Merge commit 'origin/gallium-0.2' into gallium-xlib-rework

67 files changed, 1586 insertions, 3722 deletions

diff --git a/src/gallium/auxiliary/draw/draw_vbuf.h b/src/gallium/auxiliary/draw/draw_vbuf.h
index 7e1df88f0b..a1c4c14445 100644
--- a/src/gallium/auxiliary/draw/draw_vbuf.h
+++ b/src/gallium/auxiliary/draw/draw_vbuf.h
@@ -37,6 +37,8 @@
 #define DRAW_VBUF_H_
 
 
+#include "pipe/p_compiler.h"
+
 
 struct pipe_rasterizer_state;
 struct draw_context;
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index 8b75136144..d35db57d57 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -199,6 +199,11 @@ draw_create_vs_ppc(struct draw_context *draw,
 
    ppc_init_func( &vs->ppc_program );
 
+#if 0
+   ppc_print_code(&vs->ppc_program, TRUE);
+   ppc_indent(&vs->ppc_program, 8);
+#endif
+
    if (!tgsi_emit_ppc( (struct tgsi_token *) vs->base.state.tokens,
 			&vs->ppc_program, 
                        (float (*)[4]) vs->base.immediates, 
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc.c b/src/gallium/auxiliary/rtasm/rtasm_ppc.c
index e9015ec2eb..1bb9026205 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc.c
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright (C) 2008 Tungsten Graphics, Inc.   All Rights Reserved.
+ * Copyright (C) 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"),
@@ -47,6 +48,8 @@ ppc_init_func(struct ppc_function *p)
 {
    uint i;
 
+   memset(p, 0, sizeof(*p));
+
    p->num_inst = 0;
    p->max_inst = 100; /* first guess at buffer size */
    p->store = rtasm_exec_malloc(p->max_inst * PPC_INST_SIZE);
@@ -54,6 +57,9 @@ ppc_init_func(struct ppc_function *p)
    p->fp_used = 0x0;
    p->vec_used = 0x0;
 
+   p->print = FALSE;
+   p->indent = 0;
+
    /* only allow using gp registers 3..12 for now */
    for (i = 0; i < 3; i++)
       ppc_reserve_register(p, i);
@@ -105,6 +111,42 @@ ppc_dump_func(const struct ppc_function *p)
 }
 
 
+void
+ppc_print_code(struct ppc_function *p, boolean enable)
+{
+   p->print = enable;
+}
+
+
+void
+ppc_indent(struct ppc_function *p, int spaces)
+{
+   p->indent += spaces;
+}
+
+
+static void
+indent(const struct ppc_function *p)
+{
+   int i;
+   for (i = 0; i < p->indent; i++) {
+      putchar(' ');
+   }
+}
+
+
+void
+ppc_comment(struct ppc_function *p, int rel_indent, const char *s)
+{
+   if (p->print) {
+      p->indent += rel_indent;
+      indent(p);
+      p->indent -= rel_indent;
+      printf("# %s\n", s);
+   }
+}
+
+
 /**
  * Mark a register as being unavailable.
  */
@@ -132,6 +174,7 @@ ppc_allocate_register(struct ppc_function *p)
          return i;
       }
    }
+   printf("OUT OF PPC registers!\n");
    return -1;
 }
 
@@ -163,6 +206,7 @@ ppc_allocate_fp_register(struct ppc_function *p)
          return i;
       }
    }
+   printf("OUT OF PPC FP registers!\n");
    return -1;
 }
 
@@ -194,6 +238,7 @@ ppc_allocate_vec_register(struct ppc_function *p)
          return i;
       }
    }
+   printf("OUT OF PPC VEC registers!\n");
    return -1;
 }
 
@@ -252,7 +297,8 @@ union vx_inst {
 };
 
 static INLINE void
-emit_vx(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB)
+emit_vx(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB,
+        const char *format, boolean transpose)
 {
    union vx_inst inst;
    inst.inst.op = 4;
@@ -261,6 +307,13 @@ emit_vx(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB)
    inst.inst.vB = vB;
    inst.inst.op2 = op2;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      if (transpose)
+         printf(format, vD, vB, vA);
+      else
+         printf(format, vD, vA, vB);
+   }
 }
 
 
@@ -277,7 +330,8 @@ union vxr_inst {
 };
 
 static INLINE void
-emit_vxr(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB)
+emit_vxr(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB,
+         const char *format)
 {
    union vxr_inst inst;
    inst.inst.op = 4;
@@ -287,6 +341,10 @@ emit_vxr(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB)
    inst.inst.rC = 0;
    inst.inst.op2 = op2;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vD, vA, vB);
+   }
 }
 
 
@@ -303,7 +361,8 @@ union va_inst {
 };
 
 static INLINE void
-emit_va(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB, uint vC)
+emit_va(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB, uint vC,
+        const char *format)
 {
    union va_inst inst;
    inst.inst.op = 4;
@@ -313,6 +372,10 @@ emit_va(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB, uint vC)
    inst.inst.vC = vC;
    inst.inst.op2 = op2;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vD, vA, vB, vC);
+   }
 }
 
 
@@ -396,7 +459,8 @@ union x_inst {
 };
 
 static INLINE void
-emit_x(struct ppc_function *p, uint op, uint vrs, uint ra, uint rb, uint op2)
+emit_x(struct ppc_function *p, uint op, uint vrs, uint ra, uint rb, uint op2,
+       const char *format)
 {
    union x_inst inst;
    inst.inst.op = op;
@@ -406,6 +470,10 @@ emit_x(struct ppc_function *p, uint op, uint vrs, uint ra, uint rb, uint op2)
    inst.inst.op2 = op2;
    inst.inst.unused = 0x0;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vrs, ra, rb);
+   }
 }
 
 
@@ -420,7 +488,8 @@ union d_inst {
 };
 
 static INLINE void
-emit_d(struct ppc_function *p, uint op, uint rt, uint ra, int si)
+emit_d(struct ppc_function *p, uint op, uint rt, uint ra, int si,
+       const char *format, boolean transpose)
 {
    union d_inst inst;
    assert(si >= -32768);
@@ -430,6 +499,13 @@ emit_d(struct ppc_function *p, uint op, uint rt, uint ra, int si)
    inst.inst.ra = ra;
    inst.inst.si = (unsigned) (si & 0xffff);
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      if (transpose)
+         printf(format, rt, si, ra);
+      else
+         printf(format, rt, ra, si);
+   }
 }
 
 
@@ -448,7 +524,7 @@ union a_inst {
 
 static INLINE void
 emit_a(struct ppc_function *p, uint op, uint frt, uint fra, uint frb, uint op2,
-       uint rc)
+       uint rc, const char *format)
 {
    union a_inst inst;
    inst.inst.op = op;
@@ -459,6 +535,10 @@ emit_a(struct ppc_function *p, uint op, uint frt, uint fra, uint frb, uint op2,
    inst.inst.op2 = op2;
    inst.inst.rc = rc;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, frt, fra, frb);
+   }
 }
 
 
@@ -477,7 +557,7 @@ union xo_inst {
 
 static INLINE void
 emit_xo(struct ppc_function *p, uint op, uint rt, uint ra, uint rb, uint oe,
-        uint op2, uint rc)
+        uint op2, uint rc, const char *format)
 {
    union xo_inst inst;
    inst.inst.op = op;
@@ -488,6 +568,10 @@ emit_xo(struct ppc_function *p, uint op, uint rt, uint ra, uint rb, uint oe,
    inst.inst.op2 = op2;
    inst.inst.rc = rc;
    emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, rt, ra, rb);
+   }
 }
 
 
@@ -502,140 +586,142 @@ emit_xo(struct ppc_function *p, uint op, uint rt, uint ra, uint rb, uint oe,
 void
 ppc_vaddfp(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 10, vD, vA, vB);
+   emit_vx(p, 10, vD, vA, vB, "vaddfp\t%u, v%u, v%u\n", FALSE);
 }
 
 /** vector float substract */
 void
 ppc_vsubfp(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 74, vD, vA, vB);
+   emit_vx(p, 74, vD, vA, vB, "vsubfp\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector float min */
 void
 ppc_vminfp(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1098, vD, vA, vB);
+   emit_vx(p, 1098, vD, vA, vB, "vminfp\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector float max */
 void
 ppc_vmaxfp(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1034, vD, vA, vB);
+   emit_vx(p, 1034, vD, vA, vB, "vmaxfp\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector float mult add: vD = vA * vB + vC */
 void
 ppc_vmaddfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
 {
-   emit_va(p, 46, vD, vA, vC, vB); /* note arg order */
+   /* note arg order */
+   emit_va(p, 46, vD, vA, vC, vB, "vmaddfp\tv%u, v%u, v%u, v%u\n");
 }
 
 /** vector float negative mult subtract: vD = vA - vB * vC */
 void
 ppc_vnmsubfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
 {
-   emit_va(p, 47, vD, vB, vA, vC); /* note arg order */
+   /* note arg order */
+   emit_va(p, 47, vD, vB, vA, vC, "vnmsubfp\tv%u, v%u, v%u, v%u\n");
 }
 
 /** vector float compare greater than */
 void
 ppc_vcmpgtfpx(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vxr(p, 710, vD, vA, vB);
+   emit_vxr(p, 710, vD, vA, vB, "vcmpgtfpx\tv%u, v%u, v%u");
 }
 
 /** vector float compare greater than or equal to */
 void
 ppc_vcmpgefpx(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vxr(p, 454, vD, vA, vB);
+   emit_vxr(p, 454, vD, vA, vB, "vcmpgefpx\tv%u, v%u, v%u");
 }
 
 /** vector float compare equal */
 void
 ppc_vcmpeqfpx(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vxr(p, 198, vD, vA, vB);
+   emit_vxr(p, 198, vD, vA, vB, "vcmpeqfpx\tv%u, v%u, v%u");
 }
 
 /** vector float 2^x */
 void
 ppc_vexptefp(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 394, vD, 0, vB);
+   emit_vx(p, 394, vD, 0, vB, "vexptefp\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float log2(x) */
 void
 ppc_vlogefp(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 458, vD, 0, vB);
+   emit_vx(p, 458, vD, 0, vB, "vlogefp\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float reciprocol */
 void
 ppc_vrefp(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 266, vD, 0, vB);
+   emit_vx(p, 266, vD, 0, vB, "vrefp\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float reciprocol sqrt estimate */
 void
 ppc_vrsqrtefp(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 330, vD, 0, vB);
+   emit_vx(p, 330, vD, 0, vB, "vrsqrtefp\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float round to negative infinity */
 void
 ppc_vrfim(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 714, vD, 0, vB);
+   emit_vx(p, 714, vD, 0, vB, "vrfim\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float round to positive infinity */
 void
 ppc_vrfip(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 650, vD, 0, vB);
+   emit_vx(p, 650, vD, 0, vB, "vrfip\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float round to nearest int */
 void
 ppc_vrfin(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 522, vD, 0, vB);
+   emit_vx(p, 522, vD, 0, vB, "vrfin\tv%u, 0%u, v%u\n", FALSE);
 }
 
 /** vector float round to int toward zero */
 void
 ppc_vrfiz(struct ppc_function *p, uint vD, uint vB)
 {
-   emit_vx(p, 586, vD, 0, vB);
+   emit_vx(p, 586, vD, 0, vB, "vrfiz\tv%u, 0%u, v%u\n", FALSE);
 }
 
-/** vector store: store vR at mem[vA+vB] */
+/** vector store: store vR at mem[rA+rB] */
 void
-ppc_stvx(struct ppc_function *p, uint vR, uint vA, uint vB)
+ppc_stvx(struct ppc_function *p, uint vR, uint rA, uint rB)
 {
-   emit_x(p, 31, vR, vA, vB, 231);
+   emit_x(p, 31, vR, rA, rB, 231, "stvx\tv%u, r%u, r%u\n");
 }
 
-/** vector load: vR = mem[vA+vB] */
+/** vector load: vR = mem[rA+rB] */
 void
-ppc_lvx(struct ppc_function *p, uint vR, uint vA, uint vB)
+ppc_lvx(struct ppc_function *p, uint vR, uint rA, uint rB)
 {
-   emit_x(p, 31, vR, vA, vB, 103);
+   emit_x(p, 31, vR, rA, rB, 103, "lvx\tv%u, r%u, r%u\n");
 }
 
 /** load vector element word: vR = mem_word[ra+rb] */
 void
-ppc_lvewx(struct ppc_function *p, uint vr, uint ra, uint rb)
+ppc_lvewx(struct ppc_function *p, uint vR, uint rA, uint rB)
 {
-   emit_x(p, 31, vr, ra, rb, 71);
+   emit_x(p, 31, vR, rA, rB, 71, "lvewx\tv%u, r%u, r%u\n");
 }
 
 
@@ -649,49 +735,63 @@ ppc_lvewx(struct ppc_function *p, uint vr, uint ra, uint rb)
 void
 ppc_vand(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1028, vD, vA, vB);
+   emit_vx(p, 1028, vD, vA, vB, "vand\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector and complement */
 void
 ppc_vandc(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1092, vD, vA, vB);
+   emit_vx(p, 1092, vD, vA, vB, "vandc\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector or */
 void
 ppc_vor(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1156, vD, vA, vB);
+   emit_vx(p, 1156, vD, vA, vB, "vor\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector nor */
 void
 ppc_vnor(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1284, vD, vA, vB);
+   emit_vx(p, 1284, vD, vA, vB, "vnor\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** vector xor */
 void
 ppc_vxor(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 1220, vD, vA, vB);
+   emit_vx(p, 1220, vD, vA, vB, "vxor\tv%u, v%u, v%u\n", FALSE);
 }
 
 /** Pseudo-instruction: vector move */
 void
 ppc_vmove(struct ppc_function *p, uint vD, uint vA)
 {
+   boolean print = p->print;
+   p->print = FALSE;
    ppc_vor(p, vD, vA, vA);
+   if (print) {
+      indent(p);
+      printf("vor\tv%u, v%u, v%u \t# v%u = v%u\n", vD, vA, vA, vD, vA);
+   }
+   p->print = print;
 }
 
 /** Set vector register to {0,0,0,0} */
 void
 ppc_vzero(struct ppc_function *p, uint vr)
 {
+   boolean print = p->print;
+   p->print = FALSE;
    ppc_vxor(p, vr, vr, vr);
+   if (print) {
+      indent(p);
+      printf("vxor\tv%u, v%u, v%u \t# v%u = {0,0,0,0}\n", vr, vr, vr, vr);
+   }
+   p->print = print;
 }
 
 
@@ -705,35 +805,35 @@ ppc_vzero(struct ppc_function *p, uint vr)
 void
 ppc_vperm(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
 {
-   emit_va(p, 43, vD, vA, vB, vC);
+   emit_va(p, 43, vD, vA, vB, vC, "vperm\tr%u, r%u, r%u, r%u");
 }
 
 /** vector select */
 void
 ppc_vsel(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
 {
-   emit_va(p, 42, vD, vA, vB, vC);
+   emit_va(p, 42, vD, vA, vB, vC, "vsel\tr%u, r%u, r%u, r%u");
 }
 
 /** vector splat byte */
 void
 ppc_vspltb(struct ppc_function *p, uint vD, uint vB, uint imm)
 {
-   emit_vx(p, 42, vD, imm, vB);
+   emit_vx(p, 42, vD, imm, vB, "vspltb\tv%u, v%u, %u\n", TRUE);
 }
 
 /** vector splat half word */
 void
 ppc_vsplthw(struct ppc_function *p, uint vD, uint vB, uint imm)
 {
-   emit_vx(p, 588, vD, imm, vB);
+   emit_vx(p, 588, vD, imm, vB, "vsplthw\tv%u, v%u, %u\n", TRUE);
 }
 
 /** vector splat word */
 void
 ppc_vspltw(struct ppc_function *p, uint vD, uint vB, uint imm)
 {
-   emit_vx(p, 652, vD, imm, vB);
+   emit_vx(p, 652, vD, imm, vB, "vspltw\tv%u, v%u, %u\n", TRUE);
 }
 
 /** vector splat signed immediate word */
@@ -742,14 +842,14 @@ ppc_vspltisw(struct ppc_function *p, uint vD, int imm)
 {
    assert(imm >= -16);
    assert(imm < 15);
-   emit_vx(p, 908, vD, imm, 0);
+   emit_vx(p, 908, vD, imm, 0, "vspltisw\tv%u, %d, %u\n", FALSE);
 }
 
 /** vector shift left word: vD[word] = vA[word] << (vB[word] & 0x1f) */
 void
 ppc_vslw(struct ppc_function *p, uint vD, uint vA, uint vB)
 {
-   emit_vx(p, 388, vD, vA, vB);
+   emit_vx(p, 388, vD, vA, vB, "vslw\tv%u, v%u, v%u\n", FALSE);
 }
 
 
@@ -763,63 +863,66 @@ ppc_vslw(struct ppc_function *p, uint vD, uint vA, uint vB)
 void
 ppc_addi(struct ppc_function *p, uint rt, uint ra, int imm)
 {
-   emit_d(p, 14, rt, ra, imm);
+   emit_d(p, 14, rt, ra, imm, "addi\tr%u, r%u, %d\n", FALSE);
 }
 
 /** rt = ra + (imm << 16) */
 void
 ppc_addis(struct ppc_function *p, uint rt, uint ra, int imm)
 {
-   emit_d(p, 15, rt, ra, imm);
+   emit_d(p, 15, rt, ra, imm, "addis\tr%u, r%u, %d\n", FALSE);
 }
 
 /** rt = ra + rb */
 void
 ppc_add(struct ppc_function *p, uint rt, uint ra, uint rb)
 {
-   emit_xo(p, 31, rt, ra, rb, 0, 266, 0);
+   emit_xo(p, 31, rt, ra, rb, 0, 266, 0, "add\tr%u, r%u, r%u\n");
 }
 
 /** rt = ra AND ra */
 void
 ppc_and(struct ppc_function *p, uint rt, uint ra, uint rb)
 {
-   emit_x(p, 31, ra, rt, rb, 28);  /* note argument order */
+   emit_x(p, 31, ra, rt, rb, 28, "and\tr%u, r%u, r%u\n");  /* note argument order */
 }
 
 /** rt = ra AND imm */
 void
 ppc_andi(struct ppc_function *p, uint rt, uint ra, int imm)
 {
-   emit_d(p, 28, ra, rt, imm);  /* note argument order */
+   /* note argument order */
+   emit_d(p, 28, ra, rt, imm, "andi\tr%u, r%u, %d\n", FALSE);
 }
 
 /** rt = ra OR ra */
 void
 ppc_or(struct ppc_function *p, uint rt, uint ra, uint rb)
 {
-   emit_x(p, 31, ra, rt, rb, 444);  /* note argument order */
+   emit_x(p, 31, ra, rt, rb, 444, "or\tr%u, r%u, r%u\n");  /* note argument order */
 }
 
 /** rt = ra OR imm */
 void
 ppc_ori(struct ppc_function *p, uint rt, uint ra, int imm)
 {
-   emit_d(p, 24, ra, rt, imm);  /* note argument order */
+   /* note argument order */
+   emit_d(p, 24, ra, rt, imm, "ori\tr%u, r%u, %d\n", FALSE);
 }
 
 /** rt = ra XOR ra */
 void
 ppc_xor(struct ppc_function *p, uint rt, uint ra, uint rb)
 {
-   emit_x(p, 31, ra, rt, rb, 316);  /* note argument order */
+   emit_x(p, 31, ra, rt, rb, 316, "xor\tr%u, r%u, r%u\n");  /* note argument order */
 }
 
 /** rt = ra XOR imm */
 void
 ppc_xori(struct ppc_function *p, uint rt, uint ra, int imm)
 {
-   emit_d(p, 26, ra, rt, imm);  /* note argument order */
+   /* note argument order */
+   emit_d(p, 26, ra, rt, imm, "xori\tr%u, r%u, %d\n", FALSE);
 }
 
 /** pseudo instruction: move: rt = ra */
@@ -833,7 +936,14 @@ ppc_mr(struct ppc_function *p, uint rt, uint ra)
 void
 ppc_li(struct ppc_function *p, uint rt, int imm)
 {
+   boolean print = p->print;
+   p->print = FALSE;
    ppc_addi(p, rt, 0, imm);
+   if (print) {
+      indent(p);
+      printf("addi\tr%u, r0, %d \t# r%u = %d\n", rt, imm, rt, imm);
+   }
+   p->print = print;
 }
 
 /** rt = imm << 16 */
@@ -864,21 +974,21 @@ ppc_load_int(struct ppc_function *p, uint rt, int imm)
 void
 ppc_stwu(struct ppc_function *p, uint rs, uint ra, int d)
 {
-   emit_d(p, 37, rs, ra, d);
+   emit_d(p, 37, rs, ra, d, "stwu\tr%u, %d(r%u)\n", TRUE);
 }
 
 /** store rs at memory[(ra)+d] */
 void
 ppc_stw(struct ppc_function *p, uint rs, uint ra, int d)
 {
-   emit_d(p, 36, rs, ra, d);
+   emit_d(p, 36, rs, ra, d, "stw\tr%u, %d(r%u)\n", TRUE);
 }
 
 /** Load rt = mem[(ra)+d];  then zero set high 32 bits to zero. */
 void
 ppc_lwz(struct ppc_function *p, uint rt, uint ra, int d)
 {
-   emit_d(p, 32, rt, ra, d);
+   emit_d(p, 32, rt, ra, d, "lwz\tr%u, %d(r%u)\n", TRUE);
 }
 
 
@@ -891,42 +1001,42 @@ ppc_lwz(struct ppc_function *p, uint rt, uint ra, int d)
 void
 ppc_fadd(struct ppc_function *p, uint frt, uint fra, uint frb)
 {
-   emit_a(p, 63, frt, fra, frb, 21, 0);
+   emit_a(p, 63, frt, fra, frb, 21, 0, "fadd\tf%u, f%u, f%u\n");
 }
 
 /** sub: frt = fra - frb */
 void
 ppc_fsub(struct ppc_function *p, uint frt, uint fra, uint frb)
 {
-   emit_a(p, 63, frt, fra, frb, 20, 0);
+   emit_a(p, 63, frt, fra, frb, 20, 0, "fsub\tf%u, f%u, f%u\n");
 }
 
 /** convert to int: rt = (int) ra */
 void
 ppc_fctiwz(struct ppc_function *p, uint rt, uint fra)
 {
-   emit_x(p, 63, rt, 0, fra, 15);
+   emit_x(p, 63, rt, 0, fra, 15, "fctiwz\tr%u, r%u, r%u\n");
 }
 
 /** store frs at mem[(ra)+offset] */
 void
 ppc_stfs(struct ppc_function *p, uint frs, uint ra, int offset)
 {
-   emit_d(p, 52, frs, ra, offset);
+   emit_d(p, 52, frs, ra, offset, "stfs\tr%u, %d(r%u)\n", TRUE);
 }
 
 /** store frs at mem[(ra)+(rb)] */
 void
 ppc_stfiwx(struct ppc_function *p, uint frs, uint ra, uint rb)
 {
-   emit_x(p, 31, frs, ra, rb, 983);
+   emit_x(p, 31, frs, ra, rb, 983, "stfiwx\tr%u, r%u, r%u\n");
 }
 
 /** load frt = mem[(ra)+offset] */
 void
 ppc_lfs(struct ppc_function *p, uint frt, uint ra, int offset)
 {
-   emit_d(p, 48, frt, ra, offset);
+   emit_d(p, 48, frt, ra, offset, "stfs\tr%u, %d(r%u)\n", TRUE);
 }
 
 
@@ -942,6 +1052,10 @@ void
 ppc_blr(struct ppc_function *p)
 {
    emit_i(p, 18, 0, 0, 1);
+   if (p->print) {
+      indent(p);
+      printf("blr\n");
+   }
 }
 
 /** Branch Conditional to Link Register (p. 36) */
@@ -949,6 +1063,10 @@ void
 ppc_bclr(struct ppc_function *p, uint condOp, uint branchHint, uint condReg)
 {
    emit_xl(p, 19, condOp, condReg, branchHint, 16, 0);
+   if (p->print) {
+      indent(p);
+      printf("bclr\t%u %u %u\n", condOp, branchHint, condReg);
+   }
 }
 
 /** Pseudo instruction: return from subroutine */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc.h b/src/gallium/auxiliary/rtasm/rtasm_ppc.h
index 08212a2a25..93e5f5187d 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc.h
@@ -1,6 +1,7 @@
 /**************************************************************************
  *
  * Copyright (C) 2008 Tungsten Graphics, Inc.   All Rights Reserved.
+ * Copyright (C) 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"),
@@ -58,6 +59,8 @@ struct ppc_function
    uint32_t reg_used;   /** used/free general-purpose registers bitmask */
    uint32_t fp_used;   /** used/free floating point registers bitmask */
    uint32_t vec_used;   /** used/free vector registers bitmask */
+   int indent;
+   boolean print;
 };
 
 
@@ -68,6 +71,10 @@ extern uint ppc_num_instructions(const struct ppc_function *p);
 extern void (*ppc_get_func( struct ppc_function *p ))( void );
 extern void ppc_dump_func(const struct ppc_function *p);
 
+extern void ppc_print_code(struct ppc_function *p, boolean enable);
+extern void ppc_indent(struct ppc_function *p, int spaces);
+extern void ppc_comment(struct ppc_function *p, int rel_indent, const char *s);
+
 extern int ppc_reserve_register(struct ppc_function *p, int reg);
 extern int ppc_allocate_register(struct ppc_function *p);
 extern void ppc_release_register(struct ppc_function *p, int reg);
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index b9a75ae559..071bc2015c 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -443,7 +443,7 @@ void spe_init_func(struct spe_function *p, unsigned code_size)
       p->regs[i] = 1;
     }
 
-    p->print = false;
+    p->print = FALSE;
     p->indent = 0;
 }
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
index a92b1902e3..1a4db47501 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ppc.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
@@ -78,15 +78,7 @@ const float ppc_builtin_constants[] ALIGN16_ATTRIB = {
  * How many TGSI temps should be implemented with real PPC vector registers
  * rather than memory.
  */
-#define MAX_PPC_TEMPS 4
-
-
-struct reg_chan_vec
-{
-   struct tgsi_full_src_register src;
-   uint chan;
-   uint vec;
-};
+#define MAX_PPC_TEMPS 3
 
 
 /**
@@ -158,6 +150,29 @@ init_gen_context(struct gen_context *gen, struct ppc_function *func)
 
 
 /**
+ * Is the given TGSI register stored as a real PPC vector register?
+ */
+static boolean
+is_ppc_vec_temporary(const struct tgsi_full_src_register *reg)
+{
+   return (reg->SrcRegister.File == TGSI_FILE_TEMPORARY &&
+           reg->SrcRegister.Index < MAX_PPC_TEMPS);
+}
+
+
+/**
+ * Is the given TGSI register stored as a real PPC vector register?
+ */
+static boolean
+is_ppc_vec_temporary_dst(const struct tgsi_full_dst_register *reg)
+{
+   return (reg->DstRegister.File == TGSI_FILE_TEMPORARY &&
+           reg->DstRegister.Index < MAX_PPC_TEMPS);
+}
+
+
+
+/**
  * All PPC vector load/store instructions form an effective address
  * by adding the contents of two registers.  For example:
  *    lvx v2,r8,r9   # v2 = memory[r8 + r9]
@@ -285,7 +300,7 @@ emit_fetch(struct gen_context *gen,
          }
          break;
       case TGSI_FILE_TEMPORARY:
-         if (reg->SrcRegister.Index < MAX_PPC_TEMPS) {
+         if (is_ppc_vec_temporary(reg)) {
             /* use PPC vec register */
             dst_vec = gen->temps_map[reg->SrcRegister.Index][swizzle];
          }
@@ -353,23 +368,33 @@ emit_fetch(struct gen_context *gen,
       uint sign_op = tgsi_util_get_full_src_register_sign_mode(reg, chan_index);
       if (sign_op != TGSI_UTIL_SIGN_KEEP) {
          int bit31_vec = gen_get_bit31_vec(gen);
+         int dst_vec2;
+
+         if (is_ppc_vec_temporary(reg)) {
+            /* need to use a new temp */
+            dst_vec2 = ppc_allocate_vec_register(gen->f);
+         }
+         else {
+            dst_vec2 = dst_vec;
+         }
 
          switch (sign_op) {
          case TGSI_UTIL_SIGN_CLEAR:
             /* vec = vec & ~bit31 */
-            ppc_vandc(gen->f, dst_vec, dst_vec, bit31_vec);
+            ppc_vandc(gen->f, dst_vec2, dst_vec, bit31_vec);
             break;
          case TGSI_UTIL_SIGN_SET:
             /* vec = vec | bit31 */
-            ppc_vor(gen->f, dst_vec, dst_vec, bit31_vec);
+            ppc_vor(gen->f, dst_vec2, dst_vec, bit31_vec);
             break;
          case TGSI_UTIL_SIGN_TOGGLE:
             /* vec = vec ^ bit31 */
-            ppc_vxor(gen->f, dst_vec, dst_vec, bit31_vec);
+            ppc_vxor(gen->f, dst_vec2, dst_vec, bit31_vec);
             break;
          default:
             assert(0);
          }
+         return dst_vec2;
       }
    }
 
@@ -452,8 +477,7 @@ release_src_vecs(struct gen_context *gen)
    uint i;
    for (i = 0; i < gen->num_regs; i++) {
       const const struct tgsi_full_src_register src = gen->regs[i].src;
-      if (!(src.SrcRegister.File == TGSI_FILE_TEMPORARY &&
-            src.SrcRegister.Index < MAX_PPC_TEMPS)) {
+      if (!is_ppc_vec_temporary(&src)) {
          ppc_release_vec_register(gen->f, gen->regs[i].vec);
       }
    }
@@ -469,8 +493,7 @@ get_dst_vec(struct gen_context *gen,
 {
    const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[0];
 
-   if (reg->DstRegister.File == TGSI_FILE_TEMPORARY &&
-       reg->DstRegister.Index < MAX_PPC_TEMPS) {
+   if (is_ppc_vec_temporary_dst(reg)) {
       int vec = gen->temps_map[reg->DstRegister.Index][chan_index];
       return vec;
    }
@@ -502,7 +525,7 @@ emit_store(struct gen_context *gen,
       }
       break;
    case TGSI_FILE_TEMPORARY:
-      if (reg->DstRegister.Index < MAX_PPC_TEMPS) {
+      if (is_ppc_vec_temporary_dst(reg)) {
          if (!free_vec) {
             int dst_vec = gen->temps_map[reg->DstRegister.Index][chan_index];
             if (dst_vec != src_vec)
@@ -584,6 +607,7 @@ static void
 emit_unaryop(struct gen_context *gen, struct tgsi_full_instruction *inst)
 {
    uint chan_index;
+
    FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan_index) {
       int v0 = get_src_vec(gen, inst, 0, chan_index);   /* v0 = srcreg[0] */
       int v1 = get_dst_vec(gen, inst, chan_index);
@@ -770,7 +794,7 @@ emit_dotprod(struct gen_context *gen, struct tgsi_full_instruction *inst)
 
    v2 = ppc_allocate_vec_register(gen->f);
 
-   ppc_vxor(gen->f, v2, v2, v2);           /* v2 = {0, 0, 0, 0} */
+   ppc_vzero(gen->f, v2);                  /* v2 = {0, 0, 0, 0} */
 
    v0 = get_src_vec(gen, inst, 0, CHAN_X); /* v0 = src0.XXXX */
    v1 = get_src_vec(gen, inst, 1, CHAN_X); /* v1 = src1.XXXX */
@@ -815,7 +839,7 @@ ppc_vec_pow(struct ppc_function *f, int vr, int va, int vb)
    ppc_vzero(f, zero_vec);
 
    ppc_vlogefp(f, t_vec, va);                   /* t = log2(va) */
-   ppc_vmaddfp(f, t_vec, t_vec, vb, zero_vec);  /* t = t * vb */
+   ppc_vmaddfp(f, t_vec, t_vec, vb, zero_vec);  /* t = t * vb + zero */
    ppc_vexptefp(f, vr, t_vec);                  /* vr = 2^t */
 
    ppc_release_vec_register(f, t_vec);
@@ -1221,9 +1245,12 @@ emit_prologue(struct ppc_function *func)
 static void
 emit_epilogue(struct ppc_function *func)
 {
+   ppc_comment(func, -4, "Epilogue:");
    ppc_return(func);
    /* XXX restore prev stack frame */
+#if 0
    debug_printf("PPC: Emitted %u instructions\n", func->num_inst);
+#endif
 }
 
 
@@ -1248,6 +1275,7 @@ tgsi_emit_ppc(const struct tgsi_token *tokens,
    unsigned ok = 1;
    uint num_immediates = 0;
    struct gen_context gen;
+   uint ic = 0;
 
    if (use_ppc_asm < 0) {
       /* If GALLIUM_NOPPC is set, don't use PPC codegen */
@@ -1280,6 +1308,12 @@ tgsi_emit_ppc(const struct tgsi_token *tokens,
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (func->print) {
+            _debug_printf("# ");
+            ic++;
+            tgsi_dump_instruction(&parse.FullToken.FullInstruction, ic);
+         }
+
          ok = emit_instruction(&gen, &parse.FullToken.FullInstruction);
 
 	 if (!ok) {
diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index 12d7ef9a37..c92f8e5cba 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -54,6 +54,9 @@ INCLUDE_DIRS = \
 	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
 
 
+.c.s:
+	$(CC) -S $(INCLUDE_DIRS) $(CFLAGS) $< -o $@
+
 
 default: $(CELL_LIB)
 
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fp.c b/src/gallium/drivers/cell/ppu/cell_gen_fp.c
index 8f3deb482e..5a889a6119 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fp.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fp.c
@@ -145,7 +145,7 @@ get_const_one_reg(struct codegen *gen)
       gen->one_reg = spe_allocate_available_register(gen->f);
 
       spe_indent(gen->f, 4);
-      spe_comment(gen->f, -4, "INIT CONSTANT 1.0:");
+      spe_comment(gen->f, -4, "init constant reg = 1.0:");
 
       /* one = {1.0, 1.0, 1.0, 1.0} */
       spe_load_float(gen->f, gen->one_reg, 1.0f);
@@ -168,7 +168,7 @@ get_address_reg(struct codegen *gen)
       gen->addr_reg = spe_allocate_available_register(gen->f);
 
       spe_indent(gen->f, 4);
-      spe_comment(gen->f, -4, "INIT CONSTANT 1.0:");
+      spe_comment(gen->f, -4, "init address reg = 0:");
 
       /* init addr = {0, 0, 0, 0} */
       spe_zero(gen->f, gen->addr_reg);
@@ -479,7 +479,7 @@ emit_prologue(struct codegen *gen)
 {
    gen->frame_size = 1024; /* XXX temporary, should be dynamic */
 
-   spe_comment(gen->f, -4, "Function prologue:");
+   spe_comment(gen->f, 0, "Function prologue:");
 
    /* save $lr on stack     # stqd $lr,16($sp) */
    spe_stqd(gen->f, SPE_REG_RA, SPE_REG_SP, 16);
@@ -515,7 +515,7 @@ emit_epilogue(struct codegen *gen)
 {
    const int return_reg = 3;
 
-   spe_comment(gen->f, -4, "Function epilogue:");
+   spe_comment(gen->f, 0, "Function epilogue:");
 
    spe_comment(gen->f, 0, "return the killed mask");
    if (gen->kill_mask_reg > 0) {
@@ -561,8 +561,6 @@ emit_ARL(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch = 0, src_reg, addr_reg;
 
-   spe_comment(gen->f, -4, "ARL:");
-
    src_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
    addr_reg = get_address_reg(gen);
 
@@ -580,8 +578,6 @@ emit_MOV(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, src_reg[4], dst_reg[4];
 
-   spe_comment(gen->f, -4, "MOV:");
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       src_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       dst_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
@@ -612,20 +608,6 @@ emit_binop(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], s2_reg[4], d_reg[4];
 
-   switch (inst->Instruction.Opcode) {
-   case TGSI_OPCODE_ADD:
-      spe_comment(gen->f, -4, "ADD:");
-      break;
-   case TGSI_OPCODE_SUB:
-      spe_comment(gen->f, -4, "SUB:");
-      break;
-   case TGSI_OPCODE_MUL:
-      spe_comment(gen->f, -4, "MUL:");
-      break;
-   default:
-      assert(0);
-   }
-
    /* Loop over Red/Green/Blue/Alpha channels, fetch src operands */
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
@@ -670,7 +652,7 @@ static boolean
 emit_MAD(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], s2_reg[4], s3_reg[4], d_reg[4];
-   spe_comment(gen->f, -4, "MAD:");
+
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
@@ -695,7 +677,7 @@ static boolean
 emit_LERP(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], s2_reg[4], s3_reg[4], d_reg[4], tmp_reg[4];
-   spe_comment(gen->f, -4, "LERP:");
+
    /* setup/get src/dst/temp regs */
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
@@ -730,14 +712,6 @@ emit_RCP_RSQ(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], d_reg[4], tmp_reg[4];
 
-   if (inst->Instruction.Opcode == TGSI_OPCODE_RCP) {
-      spe_comment(gen->f, -4, "RCP:");
-   }
-   else {
-      assert(inst->Instruction.Opcode == TGSI_OPCODE_RSQ);
-      spe_comment(gen->f, -4, "RSQ:");
-   }
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
@@ -778,8 +752,6 @@ emit_ABS(struct codegen *gen, const struct tgsi_full_instruction *inst)
    int ch, s1_reg[4], d_reg[4];
    const int bit31mask_reg = get_itemp(gen);
 
-   spe_comment(gen->f, -4, "ABS:");
-
    /* mask with bit 31 set, the rest cleared */  
    spe_load_uint(gen->f, bit31mask_reg, (1 << 31));
 
@@ -812,8 +784,6 @@ emit_DP3(struct codegen *gen, const struct tgsi_full_instruction *inst)
    int s2x_reg, s2y_reg, s2z_reg;
    int t0_reg = get_itemp(gen), t1_reg = get_itemp(gen);
 
-   spe_comment(gen->f, -4, "DP3:");
-
    s1x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
    s2x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
    s1y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
@@ -854,8 +824,6 @@ emit_DP4(struct codegen *gen, const struct tgsi_full_instruction *inst)
    int s1x_reg, s1y_reg, s1z_reg, s1w_reg;
    int t0_reg = get_itemp(gen), t1_reg = get_itemp(gen);
 
-   spe_comment(gen->f, -4, "DP4:");
-
    s0x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
    s1x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
    s0y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
@@ -898,8 +866,6 @@ emit_DPH(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    /* XXX rewrite this function to look more like DP3/DP4 */
    int ch;
-   spe_comment(gen->f, -4, "DPH:");
-
    int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
    int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
    int tmp_reg = get_itemp(gen);
@@ -941,8 +907,6 @@ emit_NRM3(struct codegen *gen, const struct tgsi_full_instruction *inst)
    int src_reg[3];
    int t0_reg = get_itemp(gen), t1_reg = get_itemp(gen);
 
-   spe_comment(gen->f, -4, "NRM3:");
-
    src_reg[0] = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
    src_reg[1] = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
    src_reg[2] = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
@@ -981,8 +945,6 @@ emit_NRM3(struct codegen *gen, const struct tgsi_full_instruction *inst)
 static boolean
 emit_XPD(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
-   spe_comment(gen->f, -4, "XPD:");
-
    int s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
    int s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
    int tmp_reg = get_itemp(gen);
@@ -1044,32 +1006,6 @@ emit_inequality(struct codegen *gen, const struct tgsi_full_instruction *inst)
    int ch, s1_reg[4], s2_reg[4], d_reg[4], one_reg;
    bool complement = FALSE;
 
-   switch (inst->Instruction.Opcode) {
-   case TGSI_OPCODE_SGT:
-      spe_comment(gen->f, -4, "SGT:");
-      break;
-   case TGSI_OPCODE_SLT:
-      spe_comment(gen->f, -4, "SLT:");
-      break;
-   case TGSI_OPCODE_SGE:
-      spe_comment(gen->f, -4, "SGE:");
-      complement = TRUE;
-      break;
-   case TGSI_OPCODE_SLE:
-      spe_comment(gen->f, -4, "SLE:");
-      complement = TRUE;
-      break;
-   case TGSI_OPCODE_SEQ:
-      spe_comment(gen->f, -4, "SEQ:");
-      break;
-   case TGSI_OPCODE_SNE:
-      spe_comment(gen->f, -4, "SNE:");
-      complement = TRUE;
-      break;
-   default:
-      ;
-   }
-
    one_reg = get_const_one_reg(gen);
 
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
@@ -1088,15 +1024,18 @@ emit_inequality(struct codegen *gen, const struct tgsi_full_instruction *inst)
          break;
       case TGSI_OPCODE_SGE:
          spe_fcgt(gen->f, d_reg[ch], s2_reg[ch], s1_reg[ch]);
+         complement = TRUE;
          break;
       case TGSI_OPCODE_SLE:
          spe_fcgt(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]);
+         complement = TRUE;
          break;
       case TGSI_OPCODE_SEQ:
          spe_fceq(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]);
          break;
       case TGSI_OPCODE_SNE:
          spe_fceq(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]);
+         complement = TRUE;
          break;
       default:
          assert(0);
@@ -1129,8 +1068,6 @@ emit_CMP(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch;
 
-   spe_comment(gen->f, -4, "CMP:");
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
@@ -1161,8 +1098,6 @@ emit_TRUNC(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], d_reg[4];
 
-   spe_comment(gen->f, -4, "TRUNC:");
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
@@ -1198,8 +1133,6 @@ emit_FLR(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], d_reg[4], tmp_reg[4], zero_reg, one_reg;
 
-   spe_comment(gen->f, -4, "FLR:");
-
    zero_reg = get_itemp(gen);
    spe_zero(gen->f, zero_reg);
    one_reg = get_const_one_reg(gen);
@@ -1248,8 +1181,6 @@ emit_FRC(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s1_reg[4], d_reg[4], tmp_reg[4], zero_reg, one_reg;
 
-   spe_comment(gen->f, -4, "FRC:");
-
    zero_reg = get_itemp(gen);
    spe_zero(gen->f, zero_reg);
    one_reg = get_const_one_reg(gen);
@@ -1577,8 +1508,6 @@ emit_MIN_MAX(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int ch, s0_reg[4], s1_reg[4], d_reg[4], tmp_reg[4];
 
-   spe_comment(gen->f, -4, "MAX:");
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       s0_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
@@ -1646,8 +1575,6 @@ emit_IF(struct codegen *gen, const struct tgsi_full_instruction *inst)
    const int channel = 0;
    int cond_reg;
 
-   spe_comment(gen->f, -4, "IF:");
-
    cond_reg = get_cond_mask_reg(gen);
 
    /* XXX push cond exec mask */
@@ -1682,8 +1609,6 @@ emit_ELSE(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    const int cond_reg = get_cond_mask_reg(gen);
 
-   spe_comment(gen->f, -4, "ELSE:");
-
    spe_comment(gen->f, 0, "cond exec mask = !cond exec mask");
    spe_complement(gen->f, cond_reg, cond_reg);
    emit_update_exec_mask(gen);
@@ -1695,8 +1620,6 @@ emit_ELSE(struct codegen *gen, const struct tgsi_full_instruction *inst)
 static boolean
 emit_ENDIF(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
-   spe_comment(gen->f, -4, "ENDIF:");
-
    /* XXX todo: pop cond exec mask */
 
    gen->if_nesting--;
@@ -1712,8 +1635,6 @@ emit_BGNLOOP(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
    int exec_reg, loop_reg;
 
-   spe_comment(gen->f, -4, "BGNLOOP:");
-
    exec_reg = get_exec_mask_reg(gen);
    loop_reg = get_loop_mask_reg(gen);
 
@@ -1736,8 +1657,6 @@ emit_ENDLOOP(struct codegen *gen, const struct tgsi_full_instruction *inst)
    const int tmp_reg = get_itemp(gen);
    int offset;
 
-   spe_comment(gen->f, -4, "ENDLOOP:");
-
    /* tmp_reg = exec[0] | exec[1] | exec[2] | exec[3] */
    spe_orx(gen->f, tmp_reg, loop_reg);
 
@@ -1762,8 +1681,6 @@ emit_BRK(struct codegen *gen, const struct tgsi_full_instruction *inst)
    const int exec_reg = get_exec_mask_reg(gen);
    const int loop_reg = get_loop_mask_reg(gen);
 
-   spe_comment(gen->f, -4, "BREAK:");
-
    assert(gen->loop_nesting > 0);
 
    spe_comment(gen->f, 0, "loop exec mask &= ~master exec mask");
@@ -1778,8 +1695,6 @@ emit_BRK(struct codegen *gen, const struct tgsi_full_instruction *inst)
 static boolean
 emit_CONT(struct codegen *gen, const struct tgsi_full_instruction *inst)
 {
-   spe_comment(gen->f, -4, "CONT:");
-
    assert(gen->loop_nesting > 0);
 
    return TRUE;
@@ -1792,8 +1707,6 @@ emit_DDX_DDY(struct codegen *gen, const struct tgsi_full_instruction *inst,
 {
    int ch;
 
-   spe_comment(gen->f, -4, ddx ? "DDX:" : "DDY:");
-
    FOR_EACH_ENABLED_CHANNEL(inst, ch) {
       int s_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
       int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
@@ -1829,7 +1742,6 @@ emit_DDX_DDY(struct codegen *gen, const struct tgsi_full_instruction *inst,
 static boolean
 emit_END(struct codegen *gen)
 {
-   spe_comment(gen->f, -4, "END:");
    emit_epilogue(gen);
    return TRUE;
 }
@@ -1962,8 +1874,6 @@ emit_immediate(struct codegen *gen, const struct tgsi_full_immediate *immed)
 
    assert(gen->num_imm < MAX_TEMPS);
 
-   spe_comment(gen->f, -4, "IMMEDIATE:");
-
    for (ch = 0; ch < 4; ch++) {
       float val = immed->u.ImmediateFloat32[ch].Float;
 
@@ -2028,7 +1938,7 @@ emit_declaration(struct cell_context *cell,
             sprintf(buf, "TGSI temp[%d] maps to SPU regs [$%d $%d $%d $%d]", i,
                     gen->temp_regs[i][0], gen->temp_regs[i][1],
                     gen->temp_regs[i][2], gen->temp_regs[i][3]);
-            spe_comment(gen->f, -4, buf);
+            spe_comment(gen->f, 0, buf);
          }
       }
       break;
@@ -2056,6 +1966,7 @@ cell_gen_fragment_program(struct cell_context *cell,
 {
    struct tgsi_parse_context parse;
    struct codegen gen;
+   uint ic = 0;
 
    memset(&gen, 0, sizeof(gen));
    gen.cell = cell;
@@ -2073,7 +1984,7 @@ cell_gen_fragment_program(struct cell_context *cell,
 
    if (cell->debug_flags & CELL_DEBUG_ASM) {
       spe_print_code(f, TRUE);
-      spe_indent(f, 8);
+      spe_indent(f, 2*8);
       printf("Begin %s\n", __FUNCTION__);
       tgsi_dump(tokens, 0);
    }
@@ -2087,16 +1998,29 @@ cell_gen_fragment_program(struct cell_context *cell,
 
       switch (parse.FullToken.Token.Type) {
       case TGSI_TOKEN_TYPE_IMMEDIATE:
+         if (f->print) {
+            _debug_printf("    # ");
+            tgsi_dump_immediate(&parse.FullToken.FullImmediate);
+         }
          if (!emit_immediate(&gen, &parse.FullToken.FullImmediate))
             gen.error = TRUE;
          break;
 
       case TGSI_TOKEN_TYPE_DECLARATION:
+         if (f->print) {
+            _debug_printf("    # ");
+            tgsi_dump_declaration(&parse.FullToken.FullDeclaration);
+         }
          if (!emit_declaration(cell, &gen, &parse.FullToken.FullDeclaration))
             gen.error = TRUE;
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (f->print) {
+            _debug_printf("    # ");
+            ic++;
+            tgsi_dump_instruction(&parse.FullToken.FullInstruction, ic);
+         }
          if (!emit_instruction(&gen, &parse.FullToken.FullInstruction))
             gen.error = TRUE;
          break;
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index aa670b9a45..37d297cc0f 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -21,7 +21,7 @@ nv30_miptree_layout(struct nv30_miptree *nv30mt)
 	} else {
 		nr_faces = 1;
 	}
-	
+
 	pitch = pt->width[0];
 	for (l = 0; l <= pt->last_level; l++) {
 		pt->width[l] = width;
@@ -76,13 +76,15 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+	else
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else {
 		switch (pt->format) {
 		/* TODO: Figure out which formats can be swizzled */
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 		case PIPE_FORMAT_X8R8G8B8_UNORM:
-		/* XXX: Re-enable when SIFM size limits are fixed */
-		/*case PIPE_FORMAT_R16_SNORM:*/
+		case PIPE_FORMAT_R16_SNORM:
 			break;
 		default:
 			mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
@@ -192,4 +194,3 @@ nv30_screen_init_miptree_functions(struct pipe_screen *pscreen)
 	pscreen->get_tex_surface = nv30_miptree_surface_new;
 	pscreen->tex_surface_release = nv30_miptree_surface_del;
 }
-
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index d754892299..29356e8c1e 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -98,7 +98,7 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen,
 	if (tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) {
 		switch (format) {
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM: 
+		case PIPE_FORMAT_R5G6B5_UNORM:
 		case PIPE_FORMAT_Z24S8_UNORM:
 		case PIPE_FORMAT_Z16_UNORM:
 			return TRUE;
@@ -110,7 +110,7 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen,
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 		case PIPE_FORMAT_A1R5G5B5_UNORM:
 		case PIPE_FORMAT_A4R4G4B4_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM: 
+		case PIPE_FORMAT_R5G6B5_UNORM:
 		case PIPE_FORMAT_L8_UNORM:
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
@@ -139,7 +139,8 @@ nv30_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
 
 		if (!mt->shadow_tex) {
 			unsigned old_tex_usage = surface->texture->tex_usage;
-			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR;
+			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR |
+			                              PIPE_TEXTURE_USAGE_DYNAMIC;
 			mt->shadow_tex = screen->texture_create(screen, surface->texture);
 			surface->texture->tex_usage = old_tex_usage;
 
@@ -326,7 +327,7 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 	so_data  (so, 3);
 	so_method(so, screen->rankine, 0x1450, 1);
 	so_data  (so, 0x00030004);
-	
+
 	/* NEW */
 	so_method(so, screen->rankine, 0x1e98, 1);
 	so_data  (so, 0);
@@ -382,4 +383,3 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	return &screen->pipe;
 }
-
diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c
index b68967c07f..00ce6be985 100644
--- a/src/gallium/drivers/nv40/nv40_miptree.c
+++ b/src/gallium/drivers/nv40/nv40_miptree.c
@@ -57,6 +57,8 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 {
 	struct pipe_winsys *ws = pscreen->winsys;
 	struct nv40_miptree *mt;
+	unsigned buf_usage = PIPE_BUFFER_USAGE_PIXEL |
+	                     NOUVEAU_BUFFER_USAGE_TEXTURE;
 
 	mt = MALLOC(sizeof(struct nv40_miptree));
 	if (!mt)
@@ -75,25 +77,27 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+	else
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else {
 		switch (pt->format) {
 		/* TODO: Figure out which formats can be swizzled */
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 		case PIPE_FORMAT_X8R8G8B8_UNORM:
-		/* XXX: Re-enable when SIFM size limits are fixed */
-		/*case PIPE_FORMAT_R16_SNORM:*/
+		case PIPE_FORMAT_R16_SNORM:
 			break;
 		default:
 			mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 		}
 	}
 
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		buf_usage |= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+
 	nv40_miptree_layout(mt);
 
-	mt->buffer = ws->buffer_create(ws, 256,
-				       PIPE_BUFFER_USAGE_PIXEL |
-				       NOUVEAU_BUFFER_USAGE_TEXTURE,
-				       mt->total_size);
+	mt->buffer = ws->buffer_create(ws, 256, buf_usage, mt->total_size);
 	if (!mt->buffer) {
 		FREE(mt);
 		return NULL;
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index 995148e03f..9657a19c50 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -148,7 +148,8 @@ nv40_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
 
 		if (!mt->shadow_tex) {
 			unsigned old_tex_usage = surface->texture->tex_usage;
-			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR;
+			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR |
+			                              PIPE_TEXTURE_USAGE_DYNAMIC;
 			mt->shadow_tex = screen->texture_create(screen, surface->texture);
 			surface->texture->tex_usage = old_tex_usage;
 
diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
index 6bfac581f9..5c6a92b53b 100644
--- a/src/gallium/include/pipe/p_defines.h
+++ b/src/gallium/include/pipe/p_defines.h
@@ -171,6 +171,7 @@ enum pipe_texture_target {
 #define PIPE_TEXTURE_USAGE_PRIMARY         0x4 /* ie a frontbuffer */
 #define PIPE_TEXTURE_USAGE_DEPTH_STENCIL   0x8
 #define PIPE_TEXTURE_USAGE_SAMPLER         0x10
+#define PIPE_TEXTURE_USAGE_DYNAMIC         0x20
 /** Pipe driver custom usage flags should be greater or equal to this value */
 #define PIPE_TEXTURE_USAGE_CUSTOM          (1 << 16)
 
diff --git a/src/gallium/state_trackers/g3dvl/Makefile b/src/gallium/state_trackers/g3dvl/Makefile
index 4f7a953484..cddfca54fe 100644
--- a/src/gallium/state_trackers/g3dvl/Makefile
+++ b/src/gallium/state_trackers/g3dvl/Makefile
@@ -1,9 +1,12 @@
 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.o vl_r16snorm_mc_buf.o
+		  vl_r16snorm_mc_buf.o
 GALLIUMDIR	= ../..
 
-CFLAGS		+= -g -Wall -fPIC -I${GALLIUMDIR}/include -I${GALLIUMDIR}/auxiliary -I${GALLIUMDIR}/winsys/g3dvl
+CFLAGS		+= -g -Wall -Werror=implicit-function-declaration -fPIC	\
+		   -I${GALLIUMDIR}/include				\
+		   -I${GALLIUMDIR}/auxiliary				\
+		   -I${GALLIUMDIR}/winsys/g3dvl				\
 
 #############################################
 
diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
index e3b3d03256..3ce93cf49d 100644
--- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
+++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
@@ -1,13 +1,13 @@
 #define VL_INTERNAL
 #include "vl_basic_csc.h"
 #include <assert.h>
-#include <stdlib.h>
 #include <pipe/p_context.h>
 #include <pipe/p_winsys.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"
@@ -237,7 +237,7 @@ static int vlDestroy
 	pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->vs_const_buf.buffer);
 	pipe->winsys->buffer_destroy(pipe->winsys, basic_csc->fs_const_buf.buffer);
 
-	free(basic_csc);
+	FREE(basic_csc);
 
 	return 0;
 }
@@ -369,7 +369,7 @@ static int vlCreateVertexShader
 	assert(context);
 
 	pipe = csc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
+	tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
 
 	/* Version */
 	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
@@ -430,7 +430,7 @@ static int vlCreateVertexShader
 
 	vs.tokens = tokens;
 	csc->vertex_shader = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
+	FREE(tokens);
 
 	return 0;
 }
@@ -456,7 +456,7 @@ static int vlCreateFragmentShader
 	assert(context);
 
 	pipe = csc->pipe;
-	tokens = (struct tgsi_token*)malloc(max_tokens * sizeof(struct tgsi_token));
+	tokens = (struct tgsi_token*)MALLOC(max_tokens * sizeof(struct tgsi_token));
 
 	/* Version */
 	*(struct tgsi_version*)&tokens[0] = tgsi_build_version();
@@ -517,7 +517,7 @@ static int vlCreateFragmentShader
 
 	fs.tokens = tokens;
 	csc->fragment_shader = pipe->create_fs_state(pipe, &fs);
-	free(tokens);
+	FREE(tokens);
 
 	return 0;
 }
@@ -626,7 +626,7 @@ static int vlCreateDataBufs
 	memcpy
 	(
 		pipe->winsys->buffer_map(pipe->winsys, csc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		&bt_601,
+		&bt_601_full,
 		sizeof(struct vlFragmentShaderConsts)
 	);
 
@@ -691,7 +691,7 @@ int vlCreateBasicCSC
 	assert(pipe);
 	assert(csc);
 
-	basic_csc = calloc(1, sizeof(struct vlBasicCSC));
+	basic_csc = CALLOC_STRUCT(vlBasicCSC);
 
 	if (!basic_csc)
 		return 1;
diff --git a/src/gallium/state_trackers/g3dvl/vl_context.c b/src/gallium/state_trackers/g3dvl/vl_context.c
index fe107e406d..fbea1363d8 100644
--- a/src/gallium/state_trackers/g3dvl/vl_context.c
+++ b/src/gallium/state_trackers/g3dvl/vl_context.c
@@ -1,11 +1,10 @@
 #define VL_INTERNAL
 #include "vl_context.h"
 #include <assert.h>
-#include <stdlib.h>
 #include <pipe/p_context.h>
 #include <pipe/p_state.h>
+#include <util/u_memory.h>
 #include "vl_render.h"
-#include "vl_r16snorm_mc.h"
 #include "vl_r16snorm_mc_buf.h"
 #include "vl_csc.h"
 #include "vl_basic_csc.h"
@@ -112,7 +111,7 @@ int vlCreateContext
 	assert(context);
 	assert(pipe);
 
-	ctx = calloc(1, sizeof(struct vlContext));
+	ctx = CALLOC_STRUCT(vlContext);
 
 	if (!ctx)
 		return 1;
@@ -127,7 +126,6 @@ int vlCreateContext
 
 	vlInitCommon(ctx);
 
-	/*vlCreateR16SNormMC(pipe, picture_width, picture_height, picture_format, &ctx->render);*/
 	vlCreateR16SNormBufferedMC(pipe, picture_width, picture_height, picture_format, &ctx->render);
 	vlCreateBasicCSC(pipe, &ctx->csc);
 
@@ -154,7 +152,7 @@ int vlDestroyContext
 	context->pipe->delete_rasterizer_state(context->pipe, context->raster);
 	context->pipe->delete_depth_stencil_alpha_state(context->pipe, context->dsa);
 
-	free(context);
+	FREE(context);
 
 	return 0;
 }
diff --git a/src/gallium/state_trackers/g3dvl/vl_display.c b/src/gallium/state_trackers/g3dvl/vl_display.c
index af80faa7f5..dce06de758 100644
--- a/src/gallium/state_trackers/g3dvl/vl_display.c
+++ b/src/gallium/state_trackers/g3dvl/vl_display.c
@@ -1,7 +1,7 @@
 #define VL_INTERNAL
 #include "vl_display.h"
 #include <assert.h>
-#include <stdlib.h>
+#include <util/u_memory.h>
 
 int vlCreateDisplay
 (
@@ -14,7 +14,7 @@ int vlCreateDisplay
 	assert(native_display);
 	assert(display);
 
-	dpy = calloc(1, sizeof(struct vlDisplay));
+	dpy = CALLOC_STRUCT(vlDisplay);
 
 	if (!dpy)
 		return 1;
@@ -32,7 +32,7 @@ int vlDestroyDisplay
 {
 	assert(display);
 
-	free(display);
+	FREE(display);
 
 	return 0;
 }
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c
deleted file mode 100644
index 3272220ef8..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.c
+++ /dev/null
@@ -1,2344 +0,0 @@
-#define VL_INTERNAL
-#include "vl_r16snorm_mc.h"
-#include <assert.h>
-#include <stdlib.h>
-#include <pipe/p_context.h>
-#include <pipe/p_winsys.h>
-#include <pipe/p_state.h>
-#include <pipe/p_inlines.h>
-#include <tgsi/tgsi_parse.h>
-#include <tgsi/tgsi_build.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"
-
-#define NUM_BUFS 4	/* Number of rotating buffers to use */
-
-struct vlVertexShaderConsts
-{
-	/*struct vlVertex4f scale;
-	struct vlVertex4f denorm;*/
-	struct vlVertex4f	scale;
-	struct vlVertex4f	mb_pos_trans;
-	struct vlVertex4f	denorm;
-	struct
-	{
-		struct vlVertex4f	top_field;
-		struct vlVertex4f	bottom_field;
-	} mb_tc_trans[2];
-};
-
-struct vlFragmentShaderConsts
-{
-	struct vlVertex4f multiplier;
-	struct vlVertex4f div;
-};
-
-struct vlR16SnormMC
-{
-	struct vlRender				base;
-
-	unsigned int				video_width, video_height;
-	enum vlFormat				video_format;
-	unsigned int				cur_buf;
-
-	struct pipe_context			*pipe;
-	struct pipe_viewport_state		viewport;
-	struct pipe_framebuffer_state		render_target;
-	struct pipe_sampler_state		*samplers[5];
-	struct pipe_texture			*textures[NUM_BUFS][5];
-	void					*i_vs, *p_vs[2], *b_vs[2];
-	void					*i_fs, *p_fs[2], *b_fs[2];
-	struct pipe_vertex_buffer 		vertex_bufs[3];
-	struct pipe_vertex_element		vertex_elems[3];
-	struct pipe_constant_buffer		vs_const_buf, fs_const_buf;
-};
-
-static int vlBegin
-(
-	struct vlRender *render
-)
-{
-	struct vlR16SnormMC	*mc;
-	struct pipe_context	*pipe;
-
-	assert(render);
-
-	mc = (struct vlR16SnormMC*)render;
-	pipe = mc->pipe;
-
-	/* Frame buffer set in vlRender*Macroblock() */
-	/* Shaders, samplers, textures set in vlRender*Macroblock() */
-	pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs);
-	pipe->set_vertex_elements(pipe, 3, mc->vertex_elems);
-	pipe->set_viewport_state(pipe, &mc->viewport);
-	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);
-
-	return 0;
-}
-
-/*static int vlGrabMacroBlock
-(
-	struct vlR16SnormMC *mc,
-	struct vlMpeg2MacroBlock *macroblock
-)
-{
-	assert(mc);
-	assert(macroblock);
-
-
-
-	return 0;
-}*/
-
-/*#define DO_IDCT*/
-
-#ifdef DO_IDCT
-static int vlTransformBlock(short *src, short *dst, short bias)
-{
-	static const float basis[8][8] =
-	{
-		{0.3536,   0.4904,   0.4619,   0.4157,   0.3536,   0.2778,   0.1913,   0.0975},
-		{0.3536,   0.4157,   0.1913,  -0.0975,  -0.3536,  -0.4904,  -0.4619,  -0.2778},
-		{0.3536,   0.2778,  -0.1913,  -0.4904,  -0.3536,   0.0975,   0.4619,   0.4157},
-		{0.3536,   0.0975,  -0.4619,  -0.2778,   0.3536,   0.4157,  -0.1913,  -0.4904},
-		{0.3536,  -0.0975,  -0.4619,   0.2778,   0.3536,  -0.4157,  -0.1913,   0.4904},
-		{0.3536,  -0.2778,  -0.1913,   0.4904,  -0.3536,  -0.0975,   0.4619,  -0.4157},
-		{0.3536,  -0.4157,   0.1913,   0.0975,  -0.3536,   0.4904,  -0.4619,   0.2778},
-		{0.3536,  -0.4904,   0.4619,  -0.4157,   0.3536,  -0.2778,   0.1913,  -0.0975}
-	};
-
-	unsigned int	x, y;
-	short		tmp[64];
-
-	for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-			tmp[y * VL_BLOCK_WIDTH + x] = (short)
-			(
-				src[y * VL_BLOCK_WIDTH + 0] * basis[x][0] +
-				src[y * VL_BLOCK_WIDTH + 1] * basis[x][1] +
-				src[y * VL_BLOCK_WIDTH + 2] * basis[x][2] +
-				src[y * VL_BLOCK_WIDTH + 3] * basis[x][3] +
-				src[y * VL_BLOCK_WIDTH + 4] * basis[x][4] +
-				src[y * VL_BLOCK_WIDTH + 5] * basis[x][5] +
-				src[y * VL_BLOCK_WIDTH + 6] * basis[x][6] +
-				src[y * VL_BLOCK_WIDTH + 7] * basis[x][7]
-			);
-
-	for (x = 0; x < VL_BLOCK_WIDTH; ++x)
-		for (y = 0; y < VL_BLOCK_HEIGHT; ++y)
-		{
-			dst[y * VL_BLOCK_WIDTH + x] = bias + (short)
-			(
-				tmp[0 * VL_BLOCK_WIDTH + x] * basis[y][0] +
-				tmp[1 * VL_BLOCK_WIDTH + x] * basis[y][1] +
-				tmp[2 * VL_BLOCK_WIDTH + x] * basis[y][2] +
-				tmp[3 * VL_BLOCK_WIDTH + x] * basis[y][3] +
-				tmp[4 * VL_BLOCK_WIDTH + x] * basis[y][4] +
-				tmp[5 * VL_BLOCK_WIDTH + x] * basis[y][5] +
-				tmp[6 * VL_BLOCK_WIDTH + x] * basis[y][6] +
-				tmp[7 * VL_BLOCK_WIDTH + x] * basis[y][7]
-			);
-			if (dst[y * VL_BLOCK_WIDTH + x] > 255)
-				dst[y * VL_BLOCK_WIDTH + x] = 255;
-			else if (bias > 0 && dst[y * VL_BLOCK_WIDTH + x] < 0)
-				dst[y * VL_BLOCK_WIDTH + x] = 0;
-		}
-	return 0;
-}
-#endif
-
-static 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 int vlGrabFieldCodedBlock(short *src, short *dst, unsigned int dst_pitch)
-{
-	unsigned int y;
-
-	for (y = 0; y < VL_BLOCK_HEIGHT / 2; ++y)
-		memcpy
-		(
-			dst + y * dst_pitch * 2,
-			src + y * VL_BLOCK_WIDTH,
-			VL_BLOCK_WIDTH * 2
-		);
-
-	dst += VL_BLOCK_HEIGHT * dst_pitch;
-
-	for (; y < VL_BLOCK_HEIGHT; ++y)
-		memcpy
-		(
-			dst + y * dst_pitch * 2,
-			src + y * VL_BLOCK_WIDTH,
-			VL_BLOCK_WIDTH * 2
-		);
-
-	return 0;
-}
-
-static 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;
-}
-
-enum vlSampleType
-{
-	vlSampleTypeFull,
-	vlSampleTypeDiff
-};
-
-static int vlGrabBlocks
-(
-	struct vlR16SnormMC *mc,
-	unsigned int coded_block_pattern,
-	enum vlDCTType dct_type,
-	enum vlSampleType sample_type,
-	short *blocks
-)
-{
-	struct pipe_surface	*tex_surface;
-	short			*texels;
-	unsigned int		tex_pitch;
-	unsigned int		tb, sb = 0;
-
-	assert(mc);
-	assert(blocks);
-
-	tex_surface = mc->pipe->screen->get_tex_surface
-	(
-		mc->pipe->screen,
-		mc->textures[mc->cur_buf % NUM_BUFS][0],
-		0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
-	);
-
-	texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE);
-	tex_pitch = tex_surface->stride / tex_surface->block.size;
-
-	for (tb = 0; tb < 4; ++tb)
-	{
-		if ((coded_block_pattern >> (5 - tb)) & 1)
-		{
-			short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
-
-#ifdef DO_IDCT
-			vlTransformBlock(cur_block, cur_block, sample_type == vlSampleTypeFull ? 128 : 0);
-#endif
-
-			if (dct_type == vlDCTTypeFrameCoded)
-				vlGrabFrameCodedBlock
-				(
-					cur_block,
-					texels + tb * tex_pitch * VL_BLOCK_HEIGHT,
-					tex_pitch
-				);
-			else
-				vlGrabFieldCodedBlock
-				(
-					cur_block,
-					texels + (tb % 2) * tex_pitch * VL_BLOCK_HEIGHT + (tb / 2) * tex_pitch,
-					tex_pitch
-				);
-
-			++sb;
-		}
-		else
-			vlGrabNoBlock(texels + tb * tex_pitch * VL_BLOCK_HEIGHT, tex_pitch);
-	}
-
-	pipe_surface_unmap(tex_surface);
-
-	/* TODO: Implement 422, 444 */
-	for (tb = 0; tb < 2; ++tb)
-	{
-		tex_surface = mc->pipe->screen->get_tex_surface
-		(
-			mc->pipe->screen,
-			mc->textures[mc->cur_buf % NUM_BUFS][tb + 1],
-			0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
-		);
-
-		texels = pipe_surface_map(tex_surface, PIPE_BUFFER_USAGE_CPU_WRITE);
-		tex_pitch = tex_surface->stride / tex_surface->block.size;
-
-		if ((coded_block_pattern >> (1 - tb)) & 1)
-		{
-			short *cur_block = blocks + sb * VL_BLOCK_WIDTH * VL_BLOCK_HEIGHT;
-
-#ifdef DO_IDCT
-			vlTransformBlock(cur_block, cur_block, sample_type == vlSampleTypeFull ? 128 : 0);
-#endif
-
-			vlGrabFrameCodedBlock
-			(
-				cur_block,
-				texels,
-				tex_pitch
-			);
-
-			++sb;
-		}
-		else
-			vlGrabNoBlock(texels, tex_pitch);
-
-		pipe_surface_unmap(tex_surface);
-	}
-
-	return 0;
-}
-
-static int vlRenderIMacroBlock
-(
-	struct vlR16SnormMC *mc,
-	enum vlPictureType picture_type,
-	enum vlFieldOrder field_order,
-	unsigned int mbx,
-	unsigned int mby,
-	unsigned int coded_block_pattern,
-	enum vlDCTType dct_type,
-	short *blocks,
-	struct vlSurface *surface
-)
-{
-	struct pipe_context		*pipe;
-	struct vlVertexShaderConsts	*vs_consts;
-
-	assert(blocks);
-	assert(surface);
-
-	/* TODO: Implement interlaced rendering */
-	if (picture_type != vlPictureTypeFrame)
-		return 0;
-
-	vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeFull, blocks);
-
-	pipe = mc->pipe;
-
-	vs_consts = pipe->winsys->buffer_map
-	(
-		pipe->winsys,
-		mc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE
-	);
-
-	vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0];
-	vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0];
-	vs_consts->scale.z = 1.0f;
-	vs_consts->scale.w = 1.0f;
-	vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0];
-	vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0];
-	vs_consts->mb_pos_trans.z = 0.0f;
-	vs_consts->mb_pos_trans.w = 0.0f;
-
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer);
-
-	mc->render_target.cbufs[0] = pipe->screen->get_tex_surface
-	(
-		pipe->screen,
-		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_sampler_textures(pipe, 3, mc->textures[mc->cur_buf % NUM_BUFS]);
-	pipe->bind_sampler_states(pipe, 3, (void**)mc->samplers);
-	pipe->bind_vs_state(pipe, mc->i_vs);
-	pipe->bind_fs_state(pipe, mc->i_fs);
-
-	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
-
-	mc->cur_buf++;
-
-	return 0;
-}
-
-static int vlRenderPMacroBlock
-(
-	struct vlR16SnormMC *mc,
-	enum vlPictureType picture_type,
-	enum vlFieldOrder field_order,
-	unsigned int mbx,
-	unsigned int mby,
-	enum vlMotionType mc_type,
-	short top_x,
-	short top_y,
-	short bottom_x,
-	short bottom_y,
-	unsigned int coded_block_pattern,
-	enum vlDCTType dct_type,
-	short *blocks,
-	struct vlSurface *ref_surface,
-	struct vlSurface *surface
-)
-{
-	struct pipe_context		*pipe;
-	struct vlVertexShaderConsts	*vs_consts;
-
-	assert(motion_vectors);
-	assert(blocks);
-	assert(ref_surface);
-	assert(surface);
-
-	/* TODO: Implement interlaced rendering */
-	if (picture_type != vlPictureTypeFrame)
-		return 0;
-	/* TODO: Implement other MC types */
-	if (mc_type != vlMotionTypeFrame && mc_type != vlMotionTypeField)
-		return 0;
-
-	vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeDiff, blocks);
-
-	pipe = mc->pipe;
-
-	vs_consts = pipe->winsys->buffer_map
-	(
-		pipe->winsys,
-		mc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE
-	);
-
-	vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0];
-	vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0];
-	vs_consts->scale.z = 1.0f;
-	vs_consts->scale.w = 1.0f;
-	vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0];
-	vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0];
-	vs_consts->mb_pos_trans.z = 0.0f;
-	vs_consts->mb_pos_trans.w = 0.0f;
-	vs_consts->mb_tc_trans[0].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_x * 0.5f) / (float)surface->texture->width[0];
-	vs_consts->mb_tc_trans[0].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_y * 0.5f) / (float)surface->texture->height[0];
-	vs_consts->mb_tc_trans[0].top_field.z = 0.0f;
-	vs_consts->mb_tc_trans[0].top_field.w = 0.0f;
-
-	if (mc_type == vlMotionTypeField)
-	{
-		vs_consts->denorm.x = (float)surface->texture->width[0];
-		vs_consts->denorm.y = (float)surface->texture->height[0];
-
-		vs_consts->mb_tc_trans[0].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_x * 0.5f) / (float)surface->texture->width[0];
-		vs_consts->mb_tc_trans[0].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_y * 0.5f) / (float)surface->texture->height[0];
-		vs_consts->mb_tc_trans[0].bottom_field.z = 0.0f;
-		vs_consts->mb_tc_trans[0].bottom_field.w = 0.0f;
-
-		pipe->bind_vs_state(pipe, mc->p_vs[1]);
-		pipe->bind_fs_state(pipe, mc->p_fs[1]);
-	}
-	else
-	{
-		pipe->bind_vs_state(pipe, mc->p_vs[0]);
-		pipe->bind_fs_state(pipe, mc->p_fs[0]);
-	}
-
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer);
-
-	mc->render_target.cbufs[0] = pipe->screen->get_tex_surface
-	(
-		pipe->screen,
-		surface->texture,
-		0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
-	);
-	pipe->set_framebuffer_state(pipe, &mc->render_target);
-
-	mc->textures[mc->cur_buf % NUM_BUFS][3] = ref_surface->texture;
-	pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUFS]);
-	pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
-
-	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
-
-	mc->cur_buf++;
-
-	return 0;
-}
-
-static int vlRenderBMacroBlock
-(
-	struct vlR16SnormMC *mc,
-	enum vlPictureType picture_type,
-	enum vlFieldOrder field_order,
-	unsigned int mbx,
-	unsigned int mby,
-	enum vlMotionType mc_type,
-	short top_past_x,
-	short top_past_y,
-	short bottom_past_x,
-	short bottom_past_y,
-	short top_future_x,
-	short top_future_y,
-	short bottom_future_x,
-	short bottom_future_y,
-	unsigned int coded_block_pattern,
-	enum vlDCTType dct_type,
-	short *blocks,
-	struct vlSurface *past_surface,
-	struct vlSurface *future_surface,
-	struct vlSurface *surface
-)
-{
-	struct pipe_context		*pipe;
-	struct vlVertexShaderConsts	*vs_consts;
-
-	assert(motion_vectors);
-	assert(blocks);
-	assert(ref_surface);
-	assert(surface);
-
-	/* TODO: Implement interlaced rendering */
-	if (picture_type != vlPictureTypeFrame)
-		return 0;
-	/* TODO: Implement other MC types */
-	if (mc_type != vlMotionTypeFrame && mc_type != vlMotionTypeField)
-		return 0;
-
-	vlGrabBlocks(mc, coded_block_pattern, dct_type, vlSampleTypeDiff, blocks);
-
-	pipe = mc->pipe;
-
-	vs_consts = pipe->winsys->buffer_map
-	(
-		pipe->winsys,
-		mc->vs_const_buf.buffer,
-		PIPE_BUFFER_USAGE_CPU_WRITE
-	);
-
-	vs_consts->scale.x = VL_MACROBLOCK_WIDTH / (float)surface->texture->width[0];
-	vs_consts->scale.y = VL_MACROBLOCK_HEIGHT / (float)surface->texture->height[0];
-	vs_consts->scale.z = 1.0f;
-	vs_consts->scale.w = 1.0f;
-	vs_consts->mb_pos_trans.x = (mbx * VL_MACROBLOCK_WIDTH) / (float)surface->texture->width[0];
-	vs_consts->mb_pos_trans.y = (mby * VL_MACROBLOCK_HEIGHT) / (float)surface->texture->height[0];
-	vs_consts->mb_pos_trans.z = 0.0f;
-	vs_consts->mb_pos_trans.w = 0.0f;
-	vs_consts->mb_tc_trans[0].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_past_x * 0.5f) / (float)surface->texture->width[0];
-	vs_consts->mb_tc_trans[0].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_past_y * 0.5f) / (float)surface->texture->height[0];
-	vs_consts->mb_tc_trans[0].top_field.z = 0.0f;
-	vs_consts->mb_tc_trans[0].top_field.w = 0.0f;
-	vs_consts->mb_tc_trans[1].top_field.x = (mbx * VL_MACROBLOCK_WIDTH + top_future_x * 0.5f) / (float)surface->texture->width[0];
-	vs_consts->mb_tc_trans[1].top_field.y = (mby * VL_MACROBLOCK_HEIGHT + top_future_y * 0.5f) / (float)surface->texture->height[0];
-	vs_consts->mb_tc_trans[1].top_field.z = 0.0f;
-	vs_consts->mb_tc_trans[1].top_field.w = 0.0f;
-
-	if (mc_type == vlMotionTypeField)
-	{
-		vs_consts->denorm.x = (float)surface->texture->width[0];
-		vs_consts->denorm.y = (float)surface->texture->height[0];
-
-		vs_consts->mb_tc_trans[0].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_past_x * 0.5f) / (float)surface->texture->width[0];
-		vs_consts->mb_tc_trans[0].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_past_y * 0.5f) / (float)surface->texture->height[0];
-		vs_consts->mb_tc_trans[0].bottom_field.z = 0.0f;
-		vs_consts->mb_tc_trans[0].bottom_field.w = 0.0f;
-		vs_consts->mb_tc_trans[1].bottom_field.x = (mbx * VL_MACROBLOCK_WIDTH + bottom_future_x * 0.5f) / (float)surface->texture->width[0];
-		vs_consts->mb_tc_trans[1].bottom_field.y = (mby * VL_MACROBLOCK_HEIGHT + bottom_future_y * 0.5f) / (float)surface->texture->height[0];
-		vs_consts->mb_tc_trans[1].bottom_field.z = 0.0f;
-		vs_consts->mb_tc_trans[1].bottom_field.w = 0.0f;
-
-		pipe->bind_vs_state(pipe, mc->b_vs[1]);
-		pipe->bind_fs_state(pipe, mc->b_fs[1]);
-	}
-	else
-	{
-		pipe->bind_vs_state(pipe, mc->b_vs[0]);
-		pipe->bind_fs_state(pipe, mc->b_fs[0]);
-	}
-
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->vs_const_buf.buffer);
-
-	mc->render_target.cbufs[0] = pipe->screen->get_tex_surface
-	(
-		pipe->screen,
-		surface->texture,
-		0, 0, 0, PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE
-	);
-	pipe->set_framebuffer_state(pipe, &mc->render_target);
-
-	mc->textures[mc->cur_buf % NUM_BUFS][3] = past_surface->texture;
-	mc->textures[mc->cur_buf % NUM_BUFS][4] = future_surface->texture;
-	pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUFS]);
-	pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers);
-
-	pipe->draw_arrays(pipe, PIPE_PRIM_TRIANGLES, 0, 24);
-
-	mc->cur_buf++;
-
-	return 0;
-}
-
-static int vlRenderMacroBlocksMpeg2R16Snorm
-(
-	struct vlRender *render,
-	struct vlMpeg2MacroBlockBatch *batch,
-	struct vlSurface *surface
-)
-{
-	struct vlR16SnormMC	*mc;
-	unsigned int		i;
-
-	assert(render);
-
-	mc = (struct vlR16SnormMC*)render;
-
-	/*for (i = 0; i < batch->num_macroblocks; ++i)
-		vlGrabMacroBlock(batch->macroblocks[i]);*/
-
-	for (i = 0; i < batch->num_macroblocks; ++i)
-	{
-		switch (batch->macroblocks[i].mb_type)
-		{
-			case vlMacroBlockTypeIntra:
-			{
-				vlRenderIMacroBlock
-				(
-					mc,
-					batch->picture_type,
-					batch->field_order,
-					batch->macroblocks[i].mbx,
-					batch->macroblocks[i].mby,
-					batch->macroblocks[i].cbp,
-					batch->macroblocks[i].dct_type,
-					batch->macroblocks[i].blocks,
-					surface
-				);
-				break;
-			}
-			case vlMacroBlockTypeFwdPredicted:
-			{
-				vlRenderPMacroBlock
-				(
-					mc,
-					batch->picture_type,
-					batch->field_order,
-					batch->macroblocks[i].mbx,
-					batch->macroblocks[i].mby,
-					batch->macroblocks[i].mo_type,
-					batch->macroblocks[i].PMV[0][0][0],
-					batch->macroblocks[i].PMV[0][0][1],
-					batch->macroblocks[i].PMV[1][0][0],
-					batch->macroblocks[i].PMV[1][0][1],
-					batch->macroblocks[i].cbp,
-					batch->macroblocks[i].dct_type,
-					batch->macroblocks[i].blocks,
-					batch->past_surface,
-					surface
-				);
-				break;
-			}
-			case vlMacroBlockTypeBkwdPredicted:
-			{
-				vlRenderPMacroBlock
-				(
-					mc,
-					batch->picture_type,
-					batch->field_order,
-					batch->macroblocks[i].mbx,
-					batch->macroblocks[i].mby,
-					batch->macroblocks[i].mo_type,
-					batch->macroblocks[i].PMV[0][1][0],
-					batch->macroblocks[i].PMV[0][1][1],
-					batch->macroblocks[i].PMV[1][1][0],
-					batch->macroblocks[i].PMV[1][1][1],
-					batch->macroblocks[i].cbp,
-					batch->macroblocks[i].dct_type,
-					batch->macroblocks[i].blocks,
-					batch->future_surface,
-					surface
-				);
-				break;
-			}
-			case vlMacroBlockTypeBiPredicted:
-			{
-				vlRenderBMacroBlock
-				(
-					mc,
-					batch->picture_type,
-					batch->field_order,
-					batch->macroblocks[i].mbx,
-					batch->macroblocks[i].mby,
-					batch->macroblocks[i].mo_type,
-					batch->macroblocks[i].PMV[0][0][0],
-					batch->macroblocks[i].PMV[0][0][1],
-					batch->macroblocks[i].PMV[1][0][0],
-					batch->macroblocks[i].PMV[1][0][1],
-					batch->macroblocks[i].PMV[0][1][0],
-					batch->macroblocks[i].PMV[0][1][1],
-					batch->macroblocks[i].PMV[1][1][0],
-					batch->macroblocks[i].PMV[1][1][1],
-					batch->macroblocks[i].cbp,
-					batch->macroblocks[i].dct_type,
-					batch->macroblocks[i].blocks,
-					batch->past_surface,
-					batch->future_surface,
-					surface
-				);
-				break;
-			}
-			default:
-				assert(0);
-		}
-	}
-
-	return 0;
-}
-
-static int vlEnd
-(
-	struct vlRender *render
-)
-{
-	assert(render);
-
-	return 0;
-}
-
-static int vlFlush
-(
-	struct vlRender *render
-)
-{
-	assert(render);
-
-	return 0;
-}
-
-static int vlDestroy
-(
-	struct vlRender *render
-)
-{
-	struct vlR16SnormMC	*mc;
-	struct pipe_context	*pipe;
-	unsigned int		i;
-
-	assert(render);
-
-	mc = (struct vlR16SnormMC*)render;
-	pipe = mc->pipe;
-
-	for (i = 0; i < 5; ++i)
-		pipe->delete_sampler_state(pipe, mc->samplers[i]);
-
-	for (i = 0; i < 3; ++i)
-		pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs[i].buffer);
-
-	/* Textures 3 & 4 are not created directly, no need to release them here */
-	for (i = 0; i < NUM_BUFS; ++i)
-	{
-		pipe_texture_release(&mc->textures[i][0]);
-		pipe_texture_release(&mc->textures[i][1]);
-		pipe_texture_release(&mc->textures[i][2]);
-	}
-
-	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->winsys->buffer_destroy(pipe->winsys, mc->vs_const_buf.buffer);
-	pipe->winsys->buffer_destroy(pipe->winsys, mc->fs_const_buf.buffer);
-
-	free(mc);
-
-	return 0;
-}
-
-/*
- * Represents 8 triangles (4 quads, 1 per block) in noormalized coords
- * that render a macroblock.
- * Need to be scaled to cover mbW*mbH macroblock pixels and translated into
- * position on target surface.
- */
-static const struct vlVertex2f macroblock_verts[24] =
-{
-	{0.0f, 0.0f}, {0.0f, 0.5f}, {0.5f, 0.0f},
-	{0.5f, 0.0f}, {0.0f, 0.5f}, {0.5f, 0.5f},
-
-	{0.5f, 0.0f}, {0.5f, 0.5f}, {1.0f, 0.0f},
-	{1.0f, 0.0f}, {0.5f, 0.5f}, {1.0f, 0.5f},
-
-	{0.0f, 0.5f}, {0.0f, 1.0f}, {0.5f, 0.5f},
-	{0.5f, 0.5f}, {0.0f, 1.0f}, {0.5f, 1.0f},
-
-	{0.5f, 0.5f}, {0.5f, 1.0f}, {1.0f, 0.5f},
-	{1.0f, 0.5f}, {0.5f, 1.0f}, {1.0f, 1.0f}
-};
-
-/*
- * Represents texcoords for the above for rendering 4 luma blocks arranged
- * in a bW*(bH*4) texture. First luma block located at 0,0->bW,bH; second at
- * 0,bH->bW,2bH; third at 0,2bH->bW,3bH; fourth at 0,3bH->bW,4bH.
- */
-static const struct vlVertex2f macroblock_luma_texcoords[24] =
-{
-	{0.0f, 0.0f}, {0.0f, 0.25f}, {1.0f, 0.0f},
-	{1.0f, 0.0f}, {0.0f, 0.25f}, {1.0f, 0.25f},
-
-	{0.0f, 0.25f}, {0.0f, 0.5f}, {1.0f, 0.25f},
-	{1.0f, 0.25f}, {0.0f, 0.5f}, {1.0f, 0.5f},
-
-	{0.0f, 0.5f}, {0.0f, 0.75f}, {1.0f, 0.5f},
-	{1.0f, 0.5f}, {0.0f, 0.75f}, {1.0f, 0.75f},
-
-	{0.0f, 0.75f}, {0.0f, 1.0f}, {1.0f, 0.75f},
-	{1.0f, 0.75f}, {0.0f, 1.0f}, {1.0f, 1.0f}
-};
-
-/*
- * Represents texcoords for the above for rendering 1 chroma block.
- * Straight forward 0,0->1,1 mapping so we can reuse the MB pos vectors.
- */
-static const struct vlVertex2f *macroblock_chroma_420_texcoords = macroblock_verts;
-
-/*
- * Represents texcoords for the above for rendering 2 chroma blocks arranged
- * in a bW*(bH*2) texture. First chroma block located at 0,0->bW,bH; second at
- * 0,bH->bW,2bH. We can render this with 0,0->1,1 mapping.
- * Straight forward 0,0->1,1 mapping so we can reuse MB pos vectors.
- */
-static const struct vlVertex2f *macroblock_chroma_422_texcoords = macroblock_verts;
-
-/*
- * Represents texcoords for the above for rendering 4 chroma blocks.
- * Same case as 4 luma blocks.
- */
-static const struct vlVertex2f *macroblock_chroma_444_texcoords = macroblock_luma_texcoords;
-
-/*
- * Used when rendering P and B macroblocks, multiplier is applied to the A channel,
- * which is then added to the L channel, then the bias is subtracted from that to
- * get back the differential. The differential is then added to the samples from the
- * reference surface(s).
- */
-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}
-};
-
-static int vlCreateVertexShaderIMB
-(
-	struct vlR16SnormMC *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 texcoords
-	 */
-	for (i = 0; i < 3; 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 unit rect to macroblock size
-	 * decl c1		; Translation vector to move macroblock into position
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma texcoords
-	 */
-	for (i = 0; i < 3; 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 */
-	decl = vl_decl_temps(0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t0, i0, c0	; Scale unit rect to normalized MB size */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, c1	; Translate rect into position */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma texcoords to output
-	 */
-	for (i = 1; i < 3; ++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 vlR16SnormMC *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			; Texcoords for s0
-	 * decl i1			; Texcoords for s1, s2
-	 */
-	for (i = 0; i < 2; ++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, i1, 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 > 0 ? 1 : 0, 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 vlR16SnormMC *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 texcoords
-	 */
-	for (i = 0; i < 3; 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 unit rect to macroblock size
-	 * decl c1		; Translation vector to move macroblock into position
-	 * decl c2		; Unused
-	 * decl c3		; Translation vector to move ref macroblock texcoords 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		; Luma texcoords
-	 * decl o2		; Chroma texcoords
-	 * decl o3		; Ref macroblock 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);
-	}
-
-	/* decl t0 */
-	decl = vl_decl_temps(0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t0, i0, c0	; Scale unit rect to normalized MB size */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, c1	; Translate rect into position */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma texcoords to output
-	 */
-	for (i = 1; i < 3; ++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 o3, t0, c3	; Translate rect into position on ref macroblock */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 3);
-	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 vlR16SnormMC *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 texcoords
-	 */
-	for (i = 0; i < 3; 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 unit rect to macroblock size
-	 * decl c1		; Translation vector to move macroblock into position
-	 * decl c2		; Denorm coefficients
-	 * decl c3		; Translation vector to move top field ref macroblock texcoords into position
-	 * decl c4		; Translation vector to move bottom field ref macroblock texcoords into position
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 4);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma texcoords
-	 * decl o3		; Top field ref macroblock texcoords
-	 * decl o4		; Bottom field ref macroblock texcoords
-	 * decl o5		; Denormalized vertex pos
-	 */
-	for (i = 0; i < 6; i++)
-	{
-		decl = vl_decl_output((i == 0 || i == 5) ? 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);
-
-	/* mul t0, i0, c0	; Scale unit rect to normalized MB size */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add t1, t0, c1	; Translate rect into position */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mov o0, t1		; Move vertex pos to output */
-	inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	mov o1, i1		; Move input luma texcoords to output
-	mov o2, i2		; Move input chroma texcoords to output
-	*/
-	for (i = 1; i < 3; ++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 o3, t0, c3	; Translate top field rect into position on ref macroblock
-	   add o4, t0, c4	; Translate bottom field rect into position on ref macroblock */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 3);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul o5, t1, c2	; Denorm vertex pos */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 5, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 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->p_vs[1] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFramePMB
-(
-	struct vlR16SnormMC *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			; Texcoords for s0
-	 * decl i1			; Texcoords for s1, s2
-	 * decl i2			; Texcoords for s3
-	 */
-	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
-	 * 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, i1, 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 > 0 ? 1 : 0, 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, i2, s3		; Read texel from ref macroblock */
-	inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 2, 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 vlR16SnormMC *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			; Texcoords for s0
-	 * decl i1			; Texcoords for s1, s2
-	 * decl i2			; Texcoords for s3
-	 * decl i3			; Texcoords for s3
-	 * decl i4			; Denormalized vertex pos
-	 */
-	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			; 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, i1, 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 > 0 ? 1 : 0, 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, i2, s3		; Read texel from ref macroblock top field
-	 * tex2d t2, i3, 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 + 2, 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, i4.y, c1.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 4, 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_FLOOR, 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_LERP, 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 vlR16SnormMC *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 texcoords
-	 */
-	for (i = 0; i < 3; 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 unit rect to macroblock size
-	 * decl c1		; Translation vector to move macroblock into position
-	 * decl c2		; Unused
-	 * decl c3		; Translation vector to move past ref macroblock texcoords into position
-	 * decl c4		; Unused
-	 * decl c5		; Translation vector to move future ref macroblock texcoords into position
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 5);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma texcoords
-	 * decl o3		; Past ref macroblock texcoords
-	 * decl o4		; Future 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);
-	}
-
-	/* decl t0 */
-	decl = vl_decl_temps(0, 0);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/* mul t0, i0, c0	; Scale unit rect to normalized MB size */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add o0, t0, c1	; Translate rect into position */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma texcoords to output
-	 */
-	for (i = 1; i < 3; ++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 o3, t0, c3	; Translate rect into position on past ref macroblock
-	   add o4, t0, c5	; Translate rect into position on future ref macroblock */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i * 2 + 3);
-		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 vlR16SnormMC *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 texcoords
-	 */
-	for (i = 0; i < 3; 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 unit rect to macroblock size
-	 * decl c1		; Translation vector to move macroblock into position
-	 * decl c2		; Denorm coefficients
-	 * decl c3		; Translation vector to move top field past ref macroblock texcoords into position
-	 * decl c4		; Translation vector to move bottom field past ref macroblock texcoords into position
-	 * decl c5		; Translation vector to move top field future ref macroblock texcoords into position
-	 * decl c6		; Translation vector to move bottom field future ref macroblock texcoords into position
-	 */
-	decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 6);
-	ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * decl o0		; Vertex pos
-	 * decl o1		; Luma texcoords
-	 * decl o2		; Chroma texcoords
-	 * decl o3		; Top field past ref macroblock texcoords
-	 * decl o4		; Bottom field past ref macroblock texcoords
-	 * decl o5		; Top field future ref macroblock texcoords
-	 * decl o6		; Bottom field future ref macroblock texcoords
-	 * decl o7		; Denormalized vertex pos
-	 */
-	for (i = 0; i < 8; i++)
-	{
-		decl = vl_decl_output((i == 0 || i == 7) ? 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);
-
-	/* mul t0, i0, c0	; Scale unit rect to normalized MB size */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_INPUT, 0, TGSI_FILE_CONSTANT, 0);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* add t1, t0, c1	; Translate rect into position */
-	inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/* mov o0, t1		; Move vertex pos to output */
-	inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 1);
-	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-
-	/*
-	 * mov o1, i1		; Move input luma texcoords to output
-	 * mov o2, i2		; Move input chroma texcoords to output
-	 */
-	for (i = 1; i < 3; ++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 o3, t0, c3	; Translate top field rect into position on past ref macroblock
-	 * add o4, t0, c4	; Translate bottom field rect into position on past ref macroblock
-	 * add o5, t0, c5	; Translate top field rect into position on future ref macroblock
-	 * add o6, t0, c6	; Translate bottom field rect into position on future ref macroblock
-	 */
-	for (i = 0; i < 4; ++i)
-	{
-		inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 3, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, i + 3);
-		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
-	}
-
-	/* mul o7, t1, c2	; Denorm vertex pos */
-	inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 7, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 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[1] = pipe->create_vs_state(pipe, &vs);
-	free(tokens);
-
-	return 0;
-}
-
-static int vlCreateFragmentShaderFrameBMB
-(
-	struct vlR16SnormMC *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			; Texcoords for s0
-	 * decl i1			; Texcoords for s1, s2
-	 * decl i2			; Texcoords for s3
-	 * decl i3			; Texcoords for s4
-	 */
-	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 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 past ref surface texture
-	 * decl s4			; Sampler for future 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, i1, 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 > 0 ? 1 : 0, 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, i2, s3		; Read texel from past ref macroblock
-	 * tex2d t2, i3, s4		; Read texel from future ref macroblock
-	 */
-	for (i = 0; i < 2; ++i)
-	{
-		inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 2, 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_LERP, 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 vlR16SnormMC *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			; Texcoords for s0
-	 * decl i1			; Texcoords for s1, s2
-	 * decl i2			; Texcoords for s3
-	 * decl i3			; Texcoords for s3
-	 * decl i4			; Texcoords for s4
-	 * decl i5			; Texcoords for s4
-	 * decl i6			; Denormalized vertex pos
-	 */
-	for (i = 0; i < 7; ++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 past ref surface texture
-	 * decl s4			; Sampler for future 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, i1, 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 > 0 ? 1 : 0, 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, i6.y, c1.x		; Sub 0.5 from denormalized pos */
-	inst = vl_inst3(TGSI_OPCODE_SUB, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_INPUT, 6, 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_FLOOR, 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, i2, s3		; Read texel from past ref macroblock top field
-	 * tex2d t2, i3, 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 + 2, 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_LERP, 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, i4, s4		; Read texel from future ref macroblock top field
-	 * tex2d t5, i5, 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 + 4, 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_LERP, 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_LERP, 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;
-}
-
-static int vlCreateDataBufs
-(
-	struct vlR16SnormMC *mc
-)
-{
-	struct pipe_context	*pipe;
-	unsigned int		i;
-
-	assert(mc);
-
-	pipe = mc->pipe;
-
-	/* Create our vertex buffer and vertex buffer element */
-	mc->vertex_bufs[0].pitch = sizeof(struct vlVertex2f);
-	mc->vertex_bufs[0].max_index = 23;
-	mc->vertex_bufs[0].buffer_offset = 0;
-	mc->vertex_bufs[0].buffer = pipe->winsys->buffer_create
-	(
-		pipe->winsys,
-		1,
-		PIPE_BUFFER_USAGE_VERTEX,
-		sizeof(struct vlVertex2f) * 24
-	);
-
-	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;
-
-	/* Create our texcoord buffers and texcoord buffer elements */
-	for (i = 1; i < 3; ++i)
-	{
-		mc->vertex_bufs[i].pitch = sizeof(struct vlVertex2f);
-		mc->vertex_bufs[i].max_index = 23;
-		mc->vertex_bufs[i].buffer_offset = 0;
-		mc->vertex_bufs[i].buffer = pipe->winsys->buffer_create
-		(
-			pipe->winsys,
-			1,
-			PIPE_BUFFER_USAGE_VERTEX,
-			sizeof(struct vlVertex2f) * 24
-		);
-
-		mc->vertex_elems[i].src_offset = 0;
-		mc->vertex_elems[i].vertex_buffer_index = i;
-		mc->vertex_elems[i].nr_components = 2;
-		mc->vertex_elems[i].src_format = PIPE_FORMAT_R32G32_FLOAT;
-	}
-
-	/* Fill buffers */
-	memcpy
-	(
-		pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[0].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		macroblock_verts,
-		sizeof(struct vlVertex2f) * 24
-	);
-	memcpy
-	(
-		pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[1].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		macroblock_luma_texcoords,
-		sizeof(struct vlVertex2f) * 24
-	);
-	/* TODO: Accomodate 422, 444 */
-	memcpy
-	(
-		pipe->winsys->buffer_map(pipe->winsys, mc->vertex_bufs[2].buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		macroblock_chroma_420_texcoords,
-		sizeof(struct vlVertex2f) * 24
-	);
-
-	for (i = 0; i < 3; ++i)
-		pipe->winsys->buffer_unmap(pipe->winsys, mc->vertex_bufs[i].buffer);
-
-	/* Create our constant buffer */
-	mc->vs_const_buf.size = sizeof(struct vlVertexShaderConsts);
-	mc->vs_const_buf.buffer = pipe->winsys->buffer_create
-	(
-		pipe->winsys,
-		1,
-		PIPE_BUFFER_USAGE_CONSTANT,
-		mc->vs_const_buf.size
-	);
-
-	mc->fs_const_buf.size = sizeof(struct vlFragmentShaderConsts);
-	mc->fs_const_buf.buffer = pipe->winsys->buffer_create
-	(
-		pipe->winsys,
-		1,
-		PIPE_BUFFER_USAGE_CONSTANT,
-		mc->fs_const_buf.size
-	);
-
-	memcpy
-	(
-		pipe->winsys->buffer_map(pipe->winsys, mc->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
-		&fs_consts,
-		sizeof(struct vlFragmentShaderConsts)
-	);
-
-	pipe->winsys->buffer_unmap(pipe->winsys, mc->fs_const_buf.buffer);
-
-	return 0;
-}
-
-static int vlInit
-(
-	struct vlR16SnormMC *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;
-
-	/* For MC we render to textures, which are rounded up to nearest POT */
-	mc->viewport.scale[0] = vlRoundUpPOT(mc->video_width);
-	mc->viewport.scale[1] = vlRoundUpPOT(mc->video_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->video_width);
-	mc->render_target.height = vlRoundUpPOT(mc->video_height);
-	mc->render_target.num_cbufs = 1;
-	/* FB for MC stage is a vlSurface, set in vlSetRenderSurface() */
-	mc->render_target.zsbuf = NULL;
-
-	filters[0] = PIPE_TEX_FILTER_NEAREST;
-	filters[1] = mc->video_format == vlFormatYCbCr444 ? PIPE_TEX_FILTER_NEAREST : PIPE_TEX_FILTER_LINEAR;
-	filters[2] = mc->video_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.shadow_ambient = ;*/
-		/*sampler.lod_bias = ;*/
-		sampler.min_lod = 0;
-		/*sampler.max_lod = ;*/
-		/*sampler.border_color[i] = ;*/
-		/*sampler.max_anisotropy = ;*/
-		mc->samplers[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] = 8;
-	template.height[0] = 8 * 4;
-	template.depth[0] = 1;
-	template.compressed = 0;
-	pf_get_block(template.format, &template.block);
-
-	for (i = 0; i < NUM_BUFS; ++i)
-		mc->textures[i][0] = pipe->screen->texture_create(pipe->screen, &template);
-
-	if (mc->video_format == vlFormatYCbCr420)
-		template.height[0] = 8;
-	else if (mc->video_format == vlFormatYCbCr422)
-		template.height[0] = 8 * 2;
-	else if (mc->video_format == vlFormatYCbCr444)
-		template.height[0] = 8 * 4;
-	else
-		assert(0);
-
-	for (i = 0; i < NUM_BUFS; ++i)
-	{
-		mc->textures[i][1] = pipe->screen->texture_create(pipe->screen, &template);
-		mc->textures[i][2] = pipe->screen->texture_create(pipe->screen, &template);
-	}
-
-	/* textures[3] & textures[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 vlCreateR16SNormMC
-(
-	struct pipe_context *pipe,
-	unsigned int video_width,
-	unsigned int video_height,
-	enum vlFormat video_format,
-	struct vlRender **render
-)
-{
-	struct vlR16SnormMC *mc;
-
-	assert(pipe);
-	assert(render);
-
-	mc = calloc(1, sizeof(struct vlR16SnormMC));
-
-	mc->base.vlBegin = &vlBegin;
-	mc->base.vlRenderMacroBlocksMpeg2 = &vlRenderMacroBlocksMpeg2R16Snorm;
-	mc->base.vlEnd = &vlEnd;
-	mc->base.vlFlush = &vlFlush;
-	mc->base.vlDestroy = &vlDestroy;
-	mc->pipe = pipe;
-	mc->video_width = video_width;
-	mc->video_height = video_height;
-	mc->cur_buf = 0;
-
-	vlInit(mc);
-
-	*render = &mc->base;
-
-	return 0;
-}
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h
deleted file mode 100644
index 9842926bf7..0000000000
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc.h
+++ /dev/null
@@ -1,18 +0,0 @@
-#ifndef vl_r16snorm_mc_h
-#define vl_r16snorm_mc_h
-
-#include "vl_types.h"
-
-struct pipe_context;
-struct vlRender;
-
-int vlCreateR16SNormMC
-(
-	struct pipe_context *pipe,
-	unsigned int video_width,
-	unsigned int video_height,
-	enum vlFormat video_format,
-	struct vlRender **render
-);
-
-#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
index 650528ed8f..c5a73b2bf2 100644
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
+++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf.c
@@ -1,7 +1,6 @@
 #define VL_INTERNAL
 #include "vl_r16snorm_mc_buf.h"
 #include <assert.h>
-#include <stdlib.h>
 #include <pipe/p_context.h>
 #include <pipe/p_winsys.h>
 #include <pipe/p_screen.h>
@@ -10,6 +9,7 @@
 #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"
@@ -17,16 +17,7 @@
 #include "vl_types.h"
 #include "vl_defs.h"
 
-/*
- * TODO: Dynamically determine number of buf sets to use, based on
- * video size and available mem, since we can easily run out of memory
- * for high res videos.
- * Note: Destroying previous frame's buffers and creating new ones
- * doesn't work, since the buffer are not actually destroyed until their
- * fence is signalled, and if we render fast enough we will create faster
- * than we destroy.
- */
-#define NUM_BUF_SETS 4	/* Number of rotating buffer sets to use */
+const unsigned int DEFAULT_BUF_ALIGNMENT = 1;
 
 enum vlMacroBlockTypeEx
 {
@@ -52,36 +43,79 @@ struct vlFragmentShaderConsts
 	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, picture_height;
+	unsigned int				picture_width;
+	unsigned int				picture_height;
 	enum vlFormat				picture_format;
+	unsigned int				macroblocks_per_picture;
 
-	unsigned int				cur_buf;
 	struct vlSurface			*buffered_surface;
-	struct vlSurface			*past_surface, *future_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_surface			*tex_surface[3];
+	short					*texels[3];
 
 	struct pipe_context			*pipe;
 	struct pipe_viewport_state		viewport;
 	struct pipe_framebuffer_state		render_target;
-	struct pipe_sampler_state		*samplers[5];
-	struct pipe_texture			*textures[NUM_BUF_SETS][5];
-	struct pipe_surface			*tex_surface[3];
-	short					*texels[3];
+
+	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_buffer 		vertex_bufs[NUM_BUF_SETS][3];
 	struct pipe_vertex_element		vertex_elems[8];
-	struct pipe_constant_buffer		vs_const_buf, fs_const_buf;
+	struct pipe_constant_buffer		vs_const_buf;
+	struct pipe_constant_buffer		fs_const_buf;
 };
 
-static int vlBegin
+static inline int vlBegin
 (
 	struct vlRender *render
 )
@@ -360,11 +394,13 @@ static inline int vlGrabMacroBlock
 		(vb)[5].cr_tc.x = (zb)[2].x + (hx);	(vb)[5].cr_tc.y = (zb)[2].y + (hy);					\
 	}
 
-static inline int vlGrabMacroBlockVB
+static inline int vlGenMacroblockVerts
 (
 	struct vlR16SnormBufferedMC *mc,
 	struct vlMpeg2MacroBlock *macroblock,
-	unsigned int pos
+	unsigned int pos,
+	struct vlMacroBlockVertexStream0 *ycbcr_vb,
+	struct vlVertex2f **ref_vb
 )
 {
 	struct vlVertex2f	mo_vec[2];
@@ -372,6 +408,7 @@ static inline int vlGrabMacroBlockVB
 
 	assert(mc);
 	assert(macroblock);
+	assert(ycbcr_vb);
 
 	switch (macroblock->mb_type)
 	{
@@ -379,12 +416,9 @@ static inline int vlGrabMacroBlockVB
 		{
 			struct vlVertex2f *vb;
 
-			vb = (struct vlVertex2f*)mc->pipe->winsys->buffer_map
-			(
-				mc->pipe->winsys,
-				mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][2].buffer,
-				PIPE_BUFFER_USAGE_CPU_WRITE
-			) + pos * 2 * 24;
+			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;
@@ -411,8 +445,6 @@ static inline int vlGrabMacroBlockVB
 				}
 			}
 
-			mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][2].buffer);
-
 			/* fall-through */
 		}
 		case vlMacroBlockTypeFwdPredicted:
@@ -420,12 +452,9 @@ static inline int vlGrabMacroBlockVB
 		{
 			struct vlVertex2f *vb;
 
-			vb = (struct vlVertex2f*)mc->pipe->winsys->buffer_map
-			(
-				mc->pipe->winsys,
-				mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][1].buffer,
-				PIPE_BUFFER_USAGE_CPU_WRITE
-			) + pos * 2 * 24;
+			assert(ref_vb && ref_vb[0]);
+
+			vb = ref_vb[0] + pos * 2 * 24;
 
 			if (macroblock->mb_type == vlMacroBlockTypeBkwdPredicted)
 			{
@@ -469,8 +498,6 @@ static inline int vlGrabMacroBlockVB
 				}
 			}
 
-			mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][1].buffer);
-
 			/* fall-through */
 		}
 		case vlMacroBlockTypeIntra:
@@ -486,20 +513,9 @@ static inline int vlGrabMacroBlockVB
 				mc->surface_tex_inv_size.y * (VL_MACROBLOCK_HEIGHT / 2)
 			};
 
-			struct vlMacroBlockVertexStream0
-			{
-				struct vlVertex2f pos;
-				struct vlVertex2f luma_tc;
-				struct vlVertex2f cb_tc;
-				struct vlVertex2f cr_tc;
-			} *vb;
+			struct vlMacroBlockVertexStream0 *vb;
 
-			vb = (struct vlMacroBlockVertexStream0*)mc->pipe->winsys->buffer_map
-			(
-				mc->pipe->winsys,
-				mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][0].buffer,
-				PIPE_BUFFER_USAGE_CPU_WRITE
-			) + pos * 24;
+			vb = ycbcr_vb + pos * 24;
 
 			SET_BLOCK
 			(
@@ -533,8 +549,6 @@ static inline int vlGrabMacroBlockVB
 				4, 2, 1, mc->zero_block
 			);
 
-			mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS][0].buffer);
-
 			break;
 		}
 		default:
@@ -555,9 +569,6 @@ static int vlFlush
 	unsigned int			num_macroblocks[vlNumMacroBlockExTypes] = {0};
 	unsigned int			offset[vlNumMacroBlockExTypes];
 	unsigned int			vb_start = 0;
-	unsigned int			mbw;
-	unsigned int			mbh;
-	unsigned int			num_mb_per_frame;
 	unsigned int			i;
 
 	assert(render);
@@ -567,11 +578,7 @@ static int vlFlush
 	if (!mc->buffered_surface)
 		return 0;
 
-	mbw = align(mc->picture_width, VL_MACROBLOCK_WIDTH) / VL_MACROBLOCK_WIDTH;
-	mbh = align(mc->picture_height, VL_MACROBLOCK_HEIGHT) / VL_MACROBLOCK_HEIGHT;
-	num_mb_per_frame = mbw * mbh;
-
-	if (mc->num_macroblocks < num_mb_per_frame)
+	if (mc->num_macroblocks < mc->macroblocks_per_picture)
 		return 0;
 
 	pipe = mc->pipe;
@@ -588,15 +595,39 @@ static int vlFlush
 	for (i = 1; i < vlNumMacroBlockExTypes; ++i)
 		offset[i] = offset[i - 1] + num_macroblocks[i - 1];
 
-	for (i = 0; i < mc->num_macroblocks; ++i)
 	{
-		enum vlMacroBlockTypeEx mb_type_ex = vlGetMacroBlockTypeEx(&mc->macroblocks[i]);
+		struct vlMacroBlockVertexStream0	*ycbcr_vb;
+		struct vlVertex2f			*ref_vb[2];
+
+		ycbcr_vb = (struct vlMacroBlockVertexStream0*)mc->pipe->winsys->buffer_map
+		(
+			mc->pipe->winsys,
+			mc->vertex_bufs.ycbcr.buffer,
+			PIPE_BUFFER_USAGE_CPU_WRITE
+		);
 
-		vlGrabMacroBlockVB(mc, &mc->macroblocks[i], offset[mb_type_ex]);
+		for (i = 0; i < 2; ++i)
+			ref_vb[i] = (struct vlVertex2f*)mc->pipe->winsys->buffer_map
+			(
+				mc->pipe->winsys,
+				mc->vertex_bufs.ref[i].buffer,
+				PIPE_BUFFER_USAGE_CPU_WRITE
+			);
 
-		offset[mb_type_ex]++;
+		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]++;
+		}
+
+		mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs.ycbcr.buffer);
+		for (i = 0; i < 2; ++i)
+			mc->pipe->winsys->buffer_unmap(mc->pipe->winsys, mc->vertex_bufs.ref[i].buffer);
 	}
-	
+
 	for (i = 0; i < 3; ++i)
 	{
 		pipe_surface_unmap(mc->tex_surface[i]);
@@ -628,10 +659,10 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeIntra] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 1, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		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[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 3, (void**)mc->samplers);
+		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);
 
@@ -641,11 +672,11 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeFwdPredictedFrame] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
+		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]);
 
@@ -655,11 +686,11 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeFwdPredictedField] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
+		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]);
 
@@ -669,11 +700,11 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedFrame] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
+		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]);
 
@@ -683,11 +714,11 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeBkwdPredictedField] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 2, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 6, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 4, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 4, (void**)mc->samplers);
+		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]);
 
@@ -697,12 +728,12 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeBiPredictedFrame] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 8, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture;
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][4] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers);
+		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]);
 
@@ -712,12 +743,12 @@ static int vlFlush
 
 	if (num_macroblocks[vlMacroBlockExTypeBiPredictedField] > 0)
 	{
-		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs[mc->cur_buf % NUM_BUF_SETS]);
+		pipe->set_vertex_buffers(pipe, 3, mc->vertex_bufs.all);
 		pipe->set_vertex_elements(pipe, 8, mc->vertex_elems);
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][3] = mc->past_surface->texture;
-		mc->textures[mc->cur_buf % NUM_BUF_SETS][4] = mc->future_surface->texture;
-		pipe->set_sampler_textures(pipe, 5, mc->textures[mc->cur_buf % NUM_BUF_SETS]);
-		pipe->bind_sampler_states(pipe, 5, (void**)mc->samplers);
+		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]);
 
@@ -726,13 +757,13 @@ static int vlFlush
 	}
 
 	pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &mc->buffered_surface->render_fence);
+	pipe->screen->tex_surface_release(pipe->screen, &mc->render_target.cbufs[0]);
 
 	for (i = 0; i < 3; ++i)
 		mc->zero_block[i].x = -1.0f;
 
 	mc->buffered_surface = NULL;
 	mc->num_macroblocks = 0;
-	mc->cur_buf++;
 
 	return 0;
 }
@@ -745,6 +776,7 @@ static int vlRenderMacroBlocksMpeg2R16SnormBuffered
 )
 {
 	struct vlR16SnormBufferedMC	*mc;
+	bool				new_surface = false;
 	unsigned int			i;
 
 	assert(render);
@@ -756,39 +788,26 @@ static int vlRenderMacroBlocksMpeg2R16SnormBuffered
 		if (mc->buffered_surface != surface)
 		{
 			vlFlush(&mc->base);
-			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_surface[i] = mc->pipe->screen->get_tex_surface
-				(
-					mc->pipe->screen,
-					mc->textures[mc->cur_buf % NUM_BUF_SETS][i],
-					0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
-				);
-
-				mc->texels[i] = pipe_surface_map(mc->tex_surface[i], PIPE_BUFFER_USAGE_CPU_WRITE);
-			}
+			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_surface[i] = mc->pipe->screen->get_tex_surface
 			(
 				mc->pipe->screen,
-				mc->textures[mc->cur_buf % NUM_BUF_SETS][i],
+				mc->textures.all[i],
 				0, 0, 0, PIPE_BUFFER_USAGE_CPU_WRITE
 			);
 
@@ -802,7 +821,7 @@ static int vlRenderMacroBlocksMpeg2R16SnormBuffered
 	return 0;
 }
 
-static int vlEnd
+static inline int vlEnd
 (
 	struct vlRender *render
 )
@@ -819,7 +838,7 @@ static int vlDestroy
 {
 	struct vlR16SnormBufferedMC	*mc;
 	struct pipe_context		*pipe;
-	unsigned int			h, i;
+	unsigned int			i;
 
 	assert(render);
 
@@ -827,19 +846,14 @@ static int vlDestroy
 	pipe = mc->pipe;
 
 	for (i = 0; i < 5; ++i)
-		pipe->delete_sampler_state(pipe, mc->samplers[i]);
+		pipe->delete_sampler_state(pipe, mc->samplers.all[i]);
 
-	for (h = 0; h < NUM_BUF_SETS; ++h)
-			for (i = 0; i < 3; ++i)
-				pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs[h][i].buffer);
+	for (i = 0; i < 3; ++i)
+		pipe->winsys->buffer_destroy(pipe->winsys, mc->vertex_bufs.all[i].buffer);
 
 	/* Textures 3 & 4 are not created directly, no need to release them here */
-	for (i = 0; i < NUM_BUF_SETS; ++i)
-	{
-		pipe_texture_release(&mc->textures[i][0]);
-		pipe_texture_release(&mc->textures[i][1]);
-		pipe_texture_release(&mc->textures[i][2]);
-	}
+	for (i = 0; i < 3; ++i)
+		pipe_texture_release(&mc->textures.all[i]);
 
 	pipe->delete_vs_state(pipe, mc->i_vs);
 	pipe->delete_fs_state(pipe, mc->i_fs);
@@ -855,8 +869,8 @@ static int vlDestroy
 	pipe->winsys->buffer_destroy(pipe->winsys, mc->vs_const_buf.buffer);
 	pipe->winsys->buffer_destroy(pipe->winsys, mc->fs_const_buf.buffer);
 
-	free(mc->macroblocks);
-	free(mc);
+	FREE(mc->macroblocks);
+	FREE(mc);
 
 	return 0;
 }
@@ -882,42 +896,39 @@ static int vlCreateDataBufs
 {
 	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;
-	const unsigned int	num_mb_per_frame = mbw * mbh;
 
 	struct pipe_context	*pipe;
-	unsigned int		h, i;
+	unsigned int		i;
 
 	assert(mc);
 
 	pipe = mc->pipe;
+	mc->macroblocks_per_picture = mbw * mbh;
 
 	/* Create our vertex buffers */
-	for (h = 0; h < NUM_BUF_SETS; ++h)
+	mc->vertex_bufs.ycbcr.pitch = 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->winsys->buffer_create
+	(
+		pipe->winsys,
+		DEFAULT_BUF_ALIGNMENT,
+		PIPE_BUFFER_USAGE_VERTEX,
+		sizeof(struct vlVertex2f) * 4 * 24 * mc->macroblocks_per_picture
+	);
+
+	for (i = 1; i < 3; ++i)
 	{
-		mc->vertex_bufs[h][0].pitch = sizeof(struct vlVertex2f) * 4;
-		mc->vertex_bufs[h][0].max_index = 24 * num_mb_per_frame - 1;
-		mc->vertex_bufs[h][0].buffer_offset = 0;
-		mc->vertex_bufs[h][0].buffer = pipe->winsys->buffer_create
+		mc->vertex_bufs.all[i].pitch = 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->winsys->buffer_create
 		(
 			pipe->winsys,
-			1,
+			DEFAULT_BUF_ALIGNMENT,
 			PIPE_BUFFER_USAGE_VERTEX,
-			sizeof(struct vlVertex2f) * 4 * 24 * num_mb_per_frame
+			sizeof(struct vlVertex2f) * 2 * 24 * mc->macroblocks_per_picture
 		);
-
-		for (i = 1; i < 3; ++i)
-		{
-			mc->vertex_bufs[h][i].pitch = sizeof(struct vlVertex2f) * 2;
-			mc->vertex_bufs[h][i].max_index = 24 * num_mb_per_frame - 1;
-			mc->vertex_bufs[h][i].buffer_offset = 0;
-			mc->vertex_bufs[h][i].buffer = pipe->winsys->buffer_create
-			(
-				pipe->winsys,
-				1,
-				PIPE_BUFFER_USAGE_VERTEX,
-				sizeof(struct vlVertex2f) * 2 * 24 * num_mb_per_frame
-			);
-		}
 	}
 
 	/* Position element */
@@ -973,7 +984,7 @@ static int vlCreateDataBufs
 	mc->vs_const_buf.buffer = pipe->winsys->buffer_create
 	(
 		pipe->winsys,
-		1,
+		DEFAULT_BUF_ALIGNMENT,
 		PIPE_BUFFER_USAGE_CONSTANT,
 		mc->vs_const_buf.size
 	);
@@ -982,7 +993,7 @@ static int vlCreateDataBufs
 	mc->fs_const_buf.buffer = pipe->winsys->buffer_create
 	(
 		pipe->winsys,
-		1,
+		DEFAULT_BUF_ALIGNMENT,
 		PIPE_BUFFER_USAGE_CONSTANT,
 		mc->fs_const_buf.size
 	);
@@ -996,7 +1007,7 @@ static int vlCreateDataBufs
 
 	pipe->winsys->buffer_unmap(pipe->winsys, mc->fs_const_buf.buffer);
 
-	mc->macroblocks = malloc(sizeof(struct vlMpeg2MacroBlock) * num_mb_per_frame);
+	mc->macroblocks = MALLOC(sizeof(struct vlMpeg2MacroBlock) * mc->macroblocks_per_picture);
 
 	return 0;
 }
@@ -1016,6 +1027,13 @@ static int vlInit
 
 	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);
@@ -1057,7 +1075,7 @@ static int vlInit
 		/*sampler.max_lod = ;*/
 		/*sampler.border_color[i] = ;*/
 		/*sampler.max_anisotropy = ;*/
-		mc->samplers[i] = pipe->create_sampler_state(pipe, &sampler);
+		mc->samplers.all[i] = pipe->create_sampler_state(pipe, &sampler);
 	}
 
 	memset(&template, 0, sizeof(struct pipe_texture));
@@ -1069,10 +1087,9 @@ static int vlInit
 	template.depth[0] = 1;
 	template.compressed = 0;
 	pf_get_block(template.format, &template.block);
-	template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER;
+	template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC;
 
-	for (i = 0; i < NUM_BUF_SETS; ++i)
-		mc->textures[i][0] = pipe->screen->texture_create(pipe->screen, &template);
+	mc->textures.y = pipe->screen->texture_create(pipe->screen, &template);
 
 	if (mc->picture_format == vlFormatYCbCr420)
 	{
@@ -1082,13 +1099,10 @@ static int vlInit
 	else if (mc->picture_format == vlFormatYCbCr422)
 		template.height[0] = vlRoundUpPOT(mc->picture_height / 2);
 
-	for (i = 0; i < NUM_BUF_SETS; ++i)
-	{
-		mc->textures[i][1] = pipe->screen->texture_create(pipe->screen, &template);
-		mc->textures[i][2] = pipe->screen->texture_create(pipe->screen, &template);
-	}
+	mc->textures.cb = pipe->screen->texture_create(pipe->screen, &template);
+	mc->textures.cr = pipe->screen->texture_create(pipe->screen, &template);
 
-	/* textures[3] & textures[4] are assigned from vlSurfaces for P and B macroblocks at render time */
+	/* textures.all[3] & textures.all[4] are assigned from vlSurfaces for P and B macroblocks at render time */
 
 	vlCreateVertexShaderIMB(mc);
 	vlCreateFragmentShaderIMB(mc);
@@ -1114,13 +1128,12 @@ int vlCreateR16SNormBufferedMC
 	struct vlRender **render
 )
 {
-	struct vlR16SnormBufferedMC	*mc;
-	unsigned int			i;
+	struct vlR16SnormBufferedMC *mc;
 
 	assert(pipe);
 	assert(render);
 
-	mc = calloc(1, sizeof(struct vlR16SnormBufferedMC));
+	mc = CALLOC_STRUCT(vlR16SnormBufferedMC);
 
 	mc->base.vlBegin = &vlBegin;
 	mc->base.vlRenderMacroBlocksMpeg2 = &vlRenderMacroBlocksMpeg2R16SnormBuffered;
@@ -1131,14 +1144,6 @@ int vlCreateR16SNormBufferedMC
 	mc->picture_width = picture_width;
 	mc->picture_height = picture_height;
 
-	mc->cur_buf = 0;
-	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;
-
 	vlInit(mc);
 
 	*render = &mc->base;
diff --git a/src/gallium/state_trackers/g3dvl/vl_screen.c b/src/gallium/state_trackers/g3dvl/vl_screen.c
index 484f63b0d4..ade8643a66 100644
--- a/src/gallium/state_trackers/g3dvl/vl_screen.c
+++ b/src/gallium/state_trackers/g3dvl/vl_screen.c
@@ -1,7 +1,7 @@
 #define VL_INTERNAL
 #include "vl_screen.h"
 #include <assert.h>
-#include <stdlib.h>
+#include <util/u_memory.h>
 
 int vlCreateScreen
 (
@@ -17,7 +17,7 @@ int vlCreateScreen
 	assert(pscreen);
 	assert(vl_screen);
 
-	scrn = calloc(1, sizeof(struct vlScreen));
+	scrn = CALLOC_STRUCT(vlScreen);
 
 	if (!scrn)
 		return 1;
@@ -37,7 +37,7 @@ int vlDestroyScreen
 {
 	assert(screen);
 
-	free(screen);
+	FREE(screen);
 
 	return 0;
 }
diff --git a/src/gallium/state_trackers/g3dvl/vl_surface.c b/src/gallium/state_trackers/g3dvl/vl_surface.c
index 076bd40d41..911469f966 100644
--- a/src/gallium/state_trackers/g3dvl/vl_surface.c
+++ b/src/gallium/state_trackers/g3dvl/vl_surface.c
@@ -1,11 +1,11 @@
 #define VL_INTERNAL
 #include "vl_surface.h"
 #include <assert.h>
-#include <stdlib.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"
@@ -28,7 +28,7 @@ int vlCreateSurface
 	assert(screen);
 	assert(surface);
 
-	sfc = calloc(1, sizeof(struct vlSurface));
+	sfc = CALLOC_STRUCT(vlSurface);
 
 	if (!sfc)
 		return 1;
@@ -64,7 +64,7 @@ int vlDestroySurface
 	assert(surface);
 
 	pipe_texture_release(&surface->texture);
-	free(surface);
+	FREE(surface);
 
 	return 0;
 }
diff --git a/src/gallium/winsys/drm/nouveau/Makefile b/src/gallium/winsys/drm/nouveau/Makefile
index 81562ca78d..b5735329ec 100644
--- a/src/gallium/winsys/drm/nouveau/Makefile
+++ b/src/gallium/winsys/drm/nouveau/Makefile
@@ -1,46 +1,25 @@
-
 TOP = ../../../../..
 include $(TOP)/configs/current
 
-LIBNAME = nouveau_dri.so
-
-MINIGLX_SOURCES =
-
-PIPE_DRIVERS = \
-	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
-	$(TOP)/src/gallium/drivers/nv04/libnv04.a \
-	$(TOP)/src/gallium/drivers/nv10/libnv10.a \
-	$(TOP)/src/gallium/drivers/nv20/libnv20.a \
-	$(TOP)/src/gallium/drivers/nv30/libnv30.a \
-	$(TOP)/src/gallium/drivers/nv40/libnv40.a \
-	$(TOP)/src/gallium/drivers/nv50/libnv50.a
-
-DRIVER_SOURCES = \
-	nouveau_bo.c \
-	nouveau_channel.c \
-	nouveau_context.c \
-	nouveau_device.c \
-	nouveau_dma.c \
-	nouveau_fence.c \
-	nouveau_grobj.c \
-	nouveau_lock.c \
-	nouveau_notifier.c \
-	nouveau_pushbuf.c \
-	nouveau_resource.c \
-	nouveau_screen.c \
-	nouveau_swapbuffers.c \
-	nouveau_winsys.c \
-	nouveau_winsys_pipe.c \
-	nouveau_winsys_softpipe.c \
-	nv04_surface.c \
-	nv50_surface.c
-
-C_SOURCES = \
-	$(COMMON_GALLIUM_SOURCES) \
-	$(DRIVER_SOURCES)
-
-ASM_SOURCES = 
-
-include ../Makefile.template
-
-symlinks:
+
+SUBDIRS = common dri
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`
+	rm -f `find . -name depend`
+
+
+# Dummy install target
+install:
diff --git a/src/gallium/winsys/drm/nouveau/common/Makefile b/src/gallium/winsys/drm/nouveau/common/Makefile
new file mode 100644
index 0000000000..06f558959d
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/common/Makefile
@@ -0,0 +1,32 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = nouveaudrm
+
+C_SOURCES = \
+	nouveau_bo.c \
+	nouveau_channel.c \
+	nouveau_context.c \
+	nouveau_device.c \
+	nouveau_dma.c \
+	nouveau_fence.c \
+	nouveau_grobj.c \
+	nouveau_lock.c \
+	nouveau_notifier.c \
+	nouveau_pushbuf.c \
+	nouveau_resource.c \
+	nouveau_screen.c \
+	nouveau_winsys.c \
+	nouveau_winsys_pipe.c \
+	nouveau_winsys_softpipe.c \
+	nv04_surface.c \
+	nv50_surface.c
+
+
+include ./Makefile.template
+
+DRIVER_DEFINES = $(shell pkg-config libdrm --cflags \
+                && pkg-config libdrm --atleast-version=2.3.1 \
+                && echo "-DDRM_VBLANK_FLIP=DRM_VBLANK_FLIP")
+symlinks:
+
diff --git a/src/gallium/winsys/drm/nouveau/common/Makefile.template b/src/gallium/winsys/drm/nouveau/common/Makefile.template
new file mode 100644
index 0000000000..e40836e0a8
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/common/Makefile.template
@@ -0,0 +1,59 @@
+# -*-makefile-*-
+
+COMMON_SOURCES = 
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+          $(CPP_SOURCES:.cpp=.o) \
+	  $(ASM_SOURCES:.S=.o) 
+
+
+### Include directories
+INCLUDES = \
+	-I. \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/auxiliary \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/include \
+        $(DRIVER_INCLUDES)
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDES) $(CXXFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES)  $< -o $@
+
+
+##### TARGETS #####
+
+default: depend symlinks $(LIBNAME)
+
+
+$(LIBNAME): $(OBJECTS) Makefile Makefile.template
+	$(TOP)/bin/mklib -o $@ -static $(OBJECTS) $(DRIVER_LIBS)
+
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) \
+		$(ASM_SOURCES) 2> /dev/null
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean::
+	-rm -f *.o */*.o *~ *.so *~ server/*.o $(SYMLINKS)
+	-rm -f depend depend.bak
+
+
+include depend
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_bo.c b/src/gallium/winsys/drm/nouveau/common/nouveau_bo.c
index b5942994d9..76b98bed67 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_bo.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_bo.c
@@ -273,6 +273,40 @@ nouveau_bo_del(struct nouveau_bo **bo)
 }
 
 int
+nouveau_bo_busy(struct nouveau_bo *bo, uint32_t flags)
+{
+	struct nouveau_bo_priv *nvbo = nouveau_bo(bo);
+	struct nouveau_fence *fence;
+
+	if (!nvbo)
+		return -EINVAL;
+
+	/* If the buffer is pending it must be busy, unless
+	 * both are RD, in which case we can allow access */
+	if (nvbo->pending) {
+		if ((nvbo->pending->flags & NOUVEAU_BO_RDWR) == NOUVEAU_BO_RD &&
+		    (flags & NOUVEAU_BO_RDWR) == NOUVEAU_BO_RD)
+			return 0;
+		else
+			return 1;
+	}
+
+	if (flags & NOUVEAU_BO_WR)
+		fence = nvbo->fence;
+	else
+		fence = nvbo->wr_fence;
+
+	/* If the buffer is not pending and doesn't have a fence
+	 * that conflicts with our flags then it can't be busy
+	 */
+	if (!fence)
+		return 0;
+	else
+		/* If the fence is signalled the buffer is not busy, else is busy */
+		return !nouveau_fence(fence)->signalled;
+}
+
+int
 nouveau_bo_map(struct nouveau_bo *bo, uint32_t flags)
 {
 	struct nouveau_bo_priv *nvbo = nouveau_bo(bo);
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_channel.c b/src/gallium/winsys/drm/nouveau/common/nouveau_channel.c
index 3b4dcd1ecf..b7298131c1 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_channel.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_channel.c
@@ -23,7 +23,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
-
+#include <util/u_memory.h>
 #include "nouveau_drmif.h"
 #include "nouveau_dma.h"
 
@@ -38,7 +38,7 @@ nouveau_channel_alloc(struct nouveau_device *dev, uint32_t fb_ctxdma,
 	if (!nvdev || !chan || *chan)
 	    return -EINVAL;
 
-	nvchan = calloc(1, sizeof(struct nouveau_channel_priv));
+	nvchan = CALLOC_STRUCT(nouveau_channel_priv);
 	if (!nvchan)
 		return -ENOMEM;
 	nvchan->base.device = dev;
@@ -48,7 +48,7 @@ nouveau_channel_alloc(struct nouveau_device *dev, uint32_t fb_ctxdma,
 	ret = drmCommandWriteRead(nvdev->fd, DRM_NOUVEAU_CHANNEL_ALLOC,
 				  &nvchan->drm, sizeof(nvchan->drm));
 	if (ret) {
-		free(nvchan);
+		FREE(nvchan);
 		return ret;
 	}
 
@@ -111,16 +111,16 @@ nouveau_channel_free(struct nouveau_channel **chan)
 	nvchan = nouveau_channel(*chan);
 	*chan = NULL;
 	nvdev = nouveau_device(nvchan->base.device);
-	
+
 	FIRE_RING_CH(&nvchan->base);
 
 	nouveau_grobj_free(&nvchan->base.vram);
 	nouveau_grobj_free(&nvchan->base.gart);
 	nouveau_grobj_free(&nvchan->base.nullobj);
 
+	FREE(nvchan->pb.buffers);
+	FREE(nvchan->pb.relocs);
 	cf.channel = nvchan->drm.channel;
 	drmCommandWrite(nvdev->fd, DRM_NOUVEAU_CHANNEL_FREE, &cf, sizeof(cf));
-	free(nvchan);
+	FREE(nvchan);
 }
-
-
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_context.c b/src/gallium/winsys/drm/nouveau/common/nouveau_context.c
index 74413c408f..2f245046d4 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_context.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_context.c
@@ -1,28 +1,13 @@
-#include "main/glheader.h"
-#include "glapi/glthread.h"
-#include <GL/internal/glcore.h>
-#include "utils.h"
-
-#include "state_tracker/st_public.h"
-#include "state_tracker/st_context.h"
-#include "pipe/p_defines.h"
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-
+#include <pipe/p_defines.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+#include <util/u_memory.h>
 #include "nouveau_context.h"
 #include "nouveau_dri.h"
 #include "nouveau_local.h"
 #include "nouveau_screen.h"
 #include "nouveau_winsys_pipe.h"
 
-#ifdef DEBUG
-static const struct dri_debug_control debug_control[] = {
-	{ "bo", DEBUG_BO },
-	{ NULL, 0 }
-};
-int __nouveau_debug = 0;
-#endif
-
 static void
 nouveau_channel_context_destroy(struct nouveau_channel_context *nvc)
 {
@@ -87,24 +72,17 @@ nouveau_channel_context_create(struct nouveau_device *dev)
 	return nvc;
 }
 
-GLboolean
-nouveau_context_create(const __GLcontextModes *glVis,
-		       __DRIcontextPrivate *driContextPriv,
-		       void *sharedContextPrivate)
+int
+nouveau_context_init(struct nouveau_screen *nv_screen,
+                     drm_context_t hHWContext, drmLock *sarea_lock,
+                     struct nouveau_context *nv_share,
+                     struct nouveau_context *nv)
 {
-	__DRIscreenPrivate *driScrnPriv = driContextPriv->driScreenPriv;
-	struct nouveau_screen  *nv_screen = driScrnPriv->private;
-	struct nouveau_context *nv = CALLOC_STRUCT(nouveau_context);
 	struct pipe_context *pipe = NULL;
-	struct st_context *st_share = NULL;
 	struct nouveau_channel_context *nvc = NULL;
 	struct nouveau_device *dev = nv_screen->device;
 	int i;
 
-	if (sharedContextPrivate) {
-		st_share = ((struct nouveau_context *)sharedContextPrivate)->st;
-	}
-
 	switch (dev->chipset & 0xf0) {
 	case 0x10:
 	case 0x20:
@@ -121,27 +99,18 @@ nouveau_context_create(const __GLcontextModes *glVis,
 		break;
 	default:
 		NOUVEAU_ERR("Unsupported chipset: NV%02x\n", dev->chipset);
-		return GL_FALSE;
+		return 1;
 	}
 
-	driContextPriv->driverPrivate = (void *)nv;
 	nv->nv_screen  = nv_screen;
-	nv->dri_screen = driScrnPriv;
 
 	{
 		struct nouveau_device_priv *nvdev = nouveau_device(dev);
 
-		nvdev->ctx  = driContextPriv->hHWContext;
-		nvdev->lock = (drmLock *)&driScrnPriv->pSAREA->lock;
+		nvdev->ctx  = hHWContext;
+		nvdev->lock = sarea_lock;
 	}
 
-	driParseConfigFiles(&nv->dri_option_cache, &nv_screen->option_cache,
-			    nv->dri_screen->myNum, "nouveau");
-#ifdef DEBUG
-	__nouveau_debug = driParseDebugString(getenv("NOUVEAU_DEBUG"),
-					      debug_control);
-#endif
-
 	/*XXX: Hack up a fake region and buffer object for front buffer.
 	 *     This will go away with TTM, replaced with a simple reference
 	 *     of the front buffer handle passed to us by the DDX.
@@ -185,12 +154,8 @@ nouveau_context_create(const __GLcontextModes *glVis,
 	 * a single process.
 	 */
 	nvc = nv_screen->nvc;
-	if (!nvc && st_share) {
-		struct nouveau_context *snv = st_share->pipe->priv;
-		if (snv) {
-			nvc = snv->nvc;
-		}
-	}
+	if (!nvc && nv_share)
+		nvc = nv_share->nvc;
 
 	/*XXX: temporary - disable multi-context/single-channel on pre-NV4x */
 	switch (dev->chipset & 0xf0) {
@@ -211,7 +176,7 @@ nouveau_context_create(const __GLcontextModes *glVis,
 		nvc = nouveau_channel_context_create(dev);
 		if (!nvc) {
 			NOUVEAU_ERR("Failed initialising GPU context\n");
-			return GL_FALSE;
+			return 1;
 		}
 		nv_screen->nvc = nvc;
 	}
@@ -241,11 +206,11 @@ nouveau_context_create(const __GLcontextModes *glVis,
 	case 0x80:
 	case 0x90:
 		if (nouveau_surface_init_nv50(nv))
-			return GL_FALSE;
+			return 1;
 		break;
 	default:
 		if (nouveau_surface_init_nv04(nv))
-			return GL_FALSE;
+			return 1;
 		break;
 	}
 
@@ -268,26 +233,22 @@ nouveau_context_create(const __GLcontextModes *glVis,
 		pipe = nouveau_create_softpipe(nv);
 		if (!pipe) {
 			NOUVEAU_ERR("Error creating pipe, bailing\n");
-			return GL_FALSE;
+			return 1;
 		}
 	}
 
 	pipe->priv = nv;
-	nv->st = st_create_context(pipe, glVis, st_share);
-	return GL_TRUE;
+
+	return 0;
 }
 
 void
-nouveau_context_destroy(__DRIcontextPrivate *driContextPriv)
+nouveau_context_cleanup(struct nouveau_context *nv)
 {
-	struct nouveau_context *nv = driContextPriv->driverPrivate;
 	struct nouveau_channel_context *nvc = nv->nvc;
 
 	assert(nv);
 
-	st_finish(nv->st);
-	st_destroy_context(nv->st);
-
 	if (nv->pctx_id >= 0) {
 		nvc->pctx[nv->pctx_id] = NULL;
 		if (--nvc->refcount <= 0) {
@@ -295,52 +256,7 @@ nouveau_context_destroy(__DRIcontextPrivate *driContextPriv)
 			nv->nv_screen->nvc = NULL;
 		}
 	}
-
-	free(nv);
-}
-
-GLboolean
-nouveau_context_bind(__DRIcontextPrivate *driContextPriv,
-		     __DRIdrawablePrivate *driDrawPriv,
-		     __DRIdrawablePrivate *driReadPriv)
-{
-	struct nouveau_context *nv;
-	struct nouveau_framebuffer *draw, *read;
-
-	if (!driContextPriv) {
-		st_make_current(NULL, NULL, NULL);
-		return GL_TRUE;
-	}
-
-	nv = driContextPriv->driverPrivate;
-	draw = driDrawPriv->driverPrivate;
-	read = driReadPriv->driverPrivate;
-
-	st_make_current(nv->st, draw->stfb, read->stfb);
-
-	if ((nv->dri_drawable != driDrawPriv) ||
-	    (nv->last_stamp != driDrawPriv->lastStamp)) {
-		nv->dri_drawable = driDrawPriv;
-		st_resize_framebuffer(draw->stfb, driDrawPriv->w,
-				      driDrawPriv->h);
-		nv->last_stamp = driDrawPriv->lastStamp;
-	}
-
-	if (driDrawPriv != driReadPriv) {
-		st_resize_framebuffer(read->stfb, driReadPriv->w,
-				      driReadPriv->h);
-	}
-
-	return GL_TRUE;
-}
-
-GLboolean
-nouveau_context_unbind(__DRIcontextPrivate *driContextPriv)
-{
-	struct nouveau_context *nv = driContextPriv->driverPrivate;
-	(void)nv;
-
-	st_flush(nv->st, 0, NULL);
-	return GL_TRUE;
+	
+	/* XXX: Who cleans up the pipe? */
 }
 
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_context.h b/src/gallium/winsys/drm/nouveau/common/nouveau_context.h
index 77e2147a2c..b1bdb01bdf 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_context.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_context.h
@@ -1,17 +1,10 @@
 #ifndef __NOUVEAU_CONTEXT_H__
 #define __NOUVEAU_CONTEXT_H__
 
-#include "dri_util.h"
-#include "xmlconfig.h"
-
 #include "nouveau/nouveau_winsys.h"
 #include "nouveau_drmif.h"
 #include "nouveau_dma.h"
 
-struct nouveau_framebuffer {
-	struct st_framebuffer *stfb;
-};
-
 struct nouveau_channel_context {
 	struct pipe_screen *pscreen;
 	int refcount;
@@ -41,16 +34,7 @@ struct nouveau_channel_context {
 };
 
 struct nouveau_context {
-	struct st_context *st;
-
-	/* DRI stuff */
-	__DRIscreenPrivate    *dri_screen;
-	__DRIdrawablePrivate  *dri_drawable;
-	unsigned int           last_stamp;
-	driOptionCache         dri_option_cache;
-	drm_context_t          drm_context;
-	drmLock                drm_lock;
-	GLboolean              locked;
+	int locked;
 	struct nouveau_screen *nv_screen;
 	struct pipe_surface *frontbuffer;
 
@@ -76,26 +60,11 @@ struct nouveau_context {
 			    unsigned, unsigned, unsigned, unsigned, unsigned);
 };
 
-extern GLboolean nouveau_context_create(const __GLcontextModes *,
-					__DRIcontextPrivate *, void *);
-extern void nouveau_context_destroy(__DRIcontextPrivate *);
-extern GLboolean nouveau_context_bind(__DRIcontextPrivate *,
-				      __DRIdrawablePrivate *draw,
-				      __DRIdrawablePrivate *read);
-extern GLboolean nouveau_context_unbind(__DRIcontextPrivate *);
-
-#ifdef DEBUG
-extern int __nouveau_debug;
-
-#define DEBUG_BO (1 << 0)
-
-#define DBG(flag, ...) do {                   \
-	if (__nouveau_debug & (DEBUG_##flag)) \
-		NOUVEAU_ERR(__VA_ARGS__);     \
-} while(0)
-#else
-#define DBG(flag, ...)
-#endif
+extern int nouveau_context_init(struct nouveau_screen *nv_screen,
+                                drm_context_t hHWContext, drmLock *sarea_lock,
+                                struct nouveau_context *nv_share,
+                                struct nouveau_context *nv);
+extern void nouveau_context_cleanup(struct nouveau_context *nv);
 
 extern void LOCK_HARDWARE(struct nouveau_context *);
 extern void UNLOCK_HARDWARE(struct nouveau_context *);
@@ -110,4 +79,8 @@ extern int nouveau_surface_init_nv50(struct nouveau_context *);
 extern uint32_t *nouveau_pipe_dma_beginp(struct nouveau_grobj *, int, int);
 extern void nouveau_pipe_dma_kickoff(struct nouveau_channel *);
 
+/* Must be provided by clients of common code */
+extern void
+nouveau_contended_lock(struct nouveau_context *nv);
+
 #endif
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_device.c b/src/gallium/winsys/drm/nouveau/common/nouveau_device.c
index 0b452fcd02..92b57b834b 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_device.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_device.c
@@ -23,7 +23,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <errno.h>
-
+#include <util/u_memory.h>
 #include "nouveau_drmif.h"
 
 int
@@ -36,7 +36,7 @@ nouveau_device_open_existing(struct nouveau_device **dev, int close,
 	if (!dev || *dev)
 	    return -EINVAL;
 
-	nvdev = calloc(1, sizeof(*nvdev));
+	nvdev = CALLOC_STRUCT(nouveau_device_priv);
 	if (!nvdev)
 	    return -ENOMEM;
 	nvdev->fd = fd;
@@ -112,7 +112,7 @@ nouveau_device_close(struct nouveau_device **dev)
 		drmDestroyContext(nvdev->fd, nvdev->ctx);
 		drmClose(nvdev->fd);
 	}
-	free(nvdev);
+	FREE(nvdev);
 }
 
 int
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_dma.c b/src/gallium/winsys/drm/nouveau/common/nouveau_dma.c
index f8a8ba04f6..f8a8ba04f6 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_dma.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_dma.c
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_dma.h b/src/gallium/winsys/drm/nouveau/common/nouveau_dma.h
index cfa6d26e82..cfa6d26e82 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_dma.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_dma.h
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_dri.h b/src/gallium/winsys/drm/nouveau/common/nouveau_dri.h
index 1207c2d609..1207c2d609 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_dri.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_dri.h
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_drmif.h b/src/gallium/winsys/drm/nouveau/common/nouveau_drmif.h
index dcd6a5eb0a..5f72800676 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_drmif.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_drmif.h
@@ -287,6 +287,9 @@ extern void
 nouveau_bo_del(struct nouveau_bo **);
 
 extern int
+nouveau_bo_busy(struct nouveau_bo *bo, uint32_t flags);
+
+extern int
 nouveau_bo_map(struct nouveau_bo *, uint32_t flags);
 
 extern void
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_fence.c b/src/gallium/winsys/drm/nouveau/common/nouveau_fence.c
index e7b0b4ff07..e7b0b4ff07 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_fence.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_fence.c
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_grobj.c b/src/gallium/winsys/drm/nouveau/common/nouveau_grobj.c
index 51523897d5..fb430a25b8 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_grobj.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_grobj.c
@@ -22,7 +22,7 @@
 
 #include <stdlib.h>
 #include <errno.h>
-
+#include <util/u_memory.h>
 #include "nouveau_drmif.h"
 
 int
@@ -37,7 +37,7 @@ nouveau_grobj_alloc(struct nouveau_channel *chan, uint32_t handle,
 	if (!nvdev || !grobj || *grobj)
 		return -EINVAL;
 
-	nvgrobj = calloc(1, sizeof(*nvgrobj));
+	nvgrobj = CALLOC_STRUCT(nouveau_grobj_priv);
 	if (!nvgrobj)
 		return -ENOMEM;
 	nvgrobj->base.channel = chan;
@@ -67,7 +67,7 @@ nouveau_grobj_ref(struct nouveau_channel *chan, uint32_t handle,
 	if (!chan || !grobj || *grobj)
 		return -EINVAL;
 
-	nvgrobj = calloc(1, sizeof(struct nouveau_grobj_priv));
+	nvgrobj = CALLOC_STRUCT(nouveau_grobj_priv);
 	if (!nvgrobj)
 		return -ENOMEM;
 	nvgrobj->base.channel = chan;
@@ -102,6 +102,6 @@ nouveau_grobj_free(struct nouveau_grobj **grobj)
 		drmCommandWrite(nvdev->fd, DRM_NOUVEAU_GPUOBJ_FREE,
 				&f, sizeof(f));	
 	}
-	free(nvgrobj);
+	FREE(nvgrobj);
 }
 
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_local.h b/src/gallium/winsys/drm/nouveau/common/nouveau_local.h
index e878a40803..877c7a8c47 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_local.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_local.h
@@ -5,8 +5,6 @@
 #include "nouveau_winsys_pipe.h"
 #include <stdio.h>
 
-struct pipe_buffer;
-
 /* Debug output */
 #define NOUVEAU_MSG(fmt, args...) do {                                         \
 	fprintf(stdout, "nouveau: "fmt, ##args);                               \
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_lock.c b/src/gallium/winsys/drm/nouveau/common/nouveau_lock.c
index 9adb9ac854..e8cf051ed9 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_lock.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_lock.c
@@ -25,34 +25,11 @@
  * 
  **************************************************************************/
 
-#include "main/glheader.h"
-#include "glapi/glthread.h"
-#include <GL/internal/glcore.h>
-
+#include <pipe/p_thread.h>
 #include "nouveau_context.h"
 #include "nouveau_screen.h"
 
-_glthread_DECLARE_STATIC_MUTEX( lockMutex );
-
-static void
-nouveau_contended_lock(struct nouveau_context *nv, GLuint flags)
-{
-	__DRIdrawablePrivate *dPriv = nv->dri_drawable;
-	__DRIscreenPrivate *sPriv = nv->dri_screen;
-	struct nouveau_screen *nv_screen = nv->nv_screen;
-	struct nouveau_device *dev = nv_screen->device;
-	struct nouveau_device_priv *nvdev = nouveau_device(dev);
-
-	drmGetLock(nvdev->fd, nvdev->ctx, flags);
-
-	/* If the window moved, may need to set a new cliprect now.
-	 *
-	 * NOTE: This releases and regains the hw lock, so all state
-	 * checking must be done *after* this call:
-	 */
-	if (dPriv)
-		DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
-}
+pipe_static_mutex(lockMutex);
 
 /* Lock the hardware and validate our state.
  */
@@ -64,20 +41,21 @@ LOCK_HARDWARE(struct nouveau_context *nv)
 	struct nouveau_device_priv *nvdev = nouveau_device(dev);
 	char __ret=0;
 
-	_glthread_LOCK_MUTEX(lockMutex);
 	assert(!nv->locked);
-	
+	pipe_mutex_lock(lockMutex);
+
 	DRM_CAS(nvdev->lock, nvdev->ctx,
 		(DRM_LOCK_HELD | nvdev->ctx), __ret);
-	
-	if (__ret)
-		nouveau_contended_lock(nv, 0);
-	nv->locked = GL_TRUE;
-}
 
+	if (__ret) {
+		drmGetLock(nvdev->fd, nvdev->ctx, 0);
+		nouveau_contended_lock(nv);
+	}
+	nv->locked = 1;
+}
 
-  /* Unlock the hardware using the global current context 
-   */
+/* Unlock the hardware using the global current context 
+ */
 void
 UNLOCK_HARDWARE(struct nouveau_context *nv)
 {
@@ -86,9 +64,9 @@ UNLOCK_HARDWARE(struct nouveau_context *nv)
 	struct nouveau_device_priv *nvdev = nouveau_device(dev);
 
 	assert(nv->locked);
-	nv->locked = GL_FALSE;
+	nv->locked = 0;
 
 	DRM_UNLOCK(nvdev->fd, nvdev->lock, nvdev->ctx);
 
-	_glthread_UNLOCK_MUTEX(lockMutex);
+	pipe_mutex_unlock(lockMutex);
 } 
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_notifier.c b/src/gallium/winsys/drm/nouveau/common/nouveau_notifier.c
index 01e8f38440..01e8f38440 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_notifier.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_notifier.c
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_pushbuf.c b/src/gallium/winsys/drm/nouveau/common/nouveau_pushbuf.c
index 815046ba85..7c094eb795 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_pushbuf.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_pushbuf.c
@@ -23,7 +23,7 @@
 #include <stdlib.h>
 #include <errno.h>
 #include <assert.h>
-
+#include <util/u_memory.h>
 #include "nouveau_drmif.h"
 #include "nouveau_dma.h"
 
@@ -97,9 +97,9 @@ nouveau_pushbuf_init(struct nouveau_channel *chan)
 	nouveau_pushbuf_space(chan, 0);
 	chan->pushbuf = &nvchan->pb.base;
 
-	nvchan->pb.buffers = calloc(NOUVEAU_PUSHBUF_MAX_BUFFERS,
+	nvchan->pb.buffers = CALLOC(NOUVEAU_PUSHBUF_MAX_BUFFERS,
 				    sizeof(struct nouveau_pushbuf_bo));
-	nvchan->pb.relocs = calloc(NOUVEAU_PUSHBUF_MAX_RELOCS,
+	nvchan->pb.relocs = CALLOC(NOUVEAU_PUSHBUF_MAX_RELOCS,
 				   sizeof(struct nouveau_pushbuf_reloc));
 	return 0;
 }
@@ -268,4 +268,3 @@ nouveau_pushbuf_emit_reloc(struct nouveau_channel *chan, void *ptr,
 		*(uint32_t *)ptr = nouveau_pushbuf_calc_reloc(bo, r);
 	return 0;
 }
-
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_resource.c b/src/gallium/winsys/drm/nouveau/common/nouveau_resource.c
index 3bbcb5c45e..766fd279fe 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_resource.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_resource.c
@@ -22,7 +22,7 @@
 
 #include <stdlib.h>
 #include <errno.h>
-
+#include <util/u_memory.h>
 #include "nouveau_drmif.h"
 #include "nouveau_local.h"
 
@@ -32,7 +32,7 @@ nouveau_resource_init(struct nouveau_resource **heap,
 {
 	struct nouveau_resource *r;
 
-	r = calloc(1, sizeof(struct nouveau_resource));
+	r = CALLOC_STRUCT(nouveau_resource);
 	if (!r)
 		return 1;
 
@@ -53,7 +53,7 @@ nouveau_resource_alloc(struct nouveau_resource *heap, int size, void *priv,
 
 	while (heap) {
 		if (!heap->in_use && heap->size >= size) {
-			r = calloc(1, sizeof(struct nouveau_resource));
+			r = CALLOC_STRUCT(nouveau_resource);
 			if (!r)
 				return 1;
 
@@ -73,7 +73,7 @@ nouveau_resource_alloc(struct nouveau_resource *heap, int size, void *priv,
 			*res = r;
 			return 0;
 		}
-			
+
 		heap = heap->next;
 	}
 
@@ -110,7 +110,7 @@ nouveau_resource_free(struct nouveau_resource **res)
 		if (r->next)
 			r->next->prev = r->prev;
 		r->prev->size += r->size;
-		free(r);
+		FREE(r);
 	}
-	
+
 }
diff --git a/src/gallium/winsys/drm/nouveau/common/nouveau_screen.c b/src/gallium/winsys/drm/nouveau/common/nouveau_screen.c
new file mode 100644
index 0000000000..422fbf0207
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_screen.c
@@ -0,0 +1,31 @@
+#include <util/u_memory.h>
+#include "nouveau_dri.h"
+#include "nouveau_local.h"
+#include "nouveau_screen.h"
+
+int
+nouveau_screen_init(struct nouveau_dri *nv_dri, int dev_fd,
+                    struct nouveau_screen *nv_screen)
+{
+	int ret;
+
+	ret = nouveau_device_open_existing(&nv_screen->device, 0,
+	                                   dev_fd, 0);
+	if (ret) {
+		NOUVEAU_ERR("Failed opening nouveau device: %d\n", ret);
+		return 1;
+	}
+
+	nv_screen->front_offset = nv_dri->front_offset;
+	nv_screen->front_pitch  = nv_dri->front_pitch * (nv_dri->bpp / 8);
+	nv_screen->front_cpp = nv_dri->bpp / 8;
+	nv_screen->front_height = nv_dri->height;
+
+	return 0;
+}
+
+void
+nouveau_screen_cleanup(struct nouveau_screen *nv_screen)
+{
+	nouveau_device_close(&nv_screen->device);
+}
diff --git a/src/gallium/winsys/drm/nouveau/common/nouveau_screen.h b/src/gallium/winsys/drm/nouveau/common/nouveau_screen.h
new file mode 100644
index 0000000000..3e68e219d8
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_screen.h
@@ -0,0 +1,27 @@
+#ifndef __NOUVEAU_SCREEN_H__
+#define __NOUVEAU_SCREEN_H__
+
+#include <stdint.h>
+
+struct nouveau_device;
+struct nouveau_dri;
+
+struct nouveau_screen {
+	struct nouveau_device *device;
+
+	uint32_t front_offset;
+	uint32_t front_pitch;
+	uint32_t front_cpp;
+	uint32_t front_height;
+
+	void *nvc;
+};
+
+int
+nouveau_screen_init(struct nouveau_dri *nv_dri, int dev_fd,
+                    struct nouveau_screen *nv_screen);
+
+void
+nouveau_screen_cleanup(struct nouveau_screen *nv_screen);
+
+#endif
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_winsys.c b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys.c
index 364340e1d3..364340e1d3 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_winsys.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys.c
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_winsys_pipe.c b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c
index 5276806de6..6895137506 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_winsys_pipe.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.c
@@ -1,55 +1,29 @@
-#include "pipe/p_winsys.h"
-#include "pipe/p_defines.h"
-#include "pipe/p_inlines.h"
-
-#include "util/u_memory.h"
-
+#include <pipe/p_winsys.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
 #include "nouveau_context.h"
 #include "nouveau_local.h"
 #include "nouveau_screen.h"
-#include "nouveau_swapbuffers.h"
 #include "nouveau_winsys_pipe.h"
 
-static void
-nouveau_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surf,
-			  void *context_private)
-{
-	struct nouveau_context *nv = context_private;
-	__DRIdrawablePrivate *dPriv = nv->dri_drawable;
-
-	nouveau_copy_buffer(dPriv, surf, NULL);
-}
-
 static const char *
 nouveau_get_name(struct pipe_winsys *pws)
 {
 	return "Nouveau/DRI";
 }
 
-static struct pipe_buffer *
-nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
-		       unsigned usage, unsigned size)
+static uint32_t
+nouveau_flags_from_usage(struct nouveau_context *nv, unsigned usage)
 {
-	struct nouveau_pipe_winsys *nvpws = (struct nouveau_pipe_winsys *)pws;
-	struct nouveau_context *nv = nvpws->nv;
 	struct nouveau_device *dev = nv->nv_screen->device;
-	struct nouveau_pipe_buffer *nvbuf;
-	uint32_t flags;
-
-	nvbuf = calloc(1, sizeof(*nvbuf));
-	if (!nvbuf)
-		return NULL;
-	nvbuf->base.refcount = 1;
-	nvbuf->base.alignment = alignment;
-	nvbuf->base.usage = usage;
-	nvbuf->base.size = size;
-
-	flags = NOUVEAU_BO_LOCAL;
+	uint32_t flags = NOUVEAU_BO_LOCAL;
 
 	if (usage & PIPE_BUFFER_USAGE_PIXEL) {
 		if (usage & NOUVEAU_BUFFER_USAGE_TEXTURE)
 			flags |= NOUVEAU_BO_GART;
-		flags |= NOUVEAU_BO_VRAM;
+		if (!(usage & PIPE_BUFFER_USAGE_CPU_READ_WRITE))
+			flags |= NOUVEAU_BO_VRAM;
 
 		switch (dev->chipset & 0xf0) {
 		case 0x50:
@@ -74,8 +48,31 @@ nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
 			flags |= NOUVEAU_BO_GART;
 	}
 
+	return flags;
+}
+
+static struct pipe_buffer *
+nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
+		       unsigned usage, unsigned size)
+{
+	struct nouveau_pipe_winsys *nvpws = (struct nouveau_pipe_winsys *)pws;
+	struct nouveau_context *nv = nvpws->nv;
+	struct nouveau_device *dev = nv->nv_screen->device;
+	struct nouveau_pipe_buffer *nvbuf;
+	uint32_t flags;
+
+	nvbuf = CALLOC_STRUCT(nouveau_pipe_buffer);
+	if (!nvbuf)
+		return NULL;
+	nvbuf->base.refcount = 1;
+	nvbuf->base.alignment = alignment;
+	nvbuf->base.usage = usage;
+	nvbuf->base.size = size;
+
+	flags = nouveau_flags_from_usage(nv, flags);
+
 	if (nouveau_bo_new(dev, flags, alignment, size, &nvbuf->bo)) {
-		free(nvbuf);
+		FREE(nvbuf);
 		return NULL;
 	}
 
@@ -89,14 +86,14 @@ nouveau_pipe_bo_user_create(struct pipe_winsys *pws, void *ptr, unsigned bytes)
 	struct nouveau_device *dev = nvpws->nv->nv_screen->device;
 	struct nouveau_pipe_buffer *nvbuf;
 
-	nvbuf = calloc(1, sizeof(*nvbuf));
+	nvbuf = CALLOC_STRUCT(nouveau_pipe_buffer);
 	if (!nvbuf)
 		return NULL;
 	nvbuf->base.refcount = 1;
 	nvbuf->base.size = bytes;
 
 	if (nouveau_bo_user(dev, ptr, bytes, &nvbuf->bo)) {
-		free(nvbuf);
+		FREE(nvbuf);
 		return NULL;
 	}
 
@@ -109,7 +106,7 @@ nouveau_pipe_bo_del(struct pipe_winsys *ws, struct pipe_buffer *buf)
 	struct nouveau_pipe_buffer *nvbuf = nouveau_buffer(buf);
 
 	nouveau_bo_del(&nvbuf->bo);
-	free(nvbuf);
+	FREE(nvbuf);
 }
 
 static void *
@@ -124,6 +121,26 @@ nouveau_pipe_bo_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
 	if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
 		map_flags |= NOUVEAU_BO_WR;
 
+	/* XXX: Technically incorrect. If the client maps a buffer for write-only
+	 * and leaves part of the buffer untouched it probably expects those parts
+	 * to remain intact. This is violated because we allocate a whole new buffer
+	 * and don't copy the previous buffer's contents, so this optimization is
+	 * only valid if the client intends to overwrite the whole buffer.
+	 */
+	if ((map_flags & NOUVEAU_BO_RDWR) == NOUVEAU_BO_WR &&
+	    !nouveau_bo_busy(nvbuf->bo, map_flags)) {
+		struct nouveau_pipe_winsys *nvpws = (struct nouveau_pipe_winsys *)pws;
+		struct nouveau_context *nv = nvpws->nv;
+		struct nouveau_device *dev = nv->nv_screen->device;
+		struct nouveau_bo *rename;
+		uint32_t flags = nouveau_flags_from_usage(nv, buf->usage);
+
+		if (!nouveau_bo_new(dev, flags, buf->alignment, buf->size, &rename)) {
+			nouveau_bo_del(&nvbuf->bo);
+			nvbuf->bo = rename;
+		}
+	}
+
 	if (nouveau_bo_map(nvbuf->bo, map_flags))
 		return NULL;
 	return nvbuf->bo->map;
@@ -175,6 +192,12 @@ nouveau_pipe_fence_finish(struct pipe_winsys *ws,
 	return nouveau_fence_wait(&ref);
 }
 
+static void
+nouveau_destroy(struct pipe_winsys *pws)
+{
+	FREE(pws);
+}
+
 struct pipe_winsys *
 nouveau_create_pipe_winsys(struct nouveau_context *nv)
 {
@@ -200,7 +223,7 @@ nouveau_create_pipe_winsys(struct nouveau_context *nv)
 	pws->fence_finish = nouveau_pipe_fence_finish;
 
 	pws->get_name = nouveau_get_name;
+	pws->destroy = nouveau_destroy;
 
 	return &nvpws->pws;
 }
-
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_winsys_pipe.h b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.h
index 6a03ac0d77..14c728690d 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_winsys_pipe.h
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_pipe.h
@@ -31,4 +31,9 @@ nouveau_create_softpipe(struct nouveau_context *nv);
 struct pipe_context *
 nouveau_pipe_create(struct nouveau_context *nv);
 
+/* Must be provided by clients of common code */
+extern void
+nouveau_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surf,
+			  void *context_private);
+
 #endif
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_winsys_softpipe.c b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_softpipe.c
index 68aade829d..04def600f4 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_winsys_softpipe.c
+++ b/src/gallium/winsys/drm/nouveau/common/nouveau_winsys_softpipe.c
@@ -29,12 +29,12 @@
  * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
  */
 
-#include "imports.h"
-
-#include "pipe/p_defines.h"
-#include "pipe/p_format.h"
-#include "softpipe/sp_winsys.h"
-
+#include <pipe/p_winsys.h>
+#include <pipe/p_screen.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_format.h>
+#include <softpipe/sp_winsys.h>
+#include <util/u_memory.h>
 #include "nouveau_context.h"
 #include "nouveau_winsys_pipe.h"
 
@@ -48,7 +48,7 @@ struct nouveau_softpipe_winsys {
  */
 static boolean
 nouveau_is_format_supported(struct softpipe_winsys *sws,
-						enum pipe_format format)
+                            enum pipe_format format)
 {
 	switch (format) {
 	case PIPE_FORMAT_A8R8G8B8_UNORM:
@@ -68,19 +68,34 @@ nouveau_create_softpipe(struct nouveau_context *nv)
 	struct nouveau_softpipe_winsys *nvsws;
 	struct pipe_screen *pscreen;
 	struct pipe_winsys *ws;
+	struct pipe_context *pipe;
 
 	ws = nouveau_create_pipe_winsys(nv);
 	if (!ws)
 		return NULL;
 	pscreen = softpipe_create_screen(ws);
-
+	if (!pscreen) {
+		ws->destroy(ws);
+		return NULL;
+	}
 	nvsws = CALLOC_STRUCT(nouveau_softpipe_winsys);
-	if (!nvsws)
+	if (!nvsws) {
+		ws->destroy(ws);
+		pscreen->destroy(pscreen);
 		return NULL;
+	}
 
 	nvsws->sws.is_format_supported = nouveau_is_format_supported;
 	nvsws->nv = nv;
 
-	return softpipe_create(pscreen, ws, &nvsws->sws);
+	pipe = softpipe_create(pscreen, ws, &nvsws->sws);
+	if (!pipe) {
+		ws->destroy(ws);
+		pscreen->destroy(pscreen);
+		FREE(nvsws);
+		return NULL;
+	}
+
+	return pipe;
 }
 
diff --git a/src/gallium/winsys/drm/nouveau/nv04_surface.c b/src/gallium/winsys/drm/nouveau/common/nv04_surface.c
index 68338eb814..e9a8a2ac1c 100644
--- a/src/gallium/winsys/drm/nouveau/nv04_surface.c
+++ b/src/gallium/winsys/drm/nouveau/common/nv04_surface.c
@@ -82,6 +82,37 @@ nv04_scaled_image_format(enum pipe_format format)
 	}
 }
 
+static INLINE unsigned
+nv04_swizzle_bits(unsigned x, unsigned y)
+{
+	unsigned u = (x & 0x001) << 0 |
+	             (x & 0x002) << 1 |
+	             (x & 0x004) << 2 |
+	             (x & 0x008) << 3 |
+	             (x & 0x010) << 4 |
+	             (x & 0x020) << 5 |
+	             (x & 0x040) << 6 |
+	             (x & 0x080) << 7 |
+	             (x & 0x100) << 8 |
+	             (x & 0x200) << 9 |
+	             (x & 0x400) << 10 |
+	             (x & 0x800) << 11;
+
+	unsigned v = (y & 0x001) << 1 |
+	             (y & 0x002) << 2 |
+	             (y & 0x004) << 3 |
+	             (y & 0x008) << 4 |
+	             (y & 0x010) << 5 |
+	             (y & 0x020) << 6 |
+	             (y & 0x040) << 7 |
+	             (y & 0x080) << 8 |
+	             (y & 0x100) << 9 |
+	             (y & 0x200) << 10 |
+	             (y & 0x400) << 11 |
+	             (y & 0x800) << 12;
+	return v | u;
+}
+
 static void
 nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
                           unsigned sx, unsigned sy, unsigned w, unsigned h)
@@ -90,19 +121,23 @@ nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
 	struct pipe_surface *dst = nv->surf_dst;
 	struct pipe_surface *src = nv->surf_src;
 
+	const unsigned max_w = 1024;
+	const unsigned max_h = 1024;
+	const unsigned sub_w = w > max_w ? max_w : w;
+	const unsigned sub_h = h > max_h ? max_h : h;
+	unsigned cx = 0;
+	unsigned cy = 0;
+
 	/* POT or GTFO */
 	assert(!(w & (w - 1)) && !(h & (h - 1)));
 
 	BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1);
 	OUT_RELOCo(chan, nouveau_buffer(dst->buffer)->bo,
 	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
-
-	BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_FORMAT, 2);
+	BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_FORMAT, 1);
 	OUT_RING  (chan, nv04_surface_format(dst->format) |
 	                 log2i(w) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_U_SHIFT |
 	                 log2i(h) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_V_SHIFT);
-	OUT_RELOCl(chan, nouveau_buffer(dst->buffer)->bo, dst->offset,
-	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
 	BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_DMA_IMAGE, 1);
 	OUT_RELOCo(chan, nouveau_buffer(src->buffer)->bo,
@@ -110,24 +145,35 @@ nv04_surface_copy_swizzle(struct nouveau_context *nv, unsigned dx, unsigned dy,
 	BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SURFACE, 1);
 	OUT_RING  (chan, nv->nvc->NvSwzSurf->handle);
 
-	BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
-	OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION_TRUNCATE);
-	OUT_RING  (chan, nv04_scaled_image_format(src->format));
-	OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_OPERATION_SRCCOPY);
-	OUT_RING  (chan, 0);
-	OUT_RING  (chan, h << 16 | w);
-	OUT_RING  (chan, 0);
-	OUT_RING  (chan, h << 16 | w);
-	OUT_RING  (chan, 1 << 20);
-	OUT_RING  (chan, 1 << 20);
-	BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
-	OUT_RING  (chan, h << 16 | w);
-	OUT_RING  (chan, src->stride |
-	                 NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
-	                 NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
-	OUT_RELOCl(chan, nouveau_buffer(src->buffer)->bo, src->offset,
-	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
-	OUT_RING  (chan, 0);
+	for (cy = 0; cy < h; cy += sub_h) {
+		for (cx = 0; cx < w; cx += sub_w) {
+			BEGIN_RING(chan, nv->nvc->NvSwzSurf, NV04_SWIZZLED_SURFACE_OFFSET, 1);
+			OUT_RELOCl(chan, nouveau_buffer(dst->buffer)->bo,
+			                 dst->offset + nv04_swizzle_bits(cx, cy) * dst->block.size,
+			                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+			BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
+			OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION_TRUNCATE);
+			OUT_RING  (chan, nv04_scaled_image_format(src->format));
+			OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_OPERATION_SRCCOPY);
+			OUT_RING  (chan, 0);
+			OUT_RING  (chan, sub_h << 16 | sub_w);
+			OUT_RING  (chan, 0);
+			OUT_RING  (chan, sub_h << 16 | sub_w);
+			OUT_RING  (chan, 1 << 20);
+			OUT_RING  (chan, 1 << 20);
+
+			BEGIN_RING(chan, nv->nvc->NvSIFM, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
+			OUT_RING  (chan, sub_h << 16 | sub_w);
+			OUT_RING  (chan, src->stride |
+			                 NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
+			                 NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
+			OUT_RELOCl(chan, nouveau_buffer(src->buffer)->bo,
+			                 src->offset + cy * src->stride + cx * src->block.size,
+			                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+			OUT_RING  (chan, 0);
+		}
+	}
 }
 
 static void
diff --git a/src/gallium/winsys/drm/nouveau/nv50_surface.c b/src/gallium/winsys/drm/nouveau/common/nv50_surface.c
index c8ab7f690f..c8ab7f690f 100644
--- a/src/gallium/winsys/drm/nouveau/nv50_surface.c
+++ b/src/gallium/winsys/drm/nouveau/common/nv50_surface.c
diff --git a/src/gallium/winsys/drm/nouveau/dri/Makefile b/src/gallium/winsys/drm/nouveau/dri/Makefile
new file mode 100644
index 0000000000..e129e42e97
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/dri/Makefile
@@ -0,0 +1,31 @@
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = nouveau_dri.so
+
+MINIGLX_SOURCES =
+
+PIPE_DRIVERS = \
+	$(TOP)/src/gallium/drivers/softpipe/libsoftpipe.a \
+	$(TOP)/src/gallium/drivers/nv04/libnv04.a \
+	$(TOP)/src/gallium/drivers/nv10/libnv10.a \
+	$(TOP)/src/gallium/drivers/nv20/libnv20.a \
+	$(TOP)/src/gallium/drivers/nv30/libnv30.a \
+	$(TOP)/src/gallium/drivers/nv40/libnv40.a \
+	$(TOP)/src/gallium/drivers/nv50/libnv50.a
+
+DRIVER_SOURCES = \
+	nouveau_context_dri.c \
+	nouveau_screen_dri.c \
+	nouveau_swapbuffers.c \
+	../common/libnouveaudrm.a
+
+C_SOURCES = \
+	$(COMMON_GALLIUM_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../../Makefile.template
+
+symlinks:
diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c
new file mode 100644
index 0000000000..006978b182
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.c
@@ -0,0 +1,124 @@
+#include <main/glheader.h>
+#include <glapi/glthread.h>
+#include <GL/internal/glcore.h>
+#include <utils.h>
+
+#include <state_tracker/st_public.h>
+#include <state_tracker/st_context.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+
+#include "../common/nouveau_winsys_pipe.h"
+#include "../common/nouveau_dri.h"
+#include "../common/nouveau_local.h"
+#include "nouveau_context_dri.h"
+#include "nouveau_screen_dri.h"
+
+#ifdef DEBUG
+static const struct dri_debug_control debug_control[] = {
+	{ "bo", DEBUG_BO },
+	{ NULL, 0 }
+};
+int __nouveau_debug = 0;
+#endif
+
+GLboolean
+nouveau_context_create(const __GLcontextModes *glVis,
+		       __DRIcontextPrivate *driContextPriv,
+		       void *sharedContextPrivate)
+{
+	__DRIscreenPrivate *driScrnPriv = driContextPriv->driScreenPriv;
+	struct nouveau_screen_dri  *nv_screen = driScrnPriv->private;
+	struct nouveau_context_dri *nv = CALLOC_STRUCT(nouveau_context_dri);
+	struct st_context *st_share = NULL;
+	struct nouveau_context_dri *nv_share = NULL;
+	struct pipe_context *pipe;
+
+	if (sharedContextPrivate) {
+		st_share = ((struct nouveau_context_dri *)sharedContextPrivate)->st;
+		nv_share = st_share->pipe->priv;
+	}
+
+	if (nouveau_context_init(&nv_screen->base, driContextPriv->hHWContext,
+	                         (drmLock *)&driScrnPriv->pSAREA->lock,
+	                         nv_share, &nv->base)) {
+		return GL_FALSE;
+	}
+
+	pipe = nv->base.nvc->pctx[nv->base.pctx_id];
+	driContextPriv->driverPrivate = (void *)nv;
+	//nv->nv_screen  = nv_screen;
+	nv->dri_screen = driScrnPriv;
+
+	driParseConfigFiles(&nv->dri_option_cache, &nv_screen->option_cache,
+			    nv->dri_screen->myNum, "nouveau");
+#ifdef DEBUG
+	__nouveau_debug = driParseDebugString(getenv("NOUVEAU_DEBUG"),
+					      debug_control);
+#endif
+
+	nv->st = st_create_context(pipe, glVis, st_share);
+	return GL_TRUE;
+}
+
+void
+nouveau_context_destroy(__DRIcontextPrivate *driContextPriv)
+{
+	struct nouveau_context_dri *nv = driContextPriv->driverPrivate;
+
+	assert(nv);
+
+	st_finish(nv->st);
+	st_destroy_context(nv->st);
+
+	nouveau_context_cleanup(&nv->base);
+
+	FREE(nv);
+}
+
+GLboolean
+nouveau_context_bind(__DRIcontextPrivate *driContextPriv,
+		     __DRIdrawablePrivate *driDrawPriv,
+		     __DRIdrawablePrivate *driReadPriv)
+{
+	struct nouveau_context_dri *nv;
+	struct nouveau_framebuffer *draw, *read;
+
+	if (!driContextPriv) {
+		st_make_current(NULL, NULL, NULL);
+		return GL_TRUE;
+	}
+
+	nv = driContextPriv->driverPrivate;
+	draw = driDrawPriv->driverPrivate;
+	read = driReadPriv->driverPrivate;
+
+	st_make_current(nv->st, draw->stfb, read->stfb);
+
+	if ((nv->dri_drawable != driDrawPriv) ||
+	    (nv->last_stamp != driDrawPriv->lastStamp)) {
+		nv->dri_drawable = driDrawPriv;
+		st_resize_framebuffer(draw->stfb, driDrawPriv->w,
+				      driDrawPriv->h);
+		nv->last_stamp = driDrawPriv->lastStamp;
+	}
+
+	if (driDrawPriv != driReadPriv) {
+		st_resize_framebuffer(read->stfb, driReadPriv->w,
+				      driReadPriv->h);
+	}
+
+	return GL_TRUE;
+}
+
+GLboolean
+nouveau_context_unbind(__DRIcontextPrivate *driContextPriv)
+{
+	struct nouveau_context_dri *nv = driContextPriv->driverPrivate;
+	(void)nv;
+
+	st_flush(nv->st, 0, NULL);
+	return GL_TRUE;
+}
+
diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.h
new file mode 100644
index 0000000000..8257790d47
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_context_dri.h
@@ -0,0 +1,49 @@
+#ifndef __NOUVEAU_CONTEXT_DRI_H__
+#define __NOUVEAU_CONTEXT_DRI_H__
+
+#include <dri_util.h>
+#include <xmlconfig.h>
+#include <nouveau/nouveau_winsys.h>
+#include "../common/nouveau_context.h"
+#include "../common/nouveau_drmif.h"
+#include "../common/nouveau_dma.h"
+
+struct nouveau_framebuffer {
+	struct st_framebuffer *stfb;
+};
+
+struct nouveau_context_dri {
+	struct nouveau_context base;
+	struct st_context *st;
+
+	/* DRI stuff */
+	__DRIscreenPrivate    *dri_screen;
+	__DRIdrawablePrivate  *dri_drawable;
+	unsigned int           last_stamp;
+	driOptionCache         dri_option_cache;
+	drm_context_t          drm_context;
+	drmLock                drm_lock;
+};
+
+extern GLboolean nouveau_context_create(const __GLcontextModes *,
+					__DRIcontextPrivate *, void *);
+extern void nouveau_context_destroy(__DRIcontextPrivate *);
+extern GLboolean nouveau_context_bind(__DRIcontextPrivate *,
+				      __DRIdrawablePrivate *draw,
+				      __DRIdrawablePrivate *read);
+extern GLboolean nouveau_context_unbind(__DRIcontextPrivate *);
+
+#ifdef DEBUG
+extern int __nouveau_debug;
+
+#define DEBUG_BO (1 << 0)
+
+#define DBG(flag, ...) do {                   \
+	if (__nouveau_debug & (DEBUG_##flag)) \
+		NOUVEAU_ERR(__VA_ARGS__);     \
+} while(0)
+#else
+#define DBG(flag, ...)
+#endif
+
+#endif
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_screen.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen_dri.c
index c6d0c53588..1d7c92376f 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_screen.c
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen_dri.c
@@ -1,16 +1,15 @@
-#include "utils.h"
-#include "vblank.h"
-#include "xmlpool.h"
-
-#include "pipe/p_context.h"
-#include "state_tracker/st_public.h"
-#include "state_tracker/st_cb_fbo.h"
-
-#include "nouveau_context.h"
-#include "nouveau_drm.h"
-#include "nouveau_dri.h"
-#include "nouveau_local.h"
-#include "nouveau_screen.h"
+#include <utils.h>
+#include <vblank.h>
+#include <xmlpool.h>
+
+#include <pipe/p_context.h>
+#include <state_tracker/st_public.h>
+#include <state_tracker/st_cb_fbo.h>
+#include <nouveau_drm.h>
+#include "../common/nouveau_dri.h"
+#include "../common/nouveau_local.h"
+#include "nouveau_context_dri.h"
+#include "nouveau_screen_dri.h"
 #include "nouveau_swapbuffers.h"
 
 #if NOUVEAU_DRM_HEADER_PATCHLEVEL != 11
@@ -183,13 +182,12 @@ static const __DRIconfig **
 nouveau_screen_create(__DRIscreenPrivate *psp)
 {
 	struct nouveau_dri *nv_dri = psp->pDevPriv;
-	struct nouveau_screen *nv_screen;
+	struct nouveau_screen_dri *nv_screen;
 	static const __DRIversion ddx_expected =
 		{ 0, 0, NOUVEAU_DRM_HEADER_PATCHLEVEL };
 	static const __DRIversion dri_expected = { 4, 0, 0 };
 	static const __DRIversion drm_expected =
 		{ 0, 0, NOUVEAU_DRM_HEADER_PATCHLEVEL };
-	int ret;
 
 	if (!driCheckDriDdxDrmVersions2("nouveau",
 					&psp->dri_version, &dri_expected,
@@ -209,28 +207,23 @@ nouveau_screen_create(__DRIscreenPrivate *psp)
 
 	if (psp->devPrivSize != sizeof(struct nouveau_dri)) {
 		NOUVEAU_ERR("DRI struct mismatch between DDX/DRI\n");
-		return GL_FALSE;
+		return NULL;
 	}
 
-	nv_screen = CALLOC_STRUCT(nouveau_screen);
+	nv_screen = CALLOC_STRUCT(nouveau_screen_dri);
 	if (!nv_screen)
-		return GL_FALSE;
-	nv_screen->driScrnPriv = psp;
-	psp->private = (void *)nv_screen;
+		return NULL;
 
 	driParseOptionInfo(&nv_screen->option_cache,
 			   __driConfigOptions, __driNConfigOptions);
 
-	if ((ret = nouveau_device_open_existing(&nv_screen->device, 0,
-						psp->fd, 0))) {
-		NOUVEAU_ERR("Failed opening nouveau device: %d\n", ret);
-		return GL_FALSE;
+	if (nouveau_screen_init(nv_dri, psp->fd, &nv_screen->base)) {
+		FREE(nv_screen);
+		return NULL;
 	}
 
-	nv_screen->front_offset = nv_dri->front_offset;
-	nv_screen->front_pitch  = nv_dri->front_pitch * (nv_dri->bpp / 8);
-	nv_screen->front_cpp = nv_dri->bpp / 8;
-	nv_screen->front_height = nv_dri->height;
+	nv_screen->driScrnPriv = psp;
+	psp->private = (void *)nv_screen;
 
 	return (const __DRIconfig **)
 		nouveau_fill_in_modes(psp, nv_dri->bpp,
@@ -241,9 +234,10 @@ nouveau_screen_create(__DRIscreenPrivate *psp)
 static void
 nouveau_screen_destroy(__DRIscreenPrivate *driScrnPriv)
 {
-	struct nouveau_screen *nv_screen = driScrnPriv->private;
+	struct nouveau_screen_dri *nv_screen = driScrnPriv->private;
 
 	driScrnPriv->private = NULL;
+	nouveau_screen_cleanup(&nv_screen->base);
 	FREE(nv_screen);
 }
 
diff --git a/src/gallium/winsys/drm/nouveau/dri/nouveau_screen_dri.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen_dri.h
new file mode 100644
index 0000000000..1498087819
--- /dev/null
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_screen_dri.h
@@ -0,0 +1,13 @@
+#ifndef __NOUVEAU_SCREEN_DRI_H__
+#define __NOUVEAU_SCREEN_DRI_H__
+
+#include "../common/nouveau_screen.h"
+#include "xmlconfig.h"
+
+struct nouveau_screen_dri {
+	struct nouveau_screen base;
+	__DRIscreenPrivate *driScrnPriv;
+	driOptionCache      option_cache;
+};
+
+#endif
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_swapbuffers.c b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.c
index 70e0104e83..38461b2b0c 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_swapbuffers.c
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.c
@@ -1,34 +1,34 @@
-#include "main/glheader.h"
-#include "glapi/glthread.h"
+#include <main/glheader.h>
+#include <glapi/glthread.h>
 #include <GL/internal/glcore.h>
 
-#include "pipe/p_context.h"
-#include "state_tracker/st_public.h"
-#include "state_tracker/st_context.h"
-#include "state_tracker/st_cb_fbo.h"
+#include <pipe/p_context.h>
+#include <state_tracker/st_public.h>
+#include <state_tracker/st_context.h>
+#include <state_tracker/st_cb_fbo.h>
 
-#include "nouveau_context.h"
-#include "nouveau_local.h"
-#include "nouveau_screen.h"
+#include "../common/nouveau_local.h"
+#include "nouveau_context_dri.h"
+#include "nouveau_screen_dri.h"
 #include "nouveau_swapbuffers.h"
 
 void
 nouveau_copy_buffer(__DRIdrawablePrivate *dPriv, struct pipe_surface *surf,
 		    const drm_clip_rect_t *rect)
 {
-	struct nouveau_context *nv = dPriv->driContextPriv->driverPrivate;
+	struct nouveau_context_dri *nv = dPriv->driContextPriv->driverPrivate;
 	drm_clip_rect_t *pbox;
 	int nbox, i;
 
-	LOCK_HARDWARE(nv);
+	LOCK_HARDWARE(&nv->base);
 	if (!dPriv->numClipRects) {
-		UNLOCK_HARDWARE(nv);
+		UNLOCK_HARDWARE(&nv->base);
 		return;
 	}
 	pbox = dPriv->pClipRects;
 	nbox = dPriv->numClipRects;
 
-	nv->surface_copy_prep(nv, nv->frontbuffer, surf);
+	nv->base.surface_copy_prep(&nv->base, nv->base.frontbuffer, surf);
 	for (i = 0; i < nbox; i++, pbox++) {
 		int sx, sy, dx, dy, w, h;
 
@@ -39,11 +39,11 @@ nouveau_copy_buffer(__DRIdrawablePrivate *dPriv, struct pipe_surface *surf,
 		w  = pbox->x2 - pbox->x1;
 		h  = pbox->y2 - pbox->y1;
 
-		nv->surface_copy(nv, dx, dy, sx, sy, w, h);
+		nv->base.surface_copy(&nv->base, dx, dy, sx, sy, w, h);
 	}
 
-	FIRE_RING(nv->nvc->channel);
-	UNLOCK_HARDWARE(nv);
+	FIRE_RING(nv->base.nvc->channel);
+	UNLOCK_HARDWARE(&nv->base);
 
 	if (nv->last_stamp != dPriv->lastStamp) {
 		struct nouveau_framebuffer *nvfb = dPriv->driverPrivate;
@@ -84,3 +84,29 @@ nouveau_swap_buffers(__DRIdrawablePrivate *dPriv)
 	}
 }
 
+void
+nouveau_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surf,
+			  void *context_private)
+{
+	struct nouveau_context_dri *nv = context_private;
+	__DRIdrawablePrivate *dPriv = nv->dri_drawable;
+
+	nouveau_copy_buffer(dPriv, surf, NULL);
+}
+
+void
+nouveau_contended_lock(struct nouveau_context *nv)
+{
+	struct nouveau_context_dri *nv_sub = (struct nouveau_context_dri*)nv;
+	__DRIdrawablePrivate *dPriv = nv_sub->dri_drawable;
+	__DRIscreenPrivate *sPriv = nv_sub->dri_screen;
+
+	/* If the window moved, may need to set a new cliprect now.
+	 *
+	 * NOTE: This releases and regains the hw lock, so all state
+	 * checking must be done *after* this call:
+	 */
+	if (dPriv)
+		DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);
+}
+
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_swapbuffers.h b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.h
index 825d3da6da..825d3da6da 100644
--- a/src/gallium/winsys/drm/nouveau/nouveau_swapbuffers.h
+++ b/src/gallium/winsys/drm/nouveau/dri/nouveau_swapbuffers.h
diff --git a/src/gallium/winsys/drm/nouveau/nouveau_screen.h b/src/gallium/winsys/drm/nouveau/nouveau_screen.h
deleted file mode 100644
index 388d6be9bb..0000000000
--- a/src/gallium/winsys/drm/nouveau/nouveau_screen.h
+++ /dev/null
@@ -1,20 +0,0 @@
-#ifndef __NOUVEAU_SCREEN_H__
-#define __NOUVEAU_SCREEN_H__
-
-#include "xmlconfig.h"
-
-struct nouveau_screen {
-	__DRIscreenPrivate *driScrnPriv;
-	driOptionCache      option_cache;
-
-	struct nouveau_device *device;
-
-	uint32_t front_offset;
-	uint32_t front_pitch;
-	uint32_t front_cpp;
-	uint32_t front_height;
-
-	void *nvc;
-};
-
-#endif
diff --git a/src/gallium/winsys/g3dvl/nouveau/Makefile b/src/gallium/winsys/g3dvl/nouveau/Makefile
index ff43327778..22d925b643 100644
--- a/src/gallium/winsys/g3dvl/nouveau/Makefile
+++ b/src/gallium/winsys/g3dvl/nouveau/Makefile
@@ -3,35 +3,33 @@ GALLIUMDIR	= ../../..
 DRMDIR		?= /usr
 DRIDIR		= ../../../../driclient
 
-OBJECTS		= nouveau_bo.o nouveau_fence.o nouveau_swapbuffers.o nouveau_channel.o		\
-		  nouveau_grobj.o nouveau_context.o nouveau_winsys.o nouveau_lock.o		\
-		  nouveau_winsys_pipe.o nouveau_device.o nouveau_notifier.o nouveau_dma.o	\
-		  nouveau_pushbuf.o nouveau_resource.o nouveau_screen.o nv04_surface.o		\
-		  nv50_surface.o #nouveau_winsys_softpipe.o
-
-CFLAGS		+= -g -Wall -fPIC			\
-		   -I${GALLIUMDIR}/include		\
-		   -I${GALLIUMDIR}/winsys/g3dvl		\
-		   -I${DRMDIR}/include			\
-		   -I${DRMDIR}/include/drm		\
-		   -I${GALLIUMDIR}/drivers		\
-		   -I${GALLIUMDIR}/auxiliary		\
+OBJECTS		= nouveau_screen_vl.o nouveau_context_vl.o nouveau_swapbuffers.o
+
+CFLAGS		+= -g -Wall -Werror=implicit-function-declaration -fPIC	\
+		   -I${GALLIUMDIR}/include				\
+		   -I${GALLIUMDIR}/winsys/g3dvl				\
+		   -I${GALLIUMDIR}/winsys/drm/nouveau			\
+		   -I${DRMDIR}/include					\
+		   -I${DRMDIR}/include/drm				\
+		   -I${GALLIUMDIR}/drivers				\
+		   -I${GALLIUMDIR}/auxiliary				\
 		   -I${DRIDIR}/include
 
-LDFLAGS		+= -L${DRMDIR}/lib			\
-		   -L${DRIDIR}/lib			\
-		   -L${GALLIUMDIR}/auxiliary/draw	\
-		   -L${GALLIUMDIR}/auxiliary/tgsi	\
-		   -L${GALLIUMDIR}/auxiliary/translate	\
-		   -L${GALLIUMDIR}/auxiliary/rtasm	\
-		   -L${GALLIUMDIR}/auxiliary/cso_cache	\
-		   -L${GALLIUMDIR}/drivers/nv10		\
-		   -L${GALLIUMDIR}/drivers/nv20		\
-		   -L${GALLIUMDIR}/drivers/nv30		\
-		   -L${GALLIUMDIR}/drivers/nv40		\
+LDFLAGS		+= -L${DRMDIR}/lib				\
+		   -L${DRIDIR}/lib				\
+		   -L${GALLIUMDIR}/winsys/drm/nouveau/common	\
+		   -L${GALLIUMDIR}/auxiliary/draw		\
+		   -L${GALLIUMDIR}/auxiliary/tgsi		\
+		   -L${GALLIUMDIR}/auxiliary/translate		\
+		   -L${GALLIUMDIR}/auxiliary/rtasm		\
+		   -L${GALLIUMDIR}/auxiliary/cso_cache		\
+		   -L${GALLIUMDIR}/drivers/nv10			\
+		   -L${GALLIUMDIR}/drivers/nv20			\
+		   -L${GALLIUMDIR}/drivers/nv30			\
+		   -L${GALLIUMDIR}/drivers/nv40			\
 		   -L${GALLIUMDIR}/drivers/nv50
 
-LIBS		+= -ldriclient -ldrm -lnv10 -lnv20 -lnv30 -lnv40 -lnv50 -ldraw -ltgsi -ltranslate -lrtasm -lcso_cache -lm
+LIBS		+= -lnouveaudrm -ldriclient -ldrm -lnv10 -lnv20 -lnv30 -lnv40 -lnv50 -ldraw -ltgsi -ltranslate -lrtasm -lcso_cache -lm
 
 #############################################
 
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_context.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_context.c
deleted file mode 100644
index 06a61fcda3..0000000000
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_context.c
+++ /dev/null
@@ -1,370 +0,0 @@
-#include "pipe/p_defines.h"
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-#include "util/u_memory.h"
-
-#include "nouveau_context.h"
-#include "nouveau_dri.h"
-#include "nouveau_local.h"
-#include "nouveau_screen.h"
-#include "nouveau_winsys_pipe.h"
-
-/*
-#ifdef DEBUG
-static const struct dri_debug_control debug_control[] = {
-	{ "bo", DEBUG_BO },
-	{ NULL, 0 }
-};
-int __nouveau_debug = 0;
-#endif
-*/
-
-/*
- * TODO: Re-examine dri_screen, dri_context, nouveau_screen, nouveau_context
- * relationships, seems like there is a lot of room for simplification there.
- */
-
-static void
-nouveau_channel_context_destroy(struct nouveau_channel_context *nvc)
-{
-	nouveau_grobj_free(&nvc->NvCtxSurf2D);
-	nouveau_grobj_free(&nvc->NvImageBlit);
-	nouveau_grobj_free(&nvc->NvGdiRect);
-	nouveau_grobj_free(&nvc->NvM2MF);
-	nouveau_grobj_free(&nvc->Nv2D);
-	nouveau_grobj_free(&nvc->NvSwzSurf);
-	nouveau_grobj_free(&nvc->NvSIFM);
-
-	nouveau_notifier_free(&nvc->sync_notifier);
-
-	nouveau_channel_free(&nvc->channel);
-
-	FREE(nvc);
-}
-
-static struct nouveau_channel_context *
-nouveau_channel_context_create(struct nouveau_device *dev)
-{
-	struct nouveau_channel_context *nvc;
-	int ret;
-
-	nvc = CALLOC_STRUCT(nouveau_channel_context);
-	if (!nvc)
-		return NULL;
-
-	if ((ret = nouveau_channel_alloc(dev, 0x8003d001, 0x8003d002,
-					 &nvc->channel))) {
-		NOUVEAU_ERR("Error creating GPU channel: %d\n", ret);
-		nouveau_channel_context_destroy(nvc);
-		return NULL;
-	}
-
-	nvc->next_handle = 0x80000000;
-
-	if ((ret = nouveau_notifier_alloc(nvc->channel, nvc->next_handle++, 1,
-					  &nvc->sync_notifier))) {
-		NOUVEAU_ERR("Error creating channel sync notifier: %d\n", ret);
-		nouveau_channel_context_destroy(nvc);
-		return NULL;
-	}
-
-	switch (dev->chipset & 0xf0) {
-	case 0x50:
-	case 0x80:
-	case 0x90:
-		ret = nouveau_surface_channel_create_nv50(nvc);
-		break;
-	default:
-		ret = nouveau_surface_channel_create_nv04(nvc);
-		break;
-	}
-
-	if (ret) {
-		NOUVEAU_ERR("Error initialising surface objects: %d\n", ret);
-		nouveau_channel_context_destroy(nvc);
-		return NULL;
-	}
-
-	return nvc;
-}
-
-int
-nouveau_context_create(dri_context_t *dri_context)
-{
-	dri_screen_t			*dri_screen = dri_context->dri_screen;
-	struct nouveau_screen 		*nv_screen = dri_screen->private;
-	struct nouveau_context		*nv = CALLOC_STRUCT(nouveau_context);
-	struct pipe_context		*pipe = NULL;
-	struct nouveau_channel_context	*nvc = NULL;
-	struct nouveau_device		*dev = nv_screen->device;
-	int				i;
-
-	switch (dev->chipset & 0xf0) {
-	case 0x10:
-	case 0x20:
-		/* NV10 */
-	case 0x30:
-		/* NV30 */
-	case 0x40:
-	case 0x60:
-		/* NV40 */
-	case 0x50:
-	case 0x80:
-	case 0x90:
-		/* G80 */
-		break;
-	default:
-		NOUVEAU_ERR("Unsupported chipset: NV%02x\n", dev->chipset);
-		return 1;
-	}
-
-	dri_context->private = (void*)nv;
-	nv->dri_context = dri_context;
-	nv->nv_screen  = nv_screen;
-
-	{
-		struct nouveau_device_priv *nvdev = nouveau_device(dev);
-
-		nvdev->ctx  = dri_context->drm_context;
-		nvdev->lock = (drmLock*)&dri_screen->sarea->lock;
-	}
-
-	/*
-	driParseConfigFiles(&nv->dri_option_cache, &nv_screen->option_cache,
-			    nv->dri_screen->myNum, "nouveau");
-#ifdef DEBUG
-	__nouveau_debug = driParseDebugString(getenv("NOUVEAU_DEBUG"),
-					      debug_control);
-#endif
-	*/
-
-	/*XXX: Hack up a fake region and buffer object for front buffer.
-	 *     This will go away with TTM, replaced with a simple reference
-	 *     of the front buffer handle passed to us by the DDX.
-	 */
-	{
-		struct pipe_surface *fb_surf;
-		struct nouveau_pipe_buffer *fb_buf;
-		struct nouveau_bo_priv *fb_bo;
-
-		fb_bo = calloc(1, sizeof(struct nouveau_bo_priv));
-		fb_bo->drm.offset = nv_screen->front_offset;
-		fb_bo->drm.flags = NOUVEAU_MEM_FB;
-		fb_bo->drm.size = nv_screen->front_pitch *
-				  nv_screen->front_height;
-		fb_bo->refcount = 1;
-		fb_bo->base.flags = NOUVEAU_BO_PIN | NOUVEAU_BO_VRAM;
-		fb_bo->base.offset = fb_bo->drm.offset;
-		fb_bo->base.handle = (unsigned long)fb_bo;
-		fb_bo->base.size = fb_bo->drm.size;
-		fb_bo->base.device = nv_screen->device;
-
-		fb_buf = calloc(1, sizeof(struct nouveau_pipe_buffer));
-		fb_buf->bo = &fb_bo->base;
-
-		fb_surf = calloc(1, sizeof(struct pipe_surface));
-		if (nv_screen->front_cpp == 2)
-			fb_surf->format = PIPE_FORMAT_R5G6B5_UNORM;
-		else
-			fb_surf->format = PIPE_FORMAT_A8R8G8B8_UNORM;
-		pf_get_block(fb_surf->format, &fb_surf->block);
-		fb_surf->width = nv_screen->front_pitch / nv_screen->front_cpp;
-		fb_surf->height = nv_screen->front_height;
-		fb_surf->stride = fb_surf->width * fb_surf->block.size;
-		fb_surf->refcount = 1;
-		fb_surf->buffer = &fb_buf->base;
-
-		nv->frontbuffer = fb_surf;
-	}
-
-	nvc = nv_screen->nvc;
-
-	if (!nvc) {
-		nvc = nouveau_channel_context_create(dev);
-		if (!nvc) {
-			NOUVEAU_ERR("Failed initialising GPU context\n");
-			return 1;
-		}
-		nv_screen->nvc = nvc;
-	}
-
-	nvc->refcount++;
-	nv->nvc = nvc;
-
-	/* Find a free slot for a pipe context, allocate a new one if needed */
-	nv->pctx_id = -1;
-	for (i = 0; i < nvc->nr_pctx; i++) {
-		if (nvc->pctx[i] == NULL) {
-			nv->pctx_id = i;
-			break;
-		}
-	}
-
-	if (nv->pctx_id < 0) {
-		nv->pctx_id = nvc->nr_pctx++;
-		nvc->pctx =
-			realloc(nvc->pctx,
-				sizeof(struct pipe_context *) * nvc->nr_pctx);
-	}
-
-	/* Create pipe */
-	switch (dev->chipset & 0xf0) {
-	case 0x50:
-	case 0x80:
-	case 0x90:
-		if (nouveau_surface_init_nv50(nv))
-			return 1;
-		break;
-	default:
-		if (nouveau_surface_init_nv04(nv))
-			return 1;
-		break;
-	}
-
-	if (!getenv("NOUVEAU_FORCE_SOFTPIPE")) {
-		struct pipe_screen *pscreen;
-
-		pipe = nouveau_pipe_create(nv);
-		if (!pipe)
-			NOUVEAU_ERR("Couldn't create hw pipe\n");
-		pscreen = nvc->pscreen;
-
-		nv->cap.hw_vertex_buffer =
-			pscreen->get_param(pscreen, NOUVEAU_CAP_HW_VTXBUF);
-		nv->cap.hw_index_buffer =
-			pscreen->get_param(pscreen, NOUVEAU_CAP_HW_IDXBUF);
-	}
-
-	/* XXX: nouveau_winsys_softpipe needs a mesa header removed before we can compile it. */
-	/*
-	if (!pipe) {
-		NOUVEAU_MSG("Using softpipe\n");
-		pipe = nouveau_create_softpipe(nv);
-		if (!pipe) {
-			NOUVEAU_ERR("Error creating pipe, bailing\n");
-			return 1;
-		}
-	}
-	*/
-	if (!pipe) {
-		NOUVEAU_ERR("Error creating pipe, bailing\n");
-		return 1;
-	}
-
-	pipe->priv = nv;
-
-	return 0;
-}
-
-void
-nouveau_context_destroy(dri_context_t *dri_context)
-{
-	struct nouveau_context *nv = dri_context->private;
-	struct nouveau_channel_context *nvc = nv->nvc;
-
-	assert(nv);
-
-	if (nv->pctx_id >= 0) {
-		nvc->pctx[nv->pctx_id] = NULL;
-		if (--nvc->refcount <= 0) {
-			nouveau_channel_context_destroy(nvc);
-			nv->nv_screen->nvc = NULL;
-		}
-	}
-
-	free(nv);
-}
-
-int
-nouveau_context_bind(struct nouveau_context *nv, dri_drawable_t *dri_drawable)
-{
-	assert(nv);
-	assert(dri_drawable);
-
-	if (nv->dri_drawable != dri_drawable)
-	{
-		nv->dri_drawable = dri_drawable;
-		dri_drawable->private = nv;
-	}
-
-	return 0;
-}
-
-int
-nouveau_context_unbind(struct nouveau_context *nv)
-{
-	assert(nv);
-
-	nv->dri_drawable = NULL;
-
-	return 0;
-}
-
-/* Show starts here */
-
-int bind_pipe_drawable(struct pipe_context *pipe, Drawable drawable)
-{
-	struct nouveau_context	*nv;
-	dri_drawable_t		*dri_drawable;
-
-	nv = pipe->priv;
-
-	driCreateDrawable(nv->nv_screen->dri_screen, drawable, &dri_drawable);
-
-	nouveau_context_bind(nv, dri_drawable);
-
-	return 0;
-}
-
-int unbind_pipe_drawable(struct pipe_context *pipe)
-{
-	nouveau_context_unbind(pipe->priv);
-
-	return 0;
-}
-
-struct pipe_context* create_pipe_context(Display *display, int screen)
-{
-	dri_screen_t		*dri_screen;
-	dri_framebuffer_t	dri_framebuf;
-	dri_context_t		*dri_context;
-	struct nouveau_context	*nv;
-
-	driCreateScreen(display, screen, &dri_screen, &dri_framebuf);
-	driCreateContext(dri_screen, XDefaultVisual(display, screen), &dri_context);
-
-	nouveau_screen_create(dri_screen, &dri_framebuf);
-	nouveau_context_create(dri_context);
-
-	nv = dri_context->private;
-
-	return nv->nvc->pctx[nv->pctx_id];
-}
-
-int destroy_pipe_context(struct pipe_context *pipe)
-{
-	struct pipe_screen	*screen;
-	struct pipe_winsys	*winsys;
-	struct nouveau_context	*nv;
-	dri_screen_t		*dri_screen;
-	dri_context_t		*dri_context;
-
-	assert(pipe);
-
-	screen = pipe->screen;
-	winsys = pipe->winsys;
-	nv = pipe->priv;
-	dri_context = nv->dri_context;
-	dri_screen = dri_context->dri_screen;
-
-	pipe->destroy(pipe);
-	screen->destroy(screen);
-	free(winsys);
-
-	nouveau_context_destroy(dri_context);
-	nouveau_screen_destroy(dri_screen);
-	driDestroyContext(dri_context);
-	driDestroyScreen(dri_screen);
-
-	return 0;
-}
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_context.h b/src/gallium/winsys/g3dvl/nouveau/nouveau_context.h
deleted file mode 100644
index 395a3ab790..0000000000
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_context.h
+++ /dev/null
@@ -1,105 +0,0 @@
-#ifndef __NOUVEAU_CONTEXT_H__
-#define __NOUVEAU_CONTEXT_H__
-
-/*#include "xmlconfig.h"*/
-
-#include <driclient.h>
-#include "nouveau/nouveau_winsys.h"
-#include "nouveau_drmif.h"
-#include "nouveau_dma.h"
-
-struct nouveau_channel_context {
-	struct pipe_screen *pscreen;
-	int refcount;
-
-	unsigned cur_pctx;
-	unsigned nr_pctx;
-	struct pipe_context **pctx;
-
-	struct nouveau_channel  *channel;
-
-	struct nouveau_notifier *sync_notifier;
-
-	/* Common */
-	struct nouveau_grobj    *NvM2MF;
-	/* NV04-NV40 */
-	struct nouveau_grobj    *NvCtxSurf2D;
-	struct nouveau_grobj	*NvSwzSurf;
-	struct nouveau_grobj    *NvImageBlit;
-	struct nouveau_grobj    *NvGdiRect;
-	struct nouveau_grobj	*NvSIFM;
-	/* G80 */
-	struct nouveau_grobj    *Nv2D;
-
-	uint32_t                 next_handle;
-	uint32_t                 next_subchannel;
-	uint32_t                 next_sequence;
-};
-
-struct nouveau_context {
-	/* DRI stuff */
-	dri_context_t		*dri_context;
-	dri_drawable_t		*dri_drawable;
-	unsigned int		last_stamp;
-	/*driOptionCache	dri_option_cache;*/
-	drm_context_t		drm_context;
-	drmLock			drm_lock;
-	int			locked;
-	struct nouveau_screen	*nv_screen;
-	struct pipe_surface	*frontbuffer;
-
-	struct {
-		int hw_vertex_buffer;
-		int hw_index_buffer;
-	} cap;
-
-	/* Hardware context */
-	struct nouveau_channel_context	*nvc;
-	int				pctx_id;
-
-	/* pipe_surface accel */
-	struct pipe_surface		*surf_src, *surf_dst;
-	unsigned			surf_src_offset, surf_dst_offset;
-	
-	int  (*surface_copy_prep)(struct nouveau_context *,
-				  struct pipe_surface *dst,
-				  struct pipe_surface *src);
-	void (*surface_copy)(struct nouveau_context *, unsigned dx, unsigned dy,
-			     unsigned sx, unsigned sy, unsigned w, unsigned h);
-	void (*surface_copy_done)(struct nouveau_context *);
-	int (*surface_fill)(struct nouveau_context *, struct pipe_surface *,
-			    unsigned, unsigned, unsigned, unsigned, unsigned);
-};
-
-extern int nouveau_context_create(dri_context_t *);
-extern void nouveau_context_destroy(dri_context_t *);
-extern int nouveau_context_bind(struct nouveau_context *, dri_drawable_t *);
-extern int nouveau_context_unbind(struct nouveau_context *);
-
-#ifdef DEBUG
-extern int __nouveau_debug;
-
-#define DEBUG_BO (1 << 0)
-
-#define DBG(flag, ...) do {                   \
-	if (__nouveau_debug & (DEBUG_##flag)) \
-		NOUVEAU_ERR(__VA_ARGS__);     \
-} while(0)
-#else
-#define DBG(flag, ...)
-#endif
-
-extern void LOCK_HARDWARE(struct nouveau_context *);
-extern void UNLOCK_HARDWARE(struct nouveau_context *);
-
-extern int
-nouveau_surface_channel_create_nv04(struct nouveau_channel_context *);
-extern int
-nouveau_surface_channel_create_nv50(struct nouveau_channel_context *);
-extern int nouveau_surface_init_nv04(struct nouveau_context *);
-extern int nouveau_surface_init_nv50(struct nouveau_context *);
-
-extern uint32_t *nouveau_pipe_dma_beginp(struct nouveau_grobj *, int, int);
-extern void nouveau_pipe_dma_kickoff(struct nouveau_channel *);
-
-#endif
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.c
new file mode 100644
index 0000000000..dfc4905bc0
--- /dev/null
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.c
@@ -0,0 +1,172 @@
+#include "nouveau_context_vl.h"
+#include <pipe/p_defines.h>
+#include <pipe/p_context.h>
+#include <pipe/p_screen.h>
+#include <util/u_memory.h>
+#include <common/nouveau_dri.h>
+#include <common/nouveau_local.h>
+#include <common/nouveau_winsys_pipe.h>
+#include "nouveau_screen_vl.h"
+
+/*
+#ifdef DEBUG
+static const struct dri_debug_control debug_control[] = {
+	{ "bo", DEBUG_BO },
+	{ NULL, 0 }
+};
+int __nouveau_debug = 0;
+#endif
+*/
+
+int
+nouveau_context_create(dri_context_t *dri_context)
+{
+	dri_screen_t			*dri_screen;
+	struct nouveau_screen_vl	*nv_screen;
+	struct nouveau_context_vl	*nv;
+
+	assert (dri_context);
+
+	dri_screen = dri_context->dri_screen;
+	nv_screen = dri_screen->private;
+	nv = CALLOC_STRUCT(nouveau_context_vl);
+
+	if (!nv)
+		return 1;
+
+	if (nouveau_context_init(&nv_screen->base, dri_context->drm_context,
+	                        (drmLock*)&dri_screen->sarea->lock, NULL, &nv->base))
+	{
+		FREE(nv);
+		return 1;
+	}
+
+	dri_context->private = (void*)nv;
+	nv->dri_context = dri_context;
+	nv->nv_screen  = nv_screen;
+
+	/*
+	driParseConfigFiles(&nv->dri_option_cache, &nv_screen->option_cache,
+			    nv->dri_screen->myNum, "nouveau");
+#ifdef DEBUG
+	__nouveau_debug = driParseDebugString(getenv("NOUVEAU_DEBUG"),
+					      debug_control);
+#endif
+	*/
+
+	nv->base.nvc->pctx[nv->base.pctx_id]->priv = nv;
+
+	return 0;
+}
+
+void
+nouveau_context_destroy(dri_context_t *dri_context)
+{
+	struct nouveau_context_vl *nv = dri_context->private;
+
+	assert(dri_context);
+
+	nouveau_context_cleanup(&nv->base);
+
+	FREE(nv);
+}
+
+int
+nouveau_context_bind(struct nouveau_context_vl *nv, dri_drawable_t *dri_drawable)
+{
+	assert(nv);
+	assert(dri_drawable);
+
+	if (nv->dri_drawable != dri_drawable)
+	{
+		nv->dri_drawable = dri_drawable;
+		dri_drawable->private = nv;
+	}
+
+	return 0;
+}
+
+int
+nouveau_context_unbind(struct nouveau_context_vl *nv)
+{
+	assert(nv);
+
+	nv->dri_drawable = NULL;
+
+	return 0;
+}
+
+/* Show starts here */
+
+int bind_pipe_drawable(struct pipe_context *pipe, Drawable drawable)
+{
+	struct nouveau_context_vl	*nv;
+	dri_drawable_t			*dri_drawable;
+
+	assert(pipe);
+
+	nv = pipe->priv;
+
+	driCreateDrawable(nv->nv_screen->dri_screen, drawable, &dri_drawable);
+
+	nouveau_context_bind(nv, dri_drawable);
+
+	return 0;
+}
+
+int unbind_pipe_drawable(struct pipe_context *pipe)
+{
+	assert (pipe);
+
+	nouveau_context_unbind(pipe->priv);
+
+	return 0;
+}
+
+struct pipe_context* create_pipe_context(Display *display, int screen)
+{
+	dri_screen_t			*dri_screen;
+	dri_framebuffer_t		dri_framebuf;
+	dri_context_t			*dri_context;
+	struct nouveau_context_vl	*nv;
+
+	assert(display);
+
+	driCreateScreen(display, screen, &dri_screen, &dri_framebuf);
+	driCreateContext(dri_screen, XDefaultVisual(display, screen), &dri_context);
+
+	nouveau_screen_create(dri_screen, &dri_framebuf);
+	nouveau_context_create(dri_context);
+
+	nv = dri_context->private;
+
+	return nv->base.nvc->pctx[nv->base.pctx_id];
+}
+
+int destroy_pipe_context(struct pipe_context *pipe)
+{
+	struct pipe_screen		*screen;
+	struct pipe_winsys		*winsys;
+	struct nouveau_context_vl	*nv;
+	dri_screen_t			*dri_screen;
+	dri_context_t			*dri_context;
+
+	assert(pipe);
+
+	screen = pipe->screen;
+	winsys = pipe->winsys;
+	nv = pipe->priv;
+	dri_context = nv->dri_context;
+	dri_screen = dri_context->dri_screen;
+
+	pipe->destroy(pipe);
+	screen->destroy(screen);
+	FREE(winsys);
+
+	nouveau_context_destroy(dri_context);
+	nouveau_screen_destroy(dri_screen);
+	driDestroyContext(dri_context);
+	driDestroyScreen(dri_screen);
+
+	return 0;
+}
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.h b/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.h
new file mode 100644
index 0000000000..1115c3130c
--- /dev/null
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_context_vl.h
@@ -0,0 +1,39 @@
+#ifndef __NOUVEAU_CONTEXT_VL_H__
+#define __NOUVEAU_CONTEXT_VL_H__
+
+#include <driclient.h>
+#include <nouveau/nouveau_winsys.h>
+#include <common/nouveau_context.h>
+
+/*#include "xmlconfig.h"*/
+
+struct nouveau_context_vl {
+	struct nouveau_context		base;
+	struct nouveau_screen_vl	*nv_screen;
+	dri_context_t			*dri_context;
+	dri_drawable_t			*dri_drawable;
+	unsigned int			last_stamp;
+	/*driOptionCache		dri_option_cache;*/
+	drm_context_t			drm_context;
+	drmLock				drm_lock;
+};
+
+extern int nouveau_context_create(dri_context_t *);
+extern void nouveau_context_destroy(dri_context_t *);
+extern int nouveau_context_bind(struct nouveau_context_vl *, dri_drawable_t *);
+extern int nouveau_context_unbind(struct nouveau_context_vl *);
+
+#ifdef DEBUG
+extern int __nouveau_debug;
+
+#define DEBUG_BO (1 << 0)
+
+#define DBG(flag, ...) do {                   \
+	if (__nouveau_debug & (DEBUG_##flag)) \
+		NOUVEAU_ERR(__VA_ARGS__);     \
+} while(0)
+#else
+#define DBG(flag, ...)
+#endif
+
+#endif
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_lock.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_lock.c
index 375634bd05..f292586974 100644
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_lock.c
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_lock.c
@@ -25,12 +25,12 @@
  * 
  **************************************************************************/
 
-#include <pthread.h>
+#include <pipe/p_thread.h>
 #include <driclient.h>
 #include "nouveau_context.h"
 #include "nouveau_screen.h"
 
-static pthread_mutex_t lockMutex = PTHREAD_MUTEX_INITIALIZER;
+pipe_static_mutex(lockMutex);
 
 static void
 nouveau_contended_lock(struct nouveau_context *nv, unsigned int flags)
@@ -62,7 +62,7 @@ LOCK_HARDWARE(struct nouveau_context *nv)
 	struct nouveau_device_priv	*nvdev = nouveau_device(dev);
 	char				__ret=0;
 
-	pthread_mutex_lock(&lockMutex);
+	pipe_mutex_lock(lockMutex);
 	assert(!nv->locked);
 	
 	DRM_CAS(nvdev->lock, nvdev->ctx,
@@ -88,5 +88,5 @@ UNLOCK_HARDWARE(struct nouveau_context *nv)
 
 	DRM_UNLOCK(nvdev->fd, nvdev->lock, nvdev->ctx);
 
-	pthread_mutex_unlock(&lockMutex);
+	pipe_mutex_unlock(lockMutex);
 } 
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_screen.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_screen_vl.c
index f80d00050c..658dafd910 100644
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_screen.c
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_screen_vl.c
@@ -1,11 +1,8 @@
-#include "pipe/p_context.h"
-#include "util/u_memory.h"
-#include "nouveau_context.h"
+#include "nouveau_screen_vl.h"
+#include <util/u_memory.h>
 #include <nouveau_drm.h>
-#include "nouveau_dri.h"
-#include "nouveau_local.h"
-#include "nouveau_screen.h"
-#include "nouveau_swapbuffers.h"
+#include <common/nouveau_dri.h>
+#include <common/nouveau_local.h>
 
 #if NOUVEAU_DRM_HEADER_PATCHLEVEL != 11
 #error nouveau_drm.h version does not match expected version
@@ -50,34 +47,33 @@ int nouveau_check_dri_drm_ddx(dri_version_t *dri, dri_version_t *drm, dri_versio
 int
 nouveau_screen_create(dri_screen_t *dri_screen, dri_framebuffer_t *dri_framebuf)
 {
-	struct nouveau_dri	*nv_dri = dri_framebuf->private;
-	struct nouveau_screen	*nv_screen;
-	int			ret;
+	struct nouveau_dri		*nv_dri = dri_framebuf->private;
+	struct nouveau_screen_vl	*nv_screen;
+
+	assert(dri_screen);
+	assert(dri_framebuf);
 
 	if (nouveau_check_dri_drm_ddx(&dri_screen->dri, &dri_screen->drm, &dri_screen->ddx))
 		return 1;
 
-	nv_screen = CALLOC_STRUCT(nouveau_screen);
+	nv_screen = CALLOC_STRUCT(nouveau_screen_vl);
+
 	if (!nv_screen)
 		return 1;
-	nv_screen->dri_screen = dri_screen;
-	dri_screen->private = (void*)nv_screen;
+
+	if (nouveau_screen_init(nv_dri, dri_screen->fd, &nv_screen->base))
+	{
+		FREE(nv_screen);
+		return 1;
+	}
 
 	/*
 	driParseOptionInfo(&nv_screen->option_cache,
 			   __driConfigOptions, __driNConfigOptions);
 	*/
 
-	if ((ret = nouveau_device_open_existing(&nv_screen->device, 0,
-						dri_screen->fd, 0))) {
-		NOUVEAU_ERR("Failed opening nouveau device: %d.\n", ret);
-		return 1;
-	}
-
-	nv_screen->front_offset = nv_dri->front_offset;
-	nv_screen->front_pitch  = nv_dri->front_pitch * (nv_dri->bpp / 8);
-	nv_screen->front_cpp = nv_dri->bpp / 8;
-	nv_screen->front_height = nv_dri->height;
+	nv_screen->dri_screen = dri_screen;
+	dri_screen->private = (void*)nv_screen;
 
 	return 0;
 }
@@ -85,7 +81,8 @@ nouveau_screen_create(dri_screen_t *dri_screen, dri_framebuffer_t *dri_framebuf)
 void
 nouveau_screen_destroy(dri_screen_t *dri_screen)
 {
-	struct nouveau_screen *nv_screen = dri_screen->private;
+	struct nouveau_screen_vl *nv_screen = dri_screen->private;
 
+	nouveau_screen_cleanup(&nv_screen->base);
 	FREE(nv_screen);
 }
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_screen.h b/src/gallium/winsys/g3dvl/nouveau/nouveau_screen_vl.h
index 8a58bb7556..0c1ceca6de 100644
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_screen.h
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_screen_vl.h
@@ -1,19 +1,16 @@
-#ifndef __NOUVEAU_SCREEN_H__
-#define __NOUVEAU_SCREEN_H__
+#ifndef __NOUVEAU_SCREEN_VL_H__
+#define __NOUVEAU_SCREEN_VL_H__
+
+#include <driclient.h>
+#include <common/nouveau_screen.h>
 
 /* TODO: Investigate using DRI options for interesting things */
 /*#include "xmlconfig.h"*/
 
-struct nouveau_screen {
+struct nouveau_screen_vl
+{
+	struct nouveau_screen		base;
 	dri_screen_t			*dri_screen;
-	struct nouveau_device		*device;
-	struct nouveau_channel_context	*nvc;
-
-	uint32_t			front_offset;
-	uint32_t			front_pitch;
-	uint32_t			front_cpp;
-	uint32_t			front_height;
-	
 	/*driOptionCache		option_cache;*/
 };
 
@@ -21,4 +18,3 @@ int nouveau_screen_create(dri_screen_t *dri_screen, dri_framebuffer_t *dri_frame
 void nouveau_screen_destroy(dri_screen_t *dri_screen);
 
 #endif
-
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_swapbuffers.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_swapbuffers.c
index 7916c80615..16e6d5543c 100644
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_swapbuffers.c
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_swapbuffers.c
@@ -1,26 +1,26 @@
-#include "pipe/p_context.h"
-#include "nouveau_context.h"
-#include "nouveau_local.h"
-#include "nouveau_screen.h"
+#include <driclient.h>
+#include <common/nouveau_local.h>
+#include <common/nouveau_screen.h>
+#include "nouveau_context_vl.h"
 #include "nouveau_swapbuffers.h"
 
 void
 nouveau_copy_buffer(dri_drawable_t *dri_drawable, struct pipe_surface *surf,
 		    const drm_clip_rect_t *rect)
 {
-	struct nouveau_context	*nv = dri_drawable->private;
-	drm_clip_rect_t		*pbox;
-	int			nbox, i;
+	struct nouveau_context_vl	*nv = dri_drawable->private;
+	drm_clip_rect_t			*pbox;
+	int				nbox, i;
 
-	LOCK_HARDWARE(nv);
+	LOCK_HARDWARE(&nv->base);
 	if (!dri_drawable->num_cliprects) {
-		UNLOCK_HARDWARE(nv);
+		UNLOCK_HARDWARE(&nv->base);
 		return;
 	}
 	pbox = dri_drawable->cliprects;
 	nbox = dri_drawable->num_cliprects;
 
-	nv->surface_copy_prep(nv, nv->frontbuffer, surf);
+	nv->base.surface_copy_prep(&nv->base, nv->base.frontbuffer, surf);
 	for (i = 0; i < nbox; i++, pbox++) {
 		int sx, sy, dx, dy, w, h;
 
@@ -31,14 +31,11 @@ nouveau_copy_buffer(dri_drawable_t *dri_drawable, struct pipe_surface *surf,
 		w  = pbox->x2 - pbox->x1;
 		h  = pbox->y2 - pbox->y1;
 
-		nv->surface_copy(nv, dx, dy, sx, sy, w, h);
+		nv->base.surface_copy(&nv->base, dx, dy, sx, sy, w, h);
 	}
 
-	FIRE_RING(nv->nvc->channel);
-	UNLOCK_HARDWARE(nv);
-
-	//if (nv->last_stamp != dri_drawable->last_sarea_stamp)
-		//nv->last_stamp = dri_drawable->last_sarea_stamp;
+	FIRE_RING(nv->base.nvc->channel);
+	UNLOCK_HARDWARE(&nv->base);
 }
 
 void
@@ -62,3 +59,35 @@ nouveau_swap_buffers(dri_drawable_t *dri_drawable, struct pipe_surface *surf)
 		nouveau_copy_buffer(dri_drawable, surf, NULL);
 }
 
+void
+nouveau_flush_frontbuffer(struct pipe_winsys *pws, struct pipe_surface *surf,
+			  void *context_private)
+{
+	struct nouveau_context_vl	*nv;
+	dri_drawable_t			*dri_drawable;
+
+	assert(pws);
+	assert(surf);
+	assert(context_private);
+
+	nv = context_private;
+	dri_drawable = nv->dri_drawable;
+
+	nouveau_copy_buffer(dri_drawable, surf, NULL);
+}
+
+void
+nouveau_contended_lock(struct nouveau_context *nv)
+{
+	struct nouveau_context_vl	*nv_vl = (struct nouveau_context_vl*)nv;
+	dri_drawable_t			*dri_drawable = nv_vl->dri_drawable;
+	dri_screen_t			*dri_screen = nv_vl->dri_context->dri_screen;
+
+	/* If the window moved, may need to set a new cliprect now.
+	 *
+	 * NOTE: This releases and regains the hw lock, so all state
+	 * checking must be done *after* this call:
+	 */
+	if (dri_drawable)
+		DRI_VALIDATE_DRAWABLE_INFO(dri_screen, dri_drawable);
+}
diff --git a/src/gallium/winsys/g3dvl/nouveau/nouveau_winsys_pipe.c b/src/gallium/winsys/g3dvl/nouveau/nouveau_winsys_pipe.c
index 4f6ac9cad0..17c409e1ce 100644
--- a/src/gallium/winsys/g3dvl/nouveau/nouveau_winsys_pipe.c
+++ b/src/gallium/winsys/g3dvl/nouveau/nouveau_winsys_pipe.c
@@ -89,6 +89,31 @@ nouveau_surface_release(struct pipe_winsys *ws, struct pipe_surface **s)
 	}
 }
 
+static uint32_t
+nouveau_flags_from_usage(struct nouveau_context *nv, unsigned usage)
+{
+	uint32_t flags = NOUVEAU_BO_LOCAL;
+
+	if (usage & PIPE_BUFFER_USAGE_PIXEL) {
+		if (usage & NOUVEAU_BUFFER_USAGE_TEXTURE)
+			flags |= NOUVEAU_BO_GART;
+		if (!(usage & PIPE_BUFFER_USAGE_CPU_READ_WRITE))
+			flags |= NOUVEAU_BO_VRAM;
+	}
+
+	if (usage & PIPE_BUFFER_USAGE_VERTEX) {
+		if (nv->cap.hw_vertex_buffer)
+			flags |= NOUVEAU_BO_GART;
+	}
+
+	if (usage & PIPE_BUFFER_USAGE_INDEX) {
+		if (nv->cap.hw_index_buffer)
+			flags |= NOUVEAU_BO_GART;
+	}
+
+	return flags;
+}
+
 static struct pipe_buffer *
 nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
 		       unsigned usage, unsigned size)
@@ -107,23 +132,7 @@ nouveau_pipe_bo_create(struct pipe_winsys *pws, unsigned alignment,
 	nvbuf->base.usage = usage;
 	nvbuf->base.size = size;
 
-	flags = NOUVEAU_BO_LOCAL;
-
-	if (usage & PIPE_BUFFER_USAGE_PIXEL) {
-		if (usage & NOUVEAU_BUFFER_USAGE_TEXTURE)
-			flags |= NOUVEAU_BO_GART;
-		flags |= NOUVEAU_BO_VRAM;
-	}
-
-	if (usage & PIPE_BUFFER_USAGE_VERTEX) {
-		if (nv->cap.hw_vertex_buffer)
-			flags |= NOUVEAU_BO_GART;
-	}
-
-	if (usage & PIPE_BUFFER_USAGE_INDEX) {
-		if (nv->cap.hw_index_buffer)
-			flags |= NOUVEAU_BO_GART;
-	}
+	flags = nouveau_flags_from_usage(nv, usage);
 
 	if (nouveau_bo_new(dev, flags, alignment, size, &nvbuf->bo)) {
 		free(nvbuf);
@@ -175,6 +184,26 @@ nouveau_pipe_bo_map(struct pipe_winsys *pws, struct pipe_buffer *buf,
 	if (flags & PIPE_BUFFER_USAGE_CPU_WRITE)
 		map_flags |= NOUVEAU_BO_WR;
 
+	if ((map_flags & NOUVEAU_BO_RDWR) == NOUVEAU_BO_WR &&
+	    !nouveau_bo_busy(nvbuf->bo, map_flags)) {
+		/* XXX: Technically incorrect. If the client maps a buffer for write-only
+		 * and leaves part of the buffer untouched it probably expects those parts
+		 * to remain intact. This is violated because we allocate a whole new buffer
+		 * and don't copy the previous buffer's contents, so this optimization is
+		 * only valid if the client intends to overwrite the whole buffer.
+		 */
+		struct nouveau_pipe_winsys *nvpws = (struct nouveau_pipe_winsys *)pws;
+		struct nouveau_context *nv = nvpws->nv;
+		struct nouveau_device *dev = nv->nv_screen->device;
+		struct nouveau_bo *rename;
+		uint32_t flags = nouveau_flags_from_usage(nv, buf->usage);
+
+		if (!nouveau_bo_new(dev, flags, buf->alignment, buf->size, &rename)) {
+			nouveau_bo_del(&nvbuf->bo);
+			nvbuf->bo = rename;
+		}
+	}
+
 	if (nouveau_bo_map(nvbuf->bo, map_flags))
 		return NULL;
 	return nvbuf->bo->map;