "draw/draw_context.h"
+#include "draw/draw_context.h"
 
 #include "sp_context.h"
-#include "sp_context.h"
 #include "sp_state.h"
 #include "sp_texture.h"
-#include "sp_tile_cache.h"
-#include "draw/draw_context.h"
+#include "sp_tex_sample.h"
+#include "sp_tex_tile_cache.h"
 
 
+struct sp_sampler {
+   struct pipe_sampler_state base;
+   struct sp_sampler_varient *varients;
+   struct sp_sampler_varient *current;
+};
+
+static struct sp_sampler *sp_sampler( struct pipe_sampler_state *sampler )
+{
+   return (struct sp_sampler *)sampler;
+}
+
 
 void *
 softpipe_create_sampler_state(struct pipe_context *pipe,
                               const struct pipe_sampler_state *sampler)
 {
-   return mem_dup(sampler, sizeof(*sampler));
+   struct sp_sampler *sp_sampler = CALLOC_STRUCT(sp_sampler);
+
+   sp_sampler->base = *sampler;
+   sp_sampler->varients = NULL;
+
+   return (void *)sp_sampler;
 }
 
 
@@ -97,7 +113,7 @@ softpipe_set_sampler_textures(struct pipe_context *pipe,
       struct pipe_texture *tex = i < num ? texture[i] : NULL;
 
       pipe_texture_reference(&softpipe->texture[i], tex);
-      sp_tile_cache_set_texture(pipe, softpipe->tex_cache[i], tex);
+      sp_tex_tile_cache_set_texture(softpipe->tex_cache[i], tex);
    }
 
    softpipe->num_textures = num;
@@ -106,10 +122,111 @@ softpipe_set_sampler_textures(struct pipe_context *pipe,
 }
 
 
+/**
+ * Find/create an sp_sampler_varient object for sampling the given texture,
+ * sampler and tex unit.
+ *
+ * Note that the tex unit is significant.  We can't re-use a sampler
+ * varient for multiple texture units because the sampler varient contains
+ * the texture object pointer.  If the texture object pointer were stored
+ * somewhere outside the sampler varient, we could re-use samplers for
+ * multiple texture units.
+ */
+static struct sp_sampler_varient *
+get_sampler_varient( unsigned unit,
+                     struct sp_sampler *sampler,
+                     struct pipe_texture *texture,
+                     unsigned processor )
+{
+   struct softpipe_texture *sp_texture = softpipe_texture(texture);
+   struct sp_sampler_varient *v = NULL;
+   union sp_sampler_key key;
+
+   /* if this fails, widen the key.unit field and update this assertion */
+   assert(PIPE_MAX_SAMPLERS <= 16);
+
+   key.bits.target = sp_texture->base.target;
+   key.bits.is_pot = sp_texture->pot;
+   key.bits.processor = processor;
+   key.bits.unit = unit;
+   key.bits.pad = 0;
+
+   if (sampler->current && 
+       key.value == sampler->current->key.value) {
+      v = sampler->current;
+   }
+
+   if (v == NULL) {
+      for (v = sampler->varients; v; v = v->next)
+         if (v->key.value == key.value)
+            break;
+
+      if (v == NULL) {
+         v = sp_create_sampler_varient( &sampler->base, key );
+         v->next = sampler->varients;
+         sampler->varients = v;
+      }
+   }
+   
+   sampler->current = v;
+   return v;
+}
+
+
+
+
+void
+softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
+{
+   int i;
+
+   /* It's a bit hard to build these samplers ahead of time -- don't
+    * really know which samplers are going to be used for vertex and
+    * fragment programs.
+    */
+   for (i = 0; i <= softpipe->vs->max_sampler; i++) {
+      if (softpipe->sampler[i]) {
+         softpipe->tgsi.vert_samplers_list[i] = 
+            get_sampler_varient( i,
+                                 sp_sampler(softpipe->sampler[i]),
+                                 softpipe->texture[i],
+                                 TGSI_PROCESSOR_VERTEX );
+
+         sp_sampler_varient_bind_texture( softpipe->tgsi.vert_samplers_list[i], 
+                                          softpipe->tex_cache[i],
+                                          softpipe->texture[i] );
+      }
+   }
+
+   for (i = 0; i <= softpipe->fs->info.file_max[TGSI_FILE_SAMPLER]; i++) {
+      if (softpipe->sampler[i]) {
+         softpipe->tgsi.frag_samplers_list[i] =
+            get_sampler_varient( i,
+                                 sp_sampler(softpipe->sampler[i]),
+                                 softpipe->texture[i],
+                                 TGSI_PROCESSOR_FRAGMENT );
+
+         sp_sampler_varient_bind_texture( softpipe->tgsi.frag_samplers_list[i], 
+                                          softpipe->tex_cache[i],
+                                          softpipe->texture[i] );
+      }
+   }
+}
+
+
+
 void
 softpipe_delete_sampler_state(struct pipe_context *pipe,
                               void *sampler)
 {
+   struct sp_sampler *sp_sampler = (struct sp_sampler *)sampler;
+   struct sp_sampler_varient *v, *tmp;
+
+   for (v = sp_sampler->varients; v; v = tmp) {
+      tmp = v->next;
+      sp_sampler_varient_destroy(v);
+   }
+
    FREE( sampler );
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_state_surface.c b/src/gallium/drivers/softpipe/sp_state_surface.c
index 7c06d864a7..bc0e201130 100644
--- a/src/gallium/drivers/softpipe/sp_state_surface.c
+++ b/src/gallium/drivers/softpipe/sp_state_surface.c
@@ -53,10 +53,10 @@ softpipe_set_framebuffer_state(struct pipe_context *pipe,
       /* check if changing cbuf */
       if (sp->framebuffer.cbufs[i] != fb->cbufs[i]) {
          /* flush old */
-         sp_flush_tile_cache(sp, sp->cbuf_cache[i]);
+         sp_flush_tile_cache(sp->cbuf_cache[i]);
 
          /* assign new */
-         sp->framebuffer.cbufs[i] = fb->cbufs[i];
+         pipe_surface_reference(&sp->framebuffer.cbufs[i], fb->cbufs[i]);
 
          /* update cache */
          sp_tile_cache_set_surface(sp->cbuf_cache[i], fb->cbufs[i]);
@@ -68,58 +68,28 @@ softpipe_set_framebuffer_state(struct pipe_context *pipe,
    /* zbuf changing? */
    if (sp->framebuffer.zsbuf != fb->zsbuf) {
       /* flush old */
-      sp_flush_tile_cache(sp, sp->zsbuf_cache);
+      sp_flush_tile_cache(sp->zsbuf_cache);
 
       /* assign new */
-      sp->framebuffer.zsbuf = fb->zsbuf;
+      pipe_surface_reference(&sp->framebuffer.zsbuf, fb->zsbuf);
 
       /* update cache */
       sp_tile_cache_set_surface(sp->zsbuf_cache, fb->zsbuf);
-   }
-
-#if 0
-   /* XXX combined depth/stencil here */
-
-   /* sbuf changing? */
-   if (sp->framebuffer.sbuf != fb->sbuf) {
-      /* flush old */
-      sp_flush_tile_cache(sp, sp->sbuf_cache_sep);
-
-      /* assign new */
-      sp->framebuffer.sbuf = fb->sbuf;
-
-      /* update cache */
-      if (fb->sbuf != fb->zbuf) {
-         /* separate stencil buf */
-         sp->sbuf_cache = sp->sbuf_cache_sep;
-         sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf);
-      }
-      else {
-         /* combined depth/stencil */
-         sp->sbuf_cache = sp->zbuf_cache;
-         sp_tile_cache_set_surface(sp->sbuf_cache, fb->sbuf);
-      }
-   }
-#endif
 
-   /* Tell draw module how deep the Z/depth buffer is */
-   {
-      int depth_bits;
-      double mrd;
+      /* Tell draw module how deep the Z/depth buffer is */
       if (sp->framebuffer.zsbuf) {
+         int depth_bits;
+         double mrd;
          depth_bits = pf_get_component_bits(sp->framebuffer.zsbuf->format,
                                             PIPE_FORMAT_COMP_Z);
+         if (depth_bits > 16) {
+            mrd = 0.0000001;
+         }
+         else {
+            mrd = 0.00002;
+         }
+         draw_set_mrd(sp->draw, mrd);
       }
-      else {
-         depth_bits = 0;
-      }
-      if (depth_bits > 16) {
-         mrd = 0.0000001;
-      }
-      else {
-         mrd = 0.00002;
-      }
-      draw_set_mrd(sp->draw, mrd);
    }
 
    sp->framebuffer.width = fb->width;
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index f99a30277d..c22ee86b66 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -31,29 +31,33 @@
  *
  * Authors:
  *   Brian Paul
+ *   Keith Whitwell
  */
 
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_texture.h"
-#include "sp_tex_sample.h"
-#include "sp_tile_cache.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "sp_quad.h"   /* only for #define QUAD_* tokens */
+#include "sp_tex_sample.h"
+#include "sp_tex_tile_cache.h"
 
 
 
 /*
- * Note, the FRAC macro has to work perfectly.  Otherwise you'll sometimes
- * see 1-pixel bands of improperly weighted linear-filtered textures.
+ * Return fractional part of 'f'.  Used for computing interpolation weights.
+ * Need to be careful with negative values.
+ * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
+ * of improperly weighted linear-filtered textures.
  * The tests/texwrap.c demo is a good test.
- * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0.
- * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x).
  */
-#define FRAC(f)  ((f) - util_ifloor(f))
+static INLINE float
+frac(float f)
+{
+   return f - util_ifloor(f);
+}
+
 
 
 /**
@@ -100,10 +104,16 @@ lerp_3d(float a, float b, float c,
 
 
 /**
- * If A is a signed integer, A % B doesn't give the right value for A < 0
- * (in terms of texture repeat).  Just casting to unsigned fixes that.
+ * Compute coord % size for repeat wrap modes.
+ * Note that if coord is a signed integer, coord % size doesn't give
+ * the right value for coord < 0 (in terms of texture repeat).  Just
+ * casting to unsigned fixes that.
  */
-#define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B))
+static INLINE int
+repeat(int coord, unsigned size)
+{
+   return (int) ((unsigned) coord % size);
+}
 
 
 /**
@@ -115,133 +125,153 @@ lerp_3d(float a, float b, float c,
  * \param icoord  returns the integer texcoords
  * \return  integer texture index
  */
-static INLINE void
-nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
-                   int icoord[4])
+static void
+wrap_nearest_repeat(const float s[4], unsigned size, int icoord[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_REPEAT:
-      /* s limited to [0,1) */
-      /* i limited to [0,size-1] */
-      for (ch = 0; ch < 4; ch++) {
-         int i = util_ifloor(s[ch] * size);
-         icoord[ch] = REMAINDER(i, size);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP:
+   /* s limited to [0,1) */
+   /* i limited to [0,size-1] */
+   for (ch = 0; ch < 4; ch++) {
+      int i = util_ifloor(s[ch] * size);
+      icoord[ch] = repeat(i, size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp(const float s[4], unsigned size, int icoord[4])
+{
+   uint ch;
+   /* s limited to [0,1] */
+   /* i limited to [0,size-1] */
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] <= 0.0F)
+         icoord[ch] = 0;
+      else if (s[ch] >= 1.0F)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp_to_edge(const float s[4], unsigned size, int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] < min)
+         icoord[ch] = 0;
+      else if (s[ch] > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp_to_border(const float s[4], unsigned size, int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [-1, size] */
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] <= min)
+         icoord[ch] = -1;
+      else if (s[ch] >= max)
+         icoord[ch] = size;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+
+static void
+wrap_nearest_mirror_repeat(const float s[4], unsigned size, int icoord[4])
+{
+   uint ch;
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const int flr = util_ifloor(s[ch]);
+      float u;
+      if (flr & 1)
+         u = 1.0F - (s[ch] - (float) flr);
+      else
+         u = s[ch] - (float) flr;
+      if (u < min)
+         icoord[ch] = 0;
+      else if (u > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+
+static void
+wrap_nearest_mirror_clamp(const float s[4], unsigned size, int icoord[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
       /* s limited to [0,1] */
       /* i limited to [0,size-1] */
-      for (ch = 0; ch < 4; ch++) {
-         if (s[ch] <= 0.0F)
-            icoord[ch] = 0;
-         else if (s[ch] >= 1.0F)
-            icoord[ch] = size - 1;
-         else
-            icoord[ch] = util_ifloor(s[ch] * size);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            if (s[ch] < min)
-               icoord[ch] = 0;
-            else if (s[ch] > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(s[ch] * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [-1, size] */
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            if (s[ch] <= min)
-               icoord[ch] = -1;
-            else if (s[ch] >= max)
-               icoord[ch] = size;
-            else
-               icoord[ch] = util_ifloor(s[ch] * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_REPEAT:
-      {
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const int flr = util_ifloor(s[ch]);
-            float u;
-            if (flr & 1)
-               u = 1.0F - (s[ch] - (float) flr);
-            else
-               u = s[ch] - (float) flr;
-            if (u < min)
-               icoord[ch] = 0;
-            else if (u > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         /* s limited to [0,1] */
-         /* i limited to [0,size-1] */
-         const float u = fabsf(s[ch]);
-         if (u <= 0.0F)
-            icoord[ch] = 0;
-         else if (u >= 1.0F)
-            icoord[ch] = size - 1;
-         else
-            icoord[ch] = util_ifloor(u * size);
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const float u = fabsf(s[ch]);
-            if (u < min)
-               icoord[ch] = 0;
-            else if (u > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const float u = fabsf(s[ch]);
-            if (u < min)
-               icoord[ch] = -1;
-            else if (u > max)
-               icoord[ch] = size;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   default:
-      assert(0);
+      const float u = fabsf(s[ch]);
+      if (u <= 0.0F)
+         icoord[ch] = 0;
+      else if (u >= 1.0F)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+
+static void
+wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size,
+                                  int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const float u = fabsf(s[ch]);
+      if (u < min)
+         icoord[ch] = 0;
+      else if (u > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+
+static void
+wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size,
+                                    int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const float u = fabsf(s[ch]);
+      if (u < min)
+         icoord[ch] = -1;
+      else if (u > max)
+         icoord[ch] = size;
+      else
+         icoord[ch] = util_ifloor(u * size);
    }
 }
 
@@ -256,125 +286,156 @@ nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
  * \param w  returns blend factor/weight between texture indexes
  * \param icoord  returns the computed integer texture coords
  */
-static INLINE void
-linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
+static void
+wrap_linear_repeat(const float s[4], unsigned size,
+                   int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = s[ch] * size - 0.5F;
+      icoord0[ch] = repeat(util_ifloor(u), size);
+      icoord1[ch] = repeat(icoord0[ch] + 1, size);
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_clamp(const float s[4], unsigned size,
                   int icoord0[4], int icoord1[4], float w[4])
 {
    uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.0F, 1.0F);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = frac(u);
+   }
+}
 
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_REPEAT:
-      for (ch = 0; ch < 4; ch++) {
-         float u = s[ch] * size - 0.5F;
-         icoord0[ch] = REMAINDER(util_ifloor(u), size);
-         icoord1[ch] = REMAINDER(icoord0[ch] + 1, size);
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.0F, 1.0F);
-         u = u * size - 0.5f;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.0F, 1.0F);
-         u = u * size - 0.5f;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      {
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            float u = CLAMP(s[ch], min, max);
-            u = u * size - 0.5f;
-            icoord0[ch] = util_ifloor(u);
-            icoord1[ch] = icoord0[ch] + 1;
-            w[ch] = FRAC(u);
-         }
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_REPEAT:
-      for (ch = 0; ch < 4; ch++) {
-         const int flr = util_ifloor(s[ch]);
-         float u;
-         if (flr & 1)
-            u = 1.0F - (s[ch] - (float) flr);
-         else
-            u = s[ch] - (float) flr;
-         u = u * size - 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         float u = fabsf(s[ch]);
-         if (u >= 1.0F)
-            u = (float) size;
-         else
-            u *= size;
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-      for (ch = 0; ch < 4; ch++) {
-         float u = fabsf(s[ch]);
-         if (u >= 1.0F)
-            u = (float) size;
-         else
-            u *= size;
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-      {
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            float u = fabsf(s[ch]);
-            if (u <= min)
-               u = min * size;
-            else if (u >= max)
-               u = max * size;
-            else
-               u *= size;
-            u -= 0.5F;
-            icoord0[ch] = util_ifloor(u);
-            icoord1[ch] = icoord0[ch] + 1;
-            w[ch] = FRAC(u);
-         }
-      }
-      break;;
-   default:
-      assert(0);
+
+static void
+wrap_linear_clamp_to_edge(const float s[4], unsigned size,
+                          int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.0F, 1.0F);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_clamp_to_border(const float s[4], unsigned size,
+                            int icoord0[4], int icoord1[4], float w[4])
+{
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], min, max);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_mirror_repeat(const float s[4], unsigned size,
+                          int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      const int flr = util_ifloor(s[ch]);
+      float u;
+      if (flr & 1)
+         u = 1.0F - (s[ch] - (float) flr);
+      else
+         u = s[ch] - (float) flr;
+      u = u * size - 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_mirror_clamp(const float s[4], unsigned size,
+                         int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size,
+                                 int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+static void
+wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
+                                   int icoord0[4], int icoord1[4], float w[4])
+{
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u <= min)
+         u = min * size;
+      else if (u >= max)
+         u = max * size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = frac(u);
    }
 }
 
@@ -383,27 +444,27 @@ linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
  * For RECT textures / unnormalized texcoords
  * Only a subset of wrap modes supported.
  */
-static INLINE void
-nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
-                          int icoord[4])
+static void
+wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         int i = util_ifloor(s[ch]);
-         icoord[ch]= CLAMP(i, 0, (int) size-1);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      /* fall-through */
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      for (ch = 0; ch < 4; ch++) {
-         icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
-      }
-      return;
-   default:
-      assert(0);
+   for (ch = 0; ch < 4; ch++) {
+      int i = util_ifloor(s[ch]);
+      icoord[ch]= CLAMP(i, 0, (int) size-1);
+   }
+}
+
+
+/**
+ * Handles clamp_to_edge and clamp_to_border:
+ */
+static void
+wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
+                                   int icoord[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
    }
 }
 
@@ -412,358 +473,971 @@ nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
  * For RECT textures / unnormalized texcoords.
  * Only a subset of wrap modes supported.
  */
-static INLINE void
-linear_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
-                         int icoord0[4], int icoord1[4], float w[4])
+static void
+wrap_linear_unorm_clamp(const float s[4], unsigned size,
+                        int icoord0[4], int icoord1[4], float w[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         /* Not exactly what the spec says, but it matches NVIDIA output */
-         float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      /* fall-through */
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord1[ch] > (int) size - 1)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;
-   default:
-      assert(0);
+   for (ch = 0; ch < 4; ch++) {
+      /* Not exactly what the spec says, but it matches NVIDIA output */
+      float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = frac(u);
    }
 }
 
 
-static unsigned
-choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
+static void
+wrap_linear_unorm_clamp_to_border(const float s[4], unsigned size,
+                                  int icoord0[4], int icoord1[4], float w[4])
 {
-   /*
-      major axis
-      direction     target                             sc     tc    ma
-      ----------    -------------------------------    ---    ---   ---
-       +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
-       -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
-       +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
-       -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
-       +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
-       -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
-   */
-   const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
-   unsigned face;
-   float sc, tc, ma;
-
-   if (arx >= ary && arx >= arz) {
-      if (rx >= 0.0F) {
-         face = PIPE_TEX_FACE_POS_X;
-         sc = -rz;
-         tc = -ry;
-         ma = arx;
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord1[ch] > (int) size - 1)
+         icoord1[ch] = size - 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+
+/**
+ * Examine the quad's texture coordinates to compute the partial
+ * derivatives w.r.t X and Y, then compute lambda (level of detail).
+ */
+static float
+compute_lambda_1d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
+{
+   const struct pipe_texture *texture = samp->texture;
+   const struct pipe_sampler_state *sampler = samp->sampler;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float rho = MAX2(dsdx, dsdy) * texture->width[0];
+   float lambda;
+
+   lambda = util_fast_log2(rho);
+   lambda += lodbias + sampler->lod_bias;
+   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
+
+   return lambda;
+}
+
+
+static float
+compute_lambda_2d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
+{
+   const struct pipe_texture *texture = samp->texture;
+   const struct pipe_sampler_state *sampler = samp->sampler;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+   float dtdy = fabsf(t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT]);
+   float maxx = MAX2(dsdx, dsdy) * texture->width[0];
+   float maxy = MAX2(dtdx, dtdy) * texture->height[0];
+   float rho  = MAX2(maxx, maxy);
+   float lambda;
+
+   lambda = util_fast_log2(rho);
+   lambda += lodbias + sampler->lod_bias;
+   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
+
+   return lambda;
+}
+
+
+static float
+compute_lambda_3d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
+{
+   const struct pipe_texture *texture = samp->texture;
+   const struct pipe_sampler_state *sampler = samp->sampler;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+   float dtdy = fabsf(t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT]);
+   float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
+   float dpdy = fabsf(p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT]);
+   float maxx = MAX2(dsdx, dsdy) * texture->width[0];
+   float maxy = MAX2(dtdx, dtdy) * texture->height[0];
+   float maxz = MAX2(dpdx, dpdy) * texture->depth[0];
+   float rho, lambda;
+
+   rho = MAX2(maxx, maxy);
+   rho = MAX2(rho, maxz);
+
+   lambda = util_fast_log2(rho);
+   lambda += lodbias + sampler->lod_bias;
+   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
+
+   return lambda;
+}
+
+
+/**
+ * Compute lambda for a vertex texture sampler.
+ * Since there aren't derivatives to use, just return the LOD bias.
+ */
+static float
+compute_lambda_vert(const struct sp_sampler_varient *samp,
+                    const float s[QUAD_SIZE],
+                    const float t[QUAD_SIZE],
+                    const float p[QUAD_SIZE],
+                    float lodbias)
+{
+   return lodbias;
+}
+
+
+
+/**
+ * Get a texel from a texture, using the texture tile cache.
+ *
+ * \param addr  the template tex address containing cube, z, face info.
+ * \param x  the x coord of texel within 2D image
+ * \param y  the y coord of texel within 2D image
+ * \param rgba  the quad to put the texel/color into
+ *
+ * XXX maybe move this into sp_tex_tile_cache.c and merge with the
+ * sp_get_cached_tile_tex() function.  Also, get 4 texels instead of 1...
+ */
+
+
+
+
+static INLINE const float *
+get_texel_2d_no_border(const struct sp_sampler_varient *samp,
+		       union tex_tile_address addr, int x, int y)
+{
+   const struct softpipe_tex_cached_tile *tile;
+
+   addr.bits.x = x / TILE_SIZE;
+   addr.bits.y = y / TILE_SIZE;
+   y %= TILE_SIZE;
+   x %= TILE_SIZE;
+
+   tile = sp_get_cached_tile_tex(samp->cache, addr);
+
+   return &tile->data.color[y][x][0];
+}
+
+
+static INLINE const float *
+get_texel_2d(const struct sp_sampler_varient *samp,
+	     union tex_tile_address addr, int x, int y)
+{
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level = addr.bits.level;
+
+   if (x < 0 || x >= (int) texture->width[level] ||
+       y < 0 || y >= (int) texture->height[level]) {
+      return samp->sampler->border_color;
+   }
+   else {
+      return get_texel_2d_no_border( samp, addr, x, y );
+   }
+}
+
+
+/* Gather a quad of adjacent texels within a tile:
+ */
+static INLINE void
+get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
+					union tex_tile_address addr, 
+					unsigned x, unsigned y, 
+					const float *out[4])
+{
+   const struct softpipe_tex_cached_tile *tile;
+
+   addr.bits.x = x / TILE_SIZE;
+   addr.bits.y = y / TILE_SIZE;
+   y %= TILE_SIZE;
+   x %= TILE_SIZE;
+
+   tile = sp_get_cached_tile_tex(samp->cache, addr);
+      
+   out[0] = &tile->data.color[y  ][x  ][0];
+   out[1] = &tile->data.color[y  ][x+1][0];
+   out[2] = &tile->data.color[y+1][x  ][0];
+   out[3] = &tile->data.color[y+1][x+1][0];
+}
+
+
+/* Gather a quad of potentially non-adjacent texels:
+ */
+static INLINE void
+get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
+			    union tex_tile_address addr,
+			    int x0, int y0, 
+			    int x1, int y1,
+			    const float *out[4])
+{
+   out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
+   out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
+   out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
+   out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
+}
+
+/* Can involve a lot of unnecessary checks for border color:
+ */
+static INLINE void
+get_texel_quad_2d(const struct sp_sampler_varient *samp,
+		  union tex_tile_address addr,
+		  int x0, int y0, 
+		  int x1, int y1,
+		  const float *out[4])
+{
+   out[0] = get_texel_2d( samp, addr, x0, y0 );
+   out[1] = get_texel_2d( samp, addr, x1, y0 );
+   out[3] = get_texel_2d( samp, addr, x1, y1 );
+   out[2] = get_texel_2d( samp, addr, x0, y1 );
+}
+
+
+
+/* 3d varients:
+ */
+static INLINE const float *
+get_texel_3d_no_border(const struct sp_sampler_varient *samp,
+                       union tex_tile_address addr, int x, int y, int z)
+{
+   const struct softpipe_tex_cached_tile *tile;
+
+   addr.bits.x = x / TILE_SIZE;
+   addr.bits.y = y / TILE_SIZE;
+   addr.bits.z = z;
+   y %= TILE_SIZE;
+   x %= TILE_SIZE;
+
+   tile = sp_get_cached_tile_tex(samp->cache, addr);
+
+   return &tile->data.color[y][x][0];
+}
+
+
+static INLINE const float *
+get_texel_3d(const struct sp_sampler_varient *samp,
+	     union tex_tile_address addr, int x, int y, int z)
+{
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level = addr.bits.level;
+
+   if (x < 0 || x >= (int) texture->width[level] ||
+       y < 0 || y >= (int) texture->height[level] ||
+       z < 0 || z >= (int) texture->depth[level]) {
+      return samp->sampler->border_color;
+   }
+   else {
+      return get_texel_3d_no_border( samp, addr, x, y, z );
+   }
+}
+
+
+/**
+ * Given the logbase2 of a mipmap's base level size and a mipmap level,
+ * return the size (in texels) of that mipmap level.
+ * For example, if level[0].width = 256 then base_pot will be 8.
+ * If level = 2, then we'll return 64 (the width at level=2).
+ * Return 1 if level > base_pot.
+ */
+static INLINE unsigned
+pot_level_size(unsigned base_pot, unsigned level)
+{
+   return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
+}
+
+
+/* Some image-filter fastpaths:
+ */
+static INLINE void
+img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
+                                const float s[QUAD_SIZE],
+                                const float t[QUAD_SIZE],
+                                const float p[QUAD_SIZE],
+                                float lodbias,
+                                float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   unsigned  j;
+   unsigned level = samp->level;
+   unsigned xpot = pot_level_size(samp->xpot, level);
+   unsigned ypot = pot_level_size(samp->ypot, level);
+   unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
+   unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
+   union tex_tile_address addr;
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int c;
+
+      float u = s[j] * xpot - 0.5F;
+      float v = t[j] * ypot - 0.5F;
+
+      int uflr = util_ifloor(u);
+      int vflr = util_ifloor(v);
+
+      float xw = u - (float)uflr;
+      float yw = v - (float)vflr;
+
+      int x0 = uflr & (xpot - 1);
+      int y0 = vflr & (ypot - 1);
+
+      const float *tx[4];      
+      
+      /* Can we fetch all four at once:
+       */
+      if (x0 < xmax && y0 < ymax) {
+         get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
       }
       else {
-         face = PIPE_TEX_FACE_NEG_X;
-         sc = rz;
-         tc = -ry;
-         ma = arx;
+         unsigned x1 = (x0 + 1) & (xpot - 1);
+         unsigned y1 = (y0 + 1) & (ypot - 1);
+         get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
+      }
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_2d(xw, yw, 
+                              tx[0][c], tx[1][c], 
+                              tx[2][c], tx[3][c]);
       }
    }
-   else if (ary >= arx && ary >= arz) {
-      if (ry >= 0.0F) {
-         face = PIPE_TEX_FACE_POS_Y;
-         sc = rx;
-         tc = rz;
-         ma = ary;
+}
+
+
+static INLINE void
+img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
+                                 const float s[QUAD_SIZE],
+                                 const float t[QUAD_SIZE],
+                                 const float p[QUAD_SIZE],
+                                 float lodbias,
+                                 float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   unsigned  j;
+   unsigned level = samp->level;
+   unsigned xpot = pot_level_size(samp->xpot, level);
+   unsigned ypot = pot_level_size(samp->ypot, level);
+   union tex_tile_address addr;
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int c;
+
+      float u = s[j] * xpot;
+      float v = t[j] * ypot;
+
+      int uflr = util_ifloor(u);
+      int vflr = util_ifloor(v);
+
+      int x0 = uflr & (xpot - 1);
+      int y0 = vflr & (ypot - 1);
+
+      const float *out = get_texel_2d_no_border(samp, addr, x0, y0);
+
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
       }
-      else {
-         face = PIPE_TEX_FACE_NEG_Y;
-         sc = rx;
-         tc = -rz;
-         ma = ary;
+   }
+}
+
+
+static INLINE void
+img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
+                                const float s[QUAD_SIZE],
+                                const float t[QUAD_SIZE],
+                                const float p[QUAD_SIZE],
+                                float lodbias,
+                                float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   unsigned  j;
+   unsigned level = samp->level;
+   unsigned xpot = pot_level_size(samp->xpot, level);
+   unsigned ypot = pot_level_size(samp->ypot, level);
+   union tex_tile_address addr;
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int c;
+
+      float u = s[j] * xpot;
+      float v = t[j] * ypot;
+
+      int x0, y0;
+      const float *out;
+
+      x0 = util_ifloor(u);
+      if (x0 < 0) 
+         x0 = 0;
+      else if (x0 > xpot - 1)
+         x0 = xpot - 1;
+
+      y0 = util_ifloor(v);
+      if (y0 < 0) 
+         y0 = 0;
+      else if (y0 > ypot - 1)
+         y0 = ypot - 1;
+      
+      out = get_texel_2d_no_border(samp, addr, x0, y0);
+
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
       }
    }
-   else {
-      if (rz > 0.0F) {
-         face = PIPE_TEX_FACE_POS_Z;
-         sc = rx;
-         tc = -ry;
-         ma = arz;
+}
+
+
+static void
+img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
+                        const float s[QUAD_SIZE],
+                        const float t[QUAD_SIZE],
+                        const float p[QUAD_SIZE],
+                        float lodbias,
+                        float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width;
+   int x[4];
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+
+   assert(width > 0);
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->nearest_texcoord_s(s, width, x);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *out = get_texel_2d(samp, addr, x[j], 0);
+      int c;
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
       }
-      else {
-         face = PIPE_TEX_FACE_NEG_Z;
-         sc = -rx;
-         tc = -ry;
-         ma = arz;
+   }
+}
+
+
+static void
+img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
+                      const float s[QUAD_SIZE],
+                      const float t[QUAD_SIZE],
+                      const float p[QUAD_SIZE],
+                      float lodbias,
+                      float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width, height;
+   int x[4], y[4];
+   union tex_tile_address addr;
+
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+
+   assert(width > 0);
+   assert(height > 0);
+ 
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->nearest_texcoord_s(s, width, x);
+   samp->nearest_texcoord_t(t, height, y);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *out = get_texel_2d(samp, addr, x[j], y[j]);
+      int c;
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
       }
    }
+}
 
-   *newS = ( sc / ma + 1.0F ) * 0.5F;
-   *newT = ( tc / ma + 1.0F ) * 0.5F;
 
-   return face;
+static INLINE union tex_tile_address
+face(union tex_tile_address addr, unsigned face )
+{
+   addr.bits.face = face;
+   return addr;
 }
 
 
-/**
- * Examine the quad's texture coordinates to compute the partial
- * derivatives w.r.t X and Y, then compute lambda (level of detail).
- *
- * This is only done for fragment shaders, not vertex shaders.
- */
-static float
-compute_lambda(const struct pipe_texture *tex,
-               const struct pipe_sampler_state *sampler,
-               const float s[QUAD_SIZE],
-               const float t[QUAD_SIZE],
-               const float p[QUAD_SIZE],
-               float lodbias)
+static void
+img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
+                        const float s[QUAD_SIZE],
+                        const float t[QUAD_SIZE],
+                        const float p[QUAD_SIZE],
+                        float lodbias,
+                        float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   float rho, lambda;
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   const unsigned *faces = samp->faces; /* zero when not cube-mapping */
+   unsigned level0, j;
+   int width, height;
+   int x[4], y[4];
+   union tex_tile_address addr;
 
-   assert(sampler->normalized_coords);
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
 
-   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) * tex->width[0];
-   }
-   if (t) {
-      float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
-      float dtdy = t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT];
-      float max;
-      dtdx = fabsf(dtdx);
-      dtdy = fabsf(dtdy);
-      max = MAX2(dtdx, dtdy) * tex->height[0];
-      rho = MAX2(rho, max);
+   assert(width > 0);
+   assert(height > 0);
+ 
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->nearest_texcoord_s(s, width, x);
+   samp->nearest_texcoord_t(t, height, y);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]);
+      int c;
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
+      }
    }
-   if (p) {
-      float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT];
-      float dpdy = p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT];
-      float max;
-      dpdx = fabsf(dpdx);
-      dpdy = fabsf(dpdy);
-      max = MAX2(dpdx, dpdy) * tex->depth[0];
-      rho = MAX2(rho, max);
+}
+
+
+static void
+img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
+                      const float s[QUAD_SIZE],
+                      const float t[QUAD_SIZE],
+                      const float p[QUAD_SIZE],
+                      float lodbias,
+                      float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width, height, depth;
+   int x[4], y[4], z[4];
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+   depth = texture->depth[level0];
+
+   assert(width > 0);
+   assert(height > 0);
+   assert(depth > 0);
+
+   samp->nearest_texcoord_s(s, width,  x);
+   samp->nearest_texcoord_t(t, height, y);
+   samp->nearest_texcoord_p(p, depth,  z);
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]);
+      int c;
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = out[c];
+      }      
    }
+}
 
-   lambda = util_fast_log2(rho);
-   lambda += lodbias + sampler->lod_bias;
-   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
 
-   return lambda;
+static void
+img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width;
+   int x0[4], x1[4];
+   float xw[4]; /* weights */
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+
+   assert(width > 0);
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->linear_texcoord_s(s, width, x0, x1, xw);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *tx0 = get_texel_2d(samp, addr, x0[j], 0);
+      const float *tx1 = get_texel_2d(samp, addr, x1[j], 0);
+      int c;
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
+      }
+   }
 }
 
 
-/**
- * Do several things here:
- * 1. Compute lambda from the texcoords, if needed
- * 2. Determine if we're minifying or magnifying
- * 3. If minifying, choose mipmap levels
- * 4. Return image filter to use within mipmap images
- * \param level0  Returns first mipmap level to sample from
- * \param level1  Returns second mipmap level to sample from
- * \param levelBlend  Returns blend factor between levels, in [0,1]
- * \param imgFilter  Returns either the min or mag filter, depending on lambda
- */
 static void
-choose_mipmap_levels(const struct pipe_texture *texture,
-                     const struct pipe_sampler_state *sampler,
+img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
                      const float s[QUAD_SIZE],
                      const float t[QUAD_SIZE],
                      const float p[QUAD_SIZE],
-                     boolean computeLambda,
                      float lodbias,
-                     unsigned *level0, unsigned *level1, float *levelBlend,
-                     unsigned *imgFilter)
-{
-   if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
-      /* no mipmap selection needed */
-      *level0 = *level1 = CLAMP((int) sampler->min_lod,
-                                0, (int) texture->last_level);
-
-      if (sampler->min_img_filter != sampler->mag_img_filter) {
-         /* non-mipmapped texture, but still need to determine if doing
-          * minification or magnification.
-          */
-         float lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
-         if (lambda <= 0.0) {
-            *imgFilter = sampler->mag_img_filter;
-         }
-         else {
-            *imgFilter = sampler->min_img_filter;
-         }
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width, height;
+   int x0[4], y0[4], x1[4], y1[4];
+   float xw[4], yw[4]; /* weights */
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+
+   assert(width > 0);
+   assert(height > 0);
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->linear_texcoord_s(s, width,  x0, x1, xw);
+   samp->linear_texcoord_t(t, height, y0, y1, yw);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]);
+      const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]);
+      const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]);
+      const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]);
+      int c;
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_2d(xw[j], yw[j],
+                              tx0[c], tx1[c],
+                              tx2[c], tx3[c]);
       }
-      else {
-         *imgFilter = sampler->mag_img_filter;
+   }
+}
+
+
+static void
+img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
+                       const float s[QUAD_SIZE],
+                       const float t[QUAD_SIZE],
+                       const float p[QUAD_SIZE],
+                       float lodbias,
+                       float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   const unsigned *faces = samp->faces; /* zero when not cube-mapping */
+   unsigned level0, j;
+   int width, height;
+   int x0[4], y0[4], x1[4], y1[4];
+   float xw[4], yw[4]; /* weights */
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+
+   assert(width > 0);
+   assert(height > 0);
+
+   addr.value = 0;
+   addr.bits.level = samp->level;
+
+   samp->linear_texcoord_s(s, width,  x0, x1, xw);
+   samp->linear_texcoord_t(t, height, y0, y1, yw);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      union tex_tile_address addrj = face(addr, faces[j]);
+      const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]);
+      const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]);
+      const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]);
+      const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]);
+      int c;
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_2d(xw[j], yw[j],
+                              tx0[c], tx1[c],
+                              tx2[c], tx3[c]);
       }
    }
-   else {
-      float lambda;
+}
 
-      if (computeLambda)
-         /* fragment shader */
-         lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
-      else
-         /* vertex shader */
-         lambda = lodbias; /* not really a bias, but absolute LOD */
 
-      if (lambda <= 0.0) { /* XXX threshold depends on the filter */
-         /* magnifying */
-         *imgFilter = sampler->mag_img_filter;
-         *level0 = *level1 = 0;
+static void
+img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width, height, depth;
+   int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
+   float xw[4], yw[4], zw[4]; /* interpolation weights */
+   union tex_tile_address addr;
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+   depth = texture->depth[level0];
+
+   addr.value = 0;
+   addr.bits.level = level0;
+
+   assert(width > 0);
+   assert(height > 0);
+   assert(depth > 0);
+
+   samp->linear_texcoord_s(s, width,  x0, x1, xw);
+   samp->linear_texcoord_t(t, height, y0, y1, yw);
+   samp->linear_texcoord_p(p, depth,  z0, z1, zw);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      int c;
+
+      const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]);
+      const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]);
+      const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]);
+      const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]);
+      
+      const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]);
+      const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]);
+      const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]);
+      const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]);
+      
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
+                              tx00[c], tx01[c],
+                              tx02[c], tx03[c],
+                              tx10[c], tx11[c],
+                              tx12[c], tx13[c]);
       }
-      else {
-         /* minifying */
-         *imgFilter = sampler->min_img_filter;
-
-         /* choose mipmap level(s) and compute the blend factor between them */
-         if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
-            /* Nearest mipmap level */
-            const int lvl = (int) (lambda + 0.5);
-            *level0 =
-            *level1 = CLAMP(lvl, 0, (int) texture->last_level);
-         }
-         else {
-            /* Linear interpolation between mipmap levels */
-            const int lvl = (int) lambda;
-            *level0 = CLAMP(lvl,     0, (int) texture->last_level);
-            *level1 = CLAMP(lvl + 1, 0, (int) texture->last_level);
-            *levelBlend = FRAC(lambda);  /* blending weight between levels */
+   }
+}
+
+
+static void
+mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias,
+                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   int level0;
+   float lambda;
+
+   lambda = samp->compute_lambda(samp, s, t, p, lodbias);
+   level0 = (int)lambda;
+
+   if (lambda < 0.0) { 
+      samp->level = 0;
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else if (level0 >= texture->last_level) {
+      samp->level = texture->last_level;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      float levelBlend = lambda - level0;
+      float rgba0[4][4];
+      float rgba1[4][4];
+      int c,j;
+
+      samp->level = level0;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba0 );
+
+      samp->level = level0+1;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba1 );
+
+      for (j = 0; j < QUAD_SIZE; j++) {
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
          }
       }
    }
 }
 
 
-/**
- * Get a texel from a texture, using the texture tile cache.
- *
- * \param face  the cube face in 0..5
- * \param level  the mipmap level
- * \param x  the x coord of texel within 2D image
- * \param y  the y coord of texel within 2D image
- * \param z  which slice of a 3D texture
- * \param rgba  the quad to put the texel/color into
- * \param j  which element of the rgba quad to write to
- *
- * XXX maybe move this into sp_tile_cache.c and merge with the
- * sp_get_cached_tile_tex() function.  Also, get 4 texels instead of 1...
- */
 static void
-get_texel(const struct tgsi_sampler *tgsi_sampler,
-          unsigned face, unsigned level, int x, int y, int z,
-          float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j)
+mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
+                   const float s[QUAD_SIZE],
+                   const float t[QUAD_SIZE],
+                   const float p[QUAD_SIZE],
+                   float lodbias,
+                   float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   struct softpipe_context *sp = samp->sp;
-   const uint unit = samp->unit;
-   const struct pipe_texture *texture = sp->texture[unit];
-   const struct pipe_sampler_state *sampler = sp->sampler[unit];
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   float lambda;
 
-   if (x < 0 || x >= (int) texture->width[level] ||
-       y < 0 || y >= (int) texture->height[level] ||
-       z < 0 || z >= (int) texture->depth[level]) {
-      rgba[0][j] = sampler->border_color[0];
-      rgba[1][j] = sampler->border_color[1];
-      rgba[2][j] = sampler->border_color[2];
-      rgba[3][j] = sampler->border_color[3];
+   lambda = samp->compute_lambda(samp, s, t, p, lodbias);
+
+   if (lambda < 0.0) { 
+      samp->level = 0;
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
    }
    else {
-      const int tx = x % TILE_SIZE;
-      const int ty = y % TILE_SIZE;
-      const struct softpipe_cached_tile *tile
-         = sp_get_cached_tile_tex(sp, samp->cache,
-                                  x, y, z, face, level);
-      rgba[0][j] = tile->data.color[ty][tx][0];
-      rgba[1][j] = tile->data.color[ty][tx][1];
-      rgba[2][j] = tile->data.color[ty][tx][2];
-      rgba[3][j] = tile->data.color[ty][tx][3];
-      if (0)
-      {
-         debug_printf("Get texel %f %f %f %f from %s\n",
-                      rgba[0][j], rgba[1][j], rgba[2][j], rgba[3][j],
-                      pf_name(texture->format));
-      }
+      samp->level = (int)(lambda + 0.5) ;
+      samp->level = MIN2(samp->level, (int)texture->last_level);
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+
+#if 0
+   printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
+          rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
+          rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
+          rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
+          rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
+#endif
+}
+
+
+static void
+mip_filter_none(struct tgsi_sampler *tgsi_sampler,
+                const float s[QUAD_SIZE],
+                const float t[QUAD_SIZE],
+                const float p[QUAD_SIZE],
+                float lodbias,
+                float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   float lambda = samp->compute_lambda(samp, s, t, p, lodbias);
+
+   if (lambda < 0.0) { 
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
    }
 }
 
 
+
 /**
- * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
- * When we sampled the depth texture, the depth value was put into all
- * RGBA channels.  We look at the red channel here.
- * \param rgba  quad of (depth) texel values
- * \param p  texture 'P' components for four pixels in quad
- * \param j  which pixel in the quad to test [0..3]
+ * Specialized version of mip_filter_linear with hard-wired calls to
+ * 2d lambda calculation and 2d_linear_repeat_POT img filters.
  */
-static INLINE void
-shadow_compare(const struct pipe_sampler_state *sampler,
-               float rgba[NUM_CHANNELS][QUAD_SIZE],
-               const float p[QUAD_SIZE],
-               uint j)
+static void
+mip_filter_linear_2d_linear_repeat_POT(
+   struct tgsi_sampler *tgsi_sampler,
+   const float s[QUAD_SIZE],
+   const float t[QUAD_SIZE],
+   const float p[QUAD_SIZE],
+   float lodbias,
+   float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   int k;
-   switch (sampler->compare_func) {
-   case PIPE_FUNC_LESS:
-      k = p[j] < rgba[0][j];
-      break;
-   case PIPE_FUNC_LEQUAL:
-      k = p[j] <= rgba[0][j];
-      break;
-   case PIPE_FUNC_GREATER:
-      k = p[j] > rgba[0][j];
-      break;
-   case PIPE_FUNC_GEQUAL:
-      k = p[j] >= rgba[0][j];
-      break;
-   case PIPE_FUNC_EQUAL:
-      k = p[j] == rgba[0][j];
-      break;
-   case PIPE_FUNC_NOTEQUAL:
-      k = p[j] != rgba[0][j];
-      break;
-   case PIPE_FUNC_ALWAYS:
-      k = 1;
-      break;
-   case PIPE_FUNC_NEVER:
-      k = 0;
-      break;
-   default:
-      k = 0;
-      assert(0);
-      break;
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   int level0;
+   float lambda;
+
+   lambda = compute_lambda_2d(samp, s, t, p, lodbias);
+   level0 = (int)lambda;
+
+   /* Catches both negative and large values of level0:
+    */
+   if ((unsigned)level0 >= texture->last_level) { 
+      if (level0 < 0)
+         samp->level = 0;
+      else
+         samp->level = texture->last_level;
+
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba );
    }
+   else {
+      float levelBlend = lambda - level0;
+      float rgba0[4][4];
+      float rgba1[4][4];
+      int c,j;
 
-   /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
-   rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k;
-   rgba[3][j] = 1.0F;
+      samp->level = level0;
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba0 );
+
+      samp->level = level0+1;
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba1 );
+
+      for (j = 0; j < QUAD_SIZE; j++) {
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
+         }
+      }
+   }
 }
 
 
+
 /**
- * As above, but do four z/texture comparisons.
+ * Do shadow/depth comparisons.
  */
-static INLINE void
-shadow_compare4(const struct pipe_sampler_state *sampler,
-                float rgba[NUM_CHANNELS][QUAD_SIZE],
-                const float p[QUAD_SIZE])
+static void
+sample_compare(struct tgsi_sampler *tgsi_sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               float lodbias,
+               float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_sampler_state *sampler = samp->sampler;
    int j, k0, k1, k2, k3;
    float val;
 
+   samp->mip_filter( tgsi_sampler, s, t, p, lodbias, rgba );
+
+   /**
+    * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
+    * When we sampled the depth texture, the depth value was put into all
+    * RGBA channels.  We look at the red channel here.
+    */
+
    /* compare four texcoords vs. four texture samples */
    switch (sampler->compare_func) {
    case PIPE_FUNC_LESS:
@@ -826,470 +1500,392 @@ shadow_compare4(const struct pipe_sampler_state *sampler,
 
 
 /**
- * Common code for sampling 1D/2D/cube textures.
- * Could probably extend for 3D...
+ * Compute which cube face is referenced by each texcoord and put that
+ * info into the sampler faces[] array.  Then sample the cube faces
  */
 static void
-sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler,
-                         const float s[QUAD_SIZE],
-                         const float t[QUAD_SIZE],
-                         const float p[QUAD_SIZE],
-                         boolean computeLambda,
-                         float lodbias,
-                         float rgba[NUM_CHANNELS][QUAD_SIZE],
-                         const unsigned faces[4])
-{
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   const struct softpipe_context *sp = samp->sp;
-   const uint unit = samp->unit;
-   const struct pipe_texture *texture = sp->texture[unit];
-   const struct pipe_sampler_state *sampler = sp->sampler[unit];
-   unsigned level0, level1, j, imgFilter;
-   int width, height;
-   float levelBlend;
-
-   choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
-
-   assert(sampler->normalized_coords);
-
-   width = texture->width[level0];
-   height = texture->height[level0];
-
-   assert(width > 0);
-
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4];
-         nearest_texcoord_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_4(sampler->wrap_t, t, height, y);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare(sampler, rgba, p, j);
-            }
+sample_cube(struct tgsi_sampler *tgsi_sampler,
+            const float s[QUAD_SIZE],
+            const float t[QUAD_SIZE],
+            const float p[QUAD_SIZE],
+            float lodbias,
+            float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   unsigned j;
+   float ssss[4], tttt[4];
 
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               unsigned c;
-               x[j] /= 2;
-               y[j] /= 2;
-               get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0,
-                         rgba2, j);
-               if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
-                  shadow_compare(sampler, rgba2, p, j);
-               }
-
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
+   /*
+     major axis
+     direction     target                             sc     tc    ma
+     ----------    -------------------------------    ---    ---   ---
+     +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
+     -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
+     +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
+     -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
+     +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
+     -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
+   */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      float rx = s[j];
+      float ry = t[j];
+      float rz = p[j];
+      const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
+      unsigned face;
+      float sc, tc, ma;
+
+      if (arx >= ary && arx >= arz) {
+         if (rx >= 0.0F) {
+            face = PIPE_TEX_FACE_POS_X;
+            sc = -rz;
+            tc = -ry;
+            ma = arx;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_X;
+            sc = rz;
+            tc = -ry;
+            ma = arx;
          }
       }
-      break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
+      else if (ary >= arx && ary >= arz) {
+         if (ry >= 0.0F) {
+            face = PIPE_TEX_FACE_POS_Y;
+            sc = rx;
+            tc = rz;
+            ma = ary;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_Y;
+            sc = rx;
+            tc = -rz;
+            ma = ary;
+         }
+      }
+      else {
+         if (rz > 0.0F) {
+            face = PIPE_TEX_FACE_POS_Z;
+            sc = rx;
+            tc = -ry;
+            ma = arz;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_Z;
+            sc = -rx;
+            tc = -ry;
+            ma = arz;
+         }
+      }
+
       {
-         int x0[4], y0[4], x1[4], y1[4];
-         float xw[4], yw[4]; /* weights */
-
-         linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
-         linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            float tx[4][4]; /* texels */
-            int c;
-            get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0);
-            get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1);
-            get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2);
-            get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare4(sampler, tx, p);
-            }
+	 const float ima = 1.0 / ma;
+	 ssss[j] = ( sc * ima + 1.0F ) * 0.5F;
+	 tttt[j] = ( tc * ima + 1.0F ) * 0.5F;
+	 samp->faces[j] = face;
+      }
+   }
 
-            /* interpolate R, G, B, A */
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_2d(xw[j], yw[j],
-                                    tx[c][0], tx[c][1],
-                                    tx[c][2], tx[c][3]);
-            }
+   /* In our little pipeline, the compare stage is next.  If compare
+    * is not active, this will point somewhere deeper into the
+    * pipeline, eg. to mip_filter or even img_filter.
+    */
+   samp->compare(tgsi_sampler, ssss, tttt, NULL, lodbias, rgba);
+}
 
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               x0[j] /= 2;
-               y0[j] /= 2;
-               x1[j] /= 2;
-               y1[j] /= 2;
-               get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0);
-               get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1);
-               get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2);
-               get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3);
-               if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
-                  shadow_compare4(sampler, tx, p);
-               }
-
-               /* interpolate R, G, B, A */
-               for (c = 0; c < 4; c++) {
-                  rgba2[c][j] = lerp_2d(xw[j], yw[j],
-                                        tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
-               }
-
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
-         }
-      }
-      break;
+
+
+static wrap_nearest_func
+get_nearest_unorm_wrap(unsigned mode)
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_nearest_unorm_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_nearest_unorm_clamp_to_border;
    default:
       assert(0);
+      return wrap_nearest_unorm_clamp;
    }
 }
 
 
-static INLINE void
-sp_get_samples_1d(const struct tgsi_sampler *sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  boolean computeLambda,
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+static wrap_nearest_func
+get_nearest_wrap(unsigned mode)
 {
-   static const unsigned faces[4] = {0, 0, 0, 0};
-   static const float tzero[4] = {0, 0, 0, 0};
-   sp_get_samples_2d_common(sampler, s, tzero, NULL,
-                            computeLambda, lodbias, rgba, faces);
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return wrap_nearest_repeat;
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_nearest_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_nearest_clamp_to_edge;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_nearest_clamp_to_border;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      return wrap_nearest_mirror_repeat;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      return wrap_nearest_mirror_clamp;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      return wrap_nearest_mirror_clamp_to_edge;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      return wrap_nearest_mirror_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_nearest_repeat;
+   }
 }
 
 
-static INLINE void
-sp_get_samples_2d(const struct tgsi_sampler *sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  boolean computeLambda,
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+static wrap_linear_func
+get_linear_unorm_wrap(unsigned mode)
 {
-   static const unsigned faces[4] = {0, 0, 0, 0};
-   sp_get_samples_2d_common(sampler, s, t, p,
-                            computeLambda, lodbias, rgba, faces);
+   switch (mode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_linear_unorm_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_linear_unorm_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_linear_unorm_clamp;
+   }
 }
 
 
-static INLINE void
-sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  boolean computeLambda,
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+static wrap_linear_func
+get_linear_wrap(unsigned mode)
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   const struct softpipe_context *sp = samp->sp;
-   const uint unit = samp->unit;
-   const struct pipe_texture *texture = sp->texture[unit];
-   const struct pipe_sampler_state *sampler = sp->sampler[unit];
-   /* get/map pipe_surfaces corresponding to 3D tex slices */
-   unsigned level0, level1, j, imgFilter;
-   int width, height, depth;
-   float levelBlend;
-   const uint face = 0;
-
-   choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
-
-   assert(sampler->normalized_coords);
-
-   width = texture->width[level0];
-   height = texture->height[level0];
-   depth = texture->depth[level0];
-
-   assert(width > 0);
-   assert(height > 0);
-   assert(depth > 0);
-
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4], z[4];
-         nearest_texcoord_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_4(sampler->wrap_t, t, height, y);
-         nearest_texcoord_4(sampler->wrap_r, p, depth, z);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j);
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               unsigned c;
-               x[j] /= 2;
-               y[j] /= 2;
-               z[j] /= 2;
-               get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j);
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]);
-               }
-            }
-         }
-      }
-      break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
-      {
-         int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
-         float xw[4], yw[4], zw[4]; /* interpolation weights */
-         linear_texcoord_4(sampler->wrap_s, s, width,  x0, x1, xw);
-         linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
-         linear_texcoord_4(sampler->wrap_r, p, depth,  z0, z1, zw);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int c;
-            float tx0[4][4], tx1[4][4];
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3);
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3);
-
-            /* interpolate R, G, B, A */
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
-                                    tx0[c][0], tx0[c][1],
-                                    tx0[c][2], tx0[c][3],
-                                    tx1[c][0], tx1[c][1],
-                                    tx1[c][2], tx1[c][3]);
-            }
-
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               x0[j] /= 2;
-               y0[j] /= 2;
-               z0[j] /= 2;
-               x1[j] /= 2;
-               y1[j] /= 2;
-               z1[j] /= 2;
-               get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0);
-               get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1);
-               get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2);
-               get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3);
-               get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0);
-               get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1);
-               get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2);
-               get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3);
-
-               /* interpolate R, G, B, A */
-               for (c = 0; c < 4; c++) {
-                  rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j],
-                                        tx0[c][0], tx0[c][1],
-                                        tx0[c][2], tx0[c][3],
-                                        tx1[c][0], tx1[c][1],
-                                        tx1[c][2], tx1[c][3]);
-               }
-
-               /* blend mipmap levels */
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
-         }
-      }
-      break;
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return wrap_linear_repeat;
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_linear_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_linear_clamp_to_edge;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_linear_clamp_to_border;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      return wrap_linear_mirror_repeat;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      return wrap_linear_mirror_clamp;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      return wrap_linear_mirror_clamp_to_edge;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      return wrap_linear_mirror_clamp_to_border;
    default:
       assert(0);
+      return wrap_linear_repeat;
    }
 }
 
 
-static void
-sp_get_samples_cube(const struct tgsi_sampler *sampler,
-                    const float s[QUAD_SIZE],
-                    const float t[QUAD_SIZE],
-                    const float p[QUAD_SIZE],
-                    boolean computeLambda,
-                    float lodbias,
-                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+static compute_lambda_func
+get_lambda_func(const union sp_sampler_key key)
 {
-   unsigned faces[QUAD_SIZE], j;
-   float ssss[4], tttt[4];
-   for (j = 0; j < QUAD_SIZE; j++) {
-      faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j);
+   if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
+      return compute_lambda_vert;
+   
+   switch (key.bits.target) {
+   case PIPE_TEXTURE_1D:
+      return compute_lambda_1d;
+   case PIPE_TEXTURE_2D:
+   case PIPE_TEXTURE_CUBE:
+      return compute_lambda_2d;
+   case PIPE_TEXTURE_3D:
+      return compute_lambda_3d;
+   default:
+      assert(0);
+      return compute_lambda_1d;
    }
-   sp_get_samples_2d_common(sampler, ssss, tttt, NULL,
-                            computeLambda, lodbias, rgba, faces);
 }
 
 
-static void
-sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler,
-                    const float s[QUAD_SIZE],
-                    const float t[QUAD_SIZE],
-                    const float p[QUAD_SIZE],
-                    boolean computeLambda,
-                    float lodbias,
-                    float rgba[NUM_CHANNELS][QUAD_SIZE])
-{
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   const struct softpipe_context *sp = samp->sp;
-   const uint unit = samp->unit;
-   const struct pipe_texture *texture = sp->texture[unit];
-   const struct pipe_sampler_state *sampler = sp->sampler[unit];
-   const uint face = 0;
-   unsigned level0, level1, j, imgFilter;
-   int width, height;
-   float levelBlend;
-
-   choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
-
-   /* texture RECTS cannot be mipmapped */
-   assert(level0 == level1);
-
-   width = texture->width[level0];
-   height = texture->height[level0];
-
-   assert(width > 0);
-
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4];
-         nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare(sampler, rgba, p, j);
-            }
-         }
-      }
+static filter_func
+get_img_filter(const union sp_sampler_key key,
+               unsigned filter,
+               const struct pipe_sampler_state *sampler)
+{
+   switch (key.bits.target) {
+   case PIPE_TEXTURE_1D:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_1d_nearest;
+      else
+         return img_filter_1d_linear;
       break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
+   case PIPE_TEXTURE_2D:
+      /* Try for fast path:
+       */
+      if (key.bits.is_pot &&
+          sampler->wrap_s == sampler->wrap_t &&
+          sampler->normalized_coords) 
       {
-         int x0[4], y0[4], x1[4], y1[4];
-         float xw[4], yw[4]; /* weights */
-         linear_texcoord_unnorm_4(sampler->wrap_s, s, width,  x0, x1, xw);
-         linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            float tx[4][4]; /* texels */
-            int c;
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare4(sampler, tx, p);
+         switch (sampler->wrap_s) {
+         case PIPE_TEX_WRAP_REPEAT:
+            switch (filter) {
+            case PIPE_TEX_FILTER_NEAREST:
+               return img_filter_2d_nearest_repeat_POT;
+            case PIPE_TEX_FILTER_LINEAR:
+               return img_filter_2d_linear_repeat_POT;
+            default:
+               break;
             }
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_2d(xw[j], yw[j],
-                                    tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
+            break;
+         case PIPE_TEX_WRAP_CLAMP:
+            switch (filter) {
+            case PIPE_TEX_FILTER_NEAREST:
+               return img_filter_2d_nearest_clamp_POT;
+            default:
+               break;
             }
          }
       }
+      /* Otherwise use default versions:
+       */
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_2d_nearest;
+      else
+         return img_filter_2d_linear;
+      break;
+   case PIPE_TEXTURE_CUBE:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_cube_nearest;
+      else
+         return img_filter_cube_linear;
+      break;
+   case PIPE_TEXTURE_3D:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_3d_nearest;
+      else
+         return img_filter_3d_linear;
       break;
    default:
       assert(0);
+      return img_filter_1d_nearest;
    }
 }
 
 
 /**
- * Common code for vertex/fragment program texture sampling.
+ * Bind the given texture object and texture cache to the sampler varient.
  */
-static INLINE void
-sp_get_samples(struct tgsi_sampler *tgsi_sampler,
-               const float s[QUAD_SIZE],
-               const float t[QUAD_SIZE],
-               const float p[QUAD_SIZE],
-               boolean computeLambda,
-               float lodbias,
-               float rgba[NUM_CHANNELS][QUAD_SIZE])
+void
+sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
+                                 struct softpipe_tex_tile_cache *tex_cache,
+                                 const struct pipe_texture *texture )
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   const struct softpipe_context *sp = samp->sp;
-   const uint unit = samp->unit;
-   const struct pipe_texture *texture = sp->texture[unit];
-   const struct pipe_sampler_state *sampler = sp->sampler[unit];
-
-   if (!texture)
-      return;
+   const struct pipe_sampler_state *sampler = samp->sampler;
 
-   switch (texture->target) {
-   case PIPE_TEXTURE_1D:
-      assert(sampler->normalized_coords);
-      sp_get_samples_1d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
-      break;
-   case PIPE_TEXTURE_2D:
-      if (sampler->normalized_coords)
-         sp_get_samples_2d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
-      else
-         sp_get_samples_rect(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
-      break;
-   case PIPE_TEXTURE_3D:
-      assert(sampler->normalized_coords);
-      sp_get_samples_3d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
-      break;
-   case PIPE_TEXTURE_CUBE:
-      assert(sampler->normalized_coords);
-      sp_get_samples_cube(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
-      break;
-   default:
-      assert(0);
-   }
-
-#if 0 /* DEBUG */
-   {
-      int i;
-      printf("Sampled at %f, %f, %f:\n", s[0], t[0], p[0]);
-      for (i = 0; i < 4; i++) {
-         printf("Frag %d: %f %f %f %f\n", i,
-                rgba[0][i],
-                rgba[1][i],
-                rgba[2][i],
-                rgba[3][i]);
-      }
-   }
-#endif
+   samp->texture = texture;
+   samp->cache = tex_cache;
+   samp->xpot = util_unsigned_logbase2( texture->width[0] );
+   samp->ypot = util_unsigned_logbase2( texture->height[0] );
+   samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level);
 }
 
 
-/**
- * Called via tgsi_sampler::get_samples() when running a fragment shader.
- * Get four filtered RGBA values from the sampler's texture.
- */
 void
-sp_get_samples_fragment(struct tgsi_sampler *tgsi_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_sampler_varient_destroy( struct sp_sampler_varient *samp )
 {
-   sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba);
+   FREE(samp);
 }
 
 
 /**
- * Called via tgsi_sampler::get_samples() when running a vertex shader.
- * Get four filtered RGBA values from the sampler's texture.
+ * Create a sampler varient for a given set of non-orthogonal state.
  */
-void
-sp_get_samples_vertex(struct tgsi_sampler *tgsi_sampler,
-                      const float s[QUAD_SIZE],
-                      const float t[QUAD_SIZE],
-                      const float p[QUAD_SIZE],
-                      float lodbias,
-                      float rgba[NUM_CHANNELS][QUAD_SIZE])
+struct sp_sampler_varient *
+sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+                           const union sp_sampler_key key )
 {
-   sp_get_samples(tgsi_sampler, s, t, p, FALSE, lodbias, rgba);
+   struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
+   if (!samp)
+      return NULL;
+
+   samp->sampler = sampler;
+   samp->key = key;
+
+   /* Note that (for instance) linear_texcoord_s and
+    * nearest_texcoord_s may be active at the same time, if the
+    * sampler min_img_filter differs from its mag_img_filter.
+    */
+   if (sampler->normalized_coords) {
+      samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
+      samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
+      samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
+      
+      samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
+      samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
+      samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
+   }
+   else {
+      samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
+      samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
+      samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
+      
+      samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
+      samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
+      samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
+   }
+   
+   samp->compute_lambda = get_lambda_func( key );
+
+   samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
+   samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
+
+   switch (sampler->min_mip_filter) {
+   case PIPE_TEX_MIPFILTER_NONE:
+      if (sampler->min_img_filter == sampler->mag_img_filter) 
+         samp->mip_filter = samp->min_img_filter;         
+      else
+         samp->mip_filter = mip_filter_none;
+      break;
+
+   case PIPE_TEX_MIPFILTER_NEAREST:
+      samp->mip_filter = mip_filter_nearest;
+      break;
+
+   case PIPE_TEX_MIPFILTER_LINEAR:
+      if (key.bits.is_pot &&
+          sampler->min_img_filter == sampler->mag_img_filter &&
+          sampler->normalized_coords &&
+          sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
+          sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
+          sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR)
+      {
+         samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
+      }
+      else 
+      {
+         samp->mip_filter = mip_filter_linear;
+      }
+      break;
+   }
+
+   if (sampler->compare_mode != FALSE) {
+      samp->compare = sample_compare;
+   }
+   else {
+      /* Skip compare operation by promoting the mip_filter function
+       * pointer:
+       */
+      samp->compare = samp->mip_filter;
+   }
+   
+   if (key.bits.target == PIPE_TEXTURE_CUBE) {
+      samp->base.get_samples = sample_cube;
+   }
+   else {
+      samp->faces[0] = 0;
+      samp->faces[1] = 0;
+      samp->faces[2] = 0;
+      samp->faces[3] = 0;
+
+      /* Skip cube face determination by promoting the compare
+       * function pointer:
+       */
+      samp->base.get_samples = samp->compare;
+   }
+
+   return samp;
 }
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h
index 40d8eb2c2a..b0797711d3 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.h
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.h
@@ -31,43 +31,122 @@
 
 #include "tgsi/tgsi_exec.h"
 
+struct sp_sampler_varient;
+
+typedef void (*wrap_nearest_func)(const float s[4],
+                                  unsigned size,
+                                  int icoord[4]);
+
+typedef void (*wrap_linear_func)(const float s[4], 
+                                 unsigned size,
+                                 int icoord0[4],
+                                 int icoord1[4],
+                                 float w[4]);
+
+typedef float (*compute_lambda_func)(const struct sp_sampler_varient *sampler,
+                                     const float s[QUAD_SIZE],
+                                     const float t[QUAD_SIZE],
+                                     const float p[QUAD_SIZE],
+                                     float lodbias);
+
+typedef void (*filter_func)(struct tgsi_sampler *tgsi_sampler,
+                            const float s[QUAD_SIZE],
+                            const float t[QUAD_SIZE],
+                            const float p[QUAD_SIZE],
+                            float lodbias,
+                            float rgba[NUM_CHANNELS][QUAD_SIZE]);
+
+
+union sp_sampler_key {
+   struct {
+      unsigned target:3;
+      unsigned is_pot:1;
+      unsigned processor:2;
+      unsigned unit:4;
+      unsigned pad:22;
+   } bits;
+   unsigned value;
+};
 
 /**
  * Subclass of tgsi_sampler
  */
-struct sp_shader_sampler
+struct sp_sampler_varient
 {
    struct tgsi_sampler base;  /**< base class */
 
-   uint unit;
-   struct softpipe_context *sp;
-   struct softpipe_tile_cache *cache;
+   union sp_sampler_key key;
+
+   /* The owner of this struct:
+    */
+   const struct pipe_sampler_state *sampler;
+
+
+   /* Currently bound texture:
+    */
+   const struct pipe_texture *texture;
+   struct softpipe_tex_tile_cache *cache;
+
+   unsigned processor;
+
+   /* For sp_get_samples_2d_linear_POT:
+    */
+   unsigned xpot;
+   unsigned ypot;
+   unsigned level;
+
+   unsigned faces[4];
+   
+   wrap_nearest_func nearest_texcoord_s;
+   wrap_nearest_func nearest_texcoord_t;
+   wrap_nearest_func nearest_texcoord_p;
+
+   wrap_linear_func linear_texcoord_s;
+   wrap_linear_func linear_texcoord_t;
+   wrap_linear_func linear_texcoord_p;
+
+   filter_func min_img_filter;
+   filter_func mag_img_filter;
+
+   compute_lambda_func compute_lambda;
+
+   filter_func mip_filter;
+   filter_func compare;
+   
+   /* Linked list:
+    */
+   struct sp_sampler_varient *next;
 };
 
+struct sp_sampler;
 
+/* Create a sampler varient for a given set of non-orthogonal state.  Currently the 
+ */
+struct sp_sampler_varient *
+sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+                           const union sp_sampler_key key );
 
-static INLINE const struct sp_shader_sampler *
-sp_shader_sampler(const struct tgsi_sampler *sampler)
-{
-   return (const struct sp_shader_sampler *) sampler;
-}
+void sp_sampler_varient_bind_texture( struct sp_sampler_varient *varient,
+                                      struct softpipe_tex_tile_cache *tex_cache,
+                                      const struct pipe_texture *tex );
 
+void sp_sampler_varient_destroy( struct sp_sampler_varient * );
 
-extern void
-sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler,
-                        const float s[QUAD_SIZE],
-                        const float t[QUAD_SIZE],
-                        const float p[QUAD_SIZE],
-                        float lodbias,
-                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
+
+
+static INLINE struct sp_sampler_varient *
+sp_sampler_varient(const struct tgsi_sampler *sampler)
+{
+   return (struct sp_sampler_varient *) sampler;
+}
 
 extern void
-sp_get_samples_vertex(struct tgsi_sampler *tgsi_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(struct tgsi_sampler *tgsi_sampler,
+               const float s[QUAD_SIZE],
+               const float t[QUAD_SIZE],
+               const float p[QUAD_SIZE],
+               float lodbias,
+               float rgba[NUM_CHANNELS][QUAD_SIZE]);
 
 
 #endif /* SP_TEX_SAMPLE_H */
diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.c b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
new file mode 100644
index 0000000000..407a22a9f4
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.c
@@ -0,0 +1,273 @@
+/**************************************************************************
+ * 
+ * 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.
+ * 
+ **************************************************************************/
+
+/**
+ * Texture tile caching.
+ *
+ * Author:
+ *    Brian Paul
+ */
+
+#include "pipe/p_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_tile.h"
+#include "sp_context.h"
+#include "sp_surface.h"
+#include "sp_texture.h"
+#include "sp_tex_tile_cache.h"
+
+   
+
+struct softpipe_tex_tile_cache *
+sp_create_tex_tile_cache( struct pipe_screen *screen )
+{
+   struct softpipe_tex_tile_cache *tc;
+   uint pos;
+
+   tc = CALLOC_STRUCT( softpipe_tex_tile_cache );
+   if (tc) {
+      tc->screen = screen;
+      for (pos = 0; pos < NUM_ENTRIES; pos++) {
+         tc->entries[pos].addr.bits.invalid = 1;
+      }
+      tc->last_tile = &tc->entries[0]; /* any tile */
+   }
+   return tc;
+}
+
+
+void
+sp_destroy_tex_tile_cache(struct softpipe_tex_tile_cache *tc)
+{
+   struct pipe_screen *screen;
+   uint pos;
+
+   for (pos = 0; pos < NUM_ENTRIES; pos++) {
+      /*assert(tc->entries[pos].x < 0);*/
+   }
+   if (tc->transfer) {
+      screen = tc->transfer->texture->screen;
+      screen->tex_transfer_destroy(tc->transfer);
+   }
+   if (tc->tex_trans) {
+      screen = tc->tex_trans->texture->screen;
+      screen->tex_transfer_destroy(tc->tex_trans);
+   }
+
+   FREE( tc );
+}
+
+
+
+
+void
+sp_tex_tile_cache_map_transfers(struct softpipe_tex_tile_cache *tc)
+{
+   if (tc->tex_trans && !tc->tex_trans_map)
+      tc->tex_trans_map = tc->screen->transfer_map(tc->screen, tc->tex_trans);
+}
+
+
+void
+sp_tex_tile_cache_unmap_transfers(struct softpipe_tex_tile_cache *tc)
+{
+   if (tc->tex_trans_map) {
+      tc->screen->transfer_unmap(tc->screen, tc->tex_trans);
+      tc->tex_trans_map = NULL;
+   }
+}
+
+/**
+ * Invalidate all cached tiles for the cached texture.
+ * Should be called when the texture is modified.
+ */
+void
+sp_tex_tile_cache_validate_texture(struct softpipe_tex_tile_cache *tc)
+{
+   unsigned i;
+
+   assert(tc);
+   assert(tc->texture);
+
+   for (i = 0; i < NUM_ENTRIES; i++) {
+      tc->entries[i].addr.bits.invalid = 1;
+   }
+}
+
+/**
+ * Specify the texture to cache.
+ */
+void
+sp_tex_tile_cache_set_texture(struct softpipe_tex_tile_cache *tc,
+                          struct pipe_texture *texture)
+{
+   uint i;
+
+   assert(!tc->transfer);
+
+   if (tc->texture != texture) {
+      pipe_texture_reference(&tc->texture, texture);
+
+      if (tc->tex_trans) {
+         struct pipe_screen *screen = tc->tex_trans->texture->screen;
+         
+         if (tc->tex_trans_map) {
+            screen->transfer_unmap(screen, tc->tex_trans);
+            tc->tex_trans_map = NULL;
+         }
+
+         screen->tex_transfer_destroy(tc->tex_trans);
+         tc->tex_trans = NULL;
+      }
+
+      /* mark as entries as invalid/empty */
+      /* XXX we should try to avoid this when the teximage hasn't changed */
+      for (i = 0; i < NUM_ENTRIES; i++) {
+         tc->entries[i].addr.bits.invalid = 1;
+      }
+
+      tc->tex_face = -1; /* any invalid value here */
+   }
+}
+
+
+
+
+/**
+ * Flush the tile cache: write all dirty tiles back to the transfer.
+ * any tiles "flagged" as cleared will be "really" cleared.
+ */
+void
+sp_flush_tex_tile_cache(struct softpipe_tex_tile_cache *tc)
+{
+   int pos;
+
+   if (tc->texture) {
+      /* caching a texture, mark all entries as empty */
+      for (pos = 0; pos < NUM_ENTRIES; pos++) {
+         tc->entries[pos].addr.bits.invalid = 1;
+      }
+      tc->tex_face = -1;
+   }
+
+}
+
+
+/**
+ * Given the texture face, level, zslice, x and y values, compute
+ * the cache entry position/index where we'd hope to find the
+ * cached texture tile.
+ * This is basically a direct-map cache.
+ * XXX There's probably lots of ways in which we can improve this.
+ */
+static INLINE uint
+tex_cache_pos( union tex_tile_address addr )
+{
+   uint entry = (addr.bits.x + 
+                 addr.bits.y * 9 + 
+                 addr.bits.z * 3 + 
+                 addr.bits.face + 
+                 addr.bits.level * 7);
+
+   return entry % NUM_ENTRIES;
+}
+
+/**
+ * Similar to sp_get_cached_tile() but for textures.
+ * Tiles are read-only and indexed with more params.
+ */
+const struct softpipe_tex_cached_tile *
+sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc, 
+                        union tex_tile_address addr )
+{
+   struct pipe_screen *screen = tc->screen;
+   struct softpipe_tex_cached_tile *tile;
+   
+   tile = tc->entries + tex_cache_pos( addr );
+
+   if (addr.value != tile->addr.value) {
+
+      /* cache miss.  Most misses are because we've invaldiated the
+       * texture cache previously -- most commonly on binding a new
+       * texture.  Currently we effectively flush the cache on texture
+       * bind.
+       */
+#if 0
+      _debug_printf("miss at %u:  x=%d y=%d z=%d face=%d level=%d\n"
+                    "   tile %u:  x=%d y=%d z=%d face=%d level=%d\n",
+                    pos, x/TILE_SIZE, y/TILE_SIZE, z, face, level,
+                    pos, tile->addr.bits.x, tile->addr.bits.y, tile->z, tile->face, tile->level);
+#endif
+
+      /* check if we need to get a new transfer */
+      if (!tc->tex_trans ||
+          tc->tex_face != addr.bits.face ||
+          tc->tex_level != addr.bits.level ||
+          tc->tex_z != addr.bits.z) {
+         /* get new transfer (view into texture) */
+
+         if (tc->tex_trans) {
+            if (tc->tex_trans_map) {
+               tc->screen->transfer_unmap(tc->screen, tc->tex_trans);
+               tc->tex_trans_map = NULL;
+            }
+
+            screen->tex_transfer_destroy(tc->tex_trans);
+            tc->tex_trans = NULL;
+         }
+
+         tc->tex_trans = 
+            screen->get_tex_transfer(screen, tc->texture, 
+                                     addr.bits.face, 
+                                     addr.bits.level, 
+                                     addr.bits.z, 
+                                     PIPE_TRANSFER_READ, 0, 0,
+                                     tc->texture->width[addr.bits.level],
+                                     tc->texture->height[addr.bits.level]);
+
+         tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans);
+
+         tc->tex_face = addr.bits.face;
+         tc->tex_level = addr.bits.level;
+         tc->tex_z = addr.bits.z;
+      }
+
+      /* get tile from the transfer (view into texture) */
+      pipe_get_tile_rgba(tc->tex_trans,
+                         addr.bits.x * TILE_SIZE, 
+                         addr.bits.y * TILE_SIZE,
+                         TILE_SIZE, TILE_SIZE,
+                         (float *) tile->data.color);
+      tile->addr = addr;
+   }
+
+   tc->last_tile = tile;
+   return tile;
+}
+
+
+
diff --git a/src/gallium/drivers/softpipe/sp_tex_tile_cache.h b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h
new file mode 100644
index 0000000000..ac6886a3df
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_tex_tile_cache.h
@@ -0,0 +1,155 @@
+/**************************************************************************
+ * 
+ * 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 SP_TEX_TILE_CACHE_H
+#define SP_TEX_TILE_CACHE_H
+
+
+#include "pipe/p_compiler.h"
+
+
+struct softpipe_context;
+struct softpipe_tex_tile_cache;
+
+
+/**
+ * Cache tile size (width and height). This needs to be a power of two.
+ */
+#define TILE_SIZE 64
+
+
+/* If we need to support > 4096, just expand this to be a 64 bit
+ * union, or consider tiling in Z as well.
+ */
+union tex_tile_address {
+   struct {
+      unsigned x:6;             /* 4096 / TILE_SIZE */
+      unsigned y:6;             /* 4096 / TILE_SIZE */
+      unsigned z:12;            /* 4096 -- z not tiled */
+      unsigned face:3;
+      unsigned level:4;
+      unsigned invalid:1;
+   } bits;
+   unsigned value;
+};
+
+
+struct softpipe_tex_cached_tile
+{
+   union tex_tile_address addr;
+   union {
+      float color[TILE_SIZE][TILE_SIZE][4];
+   } data;
+};
+
+#define NUM_ENTRIES 50
+
+struct softpipe_tex_tile_cache
+{
+   struct pipe_screen *screen;
+   struct pipe_transfer *transfer;
+   void *transfer_map;
+
+   struct pipe_texture *texture;  /**< if caching a texture */
+   unsigned timestamp;
+
+   struct softpipe_tex_cached_tile entries[NUM_ENTRIES];
+
+   struct pipe_transfer *tex_trans;
+   void *tex_trans_map;
+   int tex_face, tex_level, tex_z;
+
+   struct softpipe_tex_cached_tile *last_tile;  /**< most recently retrieved tile */
+};
+
+
+extern struct softpipe_tex_tile_cache *
+sp_create_tex_tile_cache( struct pipe_screen *screen );
+
+extern void
+sp_destroy_tex_tile_cache(struct softpipe_tex_tile_cache *tc);
+
+
+extern void
+sp_tex_tile_cache_map_transfers(struct softpipe_tex_tile_cache *tc);
+
+extern void
+sp_tex_tile_cache_unmap_transfers(struct softpipe_tex_tile_cache *tc);
+
+extern void
+sp_tex_tile_cache_set_texture(struct softpipe_tex_tile_cache *tc,
+                          struct pipe_texture *texture);
+
+void
+sp_tex_tile_cache_validate_texture(struct softpipe_tex_tile_cache *tc);
+
+extern void
+sp_flush_tex_tile_cache(struct softpipe_tex_tile_cache *tc);
+
+
+
+extern const struct softpipe_tex_cached_tile *
+sp_find_cached_tile_tex(struct softpipe_tex_tile_cache *tc, 
+                         union tex_tile_address addr );
+
+static INLINE union tex_tile_address
+tex_tile_address( unsigned x,
+		  unsigned y,
+		  unsigned z,
+		  unsigned face,
+		  unsigned level )
+{
+   union tex_tile_address addr;
+
+   addr.value = 0;
+   addr.bits.x = x / TILE_SIZE;
+   addr.bits.y = y / TILE_SIZE;
+   addr.bits.z = z;
+   addr.bits.face = face;
+   addr.bits.level = level;
+      
+   return addr;
+}
+
+/* Quickly retrieve tile if it matches last lookup.
+ */
+static INLINE const struct softpipe_tex_cached_tile *
+sp_get_cached_tile_tex(struct softpipe_tex_tile_cache *tc, 
+                         union tex_tile_address addr )
+{
+   if (tc->last_tile->addr.value == addr.value)
+      return tc->last_tile;
+
+   return sp_find_cached_tile_tex( tc, addr );
+}
+
+
+
+
+
+#endif /* SP_TEX_TILE_CACHE_H */
+
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 548e40c4a8..2e6c43c7ef 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -113,16 +113,20 @@ softpipe_displaytarget_layout(struct pipe_screen *screen,
 
 static struct pipe_texture *
 softpipe_texture_create(struct pipe_screen *screen,
-                        const struct pipe_texture *templat)
+                        const struct pipe_texture *template)
 {
    struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture);
    if (!spt)
       return NULL;
 
-   spt->base = *templat;
+   spt->base = *template;
    pipe_reference_init(&spt->base.reference, 1);
    spt->base.screen = screen;
 
+   spt->pot = (util_is_power_of_two(template->width[0]) &&
+               util_is_power_of_two(template->height[0]) &&
+               util_is_power_of_two(template->depth[0]));
+
    if (spt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
                               PIPE_TEXTURE_USAGE_PRIMARY)) {
       if (!softpipe_displaytarget_layout(screen, spt))
@@ -218,6 +222,13 @@ softpipe_get_tex_surface(struct pipe_screen *screen,
       if (ps->usage & PIPE_BUFFER_USAGE_GPU_READ)
          ps->usage |= PIPE_BUFFER_USAGE_CPU_READ;
 
+      if (ps->usage & (PIPE_BUFFER_USAGE_CPU_WRITE |
+                       PIPE_BUFFER_USAGE_GPU_WRITE)) {
+         /* Mark the surface as dirty.  The tile cache will look for this. */
+         spt->timestamp++;
+         softpipe_screen(screen)->timestamp++;
+      }
+
       ps->face = face;
       ps->level = level;
       ps->zslice = zslice;
@@ -318,27 +329,18 @@ softpipe_transfer_map( struct pipe_screen *screen,
 {
    ubyte *map, *xfer_map;
    struct softpipe_texture *spt;
-   unsigned flags = 0;
 
    assert(transfer->texture);
    spt = softpipe_texture(transfer->texture);
 
-   if (transfer->usage != PIPE_TRANSFER_READ) {
-      flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
-   }
-
-   if (transfer->usage != PIPE_TRANSFER_WRITE) {
-      flags |= PIPE_BUFFER_USAGE_CPU_READ;
-   }
-
-   map = pipe_buffer_map(screen, spt->buffer, flags);
+   map = pipe_buffer_map(screen, spt->buffer, pipe_transfer_buffer_flags(transfer));
    if (map == NULL)
       return NULL;
 
    /* May want to different things here depending on read/write nature
     * of the map:
     */
-   if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) {
+   if (transfer->texture && (transfer->usage & PIPE_TRANSFER_WRITE)) {
       /* Do something to notify sharing contexts of a texture change.
        * In softpipe, that would mean flushing the texture cache.
        */
@@ -364,12 +366,64 @@ softpipe_transfer_unmap(struct pipe_screen *screen,
 
    pipe_buffer_unmap( screen, spt->buffer );
 
-   if (transfer->usage != PIPE_TRANSFER_READ) {
+   if (transfer->usage & PIPE_TRANSFER_WRITE) {
       /* Mark the texture as dirty to expire the tile caches. */
-      spt->modified = TRUE;
+      spt->timestamp++;
    }
 }
 
+static struct pipe_video_surface*
+softpipe_video_surface_create(struct pipe_screen *screen,
+                              enum pipe_video_chroma_format chroma_format,
+                              unsigned width, unsigned height)
+{
+   struct softpipe_video_surface *sp_vsfc;
+   struct pipe_texture template;
+
+   assert(screen);
+   assert(width && height);
+
+   sp_vsfc = CALLOC_STRUCT(softpipe_video_surface);
+   if (!sp_vsfc)
+      return NULL;
+
+   pipe_reference_init(&sp_vsfc->base.reference, 1);
+   sp_vsfc->base.screen = screen;
+   sp_vsfc->base.chroma_format = chroma_format;
+   /*sp_vsfc->base.surface_format = PIPE_VIDEO_SURFACE_FORMAT_VUYA;*/
+   sp_vsfc->base.width = width;
+   sp_vsfc->base.height = height;
+
+   memset(&template, 0, sizeof(struct pipe_texture));
+   template.target = PIPE_TEXTURE_2D;
+   template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+   template.last_level = 0;
+   /* vl_mpeg12_mc_renderer expects this when it's initialized with pot_buffers=true */
+   template.width[0] = util_next_power_of_two(width);
+   template.height[0] = util_next_power_of_two(height);
+   template.depth[0] = 1;
+   pf_get_block(template.format, &template.block);
+   template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+   sp_vsfc->tex = screen->texture_create(screen, &template);
+   if (!sp_vsfc->tex) {
+      FREE(sp_vsfc);
+      return NULL;
+   }
+
+   return &sp_vsfc->base;
+}
+
+
+static void
+softpipe_video_surface_destroy(struct pipe_video_surface *vsfc)
+{
+   struct softpipe_video_surface *sp_vsfc = softpipe_video_surface(vsfc);
+
+   pipe_texture_reference(&sp_vsfc->tex, NULL);
+   FREE(sp_vsfc);
+}
+
 
 void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
@@ -385,6 +439,9 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
    screen->tex_transfer_destroy = softpipe_tex_transfer_destroy;
    screen->transfer_map = softpipe_transfer_map;
    screen->transfer_unmap = softpipe_transfer_unmap;
+
+   screen->video_surface_create = softpipe_video_surface_create;
+   screen->video_surface_destroy = softpipe_video_surface_destroy;
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index b1b6130b22..2ef64e1e7c 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -30,6 +30,7 @@
 
 
 #include "pipe/p_state.h"
+#include "pipe/p_video_state.h"
 
 
 struct pipe_context;
@@ -48,7 +49,11 @@ struct softpipe_texture
     */
    struct pipe_buffer *buffer;
 
-   boolean modified;
+   /* True if texture images are power-of-two in all dimensions:
+    */
+   boolean pot;
+
+   unsigned timestamp;
 };
 
 struct softpipe_transfer
@@ -58,6 +63,15 @@ struct softpipe_transfer
    unsigned long offset;
 };
 
+struct softpipe_video_surface
+{
+   struct pipe_video_surface base;
+
+   /* The data is held here:
+    */
+   struct pipe_texture *tex;
+};
+
 
 /** cast wrappers */
 static INLINE struct softpipe_texture *
@@ -72,6 +86,12 @@ softpipe_transfer(struct pipe_transfer *pt)
    return (struct softpipe_transfer *) pt;
 }
 
+static INLINE struct softpipe_video_surface *
+softpipe_video_surface(struct pipe_video_surface *pvs)
+{
+   return (struct softpipe_video_surface *) pvs;
+}
+
 
 extern void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen);
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index 1f9b8f1f4f..83fb4e0d15 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 /**
- * Texture tile caching.
+ * Render target tile caching.
  *
  * Author:
  *    Brian Paul
@@ -35,38 +35,8 @@
 #include "pipe/p_inlines.h"
 #include "util/u_memory.h"
 #include "util/u_tile.h"
-#include "sp_context.h"
-#include "sp_surface.h"
-#include "sp_texture.h"
 #include "sp_tile_cache.h"
 
-#define NUM_ENTRIES 50
-
-
-/** XXX move these */
-#define MAX_WIDTH 2048
-#define MAX_HEIGHT 2048
-
-
-struct softpipe_tile_cache
-{
-   struct pipe_screen *screen;
-   struct pipe_surface *surface;  /**< the surface we're caching */
-   struct pipe_transfer *transfer;
-   void *transfer_map;
-   struct pipe_texture *texture;  /**< if caching a texture */
-   struct softpipe_cached_tile entries[NUM_ENTRIES];
-   uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32];
-   float clear_color[4];  /**< for color bufs */
-   uint clear_val;        /**< for z+stencil, or packed color clear value */
-   boolean depth_stencil; /**< Is the surface a depth/stencil format? */
-
-   struct pipe_transfer *tex_trans;
-   void *tex_trans_map;
-   int tex_face, tex_level, tex_z;
-
-   struct softpipe_cached_tile tile;  /**< scratch tile for clears */
-};
 
 
 /**
@@ -76,7 +46,7 @@ struct softpipe_tile_cache
  * a LRU replacement policy.
  */
 #define CACHE_POS(x, y) \
-   (((x) / TILE_SIZE + ((y) / TILE_SIZE) * 5) % NUM_ENTRIES)
+   (((x) + (y) * 5) % NUM_ENTRIES)
 
 
 
@@ -84,12 +54,10 @@ struct softpipe_tile_cache
  * Is the tile at (x,y) in cleared state?
  */
 static INLINE uint
-is_clear_flag_set(const uint *bitvec, int x, int y)
+is_clear_flag_set(const uint *bitvec, union tile_address addr)
 {
    int pos, bit;
-   x /= TILE_SIZE;
-   y /= TILE_SIZE;
-   pos = y * (MAX_WIDTH / TILE_SIZE) + x;
+   pos = addr.bits.y * (MAX_WIDTH / TILE_SIZE) + addr.bits.x;
    assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32);
    bit = bitvec[pos / 32] & (1 << (pos & 31));
    return bit;
@@ -100,12 +68,10 @@ is_clear_flag_set(const uint *bitvec, int x, int y)
  * Mark the tile at (x,y) as not cleared.
  */
 static INLINE void
-clear_clear_flag(uint *bitvec, int x, int y)
+clear_clear_flag(uint *bitvec, union tile_address addr)
 {
    int pos;
-   x /= TILE_SIZE;
-   y /= TILE_SIZE;
-   pos = y * (MAX_WIDTH / TILE_SIZE) + x;
+   pos = addr.bits.y * (MAX_WIDTH / TILE_SIZE) + addr.bits.x;
    assert(pos / 32 < (MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32);
    bitvec[pos / 32] &= ~(1 << (pos & 31));
 }
@@ -116,14 +82,25 @@ sp_create_tile_cache( struct pipe_screen *screen )
 {
    struct softpipe_tile_cache *tc;
    uint pos;
+   int maxLevels, maxTexSize;
+
+   /* sanity checking: max sure MAX_WIDTH/HEIGHT >= largest texture image */
+   maxLevels = screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS);
+   maxTexSize = 1 << (maxLevels - 1);
+   assert(MAX_WIDTH >= maxTexSize);
 
    tc = CALLOC_STRUCT( softpipe_tile_cache );
    if (tc) {
       tc->screen = screen;
       for (pos = 0; pos < NUM_ENTRIES; pos++) {
-         tc->entries[pos].x =
-         tc->entries[pos].y = -1;
+         tc->entries[pos].addr.bits.invalid = 1;
       }
+      tc->last_tile = &tc->entries[0]; /* any tile */
+
+#if TILE_CLEAR_OPTIMIZATION
+      /* set flags to indicate all the tiles are cleared */
+      memset(tc->clear_flags, 255, sizeof(tc->clear_flags));
+#endif
    }
    return tc;
 }
@@ -142,10 +119,6 @@ sp_destroy_tile_cache(struct softpipe_tile_cache *tc)
       screen = tc->transfer->texture->screen;
       screen->tex_transfer_destroy(tc->transfer);
    }
-   if (tc->tex_trans) {
-      screen = tc->tex_trans->texture->screen;
-      screen->tex_transfer_destroy(tc->tex_trans);
-   }
 
    FREE( tc );
 }
@@ -158,8 +131,6 @@ void
 sp_tile_cache_set_surface(struct softpipe_tile_cache *tc,
                           struct pipe_surface *ps)
 {
-   assert(!tc->texture);
-
    if (tc->transfer) {
       struct pipe_screen *screen = tc->transfer->texture->screen;
 
@@ -211,9 +182,6 @@ sp_tile_cache_map_transfers(struct softpipe_tile_cache *tc)
 {
    if (tc->transfer && !tc->transfer_map)
       tc->transfer_map = tc->screen->transfer_map(tc->screen, tc->transfer);
-
-   if (tc->tex_trans && !tc->tex_trans_map)
-      tc->tex_trans_map = tc->screen->transfer_map(tc->screen, tc->tex_trans);
 }
 
 
@@ -224,47 +192,6 @@ sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc)
       tc->screen->transfer_unmap(tc->screen, tc->transfer);
       tc->transfer_map = NULL;
    }
-
-   if (tc->tex_trans_map) {
-      tc->screen->transfer_unmap(tc->screen, tc->tex_trans);
-      tc->tex_trans_map = NULL;
-   }
-}
-
-
-/**
- * Specify the texture to cache.
- */
-void
-sp_tile_cache_set_texture(struct pipe_context *pipe,
-                          struct softpipe_tile_cache *tc,
-                          struct pipe_texture *texture)
-{
-   uint i;
-
-   assert(!tc->transfer);
-
-   pipe_texture_reference(&tc->texture, texture);
-
-   if (tc->tex_trans) {
-      struct pipe_screen *screen = tc->tex_trans->texture->screen;
-
-      if (tc->tex_trans_map) {
-         screen->transfer_unmap(screen, tc->tex_trans);
-         tc->tex_trans_map = NULL;
-      }
-
-      screen->tex_transfer_destroy(tc->tex_trans);
-      tc->tex_trans = NULL;
-   }
-
-   /* mark as entries as invalid/empty */
-   /* XXX we should try to avoid this when the teximage hasn't changed */
-   for (i = 0; i < NUM_ENTRIES; i++) {
-      tc->entries[i].x = -1;
-   }
-
-   tc->tex_face = -1; /* any invalid value here */
 }
 
 
@@ -308,7 +235,7 @@ clear_tile(struct softpipe_cached_tile *tile,
 
    switch (pf_get_size(format)) {
    case 1:
-      memset(tile->data.any, 0, TILE_SIZE * TILE_SIZE);
+      memset(tile->data.any, clear_value, TILE_SIZE * TILE_SIZE);
       break;
    case 2:
       if (clear_value == 0) {
@@ -344,8 +271,7 @@ clear_tile(struct softpipe_cached_tile *tile,
  * Actually clear the tiles which were flagged as being in a clear state.
  */
 static void
-sp_tile_cache_flush_clear(struct pipe_context *pipe,
-                          struct softpipe_tile_cache *tc)
+sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc)
 {
    struct pipe_transfer *pt = tc->transfer;
    const uint w = tc->transfer->width;
@@ -359,13 +285,15 @@ sp_tile_cache_flush_clear(struct pipe_context *pipe,
    /* push the tile to all positions marked as clear */
    for (y = 0; y < h; y += TILE_SIZE) {
       for (x = 0; x < w; x += TILE_SIZE) {
-         if (is_clear_flag_set(tc->clear_flags, x, y)) {
+         union tile_address addr = tile_address(x, y);
+
+         if (is_clear_flag_set(tc->clear_flags, addr)) {
             pipe_put_tile_raw(pt,
                               x, y, TILE_SIZE, TILE_SIZE,
                               tc->tile.data.color32, 0/*STRIDE*/);
 
             /* do this? */
-            clear_clear_flag(tc->clear_flags, x, y);
+            clear_clear_flag(tc->clear_flags, addr);
 
             numCleared++;
          }
@@ -382,8 +310,7 @@ sp_tile_cache_flush_clear(struct pipe_context *pipe,
  * any tiles "flagged" as cleared will be "really" cleared.
  */
 void
-sp_flush_tile_cache(struct softpipe_context *softpipe,
-                    struct softpipe_tile_cache *tc)
+sp_flush_tile_cache(struct softpipe_tile_cache *tc)
 {
    struct pipe_transfer *pt = tc->transfer;
    int inuse = 0, pos;
@@ -392,33 +319,30 @@ sp_flush_tile_cache(struct softpipe_context *softpipe,
       /* caching a drawing transfer */
       for (pos = 0; pos < NUM_ENTRIES; pos++) {
          struct softpipe_cached_tile *tile = tc->entries + pos;
-         if (tile->x >= 0) {
+         if (!tile->addr.bits.invalid) {
             if (tc->depth_stencil) {
                pipe_put_tile_raw(pt,
-                                 tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                                 tile->addr.bits.x * TILE_SIZE, 
+                                 tile->addr.bits.y * TILE_SIZE, 
+                                 TILE_SIZE, TILE_SIZE,
                                  tile->data.depth32, 0/*STRIDE*/);
             }
             else {
                pipe_put_tile_rgba(pt,
-                                  tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                                  tile->addr.bits.x * TILE_SIZE, 
+                                  tile->addr.bits.y * TILE_SIZE, 
+                                  TILE_SIZE, TILE_SIZE,
                                   (float *) tile->data.color);
             }
-            tile->x = tile->y = -1;  /* mark as empty */
+            tile->addr.bits.invalid = 1;  /* mark as empty */
             inuse++;
          }
       }
 
 #if TILE_CLEAR_OPTIMIZATION
-      sp_tile_cache_flush_clear(&softpipe->pipe, tc);
+      sp_tile_cache_flush_clear(tc);
 #endif
    }
-   else if (tc->texture) {
-      /* caching a texture, mark all entries as empty */
-      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);
@@ -431,40 +355,39 @@ sp_flush_tile_cache(struct softpipe_context *softpipe,
  * \param x, y  position of tile, in pixels
  */
 struct softpipe_cached_tile *
-sp_get_cached_tile(struct softpipe_context *softpipe,
-                   struct softpipe_tile_cache *tc, int x, int y)
+sp_find_cached_tile(struct softpipe_tile_cache *tc, 
+                    union tile_address addr )
 {
    struct pipe_transfer *pt = tc->transfer;
-
-   /* tile pos in framebuffer: */
-   const int tile_x = x & ~(TILE_SIZE - 1);
-   const int tile_y = y & ~(TILE_SIZE - 1);
-
+   
    /* cache pos/entry: */
-   const int pos = CACHE_POS(x, y);
+   const int pos = CACHE_POS(addr.bits.x,
+                             addr.bits.y);
    struct softpipe_cached_tile *tile = tc->entries + pos;
 
-   if (tile_x != tile->x ||
-       tile_y != tile->y) {
+   if (addr.value != tile->addr.value) {
 
-      if (tile->x != -1) {
+      if (tile->addr.bits.invalid == 0) {
          /* put dirty tile back in framebuffer */
          if (tc->depth_stencil) {
             pipe_put_tile_raw(pt,
-                              tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                              tile->addr.bits.x * TILE_SIZE,
+                              tile->addr.bits.y * TILE_SIZE,
+                              TILE_SIZE, TILE_SIZE,
                               tile->data.depth32, 0/*STRIDE*/);
          }
          else {
             pipe_put_tile_rgba(pt,
-                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                               tile->addr.bits.x * TILE_SIZE,
+                               tile->addr.bits.y * TILE_SIZE,
+                               TILE_SIZE, TILE_SIZE,
                                (float *) tile->data.color);
          }
       }
 
-      tile->x = tile_x;
-      tile->y = tile_y;
+      tile->addr = addr;
 
-      if (is_clear_flag_set(tc->clear_flags, x, y)) {
+      if (is_clear_flag_set(tc->clear_flags, addr)) {
          /* don't get tile from framebuffer, just clear it */
          if (tc->depth_stencil) {
             clear_tile(tile, pt->format, tc->clear_val);
@@ -472,125 +395,33 @@ sp_get_cached_tile(struct softpipe_context *softpipe,
          else {
             clear_tile_rgba(tile, pt->format, tc->clear_color);
          }
-         clear_clear_flag(tc->clear_flags, x, y);
+         clear_clear_flag(tc->clear_flags, addr);
       }
       else {
          /* get new tile data from transfer */
          if (tc->depth_stencil) {
             pipe_get_tile_raw(pt,
-                              tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                              tile->addr.bits.x * TILE_SIZE, 
+                              tile->addr.bits.y * TILE_SIZE, 
+                              TILE_SIZE, TILE_SIZE,
                               tile->data.depth32, 0/*STRIDE*/);
          }
          else {
             pipe_get_tile_rgba(pt,
-                               tile->x, tile->y, TILE_SIZE, TILE_SIZE,
+                               tile->addr.bits.x * TILE_SIZE, 
+                               tile->addr.bits.y * TILE_SIZE,
+                               TILE_SIZE, TILE_SIZE,
                                (float *) tile->data.color);
          }
       }
    }
 
+   tc->last_tile = tile;
    return tile;
 }
 
 
-/**
- * Given the texture face, level, zslice, x and y values, compute
- * the cache entry position/index where we'd hope to find the
- * cached texture tile.
- * This is basically a direct-map cache.
- * XXX There's probably lots of ways in which we can improve this.
- */
-static INLINE uint
-tex_cache_pos(int x, int y, int z, int face, int level)
-{
-   uint entry = x + y * 9 + z * 3 + face + level * 7;
-   return entry % NUM_ENTRIES;
-}
-
-
-/**
- * Similar to sp_get_cached_tile() but for textures.
- * Tiles are read-only and indexed with more params.
- */
-const struct softpipe_cached_tile *
-sp_get_cached_tile_tex(struct softpipe_context *sp,
-                       struct softpipe_tile_cache *tc, int x, int y, int z,
-                       int face, int level)
-{
-   struct pipe_screen *screen = sp->pipe.screen;
-   /* tile pos in framebuffer: */
-   const int tile_x = x & ~(TILE_SIZE - 1);
-   const int tile_y = y & ~(TILE_SIZE - 1);
-   /* cache pos/entry: */
-   const uint pos = tex_cache_pos(x / TILE_SIZE, y / TILE_SIZE, z,
-                                  face, level);
-   struct softpipe_cached_tile *tile = tc->entries + pos;
 
-   if (tc->texture) {
-      struct softpipe_texture *spt = softpipe_texture(tc->texture);
-      if (spt->modified) {
-         /* texture was modified, invalidate all cached tiles */
-         uint p;
-         for (p = 0; p < NUM_ENTRIES; p++) {
-            tile = tc->entries + p;
-            tile->x = -1;
-         }
-         spt->modified = FALSE;
-      }
-   }
-
-   if (tile_x != tile->x ||
-       tile_y != tile->y ||
-       z != tile->z ||
-       face != tile->face ||
-       level != tile->level) {
-      /* cache miss */
-
-#if 0
-      printf("miss at %u  x=%d y=%d z=%d face=%d level=%d\n", pos,
-             x/TILE_SIZE, y/TILE_SIZE, z, face, level);
-#endif
-      /* check if we need to get a new transfer */
-      if (!tc->tex_trans ||
-          tc->tex_face != face ||
-          tc->tex_level != level ||
-          tc->tex_z != z) {
-         /* get new transfer (view into texture) */
-
-         if (tc->tex_trans) {
-            if (tc->tex_trans_map) {
-               tc->screen->transfer_unmap(tc->screen, tc->tex_trans);
-               tc->tex_trans_map = NULL;
-            }
-
-            screen->tex_transfer_destroy(tc->tex_trans);
-            tc->tex_trans = NULL;
-         }
-
-         tc->tex_trans = screen->get_tex_transfer(screen, tc->texture, face, level, z, 
-                                                  PIPE_TRANSFER_READ, 0, 0,
-                                                  tc->texture->width[level],
-                                                  tc->texture->height[level]);
-         tc->tex_trans_map = screen->transfer_map(screen, tc->tex_trans);
-
-         tc->tex_face = face;
-         tc->tex_level = level;
-         tc->tex_z = z;
-      }
-
-      /* get tile from the transfer (view into texture) */
-      pipe_get_tile_rgba(tc->tex_trans,
-                         tile_x, tile_y, TILE_SIZE, TILE_SIZE,
-                         (float *) tile->data.color);
-      tile->x = tile_x;
-      tile->y = tile_y;
-      tile->z = z;
-      tile->face = face;
-      tile->level = level;
-   }
-
-   return tile;
-}
 
 
 /**
@@ -621,6 +452,6 @@ sp_tile_cache_clear(struct softpipe_tile_cache *tc, const float *rgba,
 
    for (pos = 0; pos < NUM_ENTRIES; pos++) {
       struct softpipe_cached_tile *tile = tc->entries + pos;
-      tile->x = tile->y = -1;
+      tile->addr.bits.invalid = 1;
    }
 }
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.h b/src/gallium/drivers/softpipe/sp_tile_cache.h
index 8f247d0e58..a12092702a 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.h
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.h
@@ -34,7 +34,6 @@
 #include "pipe/p_compiler.h"
 
 
-struct softpipe_context;
 struct softpipe_tile_cache;
 
 
@@ -44,11 +43,23 @@ struct softpipe_tile_cache;
 #define TILE_SIZE 64
 
 
+/* If we need to support > 4096, just expand this to be a 64 bit
+ * union, or consider tiling in Z as well.
+ */
+union tile_address {
+   struct {
+      unsigned x:6;             /* 4096 / TILE_SIZE */
+      unsigned y:6;             /* 4096 / TILE_SIZE */
+      unsigned invalid:1;
+      unsigned pad:19;
+   } bits;
+   unsigned value;
+};
+
 
 struct softpipe_cached_tile
 {
-   int x, y;           /**< pos of tile in window coords */
-   int z, face, level; /**< Extra texture indexes */
+   union tile_address addr;
    union {
       float color[TILE_SIZE][TILE_SIZE][4];
       uint color32[TILE_SIZE][TILE_SIZE];
@@ -59,6 +70,32 @@ struct softpipe_cached_tile
    } data;
 };
 
+#define NUM_ENTRIES 50
+
+
+/** XXX move these */
+#define MAX_WIDTH 4096
+#define MAX_HEIGHT 4096
+
+
+struct softpipe_tile_cache
+{
+   struct pipe_screen *screen;
+   struct pipe_surface *surface;  /**< the surface we're caching */
+   struct pipe_transfer *transfer;
+   void *transfer_map;
+
+   struct softpipe_cached_tile entries[NUM_ENTRIES];
+   uint clear_flags[(MAX_WIDTH / TILE_SIZE) * (MAX_HEIGHT / TILE_SIZE) / 32];
+   float clear_color[4];  /**< for color bufs */
+   uint clear_val;        /**< for z+stencil, or packed color clear value */
+   boolean depth_stencil; /**< Is the surface a depth/stencil format? */
+
+   struct softpipe_cached_tile tile;  /**< scratch tile for clears */
+
+   struct softpipe_cached_tile *last_tile;  /**< most recently retrieved tile */
+};
+
 
 extern struct softpipe_tile_cache *
 sp_create_tile_cache( struct pipe_screen *screen );
@@ -80,26 +117,45 @@ extern void
 sp_tile_cache_unmap_transfers(struct softpipe_tile_cache *tc);
 
 extern void
-sp_tile_cache_set_texture(struct pipe_context *pipe,
-                          struct softpipe_tile_cache *tc,
-                          struct pipe_texture *texture);
-
-extern void
-sp_flush_tile_cache(struct softpipe_context *softpipe,
-                    struct softpipe_tile_cache *tc);
+sp_flush_tile_cache(struct softpipe_tile_cache *tc);
 
 extern void
 sp_tile_cache_clear(struct softpipe_tile_cache *tc, const float *rgba,
                     uint clearValue);
 
 extern struct softpipe_cached_tile *
-sp_get_cached_tile(struct softpipe_context *softpipe,
-                   struct softpipe_tile_cache *tc, int x, int y);
+sp_find_cached_tile(struct softpipe_tile_cache *tc, 
+                    union tile_address addr );
+
+
+static INLINE union tile_address
+tile_address( unsigned x,
+              unsigned y )
+{
+   union tile_address addr;
+
+   addr.value = 0;
+   addr.bits.x = x / TILE_SIZE;
+   addr.bits.y = y / TILE_SIZE;
+      
+   return addr;
+}
+
+/* Quickly retrieve tile if it matches last lookup.
+ */
+static INLINE struct softpipe_cached_tile *
+sp_get_cached_tile(struct softpipe_tile_cache *tc, 
+                   int x, int y )
+{
+   union tile_address addr = tile_address( x, y );
+
+   if (tc->last_tile->addr.value == addr.value)
+      return tc->last_tile;
+
+   return sp_find_cached_tile( tc, addr );
+}
+
 
-extern const struct softpipe_cached_tile *
-sp_get_cached_tile_tex(struct softpipe_context *softpipe,
-                       struct softpipe_tile_cache *tc, int x, int y, int z,
-                       int face, int level);
 
 
 #endif /* SP_TILE_CACHE_H */
diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c
new file mode 100644
index 0000000000..cae2d3efc5
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_video_context.c
@@ -0,0 +1,304 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "sp_video_context.h"
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include "softpipe/sp_winsys.h"
+#include "softpipe/sp_texture.h"
+
+static void
+sp_mpeg12_destroy(struct pipe_video_context *vpipe)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+   assert(vpipe);
+	
+   /* Asserted in softpipe_delete_fs_state() for some reason */
+   ctx->pipe->bind_vs_state(ctx->pipe, NULL);
+   ctx->pipe->bind_fs_state(ctx->pipe, NULL);
+
+   ctx->pipe->delete_blend_state(ctx->pipe, ctx->blend);
+   ctx->pipe->delete_rasterizer_state(ctx->pipe, ctx->rast);
+   ctx->pipe->delete_depth_stencil_alpha_state(ctx->pipe, ctx->dsa);
+
+   pipe_video_surface_reference(&ctx->decode_target, NULL);
+   vl_compositor_cleanup(&ctx->compositor);
+   vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+   ctx->pipe->destroy(ctx->pipe);
+
+   FREE(ctx);
+}
+
+static void
+sp_mpeg12_decode_macroblocks(struct pipe_video_context *vpipe,
+                             struct pipe_video_surface *past,
+                             struct pipe_video_surface *future,
+                             unsigned num_macroblocks,
+                             struct pipe_macroblock *macroblocks,
+                             struct pipe_fence_handle **fence)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+   struct pipe_mpeg12_macroblock *mpeg12_macroblocks = (struct pipe_mpeg12_macroblock*)macroblocks;
+
+   assert(vpipe);
+   assert(num_macroblocks);
+   assert(macroblocks);
+   assert(macroblocks->codec == PIPE_VIDEO_CODEC_MPEG12);
+   assert(ctx->decode_target);
+
+   vl_mpeg12_mc_renderer_render_macroblocks(&ctx->mc_renderer,
+                                            softpipe_video_surface(ctx->decode_target)->tex,
+                                            past ? softpipe_video_surface(past)->tex : NULL,
+                                            future ? softpipe_video_surface(future)->tex : NULL,
+                                            num_macroblocks, mpeg12_macroblocks, fence);
+}
+
+static void
+sp_mpeg12_clear_surface(struct pipe_video_context *vpipe,
+                        unsigned x, unsigned y,
+                        unsigned width, unsigned height,
+                        unsigned value,
+                        struct pipe_surface *surface)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+   assert(vpipe);
+   assert(surface);
+
+   ctx->pipe->surface_fill(ctx->pipe, surface, x, y, width, height, value);
+}
+
+static void
+sp_mpeg12_render_picture(struct pipe_video_context     *vpipe,
+                         /*struct pipe_surface         *backround,
+                         struct pipe_video_rect        *backround_area,*/
+                         struct pipe_video_surface     *src_surface,
+                         enum pipe_mpeg12_picture_type picture_type,
+                         /*unsigned                    num_past_surfaces,
+                         struct pipe_video_surface     *past_surfaces,
+                         unsigned                      num_future_surfaces,
+                         struct pipe_video_surface     *future_surfaces,*/
+                         struct pipe_video_rect        *src_area,
+                         struct pipe_surface           *dst_surface,
+                         struct pipe_video_rect        *dst_area,
+                         /*unsigned                      num_layers,
+                         struct pipe_surface           *layers,
+                         struct pipe_video_rect        *layer_src_areas,
+                         struct pipe_video_rect        *layer_dst_areas*/
+                         struct pipe_fence_handle      **fence)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+	
+   assert(vpipe);
+   assert(src_surface);
+   assert(src_area);
+   assert(dst_surface);
+   assert(dst_area);
+	
+   vl_compositor_render(&ctx->compositor, softpipe_video_surface(src_surface)->tex,
+                        picture_type, src_area, dst_surface->texture, dst_area, fence);
+}
+
+static void
+sp_mpeg12_set_decode_target(struct pipe_video_context *vpipe,
+                            struct pipe_video_surface *dt)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+   assert(vpipe);
+   assert(dt);
+
+   pipe_video_surface_reference(&ctx->decode_target, dt);
+}
+
+static void sp_mpeg12_set_csc_matrix(struct pipe_video_context *vpipe, const float *mat)
+{
+   struct sp_mpeg12_context *ctx = (struct sp_mpeg12_context*)vpipe;
+
+   assert(vpipe);
+
+   vl_compositor_set_csc_matrix(&ctx->compositor, mat);
+}
+
+static bool
+init_pipe_state(struct sp_mpeg12_context *ctx)
+{
+   struct pipe_rasterizer_state rast;
+   struct pipe_blend_state blend;
+   struct pipe_depth_stencil_alpha_state dsa;
+   unsigned i;
+
+   assert(ctx);
+	
+   rast.flatshade = 1;
+   rast.flatshade_first = 0;
+   rast.light_twoside = 0;
+   rast.front_winding = PIPE_WINDING_CCW;
+   rast.cull_mode = PIPE_WINDING_CW;
+   rast.fill_cw = PIPE_POLYGON_MODE_FILL;
+   rast.fill_ccw = PIPE_POLYGON_MODE_FILL;
+   rast.offset_cw = 0;
+   rast.offset_ccw = 0;
+   rast.scissor = 0;
+   rast.poly_smooth = 0;
+   rast.poly_stipple_enable = 0;
+   rast.point_sprite = 0;
+   rast.point_size_per_vertex = 0;
+   rast.multisample = 0;
+   rast.line_smooth = 0;
+   rast.line_stipple_enable = 0;
+   rast.line_stipple_factor = 0;
+   rast.line_stipple_pattern = 0;
+   rast.line_last_pixel = 0;
+   rast.bypass_vs_clip_and_viewport = 0;
+   rast.line_width = 1;
+   rast.point_smooth = 0;
+   rast.point_size = 1;
+   rast.offset_units = 1;
+   rast.offset_scale = 1;
+   /*rast.sprite_coord_mode[i] = ;*/
+   ctx->rast = ctx->pipe->create_rasterizer_state(ctx->pipe, &rast);
+   ctx->pipe->bind_rasterizer_state(ctx->pipe, ctx->rast);
+
+   blend.blend_enable = 0;
+   blend.rgb_func = PIPE_BLEND_ADD;
+   blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
+   blend.alpha_func = PIPE_BLEND_ADD;
+   blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   blend.alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
+   blend.logicop_enable = 0;
+   blend.logicop_func = PIPE_LOGICOP_CLEAR;
+   /* Needed to allow color writes to FB, even if blending disabled */
+   blend.colormask = PIPE_MASK_RGBA;
+   blend.dither = 0;
+   ctx->blend = ctx->pipe->create_blend_state(ctx->pipe, &blend);
+   ctx->pipe->bind_blend_state(ctx->pipe, ctx->blend);
+
+   dsa.depth.enabled = 0;
+   dsa.depth.writemask = 0;
+   dsa.depth.func = PIPE_FUNC_ALWAYS;
+   for (i = 0; i < 2; ++i) {
+      dsa.stencil[i].enabled = 0;
+      dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
+      dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
+      dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
+      dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
+      dsa.stencil[i].ref_value = 0;
+      dsa.stencil[i].valuemask = 0;
+      dsa.stencil[i].writemask = 0;
+   }
+   dsa.alpha.enabled = 0;
+   dsa.alpha.func = PIPE_FUNC_ALWAYS;
+   dsa.alpha.ref_value = 0;
+   ctx->dsa = ctx->pipe->create_depth_stencil_alpha_state(ctx->pipe, &dsa);
+   ctx->pipe->bind_depth_stencil_alpha_state(ctx->pipe, ctx->dsa);
+	
+   return true;
+}
+
+static struct pipe_video_context *
+sp_mpeg12_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                 enum pipe_video_chroma_format chroma_format,
+                 unsigned width, unsigned height)
+{
+   struct sp_mpeg12_context *ctx;
+
+   assert(u_reduce_video_profile(profile) == PIPE_VIDEO_CODEC_MPEG12);
+
+   ctx = CALLOC_STRUCT(sp_mpeg12_context);
+
+   if (!ctx)
+      return NULL;
+
+   ctx->base.profile = profile;
+   ctx->base.chroma_format = chroma_format;
+   ctx->base.width = width;
+   ctx->base.height = height;
+
+   ctx->base.screen = screen;
+   ctx->base.destroy = sp_mpeg12_destroy;
+   ctx->base.decode_macroblocks = sp_mpeg12_decode_macroblocks;
+   ctx->base.clear_surface = sp_mpeg12_clear_surface;
+   ctx->base.render_picture = sp_mpeg12_render_picture;
+   ctx->base.set_decode_target = sp_mpeg12_set_decode_target;
+   ctx->base.set_csc_matrix = sp_mpeg12_set_csc_matrix;
+
+   ctx->pipe = softpipe_create(screen);
+   if (!ctx->pipe) {
+      FREE(ctx);
+      return NULL;
+   }
+
+   /* TODO: Use slice buffering for softpipe when implemented, no advantage to buffering an entire picture */
+   if (!vl_mpeg12_mc_renderer_init(&ctx->mc_renderer, ctx->pipe,
+                                   width, height, chroma_format,
+                                   VL_MPEG12_MC_RENDERER_BUFFER_PICTURE,
+                                   /* TODO: Use XFER_NONE when implemented */
+                                   VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE,
+                                   true)) {
+      ctx->pipe->destroy(ctx->pipe);
+      FREE(ctx);
+      return NULL;
+   }
+	
+   if (!vl_compositor_init(&ctx->compositor, ctx->pipe)) {
+      vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+      ctx->pipe->destroy(ctx->pipe);
+      FREE(ctx);
+      return NULL;
+   }
+	
+   if (!init_pipe_state(ctx)) {
+      vl_compositor_cleanup(&ctx->compositor);
+      vl_mpeg12_mc_renderer_cleanup(&ctx->mc_renderer);
+      ctx->pipe->destroy(ctx->pipe);
+      FREE(ctx);
+      return NULL;
+   }
+
+   return &ctx->base;
+}
+
+struct pipe_video_context *
+sp_video_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height)
+{
+   assert(screen);
+   assert(width && height);
+
+   switch (u_reduce_video_profile(profile)) {
+      case PIPE_VIDEO_CODEC_MPEG12:
+         return sp_mpeg12_create(screen, profile,
+                                 chroma_format,
+                                 width, height);
+      default:
+         return NULL;
+   }
+}
diff --git a/src/gallium/drivers/softpipe/sp_video_context.h b/src/gallium/drivers/softpipe/sp_video_context.h
new file mode 100644
index 0000000000..ccbd1ffe4c
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_video_context.h
@@ -0,0 +1,57 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 SP_VIDEO_CONTEXT_H
+#define SP_VIDEO_CONTEXT_H
+
+#include <pipe/p_video_context.h>
+#include <vl/vl_mpeg12_mc_renderer.h>
+#include <vl/vl_compositor.h>
+
+struct pipe_screen;
+struct pipe_context;
+struct pipe_video_surface;
+
+struct sp_mpeg12_context
+{
+   struct pipe_video_context base;
+   struct pipe_context *pipe;
+   struct pipe_video_surface *decode_target;
+   struct vl_mpeg12_mc_renderer mc_renderer;
+   struct vl_compositor compositor;
+
+   void *rast;
+   void *dsa;
+   void *blend;
+};
+
+struct pipe_video_context *
+sp_video_create(struct pipe_screen *screen, enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height);
+
+#endif /* SP_VIDEO_CONTEXT_H */
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index ae0af4d055..bf470b46ae 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -125,11 +125,11 @@ trace_context_draw_block(struct trace_context *tr_ctx, int flag)
    } else if ((tr_ctx->draw_rule.blocker & flag) &&
               (tr_ctx->draw_blocker & 4)) {
       boolean block = FALSE;
-      debug_printf("%s (%lu %lu) (%lu %lu) (%lu %u) (%lu %u)\n", __FUNCTION__,
-					tr_ctx->draw_rule.fs, tr_ctx->curr.fs,
-					tr_ctx->draw_rule.vs, tr_ctx->curr.vs,
-					tr_ctx->draw_rule.surf, 0,
-					tr_ctx->draw_rule.tex, 0);
+      debug_printf("%s (%p %p) (%p %p) (%p %u) (%p %u)\n", __FUNCTION__,
+                   (void *) tr_ctx->draw_rule.fs, (void *) tr_ctx->curr.fs,
+                   (void *) tr_ctx->draw_rule.vs, (void *) tr_ctx->curr.vs,
+                   (void *) tr_ctx->draw_rule.surf, 0,
+                   (void *) tr_ctx->draw_rule.tex, 0);
       if (tr_ctx->draw_rule.fs &&
           tr_ctx->draw_rule.fs == tr_ctx->curr.fs)
          block = TRUE;
diff --git a/src/gallium/drivers/trace/tr_rbug.c b/src/gallium/drivers/trace/tr_rbug.c
index e85ac15edc..81e0a6f3b0 100644
--- a/src/gallium/drivers/trace/tr_rbug.c
+++ b/src/gallium/drivers/trace/tr_rbug.c
@@ -44,7 +44,7 @@
 
 #if defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 #  define sleep Sleep
-#elif defined(PIPE_OS_LINUX)
+#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD)
 void usleep(int);
 #  define sleep usleep
 #else
diff --git a/src/gallium/drivers/trace/tr_screen.c b/src/gallium/drivers/trace/tr_screen.c
index 26f1c04594..ab605c7fc8 100644
--- a/src/gallium/drivers/trace/tr_screen.c
+++ b/src/gallium/drivers/trace/tr_screen.c
@@ -403,7 +403,7 @@ trace_screen_transfer_map(struct pipe_screen *_screen,
 
    map = screen->transfer_map(screen, transfer);
    if(map) {
-      if(transfer->usage != PIPE_TRANSFER_READ) {
+      if(transfer->usage & PIPE_TRANSFER_WRITE) {
          assert(!tr_trans->map);
          tr_trans->map = map;
       }
diff --git a/src/gallium/include/pipe/p_config.h b/src/gallium/include/pipe/p_config.h
index de99957d9d..f6feea5f74 100644
--- a/src/gallium/include/pipe/p_config.h
+++ b/src/gallium/include/pipe/p_config.h
@@ -122,18 +122,35 @@
 
 #if defined(__linux__)
 #define PIPE_OS_LINUX
+#define PIPE_OS_UNIX
 #endif
 
 #if defined(__FreeBSD__)
+#define PIPE_OS_FREEBSD
 #define PIPE_OS_BSD
+#define PIPE_OS_UNIX
+#endif
+
+#if defined(__OpenBSD__)
+#define PIPE_OS_OPENBSD
+#define PIPE_OS_BSD
+#define PIPE_OS_UNIX
+#endif
+
+#if defined(__NetBSD__)
+#define PIPE_OS_NETBSD
+#define PIPE_OS_BSD
+#define PIPE_OS_UNIX
 #endif
 
 #if defined(__sun)
 #define PIPE_OS_SOLARIS
+#define PIPE_OS_UNIX
 #endif
 
 #if defined(__APPLE__)
 #define PIPE_OS_APPLE
+#define PIPE_OS_UNIX
 #endif
 
 #if defined(_WIN32) || defined(WIN32)
@@ -142,6 +159,7 @@
 
 #if defined(__HAIKU__)
 #define PIPE_OS_HAIKU
+#define PIPE_OS_UNIX
 #endif
 
 /*
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index f252d6df00..f8fa1e3f49 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -193,7 +193,18 @@ enum pipe_texture_target {
 enum pipe_transfer_usage {
    PIPE_TRANSFER_READ = (1 << 0),
    PIPE_TRANSFER_WRITE = (1 << 1),
-   PIPE_TRANSFER_READ_WRITE = PIPE_TRANSFER_READ | PIPE_TRANSFER_WRITE /**< Read/modify/write */
+   /** Read/modify/write */
+   PIPE_TRANSFER_READ_WRITE = PIPE_TRANSFER_READ | PIPE_TRANSFER_WRITE,
+   /** 
+    * The transfer should map the texture storage directly. The driver may
+    * return NULL if that isn't possible, and the state tracker needs to cope
+    * with that and use an alternative path without this flag.
+    *
+    * E.g. the state tracker could have a simpler path which maps textures and
+    * does read/modify/write cycles on them directly, and a more complicated
+    * path which uses minimal read and write transfers.
+    */
+   PIPE_TRANSFER_MAP_DIRECTLY = (1 << 2)
 };
 
 
@@ -315,6 +326,31 @@ enum pipe_transfer_usage {
 #define PIPE_REFERENCED_FOR_READ  (1 << 0)
 #define PIPE_REFERENCED_FOR_WRITE (1 << 1)
 
+
+enum pipe_video_codec
+{
+   PIPE_VIDEO_CODEC_UNKNOWN = 0,
+   PIPE_VIDEO_CODEC_MPEG12,   /**< MPEG1, MPEG2 */
+   PIPE_VIDEO_CODEC_MPEG4,    /**< DIVX, XVID */
+   PIPE_VIDEO_CODEC_VC1,      /**< WMV */
+   PIPE_VIDEO_CODEC_MPEG4_AVC /**< H.264 */
+};
+
+enum pipe_video_profile
+{
+   PIPE_VIDEO_PROFILE_MPEG1,
+   PIPE_VIDEO_PROFILE_MPEG2_SIMPLE,
+   PIPE_VIDEO_PROFILE_MPEG2_MAIN,
+   PIPE_VIDEO_PROFILE_MPEG4_SIMPLE,
+   PIPE_VIDEO_PROFILE_MPEG4_ADVANCED_SIMPLE,
+   PIPE_VIDEO_PROFILE_VC1_SIMPLE,
+   PIPE_VIDEO_PROFILE_VC1_MAIN,
+   PIPE_VIDEO_PROFILE_VC1_ADVANCED,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN,
+   PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH
+};
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_format.h b/src/gallium/include/pipe/p_format.h
index c4469d4a9e..af23080920 100644
--- a/src/gallium/include/pipe/p_format.h
+++ b/src/gallium/include/pipe/p_format.h
@@ -613,6 +613,24 @@ pf_has_alpha( enum pipe_format format )
    }
 }
 
+enum pipe_video_chroma_format
+{
+   PIPE_VIDEO_CHROMA_FORMAT_420,
+   PIPE_VIDEO_CHROMA_FORMAT_422,
+   PIPE_VIDEO_CHROMA_FORMAT_444
+};
+
+#if 0
+enum pipe_video_surface_format
+{
+   PIPE_VIDEO_SURFACE_FORMAT_NV12,  /**< Planar; Y plane, UV plane */
+   PIPE_VIDEO_SURFACE_FORMAT_YV12,  /**< Planar; Y plane, U plane, V plane */
+   PIPE_VIDEO_SURFACE_FORMAT_YUYV,  /**< Interleaved; Y,U,Y,V,Y,U,Y,V */
+   PIPE_VIDEO_SURFACE_FORMAT_UYVY,  /**< Interleaved; U,Y,V,Y,U,Y,V,Y */
+   PIPE_VIDEO_SURFACE_FORMAT_VUYA   /**< Packed; A31-24|Y23-16|U15-8|V7-0 */
+};
+#endif
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_inlines.h b/src/gallium/include/pipe/p_inlines.h
index 30a4aaf409..5fbd62a03d 100644
--- a/src/gallium/include/pipe/p_inlines.h
+++ b/src/gallium/include/pipe/p_inlines.h
@@ -176,6 +176,22 @@ pipe_transfer_destroy( struct pipe_transfer *transf )
    screen->tex_transfer_destroy(transf);
 }
 
+static INLINE unsigned
+pipe_transfer_buffer_flags( struct pipe_transfer *transf )
+{
+   switch (transf->usage & PIPE_TRANSFER_READ_WRITE) {
+   case PIPE_TRANSFER_READ_WRITE:
+      return PIPE_BUFFER_USAGE_CPU_READ | PIPE_BUFFER_USAGE_CPU_WRITE;
+   case PIPE_TRANSFER_READ:
+      return PIPE_BUFFER_USAGE_CPU_READ;
+   case PIPE_TRANSFER_WRITE:
+      return PIPE_BUFFER_USAGE_CPU_WRITE;
+   default:
+      debug_assert(0);
+      return 0;
+   }
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/include/pipe/p_screen.h b/src/gallium/include/pipe/p_screen.h
index 3f30c52a16..f0a4de5df3 100644
--- a/src/gallium/include/pipe/p_screen.h
+++ b/src/gallium/include/pipe/p_screen.h
@@ -53,7 +53,10 @@ extern "C" {
 struct pipe_fence_handle;
 struct pipe_winsys;
 struct pipe_buffer;
-
+struct pipe_texture;
+struct pipe_surface;
+struct pipe_video_surface;
+struct pipe_transfer;
 
 
 /**
@@ -252,6 +255,17 @@ struct pipe_screen {
 
    void (*buffer_destroy)( struct pipe_buffer *buf );
 
+   /**
+    * Create a video surface suitable for use as a decoding target by the
+    * driver's pipe_video_context.
+    */
+   struct pipe_video_surface*
+   (*video_surface_create)( struct pipe_screen *screen,
+                            enum pipe_video_chroma_format chroma_format,
+                            unsigned width, unsigned height );
+
+   void (*video_surface_destroy)( struct pipe_video_surface *vsfc );
+
 
    /**
     * Do any special operations to ensure frontbuffer contents are
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 2187f5b367..287b424e4a 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -114,11 +114,29 @@ struct pipe_rasterizer_state
     * the vertex shader, clipping and viewport processing.  Note that
     * a vertex shader is still needed though, to indicate the mapping
     * from vertex elements to fragment shader input semantics.
+    *
+    * XXX: considered for removal.
     */
    unsigned bypass_vs_clip_and_viewport:1;
 
-   unsigned flatshade_first:1;   /**< take color attribute from the first vertex of a primitive */
-   unsigned gl_rasterization_rules:1; /**< enable tweaks for GL rasterization?  */
+   /** 
+    * Use the first vertex of a primitive as the provoking vertex for
+    * flat shading.
+    */
+   unsigned flatshade_first:1;   
+
+   /** 
+    * When true, triangle rasterization uses (0.5, 0.5) pixel centers
+    * for determining pixel ownership.
+    *
+    * When false, triangle rasterization uses (0,0) pixel centers for
+    * determining pixel ownership.
+    *
+    * Triangle rasterization always uses a 'top,left' rule for pixel
+    * ownership, this just alters which point we consider the pixel
+    * center for that test.
+    */
+   unsigned gl_rasterization_rules:1;
 
    float line_width;
    float point_size;           /**< used when no per-vertex size */
@@ -179,7 +197,6 @@ struct pipe_depth_state
    unsigned enabled:1;         /**< depth test enabled? */
    unsigned writemask:1;       /**< allow depth buffer writes? */
    unsigned func:3;            /**< depth test func (PIPE_FUNC_x) */
-   unsigned occlusion_count:1; /**< do occlusion counting? */
 };
 
 
diff --git a/src/gallium/include/pipe/p_video_context.h b/src/gallium/include/pipe/p_video_context.h
new file mode 100644
index 0000000000..6ae31418fa
--- /dev/null
+++ b/src/gallium/include/pipe/p_video_context.h
@@ -0,0 +1,121 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 PIPE_VIDEO_CONTEXT_H
+#define PIPE_VIDEO_CONTEXT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <pipe/p_video_state.h>
+
+struct pipe_screen;
+struct pipe_buffer;
+struct pipe_surface;
+struct pipe_video_surface;
+struct pipe_macroblock;
+struct pipe_picture_desc;
+struct pipe_fence_handle;
+
+/**
+ * Gallium video rendering context
+ */
+struct pipe_video_context
+{
+   struct pipe_screen *screen;
+   enum pipe_video_profile profile;
+   enum pipe_video_chroma_format chroma_format;
+   unsigned width;
+   unsigned height;
+
+   void *priv; /**< context private data (for DRI for example) */
+
+   void (*destroy)(struct pipe_video_context *vpipe);
+
+   /**
+    * Picture decoding and displaying
+    */
+   /*@{*/
+   void (*decode_bitstream)(struct pipe_video_context *vpipe,
+                            unsigned num_bufs,
+                            struct pipe_buffer **bitstream_buf);
+
+   void (*decode_macroblocks)(struct pipe_video_context *vpipe,
+                              struct pipe_video_surface *past,
+                              struct pipe_video_surface *future,
+                              unsigned num_macroblocks,
+                              struct pipe_macroblock *macroblocks,
+                              struct pipe_fence_handle **fence);
+
+   void (*clear_surface)(struct pipe_video_context *vpipe,
+                         unsigned x, unsigned y,
+                         unsigned width, unsigned height,
+                         unsigned value,
+                         struct pipe_surface *surface);
+
+   void (*render_picture)(struct pipe_video_context     *vpipe,
+                          /*struct pipe_surface         *backround,
+                          struct pipe_video_rect        *backround_area,*/
+                          struct pipe_video_surface     *src_surface,
+                          enum pipe_mpeg12_picture_type picture_type,
+                          /*unsigned                    num_past_surfaces,
+                          struct pipe_video_surface     *past_surfaces,
+                          unsigned                      num_future_surfaces,
+                          struct pipe_video_surface     *future_surfaces,*/
+                          struct pipe_video_rect        *src_area,
+                          struct pipe_surface           *dst_surface,
+                          struct pipe_video_rect        *dst_area,
+                          /*unsigned                      num_layers,
+                          struct pipe_texture           *layers,
+                          struct pipe_video_rect        *layer_src_areas,
+                          struct pipe_video_rect        *layer_dst_areas,*/
+                          struct pipe_fence_handle      **fence);
+   /*@}*/
+
+   /**
+    * Parameter-like states (or properties)
+    */
+   /*@{*/
+   void (*set_picture_desc)(struct pipe_video_context *vpipe,
+                            const struct pipe_picture_desc *desc);
+
+   void (*set_decode_target)(struct pipe_video_context *vpipe,
+                             struct pipe_video_surface *dt);
+
+   void (*set_csc_matrix)(struct pipe_video_context *vpipe, const float *mat);
+
+   /* TODO: Interface for scaling modes, post-processing, etc. */
+   /*@}*/
+};
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* PIPE_VIDEO_CONTEXT_H */
diff --git a/src/gallium/include/pipe/p_video_state.h b/src/gallium/include/pipe/p_video_state.h
new file mode 100644
index 0000000000..4da26d608c
--- /dev/null
+++ b/src/gallium/include/pipe/p_video_state.h
@@ -0,0 +1,184 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 PIPE_VIDEO_STATE_H
+#define PIPE_VIDEO_STATE_H
+
+/* u_reduce_video_profile() needs these */
+#include <pipe/p_compiler.h>
+#include <util/u_debug.h>
+
+#include <pipe/p_defines.h>
+#include <pipe/p_format.h>
+#include <pipe/p_refcnt.h>
+#include <pipe/p_screen.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct pipe_video_surface
+{
+   struct pipe_reference reference;
+   struct pipe_screen *screen;
+   enum pipe_video_chroma_format chroma_format;
+   /*enum pipe_video_surface_format surface_format;*/
+   unsigned width;
+   unsigned height;
+};
+
+static INLINE void
+pipe_video_surface_reference(struct pipe_video_surface **ptr, struct pipe_video_surface *surf)
+{
+   struct pipe_video_surface *old_surf = *ptr;
+
+   if (pipe_reference((struct pipe_reference **)ptr, &surf->reference))
+      old_surf->screen->video_surface_destroy(old_surf);
+}
+
+struct pipe_video_rect
+{
+   unsigned x, y, w, h;
+};
+
+static INLINE enum pipe_video_codec
+u_reduce_video_profile(enum pipe_video_profile profile)
+{
+   switch (profile)
+   {
+      case PIPE_VIDEO_PROFILE_MPEG1:
+      case PIPE_VIDEO_PROFILE_MPEG2_SIMPLE:
+      case PIPE_VIDEO_PROFILE_MPEG2_MAIN:
+         return PIPE_VIDEO_CODEC_MPEG12;
+
+      case PIPE_VIDEO_PROFILE_MPEG4_SIMPLE:
+      case PIPE_VIDEO_PROFILE_MPEG4_ADVANCED_SIMPLE:
+         return PIPE_VIDEO_CODEC_MPEG4;
+
+      case PIPE_VIDEO_PROFILE_VC1_SIMPLE:
+      case PIPE_VIDEO_PROFILE_VC1_MAIN:
+      case PIPE_VIDEO_PROFILE_VC1_ADVANCED:
+         return PIPE_VIDEO_CODEC_VC1;
+
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE:
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_MAIN:
+      case PIPE_VIDEO_PROFILE_MPEG4_AVC_HIGH:
+         return PIPE_VIDEO_CODEC_MPEG4_AVC;
+
+      default:
+         assert(0);
+         return PIPE_VIDEO_CODEC_UNKNOWN;
+   }
+}
+
+enum pipe_mpeg12_picture_type
+{
+   PIPE_MPEG12_PICTURE_TYPE_FIELD_TOP,
+   PIPE_MPEG12_PICTURE_TYPE_FIELD_BOTTOM,
+   PIPE_MPEG12_PICTURE_TYPE_FRAME
+};
+
+enum pipe_mpeg12_macroblock_type
+{
+   PIPE_MPEG12_MACROBLOCK_TYPE_INTRA,
+   PIPE_MPEG12_MACROBLOCK_TYPE_FWD,
+   PIPE_MPEG12_MACROBLOCK_TYPE_BKWD,
+   PIPE_MPEG12_MACROBLOCK_TYPE_BI,
+	
+   PIPE_MPEG12_MACROBLOCK_NUM_TYPES
+};
+
+enum pipe_mpeg12_motion_type
+{
+   PIPE_MPEG12_MOTION_TYPE_FIELD,
+   PIPE_MPEG12_MOTION_TYPE_FRAME,
+   PIPE_MPEG12_MOTION_TYPE_DUALPRIME,
+   PIPE_MPEG12_MOTION_TYPE_16x8
+};
+
+enum pipe_mpeg12_dct_type
+{
+   PIPE_MPEG12_DCT_TYPE_FIELD,
+   PIPE_MPEG12_DCT_TYPE_FRAME
+};
+
+struct pipe_macroblock
+{
+   enum pipe_video_codec codec;
+};
+
+struct pipe_mpeg12_macroblock
+{
+   struct pipe_macroblock base;
+
+   unsigned mbx;
+   unsigned mby;
+   enum pipe_mpeg12_macroblock_type mb_type;
+   enum pipe_mpeg12_motion_type mo_type;
+   enum pipe_mpeg12_dct_type dct_type;
+   signed pmv[2][2][2];
+   unsigned cbp;
+   void *blocks;
+};
+
+#if 0
+struct pipe_picture_desc
+{
+   enum pipe_video_format format;
+};
+
+struct pipe_mpeg12_picture_desc
+{
+   struct pipe_picture_desc base;
+
+   /* TODO: Use bitfields where possible? */
+   struct pipe_surface *forward_reference;
+   struct pipe_surface *backward_reference;
+   unsigned picture_coding_type;
+   unsigned fcode;
+   unsigned intra_dc_precision;
+   unsigned picture_structure;
+   unsigned top_field_first;
+   unsigned frame_pred_frame_dct;
+   unsigned concealment_motion_vectors;
+   unsigned q_scale_type;
+   unsigned intra_vlc_format;
+   unsigned alternate_scan;
+   unsigned full_pel_forward_vector;
+   unsigned full_pel_backward_vector;
+   struct pipe_buffer *intra_quantizer_matrix;
+   struct pipe_buffer *non_intra_quantizer_matrix;
+   struct pipe_buffer *chroma_intra_quantizer_matrix;
+   struct pipe_buffer *chroma_non_intra_quantizer_matrix;
+};
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* PIPE_VIDEO_STATE_H */
diff --git a/src/gallium/state_trackers/dri/dri_extensions.c b/src/gallium/state_trackers/dri/dri_extensions.c
index 4349a4d1d2..f39a305531 100644
--- a/src/gallium/state_trackers/dri/dri_extensions.c
+++ b/src/gallium/state_trackers/dri/dri_extensions.c
@@ -39,6 +39,7 @@
 #define need_GL_ARB_point_parameters
 #define need_GL_ARB_shader_objects
 #define need_GL_ARB_texture_compression
+#define need_GL_ARB_vertex_array_object
 #define need_GL_ARB_vertex_buffer_object
 #define need_GL_ARB_vertex_program
 #define need_GL_ARB_vertex_shader
@@ -52,6 +53,7 @@
 #define need_GL_EXT_framebuffer_object
 #define need_GL_EXT_multi_draw_arrays
 #define need_GL_EXT_secondary_color
+#define need_GL_APPLE_vertex_array_object
 #define need_GL_NV_vertex_program
 #define need_GL_VERSION_2_0
 #define need_GL_VERSION_2_1
@@ -79,6 +81,7 @@ const struct dri_extension card_extensions[] = {
    {"GL_ARB_texture_env_dot3", NULL},
    {"GL_ARB_texture_mirrored_repeat", NULL},
    {"GL_ARB_texture_rectangle", NULL},
+   {"GL_ARB_vertex_array_object", GL_ARB_vertex_array_object_functions},
    {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions},
    {"GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions},
    {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions},
@@ -103,6 +106,7 @@ const struct dri_extension card_extensions[] = {
    {"GL_EXT_texture_lod_bias", NULL},
    {"GL_3DFX_texture_compression_FXT1", NULL},
    {"GL_APPLE_client_storage", NULL},
+   {"GL_APPLE_vertex_array_object", GL_APPLE_vertex_array_object_functions},
    {"GL_MESA_pack_invert", NULL},
    {"GL_MESA_ycbcr_texture", NULL},
    {"GL_NV_blend_square", NULL},
diff --git a/src/gallium/state_trackers/egl/egl_context.c b/src/gallium/state_trackers/egl/egl_context.c
index c4f7361ca0..e21a4a1095 100644
--- a/src/gallium/state_trackers/egl/egl_context.c
+++ b/src/gallium/state_trackers/egl/egl_context.c
@@ -138,7 +138,6 @@ drm_destroy_context(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *context)
 	struct drm_context *c = lookup_drm_context(context);
 	if (!_eglIsContextBound(&c->base)) {
 		st_destroy_context(c->st);
-		c->pipe->destroy(c->pipe);
 		free(c);
 	}
 	return EGL_TRUE;
@@ -160,18 +159,12 @@ drm_make_current(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurfa
 		if (!drawSurf || !readSurf)
 			return EGL_FALSE;
 
-		drawSurf->user = ctx;
-		readSurf->user = ctx;
-
 		st_make_current(ctx->st, drawSurf->stfb, readSurf->stfb);
 
 		/* st_resize_framebuffer needs a bound context to work */
 		st_resize_framebuffer(drawSurf->stfb, drawSurf->w, drawSurf->h);
 		st_resize_framebuffer(readSurf->stfb, readSurf->w, readSurf->h);
 	} else {
-		drawSurf->user = NULL;
-		readSurf->user = NULL;
-
 		st_make_current(NULL, NULL, NULL);
 	}
 
diff --git a/src/gallium/state_trackers/egl/egl_surface.c b/src/gallium/state_trackers/egl/egl_surface.c
index 69e2d6b708..7911a8834e 100644
--- a/src/gallium/state_trackers/egl/egl_surface.c
+++ b/src/gallium/state_trackers/egl/egl_surface.c
@@ -152,7 +152,6 @@ drm_takedown_shown_screen(_EGLDisplay *dpy, struct drm_screen *screen)
 
 	pipe_surface_reference(&screen->surface, NULL);
 	pipe_texture_reference(&screen->tex, NULL);
-	pipe_buffer_reference(&screen->buffer, NULL);
 
 	screen->shown = 0;
 }
@@ -250,8 +249,8 @@ drm_show_screen_surface_mesa(_EGLDriver *drv, _EGLDisplay *dpy,
 
 
 	drm_create_texture(dpy, scrn, mode->Width, mode->Height);
-	if (!scrn->buffer)
-		return EGL_FALSE;
+	if (!scrn->tex)
+		goto err_tex;
 
 	ret = drmModeAddFB(dev->drmFD,
 	                   scrn->front.width, scrn->front.height,
@@ -325,8 +324,8 @@ err_fb:
 err_bo:
 	pipe_surface_reference(&scrn->surface, NULL);
 	pipe_texture_reference(&scrn->tex, NULL);
-	pipe_buffer_reference(&scrn->buffer, NULL);
 
+err_tex:
 	return EGL_FALSE;
 }
 
@@ -353,24 +352,21 @@ drm_swap_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw)
 	if (!surf)
 		return EGL_FALSE;
 
-	/* error checking */
-	if (!_eglSwapBuffers(drv, dpy, draw))
-		return EGL_FALSE;
-
 	st_get_framebuffer_surface(surf->stfb, ST_SURFACE_BACK_LEFT, &back_surf);
 
 	if (back_surf) {
+		struct drm_context *ctx = lookup_drm_context(draw->Binding);
 
 		st_notify_swapbuffers(surf->stfb);
 
-		if (surf->screen) {
-			surf->user->pipe->surface_copy(surf->user->pipe,
+		if (ctx && surf->screen) {
+			ctx->pipe->surface_copy(ctx->pipe,
 				surf->screen->surface,
 				0, 0,
 				back_surf,
 				0, 0,
 				surf->w, surf->h);
-			surf->user->pipe->flush(surf->user->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL);
+			ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE, NULL);
 
 #ifdef DRM_MODE_FEATURE_DIRTYFB
 			/* TODO query connector property to see if this is needed */
diff --git a/src/gallium/state_trackers/egl/egl_tracker.h b/src/gallium/state_trackers/egl/egl_tracker.h
index dd4730f957..73eb1a1226 100644
--- a/src/gallium/state_trackers/egl/egl_tracker.h
+++ b/src/gallium/state_trackers/egl/egl_tracker.h
@@ -69,7 +69,6 @@ struct drm_surface
 	 * drm
 	 */
 
-	struct drm_context *user;
 	struct drm_screen *screen;
 
 	int w;
@@ -94,7 +93,6 @@ struct drm_screen
 	 * pipe
 	 */
 
-	struct pipe_buffer *buffer;
 	struct pipe_texture *tex;
 	struct pipe_surface *surface;
 
diff --git a/src/gallium/state_trackers/g3dvl/Makefile b/src/gallium/state_trackers/g3dvl/Makefile
deleted file mode 100644
index f9f4d6be3c..0000000000
--- a/src/gallium/state_trackers/g3dvl/Makefile
+++ /dev/null
@@ -1,21 +0,0 @@
-TARGET		= libg3dvl.a
-OBJECTS		= vl_display.o vl_screen.o vl_context.o vl_surface.o vl_shader_build.o vl_util.o vl_basic_csc.o	\
-		  vl_r16snorm_mc_buf.o
-GALLIUMDIR	= ../..
-
-CFLAGS		+= -g -Wall -Werror-implicit-function-declaration -fPIC	\
-		   -I${GALLIUMDIR}/include				\
-		   -I${GALLIUMDIR}/auxiliary				\
-		   -I${GALLIUMDIR}/winsys/g3dvl				\
-
-#############################################
-
-.PHONY	= all clean
-
-all: ${TARGET}
-
-${TARGET}: ${OBJECTS}
-	ar rcs $@ $^
-
-clean:
-	rm -rf ${OBJECTS} ${TARGET}
diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
deleted file mode 100644
index b1683b891b..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
+++ /dev/null
@@ -1,720 +0,0 @@
-#define VL_INTERNAL
-#include "vl_basic_csc.h"
-#include <assert.h>
-#include <pipe/p_context.h>
-#include <pipe/p_state.h>
-#include <pipe/p_inlines.h>
-#include <tgsi/tgsi_parse.h>
-#include <tgsi/tgsi_build.h>
-#include <util/u_memory.h>
-#include "vl_csc.h"
-#include "vl_surface.h"
-#include "vl_shader_build.h"
-#include "vl_types.h"
-
-struct vlVertexShaderConsts
-{
-	struct vlVertex4f	dst_scale;
-	struct vlVertex4f	dst_trans;
-	struct vlVertex4f	src_scale;
-	struct vlVertex4f	src_trans;
-};
-
-struct vlFragmentShaderConsts
-{
-	struct vlVertex4f	bias;
-	float			matrix[16];
-};
-
-struct vlBasicCSC
-{
-	struct vlCSC				base;
-
-	struct pipe_context			*pipe;
-	struct pipe_viewport_state		viewport;
-	struct pipe_framebuffer_state		framebuffer;
-	struct pipe_texture			*framebuffer_tex;
-	void					*sampler;
-	void					*vertex_shader, *fragment_shader;
-	struct pipe_vertex_buffer 		vertex_bufs[2];
-	struct pipe_vertex_element		vertex_elems[2];
-	struct pipe_constant_buffer		vs_const_buf, fs_const_buf;
-};
-
-static int vlResizeFrameBuffer
-(
-	struct vlCSC *csc,
-	unsigned int width,
-	unsigned int height
-)
-{
-	struct vlBasicCSC	*basic_csc;
-	struct pipe_context	*pipe;
-	struct pipe_texture	template;
-	float			clear_color[4];
-
-	assert(csc);
-
-	basic_csc = (struct vlBasicCSC*)csc;
-	pipe = basic_csc->pipe;
-
-	if (basic_csc->framebuffer.width == width && basic_csc->framebuffer.height == height)
-		return 0;
-
-	basic_csc->viewport.scale[0] = width;
-	basic_csc->viewport.scale[1] = height;
-	basic_csc->viewport.scale[2] = 1;
-	basic_csc->viewport.scale[3] = 1;
-	basic_csc->viewport.translate[0] = 0;
-	basic_csc->viewport.translate[1] = 0;
-	basic_csc->viewport.translate[2] = 0;
-	basic_csc->viewport.translate[3] = 0;
-
-	clear_color[0]	= 0.0f;
-	clear_color[1]  = 0.0f;
-	clear_color[2]  = 0.0f;
-	clear_color[3]  = 0.0f;
-
-	if (basic_csc->framebuffer_tex)
-	{
-		pipe_surface_reference(&basic_csc->framebuffer.cbufs[0], NULL);
-		pipe_texture_reference(&basic_csc->framebuffer_tex, NULL);
-	}
-	
-	memset(&template, 0, sizeof(struct pipe_texture));
-	template.target = PIPE_TEXTURE_2D;
-	template.format = PIPE_FORMAT_A8R8G8B8_UNORM;
-	template.last_level = 0;
-	template.width[0] = width;
-	template.height[0] = height;
-	template.depth[0] = 1;
-	pf_get_block(template.format, &template.block);
-	template.tex_usage = PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
-
-	basic_csc->framebuffer_tex = pipe->screen->texture_create(pipe->screen, &template);
-
-	basic_csc->framebuffer.width = width;
-	basic_csc->framebuffer.height = height;
-	basic_csc->framebuffer.cbufs[0] = pipe->screen->get_tex_surface
-	(
-		pipe->screen,
-		basic_csc->framebuffer_tex,
-		0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
-	);
-
-	/* Clear to black, in case video doesn't fill the entire window */
-	pipe->set_framebuffer_state(pipe, &basic_csc->framebuffer);
-	pipe->clear(pipe, PIPE_CLEAR_COLOR, clear_color, 0.0f, 0);
-
-	return 0;
-}
-
-static int vlBegin
-(
-	struct vlCSC *csc
-)
-{
-	struct vlBasicCSC	*basic_csc;
-	struct pipe_context	*pipe;
-
-	assert(csc);
-
-	basic_csc = (struct vlBasicCSC*)csc;
-	pipe = basic_csc->pipe;
-
-	pipe->set_framebuffer_state(pipe, &basic_csc->framebuffer);
-	pipe->set_viewport_state(pipe, &basic_csc->viewport);
-	pipe->bind_sampler_states(pipe, 1, (void**)&basic_csc->sampler);
-	/* Source texture set in vlPutPictureCSC() */
-	pipe->bind_vs_state(pipe, basic_csc->vertex_shader);
-	pipe->bind_fs_state(pipe, basic_csc->fragment_shader);
-	pipe->set_vertex_buffers(pipe, 2, basic_csc->vertex_bufs);
-	pipe->set_vertex_elements(pipe, 2, basic_csc->vertex_elems);
-	pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &basic_csc->vs_const_buf);
-	pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &basic_csc->fs_const_buf);
-
-	return 0;
-}
-
-static int vlPutPictureCSC
-(
-	struct vlCSC *csc,
-	struct vlSurface *surface,
-	int srcx,
-	int srcy,
-	int srcw,
-	int srch,
-	int destx,
-	int desty,
-	int destw,
-	int desth,
-	enum vlPictureType picture_type
-)
-{
-	struct vlBasicCSC		*basic_csc;
-	struct pipe_context		*pipe;
-	struct vlVertexShaderConsts	*vs_consts;
-
-	assert(csc);
-	assert(surface);
-
-	basic_csc = (struct vlBasicCSC*)csc;
-	pipe = basic_csc->pipe;
-
-	vs_consts = pipe_buffer_map
-	(
-		pipe->screen,
-		basic_csc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
-	);
-
-	vs_consts->dst_scale.x = destw / (float)basic_csc->framebuffer.cbufs[0]->width;
-	vs_consts->dst_scale.y = desth / (float)basic_csc->framebuffer.cbufs[0]->height;
-	vs_consts->dst_scale.z = 1;
-	vs_consts->dst_scale.w = 1;
-	vs_consts->dst_trans.x = destx / (float)basic_csc->framebuffer.cbufs[0]->width;
-	vs_consts->dst_trans.y = desty / (float)basic_csc->framebuffer.cbufs[0]->height;
-	vs_consts->dst_trans.z = 0;
-	vs_consts->dst_trans.w = 0;
-
-	vs_consts->src_scale.x = srcw / (float)surface->texture->width[0];
-	vs_consts->src_scale.y = srch / (float)surface->texture->height[0];
-	vs_consts->src_scale.z = 1;
-	vs_consts->src_scale.w = 1;
-	vs_consts->src_trans.x = srcx / (float)surface->texture->width[0];
-	vs_consts->src_trans.y = srcy / (float)surface->texture->height[0];
-	vs_consts->src_trans.z = 0;
-	vs_consts->src_trans.w = 0;
-
-	pipe_buffer_unmap(pipe->screen, basic_csc->vs_const_buf.buffer);
-
-	pipe->set_sampler_textures(pipe, 1, &surface->texture);
-	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLE_STRIP, 0, 4);
-
-	return 0;
-}
-
-static int vlEnd
-(
-	struct vlCSC *csc
-)
-{
-	assert(csc);
-
-	return 0;
-}
-
-static struct pipe_surface* vlGetFrameBuffer
-(
-	struct vlCSC *csc
-)
-{
-	struct vlBasicCSC	*basic_csc;
-
-	assert(csc);
-
-	basic_csc = (struct vlBasicCSC*)csc;
-
-	return basic_csc->framebuffer.cbufs[0];
-}
-
-static int vlDestroy
-(
-	struct vlCSC *csc
-)
-{
-	struct vlBasicCSC	*basic_csc;
-	struct pipe_context	*pipe;
-	unsigned int		i;
-
-	assert(csc);
-
-	basic_csc = (struct vlBasicCSC*)csc;
-	pipe = basic_csc->pipe;
-
-	if (basic_csc->framebuffer_tex)
-	{
-		pipe_surface_reference(&basic_csc->framebuffer.cbufs[0], NULL);
-		pipe_texture_reference(&basic_csc->framebuffer_tex, NULL);
-	}
-
-	pipe->delete_sampler_state(pipe, basic_csc->sampler);
-	pipe->delete_vs_state(pipe, basic_csc->vertex_shader);
-	pipe->delete_fs_state(pipe, basic_csc->fragment_shader);
-
-	for (i = 0; i < 2; ++i)
-		pipe_buffer_reference(&basic_csc->vertex_bufs[i].buffer, NULL);
-
-	pipe_buffer_reference(&basic_csc->vs_const_buf.buffer, NULL);
-	pipe_buffer_reference(&basic_csc->fs_const_buf.buffer, NULL);
-
-	FREE(basic_csc);
-
-	return 0;
-}
-
-/*
- * Represents 2 triangles in a strip in normalized coords.
- * Used to render the surface onto the frame buffer.
- */
-static const struct vlVertex2f surface_verts[4] =
-{
-	{0.0f, 0.0f},
-	{0.0f, 1.0f},
-	{1.0f, 0.0f},
-	{1.0f, 1.0f}
-};
-
-/*
- * Represents texcoords for the above. We can use the position values directly.
- * TODO: Duplicate these in the shader, no need to create a buffer.
- */
-static const struct vlVertex2f *surface_texcoords = surface_verts;
-
-/*
- * Identity color conversion constants, for debugging
- */
-static const struct vlFragmentShaderConsts identity =
-{
-	{
-		0.0f, 0.0f, 0.0f, 0.0f
-	},
-	{
-		1.0f, 0.0f, 0.0f, 0.0f,
-		0.0f, 1.0f, 0.0f, 0.0f,
-		0.0f, 0.0f, 1.0f, 0.0f,
-		0.0f, 0.0f, 0.0f, 1.0f
-	}
-};
-
-/*
- * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
- * Y is in [16,235], Cb and Cr are in [16,240]
- * R, G, and B are in [16,235]
- */
-static const struct vlFragmentShaderConsts bt_601 =
-{
-	{
-		0.0f,		0.501960784f,	0.501960784f,	0.0f
-	},
-	{
-		1.0f,		0.0f,		1.371f,		0.0f,
-		1.0f,		-0.336f,	-0.698f,	0.0f,
-		1.0f,		1.732f,		0.0f,		0.0f,
-		0.0f,		0.0f,		0.0f,		1.0f
-	}
-};
-
-/*
- * Converts ITU-R BT.601 YCbCr pixels to RGB pixels where:
- * Y is in [16,235], Cb and Cr are in [16,240]
- * R, G, and B are in [0,255]
- */
-static const struct vlFragmentShaderConsts bt_601_full =
-{
-	{
-		0.062745098f,	0.501960784f,	0.501960784f,	0.0f
-	},
-	{
-		1.164f,		0.0f,		1.596f,		0.0f,
-		1.164f,		-0.391f,	-0.813f,	0.0f,
-		1.164f,		2.018f,		0.0f,		0.0f,
-		0.0f,		0.0f,		0.0f,		1.0f
-	}
-};
-
-/*
- * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
- * Y is in [16,235], Cb and Cr are in [16,240]
- * R, G, and B are in [16,235]
- */
-static const struct vlFragmentShaderConsts bt_709 =
-{
-	{
-		0.0f,		0.501960784f,	0.501960784f,	0.0f
-	},
-	{
-		1.0f,		0.0f,		1.540f,		0.0f,
-		1.0f,		-0.183f,	-0.459f,	0.0f,
-		1.0f,		1.816f,		0.0f,		0.0f,
-		0.0f,		0.0f,		0.0f,		1.0f
-	}
-};
-
-/*
- * Converts ITU-R BT.709 YCbCr pixels to RGB pixels where:
- * Y is in [16,235], Cb and Cr are in [16,240]
- * R, G, and B are in [0,255]
- */
-const struct vlFragmentShaderConsts bt_709_full =
-{
-	{
-		0.062745098f,	0.501960784f,	0.501960784f,	0.0f
-	},
-	{
-		1.164f,		0.0f,		1.793f,		0.0f,
-		1.164f,		-0.213f,	-0.534f,	0.0f,
-		1.164f,		2.115f,		0.0f,		0.0f,
-		0.0f,		0.0f,		0.0f,		1.0f
-	}
-};
-
-static int vlCreateVertexShader
-(
-	struct vlBasicCSC *csc
-)
-{
-	const unsigned int		max_tokens = 50;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(csc);
-
-	pipe = csc->pipe;
-	tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Vertex texcoords
-	 */
-	for (i = 0; i < 2; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl c0		; Scaling vector to scale vertex pos rect to destination size
-	 * decl c1		; Translation vector to move vertex pos rect into position
-	 * decl c2		; Scaling vector to scale texcoord rect to source size
-	 * decl c3		; Translation vector to move texcoord rect into position
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 3);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Vertex texcoords
-	 */
-	for (i = 0; i < 2; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl t0, t1 */
-	decl = vl_decl_temps(0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * madd o0, i0, c0, c1	; Scale and translate unit output rect to destination size and pos
-	 * madd o1, i1, c2, c3	; Scale and translate unit texcoord rect to source size and pos
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_inst4(TGSI_OPCODE_MAD, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i, TGSI_FILE_CONSTANT, i * 2, TGSI_FILE_CONSTANT, i * 2 + 1);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	csc->vertex_shader = pipe->create_vs_state(pipe, &vs);
-	FREE(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShader
-(
-	struct vlBasicCSC *csc
-)
-{
-	const unsigned int		max_tokens = 50;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(csc);
-
-	pipe = csc->pipe;
-	tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/* decl i0		; Texcoords for s0 */
-	decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, 1, 0, 0, TGSI_INTERPOLATE_LINEAR);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl c0		; Bias vector for CSC
-	 * decl c1-c4		; CSC matrix c1-c4
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 4);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0		; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0 */
-	decl = vl_decl_temps(0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl s0		; Sampler for tex containing picture to display */
-	decl = vl_decl_samplers(0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* tex2d t0, i0, s0	; Read src pixel */
-	inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_SAMPLER, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* sub t0, t0, c0	; Subtract bias vector from pixel */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * dp4 o0.x, t0, c1	; Multiply pixel by the color conversion matrix
-	 * dp4 o0.y, t0, c2
-	 * dp4 o0.z, t0, c3
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_DP4, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 1);
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	csc->fragment_shader = pipe->create_fs_state(pipe, &fs);
-	FREE(tokens);
-
-	return 0;
-}
-
-static int vlCreateDataBufs
-(
-	struct vlBasicCSC *csc
-)
-{
-	struct pipe_context *pipe;
-
-	assert(csc);
-
-	pipe = csc->pipe;
-
-	/*
-	 * Create our vertex buffer and vertex buffer element
-	 * VB contains 4 vertices that render a quad covering the entire window
-	 * to display a rendered surface
-	 * Quad is rendered as a tri strip
-	 */
-	csc->vertex_bufs[0].stride = sizeof(struct vlVertex2f);
-	csc->vertex_bufs[0].max_index = 3;
-	csc->vertex_bufs[0].buffer_offset = 0;
-	csc->vertex_bufs[0].buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		1,
-		PIPE_BUFFER_USAGE_VERTEX,
-		sizeof(struct vlVertex2f) * 4
-	);
-
-	memcpy
-	(
-		pipe_buffer_map(pipe->screen, csc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		surface_verts,
-		sizeof(struct vlVertex2f) * 4
-	);
-
-	pipe_buffer_unmap(pipe->screen, csc->vertex_bufs[0].buffer);
-
-	csc->vertex_elems[0].src_offset = 0;
-	csc->vertex_elems[0].vertex_buffer_index = 0;
-	csc->vertex_elems[0].nr_components = 2;
-	csc->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/*
-	 * Create our texcoord buffer and texcoord buffer element
-	 * Texcoord buffer contains the TCs for mapping the rendered surface to the 4 vertices
-	 */
-	csc->vertex_bufs[1].stride = sizeof(struct vlVertex2f);
-	csc->vertex_bufs[1].max_index = 3;
-	csc->vertex_bufs[1].buffer_offset = 0;
-	csc->vertex_bufs[1].buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		1,
-		PIPE_BUFFER_USAGE_VERTEX,
-		sizeof(struct vlVertex2f) * 4
-	);
-
-	memcpy
-	(
-		pipe_buffer_map(pipe->screen, csc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		surface_texcoords,
-		sizeof(struct vlVertex2f) * 4
-	);
-
-	pipe_buffer_unmap(pipe->screen, csc->vertex_bufs[1].buffer);
-
-	csc->vertex_elems[1].src_offset = 0;
-	csc->vertex_elems[1].vertex_buffer_index = 1;
-	csc->vertex_elems[1].nr_components = 2;
-	csc->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/*
-	 * Create our vertex shader's constant buffer
-	 * Const buffer contains scaling and translation vectors
-	 */
-	csc->vs_const_buf.buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		1,
-		PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
-		sizeof(struct vlVertexShaderConsts)
-	);
-
-	/*
-	 * Create our fragment shader's constant buffer
-	 * Const buffer contains the color conversion matrix and bias vectors
-	 */
-	csc->fs_const_buf.buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		1,
-		PIPE_BUFFER_USAGE_CONSTANT,
-		sizeof(struct vlFragmentShaderConsts)
-	);
-
-	/*
-	 * TODO: Refactor this into a seperate function,
-	 * allow changing the CSC matrix at runtime to switch between regular & full versions
-	 */
-	memcpy
-	(
-		pipe_buffer_map(pipe->screen, csc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		&bt_601_full,
-		sizeof(struct vlFragmentShaderConsts)
-	);
-
-	pipe_buffer_unmap(pipe->screen, csc->fs_const_buf.buffer);
-
-	return 0;
-}
-
-static int vlInit
-(
-	struct vlBasicCSC *csc
-)
-{
-	struct pipe_context		*pipe;
-	struct pipe_sampler_state	sampler;
-
-	assert(csc);
-
-	pipe = csc->pipe;
-
-	/* Delay creating the FB until vlPutPictureCSC() so we know window size */
-	csc->framebuffer_tex = NULL;
-	csc->framebuffer.width = 0;
-	csc->framebuffer.height = 0;
-	csc->framebuffer.nr_cbufs = 1;
-	csc->framebuffer.cbufs[0] = NULL;
-	csc->framebuffer.zsbuf = NULL;
-
-	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_img_filter = PIPE_TEX_FILTER_LINEAR;
-	sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-	sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
-	sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
-	sampler.compare_func = PIPE_FUNC_ALWAYS;
-	sampler.normalized_coords = 1;
-	/*sampler.prefilter = ;*/
-	/*sampler.lod_bias = ;*/
-	/*sampler.min_lod = ;*/
-	/*sampler.max_lod = ;*/
-	/*sampler.border_color[i] = ;*/
-	/*sampler.max_anisotropy = ;*/
-	csc->sampler = pipe->create_sampler_state(pipe, &sampler);
-
-	vlCreateVertexShader(csc);
-	vlCreateFragmentShader(csc);
-	vlCreateDataBufs(csc);
-
-	return 0;
-}
-
-int vlCreateBasicCSC
-(
-	struct pipe_context *pipe,
-	struct vlCSC **csc
-)
-{
-	struct vlBasicCSC *basic_csc;
-
-	assert(pipe);
-	assert(csc);
-
-	basic_csc = CALLOC_STRUCT(vlBasicCSC);
-
-	if (!basic_csc)
-		return 1;
-
-	basic_csc->base.vlResizeFrameBuffer = &vlResizeFrameBuffer;
-	basic_csc->base.vlBegin = &vlBegin;
-	basic_csc->base.vlPutPicture = &vlPutPictureCSC;
-	basic_csc->base.vlEnd = &vlEnd;
-	basic_csc->base.vlGetFrameBuffer = &vlGetFrameBuffer;
-	basic_csc->base.vlDestroy = &vlDestroy;
-	basic_csc->pipe = pipe;
-
-	vlInit(basic_csc);
-
-	*csc = &basic_csc->base;
-
-	return 0;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.h b/src/gallium/state_trackers/g3dvl/vl_basic_csc.h
deleted file mode 100644
index 2e17f1d814..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.h
+++ /dev/null
@@ -1,13 +0,0 @@
-#ifndef vl_basic_csc_h
-#define vl_basic_csc_h
-
-struct pipe_context;
-struct vlCSC;
-
-int vlCreateBasicCSC
-(
-	struct pipe_context *pipe,
-	struct vlCSC **csc
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_context.c b/src/gallium/state_trackers/g3dvl/vl_context.c
deleted file mode 100644
index 5cfd233c4c..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_context.c
+++ /dev/null
@@ -1,205 +0,0 @@
-#define VL_INTERNAL
-#include "vl_context.h"
-#include <assert.h>
-#include <pipe/p_context.h>
-#include <pipe/p_state.h>
-#include <util/u_memory.h>
-#include "vl_render.h"
-#include "vl_r16snorm_mc_buf.h"
-#include "vl_csc.h"
-#include "vl_basic_csc.h"
-
-static int vlInitCommon(struct vlContext *context)
-{
-	struct pipe_context			*pipe;
-	struct pipe_rasterizer_state		rast;
-	struct pipe_blend_state			blend;
-	struct pipe_depth_stencil_alpha_state	dsa;
-	unsigned int				i;
-
-	assert(context);
-
-	pipe = context->pipe;
-
-	rast.flatshade = 1;
-	rast.flatshade_first = 0;
-	rast.light_twoside = 0;
-	rast.front_winding = PIPE_WINDING_CCW;
-	rast.cull_mode = PIPE_WINDING_CW;
-	rast.fill_cw = PIPE_POLYGON_MODE_FILL;
-	rast.fill_ccw = PIPE_POLYGON_MODE_FILL;
-	rast.offset_cw = 0;
-	rast.offset_ccw = 0;
-	rast.scissor = 0;
-	rast.poly_smooth = 0;
-	rast.poly_stipple_enable = 0;
-	rast.point_sprite = 0;
-	rast.point_size_per_vertex = 0;
-	rast.multisample = 0;
-	rast.line_smooth = 0;
-	rast.line_stipple_enable = 0;
-	rast.line_stipple_factor = 0;
-	rast.line_stipple_pattern = 0;
-	rast.line_last_pixel = 0;
-	rast.bypass_vs_clip_and_viewport = 0;
-	rast.line_width = 1;
-	rast.point_smooth = 0;
-	rast.point_size = 1;
-	rast.offset_units = 1;
-	rast.offset_scale = 1;
-	/*rast.sprite_coord_mode[i] = ;*/
-	context->raster = pipe->create_rasterizer_state(pipe, &rast);
-	pipe->bind_rasterizer_state(pipe, context->raster);
-
-	blend.blend_enable = 0;
-	blend.rgb_func = PIPE_BLEND_ADD;
-	blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
-	blend.rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
-	blend.alpha_func = PIPE_BLEND_ADD;
-	blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
-	blend.alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
-	blend.logicop_enable = 0;
-	blend.logicop_func = PIPE_LOGICOP_CLEAR;
-	/* Needed to allow color writes to FB, even if blending disabled */
-	blend.colormask = PIPE_MASK_RGBA;
-	blend.dither = 0;
-	context->blend = pipe->create_blend_state(pipe, &blend);
-	pipe->bind_blend_state(pipe, context->blend);
-
-	dsa.depth.enabled = 0;
-	dsa.depth.writemask = 0;
-	dsa.depth.func = PIPE_FUNC_ALWAYS;
-	dsa.depth.occlusion_count = 0;
-	for (i = 0; i < 2; ++i)
-	{
-		dsa.stencil[i].enabled = 0;
-		dsa.stencil[i].func = PIPE_FUNC_ALWAYS;
-		dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
-		dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
-		dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
-		dsa.stencil[i].ref_value = 0;
-		dsa.stencil[i].valuemask = 0;
-		dsa.stencil[i].writemask = 0;
-	}
-	dsa.alpha.enabled = 0;
-	dsa.alpha.func = PIPE_FUNC_ALWAYS;
-	dsa.alpha.ref_value = 0;
-	context->dsa = pipe->create_depth_stencil_alpha_state(pipe, &dsa);
-	pipe->bind_depth_stencil_alpha_state(pipe, context->dsa);
-
-	return 0;
-}
-
-int vlCreateContext
-(
-	struct vlScreen *screen,
-	struct pipe_context *pipe,
-	unsigned int picture_width,
-	unsigned int picture_height,
-	enum vlFormat picture_format,
-	enum vlProfile profile,
-	enum vlEntryPoint entry_point,
-	struct vlContext **context
-)
-{
-	struct vlContext *ctx;
-
-	assert(screen);
-	assert(context);
-	assert(pipe);
-
-	ctx = CALLOC_STRUCT(vlContext);
-
-	if (!ctx)
-		return 1;
-
-	ctx->screen = screen;
-	ctx->pipe = pipe;
-	ctx->picture_width = picture_width;
-	ctx->picture_height = picture_height;
-	ctx->picture_format = picture_format;
-	ctx->profile = profile;
-	ctx->entry_point = entry_point;
-
-	vlInitCommon(ctx);
-
-	vlCreateR16SNormBufferedMC(pipe, picture_width, picture_height, picture_format, &ctx->render);
-	vlCreateBasicCSC(pipe, &ctx->csc);
-
-	*context = ctx;
-
-	return 0;
-}
-
-int vlDestroyContext
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	/* XXX: Must unbind shaders before we can delete them for some reason */
-	context->pipe->bind_vs_state(context->pipe, NULL);
-	context->pipe->bind_fs_state(context->pipe, NULL);
-
-	context->render->vlDestroy(context->render);
-	context->csc->vlDestroy(context->csc);
-
-	context->pipe->delete_blend_state(context->pipe, context->blend);
-	context->pipe->delete_rasterizer_state(context->pipe, context->raster);
-	context->pipe->delete_depth_stencil_alpha_state(context->pipe, context->dsa);
-
-	FREE(context);
-
-	return 0;
-}
-
-struct vlScreen* vlContextGetScreen
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	return context->screen;
-}
-
-struct pipe_context* vlGetPipeContext
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	return context->pipe;
-}
-
-unsigned int vlGetPictureWidth
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	return context->picture_width;
-}
-
-unsigned int vlGetPictureHeight
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	return context->picture_height;
-}
-
-enum vlFormat vlGetPictureFormat
-(
-	struct vlContext *context
-)
-{
-	assert(context);
-
-	return context->picture_format;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_context.h b/src/gallium/state_trackers/g3dvl/vl_context.h
deleted file mode 100644
index 3d14634c44..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_context.h
+++ /dev/null
@@ -1,73 +0,0 @@
-#ifndef vl_context_h
-#define vl_context_h
-
-#include "vl_types.h"
-
-struct pipe_context;
-
-#ifdef VL_INTERNAL
-struct vlRender;
-struct vlCSC;
-
-struct vlContext
-{
-	struct vlScreen		*screen;
-	struct pipe_context	*pipe;
-	unsigned int		picture_width;
-	unsigned int		picture_height;
-	enum vlFormat		picture_format;
-	enum vlProfile		profile;
-	enum vlEntryPoint	entry_point;
-
-	void			*raster;
-	void			*dsa;
-	void			*blend;
-
-	struct vlRender		*render;
-	struct vlCSC		*csc;
-};
-#endif
-
-int vlCreateContext
-(
-	struct vlScreen *screen,
-	struct pipe_context *pipe,
-	unsigned int picture_width,
-	unsigned int picture_height,
-	enum vlFormat picture_format,
-	enum vlProfile profile,
-	enum vlEntryPoint entry_point,
-	struct vlContext **context
-);
-
-int vlDestroyContext
-(
-	struct vlContext *context
-);
-
-struct vlScreen* vlContextGetScreen
-(
-	struct vlContext *context
-);
-
-struct pipe_context* vlGetPipeContext
-(
-	struct vlContext *context
-);
-
-unsigned int vlGetPictureWidth
-(
-	struct vlContext *context
-);
-
-unsigned int vlGetPictureHeight
-(
-	struct vlContext *context
-);
-
-enum vlFormat vlGetPictureFormat
-(
-	struct vlContext *context
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_csc.h b/src/gallium/state_trackers/g3dvl/vl_csc.h
deleted file mode 100644
index 36417a2792..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_csc.h
+++ /dev/null
@@ -1,53 +0,0 @@
-#ifndef vl_csc_h
-#define vl_csc_h
-
-#include "vl_types.h"
-
-struct pipe_surface;
-
-struct vlCSC
-{
-	int (*vlResizeFrameBuffer)
-	(
-		struct vlCSC *csc,
-		unsigned int width,
-		unsigned int height
-	);
-
-	int (*vlBegin)
-	(
-		struct vlCSC *csc
-	);
-
-	int (*vlPutPicture)
-	(
-		struct vlCSC *csc,
-		struct vlSurface *surface,
-		int srcx,
-		int srcy,
-		int srcw,
-		int srch,
-		int destx,
-		int desty,
-		int destw,
-		int desth,
-		enum vlPictureType picture_type
-	);
-
-	int (*vlEnd)
-	(
-		struct vlCSC *csc
-	);
-
-	struct pipe_surface* (*vlGetFrameBuffer)
-	(
-		struct vlCSC *csc
-	);
-
-	int (*vlDestroy)
-	(
-		struct vlCSC *csc
-	);
-};
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_defs.h b/src/gallium/state_trackers/g3dvl/vl_defs.h
deleted file mode 100644
index d612d02502..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_defs.h
+++ /dev/null
@@ -1,11 +0,0 @@
-#ifndef vl_defs_h
-#define vl_defs_h
-
-#define VL_BLOCK_WIDTH		8
-#define VL_BLOCK_HEIGHT		8
-#define VL_BLOCK_SIZE		(VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT)
-#define VL_MACROBLOCK_WIDTH	16
-#define VL_MACROBLOCK_HEIGHT	16
-#define VL_MACROBLOCK_SIZE	(VL_MACROBLOCK_WIDTH * VL_MACROBLOCK_HEIGHT)
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_display.c b/src/gallium/state_trackers/g3dvl/vl_display.c
deleted file mode 100644
index dce06de758..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_display.c
+++ /dev/null
@@ -1,48 +0,0 @@
-#define VL_INTERNAL
-#include "vl_display.h"
-#include <assert.h>
-#include <util/u_memory.h>
-
-int vlCreateDisplay
-(
-	vlNativeDisplay native_display,
-	struct vlDisplay **display
-)
-{
-	struct vlDisplay *dpy;
-
-	assert(native_display);
-	assert(display);
-
-	dpy = CALLOC_STRUCT(vlDisplay);
-
-	if (!dpy)
-		return 1;
-
-	dpy->native = native_display;
-	*display = dpy;
-
-	return 0;
-}
-
-int vlDestroyDisplay
-(
-	struct vlDisplay *display
-)
-{
-	assert(display);
-
-	FREE(display);
-
-	return 0;
-}
-
-vlNativeDisplay vlGetNativeDisplay
-(
-	struct vlDisplay *display
-)
-{
-	assert(display);
-
-	return display->native;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_display.h b/src/gallium/state_trackers/g3dvl/vl_display.h
deleted file mode 100644
index e11fd40799..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_display.h
+++ /dev/null
@@ -1,29 +0,0 @@
-#ifndef vl_display_h
-#define vl_display_h
-
-#include "vl_types.h"
-
-#ifdef VL_INTERNAL
-struct vlDisplay
-{
-	vlNativeDisplay native;
-};
-#endif
-
-int vlCreateDisplay
-(
-	vlNativeDisplay native_display,
-	struct vlDisplay **display
-);
-
-int vlDestroyDisplay
-(
-	struct vlDisplay *display
-);
-
-vlNativeDisplay vlGetNativeDisplay
-(
-	struct vlDisplay *display
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
deleted file mode 100644
index 23631adb69..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
+++ /dev/null
@@ -1,1155 +0,0 @@
-#define VL_INTERNAL
-#include "vl_r16snorm_mc_buf.h"
-#include <assert.h>
-#include <pipe/p_context.h>
-#include <pipe/p_screen.h>
-#include <pipe/p_state.h>
-#include <pipe/p_inlines.h>
-#include <tgsi/tgsi_parse.h>
-#include <tgsi/tgsi_build.h>
-#include <util/u_math.h>
-#include <util/u_memory.h>
-#include "vl_render.h"
-#include "vl_shader_build.h"
-#include "vl_surface.h"
-#include "vl_util.h"
-#include "vl_types.h"
-#include "vl_defs.h"
-
-const unsigned int DEFAULT_BUF_ALIGNMENT = 1;
-
-enum vlMacroBlockTypeEx
-{
-	vlMacroBlockExTypeIntra,
-	vlMacroBlockExTypeFwdPredictedFrame,
-	vlMacroBlockExTypeFwdPredictedField,
-	vlMacroBlockExTypeBkwdPredictedFrame,
-	vlMacroBlockExTypeBkwdPredictedField,
-	vlMacroBlockExTypeBiPredictedFrame,
-	vlMacroBlockExTypeBiPredictedField,
-
-	vlNumMacroBlockExTypes
-};
-
-struct vlVertexShaderConsts
-{
-	struct vlVertex4f denorm;
-};
-
-struct vlFragmentShaderConsts
-{
-	struct vlVertex4f multiplier;
-	struct vlVertex4f div;
-};
-
-struct vlMacroBlockVertexStream0
-{
-	struct vlVertex2f pos;
-	struct vlVertex2f luma_tc;
-	struct vlVertex2f cb_tc;
-	struct vlVertex2f cr_tc;
-};
-
-struct vlR16SnormBufferedMC
-{
-	struct vlRender				base;
-
-	unsigned int				picture_width;
-	unsigned int				picture_height;
-	enum vlFormat				picture_format;
-	unsigned int				macroblocks_per_picture;
-
-	struct vlSurface			*buffered_surface;
-	struct vlSurface			*past_surface;
-	struct vlSurface			*future_surface;
-	struct vlVertex2f			surface_tex_inv_size;
-	struct vlVertex2f			zero_block[3];
-	unsigned int				num_macroblocks;
-	struct vlMpeg2MacroBlock		*macroblocks;
-	struct pipe_transfer			*tex_transfer[3];
-	short					*texels[3];
-
-	struct pipe_context			*pipe;
-	struct pipe_viewport_state		viewport;
-	struct pipe_framebuffer_state		render_target;
-
-	union
-	{
-		void					*all[5];
-		struct
-		{
-			void				*y;
-			void				*cb;
-			void				*cr;
-			void				*ref[2];
-		};
-	} samplers;
-
-	union
-	{
-		struct pipe_texture			*all[5];
-		struct
-		{
-			struct pipe_texture		*y;
-			struct pipe_texture		*cb;
-			struct pipe_texture		*cr;
-			struct pipe_texture		*ref[2];
-		};
-	} textures;
-
-	union
-	{
-		struct pipe_vertex_buffer 		all[3];
-		struct
-		{
-			struct pipe_vertex_buffer	ycbcr;
-			struct pipe_vertex_buffer	ref[2];
-		};
-	} vertex_bufs;
-
-	void					*i_vs, *p_vs[2], *b_vs[2];
-	void					*i_fs, *p_fs[2], *b_fs[2];
-	struct pipe_vertex_element		vertex_elems[8];
-	struct pipe_constant_buffer		vs_const_buf;
-	struct pipe_constant_buffer		fs_const_buf;
-};
-
-static inline int vlBegin
-(
-	struct vlRender *render
-)
-{
-	assert(render);
-
-	return 0;
-}
-
-static inline int vlGrabFrameCodedBlock(short *src, short *dst, unsigned int dst_pitch)
-{
-	unsigned int y;
-
-	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		memcpy
-		(
-			dst + y * dst_pitch,
-			src + y * VL_BLOCK_WIDTH,
-			VL_BLOCK_WIDTH * 2
-		);
-
-	return 0;
-}
-
-static inline int vlGrabFieldCodedBlock(short *src, short *dst, unsigned int dst_pitch)
-{
-	unsigned int y;
-
-	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		memcpy
-		(
-			dst + y * dst_pitch * 2,
-			src + y * VL_BLOCK_WIDTH,
-			VL_BLOCK_WIDTH * 2
-		);
-
-	return 0;
-}
-
-static inline int vlGrabNoBlock(short *dst, unsigned int dst_pitch)
-{
-	unsigned int y;
-
-	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		memset
-		(
-			dst + y * dst_pitch,
-			0,
-			VL_BLOCK_WIDTH * 2
-		);
-
-	return 0;
-}
-
-static inline int vlGrabBlocks
-(
-	struct vlR16SnormBufferedMC *mc,
-	unsigned int mbx,
-	unsigned int mby,
-	enum vlDCTType dct_type,
-	unsigned int coded_block_pattern,
-	short *blocks
-)
-{
-	short			*texels;
-	unsigned int		tex_pitch;
-	unsigned int		x, y, tb = 0, sb = 0;
-	unsigned int		mbpx = mbx * VL_MACROBLOCK_WIDTH, mbpy = mby * VL_MACROBLOCK_HEIGHT;
-
-	assert(mc);
-	assert(blocks);
-
-	tex_pitch = mc->tex_transfer[0]->stride / mc->tex_transfer[0]->block.size;
-	texels = mc->texels[0] + mbpy * tex_pitch + mbpx;
-
-	for (y = 0; y < 2; ++y)
-	{
-		for (x = 0; x < 2; ++x, ++tb)
-		{
-			if ((coded_block_pattern >> (5 - tb)) & 1)
-			{
-				short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
-
-				if (dct_type == vlDCTTypeFrameCoded)
-				{
-					vlGrabFrameCodedBlock
-					(
-						cur_block,
-						texels + y * tex_pitch * VL_BLOCK_HEIGHT + x * VL_BLOCK_WIDTH,
-						tex_pitch
-					);
-				}
-				else
-				{
-					vlGrabFieldCodedBlock
-					(
-						cur_block,
-						texels + y * tex_pitch + x * VL_BLOCK_WIDTH,
-						tex_pitch
-					);
-				}
-
-				++sb;
-			}
-			else if (mc->zero_block[0].x < 0.0f)
-			{
-				vlGrabNoBlock(texels + y * tex_pitch * VL_BLOCK_HEIGHT + x * VL_BLOCK_WIDTH, tex_pitch);
-
-				mc->zero_block[0].x = (mbpx + x * 8) * mc->surface_tex_inv_size.x;
-				mc->zero_block[0].y = (mbpy + y * 8) * mc->surface_tex_inv_size.y;
-			}
-		}
-	}
-
-	/* TODO: Implement 422, 444 */
-	mbpx >>= 1;
-	mbpy >>= 1;
-
-	for (tb = 0; tb < 2; ++tb)
-	{
-		tex_pitch = mc->tex_transfer[tb + 1]->stride / mc->tex_transfer[tb + 1]->block.size;
-		texels = mc->texels[tb + 1] + mbpy * tex_pitch + mbpx;
-
-		if ((coded_block_pattern >> (1 - tb)) & 1)
-		{
-			short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
-
-			vlGrabFrameCodedBlock
-			(
-				cur_block,
-				texels,
-				tex_pitch
-			);
-
-			++sb;
-		}
-		else if (mc->zero_block[tb + 1].x < 0.0f)
-		{
-			vlGrabNoBlock(texels, tex_pitch);
-
-			mc->zero_block[tb + 1].x = (mbpx << 1) * mc->surface_tex_inv_size.x;
-			mc->zero_block[tb + 1].y = (mbpy << 1) * mc->surface_tex_inv_size.y;
-		}
-	}
-
-	return 0;
-}
-
-static inline enum vlMacroBlockTypeEx vlGetMacroBlockTypeEx(struct vlMpeg2MacroBlock *mb)
-{
-	assert(mb);
-
-	switch (mb->mb_type)
-	{
-		case vlMacroBlockTypeIntra:
-			return vlMacroBlockExTypeIntra;
-		case vlMacroBlockTypeFwdPredicted:
-			return mb->mo_type == vlMotionTypeFrame ?
-				vlMacroBlockExTypeFwdPredictedFrame : vlMacroBlockExTypeFwdPredictedField;
-		case vlMacroBlockTypeBkwdPredicted:
-			return mb->mo_type == vlMotionTypeFrame ?
-				vlMacroBlockExTypeBkwdPredictedFrame : vlMacroBlockExTypeBkwdPredictedField;
-		case vlMacroBlockTypeBiPredicted:
-			return mb->mo_type == vlMotionTypeFrame ?
-				vlMacroBlockExTypeBiPredictedFrame : vlMacroBlockExTypeBiPredictedField;
-		default:
-			assert(0);
-	}
-
-	/* Unreachable */
-	return -1;
-}
-
-static inline int vlGrabMacroBlock
-(
-	struct vlR16SnormBufferedMC *mc,
-	struct vlMpeg2MacroBlock *macroblock
-)
-{
-	assert(mc);
-	assert(macroblock);
-	assert(mc->num_macroblocks < mc->macroblocks_per_picture);
-
-	mc->macroblocks[mc->num_macroblocks].mbx = macroblock->mbx;
-	mc->macroblocks[mc->num_macroblocks].mby = macroblock->mby;
-	mc->macroblocks[mc->num_macroblocks].mb_type = macroblock->mb_type;
-	mc->macroblocks[mc->num_macroblocks].mo_type = macroblock->mo_type;
-	mc->macroblocks[mc->num_macroblocks].dct_type = macroblock->dct_type;
-	mc->macroblocks[mc->num_macroblocks].PMV[0][0][0] = macroblock->PMV[0][0][0];
-	mc->macroblocks[mc->num_macroblocks].PMV[0][0][1] = macroblock->PMV[0][0][1];
-	mc->macroblocks[mc->num_macroblocks].PMV[0][1][0] = macroblock->PMV[0][1][0];
-	mc->macroblocks[mc->num_macroblocks].PMV[0][1][1] = macroblock->PMV[0][1][1];
-	mc->macroblocks[mc->num_macroblocks].PMV[1][0][0] = macroblock->PMV[1][0][0];
-	mc->macroblocks[mc->num_macroblocks].PMV[1][0][1] = macroblock->PMV[1][0][1];
-	mc->macroblocks[mc->num_macroblocks].PMV[1][1][0] = macroblock->PMV[1][1][0];
-	mc->macroblocks[mc->num_macroblocks].PMV[1][1][1] = macroblock->PMV[1][1][1];
-	mc->macroblocks[mc->num_macroblocks].cbp = macroblock->cbp;
-	mc->macroblocks[mc->num_macroblocks].blocks = macroblock->blocks;
-
-	vlGrabBlocks
-	(
-		mc,
-		macroblock->mbx,
-		macroblock->mby,
-		macroblock->dct_type,
-		macroblock->cbp,
-		macroblock->blocks
-	);
-
-	mc->num_macroblocks++;
-
-	return 0;
-}
-
-#define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, zb)					\
-	do {															\
-	(vb)[0].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[0].pos.y = (mby) * (unity) + (ofsy);			\
-	(vb)[1].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[1].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
-	(vb)[2].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].pos.y = (mby) * (unity) + (ofsy);			\
-	(vb)[3].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].pos.y = (mby) * (unity) + (ofsy);			\
-	(vb)[4].pos.x = (mbx) * (unitx) + (ofsx);		(vb)[4].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
-	(vb)[5].pos.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].pos.y = (mby) * (unity) + (ofsy) + (hy);		\
-																\
-	if ((cbp) & (lm))													\
-	{															\
-		(vb)[0].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].luma_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[1].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[2].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].luma_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[3].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].luma_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[4].luma_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[5].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].luma_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-	}															\
-	else															\
-	{															\
-		(vb)[0].luma_tc.x = (zb)[0].x;		(vb)[0].luma_tc.y = (zb)[0].y;						\
-		(vb)[1].luma_tc.x = (zb)[0].x;		(vb)[1].luma_tc.y = (zb)[0].y + (hy);					\
-		(vb)[2].luma_tc.x = (zb)[0].x + (hx);	(vb)[2].luma_tc.y = (zb)[0].y;						\
-		(vb)[3].luma_tc.x = (zb)[0].x + (hx);	(vb)[3].luma_tc.y = (zb)[0].y;						\
-		(vb)[4].luma_tc.x = (zb)[0].x;		(vb)[4].luma_tc.y = (zb)[0].y + (hy);					\
-		(vb)[5].luma_tc.x = (zb)[0].x + (hx);	(vb)[5].luma_tc.y = (zb)[0].y + (hy);					\
-	}															\
-																\
-	if ((cbp) & (cbm))													\
-	{															\
-		(vb)[0].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cb_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[1].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[2].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cb_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[3].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cb_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[4].cb_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[5].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cb_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-	}															\
-	else															\
-	{															\
-		(vb)[0].cb_tc.x = (zb)[1].x;		(vb)[0].cb_tc.y = (zb)[1].y;						\
-		(vb)[1].cb_tc.x = (zb)[1].x;		(vb)[1].cb_tc.y = (zb)[1].y + (hy);					\
-		(vb)[2].cb_tc.x = (zb)[1].x + (hx);	(vb)[2].cb_tc.y = (zb)[1].y;						\
-		(vb)[3].cb_tc.x = (zb)[1].x + (hx);	(vb)[3].cb_tc.y = (zb)[1].y;						\
-		(vb)[4].cb_tc.x = (zb)[1].x;		(vb)[4].cb_tc.y = (zb)[1].y + (hy);					\
-		(vb)[5].cb_tc.x = (zb)[1].x + (hx);	(vb)[5].cb_tc.y = (zb)[1].y + (hy);					\
-	}															\
-																\
-	if ((cbp) & (crm))													\
-	{															\
-		(vb)[0].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[0].cr_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[1].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[1].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[2].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[2].cr_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[3].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[3].cr_tc.y = (mby) * (unity) + (ofsy);		\
-		(vb)[4].cr_tc.x = (mbx) * (unitx) + (ofsx);		(vb)[4].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-		(vb)[5].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx);	(vb)[5].cr_tc.y = (mby) * (unity) + (ofsy) + (hy);	\
-	}															\
-	else															\
-	{															\
-		(vb)[0].cr_tc.x = (zb)[2].x;		(vb)[0].cr_tc.y = (zb)[2].y;						\
-		(vb)[1].cr_tc.x = (zb)[2].x;		(vb)[1].cr_tc.y = (zb)[2].y + (hy);					\
-		(vb)[2].cr_tc.x = (zb)[2].x + (hx);	(vb)[2].cr_tc.y = (zb)[2].y;						\
-		(vb)[3].cr_tc.x = (zb)[2].x + (hx);	(vb)[3].cr_tc.y = (zb)[2].y;						\
-		(vb)[4].cr_tc.x = (zb)[2].x;		(vb)[4].cr_tc.y = (zb)[2].y + (hy);					\
-		(vb)[5].cr_tc.x = (zb)[2].x + (hx);	(vb)[5].cr_tc.y = (zb)[2].y + (hy);					\
-	}															\
-	} while (0)
-
-static inline int vlGenMacroblockVerts
-(
-	struct vlR16SnormBufferedMC *mc,
-	struct vlMpeg2MacroBlock *macroblock,
-	unsigned int pos,
-	struct vlMacroBlockVertexStream0 *ycbcr_vb,
-	struct vlVertex2f **ref_vb
-)
-{
-	struct vlVertex2f	mo_vec[2];
-	unsigned int		i;
-
-	assert(mc);
-	assert(macroblock);
-	assert(ycbcr_vb);
-	assert(pos < mc->macroblocks_per_picture);
-
-	switch (macroblock->mb_type)
-	{
-		case vlMacroBlockTypeBiPredicted:
-		{
-			struct vlVertex2f *vb;
-
-			assert(ref_vb && ref_vb[1]);
-
-			vb = ref_vb[1] + pos * 2 * 24;
-
-			mo_vec[0].x = macroblock->PMV[0][1][0] * 0.5f * mc->surface_tex_inv_size.x;
-			mo_vec[0].y = macroblock->PMV[0][1][1] * 0.5f * mc->surface_tex_inv_size.y;
-
-			if (macroblock->mo_type == vlMotionTypeFrame)
-			{
-				for (i = 0; i < 24 * 2; i += 2)
-				{
-					vb[i].x = mo_vec[0].x;
-					vb[i].y = mo_vec[0].y;
-				}
-			}
-			else
-			{
-				mo_vec[1].x = macroblock->PMV[1][1][0] * 0.5f * mc->surface_tex_inv_size.x;
-				mo_vec[1].y = macroblock->PMV[1][1][1] * 0.5f * mc->surface_tex_inv_size.y;
-
-				for (i = 0; i < 24 * 2; i += 2)
-				{
-					vb[i].x = mo_vec[0].x;
-					vb[i].y = mo_vec[0].y;
-					vb[i + 1].x = mo_vec[1].x;
-					vb[i + 1].y = mo_vec[1].y;
-				}
-			}
-
-			/* fall-through */
-		}
-		case vlMacroBlockTypeFwdPredicted:
-		case vlMacroBlockTypeBkwdPredicted:
-		{
-			struct vlVertex2f *vb;
-
-			assert(ref_vb && ref_vb[0]);
-
-			vb = ref_vb[0] + pos * 2 * 24;
-
-			if (macroblock->mb_type == vlMacroBlockTypeBkwdPredicted)
-			{
-				mo_vec[0].x = macroblock->PMV[0][1][0] * 0.5f * mc->surface_tex_inv_size.x;
-				mo_vec[0].y = macroblock->PMV[0][1][1] * 0.5f * mc->surface_tex_inv_size.y;
-
-				if (macroblock->mo_type == vlMotionTypeField)
-				{
-					mo_vec[1].x = macroblock->PMV[1][1][0] * 0.5f * mc->surface_tex_inv_size.x;
-					mo_vec[1].y = macroblock->PMV[1][1][1] * 0.5f * mc->surface_tex_inv_size.y;
-				}
-			}
-			else
-			{
-				mo_vec[0].x = macroblock->PMV[0][0][0] * 0.5f * mc->surface_tex_inv_size.x;
-				mo_vec[0].y = macroblock->PMV[0][0][1] * 0.5f * mc->surface_tex_inv_size.y;
-
-				if (macroblock->mo_type == vlMotionTypeField)
-				{
-					mo_vec[1].x = macroblock->PMV[1][0][0] * 0.5f * mc->surface_tex_inv_size.x;
-					mo_vec[1].y = macroblock->PMV[1][0][1] * 0.5f * mc->surface_tex_inv_size.y;
-				}
-			}
-
-			if (macroblock->mo_type == vlMotionTypeFrame)
-			{
-				for (i = 0; i < 24 * 2; i += 2)
-				{
-					vb[i].x = mo_vec[0].x;
-					vb[i].y = mo_vec[0].y;
-				}
-			}
-			else
-			{
-				for (i = 0; i < 24 * 2; i += 2)
-				{
-					vb[i].x = mo_vec[0].x;
-					vb[i].y = mo_vec[0].y;
-					vb[i + 1].x = mo_vec[1].x;
-					vb[i + 1].y = mo_vec[1].y;
-				}
-			}
-
-			/* fall-through */
-		}
-		case vlMacroBlockTypeIntra:
-		{
-			const struct vlVertex2f	unit =
-			{
-				mc->surface_tex_inv_size.x * VL_MACROBLOCK_WIDTH,
-				mc->surface_tex_inv_size.y * VL_MACROBLOCK_HEIGHT
-			};
-			const struct vlVertex2f half =
-			{
-				mc->surface_tex_inv_size.x * (VL_MACROBLOCK_WIDTH / 2),
-				mc->surface_tex_inv_size.y * (VL_MACROBLOCK_HEIGHT / 2)
-			};
-
-			struct vlMacroBlockVertexStream0 *vb;
-
-			vb = ycbcr_vb + pos * 24;
-
-			SET_BLOCK
-			(
-				vb,
-				macroblock->cbp, macroblock->mbx, macroblock->mby,
-				unit.x, unit.y, 0, 0, half.x, half.y,
-				32, 2, 1, mc->zero_block
-			);
-
-			SET_BLOCK
-			(
-				vb + 6,
-				macroblock->cbp, macroblock->mbx, macroblock->mby,
-				unit.x, unit.y, half.x, 0, half.x, half.y,
-				16, 2, 1, mc->zero_block
-			);
-
-			SET_BLOCK
-			(
-				vb + 12,
-				macroblock->cbp, macroblock->mbx, macroblock->mby,
-				unit.x, unit.y, 0, half.y, half.x, half.y,
-				8, 2, 1, mc->zero_block
-			);
-
-			SET_BLOCK
-			(
-				vb + 18,
-				macroblock->cbp, macroblock->mbx, macroblock->mby,
-				unit.x, unit.y, half.x, half.y, half.x, half.y,
-				4, 2, 1, mc->zero_block
-			);
-
-			break;
-		}
-		default:
-			assert(0);
-	}
-
-	return 0;
-}
-
-static int vlFlush
-(
-	struct vlRender *render
-)
-{
-	struct vlR16SnormBufferedMC	*mc;
-	struct pipe_context		*pipe;
-	struct vlVertexShaderConsts	*vs_consts;
-	unsigned int			num_macroblocks[vlNumMacroBlockExTypes] = {0};
-	unsigned int			offset[vlNumMacroBlockExTypes];
-	unsigned int			vb_start = 0;
-	unsigned int			i;
-
-	assert(render);
-
-	mc = (struct vlR16SnormBufferedMC*)render;
-
-	if (!mc->buffered_surface)
-		return 0;
-
-	if (mc->num_macroblocks < mc->macroblocks_per_picture)
-		return 0;
-
-	assert(mc->num_macroblocks <= mc->macroblocks_per_picture);
-
-	pipe = mc->pipe;
-
-	for (i = 0; i < mc->num_macroblocks; ++i)
-	{
-		enum vlMacroBlockTypeEx mb_type_ex = vlGetMacroBlockTypeEx(&mc->macroblocks[i]);
-
-		num_macroblocks[mb_type_ex]++;
-	}
-
-	offset[0] = 0;
-
-	for (i = 1; i < vlNumMacroBlockExTypes; ++i)
-		offset[i] = offset[i - 1] + num_macroblocks[i - 1];
-
-	{
-		struct vlMacroBlockVertexStream0	*ycbcr_vb;
-		struct vlVertex2f			*ref_vb[2];
-
-		ycbcr_vb = (struct vlMacroBlockVertexStream0*)pipe_buffer_map
-		(
-			pipe->screen,
-			mc->vertex_bufs.ycbcr.buffer,
-			PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
-		);
-
-		for (i = 0; i < 2; ++i)
-			ref_vb[i] = (struct vlVertex2f*)pipe_buffer_map
-			(
-				pipe->screen,
-				mc->vertex_bufs.ref[i].buffer,
-				PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
-			);
-
-		for (i = 0; i < mc->num_macroblocks; ++i)
-		{
-			enum vlMacroBlockTypeEx mb_type_ex = vlGetMacroBlockTypeEx(&mc->macroblocks[i]);
-
-			vlGenMacroblockVerts(mc, &mc->macroblocks[i], offset[mb_type_ex], ycbcr_vb, ref_vb);
-
-			offset[mb_type_ex]++;
-		}
-
-		pipe_buffer_unmap(pipe->screen, mc->vertex_bufs.ycbcr.buffer);
-		for (i = 0; i < 2; ++i)
-			pipe_buffer_unmap(pipe->screen, mc->vertex_bufs.ref[i].buffer);
-	}
-
-	for (i = 0; i < 3; ++i)
-	{
-		pipe->screen->transfer_unmap(pipe->screen, mc->tex_transfer[i]);
-		pipe->screen->tex_transfer_destroy(mc->tex_transfer[i]);
-	}
-
-	mc->render_target.cbufs[0] = pipe->screen->get_tex_surface
-	(
-		pipe->screen,
-		mc->buffered_surface->texture,
-		0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
-	);
-
-	pipe->set_framebuffer_state(pipe, &mc->render_target);
-	pipe->set_viewport_state(pipe, &mc->viewport);
-	vs_consts = pipe_buffer_map
-	(
-		pipe->screen,
-		mc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
-	);
-
-	vs_consts->denorm.x = mc->buffered_surface->texture->width[0];
-	vs_consts->denorm.y = mc->buffered_surface->texture->height[0];
-
-	pipe_buffer_unmap(pipe->screen, mc->vs_const_buf.buffer);
-	pipe->set_constant_buffer(pipe, PIPE_SHADER_VERTEX, 0, &mc->vs_const_buf);
-	pipe->set_constant_buffer(pipe, PIPE_SHADER_FRAGMENT, 0, &mc->fs_const_buf);
-
-	if (num_macroblocks[vlMacroBlockExTypeIntra] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 1, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 4, mc->vertex_elems);
-		pipe->set_sampler_textures(pipe, 3, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 3, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->i_vs);
-		pipe->bind_fs_state(pipe, mc->i_fs);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeIntra] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeIntra] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures.ref[0] = mc->past_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 4, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->p_vs[0]);
-		pipe->bind_fs_state(pipe, mc->p_fs[0]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeFwdPredictedField] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures.ref[0] = mc->past_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 4, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->p_vs[1]);
-		pipe->bind_fs_state(pipe, mc->p_fs[1]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeFwdPredictedField] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeFwdPredictedField] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures.ref[0] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 4, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->p_vs[0]);
-		pipe->bind_fs_state(pipe, mc->p_fs[0]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures.ref[0] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 4, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->p_vs[1]);
-		pipe->bind_fs_state(pipe, mc->p_fs[1]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 8, mc->vertex_elems);
-		mc->textures.ref[0] = mc->past_surface->texture;
-		mc->textures.ref[1] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 5, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 5, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->b_vs[0]);
-		pipe->bind_fs_state(pipe, mc->b_fs[0]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] * 24;
-	}
-
-	if (num_macroblocks[vlMacroBlockExTypeBiPredictedField] > 0)
-	{
-		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs.all);
-		pipe->set_vertex_elements(pipe, 8, mc->vertex_elems);
-		mc->textures.ref[0] = mc->past_surface->texture;
-		mc->textures.ref[1] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 5, mc->textures.all);
-		pipe->bind_sampler_states(pipe, 5, mc->samplers.all);
-		pipe->bind_vs_state(pipe, mc->b_vs[1]);
-		pipe->bind_fs_state(pipe, mc->b_fs[1]);
-
-		pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, vb_start, num_macroblocks[vlMacroBlockExTypeBiPredictedField] * 24);
-		vb_start += num_macroblocks[vlMacroBlockExTypeBiPredictedField] * 24;
-	}
-
-	pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &mc->buffered_surface->render_fence);
-	pipe_surface_reference(&mc->render_target.cbufs[0], NULL);
-
-	for (i = 0; i < 3; ++i)
-		mc->zero_block[i].x = -1.0f;
-
-	mc->buffered_surface = NULL;
-	mc->num_macroblocks = 0;
-
-	return 0;
-}
-
-static int vlRenderMacroBlocksMpeg2R16SnormBuffered
-(
-	struct vlRender *render,
-	struct vlMpeg2MacroBlockBatch *batch,
-	struct vlSurface *surface
-)
-{
-	struct vlR16SnormBufferedMC	*mc;
-	bool				new_surface = false;
-	unsigned int			i;
-
-	assert(render);
-
-	mc = (struct vlR16SnormBufferedMC*)render;
-
-	if (mc->buffered_surface)
-	{
-		if (mc->buffered_surface != surface)
-		{
-			vlFlush(&mc->base);
-			new_surface = true;
-		}
-	}
-	else
-		new_surface = true;
-
-	if (new_surface)
-	{
-		mc->buffered_surface = surface;
-		mc->past_surface = batch->past_surface;
-		mc->future_surface = batch->future_surface;
-		mc->surface_tex_inv_size.x = 1.0f / surface->texture->width[0];
-		mc->surface_tex_inv_size.y = 1.0f / surface->texture->height[0];
-
-		for (i = 0; i < 3; ++i)
-		{
-			mc->tex_transfer[i] = mc->pipe->screen->get_tex_transfer
-			(
-				mc->pipe->screen,
-				mc->textures.all[i],
-				0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0,
-				surface->texture->width[0],
-				surface->texture->height[0]
-			);
-
-			mc->texels[i] = mc->pipe->screen->transfer_map(mc->pipe->screen, mc->tex_transfer[i]);
-		}
-	}
-
-	for (i = 0; i < batch->num_macroblocks; ++i)
-		vlGrabMacroBlock(mc, &batch->macroblocks[i]);
-
-	return 0;
-}
-
-static inline int vlEnd
-(
-	struct vlRender *render
-)
-{
-	assert(render);
-
-	return 0;
-}
-
-static int vlDestroy
-(
-	struct vlRender *render
-)
-{
-	struct vlR16SnormBufferedMC	*mc;
-	struct pipe_context		*pipe;
-	unsigned int			i;
-
-	assert(render);
-
-	mc = (struct vlR16SnormBufferedMC*)render;
-	pipe = mc->pipe;
-
-	for (i = 0; i < 5; ++i)
-		pipe->delete_sampler_state(pipe, mc->samplers.all[i]);
-
-	for (i = 0; i < 3; ++i)
-		pipe_buffer_reference(&mc->vertex_bufs.all[i].buffer, NULL);
-
-	/* Textures 3 & 4 are not created directly, no need to release them here */
-	for (i = 0; i < 3; ++i)
-		pipe_texture_reference(&mc->textures.all[i], NULL);
-
-	pipe->delete_vs_state(pipe, mc->i_vs);
-	pipe->delete_fs_state(pipe, mc->i_fs);
-
-	for (i = 0; i < 2; ++i)
-	{
-		pipe->delete_vs_state(pipe, mc->p_vs[i]);
-		pipe->delete_fs_state(pipe, mc->p_fs[i]);
-		pipe->delete_vs_state(pipe, mc->b_vs[i]);
-		pipe->delete_fs_state(pipe, mc->b_fs[i]);
-	}
-
-	pipe_buffer_reference(&mc->vs_const_buf.buffer, NULL);
-	pipe_buffer_reference(&mc->fs_const_buf.buffer, NULL);
-
-	FREE(mc->macroblocks);
-	FREE(mc);
-
-	return 0;
-}
-
-/*
- * Muliplier renormalizes block samples from 16 bits to 12 bits.
- * Divider is used when calculating Y % 2 for choosing top or bottom
- * field for P or B macroblocks.
- * TODO: Use immediates.
- */
-static const struct vlFragmentShaderConsts fs_consts =
-{
-	{32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
-	{0.5f, 2.0f, 0.0f, 0.0f}
-};
-
-#include "vl_r16snorm_mc_buf_shaders.inc"
-
-static int vlCreateDataBufs
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int	mbw = align(mc->picture_width, VL_MACROBLOCK_WIDTH) / VL_MACROBLOCK_WIDTH;
-	const unsigned int	mbh = align(mc->picture_height, VL_MACROBLOCK_HEIGHT) / VL_MACROBLOCK_HEIGHT;
-
-	struct pipe_context	*pipe;
-	unsigned int		i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	mc->macroblocks_per_picture = mbw * mbh;
-
-	/* Create our vertex buffers */
-	mc->vertex_bufs.ycbcr.stride = sizeof(struct vlVertex2f) * 4;
-	mc->vertex_bufs.ycbcr.max_index = 24 * mc->macroblocks_per_picture - 1;
-	mc->vertex_bufs.ycbcr.buffer_offset = 0;
-	mc->vertex_bufs.ycbcr.buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		DEFAULT_BUF_ALIGNMENT,
-		PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
-		sizeof(struct vlVertex2f) * 4 * 24 * mc->macroblocks_per_picture
-	);
-
-	for (i = 1; i < 3; ++i)
-	{
-		mc->vertex_bufs.all[i].stride = sizeof(struct vlVertex2f) * 2;
-		mc->vertex_bufs.all[i].max_index = 24 * mc->macroblocks_per_picture - 1;
-		mc->vertex_bufs.all[i].buffer_offset = 0;
-		mc->vertex_bufs.all[i].buffer = pipe_buffer_create
-		(
-			pipe->screen,
-			DEFAULT_BUF_ALIGNMENT,
-			PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
-			sizeof(struct vlVertex2f) * 2 * 24 * mc->macroblocks_per_picture
-		);
-	}
-
-	/* Position element */
-	mc->vertex_elems[0].src_offset = 0;
-	mc->vertex_elems[0].vertex_buffer_index = 0;
-	mc->vertex_elems[0].nr_components = 2;
-	mc->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Luma, texcoord element */
-	mc->vertex_elems[1].src_offset = sizeof(struct vlVertex2f);
-	mc->vertex_elems[1].vertex_buffer_index = 0;
-	mc->vertex_elems[1].nr_components = 2;
-	mc->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Chroma Cr texcoord element */
-	mc->vertex_elems[2].src_offset = sizeof(struct vlVertex2f) * 2;
-	mc->vertex_elems[2].vertex_buffer_index = 0;
-	mc->vertex_elems[2].nr_components = 2;
-	mc->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Chroma Cb texcoord element */
-	mc->vertex_elems[3].src_offset = sizeof(struct vlVertex2f) * 3;
-	mc->vertex_elems[3].vertex_buffer_index = 0;
-	mc->vertex_elems[3].nr_components = 2;
-	mc->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* First ref surface top field texcoord element */
-	mc->vertex_elems[4].src_offset = 0;
-	mc->vertex_elems[4].vertex_buffer_index = 1;
-	mc->vertex_elems[4].nr_components = 2;
-	mc->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* First ref surface bottom field texcoord element */
-	mc->vertex_elems[5].src_offset = sizeof(struct vlVertex2f);
-	mc->vertex_elems[5].vertex_buffer_index = 1;
-	mc->vertex_elems[5].nr_components = 2;
-	mc->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Second ref surface top field texcoord element */
-	mc->vertex_elems[6].src_offset = 0;
-	mc->vertex_elems[6].vertex_buffer_index = 2;
-	mc->vertex_elems[6].nr_components = 2;
-	mc->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Second ref surface bottom field texcoord element */
-	mc->vertex_elems[7].src_offset = sizeof(struct vlVertex2f);
-	mc->vertex_elems[7].vertex_buffer_index = 2;
-	mc->vertex_elems[7].nr_components = 2;
-	mc->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
-
-	/* Create our constant buffer */
-	mc->vs_const_buf.buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		DEFAULT_BUF_ALIGNMENT,
-		PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
-		sizeof(struct vlVertexShaderConsts)
-	);
-
-	mc->fs_const_buf.buffer = pipe_buffer_create
-	(
-		pipe->screen,
-		DEFAULT_BUF_ALIGNMENT,
-		PIPE_BUFFER_USAGE_CONSTANT,
-		sizeof(struct vlFragmentShaderConsts)
-	);
-
-	memcpy
-	(
-		pipe_buffer_map(pipe->screen, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		&fs_consts,
-		sizeof(struct vlFragmentShaderConsts)
-	);
-
-	pipe_buffer_unmap(pipe->screen, mc->fs_const_buf.buffer);
-
-	mc->macroblocks = MALLOC(sizeof(struct vlMpeg2MacroBlock) * mc->macroblocks_per_picture);
-
-	return 0;
-}
-
-static int vlInit
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	struct pipe_context		*pipe;
-	struct pipe_sampler_state	sampler;
-	struct pipe_texture		template;
-	unsigned int			filters[5];
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-
-	mc->buffered_surface = NULL;
-	mc->past_surface = NULL;
-	mc->future_surface = NULL;
-	for (i = 0; i < 3; ++i)
-		mc->zero_block[i].x = -1.0f;
-	mc->num_macroblocks = 0;
-
-	/* For MC we render to textures, which are rounded up to nearest POT */
-	mc->viewport.scale[0] = vlRoundUpPOT(mc->picture_width);
-	mc->viewport.scale[1] = vlRoundUpPOT(mc->picture_height);
-	mc->viewport.scale[2] = 1;
-	mc->viewport.scale[3] = 1;
-	mc->viewport.translate[0] = 0;
-	mc->viewport.translate[1] = 0;
-	mc->viewport.translate[2] = 0;
-	mc->viewport.translate[3] = 0;
-
-	mc->render_target.width = vlRoundUpPOT(mc->picture_width);
-	mc->render_target.height = vlRoundUpPOT(mc->picture_height);
-	mc->render_target.nr_cbufs = 1;
-	/* FB for MC stage is a vlSurface created by the user, set at render time */
-	mc->render_target.zsbuf = NULL;
-
-	filters[0] = PIPE_TEX_FILTER_NEAREST;
-	/* FIXME: Linear causes discoloration around block edges */
-	filters[1] = /*mc->picture_format == vlFormatYCbCr444 ?*/ PIPE_TEX_FILTER_NEAREST /*: PIPE_TEX_FILTER_LINEAR*/;
-	filters[2] = /*mc->picture_format == vlFormatYCbCr444 ?*/ PIPE_TEX_FILTER_NEAREST /*: PIPE_TEX_FILTER_LINEAR*/;
-	filters[3] = PIPE_TEX_FILTER_LINEAR;
-	filters[4] = PIPE_TEX_FILTER_LINEAR;
-
-	for (i = 0; i < 5; ++i)
-	{
-		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_img_filter = filters[i];
-		sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
-		sampler.mag_img_filter = filters[i];
-		sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
-		sampler.compare_func = PIPE_FUNC_ALWAYS;
-		sampler.normalized_coords = 1;
-		/*sampler.prefilter = ;*/
-		/*sampler.lod_bias = ;*/
-		sampler.min_lod = 0;
-		/*sampler.max_lod = ;*/
-		/*sampler.border_color[i] = ;*/
-		/*sampler.max_anisotropy = ;*/
-		mc->samplers.all[i] = pipe->create_sampler_state(pipe, &sampler);
-	}
-
-	memset(&template, 0, sizeof(struct pipe_texture));
-	template.target = PIPE_TEXTURE_2D;
-	template.format = PIPE_FORMAT_R16_SNORM;
-	template.last_level = 0;
-	template.width[0] = vlRoundUpPOT(mc->picture_width);
-	template.height[0] = vlRoundUpPOT(mc->picture_height);
-	template.depth[0] = 1;
-	pf_get_block(template.format, &template.block);
-	template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC;
-
-	mc->textures.y = pipe->screen->texture_create(pipe->screen, &template);
-
-	if (mc->picture_format == vlFormatYCbCr420)
-	{
-		template.width[0] = vlRoundUpPOT(mc->picture_width / 2);
-		template.height[0] = vlRoundUpPOT(mc->picture_height / 2);
-	}
-	else if (mc->picture_format == vlFormatYCbCr422)
-		template.height[0] = vlRoundUpPOT(mc->picture_height / 2);
-
-	mc->textures.cb = pipe->screen->texture_create(pipe->screen, &template);
-	mc->textures.cr = pipe->screen->texture_create(pipe->screen, &template);
-
-	/* textures.all[3] & textures.all[4] are assigned from vlSurfaces for P and B macroblocks at render time */
-
-	vlCreateVertexShaderIMB(mc);
-	vlCreateFragmentShaderIMB(mc);
-	vlCreateVertexShaderFramePMB(mc);
-	vlCreateVertexShaderFieldPMB(mc);
-	vlCreateFragmentShaderFramePMB(mc);
-	vlCreateFragmentShaderFieldPMB(mc);
-	vlCreateVertexShaderFrameBMB(mc);
-	vlCreateVertexShaderFieldBMB(mc);
-	vlCreateFragmentShaderFrameBMB(mc);
-	vlCreateFragmentShaderFieldBMB(mc);
-	vlCreateDataBufs(mc);
-
-	return 0;
-}
-
-int vlCreateR16SNormBufferedMC
-(
-	struct pipe_context *pipe,
-	unsigned int picture_width,
-	unsigned int picture_height,
-	enum vlFormat picture_format,
-	struct vlRender **render
-)
-{
-	struct vlR16SnormBufferedMC *mc;
-
-	assert(pipe);
-	assert(render);
-
-	mc = CALLOC_STRUCT(vlR16SnormBufferedMC);
-
-	mc->base.vlBegin = &vlBegin;
-	mc->base.vlRenderMacroBlocksMpeg2 = &vlRenderMacroBlocksMpeg2R16SnormBuffered;
-	mc->base.vlEnd = &vlEnd;
-	mc->base.vlFlush = &vlFlush;
-	mc->base.vlDestroy = &vlDestroy;
-	mc->pipe = pipe;
-	mc->picture_width = picture_width;
-	mc->picture_height = picture_height;
-
-	vlInit(mc);
-
-	*render = &mc->base;
-
-	return 0;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h
deleted file mode 100644
index 27177d64ca..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.h
+++ /dev/null
@@ -1,18 +0,0 @@
-#ifndef vl_r16snorm_mc_buf_h
-#define vl_r16snorm_mc_buf_h
-
-#include "vl_types.h"
-
-struct pipe_context;
-struct vlRender;
-
-int vlCreateR16SNormBufferedMC
-(
-	struct pipe_context *pipe,
-	unsigned int picture_width,
-	unsigned int picture_height,
-	enum vlFormat picture_format,
-	struct vlRender **render
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc
deleted file mode 100644
index 34d93e1df0..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc
+++ /dev/null
@@ -1,1185 +0,0 @@
-static int vlCreateVertexShaderIMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 50;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Luma texcoords
-	 * decl i2		; Chroma Cb texcoords
-	 * decl i3		; Chroma Cr texcoords
-	 */
-	for (i = 0; i < 4; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma Cb texcoords
-	 * decl o3		; Chroma Cr texcoords
-	 */
-	for (i = 0; i < 4; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * mov o0, i0		; Move input vertex pos to output
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma Cb texcoords to output
-	 * mov o3, i3		; Move input chroma Cr texcoords to output
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	mc->i_vs = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderIMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0			; Luma texcoords
-	 * decl i1			; Chroma Cb texcoords
-	 * decl i2			; Chroma Cr texcoords
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0			; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0, t1 */
-	decl = vl_decl_temps(0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl s0			; Sampler for luma texture
-	 * decl s1			; Sampler for chroma Cb texture
-	 * decl s2			; Sampler for chroma Cr texture
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		decl = vl_decl_samplers(i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header,max_tokens - ti);
-	}
-
-	/*
-	 * tex2d t1, i0, s0		; Read texel from luma texture
-	 * mov t0.x, t1.x		; Move luma sample into .x component
-	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
-	 * mov t0.y, t1.x		; Move Cb sample into .y component
-	 * tex2d t1, i2, s2		; Read texel from chroma Cr texture
-	 * mov t0.z, t1.x		; Move Cr sample into .z component
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul o0, t0, c0		; Rescale texel to correct range */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	mc->i_fs = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateVertexShaderFramePMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Luma texcoords
-	 * decl i2		; Chroma Cb texcoords
-	 * decl i3		; Chroma Cr texcoords
-	 * decl i4		; Ref surface top field texcoords
-	 * decl i5		; Ref surface bottom field texcoords (unused, packed in the same stream)
-	 */
-	for (i = 0; i < 6; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma Cb texcoords
-	 * decl o3		; Chroma Cr texcoords
-	 * decl o4		; Ref macroblock texcoords
-	 */
-	for (i = 0; i < 5; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * mov o0, i0		; Move input vertex pos to output
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma Cb texcoords to output
-	 * mov o3, i3		; Move input chroma Cr texcoords to output
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* add o4, i0, i4	; Translate vertex pos by motion vec to form ref macroblock texcoords */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, 4);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	mc->p_vs[0] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateVertexShaderFieldPMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Luma texcoords
-	 * decl i2		; Chroma Cb texcoords
-	 * decl i3		; Chroma Cr texcoords
-	 * decl i4              ; Ref macroblock top field texcoords
-	 * decl i5              ; Ref macroblock bottom field texcoords
-	 */
-	for (i = 0; i < 6; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl c0		; Render target dimensions */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma Cb texcoords
-	 * decl o3		; Chroma Cr texcoords
-	 * decl o4		; Ref macroblock top field texcoords
-	 * decl o5		; Ref macroblock bottom field texcoords
-	 * decl o6		; Denormalized vertex pos
-	 */
-	for (i = 0; i < 7; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * mov o0, i0		; Move input vertex pos to output
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma Cb texcoords to output
-	 * mov o3, i3		; Move input chroma Cr texcoords to output
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * add o4, i0, i4	; Translate vertex pos by motion vec to form top field macroblock texcoords
-	 * add o5, i0, i5	; Translate vertex pos by motion vec to form bottom field macroblock texcoords
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, i + 4);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul o6, i0, c0	; Denorm vertex pos */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 6, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	mc->p_vs[1] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFramePMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0			; Luma texcoords
-	 * decl i1			; Chroma Cb texcoords
-	 * decl i2			; Chroma Cr texcoords
-	 * decl i3			; Ref macroblock texcoords
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0			; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0, t1 */
-	decl = vl_decl_temps(0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl s0			; Sampler for luma texture
-	 * decl s1			; Sampler for chroma Cb texture
-	 * decl s2			; Sampler for chroma Cr texture
-	 * decl s3			; Sampler for ref surface texture
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		decl = vl_decl_samplers(i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * tex2d t1, i0, s0		; Read texel from luma texture
-	 * mov t0.x, t1.x		; Move luma sample into .x component
-	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
-	 * mov t0.y, t1.x		; Move Cb sample into .y component
-	 * tex2d t1, i2, s2		; Read texel from chroma Cr texture
-	 * mov t0.z, t1.x		; Move Cr sample into .z component
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul t0, t0, c0		; Rescale texel to correct range */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* tex2d t1, i3, s3		; Read texel from ref macroblock */
-	inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 3, TGSI_FILE_SAMPLER, 3);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, t1		; Add ref and differential to form final output */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	mc->p_fs[0] = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFieldPMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 200;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0			; Luma texcoords
-	 * decl i1			; Chroma Cb texcoords
-	 * decl i2			; Chroma Cr texcoords
-	 * decl i3			; Ref macroblock top field texcoords
-	 * decl i4			; Ref macroblock bottom field texcoords
-	 * decl i5			; Denormalized vertex pos
-	 */
-	for (i = 0; i < 6; ++i)
-	{
-		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
-	 * decl c1			; Constants 1/2 & 2 in .x, .y channels for Y-mod-2 top/bottom field selection
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0			; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0-t4 */
-	decl = vl_decl_temps(0, 4);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl s0			; Sampler for luma texture
-	 * decl s1			; Sampler for chroma Cb texture
-	 * decl s2			; Sampler for chroma Cr texture
-	 * decl s3			; Sampler for ref surface texture
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		decl = vl_decl_samplers(i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * tex2d t1, i0, s0		; Read texel from luma texture
-	 * mov t0.x, t1.x		; Move luma sample into .x component
-	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
-	 * mov t0.y, t1.x		; Move Cb sample into .y component
-	 * tex2d t1, i2, s2		; Read texel from chroma Cr texture
-	 * mov t0.z, t1.x		; Move Cr sample into .z component
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul t0, t0, c0		; Rescale texel to correct range */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * tex2d t1, i3, s3		; Read texel from ref macroblock top field
-	 * tex2d t2, i4, s3		; Read texel from ref macroblock bottom field
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, 3);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* XXX: Pos values off by 0.5? */
-	/* sub t4, i5.y, c1.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 5, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t3, t4, c1.x		; Multiply pos Y-coord by 1/2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* floor t3, t3			; Get rid of fractional part */
-	inst = vl_inst2(TGSI_OPCODE_FLR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t3, t3, c1.y		; Multiply by 2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* sub t3, t4, t3		; Subtract from original Y to get Y % 2 */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
-	/* lerp t1, t3, t1, t2		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, t1		; Add ref and differential to form final output */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	mc->p_fs[1] = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateVertexShaderFrameBMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Luma texcoords
-	 * decl i2		; Chroma Cb texcoords
-	 * decl i3		; Chroma Cr texcoords
-	 * decl i4              ; First ref macroblock top field texcoords
-	 * decl i5              ; First ref macroblock bottom field texcoords (unused, packed in the same stream)
-	 * decl i6		; Second ref macroblock top field texcoords
-	 * decl i7		; Second ref macroblock bottom field texcoords (unused, packed in the same stream)
-	 */
-	for (i = 0; i < 8; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma Cb texcoords
-	 * decl o3		; Chroma Cr texcoords
-	 * decl o4		; First ref macroblock texcoords
-	 * decl o5		; Second ref macroblock texcoords
-	 */
-	for (i = 0; i < 6; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * mov o0, i0		; Move input vertex pos to output
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma Cb texcoords to output
-	 * mov o3, i3		; Move input chroma Cr texcoords to output
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * add o4, i0, i4	; Translate vertex pos by motion vec to form first ref macroblock texcoords
-	 * add o5, i0, i6	; Translate vertex pos by motion vec to form second ref macroblock texcoords
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, (i + 2) * 2);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	mc->b_vs[0] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateVertexShaderFieldBMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	vs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0		; Vertex pos
-	 * decl i1		; Luma texcoords
-	 * decl i2		; Chroma Cb texcoords
-	 * decl i3		; Chroma Cr texcoords
-	 * decl i4              ; First ref macroblock top field texcoords
-	 * decl i5              ; First ref macroblock bottom field texcoords
-	 * decl i6              ; Second ref macroblock top field texcoords
-	 * decl i7              ; Second ref macroblock bottom field texcoords
-	 */
-	for (i = 0; i < 8; i++)
-	{
-		decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl c0		; Render target dimensions */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma Cb texcoords
-	 * decl o3		; Chroma Cr texcoords
-	 * decl o4		; First ref macroblock top field texcoords
-	 * decl o5		; First ref macroblock Bottom field texcoords
-	 * decl o6		; Second ref macroblock top field texcoords
-	 * decl o7		; Second ref macroblock Bottom field texcoords
-	 * decl o8		; Denormalized vertex pos
-	 */
-	for (i = 0; i < 9; i++)
-	{
-		decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* decl t0, t1 */
-	decl = vl_decl_temps(0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * mov o0, i0		; Move input vertex pos to output
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma Cb texcoords to output
-	 * mov o3, i3		; Move input chroma Cr texcoords to output
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * add o4, i0, i4	; Translate vertex pos by motion vec to form first top field macroblock texcoords
-	 * add o5, i0, i5	; Translate vertex pos by motion vec to form first bottom field macroblock texcoords
-	 * add o6, i0, i6	; Translate vertex pos by motion vec to form second top field macroblock texcoords
-	 * add o7, i0, i7	; Translate vertex pos by motion vec to form second bottom field macroblock texcoords
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, i + 4);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul o8, i0, c0	; Denorm vertex pos */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 8, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	vs.tokens = tokens;
-	mc->b_vs[1] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFrameBMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 100;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0			; Luma texcoords
-	 * decl i1			; Chroma Cb texcoords
-	 * decl i2			; Chroma Cr texcoords
-	 * decl i3			; First ref macroblock texcoords
-	 * decl i4			; Second ref macroblock texcoords
-	 */
-	for (i = 0; i < 5; ++i)
-	{
-		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
-	 * decl c1			; Constant 1/2 in .x channel to use as weight to blend past and future texels
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0			; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0-t2 */
-	decl = vl_decl_temps(0, 2);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl s0			; Sampler for luma texture
-	 * decl s1			; Sampler for chroma Cb texture
-	 * decl s2			; Sampler for chroma Cr texture
-	 * decl s3			; Sampler for first ref surface texture
-	 * decl s4			; Sampler for second ref surface texture
-	 */
-	for (i = 0; i < 5; ++i)
-	{
-		decl = vl_decl_samplers(i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * tex2d t1, i0, s0		; Read texel from luma texture
-	 * mov t0.x, t1.x		; Move luma sample into .x component
-	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
-	 * mov t0.y, t1.x		; Move Cb sample into .y component
-	 * tex2d t1, i2, s2		; Read texel from chroma Cr texture
-	 * mov t0.z, t1.x		; Move Cr sample into .z component
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul t0, t0, c0		; Rescale texel to correct range */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * tex2d t1, i3, s3		; Read texel from first ref macroblock
-	 * tex2d t2, i4, s4		; Read texel from second ref macroblock
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, i + 3);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, t1		; Add past/future ref and differential to form final output */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	mc->b_fs[0] = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFieldBMB
-(
-	struct vlR16SnormBufferedMC *mc
-)
-{
-	const unsigned int		max_tokens = 200;
-
-	struct pipe_context		*pipe;
-	struct pipe_shader_state	fs;
-	struct tgsi_token		*tokens;
-	struct tgsi_header		*header;
-
-	struct tgsi_full_declaration	decl;
-	struct tgsi_full_instruction	inst;
-
-	unsigned int			ti;
-	unsigned int			i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
-
-	/* Version */
-	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
-	/* Header */
-	header = (struct tgsi_header*)&tokens[1];
-	*header = tgsi_build_header();
-	/* Processor */
-	*(struct tgsi_processor*)&tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
-
-	ti = 3;
-
-	/*
-	 * decl i0			; Luma texcoords
-	 * decl i1			; Chroma Cb texcoords
-	 * decl i2			; Chroma Cr texcoords
-	 * decl i3			; First ref macroblock top field texcoords
-	 * decl i4			; First ref macroblock bottom field texcoords
-	 * decl i5			; Second ref macroblock top field texcoords
-	 * decl i6			; Second ref macroblock bottom field texcoords
-	 * decl i7			; Denormalized vertex pos
-	 */
-	for (i = 0; i < 8; ++i)
-	{
-		decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * decl c0			; Scaling factor, rescales 16-bit snorm to 9-bit snorm
-	 * decl c1			; Constants 1/2 & 2 in .x, .y channels to use as weight to blend past and future texels
-	 *				; and for Y-mod-2 top/bottom field selection
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl o0			; Fragment color */
-	decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* decl t0-t5 */
-	decl = vl_decl_temps(0, 5);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl s0			; Sampler for luma texture
-	 * decl s1			; Sampler for chroma Cb texture
-	 * decl s2			; Sampler for chroma Cr texture
-	 * decl s3			; Sampler for first ref surface texture
-	 * decl s4			; Sampler for second ref surface texture
-	 */
-	for (i = 0; i < 5; ++i)
-	{
-		decl = vl_decl_samplers(i, i);
-		ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/*
-	 * tex2d t1, i0, s0		; Read texel from luma texture
-	 * mov t0.x, t1.x		; Move luma sample into .x component
-	 * tex2d t1, i1, s1		; Read texel from chroma Cb texture
-	 * mov t0.y, t1.x		; Move Cb sample into .y component
-	 * tex2d t1, i2, s2		; Read texel from chroma Cr texture
-	 * mov t0.z, t1.x		; Move Cr sample into .z component
-	 */
-	for (i = 0; i < 3; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-		inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-		inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-		inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul t0, t0, c0		; Rescale texel to correct range */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* XXX: Pos values off by 0.5? */
-	/* sub t4, i7.y, c1.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 7, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t3, t4, c1.x		; Multiply pos Y-coord by 1/2 */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* floor t3, t3			; Get rid of fractional part */
-	inst = vl_inst2(TGSI_OPCODE_FLR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t3, t3, c1.y		; Multiply by 2 */
-	inst = vl_inst3( TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_CONSTANT, 1);
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
-	inst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* sub t3, t4, t3		; Subtract from original Y to get Y % 2 */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 3);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * tex2d t1, i3, s3		; Read texel from past ref macroblock top field
-	 * tex2d t2, i4, s3		; Read texel from past ref macroblock bottom field
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, 3);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
-	/* lerp t1, t3, t1, t2		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * tex2d t4, i5, s4		; Read texel from future ref macroblock top field
-	 * tex2d t5, i6, s4		; Read texel from future ref macroblock bottom field
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 4, TGSI_FILE_INPUT, i + 5, TGSI_FILE_SAMPLER, 4);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
-	/* lerp t2, t3, t4, t5		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
-	inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, t1		; Add past/future ref and differential to form final output */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* end */
-	inst = vl_end();
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	fs.tokens = tokens;
-	mc->b_fs[1] = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
-
-	return 0;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_render.h b/src/gallium/state_trackers/g3dvl/vl_render.h
deleted file mode 100644
index 166030b498..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_render.h
+++ /dev/null
@@ -1,38 +0,0 @@
-#ifndef vl_render_h
-#define vl_render_h
-
-#include "vl_types.h"
-
-struct pipe_surface;
-
-struct vlRender
-{
-	int (*vlBegin)
-	(
-		struct vlRender *render
-	);
-
-	int (*vlRenderMacroBlocksMpeg2)
-	(
-		struct vlRender *render,
-		struct vlMpeg2MacroBlockBatch *batch,
-		struct vlSurface *surface
-	);
-
-	int (*vlEnd)
-	(
-		struct vlRender *render
-	);
-
-	int (*vlFlush)
-	(
-		struct vlRender *render
-	);
-
-	int (*vlDestroy)
-	(
-		struct vlRender *render
-	);
-};
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_screen.c b/src/gallium/state_trackers/g3dvl/vl_screen.c
deleted file mode 100644
index ade8643a66..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_screen.c
+++ /dev/null
@@ -1,115 +0,0 @@
-#define VL_INTERNAL
-#include "vl_screen.h"
-#include <assert.h>
-#include <util/u_memory.h>
-
-int vlCreateScreen
-(
-	struct vlDisplay *display,
-	int screen,
-	struct pipe_screen *pscreen,
-	struct vlScreen **vl_screen
-)
-{
-	struct vlScreen *scrn;
-
-	assert(display);
-	assert(pscreen);
-	assert(vl_screen);
-
-	scrn = CALLOC_STRUCT(vlScreen);
-
-	if (!scrn)
-		return 1;
-
-	scrn->display = display;
-	scrn->ordinal = screen;
-	scrn->pscreen = pscreen;
-	*vl_screen = scrn;
-
-	return 0;
-}
-
-int vlDestroyScreen
-(
-	struct vlScreen *screen
-)
-{
-	assert(screen);
-
-	FREE(screen);
-
-	return 0;
-}
-
-struct vlDisplay* vlGetDisplay
-(
-	struct vlScreen *screen
-)
-{
-	assert(screen);
-
-	return screen->display;
-}
-
-struct pipe_screen* vlGetPipeScreen
-(
-	struct vlScreen *screen
-)
-{
-	assert(screen);
-
-	return screen->pscreen;
-}
-
-unsigned int vlGetMaxProfiles
-(
-	struct vlScreen *screen
-)
-{
-	assert(screen);
-
-	return vlProfileCount;
-}
-
-int vlQueryProfiles
-(
-	struct vlScreen *screen,
-	enum vlProfile *profiles
-)
-{
-	assert(screen);
-	assert(profiles);
-
-	profiles[0] = vlProfileMpeg2Simple;
-	profiles[1] = vlProfileMpeg2Main;
-
-	return 0;
-}
-
-unsigned int vlGetMaxEntryPoints
-(
-	struct vlScreen *screen
-)
-{
-	assert(screen);
-
-	return vlEntryPointCount;
-}
-
-int vlQueryEntryPoints
-(
-	struct vlScreen *screen,
-	enum vlProfile profile,
-	enum vlEntryPoint *entry_points
-)
-{
-	assert(screen);
-	assert(entry_points);
-
-	entry_points[0] = vlEntryPointIDCT;
-	entry_points[1] = vlEntryPointMC;
-	entry_points[2] = vlEntryPointCSC;
-
-	return 0;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_screen.h b/src/gallium/state_trackers/g3dvl/vl_screen.h
deleted file mode 100644
index 98f3d429b6..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_screen.h
+++ /dev/null
@@ -1,63 +0,0 @@
-#ifndef vl_screen_h
-#define vl_screen_h
-
-#include "vl_types.h"
-
-struct pipe_screen;
-
-#ifdef VL_INTERNAL
-struct vlScreen
-{
-	struct vlDisplay	*display;
-	unsigned int		ordinal;
-	struct pipe_screen	*pscreen;
-};
-#endif
-
-int vlCreateScreen
-(
-	struct vlDisplay *display,
-	int screen,
-	struct pipe_screen *pscreen,
-	struct vlScreen **vl_screen
-);
-
-int vlDestroyScreen
-(
-	struct vlScreen *screen
-);
-
-struct vlDisplay* vlGetDisplay
-(
-	struct vlScreen *screen
-);
-
-struct pipe_screen* vlGetPipeScreen
-(
-	struct vlScreen *screen
-);
-
-unsigned int vlGetMaxProfiles
-(
-	struct vlScreen *screen
-);
-
-int vlQueryProfiles
-(
-	struct vlScreen *screen,
-	enum vlProfile *profiles
-);
-
-unsigned int vlGetMaxEntryPoints
-(
-	struct vlScreen *screen
-);
-
-int vlQueryEntryPoints
-(
-	struct vlScreen *screen,
-	enum vlProfile profile,
-	enum vlEntryPoint *entry_points
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_shader_build.c b/src/gallium/state_trackers/g3dvl/vl_shader_build.c
deleted file mode 100644
index 51f1721a33..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_shader_build.c
+++ /dev/null
@@ -1,215 +0,0 @@
-#include "vl_shader_build.h"
-#include <assert.h>
-#include <tgsi/tgsi_parse.h>
-#include <tgsi/tgsi_build.h>
-
-struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	decl.Declaration.File = TGSI_FILE_INPUT;
-	decl.Declaration.Semantic = 1;
-	decl.Semantic.SemanticName = name;
-	decl.Semantic.SemanticIndex = index;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_declaration vl_decl_interpolated_input
-(
-	unsigned int name,
-	unsigned int index,
-	unsigned int first,
-	unsigned int last,
-	int interpolation
-)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	assert
-	(
-		interpolation == TGSI_INTERPOLATE_CONSTANT ||
-		interpolation == TGSI_INTERPOLATE_LINEAR ||
-		interpolation == TGSI_INTERPOLATE_PERSPECTIVE
-	);
-
-	decl.Declaration.File = TGSI_FILE_INPUT;
-	decl.Declaration.Semantic = 1;
-	decl.Semantic.SemanticName = name;
-	decl.Semantic.SemanticIndex = index;
-	decl.Declaration.Interpolate = interpolation;;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	decl.Declaration.File = TGSI_FILE_CONSTANT;
-	decl.Declaration.Semantic = 1;
-	decl.Semantic.SemanticName = name;
-	decl.Semantic.SemanticIndex = index;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	decl.Declaration.File = TGSI_FILE_OUTPUT;
-	decl.Declaration.Semantic = 1;
-	decl.Semantic.SemanticName = name;
-	decl.Semantic.SemanticIndex = index;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	decl = tgsi_default_full_declaration();
-	decl.Declaration.File = TGSI_FILE_TEMPORARY;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last)
-{
-	struct tgsi_full_declaration decl = tgsi_default_full_declaration();
-
-	decl = tgsi_default_full_declaration();
-	decl.Declaration.File = TGSI_FILE_SAMPLER;
-	decl.DeclarationRange.First = first;
-	decl.DeclarationRange.Last = last;
-
-	return decl;
-}
-
-struct tgsi_full_instruction vl_inst2
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src_file,
-	unsigned int src_index
-)
-{
-	struct tgsi_full_instruction inst = tgsi_default_full_instruction();
-
-	inst.Instruction.Opcode = opcode;
-	inst.Instruction.NumDstRegs = 1;
-	inst.FullDstRegisters[0].DstRegister.File = dst_file;
-	inst.FullDstRegisters[0].DstRegister.Index = dst_index;
-	inst.Instruction.NumSrcRegs = 1;
-	inst.FullSrcRegisters[0].SrcRegister.File = src_file;
-	inst.FullSrcRegisters[0].SrcRegister.Index = src_index;
-
-	return inst;
-}
-
-struct tgsi_full_instruction vl_inst3
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index
-)
-{
-	struct tgsi_full_instruction inst = tgsi_default_full_instruction();
-
-	inst.Instruction.Opcode = opcode;
-	inst.Instruction.NumDstRegs = 1;
-	inst.FullDstRegisters[0].DstRegister.File = dst_file;
-	inst.FullDstRegisters[0].DstRegister.Index = dst_index;
-	inst.Instruction.NumSrcRegs = 2;
-	inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
-	inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
-	inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
-	inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
-
-	return inst;
-}
-
-struct tgsi_full_instruction vl_tex
-(
-	int tex,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index
-)
-{
-	struct tgsi_full_instruction inst = tgsi_default_full_instruction();
-
-	inst.Instruction.Opcode = TGSI_OPCODE_TEX;
-	inst.Instruction.NumDstRegs = 1;
-	inst.FullDstRegisters[0].DstRegister.File = dst_file;
-	inst.FullDstRegisters[0].DstRegister.Index = dst_index;
-	inst.Instruction.NumSrcRegs = 2;
-	inst.InstructionExtTexture.Texture = tex;
-	inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
-	inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
-	inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
-	inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
-
-	return inst;
-}
-
-struct tgsi_full_instruction vl_inst4
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index,
-	enum tgsi_file_type src3_file,
-	unsigned int src3_index
-)
-{
-	struct tgsi_full_instruction inst = tgsi_default_full_instruction();
-
-	inst.Instruction.Opcode = opcode;
-	inst.Instruction.NumDstRegs = 1;
-	inst.FullDstRegisters[0].DstRegister.File = dst_file;
-	inst.FullDstRegisters[0].DstRegister.Index = dst_index;
-	inst.Instruction.NumSrcRegs = 3;
-	inst.FullSrcRegisters[0].SrcRegister.File = src1_file;
-	inst.FullSrcRegisters[0].SrcRegister.Index = src1_index;
-	inst.FullSrcRegisters[1].SrcRegister.File = src2_file;
-	inst.FullSrcRegisters[1].SrcRegister.Index = src2_index;
-	inst.FullSrcRegisters[2].SrcRegister.File = src3_file;
-	inst.FullSrcRegisters[2].SrcRegister.Index = src3_index;
-
-	return inst;
-}
-
-struct tgsi_full_instruction vl_end(void)
-{
-	struct tgsi_full_instruction inst = tgsi_default_full_instruction();
-
-	inst.Instruction.Opcode = TGSI_OPCODE_END;
-	inst.Instruction.NumDstRegs = 0;
-	inst.Instruction.NumSrcRegs = 0;
-
-	return inst;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_shader_build.h b/src/gallium/state_trackers/g3dvl/vl_shader_build.h
deleted file mode 100644
index dc615cb156..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_shader_build.h
+++ /dev/null
@@ -1,61 +0,0 @@
-#ifndef vl_shader_build_h
-#define vl_shader_build_h
-
-#include <pipe/p_shader_tokens.h>
-
-struct tgsi_full_declaration vl_decl_input(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
-struct tgsi_full_declaration vl_decl_interpolated_input
-(
-	unsigned int name,
-	unsigned int index,
-	unsigned int first,
-	unsigned int last,
-	int interpolation
-);
-struct tgsi_full_declaration vl_decl_constants(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
-struct tgsi_full_declaration vl_decl_output(unsigned int name, unsigned int index, unsigned int first, unsigned int last);
-struct tgsi_full_declaration vl_decl_temps(unsigned int first, unsigned int last);
-struct tgsi_full_declaration vl_decl_samplers(unsigned int first, unsigned int last);
-struct tgsi_full_instruction vl_inst2
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src_file,
-	unsigned int src_index
-);
-struct tgsi_full_instruction vl_inst3
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index
-);
-struct tgsi_full_instruction vl_tex
-(
-	int tex,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index
-);
-struct tgsi_full_instruction vl_inst4
-(
-	int opcode,
-	enum tgsi_file_type dst_file,
-	unsigned int dst_index,
-	enum tgsi_file_type src1_file,
-	unsigned int src1_index,
-	enum tgsi_file_type src2_file,
-	unsigned int src2_index,
-	enum tgsi_file_type src3_file,
-	unsigned int src3_index
-);
-struct tgsi_full_instruction vl_end(void);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c
deleted file mode 100644
index 7f60852cae..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_surface.c
+++ /dev/null
@@ -1,242 +0,0 @@
-#define VL_INTERNAL
-#include "vl_surface.h"
-#include <assert.h>
-#include <string.h>
-#include <pipe/p_screen.h>
-#include <pipe/p_state.h>
-#include <pipe/p_inlines.h>
-#include <util/u_memory.h>
-#include <vl_winsys.h>
-#include "vl_screen.h"
-#include "vl_context.h"
-#include "vl_render.h"
-#include "vl_csc.h"
-#include "vl_util.h"
-
-int vlCreateSurface
-(
-	struct vlScreen *screen,
-	unsigned int width,
-	unsigned int height,
-	enum vlFormat format,
-	struct vlSurface **surface
-)
-{
-	struct vlSurface	*sfc;
-	struct pipe_texture	template;
-
-	assert(screen);
-	assert(surface);
-
-	sfc = CALLOC_STRUCT(vlSurface);
-
-	if (!sfc)
-		return 1;
-
-	sfc->screen = screen;
-	sfc->width = width;
-	sfc->height = height;
-	sfc->format = format;
-
-	memset(&template, 0, sizeof(struct pipe_texture));
-	template.target = PIPE_TEXTURE_2D;
-	template.format = PIPE_FORMAT_A8R8G8B8_UNORM;
-	template.last_level = 0;
-	template.width[0] = vlRoundUpPOT(sfc->width);
-	template.height[0] = vlRoundUpPOT(sfc->height);
-	template.depth[0] = 1;
-	pf_get_block(template.format, &template.block);
-	template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET;
-
-	sfc->texture = vlGetPipeScreen(screen)->texture_create(vlGetPipeScreen(screen), &template);
-
-	if (!sfc->texture)
-	{
-		FREE(sfc);
-		return 1;
-	}
-
-	*surface = sfc;
-
-	return 0;
-}
-
-int vlDestroySurface
-(
-	struct vlSurface *surface
-)
-{
-	assert(surface);
-
-	pipe_texture_reference(&surface->texture, NULL);
-	FREE(surface);
-
-	return 0;
-}
-
-int vlRenderMacroBlocksMpeg2
-(
-	struct vlMpeg2MacroBlockBatch *batch,
-	struct vlSurface *surface
-)
-{
-	assert(batch);
-	assert(surface);
-	assert(surface->context);
-
-	surface->context->render->vlBegin(surface->context->render);
-
-	surface->context->render->vlRenderMacroBlocksMpeg2
-	(
-		surface->context->render,
-		batch,
-		surface
-	);
-
-	surface->context->render->vlEnd(surface->context->render);
-
-	return 0;
-}
-
-int vlPutPicture
-(
-	struct vlSurface *surface,
-	vlNativeDrawable drawable,
-	int srcx,
-	int srcy,
-	int srcw,
-	int srch,
-	int destx,
-	int desty,
-	int destw,
-	int desth,
-	int drawable_w,
-	int drawable_h,
-	enum vlPictureType picture_type
-)
-{
-	struct vlCSC		*csc;
-	struct pipe_context	*pipe;
-
-	assert(surface);
-	assert(surface->context);
-
-	surface->context->render->vlFlush(surface->context->render);
-
-	csc = surface->context->csc;
-	pipe = surface->context->pipe;
-
-	csc->vlResizeFrameBuffer(csc, drawable_w, drawable_h);
-
-	csc->vlBegin(csc);
-
-	csc->vlPutPicture
-	(
-		csc,
-		surface,
-		srcx,
-		srcy,
-		srcw,
-		srch,
-		destx,
-		desty,
-		destw,
-		desth,
-		picture_type
-	);
-
-	csc->vlEnd(csc);
-
-	pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &surface->disp_fence);
-
-	bind_pipe_drawable(pipe, drawable);
-
-	pipe->screen->flush_frontbuffer
-	(
-		pipe->screen,
-		csc->vlGetFrameBuffer(csc),
-		pipe->priv
-	);
-
-	return 0;
-}
-
-int vlSurfaceGetStatus
-(
-	struct vlSurface *surface,
-	enum vlResourceStatus *status
-)
-{
-	assert(surface);
-	assert(surface->context);
-	assert(status);
-
-	if (surface->render_fence && !surface->context->pipe->screen->fence_signalled(surface->context->pipe->screen, surface->render_fence, 0))
-	{
-		*status = vlResourceStatusRendering;
-		return 0;
-	}
-
-	if (surface->disp_fence && !surface->context->pipe->screen->fence_signalled(surface->context->pipe->screen, surface->disp_fence, 0))
-	{
-		*status = vlResourceStatusDisplaying;
-		return 0;
-	}
-
-	*status = vlResourceStatusFree;
-
-	return 0;
-}
-
-int vlSurfaceFlush
-(
-	struct vlSurface *surface
-)
-{
-	assert(surface);
-	assert(surface->context);
-
-	surface->context->render->vlFlush(surface->context->render);
-
-	return 0;
-}
-
-int vlSurfaceSync
-(
-	struct vlSurface *surface
-)
-{
-	assert(surface);
-	assert(surface->context);
-	assert(surface->render_fence);
-
-	surface->context->pipe->screen->fence_finish(surface->context->pipe->screen, surface->render_fence, 0);
-
-	return 0;
-}
-
-struct vlScreen* vlSurfaceGetScreen
-(
-	struct vlSurface *surface
-)
-{
-	assert(surface);
-
-	return surface->screen;
-}
-
-struct vlContext* vlBindToContext
-(
-	struct vlSurface *surface,
-	struct vlContext *context
-)
-{
-	struct vlContext *old;
-
-	assert(surface);
-
-	old = surface->context;
-	surface->context = context;
-
-	return old;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.h b/src/gallium/state_trackers/g3dvl/vl_surface.h
deleted file mode 100644
index 133e1515ef..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_surface.h
+++ /dev/null
@@ -1,86 +0,0 @@
-#ifndef vl_surface_h
-#define vl_surface_h
-
-#include "vl_types.h"
-
-#ifdef VL_INTERNAL
-struct pipe_texture;
-
-struct vlSurface
-{
-	struct vlScreen			*screen;
-	struct vlContext		*context;
-	unsigned int			width;
-	unsigned int			height;
-	enum vlFormat			format;
-	struct pipe_texture		*texture;
-	struct pipe_fence_handle	*render_fence;
-	struct pipe_fence_handle	*disp_fence;
-};
-#endif
-
-int vlCreateSurface
-(
-	struct vlScreen *screen,
-	unsigned int width,
-	unsigned int height,
-	enum vlFormat format,
-	struct vlSurface **surface
-);
-
-int vlDestroySurface
-(
-	struct vlSurface *surface
-);
-
-int vlRenderMacroBlocksMpeg2
-(
-	struct vlMpeg2MacroBlockBatch *batch,
-	struct vlSurface *surface
-);
-
-int vlPutPicture
-(
-	struct vlSurface *surface,
-	vlNativeDrawable drawable,
-	int srcx,
-	int srcy,
-	int srcw,
-	int srch,
-	int destx,
-	int desty,
-	int destw,
-	int desth,
-	int drawable_w,
-	int drawable_h,
-	enum vlPictureType picture_type
-);
-
-int vlSurfaceGetStatus
-(
-	struct vlSurface *surface,
-	enum vlResourceStatus *status
-);
-
-int vlSurfaceFlush
-(
-	struct vlSurface *surface
-);
-
-int vlSurfaceSync
-(
-	struct vlSurface *surface
-);
-
-struct vlScreen* vlSurfaceGetScreen
-(
-	struct vlSurface *surface
-);
-
-struct vlContext* vlBindToContext
-(
-	struct vlSurface *surface,
-	struct vlContext *context
-);
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_types.h b/src/gallium/state_trackers/g3dvl/vl_types.h
deleted file mode 100644
index 274e1f7437..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_types.h
+++ /dev/null
@@ -1,115 +0,0 @@
-#ifndef vl_types_h
-#define vl_types_h
-
-#if 1 /*#ifdef X11*/
-#include <X11/Xlib.h>
-
-typedef Display* vlNativeDisplay;
-typedef Drawable vlNativeDrawable;
-#endif
-
-struct vlDisplay;
-struct vlScreen;
-struct vlContext;
-struct vlSurface;
-
-enum vlResourceStatus
-{
-	vlResourceStatusFree,
-	vlResourceStatusRendering,
-	vlResourceStatusDisplaying
-};
-
-enum vlProfile
-{
-	vlProfileMpeg2Simple,
-	vlProfileMpeg2Main,
-
-	vlProfileCount
-};
-
-enum vlEntryPoint
-{
-	vlEntryPointIDCT,
-	vlEntryPointMC,
-	vlEntryPointCSC,
-
-	vlEntryPointCount
-};
-
-enum vlFormat
-{
-	vlFormatYCbCr420,
-	vlFormatYCbCr422,
-	vlFormatYCbCr444
-};
-
-enum vlPictureType
-{
-	vlPictureTypeTopField,
-	vlPictureTypeBottomField,
-	vlPictureTypeFrame
-};
-
-enum vlMotionType
-{
-	vlMotionTypeField,
-	vlMotionTypeFrame,
-	vlMotionTypeDualPrime,
-	vlMotionType16x8
-};
-
-enum vlFieldOrder
-{
-	vlFieldOrderFirst,
-	vlFieldOrderSecond
-};
-
-enum vlDCTType
-{
-	vlDCTTypeFrameCoded,
-	vlDCTTypeFieldCoded
-};
-
-struct vlVertex2f
-{
-	float x, y;
-};
-
-struct vlVertex4f
-{
-	float x, y, z, w;
-};
-
-enum vlMacroBlockType
-{
-	vlMacroBlockTypeIntra,
-	vlMacroBlockTypeFwdPredicted,
-	vlMacroBlockTypeBkwdPredicted,
-	vlMacroBlockTypeBiPredicted,
-
-	vlNumMacroBlockTypes
-};
-
-struct vlMpeg2MacroBlock
-{
-	unsigned int		mbx, mby;
-	enum vlMacroBlockType	mb_type;
-	enum vlMotionType	mo_type;
-	enum vlDCTType		dct_type;
-	int			PMV[2][2][2];
-	unsigned int		cbp;
-	short			*blocks;
-};
-
-struct vlMpeg2MacroBlockBatch
-{
-	struct vlSurface		*past_surface;
-	struct vlSurface		*future_surface;
-	enum vlPictureType		picture_type;
-	enum vlFieldOrder		field_order;
-	unsigned int			num_macroblocks;
-	struct vlMpeg2MacroBlock	*macroblocks;
-};
-
-#endif
diff --git a/src/gallium/state_trackers/g3dvl/vl_util.c b/src/gallium/state_trackers/g3dvl/vl_util.c
deleted file mode 100644
index 50aa9af66f..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_util.c
+++ /dev/null
@@ -1,16 +0,0 @@
-#include "vl_util.h"
-#include <assert.h>
-
-unsigned int vlRoundUpPOT(unsigned int x)
-{
-	unsigned int i;
-
-	assert(x > 0);
-
-	--x;
-
-	for (i = 1; i < sizeof(unsigned int) * 8; i <<= 1)
-		x |= x >> i;
-
-	return x + 1;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_util.h b/src/gallium/state_trackers/g3dvl/vl_util.h
deleted file mode 100644
index bc98e79df4..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_util.h
+++ /dev/null
@@ -1,6 +0,0 @@
-#ifndef vl_util_h
-#define vl_util_h
-
-unsigned int vlRoundUpPOT(unsigned int x);
-
-#endif
diff --git a/src/gallium/state_trackers/glx/xlib/glx_api.c b/src/gallium/state_trackers/glx/xlib/glx_api.c
index d1a98f8991..65012e9253 100644
--- a/src/gallium/state_trackers/glx/xlib/glx_api.c
+++ b/src/gallium/state_trackers/glx/xlib/glx_api.c
@@ -1538,9 +1538,9 @@ get_config( XMesaVisual xmvis, int attrib, int *value, GLboolean fbconfig )
          if (!fbconfig)
             return GLX_BAD_ATTRIBUTE;
          if (xmvis->mesa_visual.rgbMode)
-            *value = GLX_RGBA_BIT;
+            *value = GLX_RGBA_TYPE;
          else
-            *value = GLX_COLOR_INDEX_BIT;
+            *value = GLX_COLOR_INDEX_TYPE;
          break;
       case GLX_X_RENDERABLE_SGIX:
          if (!fbconfig)
@@ -2120,9 +2120,9 @@ glXQueryContext( Display *dpy, GLXContext ctx, int attribute, int *value )
       break;
    case GLX_RENDER_TYPE:
       if (xmctx->xm_visual->mesa_visual.rgbMode)
-         *value = GLX_RGBA_BIT;
+         *value = GLX_RGBA_TYPE;
       else
-         *value = GLX_COLOR_INDEX_BIT;
+         *value = GLX_COLOR_INDEX_TYPE;
       break;
    case GLX_SCREEN:
       *value = 0;
diff --git a/src/gallium/state_trackers/python/retrace/interpreter.py b/src/gallium/state_trackers/python/retrace/interpreter.py
index 6f0bd6ae52..f4ed2fde4d 100755
--- a/src/gallium/state_trackers/python/retrace/interpreter.py
+++ b/src/gallium/state_trackers/python/retrace/interpreter.py
@@ -314,7 +314,7 @@ class Screen(Object):
         if texture is None:
             return None
         transfer = Transfer(texture.get_surface(face, level, zslice), x, y, w, h)
-        if transfer and usage != gallium.PIPE_TRANSFER_WRITE:
+        if transfer and usage & gallium.PIPE_TRANSFER_READ
             if self.interpreter.options.all:
                 self.interpreter.present(transfer.surface, 'transf_read', x, y, w, h)
         return transfer
diff --git a/src/gallium/state_trackers/python/st_softpipe_winsys.c b/src/gallium/state_trackers/python/st_softpipe_winsys.c
index f0a4826a00..f0abd12e3d 100644
--- a/src/gallium/state_trackers/python/st_softpipe_winsys.c
+++ b/src/gallium/state_trackers/python/st_softpipe_winsys.c
@@ -172,6 +172,7 @@ st_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
                                   unsigned width, unsigned height,
                                   enum pipe_format format,
                                   unsigned usage,
+                                  unsigned tex_usage,
                                   unsigned *stride)
 {
    const unsigned alignment = 64;
diff --git a/src/gallium/state_trackers/vega/st_inlines.h b/src/gallium/state_trackers/vega/st_inlines.h
index 1f331dfcdb..610755e063 100644
--- a/src/gallium/state_trackers/vega/st_inlines.h
+++ b/src/gallium/state_trackers/vega/st_inlines.h
@@ -57,8 +57,7 @@ st_cond_flush_get_tex_transfer(struct vg_context *st,
       pipe->is_texture_referenced(pipe, pt, face, level);
 
    if (referenced && ((referenced & PIPE_REFERENCED_FOR_WRITE) ||
-		      usage == PIPE_TRANSFER_WRITE ||
-		      usage == PIPE_TRANSFER_READ_WRITE))
+		      (usage & PIPE_TRANSFER_WRITE)))
       vgFlush();
 
    return screen->get_tex_transfer(screen, pt, face, level, zslice, usage,
diff --git a/src/gallium/state_trackers/wgl/SConscript b/src/gallium/state_trackers/wgl/SConscript
index 69b88618ec..b05944a33b 100644
--- a/src/gallium/state_trackers/wgl/SConscript
+++ b/src/gallium/state_trackers/wgl/SConscript
@@ -18,20 +18,17 @@ if env['platform'] in ['windows']:
     ])
      
     sources = [
-        'icd/stw_icd.c',
-
-        'wgl/stw_wgl.c',
-
-        'shared/stw_context.c',
-        'shared/stw_device.c',
-        'shared/stw_framebuffer.c',
-        'shared/stw_pixelformat.c',
-        'shared/stw_extensionsstring.c',
-        'shared/stw_extswapinterval.c',
-        'shared/stw_getprocaddress.c',
-        'shared/stw_extgallium.c',
-        'shared/stw_arbpixelformat.c',
-        'shared/stw_tls.c',
+        'stw_context.c',
+        'stw_device.c',
+        'stw_ext_extensionsstring.c',
+        'stw_ext_gallium.c',
+        'stw_ext_pixelformat.c',
+        'stw_ext_swapinterval.c',
+        'stw_framebuffer.c',
+        'stw_getprocaddress.c',
+        'stw_pixelformat.c',
+        'stw_tls.c',
+        'stw_wgl.c',
     ]
 
     wgl = env.ConvenienceLibrary(
diff --git a/src/gallium/state_trackers/wgl/opengl32.def b/src/gallium/state_trackers/wgl/opengl32.def
index 596417ed84..5daa6ddd41 100644
--- a/src/gallium/state_trackers/wgl/opengl32.def
+++ b/src/gallium/state_trackers/wgl/opengl32.def
@@ -376,6 +376,7 @@ EXPORTS
 	DrvDescribePixelFormat
 	DrvGetLayerPaletteEntries
 	DrvGetProcAddress
+	DrvPresentBuffers
 	DrvRealizeLayerPalette
 	DrvReleaseContext
 	DrvSetCallbackProcs
diff --git a/src/gallium/state_trackers/wgl/opengl32.mingw.def b/src/gallium/state_trackers/wgl/opengl32.mingw.def
index 1f03ea3b37..6ebb31a6f1 100644
--- a/src/gallium/state_trackers/wgl/opengl32.mingw.def
+++ b/src/gallium/state_trackers/wgl/opengl32.mingw.def
@@ -375,6 +375,7 @@ EXPORTS
 	DrvDescribePixelFormat = DrvDescribePixelFormat@16
 	DrvGetLayerPaletteEntries = DrvGetLayerPaletteEntries@20
 	DrvGetProcAddress = DrvGetProcAddress@4
+	DrvPresentBuffers = DrvPresentBuffers@8
 	DrvRealizeLayerPalette = DrvRealizeLayerPalette@12
 	DrvReleaseContext = DrvReleaseContext@4
 	DrvSetCallbackProcs = DrvSetCallbackProcs@8
diff --git a/src/gallium/state_trackers/wgl/shared/stw_context.c b/src/gallium/state_trackers/wgl/shared/stw_context.c
deleted file mode 100644
index 4968ecc692..0000000000
--- a/src/gallium/state_trackers/wgl/shared/stw_context.c
+++ /dev/null
@@ -1,382 +0,0 @@
-/**************************************************************************
- *
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- *
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
-
-#include <windows.h>
-
-#include "main/mtypes.h"
-#include "main/context.h"
-#include "pipe/p_compiler.h"
-#include "pipe/p_context.h"
-#include "state_tracker/st_context.h"
-#include "state_tracker/st_public.h"
-
-#ifdef DEBUG
-#include "trace/tr_screen.h"
-#include "trace/tr_context.h"
-#endif
-
-#include "shared/stw_device.h"
-#include "shared/stw_winsys.h"
-#include "shared/stw_framebuffer.h"
-#include "shared/stw_pixelformat.h"
-#include "stw_public.h"
-#include "stw_context.h"
-#include "stw_tls.h"
-
-
-static INLINE struct stw_context *
-stw_context(GLcontext *glctx)
-{
-   if(!glctx)
-      return NULL;
-   assert(glctx->DriverCtx);
-   return (struct stw_context *)glctx->DriverCtx;
-}
-
-static INLINE struct stw_context *
-stw_current_context(void)
-{
-   /* We must check if multiple threads are being used or GET_CURRENT_CONTEXT 
-    * might return the current context of the thread first seen. */
-   _glapi_check_multithread();
-
-   {
-      GET_CURRENT_CONTEXT( glctx );
-      return stw_context(glctx);
-   }
-}
-
-BOOL
-stw_copy_context(
-   UINT_PTR hglrcSrc,
-   UINT_PTR hglrcDst,
-   UINT mask )
-{
-   struct stw_context *src;
-   struct stw_context *dst;
-   BOOL ret = FALSE;
-
-   pipe_mutex_lock( stw_dev->ctx_mutex );
-   
-   src = stw_lookup_context_locked( hglrcSrc );
-   dst = stw_lookup_context_locked( hglrcDst );
-
-   if (src && dst) { 
-      /* FIXME */
-      assert(0);
-      (void) src;
-      (void) dst;
-      (void) mask;
-   }
-
-   pipe_mutex_unlock( stw_dev->ctx_mutex );
-   
-   return ret;
-}
-
-BOOL
-stw_share_lists(
-   UINT_PTR hglrc1, 
-   UINT_PTR hglrc2 )
-{
-   struct stw_context *ctx1;
-   struct stw_context *ctx2;
-   BOOL ret = FALSE;
-
-   pipe_mutex_lock( stw_dev->ctx_mutex );
-   
-   ctx1 = stw_lookup_context_locked( hglrc1 );
-   ctx2 = stw_lookup_context_locked( hglrc2 );
-
-   if (ctx1 && ctx2 &&
-       ctx1->iPixelFormat == ctx2->iPixelFormat) { 
-      ret = _mesa_share_state(ctx2->st->ctx, ctx1->st->ctx);
-   }
-
-   pipe_mutex_unlock( stw_dev->ctx_mutex );
-   
-   return ret;
-}
-
-static void
-stw_viewport(GLcontext * glctx, GLint x, GLint y,
-             GLsizei width, GLsizei height)
-{
-   struct stw_context *ctx = (struct stw_context *)glctx->DriverCtx;
-   struct stw_framebuffer *fb;
-   
-   fb = stw_framebuffer_from_hdc( ctx->hdc );
-   if(fb) {
-      stw_framebuffer_update(fb);
-      stw_framebuffer_release(fb);
-   }
-}
-
-UINT_PTR
-stw_create_layer_context(
-   HDC hdc,
-   int iLayerPlane )
-{
-   int iPixelFormat;
-   const struct stw_pixelformat_info *pfi;
-   GLvisual visual;
-   struct stw_context *ctx = NULL;
-   struct pipe_screen *screen = NULL;
-   struct pipe_context *pipe = NULL;
-   
-   if(!stw_dev)
-      return 0;
-   
-   if (iLayerPlane != 0)
-      return 0;
-
-   iPixelFormat = GetPixelFormat(hdc);
-   if(!iPixelFormat)
-      return 0;
-   
-   pfi = stw_pixelformat_get_info( iPixelFormat - 1 );
-   stw_pixelformat_visual(&visual, pfi);
-   
-   ctx = CALLOC_STRUCT( stw_context );
-   if (ctx == NULL)
-      goto no_ctx;
-
-   ctx->hdc = hdc;
-   ctx->iPixelFormat = iPixelFormat;
-
-   screen = stw_dev->screen;
-
-#ifdef DEBUG
-   /* Unwrap screen */
-   if(stw_dev->trace_running)
-      screen = trace_screen(screen)->screen;
-#endif
-
-   pipe = stw_dev->stw_winsys->create_context( screen );
-   if (pipe == NULL) 
-      goto no_pipe;
-
-#ifdef DEBUG
-   /* Wrap context */
-   if(stw_dev->trace_running)
-      pipe = trace_context_create(stw_dev->screen, pipe);
-#endif
-
-   /* pass to stw_flush_frontbuffer as context_private */
-   assert(!pipe->priv);
-   pipe->priv = hdc;
-
-   ctx->st = st_create_context( pipe, &visual, NULL );
-   if (ctx->st == NULL) 
-      goto no_st_ctx;
-
-   ctx->st->ctx->DriverCtx = ctx;
-   ctx->st->ctx->Driver.Viewport = stw_viewport;
-
-   pipe_mutex_lock( stw_dev->ctx_mutex );
-   ctx->hglrc = handle_table_add(stw_dev->ctx_table, ctx);
-   pipe_mutex_unlock( stw_dev->ctx_mutex );
-   if (!ctx->hglrc)
-      goto no_hglrc;
-
-   return ctx->hglrc;
-
-no_hglrc:
-   st_destroy_context(ctx->st);
-   goto no_pipe; /* st_context_destroy already destroys pipe */
-no_st_ctx:
-   pipe->destroy( pipe );
-no_pipe:
-   FREE(ctx);
-no_ctx:
-   return 0;
-}
-
-BOOL
-stw_delete_context(
-   UINT_PTR hglrc )
-{
-   struct stw_context *ctx ;
-   BOOL ret = FALSE;
-   
-   if (!stw_dev)
-      return FALSE;
-
-   pipe_mutex_lock( stw_dev->ctx_mutex );
-   ctx = stw_lookup_context_locked(hglrc);
-   handle_table_remove(stw_dev->ctx_table, hglrc);
-   pipe_mutex_unlock( stw_dev->ctx_mutex );
-
-   if (ctx) {
-      struct stw_context *curctx = stw_current_context();
-      
-      /* Unbind current if deleting current context. */
-      if (curctx == ctx)
-         st_make_current( NULL, NULL, NULL );
-
-      st_destroy_context(ctx->st);
-      FREE(ctx);
-
-      ret = TRUE;
-   }
-
-   return ret;
-}
-
-BOOL
-stw_release_context(
-   UINT_PTR hglrc )
-{
-   struct stw_context *ctx;
-
-   if (!stw_dev)
-      return FALSE;
-
-   pipe_mutex_lock( stw_dev->ctx_mutex );
-   ctx = stw_lookup_context_locked( hglrc );
-   pipe_mutex_unlock( stw_dev->ctx_mutex );
-
-   if (!ctx)
-      return FALSE;
-   
-   /* The expectation is that ctx is the same context which is
-    * current for this thread.  We should check that and return False
-    * if not the case.
-    */
-   if (ctx != stw_current_context())
-      return FALSE;
-
-   if (stw_make_current( NULL, 0 ) == FALSE)
-      return FALSE;
-
-   return TRUE;
-}
-
-
-UINT_PTR
-stw_get_current_context( void )
-{
-   struct stw_context *ctx;
-
-   ctx = stw_current_context();
-   if(!ctx)
-      return 0;
-   
-   return ctx->hglrc;
-}
-
-HDC
-stw_get_current_dc( void )
-{
-   struct stw_context *ctx;
-
-   ctx = stw_current_context();
-   if(!ctx)
-      return NULL;
-   
-   return ctx->hdc;
-}
-
-BOOL
-stw_make_current(
-   HDC hdc,
-   UINT_PTR hglrc )
-{
-   struct stw_context *curctx = NULL;
-   struct stw_context *ctx = NULL;
-   struct stw_framebuffer *fb = NULL;
-
-   if (!stw_dev)
-      goto fail;
-
-   curctx = stw_current_context();
-   if (curctx != NULL) {
-      if (curctx->hglrc != hglrc)
-	 st_flush(curctx->st, PIPE_FLUSH_RENDER_CACHE, NULL);
-      
-      /* Return if already current. */
-      if (curctx->hglrc == hglrc && curctx->hdc == hdc) {
-         ctx = curctx;
-         fb = stw_framebuffer_from_hdc( hdc );
-         goto success;
-      }
-   }
-
-   if (hdc == NULL || hglrc == 0) {
-      return st_make_current( NULL, NULL, NULL );
-   }
-
-   pipe_mutex_lock( stw_dev->ctx_mutex ); 
-   ctx = stw_lookup_context_locked( hglrc );
-   pipe_mutex_unlock( stw_dev->ctx_mutex ); 
-   if(!ctx)
-      goto fail;
-
-   fb = stw_framebuffer_from_hdc( hdc );
-   if(!fb) { 
-      /* Applications should call SetPixelFormat before creating a context,
-       * but not all do, and the opengl32 runtime seems to use a default pixel
-       * format in some cases, so we must create a framebuffer for those here
-       */
-      int iPixelFormat = GetPixelFormat(hdc);
-      if(iPixelFormat)
-         fb = stw_framebuffer_create( hdc, iPixelFormat );
-      if(!fb) 
-         goto fail;
-   }
-   
-   if(fb->iPixelFormat != ctx->iPixelFormat)
-      goto fail;
-
-   /* Lazy allocation of the frame buffer */
-   if(!stw_framebuffer_allocate(fb))
-      goto fail;
-
-   /* Bind the new framebuffer */
-   ctx->hdc = hdc;
-   
-   /* pass to stw_flush_frontbuffer as context_private */
-   ctx->st->pipe->priv = hdc;
-   
-   if(!st_make_current( ctx->st, fb->stfb, fb->stfb ))
-      goto fail;
-
-success:
-   assert(fb);
-   if(fb) {
-      stw_framebuffer_update(fb);
-      stw_framebuffer_release(fb);
-   }
-   
-   return TRUE;
-
-fail:
-   if(fb)
-      stw_framebuffer_release(fb);
-   st_make_current( NULL, NULL, NULL );
-   return FALSE;
-}
diff --git a/src/gallium/state_trackers/wgl/icd/stw_icd.c b/src/gallium/state_trackers/wgl/stw_context.c
index 347f40aa06..f2f0264844 100644
--- a/src/gallium/state_trackers/wgl/icd/stw_icd.c
+++ b/src/gallium/state_trackers/wgl/stw_context.c
@@ -26,18 +26,49 @@
  **************************************************************************/
 
 #include <windows.h>
-#include <stdio.h>
 
-#include "GL/gl.h"
-
-#include "util/u_debug.h"
-#include "pipe/p_thread.h"
-
-#include "shared/stw_public.h"
-#include "icd/stw_icd.h"
+#include "main/mtypes.h"
+#include "main/context.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_context.h"
+#include "state_tracker/st_context.h"
+#include "state_tracker/st_public.h"
+
+#ifdef DEBUG
+#include "trace/tr_screen.h"
+#include "trace/tr_context.h"
+#endif
+
+#include "stw_icd.h"
+#include "stw_device.h"
+#include "stw_winsys.h"
+#include "stw_framebuffer.h"
+#include "stw_pixelformat.h"
+#include "stw_context.h"
+#include "stw_tls.h"
+
+
+static INLINE struct stw_context *
+stw_context(GLcontext *glctx)
+{
+   if(!glctx)
+      return NULL;
+   assert(glctx->DriverCtx);
+   return (struct stw_context *)glctx->DriverCtx;
+}
 
-#define DBG 0
+static INLINE struct stw_context *
+stw_current_context(void)
+{
+   /* We must check if multiple threads are being used or GET_CURRENT_CONTEXT 
+    * might return the current context of the thread first seen. */
+   _glapi_check_multithread();
 
+   {
+      GET_CURRENT_CONTEXT( glctx );
+      return stw_context(glctx);
+   }
+}
 
 BOOL APIENTRY
 DrvCopyContext(
@@ -45,24 +76,64 @@ DrvCopyContext(
    DHGLRC dhrcDest,
    UINT fuMask )
 {
-   return stw_copy_context(dhrcSource, dhrcDest, fuMask);
-}
+   struct stw_context *src;
+   struct stw_context *dst;
+   BOOL ret = FALSE;
 
+   pipe_mutex_lock( stw_dev->ctx_mutex );
+   
+   src = stw_lookup_context_locked( dhrcSource );
+   dst = stw_lookup_context_locked( dhrcDest );
+
+   if (src && dst) { 
+      /* FIXME */
+      assert(0);
+      (void) src;
+      (void) dst;
+      (void) fuMask;
+   }
 
-DHGLRC APIENTRY
-DrvCreateLayerContext(
-   HDC hdc,
-   INT iLayerPlane )
+   pipe_mutex_unlock( stw_dev->ctx_mutex );
+   
+   return ret;
+}
+
+BOOL APIENTRY
+DrvShareLists(
+   DHGLRC dhglrc1,
+   DHGLRC dhglrc2 )
 {
-   DHGLRC r;
+   struct stw_context *ctx1;
+   struct stw_context *ctx2;
+   BOOL ret = FALSE;
+
+   pipe_mutex_lock( stw_dev->ctx_mutex );
    
-   r = stw_create_layer_context( hdc, iLayerPlane );
+   ctx1 = stw_lookup_context_locked( dhglrc1 );
+   ctx2 = stw_lookup_context_locked( dhglrc2 );
+
+   if (ctx1 && ctx2 &&
+       ctx1->iPixelFormat == ctx2->iPixelFormat) { 
+      ret = _mesa_share_state(ctx2->st->ctx, ctx1->st->ctx);
+   }
+
+   pipe_mutex_unlock( stw_dev->ctx_mutex );
    
-   if (DBG)
-      debug_printf( "%s( %p, %i ) = %u\n",
-                    __FUNCTION__, hdc, iLayerPlane, r );
+   return ret;
+}
+
+static void
+stw_viewport(GLcontext * glctx, GLint x, GLint y,
+             GLsizei width, GLsizei height)
+{
+   struct stw_context *ctx = (struct stw_context *)glctx->DriverCtx;
+   struct stw_framebuffer *fb;
    
-   return r;
+   fb = stw_framebuffer_from_hdc( ctx->hdc );
+   if(fb) {
+      stw_framebuffer_update(fb);
+      stw_framebuffer_release(fb);
+   }
 }
 
 DHGLRC APIENTRY
@@ -72,114 +143,253 @@ DrvCreateContext(
    return DrvCreateLayerContext( hdc, 0 );
 }
 
-BOOL APIENTRY
-DrvDeleteContext(
-   DHGLRC dhglrc )
+DHGLRC APIENTRY
+DrvCreateLayerContext(
+   HDC hdc,
+   INT iLayerPlane )
 {
-   BOOL r;
+   int iPixelFormat;
+   const struct stw_pixelformat_info *pfi;
+   GLvisual visual;
+   struct stw_context *ctx = NULL;
+   struct pipe_screen *screen = NULL;
+   struct pipe_context *pipe = NULL;
    
-   r = stw_delete_context( dhglrc );
+   if(!stw_dev)
+      return 0;
    
-   if (DBG)
-      debug_printf( "%s( %u ) = %u\n",
-                    __FUNCTION__, dhglrc, r );
+   if (iLayerPlane != 0)
+      return 0;
+
+   iPixelFormat = GetPixelFormat(hdc);
+   if(!iPixelFormat)
+      return 0;
    
-   return r;
+   pfi = stw_pixelformat_get_info( iPixelFormat - 1 );
+   stw_pixelformat_visual(&visual, pfi);
+   
+   ctx = CALLOC_STRUCT( stw_context );
+   if (ctx == NULL)
+      goto no_ctx;
+
+   ctx->hdc = hdc;
+   ctx->iPixelFormat = iPixelFormat;
+
+   screen = stw_dev->screen;
+
+#ifdef DEBUG
+   /* Unwrap screen */
+   if(stw_dev->trace_running)
+      screen = trace_screen(screen)->screen;
+#endif
+
+   pipe = stw_dev->stw_winsys->create_context( screen );
+   if (pipe == NULL) 
+      goto no_pipe;
+
+#ifdef DEBUG
+   /* Wrap context */
+   if(stw_dev->trace_running)
+      pipe = trace_context_create(stw_dev->screen, pipe);
+#endif
+
+   /* pass to stw_flush_frontbuffer as context_private */
+   assert(!pipe->priv);
+   pipe->priv = hdc;
+
+   ctx->st = st_create_context( pipe, &visual, NULL );
+   if (ctx->st == NULL) 
+      goto no_st_ctx;
+
+   ctx->st->ctx->DriverCtx = ctx;
+   ctx->st->ctx->Driver.Viewport = stw_viewport;
+
+   pipe_mutex_lock( stw_dev->ctx_mutex );
+   ctx->dhglrc = handle_table_add(stw_dev->ctx_table, ctx);
+   pipe_mutex_unlock( stw_dev->ctx_mutex );
+   if (!ctx->dhglrc)
+      goto no_hglrc;
+
+   return ctx->dhglrc;
+
+no_hglrc:
+   st_destroy_context(ctx->st);
+   goto no_pipe; /* st_context_destroy already destroys pipe */
+no_st_ctx:
+   pipe->destroy( pipe );
+no_pipe:
+   FREE(ctx);
+no_ctx:
+   return 0;
 }
 
 BOOL APIENTRY
-DrvDescribeLayerPlane(
-   HDC hdc,
-   INT iPixelFormat,
-   INT iLayerPlane,
-   UINT nBytes,
-   LPLAYERPLANEDESCRIPTOR plpd )
+DrvDeleteContext(
+   DHGLRC dhglrc )
 {
-   if (DBG) 
-      debug_printf( "%s\n", __FUNCTION__ );
+   struct stw_context *ctx ;
+   BOOL ret = FALSE;
+   
+   if (!stw_dev)
+      return FALSE;
 
-   return FALSE;
-}
+   pipe_mutex_lock( stw_dev->ctx_mutex );
+   ctx = stw_lookup_context_locked(dhglrc);
+   handle_table_remove(stw_dev->ctx_table, dhglrc);
+   pipe_mutex_unlock( stw_dev->ctx_mutex );
 
-LONG APIENTRY
-DrvDescribePixelFormat(
-   HDC hdc,
-   INT iPixelFormat,
-   ULONG cjpfd,
-   PIXELFORMATDESCRIPTOR *ppfd )
-{
-   LONG r;
+   if (ctx) {
+      struct stw_context *curctx = stw_current_context();
+      
+      /* Unbind current if deleting current context. */
+      if (curctx == ctx)
+         st_make_current( NULL, NULL, NULL );
 
-   r = stw_pixelformat_describe( hdc, iPixelFormat, cjpfd, ppfd );
+      st_destroy_context(ctx->st);
+      FREE(ctx);
 
-   if (DBG)
-      debug_printf( "%s( %p, %d, %u, %p ) = %d\n",
-                    __FUNCTION__, hdc, iPixelFormat, cjpfd, ppfd, r );
+      ret = TRUE;
+   }
 
-   return r;
+   return ret;
 }
 
-int APIENTRY
-DrvGetLayerPaletteEntries(
-   HDC hdc,
-   INT iLayerPlane,
-   INT iStart,
-   INT cEntries,
-   COLORREF *pcr )
+BOOL APIENTRY
+DrvReleaseContext(
+   DHGLRC dhglrc )
 {
-   if (DBG)
-      debug_printf( "%s\n", __FUNCTION__ );
+   struct stw_context *ctx;
 
-   return 0;
-}
+   if (!stw_dev)
+      return FALSE;
 
-PROC APIENTRY
-DrvGetProcAddress(
-   LPCSTR lpszProc )
-{
-   PROC r;
+   pipe_mutex_lock( stw_dev->ctx_mutex );
+   ctx = stw_lookup_context_locked( dhglrc );
+   pipe_mutex_unlock( stw_dev->ctx_mutex );
 
-   r = stw_get_proc_address( lpszProc );
+   if (!ctx)
+      return FALSE;
+   
+   /* The expectation is that ctx is the same context which is
+    * current for this thread.  We should check that and return False
+    * if not the case.
+    */
+   if (ctx != stw_current_context())
+      return FALSE;
 
-   if (DBG)
-      debug_printf( "%s( \"%s\" ) = %p\n", __FUNCTION__, lpszProc, r );
+   if (stw_make_current( NULL, 0 ) == FALSE)
+      return FALSE;
 
-   return r;
+   return TRUE;
 }
 
-BOOL APIENTRY
-DrvRealizeLayerPalette(
-   HDC hdc,
-   INT iLayerPlane,
-   BOOL bRealize )
+
+DHGLRC
+stw_get_current_context( void )
 {
-   if (DBG)
-      debug_printf( "%s\n", __FUNCTION__ );
+   struct stw_context *ctx;
 
-   return FALSE;
+   ctx = stw_current_context();
+   if(!ctx)
+      return 0;
+   
+   return ctx->dhglrc;
 }
 
-BOOL APIENTRY
-DrvReleaseContext(
-   DHGLRC dhglrc )
+HDC
+stw_get_current_dc( void )
 {
-   return stw_release_context(dhglrc);
+   struct stw_context *ctx;
+
+   ctx = stw_current_context();
+   if(!ctx)
+      return NULL;
+   
+   return ctx->hdc;
 }
 
-void APIENTRY
-DrvSetCallbackProcs(
-   INT nProcs,
-   PROC *pProcs )
+BOOL
+stw_make_current(
+   HDC hdc,
+   DHGLRC dhglrc )
 {
-   if (DBG)
-      debug_printf( "%s( %d, %p )\n", __FUNCTION__, nProcs, pProcs );
+   struct stw_context *curctx = NULL;
+   struct stw_context *ctx = NULL;
+   struct stw_framebuffer *fb = NULL;
 
-   return;
-}
+   if (!stw_dev)
+      goto fail;
+
+   curctx = stw_current_context();
+   if (curctx != NULL) {
+      if (curctx->dhglrc != dhglrc)
+	 st_flush(curctx->st, PIPE_FLUSH_RENDER_CACHE, NULL);
+      
+      /* Return if already current. */
+      if (curctx->dhglrc == dhglrc && curctx->hdc == hdc) {
+         ctx = curctx;
+         fb = stw_framebuffer_from_hdc( hdc );
+         goto success;
+      }
+   }
+
+   if (hdc == NULL || dhglrc == 0) {
+      return st_make_current( NULL, NULL, NULL );
+   }
+
+   pipe_mutex_lock( stw_dev->ctx_mutex ); 
+   ctx = stw_lookup_context_locked( dhglrc );
+   pipe_mutex_unlock( stw_dev->ctx_mutex ); 
+   if(!ctx)
+      goto fail;
+
+   fb = stw_framebuffer_from_hdc( hdc );
+   if(!fb) { 
+      /* Applications should call SetPixelFormat before creating a context,
+       * but not all do, and the opengl32 runtime seems to use a default pixel
+       * format in some cases, so we must create a framebuffer for those here
+       */
+      int iPixelFormat = GetPixelFormat(hdc);
+      if(iPixelFormat)
+         fb = stw_framebuffer_create( hdc, iPixelFormat );
+      if(!fb) 
+         goto fail;
+   }
+   
+   if(fb->iPixelFormat != ctx->iPixelFormat)
+      goto fail;
+
+   /* Lazy allocation of the frame buffer */
+   if(!stw_framebuffer_allocate(fb))
+      goto fail;
+
+   /* Bind the new framebuffer */
+   ctx->hdc = hdc;
+   
+   /* pass to stw_flush_frontbuffer as context_private */
+   ctx->st->pipe->priv = hdc;
+   
+   if(!st_make_current( ctx->st, fb->stfb, fb->stfb ))
+      goto fail;
+
+success:
+   assert(fb);
+   if(fb) {
+      stw_framebuffer_update(fb);
+      stw_framebuffer_release(fb);
+   }
+   
+   return TRUE;
 
+fail:
+   if(fb)
+      stw_framebuffer_release(fb);
+   st_make_current( NULL, NULL, NULL );
+   return FALSE;
+}
 
 /**
- * Although WGL allows different dispatch entrypoints per context 
+ * Although WGL allows different dispatch entrypoints per context
  */
 static const GLCLTPROCTABLE cpt =
 {
@@ -524,7 +734,6 @@ static const GLCLTPROCTABLE cpt =
    }
 };
 
-
 PGLCLTPROCTABLE APIENTRY
 DrvSetContext(
    HDC hdc,
@@ -532,86 +741,9 @@ DrvSetContext(
    PFN_SETPROCTABLE pfnSetProcTable )
 {
    PGLCLTPROCTABLE r = (PGLCLTPROCTABLE)&cpt;
-   
+
    if (!stw_make_current( hdc, dhglrc ))
       r = NULL;
-      
-   if (DBG)
-      debug_printf( "%s( 0x%p, %u, 0x%p ) = %p\n", 
-                    __FUNCTION__, hdc, dhglrc, pfnSetProcTable, r );
-
-   return r;
-}
-
-int APIENTRY
-DrvSetLayerPaletteEntries(
-   HDC hdc,
-   INT iLayerPlane,
-   INT iStart,
-   INT cEntries,
-   CONST COLORREF *pcr )
-{
-   if (DBG)
-      debug_printf( "%s\n", __FUNCTION__ );
-
-   return 0;
-}
-
-BOOL APIENTRY
-DrvSetPixelFormat(
-   HDC hdc,
-   LONG iPixelFormat )
-{
-   BOOL r;
-
-   r = stw_pixelformat_set( hdc, iPixelFormat );
-
-   if (DBG)
-      debug_printf( "%s( %p, %d ) = %s\n", __FUNCTION__, hdc, iPixelFormat, r ? "TRUE" : "FALSE" );
 
    return r;
 }
-
-BOOL APIENTRY
-DrvShareLists(
-   DHGLRC dhglrc1,
-   DHGLRC dhglrc2 )
-{
-   if (DBG)
-      debug_printf( "%s\n", __FUNCTION__ );
-
-   return stw_share_lists(dhglrc1, dhglrc2);
-}
-
-BOOL APIENTRY
-DrvSwapBuffers(
-   HDC hdc )
-{
-   if (DBG)
-      debug_printf( "%s( %p )\n", __FUNCTION__, hdc );
-
-   return stw_swap_buffers( hdc );
-}
-
-BOOL APIENTRY
-DrvSwapLayerBuffers(
-   HDC hdc,
-   UINT fuPlanes )
-{
-   if (DBG)
-      debug_printf( "%s\n", __FUNCTION__ );
-
-   return stw_swap_layer_buffers( hdc, fuPlanes );
-}
-
-BOOL APIENTRY
-DrvValidateVersion(
-   ULONG ulVersion )
-{
-   if (DBG)
-      debug_printf( "%s( %u )\n", __FUNCTION__, ulVersion );
-
-   /* TODO: get the expected version from the winsys */
-   
-   return ulVersion == 1;
-}
diff --git a/src/gallium/state_trackers/wgl/shared/stw_context.h b/src/gallium/state_trackers/wgl/stw_context.h
index 166471de5e..256c27e21e 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_context.h
+++ b/src/gallium/state_trackers/wgl/stw_context.h
@@ -35,9 +35,15 @@ struct st_context;
 struct stw_context
 {
    struct st_context *st;
-   UINT_PTR hglrc;
+   DHGLRC dhglrc;
    int iPixelFormat;
    HDC hdc;
 };
 
+DHGLRC stw_get_current_context( void );
+
+HDC stw_get_current_dc( void );
+
+BOOL stw_make_current( HDC hdc, DHGLRC dhglrc );
+
 #endif /* STW_CONTEXT_H */
diff --git a/src/gallium/state_trackers/wgl/shared/stw_device.c b/src/gallium/state_trackers/wgl/stw_device.c
index 0b6954915a..985b8f0456 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_device.c
+++ b/src/gallium/state_trackers/wgl/stw_device.c
@@ -29,6 +29,7 @@
 
 #include "glapi/glthread.h"
 #include "util/u_debug.h"
+#include "util/u_math.h"
 #include "pipe/p_screen.h"
 #include "state_tracker/st_public.h"
 
@@ -37,12 +38,12 @@
 #include "trace/tr_texture.h"
 #endif
 
-#include "shared/stw_device.h"
-#include "shared/stw_winsys.h"
-#include "shared/stw_pixelformat.h"
-#include "shared/stw_public.h"
-#include "shared/stw_tls.h"
-#include "shared/stw_framebuffer.h"
+#include "stw_device.h"
+#include "stw_winsys.h"
+#include "stw_pixelformat.h"
+#include "stw_icd.h"
+#include "stw_tls.h"
+#include "stw_framebuffer.h"
 
 #ifdef WIN32_THREADS
 extern _glthread_Mutex OneTimeLock;
@@ -62,38 +63,28 @@ stw_flush_frontbuffer(struct pipe_screen *screen,
                      struct pipe_surface *surface,
                      void *context_private )
 {
-   const struct stw_winsys *stw_winsys = stw_dev->stw_winsys;
    HDC hdc = (HDC)context_private;
    struct stw_framebuffer *fb;
    
    fb = stw_framebuffer_from_hdc( hdc );
-   /* fb can be NULL if window was destroyed already */
-   if (fb) {
+   if (!fb) {
+      /* fb can be NULL if window was destroyed already */
+      return;
+   }
+
 #if DEBUG
-      {
-         struct pipe_surface *surface2;
-   
-         if(!st_get_framebuffer_surface( fb->stfb, ST_SURFACE_FRONT_LEFT, &surface2 ))
-            assert(0);
-         else
-            assert(surface2 == surface);
-      }
-#endif
+   {
+      /* ensure that a random surface was not passed to us */
+      struct pipe_surface *surface2;
 
-#ifdef DEBUG
-      if(stw_dev->trace_running) {
-         screen = trace_screen(screen)->screen;
-         surface = trace_surface(surface)->surface;
-      }
-#endif
-   }
-   
-   stw_winsys->flush_frontbuffer(screen, surface, hdc);
-   
-   if(fb) {
-      stw_framebuffer_update(fb);
-      stw_framebuffer_release(fb);
+      if(!st_get_framebuffer_surface( fb->stfb, ST_SURFACE_FRONT_LEFT, &surface2 ))
+         assert(0);
+      else
+         assert(surface2 == surface);
    }
+#endif
+
+   stw_framebuffer_present_locked(hdc, fb, ST_SURFACE_FRONT_LEFT);
 }
 
 
@@ -126,6 +117,9 @@ stw_init(const struct stw_winsys *stw_winsys)
    if(!screen)
       goto error1;
 
+   if(stw_winsys->get_adapter_luid)
+      stw_winsys->get_adapter_luid(screen, &stw_dev->AdapterLuid);
+
 #ifdef DEBUG
    stw_dev->screen = trace_screen_create(screen);
    stw_dev->trace_running = stw_dev->screen != screen ? TRUE : FALSE;
@@ -182,7 +176,7 @@ stw_cleanup(void)
       /* Ensure all contexts are destroyed */
       i = handle_table_get_first_handle(stw_dev->ctx_table);
       while (i) {
-         stw_delete_context(i);
+         DrvDeleteContext(i);
          i = handle_table_get_next_handle(stw_dev->ctx_table, i);
       }
       handle_table_destroy(stw_dev->ctx_table);
@@ -212,7 +206,7 @@ stw_cleanup(void)
 
 
 struct stw_context *
-stw_lookup_context_locked( UINT_PTR dhglrc )
+stw_lookup_context_locked( DHGLRC dhglrc )
 {
    if (dhglrc == 0)
       return NULL;
@@ -223,3 +217,28 @@ stw_lookup_context_locked( UINT_PTR dhglrc )
    return (struct stw_context *) handle_table_get(stw_dev->ctx_table, dhglrc);
 }
 
+
+void APIENTRY
+DrvSetCallbackProcs(
+   INT nProcs,
+   PROC *pProcs )
+{
+   size_t size;
+
+   if (stw_dev == NULL)
+      return;
+
+   size = MIN2(nProcs * sizeof *pProcs, sizeof stw_dev->callbacks);
+   memcpy(&stw_dev->callbacks, pProcs, size);
+
+   return;
+}
+
+
+BOOL APIENTRY
+DrvValidateVersion(
+   ULONG ulVersion )
+{
+   /* TODO: get the expected version from the winsys */
+   return ulVersion == 1;
+}
diff --git a/src/gallium/state_trackers/wgl/shared/stw_device.h b/src/gallium/state_trackers/wgl/stw_device.h
index e1bb9518dd..0bf3b0da82 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_device.h
+++ b/src/gallium/state_trackers/wgl/stw_device.h
@@ -29,11 +29,10 @@
 #define STW_DEVICE_H_
 
 
-#include <windows.h>
-
 #include "pipe/p_compiler.h"
 #include "pipe/p_thread.h"
 #include "util/u_handle_table.h"
+#include "stw_icd.h"
 #include "stw_pixelformat.h"
 
 
@@ -53,10 +52,14 @@ struct stw_device
    boolean trace_running;
 #endif
 
+   LUID AdapterLuid;
+
    struct stw_pixelformat_info pixelformats[STW_MAX_PIXELFORMATS];
    unsigned pixelformat_count;
    unsigned pixelformat_extended_count;
 
+   GLCALLBACKTABLE callbacks;
+
    pipe_mutex ctx_mutex;
    struct handle_table *ctx_table;
    
@@ -69,7 +72,7 @@ struct stw_device
 };
 
 struct stw_context *
-stw_lookup_context_locked( UINT_PTR hglrc );
+stw_lookup_context_locked( DHGLRC hglrc );
 
 extern struct stw_device *stw_dev;
 
diff --git a/src/gallium/state_trackers/wgl/shared/stw_extensionsstring.c b/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c
index 62c859e1f9..62c859e1f9 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_extensionsstring.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_extensionsstring.c
diff --git a/src/gallium/state_trackers/wgl/shared/stw_extgallium.c b/src/gallium/state_trackers/wgl/stw_ext_gallium.c
index fc22737d7e..fb30ec5dba 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_extgallium.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_gallium.c
@@ -27,9 +27,9 @@
 
 
 #include "pipe/p_screen.h"
-#include "stw_public.h"
 #include "stw_device.h"
 #include "stw_winsys.h"
+#include "stw_ext_gallium.h"
 
 #ifdef DEBUG
 #include "trace/tr_screen.h"
diff --git a/src/gallium/state_trackers/wgl/shared/stw_extgallium.h b/src/gallium/state_trackers/wgl/stw_ext_gallium.h
index cc35f2bb7f..cc35f2bb7f 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_extgallium.h
+++ b/src/gallium/state_trackers/wgl/stw_ext_gallium.h
diff --git a/src/gallium/state_trackers/wgl/shared/stw_arbpixelformat.c b/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c
index 0e2d407699..8a9995aba8 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_arbpixelformat.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_pixelformat.c
@@ -43,7 +43,6 @@
 
 #include "pipe/p_compiler.h"
 #include "util/u_memory.h"
-#include "stw_public.h"
 #include "stw_pixelformat.h"
 
 
diff --git a/src/gallium/state_trackers/wgl/shared/stw_extswapinterval.c b/src/gallium/state_trackers/wgl/stw_ext_swapinterval.c
index 9eac6a1d09..9eac6a1d09 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_extswapinterval.c
+++ b/src/gallium/state_trackers/wgl/stw_ext_swapinterval.c
diff --git a/src/gallium/state_trackers/wgl/shared/stw_framebuffer.c b/src/gallium/state_trackers/wgl/stw_framebuffer.c
index b8956bb550..8a3e11b6b4 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_framebuffer.c
+++ b/src/gallium/state_trackers/wgl/stw_framebuffer.c
@@ -1,8 +1,8 @@
 /**************************************************************************
- * 
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ *
+ * Copyright 2008-2009 Vmware, Inc.
  * All Rights Reserved.
- * 
+ *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the
  * "Software"), to deal in the Software without restriction, including
@@ -10,19 +10,19 @@
  * 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
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
+ *
  **************************************************************************/
 
 #include <windows.h>
@@ -38,9 +38,9 @@
 #include "trace/tr_texture.h"
 #endif
 
+#include "stw_icd.h"
 #include "stw_framebuffer.h"
 #include "stw_device.h"
-#include "stw_public.h"
 #include "stw_winsys.h"
 #include "stw_tls.h"
 
@@ -83,6 +83,9 @@ stw_framebuffer_destroy_locked(
    *link = fb->next;
    fb->next = NULL;
 
+   if(fb->shared_surface)
+      stw_dev->stw_winsys->shared_surface_close(stw_dev->screen, fb->shared_surface);
+
    st_unreference_framebuffer(fb->stfb);
    
    pipe_mutex_unlock( fb->mutex );
@@ -106,13 +109,18 @@ static INLINE void
 stw_framebuffer_get_size( struct stw_framebuffer *fb )
 {
    unsigned width, height;
-   RECT rect;
+   RECT client_rect;
+   RECT window_rect;
+   POINT client_pos;
 
    assert(fb->hWnd);
    
-   GetClientRect( fb->hWnd, &rect );
-   width = rect.right - rect.left;
-   height = rect.bottom - rect.top;
+   /* Get the client area size. */
+   GetClientRect( fb->hWnd, &client_rect );
+   assert(client_rect.left == 0);
+   assert(client_rect.top == 0);
+   width = client_rect.right - client_rect.left;
+   height = client_rect.bottom - client_rect.top;
 
    if(width < 1)
       width = 1;
@@ -124,6 +132,31 @@ stw_framebuffer_get_size( struct stw_framebuffer *fb )
       fb->width = width; 
       fb->height = height; 
    }
+
+   client_pos.x = 0;
+   client_pos.y = 0;
+   ClientToScreen(fb->hWnd, &client_pos);
+
+   GetWindowRect(fb->hWnd, &window_rect);
+
+   fb->client_rect.left = client_pos.x - window_rect.left;
+   fb->client_rect.top =  client_pos.y - window_rect.top;
+   fb->client_rect.right = fb->client_rect.left + fb->width;
+   fb->client_rect.bottom = fb->client_rect.top + fb->height;
+
+#if 0
+   debug_printf("\n");
+   debug_printf("%s: client_position = (%i, %i)\n",
+                __FUNCTION__, client_pos.x, client_pos.y);
+   debug_printf("%s: window_rect = (%i, %i) - (%i, %i)\n",
+                __FUNCTION__,
+                window_rect.left, window_rect.top,
+                window_rect.right, window_rect.bottom);
+   debug_printf("%s: client_rect = (%i, %i) - (%i, %i)\n",
+                __FUNCTION__,
+                fb->client_rect.left, fb->client_rect.top,
+                fb->client_rect.right, fb->client_rect.bottom);
+#endif
 }
 
 
@@ -155,6 +188,7 @@ stw_call_window_proc(
        * can be masked out by the application. */
       LPWINDOWPOS lpWindowPos = (LPWINDOWPOS)pParams->lParam;
       if((lpWindowPos->flags & SWP_SHOWWINDOW) || 
+         !(lpWindowPos->flags & SWP_NOMOVE) ||
          !(lpWindowPos->flags & SWP_NOSIZE)) {
          fb = stw_framebuffer_from_hwnd( pParams->hwnd );
          if(fb) {
@@ -379,10 +413,10 @@ stw_framebuffer_from_hwnd(
 }
 
 
-BOOL
-stw_pixelformat_set(
+BOOL APIENTRY
+DrvSetPixelFormat(
    HDC hdc,
-   int iPixelFormat )
+   LONG iPixelFormat )
 {
    uint count;
    uint index;
@@ -435,35 +469,24 @@ stw_pixelformat_get(
 }
 
 
-BOOL
-stw_swap_buffers(
-   HDC hdc )
+BOOL APIENTRY
+DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data)
 {
    struct stw_framebuffer *fb;
    struct pipe_screen *screen;
    struct pipe_surface *surface;
+   unsigned surface_index;
+   BOOL ret = FALSE;
 
    fb = stw_framebuffer_from_hdc( hdc );
    if (fb == NULL)
       return FALSE;
 
-   if (!(fb->pfi->pfd.dwFlags & PFD_DOUBLEBUFFER)) {
-      stw_framebuffer_release(fb);
-      return TRUE;
-   }
-
-   /* If we're swapping the buffer associated with the current context
-    * we have to flush any pending rendering commands first.
-    */
-   st_notify_swapbuffers( fb->stfb );
-
    screen = stw_dev->screen;
-   
-   if(!st_get_framebuffer_surface( fb->stfb, ST_SURFACE_BACK_LEFT, &surface )) {
-      /* FIXME: this shouldn't happen, but does on glean */
-      stw_framebuffer_release(fb);
-      return FALSE;
-   }
+
+   surface_index = (unsigned)(uintptr_t)data->pPrivateData;
+   if(!st_get_framebuffer_surface( fb->stfb, surface_index, &surface ))
+      goto fail;
 
 #ifdef DEBUG
    if(stw_dev->trace_running) {
@@ -472,22 +495,127 @@ stw_swap_buffers(
    }
 #endif
 
-   stw_dev->stw_winsys->flush_frontbuffer( screen, surface, hdc );
-   
+   if(data->hSharedSurface != fb->hSharedSurface) {
+      if(fb->shared_surface) {
+         stw_dev->stw_winsys->shared_surface_close(screen, fb->shared_surface);
+         fb->shared_surface = NULL;
+      }
+
+      fb->hSharedSurface = data->hSharedSurface;
+
+      if(data->hSharedSurface &&
+         stw_dev->stw_winsys->shared_surface_open) {
+         fb->shared_surface = stw_dev->stw_winsys->shared_surface_open(screen, fb->hSharedSurface);
+      }
+   }
+
+   if(fb->shared_surface) {
+      stw_dev->stw_winsys->compose(screen,
+                                   surface,
+                                   fb->shared_surface,
+                                   &fb->client_rect,
+                                   data->PresentHistoryToken);
+   }
+   else {
+      stw_dev->stw_winsys->present( screen, surface, hdc );
+   }
+
+   ret = TRUE;
+
+fail:
+
    stw_framebuffer_update(fb);
+
    stw_framebuffer_release(fb);
-   
-   return TRUE;
+
+   return ret;
 }
 
 
+/**
+ * Queue a composition.
+ *
+ * It will drop the lock on success.
+ */
 BOOL
-stw_swap_layer_buffers(
+stw_framebuffer_present_locked(HDC hdc,
+                               struct stw_framebuffer *fb,
+                               unsigned surface_index)
+{
+   if(stw_dev->callbacks.wglCbPresentBuffers &&
+      stw_dev->stw_winsys->compose) {
+      GLCBPRESENTBUFFERSDATA data;
+
+      memset(&data, 0, sizeof data);
+      data.magic1 = 2;
+      data.magic2 = 0;
+      data.AdapterLuid = stw_dev->AdapterLuid;
+      data.rect = fb->client_rect;
+      data.pPrivateData = (void *)(uintptr_t)surface_index;
+
+      stw_framebuffer_release(fb);
+
+      return stw_dev->callbacks.wglCbPresentBuffers(hdc, &data);
+   }
+   else {
+      struct pipe_screen *screen = stw_dev->screen;
+      struct pipe_surface *surface;
+
+      if(!st_get_framebuffer_surface( fb->stfb, surface_index, &surface )) {
+         /* FIXME: this shouldn't happen, but does on glean */
+         stw_framebuffer_release(fb);
+         return FALSE;
+      }
+
+#ifdef DEBUG
+      if(stw_dev->trace_running) {
+         screen = trace_screen(screen)->screen;
+         surface = trace_surface(surface)->surface;
+      }
+#endif
+
+      stw_dev->stw_winsys->present( screen, surface, hdc );
+
+      stw_framebuffer_update(fb);
+
+      stw_framebuffer_release(fb);
+
+      return TRUE;
+   }
+}
+
+
+BOOL APIENTRY
+DrvSwapBuffers(
+   HDC hdc )
+{
+   struct stw_framebuffer *fb;
+
+   fb = stw_framebuffer_from_hdc( hdc );
+   if (fb == NULL)
+      return FALSE;
+
+   if (!(fb->pfi->pfd.dwFlags & PFD_DOUBLEBUFFER)) {
+      stw_framebuffer_release(fb);
+      return TRUE;
+   }
+
+   /* If we're swapping the buffer associated with the current context
+    * we have to flush any pending rendering commands first.
+    */
+   st_notify_swapbuffers( fb->stfb );
+
+   return stw_framebuffer_present_locked(hdc, fb, ST_SURFACE_BACK_LEFT);
+}
+
+
+BOOL APIENTRY
+DrvSwapLayerBuffers(
    HDC hdc,
    UINT fuPlanes )
 {
    if(fuPlanes & WGL_SWAP_MAIN_PLANE)
-      return stw_swap_buffers(hdc);
+      return DrvSwapBuffers(hdc);
 
    return FALSE;
 }
diff --git a/src/gallium/state_trackers/wgl/shared/stw_framebuffer.h b/src/gallium/state_trackers/wgl/stw_framebuffer.h
index 13d29f37e4..5afbe74908 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_framebuffer.h
+++ b/src/gallium/state_trackers/wgl/stw_framebuffer.h
@@ -73,9 +73,20 @@ struct stw_framebuffer
    
    /* FIXME: Make this work for multiple contexts bound to the same framebuffer */
    boolean must_resize;
+
    unsigned width;
    unsigned height;
    
+   /**
+    * Client area rectangle, relative to the window upper-left corner.
+    *
+    * @sa GLCBPRESENTBUFFERSDATA::rect.
+    */
+   RECT client_rect;
+
+   HANDLE hSharedSurface;
+   struct stw_shared_surface *shared_surface;
+
    /** 
     * This is protected by stw_device::fb_mutex, not the mutex above.
     * 
@@ -126,6 +137,11 @@ BOOL
 stw_framebuffer_allocate(
    struct stw_framebuffer *fb );
 
+BOOL
+stw_framebuffer_present_locked(HDC hdc,
+                               struct stw_framebuffer *fb,
+                               unsigned surface_index);
+
 void
 stw_framebuffer_update(
    struct stw_framebuffer *fb);
diff --git a/src/gallium/state_trackers/wgl/shared/stw_getprocaddress.c b/src/gallium/state_trackers/wgl/stw_getprocaddress.c
index 879ced925a..8875dc22f3 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_getprocaddress.c
+++ b/src/gallium/state_trackers/wgl/stw_getprocaddress.c
@@ -33,8 +33,7 @@
 #include <GL/wglext.h>
 
 #include "glapi/glapi.h"
-#include "stw_public.h"
-#include "stw_extgallium.h"
+#include "stw_ext_gallium.h"
 
 struct stw_extension_entry
 {
@@ -68,8 +67,8 @@ static const struct stw_extension_entry stw_extension_entries[] = {
    { NULL, NULL }
 };
 
-PROC
-stw_get_proc_address(
+PROC APIENTRY
+DrvGetProcAddress(
    LPCSTR lpszProc )
 {
    const struct stw_extension_entry *entry;
diff --git a/src/gallium/state_trackers/wgl/icd/stw_icd.h b/src/gallium/state_trackers/wgl/stw_icd.h
index cbc1a66548..02eb543fef 100644
--- a/src/gallium/state_trackers/wgl/icd/stw_icd.h
+++ b/src/gallium/state_trackers/wgl/stw_icd.h
@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2008-2009 Vmware, Inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * 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
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
@@ -388,6 +388,113 @@ typedef struct _GLCLTPROCTABLE
 
 typedef VOID (APIENTRY * PFN_SETPROCTABLE)(PGLCLTPROCTABLE);
 
+/**
+ * Presentation data passed to opengl32!wglCbPresentBuffers.
+ *
+ * Pure software drivers don't need to worry about this -- if they stick to the
+ * GDI API then will integrate with the Desktop Window Manager (DWM) without
+ * problems. Hardware drivers, however, cannot present directly to the primary
+ * surface while the DWM is active, as DWM gets exclusive access to the primary
+ * surface.
+ *
+ * Proper DWM integration requires:
+ * - advertise the PFD_SUPPORT_COMPOSITION flag
+ * - redirect glFlush/glfinish/wglSwapBuffers into a surface shared with the
+ * DWM process.
+ *
+ * @sa http://www.opengl.org/pipeline/article/vol003_7/
+ * @sa http://blogs.msdn.com/greg_schechter/archive/2006/05/02/588934.aspx
+ */
+typedef struct _GLCBPRESENTBUFFERSDATA
+{
+   /**
+    * wglCbPresentBuffers enforces this to be 2.
+    */
+   DWORD magic1;
+
+   /**
+    * wglCbPresentBuffers enforces to be 0 or 1, but it is most commonly
+    * set to 0.
+    */
+   DWORD magic2;
+
+   /**
+    * Locally unique identifier (LUID) of the graphics adapter.
+    *
+    * This should contain the value returned by D3DKMTOpenAdapterFromHdc. It
+    * is passed to dwmapi!DwmpDxGetWindowSharedSurface in order to obtain
+    * the shared surface handle for the bound drawable (window).
+    *
+    * @sa http://msdn.microsoft.com/en-us/library/ms799177.aspx
+    */
+   LUID AdapterLuid;
+
+   /**
+    * This is passed unmodified to DrvPresentBuffers
+    */
+   LPVOID pPrivateData;
+
+   /**
+    * Client area rectangle to update, relative to the window upper-left corner.
+    */
+   RECT rect;
+} GLCBPRESENTBUFFERSDATA, *PGLCBPRESENTBUFFERSDATA;
+
+/**
+ * Callbacks supplied to DrvSetCallbackProcs by the OpenGL runtime.
+ *
+ * Pointers to several callback functions in opengl32.dll.
+ */
+typedef struct _GLCALLBACKTABLE
+{
+   /** Unused */
+   PROC wglCbSetCurrentValue;
+
+   /** Unused */
+   PROC wglCbGetCurrentValue;
+
+   /** Unused */
+   PROC wglCbGetDhglrc;
+
+   /** Unused */
+   PROC wglCbGetDdHandle;
+
+   /**
+    * Queue a present composition.
+    *
+    * Makes the runtime call DrvPresentBuffers with the composition information.
+    */
+   BOOL (APIENTRY *wglCbPresentBuffers)(HDC hdc, PGLCBPRESENTBUFFERSDATA data);
+
+} GLCALLBACKTABLE;
+
+typedef struct _GLPRESENTBUFFERSDATA
+{
+   /**
+    * The shared surface handle.
+    *
+    * Return by dwmapi!DwmpDxGetWindowSharedSurface.
+    *
+    * @sa http://channel9.msdn.com/forums/TechOff/251261-Help-Getting-the-shared-window-texture-out-of-DWM-/
+    */
+   HANDLE hSharedSurface;
+
+   LUID AdapterLuid;
+
+   /**
+    * Present history token.
+    *
+    * This is returned by dwmapi!DwmpDxGetWindowSharedSurface and
+    * should be passed to D3DKMTRender in D3DKMT_RENDER::PresentHistoryToken.
+    *
+    * @sa http://msdn.microsoft.com/en-us/library/ms799176.aspx
+    */
+   ULONGLONG PresentHistoryToken;
+
+   /** Same as GLCBPRESENTBUFFERSDATA::pPrivateData */
+   LPVOID pPrivateData;
+} GLPRESENTBUFFERSDATA, *PGLPRESENTBUFFERSDATA;
+
 BOOL APIENTRY
 DrvCopyContext(
    DHGLRC dhrcSource,
@@ -435,6 +542,9 @@ DrvGetProcAddress(
    LPCSTR lpszProc );
 
 BOOL APIENTRY
+DrvPresentBuffers(HDC hdc, PGLPRESENTBUFFERSDATA data);
+
+BOOL APIENTRY
 DrvRealizeLayerPalette(
    HDC hdc,
    INT iLayerPlane,
diff --git a/src/gallium/state_trackers/wgl/shared/stw_pixelformat.c b/src/gallium/state_trackers/wgl/stw_pixelformat.c
index c296744838..7abe5d9f7f 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_pixelformat.c
+++ b/src/gallium/state_trackers/wgl/stw_pixelformat.c
@@ -34,9 +34,9 @@
 
 #include "util/u_debug.h"
 
+#include "stw_icd.h"
 #include "stw_device.h"
 #include "stw_pixelformat.h"
-#include "stw_public.h"
 #include "stw_tls.h"
 
 
@@ -154,8 +154,11 @@ stw_pixelformat_add(
    pfi->pfd.dwFlags = PFD_SUPPORT_OPENGL;
    
    /* TODO: also support non-native pixel formats */
-   pfi->pfd.dwFlags |= PFD_DRAW_TO_WINDOW ;
-   
+   pfi->pfd.dwFlags |= PFD_DRAW_TO_WINDOW;
+
+   /* See http://www.opengl.org/pipeline/article/vol003_7/ */
+   pfi->pfd.dwFlags |= PFD_SUPPORT_COMPOSITION;
+
    if (doublebuffer)
       pfi->pfd.dwFlags |= PFD_DOUBLEBUFFER | PFD_SWAP_COPY;
    
@@ -288,12 +291,12 @@ stw_pixelformat_visual(GLvisual *visual,
 }
 
 
-int
-stw_pixelformat_describe(
+LONG APIENTRY
+DrvDescribePixelFormat(
    HDC hdc,
-   int iPixelFormat,
-   UINT nBytes,
-   LPPIXELFORMATDESCRIPTOR ppfd )
+   INT iPixelFormat,
+   ULONG cjpfd,
+   PIXELFORMATDESCRIPTOR *ppfd )
 {
    uint count;
    uint index;
@@ -306,7 +309,7 @@ stw_pixelformat_describe(
 
    if (ppfd == NULL)
       return count;
-   if (index >= count || nBytes != sizeof( PIXELFORMATDESCRIPTOR ))
+   if (index >= count || cjpfd != sizeof( PIXELFORMATDESCRIPTOR ))
       return 0;
 
    pfi = stw_pixelformat_get_info( index );
@@ -316,6 +319,52 @@ stw_pixelformat_describe(
    return count;
 }
 
+BOOL APIENTRY
+DrvDescribeLayerPlane(
+   HDC hdc,
+   INT iPixelFormat,
+   INT iLayerPlane,
+   UINT nBytes,
+   LPLAYERPLANEDESCRIPTOR plpd )
+{
+   assert(0);
+   return FALSE;
+}
+
+int APIENTRY
+DrvGetLayerPaletteEntries(
+   HDC hdc,
+   INT iLayerPlane,
+   INT iStart,
+   INT cEntries,
+   COLORREF *pcr )
+{
+   assert(0);
+   return 0;
+}
+
+int APIENTRY
+DrvSetLayerPaletteEntries(
+   HDC hdc,
+   INT iLayerPlane,
+   INT iStart,
+   INT cEntries,
+   CONST COLORREF *pcr )
+{
+   assert(0);
+   return 0;
+}
+
+BOOL APIENTRY
+DrvRealizeLayerPalette(
+   HDC hdc,
+   INT iLayerPlane,
+   BOOL bRealize )
+{
+   assert(0);
+   return FALSE;
+}
+
 /* Only used by the wgl code, but have it here to avoid exporting the
  * pixelformat.h functionality.
  */
diff --git a/src/gallium/state_trackers/wgl/shared/stw_pixelformat.h b/src/gallium/state_trackers/wgl/stw_pixelformat.h
index bec429231b..3a690b35ba 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_pixelformat.h
+++ b/src/gallium/state_trackers/wgl/stw_pixelformat.h
@@ -30,6 +30,10 @@
 
 #include <windows.h>
 
+#ifndef PFD_SUPPORT_COMPOSITION
+#define PFD_SUPPORT_COMPOSITION 0x00008000
+#endif
+
 #include "main/mtypes.h"
 
 #include "pipe/p_compiler.h"
@@ -62,4 +66,11 @@ void
 stw_pixelformat_visual(GLvisual *visual, 
                        const struct stw_pixelformat_info *pfi );
 
+int
+stw_pixelformat_choose( HDC hdc,
+                        CONST PIXELFORMATDESCRIPTOR *ppfd );
+
+int
+stw_pixelformat_get(HDC hdc);
+
 #endif /* STW_PIXELFORMAT_H */
diff --git a/src/gallium/state_trackers/wgl/shared/stw_tls.c b/src/gallium/state_trackers/wgl/stw_tls.c
index 4bd6a9289c..4bd6a9289c 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_tls.c
+++ b/src/gallium/state_trackers/wgl/stw_tls.c
diff --git a/src/gallium/state_trackers/wgl/shared/stw_tls.h b/src/gallium/state_trackers/wgl/stw_tls.h
index fbf8b1cbee..fbf8b1cbee 100644
--- a/src/gallium/state_trackers/wgl/shared/stw_tls.h
+++ b/src/gallium/state_trackers/wgl/stw_tls.h
diff --git a/src/gallium/state_trackers/wgl/wgl/stw_wgl.c b/src/gallium/state_trackers/wgl/stw_wgl.c
index a131292f7a..bb199fdd25 100644
--- a/src/gallium/state_trackers/wgl/wgl/stw_wgl.c
+++ b/src/gallium/state_trackers/wgl/stw_wgl.c
@@ -28,7 +28,9 @@
 #include <windows.h>
 
 #include "util/u_debug.h"
-#include "shared/stw_public.h"
+#include "stw_icd.h"
+#include "stw_context.h"
+#include "stw_pixelformat.h"
 #include "stw_wgl.h"
 
 
@@ -38,16 +40,16 @@ wglCopyContext(
    HGLRC hglrcDst,
    UINT mask )
 {
-   return stw_copy_context( (UINT_PTR)hglrcSrc, 
-                            (UINT_PTR)hglrcDst, 
-                            mask );
+   return DrvCopyContext( (DHGLRC)(UINT_PTR)hglrcSrc,
+                          (DHGLRC)(UINT_PTR)hglrcDst,
+                          mask );
 }
 
 WINGDIAPI HGLRC APIENTRY
 wglCreateContext(
    HDC hdc )
 {
-   return wglCreateLayerContext(hdc, 0);
+   return (HGLRC) DrvCreateContext(hdc);
 }
 
 WINGDIAPI HGLRC APIENTRY
@@ -55,21 +57,21 @@ wglCreateLayerContext(
    HDC hdc,
    int iLayerPlane )
 {
-   return (HGLRC) stw_create_layer_context( hdc, iLayerPlane );
+   return (HGLRC) DrvCreateLayerContext( hdc, iLayerPlane );
 }
 
 WINGDIAPI BOOL APIENTRY
 wglDeleteContext(
    HGLRC hglrc )
 {
-   return stw_delete_context( (UINT_PTR)hglrc );
+   return DrvDeleteContext((DHGLRC)(UINT_PTR)hglrc );
 }
 
 
 WINGDIAPI HGLRC APIENTRY
 wglGetCurrentContext( VOID )
 {
-   return (HGLRC)stw_get_current_context();
+   return (HGLRC)(UINT_PTR)stw_get_current_context();
 }
 
 WINGDIAPI HDC APIENTRY
@@ -83,7 +85,7 @@ wglMakeCurrent(
    HDC hdc,
    HGLRC hglrc )
 {
-   return stw_make_current( hdc, (UINT_PTR)hglrc );
+   return DrvSetContext( hdc, (DHGLRC)(UINT_PTR)hglrc, NULL ) ? TRUE : FALSE;
 }
 
 
@@ -91,7 +93,7 @@ WINGDIAPI BOOL APIENTRY
 wglSwapBuffers(
    HDC hdc )
 {
-   return stw_swap_buffers( hdc );
+   return DrvSwapBuffers( hdc );
 }
 
 
@@ -100,14 +102,14 @@ wglSwapLayerBuffers(
    HDC hdc,
    UINT fuPlanes )
 {
-   return stw_swap_layer_buffers( hdc, fuPlanes );
+   return DrvSwapLayerBuffers( hdc, fuPlanes );
 }
 
 WINGDIAPI PROC APIENTRY
 wglGetProcAddress(
     LPCSTR lpszProc )
 {
-   return stw_get_proc_address( lpszProc );
+   return DrvGetProcAddress( lpszProc );
 }
 
 
@@ -141,7 +143,7 @@ wglDescribePixelFormat(
    UINT nBytes,
    LPPIXELFORMATDESCRIPTOR ppfd )
 {
-   return stw_pixelformat_describe( hdc, iPixelFormat, nBytes, ppfd );
+   return DrvDescribePixelFormat( hdc, iPixelFormat, nBytes, ppfd );
 }
 
 WINGDIAPI int APIENTRY
@@ -160,7 +162,7 @@ wglSetPixelFormat(
    if (ppfd->nSize != sizeof( PIXELFORMATDESCRIPTOR ))
       return FALSE;
 
-   return stw_pixelformat_set( hdc, iPixelFormat );
+   return DrvSetPixelFormat( hdc, iPixelFormat );
 }
 
 
@@ -186,7 +188,8 @@ wglShareLists(
    HGLRC hglrc1,
    HGLRC hglrc2 )
 {
-   return stw_share_lists( (UINT_PTR)hglrc1, (UINT_PTR)hglrc2);;
+   return DrvShareLists((DHGLRC)(UINT_PTR)hglrc1,
+                        (DHGLRC)(UINT_PTR)hglrc2);
 }
 
 WINGDIAPI BOOL APIENTRY
@@ -264,15 +267,7 @@ wglDescribeLayerPlane(
    UINT nBytes,
    LPLAYERPLANEDESCRIPTOR plpd )
 {
-   (void) hdc;
-   (void) iPixelFormat;
-   (void) iLayerPlane;
-   (void) nBytes;
-   (void) plpd;
-
-   assert( 0 );
-
-   return FALSE;
+   return DrvDescribeLayerPlane(hdc, iPixelFormat, iLayerPlane, nBytes, plpd);
 }
 
 WINGDIAPI int APIENTRY
@@ -283,15 +278,7 @@ wglSetLayerPaletteEntries(
    int cEntries,
    CONST COLORREF *pcr )
 {
-   (void) hdc;
-   (void) iLayerPlane;
-   (void) iStart;
-   (void) cEntries;
-   (void) pcr;
-
-   assert( 0 );
-
-   return 0;
+   return DrvSetLayerPaletteEntries(hdc, iLayerPlane, iStart, cEntries, pcr);
 }
 
 WINGDIAPI int APIENTRY
@@ -302,15 +289,7 @@ wglGetLayerPaletteEntries(
    int cEntries,
    COLORREF *pcr )
 {
-   (void) hdc;
-   (void) iLayerPlane;
-   (void) iStart;
-   (void) cEntries;
-   (void) pcr;
-
-   assert( 0 );
-
-   return 0;
+   return DrvGetLayerPaletteEntries(hdc, iLayerPlane, iStart, cEntries, pcr);
 }
 
 WINGDIAPI BOOL APIENTRY
diff --git a/src/gallium/state_trackers/wgl/wgl/stw_wgl.h b/src/gallium/state_trackers/wgl/stw_wgl.h
index a98179944a..a98179944a 100644
--- a/src/gallium/state_trackers/wgl/wgl/stw_wgl.h
+++ b/src/gallium/state_trackers/wgl/stw_wgl.h
diff --git a/src/gallium/state_trackers/wgl/stw_winsys.h b/src/gallium/state_trackers/wgl/stw_winsys.h
new file mode 100644
index 0000000000..1ead47d6e6
--- /dev/null
+++ b/src/gallium/state_trackers/wgl/stw_winsys.h
@@ -0,0 +1,109 @@
+/**************************************************************************
+ *
+ * Copyright 2008-2009 Vmware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef STW_WINSYS_H
+#define STW_WINSYS_H
+
+#include <windows.h> /* for HDC */
+
+#include "pipe/p_compiler.h"
+
+struct pipe_screen;
+struct pipe_context;
+struct pipe_surface;
+
+struct stw_shared_surface;
+
+struct stw_winsys
+{
+   struct pipe_screen *
+   (*create_screen)( void );
+
+   struct pipe_context *
+   (*create_context)( struct pipe_screen *screen );
+
+   /**
+    * Present the color buffer to the window associated with the device context.
+    */
+   void
+   (*present)( struct pipe_screen *screen,
+               struct pipe_surface *surf,
+               HDC hDC );
+
+   /**
+    * Locally unique identifier (LUID) of the graphics adapter.
+    *
+    * @sa GLCBPRESENTBUFFERSDATA::AdapterLuid;
+    */
+   boolean
+   (*get_adapter_luid)( struct pipe_screen *screen,
+                        LUID *pAdapterLuid );
+
+   /**
+    * Open a shared surface (optional).
+    *
+    * @sa GLCBPRESENTBUFFERSDATA::hSharedSurface;
+    */
+   struct stw_shared_surface *
+   (*shared_surface_open)(struct pipe_screen *screen,
+                          HANDLE hSharedSurface);
+
+   /**
+    * Open a shared surface (optional).
+    */
+   void
+   (*shared_surface_close)(struct pipe_screen *screen,
+                           struct stw_shared_surface *surface);
+
+   /**
+    * Compose into a shared (optional).
+    *
+    * Blit the color buffer into a shared surface.
+    *
+    * @sa GLPRESENTBUFFERSDATA::PresentHistoryToken.
+    */
+   void
+   (*compose)( struct pipe_screen *screen,
+               struct pipe_surface *src,
+               struct stw_shared_surface *dest,
+               LPCRECT pRect,
+               ULONGLONG PresentHistoryToken );
+};
+
+boolean
+stw_init(const struct stw_winsys *stw_winsys);
+
+boolean
+stw_init_thread(void);
+
+void
+stw_cleanup_thread(void);
+
+void
+stw_cleanup(void);
+
+#endif /* STW_WINSYS_H */
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c
index 66ca4cb590..7037d17e43 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.c
+++ b/src/gallium/state_trackers/xorg/xorg_composite.c
@@ -57,24 +57,6 @@ pixel_to_float4(Pixel pixel, float *color)
    color[3] = ((float)a) / 255.;
 }
 
-static INLINE void
-render_pixel_to_float4(PictFormatPtr format,
-                       CARD32 pixel, float *color)
-{
-   CARD32	    r, g, b, a;
-
-   debug_assert(format->type == PictTypeDirect);
-
-   r = (pixel >> format->direct.red) & format->direct.redMask;
-   g = (pixel >> format->direct.green) & format->direct.greenMask;
-   b = (pixel >> format->direct.blue) & format->direct.blueMask;
-   a = (pixel >> format->direct.alpha) & format->direct.alphaMask;
-   color[0] = ((float)r) / ((float)format->direct.redMask);
-   color[1] = ((float)g) / ((float)format->direct.greenMask);
-   color[2] = ((float)b) / ((float)format->direct.blueMask);
-   color[3] = ((float)a) / ((float)format->direct.alphaMask);
-}
-
 struct acceleration_info {
    int op : 16;
    int with_mask : 1;
@@ -150,24 +132,22 @@ setup_vertex_data0(struct exa_context *ctx,
                    int srcX, int srcY, int maskX, int maskY,
                    int dstX, int dstY, int width, int height)
 {
-   float vertices[4][2][4];
-
    /* 1st vertex */
-   setup_vertex0(vertices[0], dstX, dstY,
+   setup_vertex0(ctx->vertices2[0], dstX, dstY,
                  ctx->solid_color);
    /* 2nd vertex */
-   setup_vertex0(vertices[1], dstX + width, dstY,
+   setup_vertex0(ctx->vertices2[1], dstX + width, dstY,
                  ctx->solid_color);
    /* 3rd vertex */
-   setup_vertex0(vertices[2], dstX + width, dstY + height,
+   setup_vertex0(ctx->vertices2[2], dstX + width, dstY + height,
                  ctx->solid_color);
    /* 4th vertex */
-   setup_vertex0(vertices[3], dstX, dstY + height,
+   setup_vertex0(ctx->vertices2[3], dstX, dstY + height,
                  ctx->solid_color);
 
    return pipe_user_buffer_create(ctx->pipe->screen,
-                                  vertices,
-                                  sizeof(vertices));
+                                  ctx->vertices2,
+                                  sizeof(ctx->vertices2));
 }
 
 static INLINE void
@@ -189,7 +169,6 @@ setup_vertex_data1(struct exa_context *ctx,
                    int srcX, int srcY, int maskX, int maskY,
                    int dstX, int dstY, int width, int height)
 {
-   float vertices[4][2][4];
    float s0, t0, s1, t1;
    struct pipe_texture *src = ctx->bound_textures[0];
 
@@ -199,21 +178,21 @@ setup_vertex_data1(struct exa_context *ctx,
    t1 = srcY + height / src->height[0];
 
    /* 1st vertex */
-   setup_vertex1(vertices[0], dstX, dstY,
+   setup_vertex1(ctx->vertices2[0], dstX, dstY,
                  s0, t0);
    /* 2nd vertex */
-   setup_vertex1(vertices[1], dstX + width, dstY,
+   setup_vertex1(ctx->vertices2[1], dstX + width, dstY,
                  s1, t0);
    /* 3rd vertex */
-   setup_vertex1(vertices[2], dstX + width, dstY + height,
+   setup_vertex1(ctx->vertices2[2], dstX + width, dstY + height,
                  s1, t1);
    /* 4th vertex */
-   setup_vertex1(vertices[3], dstX, dstY + height,
+   setup_vertex1(ctx->vertices2[3], dstX, dstY + height,
                  s0, t1);
 
    return pipe_user_buffer_create(ctx->pipe->screen,
-                                  vertices,
-                                  sizeof(vertices));
+                                  ctx->vertices2,
+                                  sizeof(ctx->vertices2));
 }
 
 static struct pipe_buffer *
@@ -222,24 +201,22 @@ setup_vertex_data_tex(struct exa_context *ctx,
                       float s0, float t0, float s1, float t1,
                       float z)
 {
-   float vertices[4][2][4];
-
    /* 1st vertex */
-   setup_vertex1(vertices[0], x0, y0,
+   setup_vertex1(ctx->vertices2[0], x0, y0,
                  s0, t0);
    /* 2nd vertex */
-   setup_vertex1(vertices[1], x1, y0,
+   setup_vertex1(ctx->vertices2[1], x1, y0,
                  s1, t0);
    /* 3rd vertex */
-   setup_vertex1(vertices[2], x1, y1,
+   setup_vertex1(ctx->vertices2[2], x1, y1,
                  s1, t1);
    /* 4th vertex */
-   setup_vertex1(vertices[3], x0, y1,
+   setup_vertex1(ctx->vertices2[3], x0, y1,
                  s0, t1);
 
    return pipe_user_buffer_create(ctx->pipe->screen,
-                                  vertices,
-                                  sizeof(vertices));
+                                  ctx->vertices2,
+                                  sizeof(ctx->vertices2));
 }
 
 
@@ -269,7 +246,6 @@ setup_vertex_data2(struct exa_context *ctx,
                    int srcX, int srcY, int maskX, int maskY,
                    int dstX, int dstY, int width, int height)
 {
-   float vertices[4][3][4];
    float st0[4], st1[4];
    struct pipe_texture *src = ctx->bound_textures[0];
    struct pipe_texture *mask = ctx->bound_textures[0];
@@ -285,21 +261,21 @@ setup_vertex_data2(struct exa_context *ctx,
    st1[3] = maskY + height / mask->height[0];
 
    /* 1st vertex */
-   setup_vertex2(vertices[0], dstX, dstY,
+   setup_vertex2(ctx->vertices3[0], dstX, dstY,
                  st0[0], st0[1], st1[0], st1[1]);
    /* 2nd vertex */
-   setup_vertex2(vertices[1], dstX + width, dstY,
+   setup_vertex2(ctx->vertices3[1], dstX + width, dstY,
                  st0[2], st0[1], st1[2], st1[1]);
    /* 3rd vertex */
-   setup_vertex2(vertices[2], dstX + width, dstY + height,
+   setup_vertex2(ctx->vertices3[2], dstX + width, dstY + height,
                  st0[2], st0[3], st1[2], st1[3]);
    /* 4th vertex */
-   setup_vertex2(vertices[3], dstX, dstY + height,
+   setup_vertex2(ctx->vertices3[3], dstX, dstY + height,
                  st0[0], st0[3], st1[0], st1[3]);
 
    return pipe_user_buffer_create(ctx->pipe->screen,
-                                  vertices,
-                                  sizeof(vertices));
+                                  ctx->vertices3,
+                                  sizeof(ctx->vertices3));
 }
 
 boolean xorg_composite_accelerated(int op,
@@ -311,22 +287,25 @@ boolean xorg_composite_accelerated(int op,
    unsigned accel_ops_count =
       sizeof(accelerated_ops)/sizeof(struct acceleration_info);
 
-
-   /*FIXME: currently accel is disabled */
-   return FALSE;
-
-   if (pSrcPicture) {
-      /* component alpha not supported */
-      if (pSrcPicture->componentAlpha)
+   if (pSrcPicture->pSourcePict) {
+      /* Gradients not yet supported */
+      if (pSrcPicture->pSourcePict->type != SourcePictTypeSolidFill)
          return FALSE;
-      /* fills not supported */
-      if (pSrcPicture->pSourcePict)
+
+      /* Solid source with mask not yet handled properly */
+      if (pMaskPicture)
          return FALSE;
    }
 
    for (i = 0; i < accel_ops_count; ++i) {
       if (op == accelerated_ops[i].op) {
-         if (pMaskPicture && !accelerated_ops[i].with_mask)
+         /* Check for unsupported component alpha */
+         if ((pSrcPicture->componentAlpha &&
+              !accelerated_ops[i].component_alpha) ||
+             (pMaskPicture &&
+              (!accelerated_ops[i].with_mask ||
+               (pMaskPicture->componentAlpha &&
+                !accelerated_ops[i].component_alpha))))
             return FALSE;
          return TRUE;
       }
@@ -359,6 +338,9 @@ bind_framebuffer_state(struct exa_context *exa, struct exa_pixmap_priv *pDst)
    state.zsbuf = 0;
 
    cso_set_framebuffer(exa->cso, &state);
+
+   /* we do fire and forget for the framebuffer, this is the forget part */
+   pipe_surface_reference(&surface, NULL);
 }
 
 enum AxisOrientation {
@@ -391,7 +373,7 @@ bind_viewport_state(struct exa_context *exa, struct exa_pixmap_priv *pDst)
    int width = pDst->tex->width[0];
    int height = pDst->tex->height[0];
 
-   debug_printf("Bind viewport (%d, %d)\n", width, height);
+   /*debug_printf("Bind viewport (%d, %d)\n", width, height);*/
 
    set_viewport(exa, width, height, Y0_TOP);
 }
@@ -400,14 +382,9 @@ static void
 bind_blend_state(struct exa_context *exa, int op,
                  PicturePtr pSrcPicture, PicturePtr pMaskPicture)
 {
-   boolean component_alpha = (pSrcPicture) ?
-                             pSrcPicture->componentAlpha : FALSE;
    struct xorg_composite_blend blend_opt;
    struct pipe_blend_state blend;
 
-   if (component_alpha) {
-      op = PictOpOver;
-   }
    blend_opt = blend_for_op(op);
 
    memset(&blend, 0, sizeof(struct pipe_blend_state));
@@ -448,9 +425,9 @@ bind_shaders(struct exa_context *exa, int op,
          if (pSrcPicture->pSourcePict->type == SourcePictTypeSolidFill) {
             fs_traits |= FS_SOLID_FILL;
             vs_traits |= VS_SOLID_FILL;
-            render_pixel_to_float4(pSrcPicture->pFormat,
-                                   pSrcPicture->pSourcePict->solidFill.color,
-                                   exa->solid_color);
+            debug_assert(pSrcPicture->format == PICT_a8r8g8b8);
+            pixel_to_float4(pSrcPicture->pSourcePict->solidFill.color,
+                            exa->solid_color);
             exa->has_solid_color = TRUE;
          } else {
             debug_assert("!gradients not supported");
@@ -488,6 +465,12 @@ bind_samplers(struct exa_context *exa, int op,
    memset(&src_sampler, 0, sizeof(struct pipe_sampler_state));
    memset(&mask_sampler, 0, sizeof(struct pipe_sampler_state));
 
+   if ((pSrc && exa->pipe->is_texture_referenced(exa->pipe, pSrc->tex, 0, 0) &
+        PIPE_REFERENCED_FOR_WRITE) ||
+       (pMask && exa->pipe->is_texture_referenced(exa->pipe, pMask->tex, 0, 0) &
+        PIPE_REFERENCED_FOR_WRITE))
+      exa->pipe->flush(exa->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
+
    if (pSrcPicture && pSrc) {
       unsigned src_wrap = render_repeat_to_gallium(
          pSrcPicture->repeatType);
@@ -675,7 +658,7 @@ boolean xorg_solid_bind_state(struct exa_context *exa,
    cso_set_vertex_shader_handle(exa->cso, shader.vs);
    cso_set_fragment_shader_handle(exa->cso, shader.fs);
 
-   return FALSE;
+   return TRUE;
 }
 
 void xorg_solid(struct exa_context *exa,
@@ -684,28 +667,26 @@ void xorg_solid(struct exa_context *exa,
 {
    struct pipe_context *pipe = exa->pipe;
    struct pipe_buffer *buf = 0;
-   float vertices[4][2][4];
 
    /* 1st vertex */
-   setup_vertex0(vertices[0], x0, y0,
+   setup_vertex0(exa->vertices2[0], x0, y0,
                  exa->solid_color);
    /* 2nd vertex */
-   setup_vertex0(vertices[1], x1, y0,
+   setup_vertex0(exa->vertices2[1], x1, y0,
                  exa->solid_color);
    /* 3rd vertex */
-   setup_vertex0(vertices[2], x1, y1,
+   setup_vertex0(exa->vertices2[2], x1, y1,
                  exa->solid_color);
    /* 4th vertex */
-   setup_vertex0(vertices[3], x0, y1,
+   setup_vertex0(exa->vertices2[3], x0, y1,
                  exa->solid_color);
 
    buf = pipe_user_buffer_create(exa->pipe->screen,
-                                 vertices,
-                                 sizeof(vertices));
+                                 exa->vertices2,
+                                 sizeof(exa->vertices2));
 
 
    if (buf) {
-      debug_printf("Drawing buf is %p\n", buf);
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
@@ -848,7 +829,7 @@ static void renderer_copy_texture(struct exa_context *exa,
    t1 = 1;
 #endif
 
-#if 1
+#if 0
    debug_printf("copy texture src=[%f, %f, %f, %f], dst=[%f, %f, %f, %f], tex=[%f, %f, %f, %f]\n",
                 sx1, sy1, sx2, sy2, dx1, dy1, dx2, dy2,
                 s0, t0, s1, t1);
@@ -902,6 +883,8 @@ static void renderer_copy_texture(struct exa_context *exa,
    /* texture */
    cso_set_sampler_textures(exa->cso, 1, &src);
 
+   bind_rasterizer_state(exa);
+
    /* shaders */
    shader = xorg_shaders_get(exa->shaders,
                              VS_COMPOSITE,
@@ -1018,7 +1001,9 @@ void xorg_copy_pixmap(struct exa_context *ctx,
    struct pipe_texture *dst = dst_priv->tex;
    struct pipe_texture *src = src_priv->tex;
 
-   xorg_exa_finish(ctx);
+   if (ctx->pipe->is_texture_referenced(ctx->pipe, src, 0, 0) &
+       PIPE_REFERENCED_FOR_WRITE)
+      ctx->pipe->flush(ctx->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 
    dst_loc[0] = dx;
    dst_loc[1] = dy;
diff --git a/src/gallium/state_trackers/xorg/xorg_driver.c b/src/gallium/state_trackers/xorg/xorg_driver.c
index 643b6b3b9e..4bc87aa613 100644
--- a/src/gallium/state_trackers/xorg/xorg_driver.c
+++ b/src/gallium/state_trackers/xorg/xorg_driver.c
@@ -558,6 +558,7 @@ ScreenInit(int scrnIndex, ScreenPtr pScreen, int argc, char **argv)
     xf86SetBlackWhitePixels(pScreen);
 
     ms->exa = xorg_exa_init(pScrn);
+    ms->debug_fallback = debug_get_bool_option("XORG_DEBUG_FALLBACK", TRUE);
 
     miInitializeBackingStore(pScreen);
     xf86SetBackingStore(pScreen);
@@ -725,8 +726,10 @@ CloseScreen(int scrnIndex, ScreenPtr pScreen)
     if (ms->exa)
 	xorg_exa_close(pScrn);
 
+    if (ms->api->destroy)
 	ms->api->destroy(ms->api);
-	ms->api = NULL;
+    ms->api = NULL;
+
     drmClose(ms->fd);
     ms->fd = -1;
 
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index dea9f4c2bc..f7949bafaa 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -47,7 +47,9 @@
 
 #include "util/u_rect.h"
 
+#define DEBUG_PRINT 0
 #define DEBUG_SOLID 0
+#define DISABLE_ACCEL 0
 
 /*
  * Helper functions
@@ -74,7 +76,7 @@ exa_get_pipe_format(int depth, enum pipe_format *format, int *bbp)
 	assert(*bbp == 16);
 	break;
     case 8:
-	*format = PIPE_FORMAT_A8_UNORM;
+	*format = PIPE_FORMAT_I8_UNORM;
 	assert(*bbp == 8);
 	break;
     case 4:
@@ -200,7 +202,7 @@ ExaPrepareAccess(PixmapPtr pPix, int index)
     if (!priv->tex)
 	return FALSE;
 
-    if (priv->map_count++ == 0)
+    if (priv->map_count == 0)
     {
 	if (exa->pipe->is_texture_referenced(exa->pipe, priv->tex, 0, 0) &
 	    PIPE_REFERENCED_FOR_WRITE)
@@ -208,14 +210,21 @@ ExaPrepareAccess(PixmapPtr pPix, int index)
 
 	priv->map_transfer =
 	    exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0,
+#ifdef EXA_MIXED_PIXMAPS
+					PIPE_TRANSFER_MAP_DIRECTLY |
+#endif
 					PIPE_TRANSFER_READ_WRITE,
 					0, 0, priv->tex->width[0], priv->tex->height[0]);
+        if (!priv->map_transfer)
+	    return FALSE;
 
 	pPix->devPrivate.ptr =
 	    exa->scrn->transfer_map(exa->scrn, priv->map_transfer);
 	pPix->devKind = priv->map_transfer->stride;
     }
 
+    priv->map_count++;
+
     return TRUE;
 }
 
@@ -281,36 +290,36 @@ ExaPrepareSolid(PixmapPtr pPixmap, int alu, Pixel planeMask, Pixel fg)
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap);
     struct exa_context *exa = ms->exa;
 
-#if 0
-    debug_printf("ExaPrepareSolid - test\n");
+#if DEBUG_PRINT
+    debug_printf("ExaPrepareSolid(0x%x)\n", fg);
 #endif
-    if (!EXA_PM_IS_SOLID(&pPixmap->drawable, planeMask))
-	return FALSE;
+    if (!exa->pipe)
+	XORG_FALLBACK("solid accle not enabled");
 
     if (!priv || !priv->tex)
-	return FALSE;
+	XORG_FALLBACK("solid !priv || !priv->tex");
 
-    if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
-                                        priv->tex->target,
-                                        PIPE_TEXTURE_USAGE_RENDER_TARGET, 0))
-	return FALSE;
+    if (!EXA_PM_IS_SOLID(&pPixmap->drawable, planeMask))
+	XORG_FALLBACK("solid planeMask is not solid");
 
     if (alu != GXcopy)
-	return FALSE;
-
-    if (!exa->pipe)
-	return FALSE;
+	XORG_FALLBACK("solid not GXcopy");
 
+    if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
+                                        priv->tex->target,
+                                        PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) {
+	XORG_FALLBACK("solid bad format %s", pf_name(priv->tex->format));
+    }
 
 #if DEBUG_SOLID
     fg = 0xffff0000;
 #endif
 
-#if 1
-    debug_printf("  ExaPrepareSolid(0x%x)\n", fg);
-#endif
-
+#if DISABLE_ACCEL
+    return FALSE;
+#else
     return xorg_solid_bind_state(exa, priv, fg);
+#endif
 }
 
 static void
@@ -321,7 +330,9 @@ ExaSolid(PixmapPtr pPixmap, int x0, int y0, int x1, int y1)
     struct exa_context *exa = ms->exa;
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap);
 
+#if DEBUG_PRINT
     debug_printf("\tExaSolid(%d, %d, %d, %d)\n", x0, y0, x1, y1);
+#endif
 
 #if 0
     if (x0 == 0 && y0 == 0 &&
@@ -375,36 +386,42 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap);
     struct exa_pixmap_priv *src_priv = exaGetPixmapDriverPrivate(pSrcPixmap);
 
+#if DEBUG_PRINT
     debug_printf("ExaPrepareCopy\n");
+#endif
+    if (!exa->pipe)
+	XORG_FALLBACK("copy accle not enabled");
 
-    if (alu != GXcopy)
-	return FALSE;
+    if (!priv || !src_priv)
+	XORG_FALLBACK("copy !priv || !src_priv");
+
+    if (!priv->tex || !src_priv->tex)
+	XORG_FALLBACK("copy !priv->tex || !src_priv->tex");
 
     if (!EXA_PM_IS_SOLID(&pSrcPixmap->drawable, planeMask))
-	return FALSE;
+	XORG_FALLBACK("copy planeMask is not solid");
 
-    if (!priv || !src_priv)
-	return FALSE;
+    if (alu != GXcopy)
+	XORG_FALLBACK("copy alu not GXcopy");
 
     if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
                                         priv->tex->target,
-                                        PIPE_TEXTURE_USAGE_RENDER_TARGET, 0) ||
-        !exa->scrn->is_format_supported(exa->scrn, src_priv->tex->format,
+                                        PIPE_TEXTURE_USAGE_RENDER_TARGET, 0))
+	XORG_FALLBACK("copy pDst format %s", pf_name(priv->tex->format));
+
+    if (!exa->scrn->is_format_supported(exa->scrn, src_priv->tex->format,
                                         src_priv->tex->target,
                                         PIPE_TEXTURE_USAGE_SAMPLER, 0))
-	return FALSE;
-
-    if (!priv->tex || !src_priv->tex)
-	return FALSE;
-
-    if (!exa->pipe)
-	return FALSE;
+	XORG_FALLBACK("copy pSrc format %s", pf_name(src_priv->tex->format));
 
     exa->copy.src = src_priv;
     exa->copy.dst = priv;
 
-    /*XXX disabled until some issues with syncing are fixed */
+#if DISABLE_ACCEL
     return FALSE;
+#else
+    return TRUE;
+#endif
 }
 
 static void
@@ -416,8 +433,10 @@ ExaCopy(PixmapPtr pDstPixmap, int srcX, int srcY, int dstX, int dstY,
    struct exa_context *exa = ms->exa;
    struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap);
 
+#if DEBUG_PRINT
    debug_printf("\tExaCopy(srcx=%d, srcy=%d, dstX=%d, dstY=%d, w=%d, h=%d)\n",
                 srcX, srcY, dstX, dstY, width, height);
+#endif
 
    debug_assert(priv == exa->copy.dst);
 
@@ -434,14 +453,55 @@ ExaPrepareComposite(int op, PicturePtr pSrcPicture,
    ScrnInfoPtr pScrn = xf86Screens[pDst->drawable.pScreen->myNum];
    modesettingPtr ms = modesettingPTR(pScrn);
    struct exa_context *exa = ms->exa;
+   struct exa_pixmap_priv *priv;
 
+#if DEBUG_PRINT
    debug_printf("ExaPrepareComposite\n");
+#endif
+   if (!exa->pipe)
+      XORG_FALLBACK("comp accle not enabled");
+
+   priv = exaGetPixmapDriverPrivate(pDst);
+   if (!priv || !priv->tex)
+      XORG_FALLBACK("comp pDst %s", !priv ? "!priv" : "!priv->tex");
+
+   if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
+                                       priv->tex->target,
+                                       PIPE_TEXTURE_USAGE_RENDER_TARGET, 0))
+      XORG_FALLBACK("copy pDst format: %s", pf_name(priv->tex->format));
+
+   if (pSrc) {
+      priv = exaGetPixmapDriverPrivate(pSrc);
+      if (!priv || !priv->tex)
+         XORG_FALLBACK("comp pSrc %s", !priv ? "!priv" : "!priv->tex");
+
+      if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
+                                          priv->tex->target,
+                                          PIPE_TEXTURE_USAGE_SAMPLER, 0))
+         XORG_FALLBACK("copy pSrc format: %s", pf_name(priv->tex->format));
+   }
+
+   if (pMask) {
+      priv = exaGetPixmapDriverPrivate(pMask);
+      if (!priv || !priv->tex)
+         XORG_FALLBACK("comp pMask %s", !priv ? "!priv" : "!priv->tex");
+
+      if (!exa->scrn->is_format_supported(exa->scrn, priv->tex->format,
+                                          priv->tex->target,
+                                          PIPE_TEXTURE_USAGE_SAMPLER, 0))
+         XORG_FALLBACK("copy pMask format: %s", pf_name(priv->tex->format));
+   }
 
+#if DISABLE_ACCEL
+   (void) exa;
+   return FALSE;
+#else
    return xorg_composite_bind_state(exa, op, pSrcPicture, pMaskPicture,
                                     pDstPicture,
-                                    exaGetPixmapDriverPrivate(pSrc),
-                                    exaGetPixmapDriverPrivate(pMask),
+                                    pSrc ? exaGetPixmapDriverPrivate(pSrc) : NULL,
+                                    pMask ? exaGetPixmapDriverPrivate(pMask) : NULL,
                                     exaGetPixmapDriverPrivate(pDst));
+#endif
 }
 
 static void
@@ -453,7 +513,9 @@ ExaComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY,
    struct exa_context *exa = ms->exa;
    struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDst);
 
+#if DEBUG_PRINT
    debug_printf("\tExaComposite\n");
+#endif
 
    xorg_composite(exa, priv, srcX, srcY, maskX, maskY,
                   dstX, dstY, width, height);
@@ -464,10 +526,15 @@ ExaCheckComposite(int op,
 		  PicturePtr pSrcPicture, PicturePtr pMaskPicture,
 		  PicturePtr pDstPicture)
 {
-   return xorg_composite_accelerated(op,
-                                     pSrcPicture,
-                                     pMaskPicture,
-                                     pDstPicture);
+   boolean accelerated = xorg_composite_accelerated(op,
+                                                    pSrcPicture,
+                                                    pMaskPicture,
+                                                    pDstPicture);
+#if DEBUG_PRINT
+   debug_printf("ExaCheckComposite(%d, %p, %p, %p) = %d\n",
+                op, pSrcPicture, pMaskPicture, pDstPicture, accelerated);
+#endif
+   return accelerated;
 }
 
 static void *
@@ -610,67 +677,37 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height,
          priv->tex->height[0] != height ||
          priv->tex_flags != priv->flags)) {
 	struct pipe_texture *texture = NULL;
-
-#ifdef DRM_MODE_FEATURE_DIRTYFB
-	if (priv->flags)
-#endif
-	{
-	    struct pipe_texture template;
-
-	    memset(&template, 0, sizeof(template));
-	    template.target = PIPE_TEXTURE_2D;
-	    exa_get_pipe_format(depth, &template.format, &bitsPerPixel);
-	    pf_get_block(template.format, &template.block);
-	    template.width[0] = width;
-	    template.height[0] = height;
-	    template.depth[0] = 1;
-	    template.last_level = 0;
-	    template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET | priv->flags;
-	    priv->tex_flags = priv->flags;
-	    texture = exa->scrn->texture_create(exa->scrn, &template);
-
-	    if (priv->tex) {
-		struct pipe_surface *dst_surf;
-                struct pipe_surface *src_surf;
-
-		dst_surf = exa->scrn->get_tex_surface(
-                   exa->scrn, texture, 0, 0, 0, PIPE_BUFFER_USAGE_GPU_WRITE);
-		src_surf = exa_gpu_surface(exa, priv);
-		exa->pipe->surface_copy(exa->pipe, dst_surf, 0, 0, src_surf,
-                                        0, 0, min(width, texture->width[0]),
-                                        min(height, texture->height[0]));
-		exa->scrn->tex_surface_destroy(dst_surf);
-		exa->scrn->tex_surface_destroy(src_surf);
-	    } else if (pPixmap->devPrivate.ptr) {
-		struct pipe_transfer *transfer;
-
-		if (priv->map_count != 0)
-		     FatalError("doing ExaModifyPixmapHeader on mapped buffer\n");
-
-		transfer =
-		    exa->scrn->get_tex_transfer(exa->scrn, texture, 0, 0, 0,
-						PIPE_TRANSFER_WRITE,
-						0, 0, width, height);
-		util_copy_rect(exa->scrn->transfer_map(exa->scrn, transfer),
-			       &texture->block, transfer->stride, 0, 0,
-			       width, height, pPixmap->devPrivate.ptr,
-			       pPixmap->devKind, 0, 0);
-		exa->scrn->transfer_unmap(exa->scrn, transfer);
-		exa->scrn->tex_transfer_destroy(transfer);
-
-		xfree(pPixmap->devPrivate.ptr);
-		pPixmap->devPrivate.ptr = NULL;
-	    }
+	struct pipe_texture template;
+
+	memset(&template, 0, sizeof(template));
+	template.target = PIPE_TEXTURE_2D;
+	exa_get_pipe_format(depth, &template.format, &bitsPerPixel);
+	pf_get_block(template.format, &template.block);
+	template.width[0] = width;
+	template.height[0] = height;
+	template.depth[0] = 1;
+	template.last_level = 0;
+	template.tex_usage = PIPE_TEXTURE_USAGE_RENDER_TARGET | priv->flags;
+	priv->tex_flags = priv->flags;
+	texture = exa->scrn->texture_create(exa->scrn, &template);
+
+	if (priv->tex) {
+	    struct pipe_surface *dst_surf;
+	    struct pipe_surface *src_surf;
+
+	    dst_surf = exa->scrn->get_tex_surface(
+		exa->scrn, texture, 0, 0, 0, PIPE_BUFFER_USAGE_GPU_WRITE);
+	    src_surf = exa_gpu_surface(exa, priv);
+	    exa->pipe->surface_copy(exa->pipe, dst_surf, 0, 0, src_surf,
+				    0, 0, min(width, texture->width[0]),
+				    min(height, texture->height[0]));
+	    exa->scrn->tex_surface_destroy(dst_surf);
+	    exa->scrn->tex_surface_destroy(src_surf);
 	}
-#ifdef DRM_MODE_FEATURE_DIRTYFB
-	else {
-	    xfree(pPixmap->devPrivate.ptr);
-	    pPixmap->devPrivate.ptr = xalloc(pPixmap->drawable.height *
-					     pPixmap->devKind);
-	}
-#endif
 
 	pipe_texture_reference(&priv->tex, texture);
+	/* the texture we create has one reference */
+	pipe_texture_reference(&texture, NULL);
     }
 
     return TRUE;
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.h b/src/gallium/state_trackers/xorg/xorg_exa.h
index 43949b04a4..d3f25ca844 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.h
+++ b/src/gallium/state_trackers/xorg/xorg_exa.h
@@ -32,8 +32,11 @@ struct exa_context
       struct exa_pixmap_priv *src;
       struct exa_pixmap_priv *dst;
    } copy;
-};
 
+   /* we should combine these two */
+   float vertices2[4][2][4];
+   float vertices3[4][3][4];
+};
 
 struct exa_pixmap_priv
 {
@@ -47,6 +50,15 @@ struct exa_pixmap_priv
    unsigned map_count;
 };
 
+#define XORG_FALLBACK(s, arg...)              \
+do {                                          \
+   if (ms->debug_fallback) {                  \
+      xf86DrvMsg(pScrn->scrnIndex, X_INFO,    \
+                 "fallback: " s "\n", ##arg); \
+   }                                          \
+   return FALSE;                              \
+} while(0)
+
 struct pipe_surface *
 exa_gpu_surface(struct exa_context *exa, struct exa_pixmap_priv *priv);
 
diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
index 2daa5b5628..bb5a42af37 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
@@ -271,7 +271,7 @@ create_vs(struct pipe_context *pipe,
 
    if (has_mask) {
       src = ureg_DECL_vs_input(ureg, input_slot++);
-      dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 2);
+      dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 2);
       ureg_MOV(ureg, dst, src);
    }
 
@@ -330,7 +330,7 @@ create_fs(struct pipe_context *pipe,
    if (has_mask) {
       mask_sampler = ureg_DECL_sampler(ureg, 1);
       mask_pos = ureg_DECL_fs_input(ureg,
-                                    TGSI_SEMANTIC_POSITION,
+                                    TGSI_SEMANTIC_GENERIC,
                                     1,
                                     TGSI_INTERPOLATE_PERSPECTIVE);
    }
@@ -465,7 +465,7 @@ struct xorg_shader xorg_shaders_get(struct xorg_shaders *sc,
                                     unsigned vs_traits,
                                     unsigned fs_traits)
 {
-   struct xorg_shader shader = {0};
+   struct xorg_shader shader = { NULL, NULL };
    void *vs, *fs;
 
    vs = shader_from_cache(sc->exa->pipe, PIPE_SHADER_VERTEX,
diff --git a/src/gallium/state_trackers/xorg/xorg_tracker.h b/src/gallium/state_trackers/xorg/xorg_tracker.h
index b1ab783a15..db2f16f63e 100644
--- a/src/gallium/state_trackers/xorg/xorg_tracker.h
+++ b/src/gallium/state_trackers/xorg/xorg_tracker.h
@@ -94,6 +94,7 @@ typedef struct _modesettingRec
     /* exa */
     void *exa;
     Bool noEvict;
+    Bool debug_fallback;
 
 #ifdef DRM_MODE_FEATURE_DIRTYFB
     DamagePtr damage;
@@ -151,5 +152,11 @@ crtc_cursor_destroy(xf86CrtcPtr crtc);
 void
 output_init(ScrnInfoPtr pScrn);
 
+/***********************************************************************
+ * xorg_xv.c
+ */
+void
+xorg_init_video(ScreenPtr pScreen);
+
 
 #endif /* _XORG_TRACKER_H_ */
diff --git a/src/gallium/state_trackers/xorg/xorg_xv.c b/src/gallium/state_trackers/xorg/xorg_xv.c
new file mode 100644
index 0000000000..88955d47fd
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xorg_xv.c
@@ -0,0 +1,212 @@
+#include "xorg_tracker.h"
+
+#include <xf86xv.h>
+#include <X11/extensions/Xv.h>
+#include <fourcc.h>
+
+/*XXX get these from pipe's texture limits */
+#define IMAGE_MAX_WIDTH		2048
+#define IMAGE_MAX_HEIGHT	2048
+
+#define MAKE_ATOM(a) MakeAtom(a, sizeof(a) - 1, TRUE)
+
+static Atom xvBrightness, xvContrast;
+
+#define NUM_TEXTURED_ATTRIBUTES 2
+static XF86AttributeRec TexturedAttributes[NUM_TEXTURED_ATTRIBUTES] = {
+   {XvSettable | XvGettable, -128, 127, "XV_BRIGHTNESS"},
+   {XvSettable | XvGettable, 0, 255, "XV_CONTRAST"}
+};
+
+#define NUM_FORMATS 3
+static XF86VideoFormatRec Formats[NUM_FORMATS] = {
+   {15, TrueColor}, {16, TrueColor}, {24, TrueColor}
+};
+
+static XF86VideoEncodingRec DummyEncoding[1] = {
+   {
+      0,
+      "XV_IMAGE",
+      IMAGE_MAX_WIDTH, IMAGE_MAX_HEIGHT,
+      {1, 1}
+   }
+};
+
+#define NUM_IMAGES 2
+static XF86ImageRec Images[NUM_IMAGES] = {
+   XVIMAGE_UYVY,
+   XVIMAGE_YUY2,
+};
+
+struct xorg_xv_port_priv {
+   RegionRec clip;
+};
+
+
+static void
+stop_video(ScrnInfoPtr pScrn, pointer data, Bool shutdown)
+{
+}
+
+static int
+set_port_attribute(ScrnInfoPtr pScrn,
+                   Atom attribute, INT32 value, pointer data)
+{
+   return 0;
+}
+
+static int
+get_port_attribute(ScrnInfoPtr pScrn,
+                   Atom attribute, INT32 * value, pointer data)
+{
+   return 0;
+}
+
+static void
+query_best_size(ScrnInfoPtr pScrn,
+                Bool motion,
+                short vid_w, short vid_h,
+                short drw_w, short drw_h,
+                unsigned int *p_w, unsigned int *p_h, pointer data)
+{
+}
+
+static int
+put_image(ScrnInfoPtr pScrn,
+          short src_x, short src_y,
+          short drw_x, short drw_y,
+          short src_w, short src_h,
+          short drw_w, short drw_h,
+          int id, unsigned char *buf,
+          short width, short height,
+          Bool sync, RegionPtr clipBoxes, pointer data,
+          DrawablePtr pDraw)
+{
+   return 0;
+}
+
+static int
+query_image_attributes(ScrnInfoPtr pScrn,
+                       int id,
+                       unsigned short *w, unsigned short *h,
+                       int *pitches, int *offsets)
+{
+   return 0;
+}
+
+static struct xorg_xv_port_priv *
+port_priv_create(ScreenPtr pScreen)
+{
+   /*ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];*/
+   /*modesettingPtr ms = modesettingPTR(pScrn);*/
+   struct xorg_xv_port_priv *priv = NULL;
+
+   priv = calloc(1, sizeof(struct xorg_xv_port_priv));
+
+   if (!priv)
+      return NULL;
+
+    REGION_NULL(pScreen, &priv->clip);
+
+   return priv;
+}
+
+static XF86VideoAdaptorPtr
+xorg_setup_textured_adapter(ScreenPtr pScreen)
+{
+   /*ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];*/
+   /*modesettingPtr ms = modesettingPTR(pScrn);*/
+   XF86VideoAdaptorPtr adapt;
+   XF86AttributePtr attrs;
+   DevUnion *dev_unions;
+   int nports = 16, i;
+   int nattributes;
+
+   nattributes = NUM_TEXTURED_ATTRIBUTES;
+
+   adapt = calloc(1, sizeof(XF86VideoAdaptorRec));
+   dev_unions = calloc(nports, sizeof(DevUnion));
+   attrs = calloc(nattributes, sizeof(XF86AttributeRec));
+   if (adapt == NULL || dev_unions == NULL || attrs == NULL) {
+      free(adapt);
+      free(dev_unions);
+      free(attrs);
+      return NULL;
+   }
+
+   adapt->type = XvWindowMask | XvInputMask | XvImageMask;
+   adapt->flags = 0;
+   adapt->name = "Gallium3D Textured Video";
+   adapt->nEncodings = 1;
+   adapt->pEncodings = DummyEncoding;
+   adapt->nFormats = NUM_FORMATS;
+   adapt->pFormats = Formats;
+   adapt->nPorts = 0;
+   adapt->pPortPrivates = dev_unions;
+   adapt->nAttributes = nattributes;
+   adapt->pAttributes = attrs;
+   memcpy(attrs, TexturedAttributes, nattributes * sizeof(XF86AttributeRec));
+   adapt->nImages = NUM_IMAGES;
+   adapt->pImages = Images;
+   adapt->PutVideo = NULL;
+   adapt->PutStill = NULL;
+   adapt->GetVideo = NULL;
+   adapt->GetStill = NULL;
+   adapt->StopVideo = stop_video;
+   adapt->SetPortAttribute = set_port_attribute;
+   adapt->GetPortAttribute = get_port_attribute;
+   adapt->QueryBestSize = query_best_size;
+   adapt->PutImage = put_image;
+   adapt->QueryImageAttributes = query_image_attributes;
+
+   for (i = 0; i < nports; i++) {
+      struct xorg_xv_port_priv *priv =
+         port_priv_create(pScreen);
+
+      adapt->pPortPrivates[i].ptr = (pointer) (priv);
+      adapt->nPorts++;
+   }
+
+   return adapt;
+}
+
+void
+xorg_init_video(ScreenPtr pScreen)
+{
+   ScrnInfoPtr pScrn = xf86Screens[pScreen->myNum];
+   /*modesettingPtr ms = modesettingPTR(pScrn);*/
+   XF86VideoAdaptorPtr *adaptors, *new_adaptors = NULL;
+   XF86VideoAdaptorPtr textured_adapter;
+   int num_adaptors;
+
+   num_adaptors = xf86XVListGenericAdaptors(pScrn, &adaptors);
+   new_adaptors = malloc((num_adaptors + 1) * sizeof(XF86VideoAdaptorPtr *));
+   if (new_adaptors == NULL)
+      return;
+
+   memcpy(new_adaptors, adaptors, num_adaptors * sizeof(XF86VideoAdaptorPtr));
+   adaptors = new_adaptors;
+
+   /* Add the adaptors supported by our hardware.  First, set up the atoms
+    * that will be used by both output adaptors.
+    */
+   xvBrightness = MAKE_ATOM("XV_BRIGHTNESS");
+   xvContrast = MAKE_ATOM("XV_CONTRAST");
+
+   textured_adapter = xorg_setup_textured_adapter(pScreen);
+
+   debug_assert(textured_adapter);
+
+   if (textured_adapter) {
+      adaptors[num_adaptors++] = textured_adapter;
+   }
+
+   if (num_adaptors) {
+      xf86XVScreenInit(pScreen, adaptors, num_adaptors);
+   } else {
+      xf86DrvMsg(pScrn->scrnIndex, X_WARNING,
+                 "Disabling Xv because no adaptors could be initialized.\n");
+   }
+
+   free(adaptors);
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/Makefile b/src/gallium/state_trackers/xorg/xvmc/Makefile
new file mode 100644
index 0000000000..126dc6d58f
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/Makefile
@@ -0,0 +1,16 @@
+TOP = ../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = xvmctracker
+
+LIBRARY_INCLUDES = \
+	$(shell pkg-config --cflags-only-I xvmc) \
+	-I$(TOP)/src/gallium/winsys/g3dvl
+
+C_SOURCES = block.c \
+            surface.c \
+            context.c \
+            subpicture.c \
+            attributes.c
+
+include ../../../Makefile.template
diff --git a/src/gallium/state_trackers/xorg/xvmc/SConscript b/src/gallium/state_trackers/xorg/xvmc/SConscript
new file mode 100644
index 0000000000..cb25d68bd8
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/SConscript
@@ -0,0 +1,27 @@
+#######################################################################
+# SConscript for xvmc state_tracker
+
+Import('*')
+
+if 'xorg/xvmc' in env['statetrackers']:
+
+    env = env.Clone()
+    
+    env.Append(CPPPATH = [
+	'#/src/gallium/include',
+	'#/src/gallium/auxiliary',
+	'#/src/gallium/winsys/g3dvl',
+    ])
+
+    env.ParseConfig('pkg-config --cflags --libs xvmc')
+
+    st_xvmc = env.ConvenienceLibrary(
+	target = 'st_xvmc',
+	source = [ 'block.c',
+		'surface.c',
+		'context.c',
+		'subpicture.c',
+		'attributes.c',
+		]
+    )
+    Export('st_xvmc')
diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/state_trackers/xorg/xvmc/attributes.c
index dfa7ff3b1d..79a67838e6 100644
--- a/src/gallium/drivers/softpipe/sp_quad_occlusion.c
+++ b/src/gallium/state_trackers/xorg/xvmc/attributes.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2009 Younes Manton.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -25,61 +25,22 @@
  * 
  **************************************************************************/
 
+#include <assert.h>
+#include <X11/Xlib.h>
+#include <X11/extensions/Xvlib.h>
+#include <X11/extensions/XvMClib.h>
 
-/**
- * \brief  Quad occlusion counter stage
- * \author  Brian Paul
- */
-
-
-#include "pipe/p_defines.h"
-#include "util/u_memory.h"
-#include "sp_context.h"
-#include "sp_quad.h"
-#include "sp_surface.h"
-#include "sp_quad_pipe.h"
-
-static unsigned count_bits( unsigned val )
-{
-   unsigned i;
-
-   for (i = 0; val ; val >>= 1)
-      i += (val & 1);
-
-   return i;
-}
-
-static void
-occlusion_count_quad(struct quad_stage *qs, struct quad_header *quad)
-{
-   struct softpipe_context *softpipe = qs->softpipe;
-
-   softpipe->occlusion_count += count_bits(quad->inout.mask);
-
-   qs->next->run(qs->next, quad);
-}
-
-
-static void occlusion_begin(struct quad_stage *qs)
+XvAttribute* XvMCQueryAttributes(Display *dpy, XvMCContext *context, int *number)
 {
-   qs->next->begin(qs->next);
+   return NULL;
 }
 
-
-static void occlusion_destroy(struct quad_stage *qs)
+Status XvMCSetAttribute(Display *dpy, XvMCContext *context, Atom attribute, int value)
 {
-   FREE( qs );
+   return BadImplementation;
 }
 
-
-struct quad_stage *sp_quad_occlusion_stage( struct softpipe_context *softpipe )
+Status XvMCGetAttribute(Display *dpy, XvMCContext *context, Atom attribute, int *value)
 {
-   struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
-
-   stage->softpipe = softpipe;
-   stage->begin = occlusion_begin;
-   stage->run = occlusion_count_quad;
-   stage->destroy = occlusion_destroy;
-
-   return stage;
+   return BadImplementation;
 }
diff --git a/src/gallium/state_trackers/xorg/xvmc/block.c b/src/gallium/state_trackers/xorg/xvmc/block.c
new file mode 100644
index 0000000000..5102375fcf
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/block.c
@@ -0,0 +1,88 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <X11/Xlib.h>
+#include <X11/extensions/XvMClib.h>
+#include <util/u_memory.h>
+#include "xvmc_private.h"
+
+Status XvMCCreateBlocks(Display *dpy, XvMCContext *context, unsigned int num_blocks, XvMCBlockArray *blocks)
+{
+   assert(dpy);
+
+   if (!context)
+      return XvMCBadContext;
+   if (num_blocks == 0)
+      return BadValue;
+
+   assert(blocks);
+
+   blocks->context_id = context->context_id;
+   blocks->num_blocks = num_blocks;
+   blocks->blocks = MALLOC(BLOCK_SIZE_BYTES * num_blocks);
+   blocks->privData = NULL;
+
+   return Success;
+}
+
+Status XvMCDestroyBlocks(Display *dpy, XvMCBlockArray *blocks)
+{
+   assert(dpy);
+   assert(blocks);
+   FREE(blocks->blocks);
+
+   return Success;
+}
+
+Status XvMCCreateMacroBlocks(Display *dpy, XvMCContext *context, unsigned int num_blocks, XvMCMacroBlockArray *blocks)
+{
+   assert(dpy);
+
+   if (!context)
+      return XvMCBadContext;
+   if (num_blocks == 0)
+      return BadValue;
+
+   assert(blocks);
+
+   blocks->context_id = context->context_id;
+   blocks->num_blocks = num_blocks;
+   blocks->macro_blocks = MALLOC(sizeof(XvMCMacroBlock) * num_blocks);
+   blocks->privData = NULL;
+
+   return Success;
+}
+
+Status XvMCDestroyMacroBlocks(Display *dpy, XvMCMacroBlockArray *blocks)
+{
+   assert(dpy);
+   assert(blocks);
+   FREE(blocks->macro_blocks);
+
+   return Success;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/context.c b/src/gallium/state_trackers/xorg/xvmc/context.c
new file mode 100644
index 0000000000..942692d1bb
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/context.c
@@ -0,0 +1,252 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <X11/Xlibint.h>
+#include <X11/extensions/XvMClib.h>
+#include <pipe/p_screen.h>
+#include <pipe/p_video_context.h>
+#include <pipe/p_video_state.h>
+#include <pipe/p_state.h>
+#include <vl_winsys.h>
+#include <util/u_memory.h>
+#include <util/u_debug.h>
+#include <vl/vl_csc.h>
+#include "xvmc_private.h"
+
+static Status Validate(Display *dpy, XvPortID port, int surface_type_id,
+                       unsigned int width, unsigned int height, int flags,
+                       bool *found_port, int *screen, int *chroma_format,
+                       int *mc_type, int *surface_flags)
+{
+   bool found_surface = false;
+   XvAdaptorInfo *adaptor_info;
+   unsigned int num_adaptors;
+   int num_types;
+   unsigned int max_width, max_height;
+   Status ret;
+
+   assert(dpy);
+   assert(found_port);
+   assert(screen);
+   assert(chroma_format);
+   assert(mc_type);
+   assert(surface_flags);
+
+   *found_port = false;
+
+   for (unsigned int i = 0; i < XScreenCount(dpy); ++i) {
+      ret = XvQueryAdaptors(dpy, XRootWindow(dpy, i), &num_adaptors, &adaptor_info);
+      if (ret != Success)
+         return ret;
+
+      for (unsigned int j = 0; j < num_adaptors && !*found_port; ++j) {
+         for (unsigned int k = 0; k < adaptor_info[j].num_ports && !*found_port; ++k) {
+            XvMCSurfaceInfo *surface_info;
+
+            if (adaptor_info[j].base_id + k != port)
+               continue;
+
+            *found_port = true;
+
+            surface_info = XvMCListSurfaceTypes(dpy, adaptor_info[j].base_id, &num_types);
+            if (!surface_info) {
+               XvFreeAdaptorInfo(adaptor_info);
+               return BadAlloc;
+            }
+
+            for (unsigned int l = 0; l < num_types && !found_surface; ++l) {
+               if (surface_info[l].surface_type_id != surface_type_id)
+                  continue;
+
+               found_surface = true;
+               max_width = surface_info[l].max_width;
+               max_height = surface_info[l].max_height;
+               *chroma_format = surface_info[l].chroma_format;
+               *mc_type = surface_info[l].mc_type;
+               *surface_flags = surface_info[l].flags;
+               *screen = i;
+            }
+
+            XFree(surface_info);
+         }
+      }
+
+      XvFreeAdaptorInfo(adaptor_info);
+   }
+
+   if (!*found_port)
+      return XvBadPort;
+   if (!found_surface)
+      return BadMatch;
+   if (width > max_width || height > max_height)
+      return BadValue;
+   if (flags != XVMC_DIRECT && flags != 0)
+      return BadValue;
+
+   return Success;
+}
+
+static enum pipe_video_profile ProfileToPipe(int xvmc_profile)
+{
+   if (xvmc_profile & XVMC_MPEG_1)
+      assert(0);
+   if (xvmc_profile & XVMC_MPEG_2)
+      return PIPE_VIDEO_PROFILE_MPEG2_MAIN;
+   if (xvmc_profile & XVMC_H263)
+      assert(0);
+   if (xvmc_profile & XVMC_MPEG_4)
+      assert(0);
+	
+   assert(0);
+
+   return -1;
+}
+
+static enum pipe_video_chroma_format FormatToPipe(int xvmc_format)
+{
+   switch (xvmc_format) {
+      case XVMC_CHROMA_FORMAT_420:
+         return PIPE_VIDEO_CHROMA_FORMAT_420;
+      case XVMC_CHROMA_FORMAT_422:
+         return PIPE_VIDEO_CHROMA_FORMAT_422;
+      case XVMC_CHROMA_FORMAT_444:
+         return PIPE_VIDEO_CHROMA_FORMAT_444;
+      default:
+         assert(0);
+   }
+
+   return -1;
+}
+
+Status XvMCCreateContext(Display *dpy, XvPortID port, int surface_type_id,
+                         int width, int height, int flags, XvMCContext *context)
+{
+   bool found_port;
+   int scrn;
+   int chroma_format;
+   int mc_type;
+   int surface_flags;
+   Status ret;
+   struct pipe_screen *screen;
+   struct pipe_video_context *vpipe;
+   XvMCContextPrivate *context_priv;
+   float csc[16];
+
+   assert(dpy);
+
+   if (!context)
+      return XvMCBadContext;
+
+   ret = Validate(dpy, port, surface_type_id, width, height, flags,
+                  &found_port, &scrn, &chroma_format, &mc_type, &surface_flags);
+
+   /* Success and XvBadPort have the same value */
+   if (ret != Success || !found_port)
+      return ret;
+
+   /* XXX: Current limits */
+   if (chroma_format != XVMC_CHROMA_FORMAT_420) {
+      debug_printf("[XvMCg3dvl] Cannot decode requested surface type. Unsupported chroma format.\n");
+      return BadImplementation;
+   }
+   if (mc_type != (XVMC_MOCOMP | XVMC_MPEG_2)) {
+      debug_printf("[XvMCg3dvl] Cannot decode requested surface type. Non-MPEG2/Mocomp acceleration unsupported.\n");
+      return BadImplementation;
+   }
+   if (!(surface_flags & XVMC_INTRA_UNSIGNED)) {
+      debug_printf("[XvMCg3dvl] Cannot decode requested surface type. Signed intra unsupported.\n");
+      return BadImplementation;
+   }
+
+   context_priv = CALLOC(1, sizeof(XvMCContextPrivate));
+   if (!context_priv)
+      return BadAlloc;
+
+   /* TODO: Reuse screen if process creates another context */
+   screen = vl_screen_create(dpy, scrn);
+
+   if (!screen) {
+      FREE(context_priv);
+      return BadAlloc;
+   }
+
+   vpipe = vl_video_create(screen, ProfileToPipe(mc_type),
+                           FormatToPipe(chroma_format), width, height);
+
+   if (!vpipe) {
+      screen->destroy(screen);
+      FREE(context_priv);
+      return BadAlloc;
+   }
+
+   /* TODO: Define some Xv attribs to allow users to specify color standard, procamp */
+   vl_csc_get_matrix
+   (
+      debug_get_bool_option("G3DVL_NO_CSC", FALSE) ?
+      VL_CSC_COLOR_STANDARD_IDENTITY : VL_CSC_COLOR_STANDARD_BT_601,
+      NULL, true, csc
+   );
+   vpipe->set_csc_matrix(vpipe, csc);
+
+   context_priv->vpipe = vpipe;
+
+   context->context_id = XAllocID(dpy);
+   context->surface_type_id = surface_type_id;
+   context->width = width;
+   context->height = height;
+   context->flags = flags;
+   context->port = port;
+   context->privData = context_priv;
+	
+   SyncHandle();
+
+   return Success;
+}
+
+Status XvMCDestroyContext(Display *dpy, XvMCContext *context)
+{
+   struct pipe_screen *screen;
+   struct pipe_video_context *vpipe;
+   XvMCContextPrivate *context_priv;
+
+   assert(dpy);
+
+   if (!context || !context->privData)
+      return XvMCBadContext;
+
+   context_priv = context->privData;
+   vpipe = context_priv->vpipe;
+   pipe_surface_reference(&context_priv->backbuffer, NULL);
+   screen = vpipe->screen;
+   vpipe->destroy(vpipe);
+   screen->destroy(screen);
+   FREE(context_priv);
+   context->privData = NULL;
+
+   return Success;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/subpicture.c b/src/gallium/state_trackers/xorg/xvmc/subpicture.c
new file mode 100644
index 0000000000..69898d5fcd
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/subpicture.c
@@ -0,0 +1,195 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <X11/Xlibint.h>
+#include <X11/extensions/XvMClib.h>
+
+Status XvMCCreateSubpicture(Display *dpy, XvMCContext *context, XvMCSubpicture *subpicture,
+                            unsigned short width, unsigned short height, int xvimage_id)
+{
+   assert(dpy);
+
+   if (!context)
+      return XvMCBadContext;
+
+   assert(subpicture);
+
+   /*if (width > || height > )
+      return BadValue;*/
+
+   /*if (xvimage_id != )
+      return BadMatch;*/
+
+   subpicture->subpicture_id = XAllocID(dpy);
+   subpicture->context_id = context->context_id;
+   subpicture->xvimage_id = xvimage_id;
+   subpicture->width = width;
+   subpicture->height = height;
+   subpicture->num_palette_entries = 0;
+   subpicture->entry_bytes = 0;
+   subpicture->component_order[0] = 0;
+   subpicture->component_order[1] = 0;
+   subpicture->component_order[2] = 0;
+   subpicture->component_order[3] = 0;
+   /* TODO: subpicture->privData = ;*/
+
+   SyncHandle();
+
+   return Success;
+}
+
+Status XvMCClearSubpicture(Display *dpy, XvMCSubpicture *subpicture, short x, short y,
+                           unsigned short width, unsigned short height, unsigned int color)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   /* TODO: Assert clear rect is within bounds? Or clip? */
+
+   return Success;
+}
+
+Status XvMCCompositeSubpicture(Display *dpy, XvMCSubpicture *subpicture, XvImage *image,
+                               short srcx, short srcy, unsigned short width, unsigned short height,
+                               short dstx, short dsty)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   assert(image);
+
+   if (subpicture->xvimage_id != image->id)
+      return BadMatch;
+
+   /* TODO: Assert rects are within bounds? Or clip? */
+
+   return Success;
+}
+
+Status XvMCDestroySubpicture(Display *dpy, XvMCSubpicture *subpicture)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   return BadImplementation;
+}
+
+Status XvMCSetSubpicturePalette(Display *dpy, XvMCSubpicture *subpicture, unsigned char *palette)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   assert(palette);
+
+   /* We don't support paletted subpictures */
+   return BadMatch;
+}
+
+Status XvMCBlendSubpicture(Display *dpy, XvMCSurface *target_surface, XvMCSubpicture *subpicture,
+                           short subx, short suby, unsigned short subw, unsigned short subh,
+                           short surfx, short surfy, unsigned short surfw, unsigned short surfh)
+{
+   assert(dpy);
+
+   if (!target_surface)
+      return XvMCBadSurface;
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   if (target_surface->context_id != subpicture->context_id)
+      return BadMatch;
+
+   /* TODO: Assert rects are within bounds? Or clip? */
+   return Success;
+}
+
+Status XvMCBlendSubpicture2(Display *dpy, XvMCSurface *source_surface, XvMCSurface *target_surface,
+                            XvMCSubpicture *subpicture, short subx, short suby, unsigned short subw, unsigned short subh,
+                            short surfx, short surfy, unsigned short surfw, unsigned short surfh)
+{
+   assert(dpy);
+
+   if (!source_surface || !target_surface)
+      return XvMCBadSurface;
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   if (source_surface->context_id != subpicture->context_id)
+      return BadMatch;
+
+   if (source_surface->context_id != subpicture->context_id)
+      return BadMatch;
+
+   /* TODO: Assert rects are within bounds? Or clip? */
+   return Success;
+}
+
+Status XvMCSyncSubpicture(Display *dpy, XvMCSubpicture *subpicture)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   return Success;
+}
+
+Status XvMCFlushSubpicture(Display *dpy, XvMCSubpicture *subpicture)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   return Success;
+}
+
+Status XvMCGetSubpictureStatus(Display *dpy, XvMCSubpicture *subpicture, int *status)
+{
+   assert(dpy);
+
+   if (!subpicture)
+      return XvMCBadSubpicture;
+
+   assert(status);
+
+   /* TODO */
+   *status = 0;
+
+   return Success;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/surface.c b/src/gallium/state_trackers/xorg/xvmc/surface.c
new file mode 100644
index 0000000000..bf9038f356
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/surface.c
@@ -0,0 +1,409 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <X11/Xlibint.h>
+#include <pipe/p_video_context.h>
+#include <pipe/p_video_state.h>
+#include <pipe/p_state.h>
+#include <util/u_memory.h>
+#include "xvmc_private.h"
+
+static enum pipe_mpeg12_macroblock_type TypeToPipe(int xvmc_mb_type)
+{
+   if (xvmc_mb_type & XVMC_MB_TYPE_INTRA)
+      return PIPE_MPEG12_MACROBLOCK_TYPE_INTRA;
+   if ((xvmc_mb_type & (XVMC_MB_TYPE_MOTION_FORWARD | XVMC_MB_TYPE_MOTION_BACKWARD)) == XVMC_MB_TYPE_MOTION_FORWARD)
+      return PIPE_MPEG12_MACROBLOCK_TYPE_FWD;
+   if ((xvmc_mb_type & (XVMC_MB_TYPE_MOTION_FORWARD | XVMC_MB_TYPE_MOTION_BACKWARD)) == XVMC_MB_TYPE_MOTION_BACKWARD)
+      return PIPE_MPEG12_MACROBLOCK_TYPE_BKWD;
+   if ((xvmc_mb_type & (XVMC_MB_TYPE_MOTION_FORWARD | XVMC_MB_TYPE_MOTION_BACKWARD)) == (XVMC_MB_TYPE_MOTION_FORWARD | XVMC_MB_TYPE_MOTION_BACKWARD))
+      return PIPE_MPEG12_MACROBLOCK_TYPE_BI;
+
+   assert(0);
+
+   return -1;
+}
+
+static enum pipe_mpeg12_picture_type PictureToPipe(int xvmc_pic)
+{
+   switch (xvmc_pic) {
+      case XVMC_TOP_FIELD:
+         return PIPE_MPEG12_PICTURE_TYPE_FIELD_TOP;
+      case XVMC_BOTTOM_FIELD:
+         return PIPE_MPEG12_PICTURE_TYPE_FIELD_BOTTOM;
+      case XVMC_FRAME_PICTURE:
+         return PIPE_MPEG12_PICTURE_TYPE_FRAME;
+      default:
+         assert(0);
+   }
+
+   return -1;
+}
+
+static enum pipe_mpeg12_motion_type MotionToPipe(int xvmc_motion_type, int xvmc_dct_type)
+{
+   switch (xvmc_motion_type) {
+      case XVMC_PREDICTION_FRAME:
+         return xvmc_dct_type == XVMC_DCT_TYPE_FIELD ?
+            PIPE_MPEG12_MOTION_TYPE_16x8 : PIPE_MPEG12_MOTION_TYPE_FRAME;
+      case XVMC_PREDICTION_FIELD:
+         return PIPE_MPEG12_MOTION_TYPE_FIELD;
+      case XVMC_PREDICTION_DUAL_PRIME:
+         return PIPE_MPEG12_MOTION_TYPE_DUALPRIME;
+      default:
+         assert(0);
+   }
+
+   return -1;
+}
+
+static bool
+CreateOrResizeBackBuffer(struct pipe_video_context *vpipe, unsigned int width, unsigned int height,
+                         struct pipe_surface **backbuffer)
+{
+   struct pipe_texture template;
+   struct pipe_texture *tex;
+
+   assert(vpipe);
+
+   if (*backbuffer) {
+      if ((*backbuffer)->width != width || (*backbuffer)->height != height)
+         pipe_surface_reference(backbuffer, NULL);
+      else
+         return true;
+   }
+
+   memset(&template, 0, sizeof(struct pipe_texture));
+   template.target = PIPE_TEXTURE_2D;
+   /* XXX: Needs to match the drawable's format? */
+   template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+   template.last_level = 0;
+   template.width[0] = width;
+   template.height[0] = height;
+   template.depth[0] = 1;
+   pf_get_block(template.format, &template.block);
+   template.tex_usage = PIPE_TEXTURE_USAGE_DISPLAY_TARGET;
+
+   tex = vpipe->screen->texture_create(vpipe->screen, &template);
+   if (!tex)
+      return false;
+
+   *backbuffer = vpipe->screen->get_tex_surface(vpipe->screen, tex, 0, 0, 0,
+                                                PIPE_BUFFER_USAGE_GPU_READ |
+                                                PIPE_BUFFER_USAGE_GPU_WRITE);
+   pipe_texture_reference(&tex, NULL);
+
+   if (!*backbuffer)
+      return false;
+
+   /* Clear the backbuffer in case the video doesn't cover the whole window */
+   /* FIXME: Need to clear every time a frame moves and leaves dirty rects */
+   vpipe->clear_surface(vpipe, 0, 0, width, height, 0, *backbuffer);
+
+   return true;
+}
+
+static void
+MacroBlocksToPipe(const XvMCMacroBlockArray *xvmc_macroblocks,
+                  const XvMCBlockArray *xvmc_blocks,
+                  unsigned int first_macroblock,
+                  unsigned int num_macroblocks,
+                  struct pipe_mpeg12_macroblock *pipe_macroblocks)
+{
+   unsigned int i, j, k, l;
+   XvMCMacroBlock *xvmc_mb;
+
+   assert(xvmc_macroblocks);
+   assert(xvmc_blocks);
+   assert(pipe_macroblocks);
+   assert(num_macroblocks);
+
+   xvmc_mb = xvmc_macroblocks->macro_blocks + first_macroblock;
+
+   for (i = 0; i < num_macroblocks; ++i) {
+      pipe_macroblocks->base.codec = PIPE_VIDEO_CODEC_MPEG12;
+      pipe_macroblocks->mbx = xvmc_mb->x;
+      pipe_macroblocks->mby = xvmc_mb->y;
+      pipe_macroblocks->mb_type = TypeToPipe(xvmc_mb->macroblock_type);
+      if (pipe_macroblocks->mb_type != PIPE_MPEG12_MACROBLOCK_TYPE_INTRA)
+         pipe_macroblocks->mo_type = MotionToPipe(xvmc_mb->motion_type, xvmc_mb->dct_type);
+      /* Get rid of Valgrind 'undefined' warnings */
+      else
+         pipe_macroblocks->mo_type = -1;
+      pipe_macroblocks->dct_type = xvmc_mb->dct_type == XVMC_DCT_TYPE_FIELD ?
+         PIPE_MPEG12_DCT_TYPE_FIELD : PIPE_MPEG12_DCT_TYPE_FRAME;
+
+      for (j = 0; j < 2; ++j)
+         for (k = 0; k < 2; ++k)
+            for (l = 0; l < 2; ++l)
+               pipe_macroblocks->pmv[j][k][l] = xvmc_mb->PMV[j][k][l];
+
+      pipe_macroblocks->cbp = xvmc_mb->coded_block_pattern;
+      pipe_macroblocks->blocks = xvmc_blocks->blocks + xvmc_mb->index * BLOCK_SIZE_SAMPLES;
+
+      ++pipe_macroblocks;
+      ++xvmc_mb;
+   }
+}
+
+Status XvMCCreateSurface(Display *dpy, XvMCContext *context, XvMCSurface *surface)
+{
+   XvMCContextPrivate *context_priv;
+   struct pipe_video_context *vpipe;
+   XvMCSurfacePrivate *surface_priv;
+   struct pipe_video_surface *vsfc;
+
+   assert(dpy);
+
+   if (!context)
+      return XvMCBadContext;
+   if (!surface)
+      return XvMCBadSurface;
+
+   context_priv = context->privData;
+   vpipe = context_priv->vpipe;
+
+   surface_priv = CALLOC(1, sizeof(XvMCSurfacePrivate));
+   if (!surface_priv)
+      return BadAlloc;
+
+   vsfc = vpipe->screen->video_surface_create(vpipe->screen, vpipe->chroma_format,
+                                              vpipe->width, vpipe->height);
+   if (!vsfc) {
+      FREE(surface_priv);
+      return BadAlloc;
+   }
+
+   surface_priv->pipe_vsfc = vsfc;
+   surface_priv->context = context;
+
+   surface->surface_id = XAllocID(dpy);
+   surface->context_id = context->context_id;
+   surface->surface_type_id = context->surface_type_id;
+   surface->width = context->width;
+   surface->height = context->height;
+   surface->privData = surface_priv;
+
+   SyncHandle();
+
+   return Success;
+}
+
+Status XvMCRenderSurface(Display *dpy, XvMCContext *context, unsigned int picture_structure,
+                         XvMCSurface *target_surface, XvMCSurface *past_surface, XvMCSurface *future_surface,
+                         unsigned int flags, unsigned int num_macroblocks, unsigned int first_macroblock,
+                         XvMCMacroBlockArray *macroblocks, XvMCBlockArray *blocks
+)
+{
+   struct pipe_video_context *vpipe;
+   struct pipe_surface *t_vsfc;
+   struct pipe_surface *p_vsfc;
+   struct pipe_surface *f_vsfc;
+   XvMCContextPrivate *context_priv;
+   XvMCSurfacePrivate *target_surface_priv;
+   XvMCSurfacePrivate *past_surface_priv;
+   XvMCSurfacePrivate *future_surface_priv;
+   struct pipe_mpeg12_macroblock pipe_macroblocks[num_macroblocks];
+
+   assert(dpy);
+
+   if (!context || !context->privData)
+      return XvMCBadContext;
+   if (!target_surface || !target_surface->privData)
+      return XvMCBadSurface;
+
+   if (picture_structure != XVMC_TOP_FIELD &&
+       picture_structure != XVMC_BOTTOM_FIELD &&
+       picture_structure != XVMC_FRAME_PICTURE)
+      return BadValue;
+   /* Bkwd pred equivalent to fwd (past && !future) */
+   if (future_surface && !past_surface)
+      return BadMatch;
+
+   assert(context->context_id == target_surface->context_id);
+   assert(!past_surface || context->context_id == past_surface->context_id);
+   assert(!future_surface || context->context_id == future_surface->context_id);
+
+   assert(macroblocks);
+   assert(blocks);
+
+   assert(macroblocks->context_id == context->context_id);
+   assert(blocks->context_id == context->context_id);
+
+   assert(flags == 0 || flags == XVMC_SECOND_FIELD);
+
+   target_surface_priv = target_surface->privData;
+   past_surface_priv = past_surface ? past_surface->privData : NULL;
+   future_surface_priv = future_surface ? future_surface->privData : NULL;
+
+   assert(target_surface_priv->context == context);
+   assert(!past_surface || past_surface_priv->context == context);
+   assert(!future_surface || future_surface_priv->context == context);
+
+   context_priv = context->privData;
+   vpipe = context_priv->vpipe;
+
+   t_vsfc = target_surface_priv->pipe_vsfc;
+   p_vsfc = past_surface ? past_surface_priv->pipe_vsfc : NULL;
+   f_vsfc = future_surface ? future_surface_priv->pipe_vsfc : NULL;
+
+   MacroBlocksToPipe(macroblocks, blocks, first_macroblock,
+                     num_macroblocks, pipe_macroblocks);
+
+   vpipe->set_decode_target(vpipe, t_vsfc);
+   vpipe->decode_macroblocks(vpipe, p_vsfc, f_vsfc, num_macroblocks,
+                             &pipe_macroblocks->base, target_surface_priv->render_fence);
+
+   return Success;
+}
+
+Status XvMCFlushSurface(Display *dpy, XvMCSurface *surface)
+{
+   assert(dpy);
+
+   if (!surface)
+      return XvMCBadSurface;
+
+   return Success;
+}
+
+Status XvMCSyncSurface(Display *dpy, XvMCSurface *surface)
+{
+   assert(dpy);
+
+   if (!surface)
+      return XvMCBadSurface;
+
+   return Success;
+}
+
+Status XvMCPutSurface(Display *dpy, XvMCSurface *surface, Drawable drawable,
+                      short srcx, short srcy, unsigned short srcw, unsigned short srch,
+                      short destx, short desty, unsigned short destw, unsigned short desth,
+                      int flags)
+{
+   Window root;
+   int x, y;
+   unsigned int width, height;
+   unsigned int border_width;
+   unsigned int depth;
+   struct pipe_video_context *vpipe;
+   XvMCSurfacePrivate *surface_priv;
+   XvMCContextPrivate *context_priv;
+   XvMCContext *context;
+   struct pipe_video_rect src_rect = {srcx, srcy, srcw, srch};
+   struct pipe_video_rect dst_rect = {destx, desty, destw, desth};
+
+   assert(dpy);
+
+   if (!surface || !surface->privData)
+      return XvMCBadSurface;
+
+   if (XGetGeometry(dpy, drawable, &root, &x, &y, &width, &height, &border_width, &depth) == BadDrawable)
+      return BadDrawable;
+
+   assert(flags == XVMC_TOP_FIELD || flags == XVMC_BOTTOM_FIELD || flags == XVMC_FRAME_PICTURE);
+   assert(srcx + srcw - 1 < surface->width);
+   assert(srcy + srch - 1 < surface->height);
+   /*
+    * Some apps (mplayer) hit these asserts because they call
+    * this function after the window has been resized by the WM
+    * but before they've handled the corresponding XEvent and
+    * know about the new dimensions. The output should be clipped
+    * until the app updates destw and desth.
+    */
+   /*
+   assert(destx + destw - 1 < width);
+   assert(desty + desth - 1 < height);
+    */
+
+   surface_priv = surface->privData;
+   context = surface_priv->context;
+   context_priv = context->privData;
+   vpipe = context_priv->vpipe;
+
+   if (!CreateOrResizeBackBuffer(vpipe, width, height, &context_priv->backbuffer))
+      return BadAlloc;
+
+   vpipe->render_picture(vpipe, surface_priv->pipe_vsfc, PictureToPipe(flags), &src_rect,
+                         context_priv->backbuffer, &dst_rect, surface_priv->disp_fence);
+
+   vl_video_bind_drawable(vpipe, drawable);
+	
+   vpipe->screen->flush_frontbuffer
+   (
+      vpipe->screen,
+      context_priv->backbuffer,
+      vpipe->priv
+   );
+
+   return Success;
+}
+
+Status XvMCGetSurfaceStatus(Display *dpy, XvMCSurface *surface, int *status)
+{
+   assert(dpy);
+
+   if (!surface)
+      return XvMCBadSurface;
+
+   assert(status);
+
+   *status = 0;
+
+   return Success;
+}
+
+Status XvMCDestroySurface(Display *dpy, XvMCSurface *surface)
+{
+   XvMCSurfacePrivate *surface_priv;
+
+   assert(dpy);
+
+   if (!surface || !surface->privData)
+      return XvMCBadSurface;
+
+   surface_priv = surface->privData;
+   pipe_video_surface_reference(&surface_priv->pipe_vsfc, NULL);
+   FREE(surface_priv);
+   surface->privData = NULL;
+
+   return Success;
+}
+
+Status XvMCHideSurface(Display *dpy, XvMCSurface *surface)
+{
+   assert(dpy);
+
+   if (!surface || !surface->privData)
+      return XvMCBadSurface;
+
+   /* No op, only for overlaid rendering */
+
+   return Success;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/.gitignore b/src/gallium/state_trackers/xorg/xvmc/tests/.gitignore
new file mode 100644
index 0000000000..e1d2f9023d
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/.gitignore
@@ -0,0 +1,5 @@
+test_context
+test_surface
+test_blocks
+test_rendering
+xvmc_bench
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/Makefile b/src/gallium/state_trackers/xorg/xvmc/tests/Makefile
new file mode 100644
index 0000000000..c875dd7605
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/Makefile
@@ -0,0 +1,28 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBS = -lXvMCW -lXvMC -lXv -lX11
+
+#############################################
+
+.PHONY: default clean
+
+default: test_context test_surface test_blocks test_rendering xvmc_bench
+
+test_context: test_context.o testlib.o
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+test_surface: test_surface.o testlib.o
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+test_blocks: test_blocks.o testlib.o
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+test_rendering: test_rendering.o testlib.o
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+xvmc_bench: xvmc_bench.o testlib.o
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS)
+
+clean:
+	$(RM) -rf *.o test_context test_surface test_blocks test_rendering xvmc_bench
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/test_blocks.c b/src/gallium/state_trackers/xorg/xvmc/tests/test_blocks.c
new file mode 100644
index 0000000000..994e3ca4d1
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/test_blocks.c
@@ -0,0 +1,111 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <error.h>
+#include "testlib.h"
+
+int main(int argc, char **argv)
+{
+	const unsigned int	width = 16, height = 16;
+	const unsigned int	min_required_blocks = 1, min_required_macroblocks = 1;
+	const unsigned int	mc_types[2] = {XVMC_MOCOMP | XVMC_MPEG_2, XVMC_IDCT | XVMC_MPEG_2};
+
+	Display			*display;
+	XvPortID		port_num;
+	int			surface_type_id;
+	unsigned int		is_overlay, intra_unsigned;
+	int			colorkey;
+	XvMCContext		context;
+	XvMCSurface		surface;
+	XvMCBlockArray		blocks = {0};
+	XvMCMacroBlockArray	macroblocks = {0};
+
+	display = XOpenDisplay(NULL);
+
+	if (!GetPort
+	(
+		display,
+		width,
+		height,
+		XVMC_CHROMA_FORMAT_420,
+    		mc_types,
+    		2,
+    		&port_num,
+    		&surface_type_id,
+    		&is_overlay,
+    		&intra_unsigned
+	))
+	{
+		XCloseDisplay(display);
+		error(1, 0, "Error, unable to find a good port.\n");
+	}
+
+	if (is_overlay)
+	{
+		Atom xv_colorkey = XInternAtom(display, "XV_COLORKEY", 0);
+		XvGetPortAttribute(display, port_num, xv_colorkey, &colorkey);
+	}
+
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height, XVMC_DIRECT, &context) == Success);
+	assert(XvMCCreateSurface(display, &context, &surface) == Success);
+
+	/* Test NULL context */
+	assert(XvMCCreateBlocks(display, NULL, 1, &blocks) == XvMCBadContext);
+	/* Test 0 blocks */
+	assert(XvMCCreateBlocks(display, &context, 0, &blocks) == BadValue);
+	/* Test valid params */
+	assert(XvMCCreateBlocks(display, &context, min_required_blocks, &blocks) == Success);
+	/* Test context id assigned and correct */
+	assert(blocks.context_id == context.context_id);
+	/* Test number of blocks assigned and correct */
+	assert(blocks.num_blocks == min_required_blocks);
+	/* Test block pointer valid */
+	assert(blocks.blocks != NULL);
+	/* Test NULL context */
+	assert(XvMCCreateMacroBlocks(display, NULL, 1, &macroblocks) == XvMCBadContext);
+	/* Test 0 macroblocks */
+	assert(XvMCCreateMacroBlocks(display, &context, 0, &macroblocks) == BadValue);
+	/* Test valid params */
+	assert(XvMCCreateMacroBlocks(display, &context, min_required_macroblocks, &macroblocks) == Success);
+	/* Test context id assigned and correct */
+	assert(macroblocks.context_id == context.context_id);
+	/* Test macroblock pointer valid */
+	assert(macroblocks.macro_blocks != NULL);
+	/* Test valid params */
+	assert(XvMCDestroyMacroBlocks(display, &macroblocks) == Success);
+	/* Test valid params */
+	assert(XvMCDestroyBlocks(display, &blocks) == Success);
+
+	assert(XvMCDestroySurface(display, &surface) == Success);
+	assert(XvMCDestroyContext(display, &context) == Success);
+
+	XvUngrabPort(display, port_num, CurrentTime);
+	XCloseDisplay(display);
+
+	return 0;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/test_context.c b/src/gallium/state_trackers/xorg/xvmc/tests/test_context.c
new file mode 100644
index 0000000000..3da957c933
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/test_context.c
@@ -0,0 +1,119 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <error.h>
+#include "testlib.h"
+
+int main(int argc, char **argv)
+{
+	const unsigned int	width = 16, height = 16;
+	const unsigned int	mc_types[2] = {XVMC_MOCOMP | XVMC_MPEG_2, XVMC_IDCT | XVMC_MPEG_2};
+
+	Display			*display;
+	XvPortID		port_num;
+	int			surface_type_id;
+	unsigned int		is_overlay, intra_unsigned;
+	int			colorkey;
+	XvMCContext		context = {0};
+
+	display = XOpenDisplay(NULL);
+
+	if (!GetPort
+	(
+		display,
+		width,
+		height,
+		XVMC_CHROMA_FORMAT_420,
+    		mc_types,
+    		2,
+    		&port_num,
+    		&surface_type_id,
+    		&is_overlay,
+    		&intra_unsigned
+	))
+	{
+		XCloseDisplay(display);
+		error(1, 0, "Error, unable to find a good port.\n");
+	}
+
+	if (is_overlay)
+	{
+		Atom xv_colorkey = XInternAtom(display, "XV_COLORKEY", 0);
+		XvGetPortAttribute(display, port_num, xv_colorkey, &colorkey);
+	}
+
+	/* Test NULL context */
+	/* XXX: XvMCBadContext not a valid return for XvMCCreateContext in the XvMC API, but openChrome driver returns it */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height, XVMC_DIRECT, NULL) == XvMCBadContext);
+	/* Test invalid port */
+	/* XXX: Success and XvBadPort have the same value, if this call actually gets passed the validation step as of now we'll crash later */
+	assert(XvMCCreateContext(display, -1, surface_type_id, width, height, XVMC_DIRECT, &context) == XvBadPort);
+	/* Test invalid surface */
+	assert(XvMCCreateContext(display, port_num, -1, width, height, XVMC_DIRECT, &context) == BadMatch);
+	/* Test invalid flags */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height, -1, &context) == BadValue);
+	/* Test huge width */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, 16384, height, XVMC_DIRECT, &context) == BadValue);
+	/* Test huge height */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, 16384, XVMC_DIRECT, &context) == BadValue);
+	/* Test huge width & height */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, 16384, 16384, XVMC_DIRECT, &context) == BadValue);
+	/* Test valid params */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height, XVMC_DIRECT, &context) == Success);
+	/* Test context id assigned */
+	assert(context.context_id != 0);
+	/* Test surface type id assigned and correct */
+	assert(context.surface_type_id == surface_type_id);
+	/* Test width & height assigned and correct */
+	assert(context.width == width && context.height == height);
+	/* Test port assigned and correct */
+	assert(context.port == port_num);
+	/* Test flags assigned and correct */
+	assert(context.flags == XVMC_DIRECT);
+	/* Test NULL context */
+	assert(XvMCDestroyContext(display, NULL) == XvMCBadContext);
+	/* Test valid params */
+	assert(XvMCDestroyContext(display, &context) == Success);
+	/* Test awkward but valid width */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width + 1, height, XVMC_DIRECT, &context) == Success);
+	assert(context.width >= width + 1);
+	assert(XvMCDestroyContext(display, &context) == Success);
+	/* Test awkward but valid height */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height + 1, XVMC_DIRECT, &context) == Success);
+	assert(context.height >= height + 1);
+	assert(XvMCDestroyContext(display, &context) == Success);
+	/* Test awkward but valid width & height */
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width + 1, height + 1, XVMC_DIRECT, &context) == Success);
+	assert(context.width >= width + 1 && context.height >= height + 1);
+	assert(XvMCDestroyContext(display, &context) == Success);
+
+	XvUngrabPort(display, port_num, CurrentTime);
+	XCloseDisplay(display);
+
+	return 0;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/test_rendering.c b/src/gallium/state_trackers/xorg/xvmc/tests/test_rendering.c
new file mode 100644
index 0000000000..6058783a79
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/test_rendering.c
@@ -0,0 +1,317 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <error.h>
+#include "testlib.h"
+
+#define BLOCK_WIDTH			8
+#define BLOCK_HEIGHT			8
+#define BLOCK_SIZE			(BLOCK_WIDTH * BLOCK_HEIGHT)
+#define MACROBLOCK_WIDTH		16
+#define MACROBLOCK_HEIGHT		16
+#define MACROBLOCK_WIDTH_IN_BLOCKS	(MACROBLOCK_WIDTH / BLOCK_WIDTH)
+#define MACROBLOCK_HEIGHT_IN_BLOCKS	(MACROBLOCK_HEIGHT / BLOCK_HEIGHT)
+#define BLOCKS_PER_MACROBLOCK		6
+
+#define INPUT_WIDTH			16
+#define INPUT_HEIGHT			16
+#define INPUT_WIDTH_IN_MACROBLOCKS	(INPUT_WIDTH / MACROBLOCK_WIDTH)
+#define INPUT_HEIGHT_IN_MACROBLOCKS	(INPUT_HEIGHT / MACROBLOCK_HEIGHT)
+#define NUM_MACROBLOCKS			(INPUT_WIDTH_IN_MACROBLOCKS * INPUT_HEIGHT_IN_MACROBLOCKS)
+
+#define DEFAULT_OUTPUT_WIDTH		INPUT_WIDTH
+#define DEFAULT_OUTPUT_HEIGHT		INPUT_HEIGHT
+#define DEFAULT_ACCEPTABLE_ERR		0.01
+
+void ParseArgs(int argc, char **argv, unsigned int *output_width, unsigned int *output_height, double *acceptable_error, int *prompt);
+void Gradient(short *block, unsigned int start, unsigned int stop, int horizontal);
+
+void ParseArgs(int argc, char **argv, unsigned int *output_width, unsigned int *output_height, double *acceptable_error, int *prompt)
+{
+	int fail = 0;
+	int i;
+
+	*output_width = DEFAULT_OUTPUT_WIDTH;
+	*output_height = DEFAULT_OUTPUT_WIDTH;
+	*acceptable_error = DEFAULT_ACCEPTABLE_ERR;
+	*prompt = 1;
+
+	for (i = 1; i < argc && !fail; ++i)
+	{
+		if (!strcmp(argv[i], "-w"))
+		{
+			if (sscanf(argv[++i], "%u", output_width) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-h"))
+		{
+			if (sscanf(argv[++i], "%u", output_height) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-e"))
+		{
+			if (sscanf(argv[++i], "%lf", acceptable_error) != 1)
+				fail = 1;
+		}
+		else if (strcmp(argv[i], "-n"))
+			*prompt = 0;
+		else
+			fail = 1;
+	}
+
+	if (fail)
+		error
+		(
+			1, 0,
+			"Bad argument.\n"
+			"\n"
+			"Usage: %s [options]\n"
+			"\t-w <width>\tOutput width\n"
+			"\t-h <height>\tOutput height\n"
+			"\t-e <error>\tAcceptable margin of error per pixel, from 0 to 1\n"
+			"\t-n\tDon't prompt for quit\n",
+			argv[0]
+		);
+}
+
+void Gradient(short *block, unsigned int start, unsigned int stop, int horizontal)
+{
+	unsigned int x, y;
+	unsigned int range = stop - start;
+
+	if (horizontal)
+	{
+		for (y = 0; y < BLOCK_HEIGHT; ++y)
+			for (x = 0; x < BLOCK_WIDTH; ++x)
+				block[y * BLOCK_WIDTH + x] = (short)(start + range * (x / (float)(BLOCK_WIDTH - 1)));
+	}
+	else
+	{
+		for (y = 0; y < BLOCK_HEIGHT; ++y)
+			for (x = 0; x < BLOCK_WIDTH; ++x)
+				block[y * BLOCK_WIDTH + x] = (short)(start + range * (y / (float)(BLOCK_HEIGHT - 1)));
+	}
+}
+
+int main(int argc, char **argv)
+{
+	unsigned int		output_width;
+	unsigned int		output_height;
+	double			acceptable_error;
+	int			prompt;
+	Display			*display;
+	Window			root, window;
+	const unsigned int	mc_types[2] = {XVMC_MOCOMP | XVMC_MPEG_2, XVMC_IDCT | XVMC_MPEG_2};
+	XvPortID		port_num;
+	int			surface_type_id;
+	unsigned int		is_overlay, intra_unsigned;
+	int			colorkey;
+	XvMCContext		context;
+	XvMCSurface		surface;
+	XvMCBlockArray		block_array;
+	XvMCMacroBlockArray	mb_array;
+	int			mbx, mby, bx, by;
+	XvMCMacroBlock		*mb;
+	short			*blocks;
+	int			quit = 0;
+
+	ParseArgs(argc, argv, &output_width, &output_height, &acceptable_error, &prompt);
+
+	display = XOpenDisplay(NULL);
+
+	if (!GetPort
+	(
+		display,
+		INPUT_WIDTH,
+		INPUT_HEIGHT,
+		XVMC_CHROMA_FORMAT_420,
+    		mc_types,
+    		2,
+    		&port_num,
+    		&surface_type_id,
+    		&is_overlay,
+    		&intra_unsigned
+	))
+	{
+		XCloseDisplay(display);
+		error(1, 0, "Error, unable to find a good port.\n");
+	}
+
+	if (is_overlay)
+	{
+		Atom xv_colorkey = XInternAtom(display, "XV_COLORKEY", 0);
+		XvGetPortAttribute(display, port_num, xv_colorkey, &colorkey);
+	}
+
+	root = XDefaultRootWindow(display);
+	window = XCreateSimpleWindow(display, root, 0, 0, output_width, output_height, 0, 0, colorkey);
+
+	assert(XvMCCreateContext(display, port_num, surface_type_id, INPUT_WIDTH, INPUT_HEIGHT, XVMC_DIRECT, &context) == Success);
+	assert(XvMCCreateSurface(display, &context, &surface) == Success);
+	assert(XvMCCreateBlocks(display, &context, NUM_MACROBLOCKS * BLOCKS_PER_MACROBLOCK, &block_array) == Success);
+	assert(XvMCCreateMacroBlocks(display, &context, NUM_MACROBLOCKS, &mb_array) == Success);
+
+	mb = mb_array.macro_blocks;
+	blocks = block_array.blocks;
+
+	for (mby = 0; mby < INPUT_HEIGHT_IN_MACROBLOCKS; ++mby)
+		for (mbx = 0; mbx < INPUT_WIDTH_IN_MACROBLOCKS; ++mbx)
+		{
+			mb->x = mbx;
+			mb->y = mby;
+			mb->macroblock_type = XVMC_MB_TYPE_INTRA;
+			/*mb->motion_type = ;*/
+			/*mb->motion_vertical_field_select = ;*/
+			mb->dct_type = XVMC_DCT_TYPE_FRAME;
+			/*mb->PMV[0][0][0] = ;
+			mb->PMV[0][0][1] = ;
+			mb->PMV[0][1][0] = ;
+			mb->PMV[0][1][1] = ;
+			mb->PMV[1][0][0] = ;
+			mb->PMV[1][0][1] = ;
+			mb->PMV[1][1][0] = ;
+			mb->PMV[1][1][1] = ;*/
+			mb->index = (mby * INPUT_WIDTH_IN_MACROBLOCKS + mbx) * BLOCKS_PER_MACROBLOCK;
+			mb->coded_block_pattern = 0x3F;
+
+			mb++;
+
+			for (by = 0; by < MACROBLOCK_HEIGHT_IN_BLOCKS; ++by)
+				for (bx = 0; bx < MACROBLOCK_WIDTH_IN_BLOCKS; ++bx)
+				{
+					const int start = 16, stop = 235, range = stop - start;
+
+					Gradient
+					(
+						blocks,
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH) / (float)(INPUT_WIDTH - 1))),
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH + BLOCK_WIDTH - 1) / (float)(INPUT_WIDTH - 1))),
+						1
+					);
+
+					blocks += BLOCK_SIZE;
+				}
+
+			for (by = 0; by < MACROBLOCK_HEIGHT_IN_BLOCKS / 2; ++by)
+				for (bx = 0; bx < MACROBLOCK_WIDTH_IN_BLOCKS / 2; ++bx)
+				{
+					const int start = 16, stop = 240, range = stop - start;
+
+					Gradient
+					(
+						blocks,
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH) / (float)(INPUT_WIDTH - 1))),
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH + BLOCK_WIDTH - 1) / (float)(INPUT_WIDTH - 1))),
+						1
+					);
+
+					blocks += BLOCK_SIZE;
+
+					Gradient
+					(
+						blocks,
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH) / (float)(INPUT_WIDTH - 1))),
+						(short)(start + range * ((mbx * MACROBLOCK_WIDTH + bx * BLOCK_WIDTH + BLOCK_WIDTH - 1) / (float)(INPUT_WIDTH - 1))),
+						1
+					);
+
+					blocks += BLOCK_SIZE;
+				}
+		}
+
+	XSelectInput(display, window, ExposureMask | KeyPressMask);
+	XMapWindow(display, window);
+	XSync(display, 0);
+
+	/* Test NULL context */
+	assert(XvMCRenderSurface(display, NULL, XVMC_FRAME_PICTURE, &surface, NULL, NULL, 0, NUM_MACROBLOCKS, 0, &mb_array, &block_array) == XvMCBadContext);
+	/* Test NULL surface */
+	assert(XvMCRenderSurface(display, &context, XVMC_FRAME_PICTURE, NULL, NULL, NULL, 0, NUM_MACROBLOCKS, 0, &mb_array, &block_array) == XvMCBadSurface);
+	/* Test bad picture structure */
+	assert(XvMCRenderSurface(display, &context, 0, &surface, NULL, NULL, 0, NUM_MACROBLOCKS, 0, &mb_array, &block_array) == BadValue);
+	/* Test valid params */
+	assert(XvMCRenderSurface(display, &context, XVMC_FRAME_PICTURE, &surface, NULL, NULL, 0, NUM_MACROBLOCKS, 0, &mb_array, &block_array) == Success);
+
+	/* Test NULL surface */
+	assert(XvMCPutSurface(display, NULL, window, 0, 0, INPUT_WIDTH, INPUT_HEIGHT, 0, 0, output_width, output_height, XVMC_FRAME_PICTURE) == XvMCBadSurface);
+	/* Test bad window */
+	/* XXX: X halts with a bad drawable for some reason, doesn't return BadDrawable as expected */
+	/*assert(XvMCPutSurface(display, &surface, 0, 0, 0, width, height, 0, 0, width, height, XVMC_FRAME_PICTURE) == BadDrawable);*/
+
+	if (prompt)
+	{
+		puts("Press any button to quit...");
+
+		while (!quit)
+		{
+			if (XPending(display) > 0)
+			{
+				XEvent event;
+
+				XNextEvent(display, &event);
+
+				switch (event.type)
+				{
+					case Expose:
+					{
+						/* Test valid params */
+						assert
+						(
+							XvMCPutSurface
+							(
+								display, &surface, window,
+								0, 0, INPUT_WIDTH, INPUT_HEIGHT,
+								0, 0, output_width, output_height,
+								XVMC_FRAME_PICTURE
+							) == Success
+						);
+						break;
+					}
+					case KeyPress:
+					{
+						quit = 1;
+						break;
+					}
+				}
+			}
+		}
+	}
+
+	assert(XvMCDestroyBlocks(display, &block_array) == Success);
+	assert(XvMCDestroyMacroBlocks(display, &mb_array) == Success);
+	assert(XvMCDestroySurface(display, &surface) == Success);
+	assert(XvMCDestroyContext(display, &context) == Success);
+
+	XvUngrabPort(display, port_num, CurrentTime);
+	XDestroyWindow(display, window);
+	XCloseDisplay(display);
+
+	return 0;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/test_surface.c b/src/gallium/state_trackers/xorg/xvmc/tests/test_surface.c
new file mode 100644
index 0000000000..b65eb265c0
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/test_surface.c
@@ -0,0 +1,98 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <error.h>
+#include "testlib.h"
+
+int main(int argc, char **argv)
+{
+	const unsigned int	width = 16, height = 16;
+	const unsigned int	mc_types[2] = {XVMC_MOCOMP | XVMC_MPEG_2, XVMC_IDCT | XVMC_MPEG_2};
+
+	Display			*display;
+	XvPortID		port_num;
+	int			surface_type_id;
+	unsigned int		is_overlay, intra_unsigned;
+	int			colorkey;
+	XvMCContext		context;
+	XvMCSurface		surface = {0};
+
+	display = XOpenDisplay(NULL);
+
+	if (!GetPort
+	(
+		display,
+		width,
+		height,
+		XVMC_CHROMA_FORMAT_420,
+    		mc_types,
+    		2,
+    		&port_num,
+    		&surface_type_id,
+    		&is_overlay,
+    		&intra_unsigned
+	))
+	{
+		XCloseDisplay(display);
+		error(1, 0, "Error, unable to find a good port.\n");
+	}
+
+	if (is_overlay)
+	{
+		Atom xv_colorkey = XInternAtom(display, "XV_COLORKEY", 0);
+		XvGetPortAttribute(display, port_num, xv_colorkey, &colorkey);
+	}
+
+	assert(XvMCCreateContext(display, port_num, surface_type_id, width, height, XVMC_DIRECT, &context) == Success);
+
+	/* Test NULL context */
+	assert(XvMCCreateSurface(display, NULL, &surface) == XvMCBadContext);
+	/* Test NULL surface */
+	assert(XvMCCreateSurface(display, &context, NULL) == XvMCBadSurface);
+	/* Test valid params */
+	assert(XvMCCreateSurface(display, &context, &surface) == Success);
+	/* Test surface id assigned */
+	assert(surface.surface_id != 0);
+	/* Test context id assigned and correct */
+	assert(surface.context_id == context.context_id);
+	/* Test surface type id assigned and correct */
+	assert(surface.surface_type_id == surface_type_id);
+	/* Test width & height assigned and correct */
+	assert(surface.width == width && surface.height == height);
+	/* Test valid params */
+	assert(XvMCDestroySurface(display, &surface) == Success);
+	/* Test NULL surface */
+	assert(XvMCDestroySurface(display, NULL) == XvMCBadSurface);
+
+	assert(XvMCDestroyContext(display, &context) == Success);
+
+	XvUngrabPort(display, port_num, CurrentTime);
+	XCloseDisplay(display);
+
+	return 0;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/testlib.c b/src/gallium/state_trackers/xorg/xvmc/tests/testlib.c
new file mode 100644
index 0000000000..142c09bb59
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/testlib.c
@@ -0,0 +1,146 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 "testlib.h"
+#include <stdio.h>
+
+/*
+void test(int pred, const char *pred_string, const char *doc_string, const char *file, unsigned int line)
+{
+	fputs(doc_string, stderr);
+	if (!pred)
+		fprintf(stderr, " FAIL!\n\t\"%s\" at %s:%u\n", pred_string, file, line);
+	else
+		fputs(" PASS!\n", stderr);
+}
+*/
+
+int GetPort
+(
+	Display *display,
+	unsigned int width,
+	unsigned int height,
+	unsigned int chroma_format,
+	const unsigned int *mc_types,
+	unsigned int num_mc_types,
+	XvPortID *port_id,
+	int *surface_type_id,
+	unsigned int *is_overlay,
+	unsigned int *intra_unsigned
+)
+{
+	unsigned int	found_port = 0;
+	XvAdaptorInfo	*adaptor_info;
+	unsigned int	num_adaptors;
+	int		num_types;
+	int		ev_base, err_base;
+	unsigned int	i, j, k, l;
+
+	if (!XvMCQueryExtension(display, &ev_base, &err_base))
+		return 0;
+	if (XvQueryAdaptors(display, XDefaultRootWindow(display), &num_adaptors, &adaptor_info) != Success)
+		return 0;
+
+	for (i = 0; i < num_adaptors && !found_port; ++i)
+	{
+		if (adaptor_info[i].type & XvImageMask)
+		{
+			XvMCSurfaceInfo *surface_info = XvMCListSurfaceTypes(display, adaptor_info[i].base_id, &num_types);
+
+			if (surface_info)
+			{
+				for (j = 0; j < num_types && !found_port; ++j)
+				{
+					if
+					(
+						surface_info[j].chroma_format == chroma_format &&
+						surface_info[j].max_width >= width &&
+						surface_info[j].max_height >= height
+					)
+					{
+						for (k = 0; k < num_mc_types && !found_port; ++k)
+						{
+							if (surface_info[j].mc_type == mc_types[k])
+							{
+								for (l = 0; l < adaptor_info[i].num_ports && !found_port; ++l)
+								{
+									if (XvGrabPort(display, adaptor_info[i].base_id + l, CurrentTime) == Success)
+									{
+										*port_id = adaptor_info[i].base_id + l;
+										*surface_type_id = surface_info[j].surface_type_id;
+										*is_overlay = surface_info[j].flags & XVMC_OVERLAID_SURFACE;
+										*intra_unsigned = surface_info[j].flags & XVMC_INTRA_UNSIGNED;
+										found_port = 1;
+									}
+								}
+							}
+						}
+					}
+				}
+
+				XFree(surface_info);
+			}
+		}
+	}
+
+	XvFreeAdaptorInfo(adaptor_info);
+
+	return found_port;
+}
+
+unsigned int align(unsigned int value, unsigned int alignment)
+{
+	return (value + alignment - 1) & ~(alignment - 1);
+}
+
+/* From the glibc manual */
+int timeval_subtract(struct timeval *result, struct timeval *x, struct timeval *y)
+{
+	/* Perform the carry for the later subtraction by updating y. */
+	if (x->tv_usec < y->tv_usec)
+	{
+		int nsec = (y->tv_usec - x->tv_usec) / 1000000 + 1;
+		y->tv_usec -= 1000000 * nsec;
+		y->tv_sec += nsec;
+	}
+	if (x->tv_usec - y->tv_usec > 1000000)
+	{
+		int nsec = (x->tv_usec - y->tv_usec) / 1000000;
+		y->tv_usec += 1000000 * nsec;
+		y->tv_sec -= nsec;
+	}
+
+	/*
+	 * Compute the time remaining to wait.
+	 * tv_usec is certainly positive.
+	 */
+	result->tv_sec = x->tv_sec - y->tv_sec;
+	result->tv_usec = x->tv_usec - y->tv_usec;
+
+	/* Return 1 if result is negative. */
+	return x->tv_sec < y->tv_sec;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/testlib.h b/src/gallium/state_trackers/xorg/xvmc/tests/testlib.h
new file mode 100644
index 0000000000..0438e52928
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/testlib.h
@@ -0,0 +1,69 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 testlib_h
+#define testlib_h
+
+/*
+#define TEST(pred, doc)	test(pred, #pred, doc, __FILE__, __LINE__)
+
+void test(int pred, const char *pred_string, const char *doc_string, const char *file, unsigned int line);
+*/
+
+#include <sys/time.h>
+#include <X11/Xlib.h>
+#include <X11/extensions/XvMClib.h>
+
+/*
+ * display: IN			A valid X display
+ * width, height: IN		Surface size that the port must display
+ * chroma_format: IN		Chroma format that the port must display
+ * mc_types, num_mc_types: IN	List of MC types that the port must support, first port that matches the first mc_type will be returned
+ * port_id: OUT			Your port's ID
+ * surface_type_id: OUT		Your port's surface ID
+ * is_overlay: OUT		If 1, port uses overlay surfaces, you need to set a colorkey
+ * intra_unsigned: OUT		If 1, port uses unsigned values for intra-coded blocks
+ */
+int GetPort
+(
+	Display *display,
+	unsigned int width,
+	unsigned int height,
+	unsigned int chroma_format,
+	const unsigned int *mc_types,
+	unsigned int num_mc_types,
+	XvPortID *port_id,
+	int *surface_type_id,
+	unsigned int *is_overlay,
+	unsigned int *intra_unsigned
+);
+
+unsigned int align(unsigned int value, unsigned int alignment);
+
+int timeval_subtract(struct timeval *result, struct timeval *x, struct timeval *y);
+
+#endif
diff --git a/src/gallium/state_trackers/xorg/xvmc/tests/xvmc_bench.c b/src/gallium/state_trackers/xorg/xvmc/tests/xvmc_bench.c
new file mode 100644
index 0000000000..bf94d85623
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/tests/xvmc_bench.c
@@ -0,0 +1,300 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <error.h>
+#include <sys/time.h>
+#include "testlib.h"
+
+#define MACROBLOCK_WIDTH		16
+#define MACROBLOCK_HEIGHT		16
+#define BLOCKS_PER_MACROBLOCK		6
+
+#define DEFAULT_INPUT_WIDTH		720
+#define DEFAULT_INPUT_HEIGHT		480
+#define DEFAULT_REPS			100
+
+#define PIPELINE_STEP_MC		1
+#define PIPELINE_STEP_CSC		2
+#define PIPELINE_STEP_SWAP		4
+
+#define MB_TYPE_I			1
+#define MB_TYPE_P			2
+#define MB_TYPE_B			4
+
+struct Config
+{
+	unsigned int input_width;
+	unsigned int input_height;
+	unsigned int output_width;
+	unsigned int output_height;
+	unsigned int pipeline;
+	unsigned int mb_types;
+	unsigned int reps;
+};
+
+void ParseArgs(int argc, char **argv, struct Config *config);
+
+void ParseArgs(int argc, char **argv, struct Config *config)
+{
+	int fail = 0;
+	int i;
+
+	config->input_width = DEFAULT_INPUT_WIDTH;
+	config->input_height = DEFAULT_INPUT_HEIGHT;
+	config->output_width = 0;
+	config->output_height = 0;
+	config->pipeline = 0;
+	config->mb_types = 0;
+	config->reps = DEFAULT_REPS;
+
+	for (i = 1; i < argc && !fail; ++i)
+	{
+		if (!strcmp(argv[i], "-iw"))
+		{
+			if (sscanf(argv[++i], "%u", &config->input_width) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-ih"))
+		{
+			if (sscanf(argv[++i], "%u", &config->input_height) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-ow"))
+		{
+			if (sscanf(argv[++i], "%u", &config->output_width) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-oh"))
+		{
+			if (sscanf(argv[++i], "%u", &config->output_height) != 1)
+				fail = 1;
+		}
+		else if (!strcmp(argv[i], "-p"))
+		{
+			char *token = strtok(argv[++i], ",");
+
+			while (token && !fail)
+			{
+				if (!strcmp(token, "mc"))
+					config->pipeline |= PIPELINE_STEP_MC;
+				else if (!strcmp(token, "csc"))
+					config->pipeline |= PIPELINE_STEP_CSC;
+				else if (!strcmp(token, "swp"))
+					config->pipeline |= PIPELINE_STEP_SWAP;
+				else
+					fail = 1;
+
+				if (!fail)
+					token = strtok(NULL, ",");
+			}
+		}
+		else if (!strcmp(argv[i], "-mb"))
+		{
+			char *token = strtok(argv[++i], ",");
+
+			while (token && !fail)
+			{
+				if (strcmp(token, "i"))
+					config->mb_types |= MB_TYPE_I;
+				else if (strcmp(token, "p"))
+					config->mb_types |= MB_TYPE_P;
+				else if (strcmp(token, "b"))
+					config->mb_types |= MB_TYPE_B;
+				else
+					fail = 1;
+
+				if (!fail)
+					token = strtok(NULL, ",");
+			}
+		}
+		else if (!strcmp(argv[i], "-r"))
+		{
+			if (sscanf(argv[++i], "%u", &config->reps) != 1)
+				fail = 1;
+		}
+		else
+			fail = 1;
+	}
+
+	if (fail)
+		error
+		(
+			1, 0,
+			"Bad argument.\n"
+			"\n"
+			"Usage: %s [options]\n"
+			"\t-iw <width>\tInput width\n"
+			"\t-ih <height>\tInput height\n"
+			"\t-ow <width>\tOutput width\n"
+			"\t-oh <height>\tOutput height\n"
+			"\t-p <pipeline>\tPipeline to test\n"
+			"\t-mb <mb type>\tMacroBlock types to use\n"
+			"\t-r <reps>\tRepetitions\n\n"
+			"\tPipeline steps: mc,csc,swap\n"
+			"\tMB types: i,p,b\n",
+			argv[0]
+		);
+
+	if (config->output_width == 0)
+		config->output_width = config->input_width;
+	if (config->output_height == 0)
+		config->output_height = config->input_height;
+	if (!config->pipeline)
+		config->pipeline = PIPELINE_STEP_MC | PIPELINE_STEP_CSC | PIPELINE_STEP_SWAP;
+	if (!config->mb_types)
+		config->mb_types = MB_TYPE_I | MB_TYPE_P | MB_TYPE_B;
+}
+
+int main(int argc, char **argv)
+{
+	struct Config		config;
+	Display			*display;
+	Window			root, window;
+	const unsigned int	mc_types[2] = {XVMC_MOCOMP | XVMC_MPEG_2, XVMC_IDCT | XVMC_MPEG_2};
+	XvPortID		port_num;
+	int			surface_type_id;
+	unsigned int		is_overlay, intra_unsigned;
+	int			colorkey;
+	XvMCContext		context;
+	XvMCSurface		surface;
+	XvMCBlockArray		block_array;
+	XvMCMacroBlockArray	mb_array;
+	unsigned int		mbw, mbh;
+	unsigned int		mbx, mby;
+	unsigned int		reps;
+	struct timeval		start, stop, diff;
+	double			diff_secs;
+
+	ParseArgs(argc, argv, &config);
+
+	mbw = align(config.input_width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
+	mbh = align(config.input_height, MACROBLOCK_HEIGHT) / MACROBLOCK_HEIGHT;
+
+	display = XOpenDisplay(NULL);
+
+	if (!GetPort
+	(
+		display,
+		config.input_width,
+		config.input_height,
+		XVMC_CHROMA_FORMAT_420,
+    		mc_types,
+    		2,
+    		&port_num,
+    		&surface_type_id,
+    		&is_overlay,
+    		&intra_unsigned
+	))
+	{
+		XCloseDisplay(display);
+		error(1, 0, "Error, unable to find a good port.\n");
+	}
+
+	if (is_overlay)
+	{
+		Atom xv_colorkey = XInternAtom(display, "XV_COLORKEY", 0);
+		XvGetPortAttribute(display, port_num, xv_colorkey, &colorkey);
+	}
+
+	root = XDefaultRootWindow(display);
+	window = XCreateSimpleWindow(display, root, 0, 0, config.output_width, config.output_height, 0, 0, colorkey);
+
+	assert(XvMCCreateContext(display, port_num, surface_type_id, config.input_width, config.input_height, XVMC_DIRECT, &context) == Success);
+	assert(XvMCCreateSurface(display, &context, &surface) == Success);
+	assert(XvMCCreateBlocks(display, &context, mbw * mbh * BLOCKS_PER_MACROBLOCK, &block_array) == Success);
+	assert(XvMCCreateMacroBlocks(display, &context, mbw * mbh, &mb_array) == Success);
+
+	for (mby = 0; mby < mbh; ++mby)
+		for (mbx = 0; mbx < mbw; ++mbx)
+		{
+			mb_array.macro_blocks[mby * mbw + mbx].x = mbx;
+			mb_array.macro_blocks[mby * mbw + mbx].y = mby;
+			mb_array.macro_blocks[mby * mbw + mbx].macroblock_type = XVMC_MB_TYPE_INTRA;
+			/*mb->motion_type = ;*/
+			/*mb->motion_vertical_field_select = ;*/
+			mb_array.macro_blocks[mby * mbw + mbx].dct_type = XVMC_DCT_TYPE_FRAME;
+			/*mb->PMV[0][0][0] = ;
+			mb->PMV[0][0][1] = ;
+			mb->PMV[0][1][0] = ;
+			mb->PMV[0][1][1] = ;
+			mb->PMV[1][0][0] = ;
+			mb->PMV[1][0][1] = ;
+			mb->PMV[1][1][0] = ;
+			mb->PMV[1][1][1] = ;*/
+			mb_array.macro_blocks[mby * mbw + mbx].index = (mby * mbw + mbx) * BLOCKS_PER_MACROBLOCK;
+			mb_array.macro_blocks[mby * mbw + mbx].coded_block_pattern = 0x3F;
+		}
+
+	XSelectInput(display, window, ExposureMask | KeyPressMask);
+	XMapWindow(display, window);
+	XSync(display, 0);
+
+	gettimeofday(&start, NULL);
+
+	for (reps = 0; reps < config.reps; ++reps)
+	{
+		if (config.pipeline & PIPELINE_STEP_MC)
+		{
+			assert(XvMCRenderSurface(display, &context, XVMC_FRAME_PICTURE, &surface, NULL, NULL, 0, mbw * mbh, 0, &mb_array, &block_array) == Success);
+			assert(XvMCFlushSurface(display, &surface) == Success);
+		}
+		if (config.pipeline & PIPELINE_STEP_CSC)
+			assert(XvMCPutSurface(display, &surface, window, 0, 0, config.input_width, config.input_height, 0, 0, config.output_width, config.output_height, XVMC_FRAME_PICTURE) == Success);
+	}
+
+	gettimeofday(&stop, NULL);
+
+	timeval_subtract(&diff, &stop, &start);
+	diff_secs = (double)diff.tv_sec + (double)diff.tv_usec / 1000000.0;
+
+	printf("XvMC Benchmark\n");
+	printf("Input: %u,%u\nOutput: %u,%u\n", config.input_width, config.input_height, config.output_width, config.output_height);
+	printf("Pipeline: ");
+	if (config.pipeline & PIPELINE_STEP_MC)
+		printf("|mc|");
+	if (config.pipeline & PIPELINE_STEP_CSC)
+		printf("|csc|");
+	if (config.pipeline & PIPELINE_STEP_SWAP)
+		printf("|swap|");
+	printf("\n");
+	printf("Reps: %u\n", config.reps);
+	printf("Total time: %.2lf (%.2lf reps / sec)\n", diff_secs, config.reps / diff_secs);
+
+	assert(XvMCDestroyBlocks(display, &block_array) == Success);
+	assert(XvMCDestroyMacroBlocks(display, &mb_array) == Success);
+	assert(XvMCDestroySurface(display, &surface) == Success);
+	assert(XvMCDestroyContext(display, &context) == Success);
+
+	XvUngrabPort(display, port_num, CurrentTime);
+	XDestroyWindow(display, window);
+	XCloseDisplay(display);
+
+	return 0;
+}
diff --git a/src/gallium/state_trackers/xorg/xvmc/xvmc_private.h b/src/gallium/state_trackers/xorg/xvmc/xvmc_private.h
new file mode 100644
index 0000000000..42337631ca
--- /dev/null
+++ b/src/gallium/state_trackers/xorg/xvmc/xvmc_private.h
@@ -0,0 +1,58 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 xvmc_private_h
+#define xvmc_private_h
+
+#include <X11/Xlib.h>
+#include <X11/extensions/XvMClib.h>
+
+#define BLOCK_SIZE_SAMPLES 64
+#define BLOCK_SIZE_BYTES (BLOCK_SIZE_SAMPLES * 2)
+
+struct pipe_video_context;
+struct pipe_surface;
+struct pipe_fence_handle;
+
+typedef struct
+{
+	struct pipe_video_context *vpipe;
+	struct pipe_surface *backbuffer;
+} XvMCContextPrivate;
+
+typedef struct
+{
+	struct pipe_video_surface *pipe_vsfc;
+	struct pipe_fence_handle *render_fence;
+	struct pipe_fence_handle *disp_fence;
+	
+	/* Some XvMC functions take a surface but not a context,
+	   so we keep track of which context each surface belongs to. */
+	XvMCContext *context;
+} XvMCSurfacePrivate;
+
+#endif /* xvmc_private_h */
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
index 4c5a1d2ea8..8b647a769b 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_api.c
@@ -7,6 +7,7 @@
 #include "i915simple/i915_context.h"
 #include "i915simple/i915_screen.h"
 
+#include "trace/tr_drm.h"
 
 /*
  * Helper functions
@@ -40,6 +41,7 @@ intel_drm_buffer_from_handle(struct intel_drm_winsys *idws,
                              const char* name, unsigned handle)
 {
    struct intel_drm_buffer *buf = CALLOC_STRUCT(intel_drm_buffer);
+   uint32_t tile = 0, swizzle = 0;
 
    if (!buf)
       return NULL;
@@ -52,6 +54,10 @@ intel_drm_buffer_from_handle(struct intel_drm_winsys *idws,
    if (!buf->bo)
       goto err;
 
+   drm_intel_bo_get_tiling(buf->bo, &tile, &swizzle);
+   if (tile != INTEL_TILE_NONE)
+      buf->map_gtt = TRUE;
+
    return (struct intel_buffer *)buf;
 
 err:
@@ -166,6 +172,7 @@ intel_drm_create_screen(struct drm_api *api, int drmFD,
    idws->base.destroy = intel_drm_winsys_destroy;
 
    idws->pools.gem = drm_intel_bufmgr_gem_init(idws->fd, idws->max_batch_size);
+   drm_intel_bufmgr_gem_enable_reuse(idws->pools.gem);
 
    idws->softpipe = FALSE;
    idws->dump_cmd = debug_get_bool_option("INTEL_DUMP_CMD", FALSE);
@@ -198,5 +205,5 @@ struct drm_api intel_drm_api =
 struct drm_api *
 drm_api_create()
 {
-   return &intel_drm_api;
+   return trace_drm_create(&intel_drm_api);
 }
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
index 0030f915a3..327e19fcd6 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
@@ -58,11 +58,17 @@ intel_drm_buffer_set_fence_reg(struct intel_winsys *iws,
                                unsigned stride,
                                enum intel_buffer_tile tile)
 {
+   struct intel_drm_buffer *buf = intel_drm_buffer(buffer);
    assert(I915_TILING_NONE == INTEL_TILE_NONE);
    assert(I915_TILING_X == INTEL_TILE_X);
    assert(I915_TILING_Y == INTEL_TILE_Y);
 
-   return drm_intel_bo_set_tiling(intel_bo(buffer), &tile, stride);
+   if (tile != INTEL_TILE_NONE) {
+      assert(buf->map_count == 0);
+      buf->map_gtt = TRUE;
+   }
+
+   return drm_intel_bo_set_tiling(buf->bo, &tile, stride);
 }
 
 static void *
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_drm.c b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
index a4011db0b8..caab33de1c 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_drm.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_drm.c
@@ -107,13 +107,18 @@ radeon_texture_from_shared_handle(struct drm_api *api,
                                   unsigned handle)
 {
     struct pipe_buffer *buffer;
+    struct pipe_texture *blanket;
 
     buffer = radeon_buffer_from_handle(api, screen, name, handle);
     if (!buffer) {
         return NULL;
     }
 
-    return screen->texture_blanket(screen, templ, &stride, buffer);
+    blanket = screen->texture_blanket(screen, templ, &stride, buffer);
+
+    pipe_buffer_reference(&buffer, NULL);
+
+    return blanket;
 }
 
 static boolean radeon_shared_handle_from_texture(struct drm_api *api,
diff --git a/src/gallium/winsys/drm/radeon/xorg/Makefile b/src/gallium/winsys/drm/radeon/xorg/Makefile
index 0241625f69..9fa16dab24 100644
--- a/src/gallium/winsys/drm/radeon/xorg/Makefile
+++ b/src/gallium/winsys/drm/radeon/xorg/Makefile
@@ -20,6 +20,8 @@ LIBS = \
 	$(GALLIUMDIR)/state_trackers/xorg/libxorgtracker.a \
 	$(GALLIUMDIR)/winsys/drm/radeon/core/libradeonwinsys.a \
 	$(TOP)/src/gallium/drivers/r300/libr300.a \
+	$(TOP)/src/gallium/drivers/trace/libtrace.a \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
 	$(GALLIUM_AUXILIARIES)
 
 #############################################
diff --git a/src/gallium/winsys/g3dvl/Makefile b/src/gallium/winsys/g3dvl/Makefile
new file mode 100644
index 0000000000..424ddea87a
--- /dev/null
+++ b/src/gallium/winsys/g3dvl/Makefile
@@ -0,0 +1,11 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+SUBDIRS = $(GALLIUM_WINSYS_DIRS)
+
+default install clean:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE) $@) || exit 1; \
+		fi \
+	done
diff --git a/src/gallium/winsys/g3dvl/vl_winsys.h b/src/gallium/winsys/g3dvl/vl_winsys.h
index c83db28dd9..22119f9559 100644
--- a/src/gallium/winsys/g3dvl/vl_winsys.h
+++ b/src/gallium/winsys/g3dvl/vl_winsys.h
@@ -1,14 +1,50 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 vl_winsys_h
 #define vl_winsys_h
 
 #include <X11/Xlib.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_format.h>
 
-struct pipe_context;
+struct pipe_screen;
+struct pipe_video_context;
 
-struct pipe_context* create_pipe_context(Display *display, int screen);
-int destroy_pipe_context(struct pipe_context *pipe);
-int bind_pipe_drawable(struct pipe_context *pipe, Drawable drawable);
-int unbind_pipe_drawable(struct pipe_context *pipe);
+struct pipe_screen*
+vl_screen_create(Display *display, int screen);
 
-#endif
+struct pipe_video_context*
+vl_video_create(struct pipe_screen *screen,
+                enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height);
 
+Drawable
+vl_video_bind_drawable(struct pipe_video_context *vpipe, Drawable drawable);
+
+#endif
diff --git a/src/gallium/winsys/g3dvl/xlib/Makefile b/src/gallium/winsys/g3dvl/xlib/Makefile
new file mode 100644
index 0000000000..cf765ef51a
--- /dev/null
+++ b/src/gallium/winsys/g3dvl/xlib/Makefile
@@ -0,0 +1,74 @@
+# This makefile produces a "stand-alone" libXvMCg3dvl.so which is
+# based on Xlib (no DRI HW acceleration)
+
+TOP = ../../../../..
+include $(TOP)/configs/current
+
+XVMC_MAJOR = 1
+XVMC_MINOR = 0
+XVMC_LIB = XvMCg3dvl
+XVMC_LIB_NAME = lib$(XVMC_LIB).so
+XVMC_LIB_DEPS = $(EXTRA_LIB_PATH) -lXvMC -lXv -lX11 -lm
+
+INCLUDES = -I$(TOP)/src/gallium/include \
+           -I$(TOP)/src/gallium/auxiliary \
+           -I$(TOP)/src/gallium/drivers \
+           -I$(TOP)/src/gallium/winsys/g3dvl
+
+DEFINES += -DGALLIUM_SOFTPIPE \
+	   -DGALLIUM_TRACE
+
+SOURCES = xsp_winsys.c
+
+# XXX: Hack, if we include libxvmctracker.a in LIBS none of the symbols are
+# pulled in by the linker because xsp_winsys.c doesn't refer to them
+OBJECTS = $(SOURCES:.c=.o) $(TOP)/src/gallium/state_trackers/xorg/xvmc/*.o
+
+LIBS = $(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+       $(TOP)/src/gallium/auxiliary/vl/libvl.a \
+       $(TOP)/src/gallium/auxiliary/tgsi/libtgsi.a \
+       $(TOP)/src/gallium/auxiliary/draw/libdraw.a \
+       $(TOP)/src/gallium/auxiliary/translate/libtranslate.a \
+       $(TOP)/src/gallium/auxiliary/cso_cache/libcso_cache.a \
+       $(TOP)/src/gallium/auxiliary/rtasm/librtasm.a \
+       $(TOP)/src/gallium/auxiliary/util/libutil.a
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(DEFINES) $(CFLAGS) $< -o $@
+
+.PHONY: default $(TOP)/$(LIB_DIR)/gallium clean
+
+default: depend $(TOP)/$(LIB_DIR)/gallium $(TOP)/$(LIB_DIR)/gallium/$(XVMC_LIB_NAME)
+
+$(TOP)/$(LIB_DIR)/gallium:
+	@mkdir -p $(TOP)/$(LIB_DIR)/gallium
+
+# Make the libXvMCg3dvl.so library
+$(TOP)/$(LIB_DIR)/gallium/$(XVMC_LIB_NAME): $(OBJECTS) $(LIBS) Makefile
+	$(MKLIB) -o $(XVMC_LIB) -linker '$(CC)' -ldflags '$(LDFLAGS)' \
+		-major $(XVMC_MAJOR) -minor $(XVMC_MINOR) $(MKLIB_OPTIONS) \
+		-install $(TOP)/$(LIB_DIR)/gallium -id $(INSTALL_LIB_DIR)/lib$(XVMC_LIB).1.dylib \
+		$(XVMC_LIB_DEPS) $(OBJECTS) $(LIBS)
+
+depend: $(SOURCES) Makefile
+	$(RM) depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DEFINES) $(INCLUDES) $(SOURCES)
+
+#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
+
+clean: Makefile
+	$(RM) $(TOP)/$(LIB_DIR)/gallium/$(XVMC_LIB_NAME)
+	$(RM) *.o *~
+	$(RM) depend depend.bak
+
+-include depend
diff --git a/src/gallium/winsys/g3dvl/xlib/xsp_winsys.c b/src/gallium/winsys/g3dvl/xlib/xsp_winsys.c
new file mode 100644
index 0000000000..0e5f5a587b
--- /dev/null
+++ b/src/gallium/winsys/g3dvl/xlib/xsp_winsys.c
@@ -0,0 +1,331 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 Younes Manton.
+ * 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 <vl_winsys.h>
+#include <X11/Xutil.h>
+#include <pipe/internal/p_winsys_screen.h>
+#include <pipe/p_state.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <util/u_math.h>
+#include <softpipe/sp_winsys.h>
+#include <softpipe/sp_video_context.h>
+#include <softpipe/sp_texture.h>
+
+/* pipe_winsys implementation */
+
+struct xsp_pipe_winsys
+{
+   struct pipe_winsys base;
+   Display *display;
+   int screen;
+   XImage *fbimage;
+};
+
+struct xsp_context
+{
+   Drawable drawable;
+
+   void (*pipe_destroy)(struct pipe_video_context *vpipe);
+};
+
+struct xsp_buffer
+{
+   struct pipe_buffer base;
+   boolean is_user_buffer;
+   void *data;
+   void *mapped_data;
+};
+
+static struct pipe_buffer* xsp_buffer_create(struct pipe_winsys *pws, unsigned alignment, unsigned usage, unsigned size)
+{
+   struct xsp_buffer *buffer;
+
+   assert(pws);
+
+   buffer = calloc(1, sizeof(struct xsp_buffer));
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.alignment = alignment;
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+   buffer->data = align_malloc(size, alignment);
+
+   return (struct pipe_buffer*)buffer;
+}
+
+static struct pipe_buffer* xsp_user_buffer_create(struct pipe_winsys *pws, void *data, unsigned size)
+{
+   struct xsp_buffer *buffer;
+
+   assert(pws);
+
+   buffer = calloc(1, sizeof(struct xsp_buffer));
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.size = size;
+   buffer->is_user_buffer = TRUE;
+   buffer->data = data;
+
+   return (struct pipe_buffer*)buffer;
+}
+
+static void* xsp_buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buffer, unsigned flags)
+{
+   struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
+
+   assert(pws);
+   assert(buffer);
+
+   xsp_buf->mapped_data = xsp_buf->data;
+
+   return xsp_buf->mapped_data;
+}
+
+static void xsp_buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buffer)
+{
+   struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
+
+   assert(pws);
+   assert(buffer);
+
+   xsp_buf->mapped_data = NULL;
+}
+
+static void xsp_buffer_destroy(struct pipe_buffer *buffer)
+{
+   struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
+
+   assert(buffer);
+
+   if (!xsp_buf->is_user_buffer)
+      align_free(xsp_buf->data);
+
+   free(xsp_buf);
+}
+
+static struct pipe_buffer* xsp_surface_buffer_create
+(
+   struct pipe_winsys *pws,
+   unsigned width,
+   unsigned height,
+   enum pipe_format format,
+   unsigned usage,
+   unsigned tex_usage,
+   unsigned *stride
+)
+{
+   const unsigned int ALIGNMENT = 1;
+   struct pipe_format_block block;
+   unsigned nblocksx, nblocksy;
+
+   pf_get_block(format, &block);
+   nblocksx = pf_get_nblocksx(&block, width);
+   nblocksy = pf_get_nblocksy(&block, height);
+   *stride = align(nblocksx * block.size, ALIGNMENT);
+
+   return pws->buffer_create(pws, ALIGNMENT, usage,
+                             *stride * nblocksy);
+}
+
+static void xsp_fence_reference(struct pipe_winsys *pws, struct pipe_fence_handle **ptr, struct pipe_fence_handle *fence)
+{
+   assert(pws);
+   assert(ptr);
+   assert(fence);
+}
+
+static int xsp_fence_signalled(struct pipe_winsys *pws, struct pipe_fence_handle *fence, unsigned flag)
+{
+   assert(pws);
+   assert(fence);
+
+   return 0;
+}
+
+static int xsp_fence_finish(struct pipe_winsys *pws, struct pipe_fence_handle *fence, unsigned flag)
+{
+   assert(pws);
+   assert(fence);
+
+   return 0;
+}
+
+static void xsp_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surface, void *context_private)
+{
+   struct xsp_pipe_winsys *xsp_winsys;
+   struct xsp_context *xsp_context;
+
+   assert(pws);
+   assert(surface);
+   assert(context_private);
+
+   xsp_winsys = (struct xsp_pipe_winsys*)pws;
+   xsp_context = (struct xsp_context*)context_private;
+   xsp_winsys->fbimage->width = surface->width;
+   xsp_winsys->fbimage->height = surface->height;
+   xsp_winsys->fbimage->bytes_per_line = surface->width * (xsp_winsys->fbimage->bits_per_pixel >> 3);
+   xsp_winsys->fbimage->data = (char*)((struct xsp_buffer *)softpipe_texture(surface->texture)->buffer)->data + surface->offset;
+
+   XPutImage
+   (
+      xsp_winsys->display, xsp_context->drawable,
+      XDefaultGC(xsp_winsys->display, xsp_winsys->screen),
+      xsp_winsys->fbimage, 0, 0, 0, 0,
+      surface->width, surface->height
+   );
+   XFlush(xsp_winsys->display);
+}
+
+static const char* xsp_get_name(struct pipe_winsys *pws)
+{
+   assert(pws);
+   return "X11 SoftPipe";
+}
+
+static void xsp_destroy(struct pipe_winsys *pws)
+{
+   struct xsp_pipe_winsys *xsp_winsys = (struct xsp_pipe_winsys*)pws;
+
+   assert(pws);
+
+   /* XDestroyImage() wants to free the data as well */
+   xsp_winsys->fbimage->data = NULL;
+
+   XDestroyImage(xsp_winsys->fbimage);
+   FREE(xsp_winsys);
+}
+
+/* Called through pipe_video_context::destroy() */
+static void xsp_pipe_destroy(struct pipe_video_context *vpipe)
+{
+   struct xsp_context *xsp_context;
+
+   assert(vpipe);
+
+   xsp_context = vpipe->priv;
+
+   /* Call the original destroy */
+   xsp_context->pipe_destroy(vpipe);
+
+   FREE(xsp_context);
+}
+
+/* Show starts here */
+
+Drawable
+vl_video_bind_drawable(struct pipe_video_context *vpipe, Drawable drawable)
+{
+   struct xsp_context *xsp_context;
+   Drawable old_drawable;
+
+   assert(vpipe);
+
+   xsp_context = vpipe->priv;
+   old_drawable = xsp_context->drawable;
+   xsp_context->drawable = drawable;
+
+   return old_drawable;
+}
+
+struct pipe_screen*
+vl_screen_create(Display *display, int screen)
+{
+   struct xsp_pipe_winsys *xsp_winsys;
+
+   assert(display);
+
+   xsp_winsys = CALLOC_STRUCT(xsp_pipe_winsys);
+   if (!xsp_winsys)
+      return NULL;
+
+   xsp_winsys->base.buffer_create = xsp_buffer_create;
+   xsp_winsys->base.user_buffer_create = xsp_user_buffer_create;
+   xsp_winsys->base.buffer_map = xsp_buffer_map;
+   xsp_winsys->base.buffer_unmap = xsp_buffer_unmap;
+   xsp_winsys->base.buffer_destroy = xsp_buffer_destroy;
+   xsp_winsys->base.surface_buffer_create = xsp_surface_buffer_create;
+   xsp_winsys->base.fence_reference = xsp_fence_reference;
+   xsp_winsys->base.fence_signalled = xsp_fence_signalled;
+   xsp_winsys->base.fence_finish = xsp_fence_finish;
+   xsp_winsys->base.flush_frontbuffer = xsp_flush_frontbuffer;
+   xsp_winsys->base.get_name = xsp_get_name;
+   xsp_winsys->base.destroy = xsp_destroy;
+   xsp_winsys->display = display;
+   xsp_winsys->screen = screen;
+   xsp_winsys->fbimage = XCreateImage
+   (
+      display,
+      XDefaultVisual(display, screen),
+      XDefaultDepth(display, screen),
+      ZPixmap,
+      0,
+      NULL,
+      0, /* Don't know the width and height until flush_frontbuffer */
+      0,
+      32,
+      0
+   );
+
+   if (!xsp_winsys->fbimage) {
+      FREE(xsp_winsys);
+      return NULL;
+   }
+
+   XInitImage(xsp_winsys->fbimage);
+
+   return softpipe_create_screen(&xsp_winsys->base);
+}
+
+struct pipe_video_context*
+vl_video_create(struct pipe_screen *screen,
+                enum pipe_video_profile profile,
+                enum pipe_video_chroma_format chroma_format,
+                unsigned width, unsigned height)
+{
+   struct pipe_video_context *vpipe;
+   struct xsp_context *xsp_context;
+
+   assert(screen);
+   assert(width && height);
+
+   vpipe = sp_video_create(screen, profile, chroma_format, width, height);
+   if (!vpipe)
+      return NULL;
+
+   xsp_context = CALLOC_STRUCT(xsp_context);
+   if (!xsp_context) {
+      vpipe->destroy(vpipe);
+      return NULL;
+   }
+
+   /* Override this so we can free our xsp_context when the pipe is freed */
+   xsp_context->pipe_destroy = vpipe->destroy;
+   vpipe->destroy = xsp_pipe_destroy;
+
+   vpipe->priv = xsp_context;
+
+   return vpipe;
+}
diff --git a/src/gallium/winsys/g3dvl/xsp_winsys.c b/src/gallium/winsys/g3dvl/xsp_winsys.c
deleted file mode 100644
index 37d60ce540..0000000000
--- a/src/gallium/winsys/g3dvl/xsp_winsys.c
+++ /dev/null
@@ -1,291 +0,0 @@
-#include "vl_winsys.h"
-#include <X11/Xutil.h>
-#include <pipe/internal/p_winsys_screen.h>
-#include <pipe/p_state.h>
-#include <pipe/p_inlines.h>
-#include <util/u_memory.h>
-#include <util/u_math.h>
-#include <softpipe/sp_winsys.h>
-#include <softpipe/sp_texture.h>
-
-/* pipe_winsys implementation */
-
-struct xsp_pipe_winsys
-{
-	struct pipe_winsys	base;
-	XImage			fbimage;
-};
-
-struct xsp_context
-{
-	Display			*display;
-	int			screen;
-	Drawable		drawable;
-	int			drawable_bound;
-};
-
-struct xsp_buffer
-{
-	struct pipe_buffer	base;
-	boolean			is_user_buffer;
-	void			*data;
-	void			*mapped_data;
-};
-
-static struct pipe_buffer* xsp_buffer_create(struct pipe_winsys *pws, unsigned alignment, unsigned usage, unsigned size)
-{
-	struct xsp_buffer *buffer;
-
-	assert(pws);
-
-	buffer = calloc(1, sizeof(struct xsp_buffer));
-	pipe_reference_init(&buffer->base.reference, 1);
-	buffer->base.alignment = alignment;
-	buffer->base.usage = usage;
-	buffer->base.size = size;
-	buffer->data = align_malloc(size, alignment);
-
-	return (struct pipe_buffer*)buffer;
-}
-
-static struct pipe_buffer* xsp_user_buffer_create(struct pipe_winsys *pws, void *data, unsigned size)
-{
-	struct xsp_buffer *buffer;
-
-	assert(pws);
-
-	buffer = calloc(1, sizeof(struct xsp_buffer));
-	pipe_reference_init(&buffer->base.reference, 1);
-	buffer->base.size = size;
-	buffer->is_user_buffer = TRUE;
-	buffer->data = data;
-
-	return (struct pipe_buffer*)buffer;
-}
-
-static void* xsp_buffer_map(struct pipe_winsys *pws, struct pipe_buffer *buffer, unsigned flags)
-{
-	struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
-
-	assert(pws);
-	assert(buffer);
-
-	xsp_buf->mapped_data = xsp_buf->data;
-
-	return xsp_buf->mapped_data;
-}
-
-static void xsp_buffer_unmap(struct pipe_winsys *pws, struct pipe_buffer *buffer)
-{
-	struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
-
-	assert(pws);
-	assert(buffer);
-
-	xsp_buf->mapped_data = NULL;
-}
-
-static void xsp_buffer_destroy(struct pipe_winsys *pws, struct pipe_buffer *buffer)
-{
-	struct xsp_buffer *xsp_buf = (struct xsp_buffer*)buffer;
-
-	assert(pws);
-	assert(buffer);
-
-	if (!xsp_buf->is_user_buffer)
-		align_free(xsp_buf->data);
-
-	free(xsp_buf);
-}
-
-static struct pipe_buffer* xsp_surface_buffer_create
-(
-	struct pipe_winsys *pws,
-	unsigned width,
-	unsigned height,
-	enum pipe_format format,
-	unsigned usage,
-	unsigned tex_usage,
-	unsigned *stride
-)
-{
-	const unsigned int ALIGNMENT = 1;
-	struct pipe_format_block block;
-	unsigned nblocksx, nblocksy;
-
-	pf_get_block(format, &block);
-	nblocksx = pf_get_nblocksx(&block, width);
-	nblocksy = pf_get_nblocksy(&block, height);
-	*stride = align(nblocksx * block.size, ALIGNMENT);
-
-	return pws->buffer_create(pws, ALIGNMENT,
-				  usage,
-				  *stride * nblocksy);
-}
-
-static void xsp_fence_reference(struct pipe_winsys *pws, struct pipe_fence_handle **ptr, struct pipe_fence_handle *fence)
-{
-	assert(pws);
-	assert(ptr);
-	assert(fence);
-}
-
-static int xsp_fence_signalled(struct pipe_winsys *pws, struct pipe_fence_handle *fence, unsigned flag)
-{
-	assert(pws);
-	assert(fence);
-
-	return 0;
-}
-
-static int xsp_fence_finish(struct pipe_winsys *pws, struct pipe_fence_handle *fence, unsigned flag)
-{
-	assert(pws);
-	assert(fence);
-
-	return 0;
-}
-
-static void xsp_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surface, void *context_private)
-{
-	struct xsp_pipe_winsys	*xsp_winsys;
-	struct xsp_context	*xsp_context;
-
-	assert(pws);
-	assert(surface);
-	assert(context_private);
-
-	xsp_winsys = (struct xsp_pipe_winsys*)pws;
-	xsp_context = (struct xsp_context*)context_private;
-
-	if (!xsp_context->drawable_bound)
-		return;
-
-	xsp_winsys->fbimage.width = surface->width;
-	xsp_winsys->fbimage.height = surface->height;
-	xsp_winsys->fbimage.bytes_per_line = surface->width * (xsp_winsys->fbimage.bits_per_pixel >> 3);
-	xsp_winsys->fbimage.data = ((struct xsp_buffer *)softpipe_texture(surface->texture)->buffer)->data + surface->offset;
-
-	XPutImage
-	(
-		xsp_context->display,
-		xsp_context->drawable,
-		XDefaultGC(xsp_context->display, xsp_context->screen),
-		&xsp_winsys->fbimage,
-		0,
-		0,
-		0,
-		0,
-		surface->width,
-		surface->height
-	);
-	XFlush(xsp_context->display);
-}
-
-static const char* xsp_get_name(struct pipe_winsys *pws)
-{
-	assert(pws);
-	return "X11 SoftPipe";
-}
-
-/* Show starts here */
-
-int bind_pipe_drawable(struct pipe_context *pipe, Drawable drawable)
-{
-	struct xsp_context *xsp_context;
-
-	assert(pipe);
-
-	xsp_context = pipe->priv;
-	xsp_context->drawable = drawable;
-	xsp_context->drawable_bound = 1;
-
-	return 0;
-}
-
-int unbind_pipe_drawable(struct pipe_context *pipe)
-{
-	struct xsp_context *xsp_context;
-
-	assert(pipe);
-
-	xsp_context = pipe->priv;
-	xsp_context->drawable_bound = 0;
-
-	return 0;
-}
-
-struct pipe_context* create_pipe_context(Display *display, int screen)
-{
-	struct xsp_pipe_winsys	*xsp_winsys;
-	struct xsp_context	*xsp_context;
-	struct pipe_screen	*sp_screen;
-	struct pipe_context	*sp_pipe;
-
-	assert(display);
-
-	xsp_winsys = calloc(1, sizeof(struct xsp_pipe_winsys));
-	xsp_winsys->base.buffer_create = xsp_buffer_create;
-	xsp_winsys->base.user_buffer_create = xsp_user_buffer_create;
-	xsp_winsys->base.buffer_map = xsp_buffer_map;
-	xsp_winsys->base.buffer_unmap = xsp_buffer_unmap;
-	xsp_winsys->base.buffer_destroy = xsp_buffer_destroy;
-	xsp_winsys->base.surface_buffer_create = xsp_surface_buffer_create;
-	xsp_winsys->base.fence_reference = xsp_fence_reference;
-	xsp_winsys->base.fence_signalled = xsp_fence_signalled;
-	xsp_winsys->base.fence_finish = xsp_fence_finish;
-	xsp_winsys->base.flush_frontbuffer = xsp_flush_frontbuffer;
-	xsp_winsys->base.get_name = xsp_get_name;
-
-	{
-		/* XXX: Can't use the returned XImage* directly,
-		since we don't have control over winsys destruction
-		and we wouldn't be able to free it */
-		XImage *template = XCreateImage
-		(
-			display,
-			XDefaultVisual(display, XDefaultScreen(display)),
-			XDefaultDepth(display, XDefaultScreen(display)),
-			ZPixmap,
-			0,
-			NULL,
-			0,	/* Don't know the width and height until flush_frontbuffer */
-			0,
-			32,
-			0
-		);
-
-		memcpy(&xsp_winsys->fbimage, template, sizeof(XImage));
-		XInitImage(&xsp_winsys->fbimage);
-
-		XDestroyImage(template);
-	}
-
-	sp_screen = softpipe_create_screen((struct pipe_winsys*)xsp_winsys);
-	sp_pipe = softpipe_create(sp_screen);
-
-	xsp_context = calloc(1, sizeof(struct xsp_context));
-	xsp_context->display = display;
-	xsp_context->screen = screen;
-
-	sp_pipe->priv = xsp_context;
-
-	return sp_pipe;
-}
-
-int destroy_pipe_context(struct pipe_context *pipe)
-{
-	struct pipe_screen *screen;
-	struct pipe_winsys *winsys;
-
-	assert(pipe);
-
-	screen = pipe->screen;
-	winsys = pipe->winsys;
-	free(pipe->priv);
-	pipe->destroy(pipe);
-	screen->destroy(screen);
-	free(winsys);
-
-	return 0;
-}
diff --git a/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c b/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c
index c0c33b45d5..e8bc0f55ac 100644
--- a/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c
+++ b/src/gallium/winsys/gdi/gdi_llvmpipe_winsys.c
@@ -43,7 +43,7 @@
 #include "util/u_memory.h"
 #include "llvmpipe/lp_winsys.h"
 #include "llvmpipe/lp_texture.h"
-#include "shared/stw_winsys.h"
+#include "stw_winsys.h"
 
 
 struct gdi_llvmpipe_displaytarget
@@ -234,9 +234,9 @@ gdi_llvmpipe_context_create(struct pipe_screen *screen)
 
 
 static void
-gdi_llvmpipe_flush_frontbuffer(struct pipe_screen *screen,
-                               struct pipe_surface *surface,
-                               HDC hDC)
+gdi_llvmpipe_present(struct pipe_screen *screen,
+                     struct pipe_surface *surface,
+                     HDC hDC)
 {
     struct llvmpipe_texture *texture;
     struct gdi_llvmpipe_displaytarget *gdt;
@@ -254,7 +254,11 @@ gdi_llvmpipe_flush_frontbuffer(struct pipe_screen *screen,
 static const struct stw_winsys stw_winsys = {
    &gdi_llvmpipe_screen_create,
    &gdi_llvmpipe_context_create,
-   &gdi_llvmpipe_flush_frontbuffer
+   &gdi_llvmpipe_present,
+   NULL, /* get_adapter_luid */
+   NULL, /* shared_surface_open */
+   NULL, /* shared_surface_close */
+   NULL  /* compose */
 };
 
 
diff --git a/src/gallium/winsys/gdi/gdi_softpipe_winsys.c b/src/gallium/winsys/gdi/gdi_softpipe_winsys.c
index 66120a6a98..5e0ccf32f4 100644
--- a/src/gallium/winsys/gdi/gdi_softpipe_winsys.c
+++ b/src/gallium/winsys/gdi/gdi_softpipe_winsys.c
@@ -46,7 +46,7 @@
 #include "util/u_memory.h"
 #include "softpipe/sp_winsys.h"
 #include "softpipe/sp_texture.h"
-#include "shared/stw_winsys.h"
+#include "stw_winsys.h"
 
 
 struct gdi_softpipe_buffer
@@ -269,9 +269,9 @@ gdi_softpipe_context_create(struct pipe_screen *screen)
 
 
 static void
-gdi_softpipe_flush_frontbuffer(struct pipe_screen *screen,
-                               struct pipe_surface *surface,
-                               HDC hDC)
+gdi_softpipe_present(struct pipe_screen *screen,
+                     struct pipe_surface *surface,
+                     HDC hDC)
 {
     struct softpipe_texture *texture;
     struct gdi_softpipe_buffer *buffer;
@@ -304,7 +304,11 @@ gdi_softpipe_flush_frontbuffer(struct pipe_screen *screen,
 static const struct stw_winsys stw_winsys = {
    &gdi_softpipe_screen_create,
    &gdi_softpipe_context_create,
-   &gdi_softpipe_flush_frontbuffer
+   &gdi_softpipe_present,
+   NULL, /* get_adapter_luid */
+   NULL, /* shared_surface_open */
+   NULL, /* shared_surface_close */
+   NULL  /* compose */
 };
 
 
diff --git a/src/gallium/winsys/xlib/xlib_softpipe.c b/src/gallium/winsys/xlib/xlib_softpipe.c
index 67fea023a3..260b39e2a0 100644
--- a/src/gallium/winsys/xlib/xlib_softpipe.c
+++ b/src/gallium/winsys/xlib/xlib_softpipe.c
@@ -75,9 +75,6 @@ struct xmesa_pipe_winsys
 {
    struct pipe_winsys base;
 /*   struct xmesa_visual *xm_visual; */
-#ifdef USE_XSHM
-   int shm;
-#endif
 };
 
 
@@ -93,11 +90,6 @@ xm_buffer( struct pipe_buffer *buf )
 /**
  * X Shared Memory Image extension code
  */
-#ifdef USE_XSHM
-#define XSHM_ENABLED(b) ((b)->shm)
-#else
-#define XSHM_ENABLED(b) 0
-#endif
 
 #ifdef USE_XSHM
 
@@ -116,23 +108,23 @@ mesaHandleXError(Display *dpy, XErrorEvent *event)
 }
 
 
-static GLboolean alloc_shm(struct xm_buffer *buf, unsigned size)
+static char *alloc_shm(struct xm_buffer *buf, unsigned size)
 {
    XShmSegmentInfo *const shminfo = & buf->shminfo;
 
    shminfo->shmid = shmget(IPC_PRIVATE, size, IPC_CREAT|0777);
    if (shminfo->shmid < 0) {
-      return GL_FALSE;
+      return NULL;
    }
 
    shminfo->shmaddr = (char *) shmat(shminfo->shmid, 0, 0);
    if (shminfo->shmaddr == (char *) -1) {
       shmctl(shminfo->shmid, IPC_RMID, 0);
-      return GL_FALSE;
+      return NULL;
    }
 
    shminfo->readOnly = False;
-   return GL_TRUE;
+   return shminfo->shmaddr;
 }
 
 
@@ -258,25 +250,30 @@ xlib_softpipe_display_surface(struct xmesa_buffer *b,
       return;
 
 #ifdef USE_XSHM
-   if (XSHM_ENABLED(xm_buf) && (xm_buf->tempImage == NULL)) {
-      assert(surf->texture->block.width == 1);
-      assert(surf->texture->block.height == 1);
-      alloc_shm_ximage(xm_buf, b, spt->stride[surf->level] /
-                       surf->texture->block.size, surf->height);
-   }
-#endif
+   if (xm_buf->shm)
+   {
+      if (xm_buf->tempImage == NULL) 
+      {
+         assert(surf->texture->block.width == 1);
+         assert(surf->texture->block.height == 1);
+         alloc_shm_ximage(xm_buf, b, spt->stride[surf->level] /
+                          surf->texture->block.size, surf->height);
+      }
 
-   ximage = (XSHM_ENABLED(xm_buf)) ? xm_buf->tempImage : b->tempImage;
-   ximage->data = xm_buf->data;
+      ximage = xm_buf->tempImage;
+      ximage->data = xm_buf->data;
 
-   /* display image in Window */
-#ifdef USE_XSHM
-   if (XSHM_ENABLED(xm_buf)) {
+      /* _debug_printf("XSHM\n"); */
       XShmPutImage(b->xm_visual->display, b->drawable, b->gc,
                    ximage, 0, 0, 0, 0, surf->width, surf->height, False);
-   } else
+   }
+   else
 #endif
    {
+      /* display image in Window */
+      ximage = b->tempImage;
+      ximage->data = xm_buf->data;
+
       /* check that the XImage has been previously initialized */
       assert(ximage->format);
       assert(ximage->bitmap_unit);
@@ -286,6 +283,7 @@ xlib_softpipe_display_surface(struct xmesa_buffer *b,
       ximage->height = surf->height;
       ximage->bytes_per_line = spt->stride[surf->level];
 
+      /* _debug_printf("XPUT\n"); */
       XPutImage(b->xm_visual->display, b->drawable, b->gc,
                 ximage, 0, 0, 0, 0, surf->width, surf->height);
    }
@@ -322,21 +320,6 @@ xm_buffer_create(struct pipe_winsys *pws,
                  unsigned size)
 {
    struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
-#ifdef USE_XSHM
-   struct xmesa_pipe_winsys *xpws = (struct xmesa_pipe_winsys *) pws;
-
-   buffer->shminfo.shmid = -1;
-   buffer->shminfo.shmaddr = (char *) -1;
-
-   if (xpws->shm && (usage & PIPE_BUFFER_USAGE_PIXEL) != 0) {
-      buffer->shm = xpws->shm;
-
-      if (alloc_shm(buffer, size)) {
-         buffer->data = buffer->shminfo.shmaddr;
-         buffer->shm = 1;
-      }
-   }
-#endif
 
    pipe_reference_init(&buffer->base.reference, 1);
    buffer->base.alignment = alignment;
@@ -363,9 +346,6 @@ xm_user_buffer_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
    buffer->base.size = bytes;
    buffer->userBuffer = TRUE;
    buffer->data = ptr;
-#ifdef USE_XSHM
-   buffer->shm = 0;
-#endif
 
    return &buffer->base;
 }
@@ -381,16 +361,44 @@ xm_surface_buffer_create(struct pipe_winsys *winsys,
 {
    const unsigned alignment = 64;
    struct pipe_format_block block;
-   unsigned nblocksx, nblocksy;
+   unsigned nblocksx, nblocksy, size;
 
    pf_get_block(format, &block);
    nblocksx = pf_get_nblocksx(&block, width);
    nblocksy = pf_get_nblocksy(&block, height);
    *stride = align(nblocksx * block.size, alignment);
+   size = *stride * nblocksy;
+
+#ifdef USE_XSHM
+   if (!debug_get_bool_option("XLIB_NO_SHM", FALSE))
+   {
+      struct xm_buffer *buffer = CALLOC_STRUCT(xm_buffer);
+
+      pipe_reference_init(&buffer->base.reference, 1);
+      buffer->base.alignment = alignment;
+      buffer->base.usage = usage;
+      buffer->base.size = size;
+      buffer->userBuffer = FALSE;
+      buffer->shminfo.shmid = -1;
+      buffer->shminfo.shmaddr = (char *) -1;
+      buffer->shm = TRUE;
+         
+      buffer->data = alloc_shm(buffer, size);
+      if (!buffer->data)
+         goto out;
+
+      return &buffer->base;
+         
+   out:
+      if (buffer)
+         FREE(buffer);
+   }
+#endif
+   
 
    return winsys->buffer_create(winsys, alignment,
                                 usage,
-                                *stride * nblocksy);
+                                size);
 }