Merge branch 'master' into asm-shader-rework-2

Conflicts:
	src/mesa/shader/lex.yy.c
	src/mesa/shader/program_parse.tab.c
	src/mesa/shader/program_parse.tab.h

12 files changed, 480 insertions, 209 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
index 802ec371189..eb492076b7d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
+++ b/src/gallium/auxiliary/tgsi/tgsi-instruction-set.txt
@@ -187,11 +187,7 @@ TGSI Instruction Specification
 
 1.2.6  CND0 - Condition Zero
 
-  dst.x = (src2.x >= 0.0) ? src0.x : src1.x
-  dst.y = (src2.y >= 0.0) ? src0.y : src1.y
-  dst.z = (src2.z >= 0.0) ? src0.z : src1.z
-  dst.w = (src2.w >= 0.0) ? src0.w : src1.w
-
+       Removed.  Use (CMP src2, src1, src0) instead.
 
 1.2.7  DOT2ADD - 2-component Dot Product And Add
 
@@ -1031,12 +1027,12 @@ TGSI Instruction Specification
 
 1.18.1  EXPP - Approximate Exponential Base 2
 
-  Alias for EXP.
+  Use EXP. See also 1.19.3.
 
 
 1.18.2  LOGP - Logarithm Base 2
 
-  Alias for LG2.
+  Use LOG. See also 1.19.4.
 
 
 1.19  vs_2_0
@@ -1053,6 +1049,16 @@ TGSI Instruction Specification
   Alias for ARR.
 
 
+1.19.3  EXPP - Approximate Exponential Base 2
+
+  Use EX2.
+
+
+1.19.4  LOGP - Logarithm Base 2
+
+  Use LG2.
+
+
 2  Explanation of symbols used
 ==============================
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index 05b07a3a73e..111d95b6665 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -43,6 +43,7 @@ struct dump_ctx
    struct tgsi_iterate_context iter;
 
    uint instno;
+   int indent;
    
    uint indentation;
 
@@ -335,14 +336,6 @@ tgsi_dump_immediate(
    iter_immediate( &ctx.iter, (struct tgsi_full_immediate *)imm );
 }
 
-static void
-indent(struct dump_ctx *ctx)
-{
-   uint i;
-   for (i = 0; i < ctx->indentation; i++)
-      TXT(" ");
-}
-
 static boolean
 iter_instruction(
    struct tgsi_iterate_context *iter,
@@ -350,22 +343,19 @@ iter_instruction(
 {
    struct dump_ctx *ctx = (struct dump_ctx *) iter;
    uint instno = ctx->instno++;
-   
+   const struct tgsi_opcode_info *info = tgsi_get_opcode_info( inst->Instruction.Opcode );
    uint i;
    boolean first_reg = TRUE;
 
    INSTID( instno );
    TXT( ": " );
-
-   /* update indentation */
-   if (inst->Instruction.Opcode == TGSI_OPCODE_ENDIF ||
-       inst->Instruction.Opcode == TGSI_OPCODE_ENDFOR ||
-       inst->Instruction.Opcode == TGSI_OPCODE_ENDLOOP) {
-      ctx->indentation -= indent_spaces;
-   }
-   indent(ctx);
-
-   TXT( tgsi_get_opcode_info( inst->Instruction.Opcode )->mnemonic );
+   
+   ctx->indent -= info->pre_dedent;
+   for(i = 0; (int)i < ctx->indent; ++i)
+      TXT( "  " );
+   ctx->indent += info->post_indent;
+   
+   TXT( info->mnemonic );
 
    switch (inst->Instruction.Saturate) {
    case TGSI_SAT_NONE:
@@ -526,6 +516,7 @@ tgsi_dump_instruction(
    struct dump_ctx ctx;
 
    ctx.instno = instno;
+   ctx.indent = 0;
    ctx.printf = dump_ctx_printf;
    ctx.indentation = 0;
 
@@ -559,6 +550,7 @@ tgsi_dump(
    ctx.iter.epilog = NULL;
 
    ctx.instno = 0;
+   ctx.indent = 0;
    ctx.printf = dump_ctx_printf;
    ctx.indentation = 0;
 
@@ -612,6 +604,7 @@ tgsi_dump_str(
    ctx.base.iter.epilog = NULL;
 
    ctx.base.instno = 0;
+   ctx.base.indent = 0;
    ctx.base.printf = &str_dump_ctx_printf;
    ctx.base.indentation = 0;
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 711e86d6edf..c79c56debd6 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -2329,16 +2329,6 @@ exec_instruction(
       }
       break;
 
-   case TGSI_OPCODE_CND0:
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-         FETCH(&r[2], 2, chan_index);
-         micro_le(&r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[2], &r[0], &r[1]);
-         STORE(&r[0], 0, chan_index);
-      }
-      break;
-
    case TGSI_OPCODE_DP2A:
       FETCH( &r[0], 0, CHAN_X );
       FETCH( &r[1], 1, CHAN_X );
@@ -2766,19 +2756,32 @@ exec_instruction(
       if (mach->ExecMask) {
          /* do the call */
 
-         /* push the Cond, Loop, Cont stacks */
+         /* First, record the depths of the execution stacks.
+          * This is important for deeply nested/looped return statements.
+          * We have to unwind the stacks by the correct amount.  For a
+          * real code generator, we could determine the number of entries
+          * to pop off each stack with simple static analysis and avoid
+          * implementing this data structure at run time.
+          */
+         mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
+         mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
+         mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
+         /* note that PC was already incremented above */
+         mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
+
+         mach->CallStackTop++;
+
+         /* Second, push the Cond, Loop, Cont, Func stacks */
          assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
          mach->CondStack[mach->CondStackTop++] = mach->CondMask;
          assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
          assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->ContStack[mach->ContStackTop++] = mach->ContMask;
-
          assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
          mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
 
-         /* note that PC was already incremented above */
-         mach->CallStack[mach->CallStackTop++] = *pc;
+         /* Finally, jump to the subroutine */
          *pc = inst->InstructionExtLabel.Label;
       }
       break;
@@ -2795,18 +2798,24 @@ exec_instruction(
             *pc = -1;
             return;
          }
-         *pc = mach->CallStack[--mach->CallStackTop];
 
-         /* pop the Cond, Loop, Cont stacks */
-         assert(mach->CondStackTop > 0);
-         mach->CondMask = mach->CondStack[--mach->CondStackTop];
-         assert(mach->LoopStackTop > 0);
-         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
-         assert(mach->ContStackTop > 0);
-         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->CallStackTop > 0);
+         mach->CallStackTop--;
+
+         mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
+         mach->CondMask = mach->CondStack[mach->CondStackTop];
+
+         mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
+         mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
+
+         mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
+         mach->ContMask = mach->ContStack[mach->ContStackTop];
+
          assert(mach->FuncStackTop > 0);
          mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
 
+         *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
+
          UPDATE_EXEC_MASK(mach);
       }
       break;
@@ -3104,6 +3113,12 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_BGNFOR:
+      assert(mach->LoopCounterStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      for (chan_index = 0; chan_index < 3; chan_index++) {
+         FETCH( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[chan_index], 0, chan_index );
+      }
+      STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X );
+      ++mach->LoopCounterStackTop;
       /* fall-through (for now) */
    case TGSI_OPCODE_BGNLOOP:
       /* push LoopMask and ContMasks */
@@ -3111,10 +3126,58 @@ exec_instruction(
       mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
       assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
       mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+      assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1;
       break;
 
    case TGSI_OPCODE_ENDFOR:
-      /* fall-through (for now at least) */
+      assert(mach->LoopCounterStackTop > 0);
+      micro_sub( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 
+                 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X],
+                 &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+      /* update LoopMask */
+      if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[0] <= 0) {
+         mach->LoopMask &= ~0x1;
+      }
+      if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[1] <= 0 ) {
+         mach->LoopMask &= ~0x2;
+      }
+      if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[2] <= 0 ) {
+         mach->LoopMask &= ~0x4;
+      }
+      if( mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X].f[3] <= 0 ) {
+         mach->LoopMask &= ~0x8;
+      }
+      micro_add( &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], 
+                 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], 
+                 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]);
+      assert(mach->LoopLabelStackTop > 0);
+      inst = mach->Instructions + mach->LoopLabelStack[mach->LoopLabelStackTop - 1];
+      STORE( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_Y], 0, CHAN_X );
+      /* Restore ContMask, but don't pop */
+      assert(mach->ContStackTop > 0);
+      mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
+      UPDATE_EXEC_MASK(mach);
+      if (mach->ExecMask) {
+         /* repeat loop: jump to instruction just past BGNLOOP */
+         assert(mach->LoopLabelStackTop > 0);
+         *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
+      }
+      else {
+         /* exit loop: pop LoopMask */
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         /* pop ContMask */
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->LoopLabelStackTop > 0);
+         --mach->LoopLabelStackTop;
+         assert(mach->LoopCounterStackTop > 0);
+         --mach->LoopCounterStackTop;
+      }
+      UPDATE_EXEC_MASK(mach);
+      break;
+      
    case TGSI_OPCODE_ENDLOOP:
       /* Restore ContMask, but don't pop */
       assert(mach->ContStackTop > 0);
@@ -3122,7 +3185,8 @@ exec_instruction(
       UPDATE_EXEC_MASK(mach);
       if (mach->ExecMask) {
          /* repeat loop: jump to instruction just past BGNLOOP */
-         *pc = inst->InstructionExtLabel.Label + 1;
+         assert(mach->LoopLabelStackTop > 0);
+         *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
       }
       else {
          /* exit loop: pop LoopMask */
@@ -3131,6 +3195,8 @@ exec_instruction(
          /* pop ContMask */
          assert(mach->ContStackTop > 0);
          mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->LoopLabelStackTop > 0);
+         --mach->LoopLabelStackTop;
       }
       UPDATE_EXEC_MASK(mach);
       break;
@@ -3198,7 +3264,6 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    mach->FuncMask = 0xf;
    mach->ExecMask = 0xf;
 
-   mach->CondStackTop = 0; /* temporarily subvert this assertion */
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index fd9ef6f35df..c72f76809d4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -186,6 +186,17 @@ struct tgsi_exec_labels
  */
 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 
+
+/** function call/activation record */
+struct tgsi_call_record
+{
+   uint CondStackTop;
+   uint LoopStackTop;
+   uint ContStackTop;
+   uint ReturnAddr;
+};
+
+
 /**
  * Run-time virtual machine state for executing TGSI shader.
  */
@@ -232,6 +243,14 @@ struct tgsi_exec_machine
    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
    int LoopStackTop;
 
+   /** Loop label stack */
+   uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int LoopLabelStackTop;
+
+   /** Loop counter stack (x = count, y = current, z = step) */
+   struct tgsi_exec_vector LoopCounterStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int LoopCounterStackTop;
+   
    /** Loop continue mask stack (see comments in tgsi_exec.c) */
    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
    int ContStackTop;
@@ -241,7 +260,7 @@ struct tgsi_exec_machine
    int FuncStackTop;
 
    /** Function call stack for saving/restoring the program counter */
-   uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+   struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
    int CallStackTop;
 
    struct tgsi_full_instruction *Instructions;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
index ccf4b205ffb..17af4cb7ad2 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -31,125 +31,125 @@
 
 static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
 {
-   { 1, 1, 0, 0, "ARL", TGSI_OPCODE_ARL },
-   { 1, 1, 0, 0, "MOV", TGSI_OPCODE_MOV },
-   { 1, 1, 0, 0, "LIT", TGSI_OPCODE_LIT },
-   { 1, 1, 0, 0, "RCP", TGSI_OPCODE_RCP },
-   { 1, 1, 0, 0, "RSQ", TGSI_OPCODE_RSQ },
-   { 1, 1, 0, 0, "EXP", TGSI_OPCODE_EXP },
-   { 1, 1, 0, 0, "LOG", TGSI_OPCODE_LOG },
-   { 1, 2, 0, 0, "MUL", TGSI_OPCODE_MUL },
-   { 1, 2, 0, 0, "ADD", TGSI_OPCODE_ADD },
-   { 1, 2, 0, 0, "DP3", TGSI_OPCODE_DP3 },
-   { 1, 2, 0, 0, "DP4", TGSI_OPCODE_DP4 },
-   { 1, 2, 0, 0, "DST", TGSI_OPCODE_DST },
-   { 1, 2, 0, 0, "MIN", TGSI_OPCODE_MIN },
-   { 1, 2, 0, 0, "MAX", TGSI_OPCODE_MAX },
-   { 1, 2, 0, 0, "SLT", TGSI_OPCODE_SLT },
-   { 1, 2, 0, 0, "SGE", TGSI_OPCODE_SGE },
-   { 1, 3, 0, 0, "MAD", TGSI_OPCODE_MAD },
-   { 1, 2, 0, 0, "SUB", TGSI_OPCODE_SUB },
-   { 1, 3, 0, 0, "LRP", TGSI_OPCODE_LRP },
-   { 1, 3, 0, 0, "CND", TGSI_OPCODE_CND },
-   { 1, 3, 0, 0, "CND0", TGSI_OPCODE_CND0 },
-   { 1, 3, 0, 0, "DP2A", TGSI_OPCODE_DP2A },
-   { 0, 0, 0, 0, "", 22 },      /* removed */
-   { 0, 0, 0, 0, "", 23 },      /* removed */
-   { 1, 1, 0, 0, "FRC", TGSI_OPCODE_FRC },
-   { 1, 3, 0, 0, "CLAMP", TGSI_OPCODE_CLAMP },
-   { 1, 1, 0, 0, "FLR", TGSI_OPCODE_FLR },
-   { 1, 1, 0, 0, "ROUND", TGSI_OPCODE_ROUND },
-   { 1, 1, 0, 0, "EX2", TGSI_OPCODE_EX2 },
-   { 1, 1, 0, 0, "LG2", TGSI_OPCODE_LG2 },
-   { 1, 2, 0, 0, "POW", TGSI_OPCODE_POW },
-   { 1, 2, 0, 0, "XPD", TGSI_OPCODE_XPD },
-   { 0, 0, 0, 0, "", 32 },      /* removed */
-   { 1, 1, 0, 0, "ABS", TGSI_OPCODE_ABS },
-   { 1, 1, 0, 0, "RCC", TGSI_OPCODE_RCC },
-   { 1, 2, 0, 0, "DPH", TGSI_OPCODE_DPH },
-   { 1, 1, 0, 0, "COS", TGSI_OPCODE_COS },
-   { 1, 1, 0, 0, "DDX", TGSI_OPCODE_DDX },
-   { 1, 1, 0, 0, "DDY", TGSI_OPCODE_DDY },
-   { 0, 0, 0, 0, "KILP", TGSI_OPCODE_KILP },
-   { 1, 1, 0, 0, "PK2H", TGSI_OPCODE_PK2H },
-   { 1, 1, 0, 0, "PK2US", TGSI_OPCODE_PK2US },
-   { 1, 1, 0, 0, "PK4B", TGSI_OPCODE_PK4B },
-   { 1, 1, 0, 0, "PK4UB", TGSI_OPCODE_PK4UB },
-   { 1, 2, 0, 0, "RFL", TGSI_OPCODE_RFL },
-   { 1, 2, 0, 0, "SEQ", TGSI_OPCODE_SEQ },
-   { 1, 2, 0, 0, "SFL", TGSI_OPCODE_SFL },
-   { 1, 2, 0, 0, "SGT", TGSI_OPCODE_SGT },
-   { 1, 1, 0, 0, "SIN", TGSI_OPCODE_SIN },
-   { 1, 2, 0, 0, "SLE", TGSI_OPCODE_SLE },
-   { 1, 2, 0, 0, "SNE", TGSI_OPCODE_SNE },
-   { 1, 2, 0, 0, "STR", TGSI_OPCODE_STR },
-   { 1, 2, 1, 0, "TEX", TGSI_OPCODE_TEX },
-   { 1, 4, 1, 0, "TXD", TGSI_OPCODE_TXD },
-   { 1, 2, 1, 0, "TXP", TGSI_OPCODE_TXP },
-   { 1, 1, 0, 0, "UP2H", TGSI_OPCODE_UP2H },
-   { 1, 1, 0, 0, "UP2US", TGSI_OPCODE_UP2US },
-   { 1, 1, 0, 0, "UP4B", TGSI_OPCODE_UP4B },
-   { 1, 1, 0, 0, "UP4UB", TGSI_OPCODE_UP4UB },
-   { 1, 3, 0, 0, "X2D", TGSI_OPCODE_X2D },
-   { 1, 1, 0, 0, "ARA", TGSI_OPCODE_ARA },
-   { 1, 1, 0, 0, "ARR", TGSI_OPCODE_ARR },
-   { 0, 1, 0, 0, "BRA", TGSI_OPCODE_BRA },
-   { 0, 0, 0, 1, "CAL", TGSI_OPCODE_CAL },
-   { 0, 0, 0, 0, "RET", TGSI_OPCODE_RET },
-   { 1, 1, 0, 0, "SSG", TGSI_OPCODE_SSG },
-   { 1, 3, 0, 0, "CMP", TGSI_OPCODE_CMP },
-   { 1, 1, 0, 0, "SCS", TGSI_OPCODE_SCS },
-   { 1, 2, 1, 0, "TXB", TGSI_OPCODE_TXB },
-   { 1, 1, 0, 0, "NRM", TGSI_OPCODE_NRM },
-   { 1, 2, 0, 0, "DIV", TGSI_OPCODE_DIV },
-   { 1, 2, 0, 0, "DP2", TGSI_OPCODE_DP2 },
-   { 1, 2, 1, 0, "TXL", TGSI_OPCODE_TXL },
-   { 0, 0, 0, 0, "BRK", TGSI_OPCODE_BRK },
-   { 0, 1, 0, 1, "IF", TGSI_OPCODE_IF },
-   { 1, 1, 0, 0, "BGNFOR", TGSI_OPCODE_BGNFOR },
-   { 0, 1, 0, 0, "REP", TGSI_OPCODE_REP },
-   { 0, 0, 0, 1, "ELSE", TGSI_OPCODE_ELSE },
-   { 0, 0, 0, 0, "ENDIF", TGSI_OPCODE_ENDIF },
-   { 1, 0, 0, 0, "ENDFOR", TGSI_OPCODE_ENDFOR },
-   { 0, 0, 0, 0, "ENDREP", TGSI_OPCODE_ENDREP },
-   { 0, 1, 0, 0, "PUSHA", TGSI_OPCODE_PUSHA },
-   { 1, 0, 0, 0, "POPA", TGSI_OPCODE_POPA },
-   { 1, 1, 0, 0, "CEIL", TGSI_OPCODE_CEIL },
-   { 1, 1, 0, 0, "I2F", TGSI_OPCODE_I2F },
-   { 1, 1, 0, 0, "NOT", TGSI_OPCODE_NOT },
-   { 1, 1, 0, 0, "TRUNC", TGSI_OPCODE_TRUNC },
-   { 1, 2, 0, 0, "SHL", TGSI_OPCODE_SHL },
-   { 1, 2, 0, 0, "SHR", TGSI_OPCODE_SHR },
-   { 1, 2, 0, 0, "AND", TGSI_OPCODE_AND },
-   { 1, 2, 0, 0, "OR", TGSI_OPCODE_OR },
-   { 1, 2, 0, 0, "MOD", TGSI_OPCODE_MOD },
-   { 1, 2, 0, 0, "XOR", TGSI_OPCODE_XOR },
-   { 1, 3, 0, 0, "SAD", TGSI_OPCODE_SAD },
-   { 1, 2, 1, 0, "TXF", TGSI_OPCODE_TXF },
-   { 1, 2, 1, 0, "TXQ", TGSI_OPCODE_TXQ },
-   { 0, 0, 0, 0, "CONT", TGSI_OPCODE_CONT },
-   { 0, 0, 0, 0, "EMIT", TGSI_OPCODE_EMIT },
-   { 0, 0, 0, 0, "ENDPRIM", TGSI_OPCODE_ENDPRIM },
-   { 0, 0, 0, 1, "BGNLOOP", TGSI_OPCODE_BGNLOOP },
-   { 0, 0, 0, 0, "BGNSUB", TGSI_OPCODE_BGNSUB },
-   { 0, 0, 0, 1, "ENDLOOP", TGSI_OPCODE_ENDLOOP },
-   { 0, 0, 0, 0, "ENDSUB", TGSI_OPCODE_ENDSUB },
-   { 1, 1, 0, 0, "NOISE1", TGSI_OPCODE_NOISE1 },
-   { 1, 1, 0, 0, "NOISE2", TGSI_OPCODE_NOISE2 },
-   { 1, 1, 0, 0, "NOISE3", TGSI_OPCODE_NOISE3 },
-   { 1, 1, 0, 0, "NOISE4", TGSI_OPCODE_NOISE4 },
-   { 0, 0, 0, 0, "NOP", TGSI_OPCODE_NOP },
-   { 0, 0, 0, 0, "", 108 },     /* removed */
-   { 0, 0, 0, 0, "", 109 },     /* removed */
-   { 0, 0, 0, 0, "", 110 },     /* removed */
-   { 0, 0, 0, 0, "", 111 },     /* removed */
-   { 1, 1, 0, 0, "NRM4", TGSI_OPCODE_NRM4 },
-   { 0, 1, 0, 0, "CALLNZ", TGSI_OPCODE_CALLNZ },
-   { 0, 1, 0, 0, "IFC", TGSI_OPCODE_IFC },
-   { 0, 1, 0, 0, "BREAKC", TGSI_OPCODE_BREAKC },
-   { 0, 1, 0, 0, "KIL", TGSI_OPCODE_KIL },
-   { 0, 0, 0, 0, "END", TGSI_OPCODE_END },
-   { 1, 1, 0, 0, "SWZ", TGSI_OPCODE_SWZ }
+   { 1, 1, 0, 0, 0, 0, "ARL", TGSI_OPCODE_ARL },
+   { 1, 1, 0, 0, 0, 0, "MOV", TGSI_OPCODE_MOV },
+   { 1, 1, 0, 0, 0, 0, "LIT", TGSI_OPCODE_LIT },
+   { 1, 1, 0, 0, 0, 0, "RCP", TGSI_OPCODE_RCP },
+   { 1, 1, 0, 0, 0, 0, "RSQ", TGSI_OPCODE_RSQ },
+   { 1, 1, 0, 0, 0, 0, "EXP", TGSI_OPCODE_EXP },
+   { 1, 1, 0, 0, 0, 0, "LOG", TGSI_OPCODE_LOG },
+   { 1, 2, 0, 0, 0, 0, "MUL", TGSI_OPCODE_MUL },
+   { 1, 2, 0, 0, 0, 0, "ADD", TGSI_OPCODE_ADD },
+   { 1, 2, 0, 0, 0, 0, "DP3", TGSI_OPCODE_DP3 },
+   { 1, 2, 0, 0, 0, 0, "DP4", TGSI_OPCODE_DP4 },
+   { 1, 2, 0, 0, 0, 0, "DST", TGSI_OPCODE_DST },
+   { 1, 2, 0, 0, 0, 0, "MIN", TGSI_OPCODE_MIN },
+   { 1, 2, 0, 0, 0, 0, "MAX", TGSI_OPCODE_MAX },
+   { 1, 2, 0, 0, 0, 0, "SLT", TGSI_OPCODE_SLT },
+   { 1, 2, 0, 0, 0, 0, "SGE", TGSI_OPCODE_SGE },
+   { 1, 3, 0, 0, 0, 0, "MAD", TGSI_OPCODE_MAD },
+   { 1, 2, 0, 0, 0, 0, "SUB", TGSI_OPCODE_SUB },
+   { 1, 3, 0, 0, 0, 0, "LRP", TGSI_OPCODE_LRP },
+   { 1, 3, 0, 0, 0, 0, "CND", TGSI_OPCODE_CND },
+   { 0, 0, 0, 0, 0, 0, "", 20 },      /* removed */
+   { 1, 3, 0, 0, 0, 0, "DP2A", TGSI_OPCODE_DP2A },
+   { 0, 0, 0, 0, 0, 0, "", 22 },      /* removed */
+   { 0, 0, 0, 0, 0, 0, "", 23 },      /* removed */
+   { 1, 1, 0, 0, 0, 0, "FRC", TGSI_OPCODE_FRC },
+   { 1, 3, 0, 0, 0, 0, "CLAMP", TGSI_OPCODE_CLAMP },
+   { 1, 1, 0, 0, 0, 0, "FLR", TGSI_OPCODE_FLR },
+   { 1, 1, 0, 0, 0, 0, "ROUND", TGSI_OPCODE_ROUND },
+   { 1, 1, 0, 0, 0, 0, "EX2", TGSI_OPCODE_EX2 },
+   { 1, 1, 0, 0, 0, 0, "LG2", TGSI_OPCODE_LG2 },
+   { 1, 2, 0, 0, 0, 0, "POW", TGSI_OPCODE_POW },
+   { 1, 2, 0, 0, 0, 0, "XPD", TGSI_OPCODE_XPD },
+   { 0, 0, 0, 0, 0, 0, "", 32 },      /* removed */
+   { 1, 1, 0, 0, 0, 0, "ABS", TGSI_OPCODE_ABS },
+   { 1, 1, 0, 0, 0, 0, "RCC", TGSI_OPCODE_RCC },
+   { 1, 2, 0, 0, 0, 0, "DPH", TGSI_OPCODE_DPH },
+   { 1, 1, 0, 0, 0, 0, "COS", TGSI_OPCODE_COS },
+   { 1, 1, 0, 0, 0, 0, "DDX", TGSI_OPCODE_DDX },
+   { 1, 1, 0, 0, 0, 0, "DDY", TGSI_OPCODE_DDY },
+   { 0, 0, 0, 0, 0, 0, "KILP", TGSI_OPCODE_KILP },
+   { 1, 1, 0, 0, 0, 0, "PK2H", TGSI_OPCODE_PK2H },
+   { 1, 1, 0, 0, 0, 0, "PK2US", TGSI_OPCODE_PK2US },
+   { 1, 1, 0, 0, 0, 0, "PK4B", TGSI_OPCODE_PK4B },
+   { 1, 1, 0, 0, 0, 0, "PK4UB", TGSI_OPCODE_PK4UB },
+   { 1, 2, 0, 0, 0, 0, "RFL", TGSI_OPCODE_RFL },
+   { 1, 2, 0, 0, 0, 0, "SEQ", TGSI_OPCODE_SEQ },
+   { 1, 2, 0, 0, 0, 0, "SFL", TGSI_OPCODE_SFL },
+   { 1, 2, 0, 0, 0, 0, "SGT", TGSI_OPCODE_SGT },
+   { 1, 1, 0, 0, 0, 0, "SIN", TGSI_OPCODE_SIN },
+   { 1, 2, 0, 0, 0, 0, "SLE", TGSI_OPCODE_SLE },
+   { 1, 2, 0, 0, 0, 0, "SNE", TGSI_OPCODE_SNE },
+   { 1, 2, 0, 0, 0, 0, "STR", TGSI_OPCODE_STR },
+   { 1, 2, 1, 0, 0, 0, "TEX", TGSI_OPCODE_TEX },
+   { 1, 4, 1, 0, 0, 0, "TXD", TGSI_OPCODE_TXD },
+   { 1, 2, 1, 0, 0, 0, "TXP", TGSI_OPCODE_TXP },
+   { 1, 1, 0, 0, 0, 0, "UP2H", TGSI_OPCODE_UP2H },
+   { 1, 1, 0, 0, 0, 0, "UP2US", TGSI_OPCODE_UP2US },
+   { 1, 1, 0, 0, 0, 0, "UP4B", TGSI_OPCODE_UP4B },
+   { 1, 1, 0, 0, 0, 0, "UP4UB", TGSI_OPCODE_UP4UB },
+   { 1, 3, 0, 0, 0, 0, "X2D", TGSI_OPCODE_X2D },
+   { 1, 1, 0, 0, 0, 0, "ARA", TGSI_OPCODE_ARA },
+   { 1, 1, 0, 0, 0, 0, "ARR", TGSI_OPCODE_ARR },
+   { 0, 1, 0, 0, 0, 0, "BRA", TGSI_OPCODE_BRA },
+   { 0, 0, 0, 1, 0, 0, "CAL", TGSI_OPCODE_CAL },
+   { 0, 0, 0, 0, 0, 0, "RET", TGSI_OPCODE_RET },
+   { 1, 1, 0, 0, 0, 0, "SSG", TGSI_OPCODE_SSG },
+   { 1, 3, 0, 0, 0, 0, "CMP", TGSI_OPCODE_CMP },
+   { 1, 1, 0, 0, 0, 0, "SCS", TGSI_OPCODE_SCS },
+   { 1, 2, 1, 0, 0, 0, "TXB", TGSI_OPCODE_TXB },
+   { 1, 1, 0, 0, 0, 0, "NRM", TGSI_OPCODE_NRM },
+   { 1, 2, 0, 0, 0, 0, "DIV", TGSI_OPCODE_DIV },
+   { 1, 2, 0, 0, 0, 0, "DP2", TGSI_OPCODE_DP2 },
+   { 1, 2, 1, 0, 0, 0, "TXL", TGSI_OPCODE_TXL },
+   { 0, 0, 0, 0, 0, 0, "BRK", TGSI_OPCODE_BRK },
+   { 0, 1, 0, 1, 0, 1, "IF", TGSI_OPCODE_IF },
+   { 1, 1, 0, 0, 0, 1, "BGNFOR", TGSI_OPCODE_BGNFOR },
+   { 0, 1, 0, 0, 0, 1, "REP", TGSI_OPCODE_REP },
+   { 0, 0, 0, 1, 1, 1, "ELSE", TGSI_OPCODE_ELSE },
+   { 0, 0, 0, 0, 1, 0, "ENDIF", TGSI_OPCODE_ENDIF },
+   { 1, 0, 0, 0, 1, 0, "ENDFOR", TGSI_OPCODE_ENDFOR },
+   { 0, 0, 0, 0, 1, 0, "ENDREP", TGSI_OPCODE_ENDREP },
+   { 0, 1, 0, 0, 0, 0, "PUSHA", TGSI_OPCODE_PUSHA },
+   { 1, 0, 0, 0, 0, 0, "POPA", TGSI_OPCODE_POPA },
+   { 1, 1, 0, 0, 0, 0, "CEIL", TGSI_OPCODE_CEIL },
+   { 1, 1, 0, 0, 0, 0, "I2F", TGSI_OPCODE_I2F },
+   { 1, 1, 0, 0, 0, 0, "NOT", TGSI_OPCODE_NOT },
+   { 1, 1, 0, 0, 0, 0, "TRUNC", TGSI_OPCODE_TRUNC },
+   { 1, 2, 0, 0, 0, 0, "SHL", TGSI_OPCODE_SHL },
+   { 1, 2, 0, 0, 0, 0, "SHR", TGSI_OPCODE_SHR },
+   { 1, 2, 0, 0, 0, 0, "AND", TGSI_OPCODE_AND },
+   { 1, 2, 0, 0, 0, 0, "OR", TGSI_OPCODE_OR },
+   { 1, 2, 0, 0, 0, 0, "MOD", TGSI_OPCODE_MOD },
+   { 1, 2, 0, 0, 0, 0, "XOR", TGSI_OPCODE_XOR },
+   { 1, 3, 0, 0, 0, 0, "SAD", TGSI_OPCODE_SAD },
+   { 1, 2, 1, 0, 0, 0, "TXF", TGSI_OPCODE_TXF },
+   { 1, 2, 1, 0, 0, 0, "TXQ", TGSI_OPCODE_TXQ },
+   { 0, 0, 0, 0, 0, 0, "CONT", TGSI_OPCODE_CONT },
+   { 0, 0, 0, 0, 0, 0, "EMIT", TGSI_OPCODE_EMIT },
+   { 0, 0, 0, 0, 0, 0, "ENDPRIM", TGSI_OPCODE_ENDPRIM },
+   { 0, 0, 0, 1, 0, 1, "BGNLOOP", TGSI_OPCODE_BGNLOOP },
+   { 0, 0, 0, 0, 0, 1, "BGNSUB", TGSI_OPCODE_BGNSUB },
+   { 0, 0, 0, 1, 1, 0, "ENDLOOP", TGSI_OPCODE_ENDLOOP },
+   { 0, 0, 0, 0, 1, 0, "ENDSUB", TGSI_OPCODE_ENDSUB },
+   { 1, 1, 0, 0, 0, 0, "NOISE1", TGSI_OPCODE_NOISE1 },
+   { 1, 1, 0, 0, 0, 0, "NOISE2", TGSI_OPCODE_NOISE2 },
+   { 1, 1, 0, 0, 0, 0, "NOISE3", TGSI_OPCODE_NOISE3 },
+   { 1, 1, 0, 0, 0, 0, "NOISE4", TGSI_OPCODE_NOISE4 },
+   { 0, 0, 0, 0, 0, 0, "NOP", TGSI_OPCODE_NOP },
+   { 0, 0, 0, 0, 0, 0, "", 108 },     /* removed */
+   { 0, 0, 0, 0, 0, 0, "", 109 },     /* removed */
+   { 0, 0, 0, 0, 0, 0, "", 110 },     /* removed */
+   { 0, 0, 0, 0, 0, 0, "", 111 },     /* removed */
+   { 1, 1, 0, 0, 0, 0, "NRM4", TGSI_OPCODE_NRM4 },
+   { 0, 1, 0, 0, 0, 0, "CALLNZ", TGSI_OPCODE_CALLNZ },
+   { 0, 1, 0, 0, 0, 0, "IFC", TGSI_OPCODE_IFC },
+   { 0, 1, 0, 0, 0, 0, "BREAKC", TGSI_OPCODE_BREAKC },
+   { 0, 1, 0, 0, 0, 0, "KIL", TGSI_OPCODE_KIL },
+   { 0, 0, 0, 0, 0, 0, "END", TGSI_OPCODE_END },
+   { 1, 1, 0, 0, 0, 0, "SWZ", TGSI_OPCODE_SWZ }
 };
 
 const struct tgsi_opcode_info *
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.h b/src/gallium/auxiliary/tgsi/tgsi_info.h
index b2375c69710..74713c3b98a 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_info.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.h
@@ -36,10 +36,12 @@ extern "C" {
 
 struct tgsi_opcode_info
 {
-   uint num_dst;
-   uint num_src;
-   boolean is_tex;
-   boolean is_branch;
+   unsigned num_dst:3;
+   unsigned num_src:3;
+   unsigned is_tex:1;
+   unsigned is_branch:1;
+   int pre_dedent:2;
+   int post_indent:2;
    const char *mnemonic;
    uint opcode;
 };
diff --git a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
index ed594a3e2c7..e7bcf4bf754 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_opcode_tmp.h
@@ -60,7 +60,6 @@ OP13(MAD)
 OP12(SUB)
 OP13(LRP)
 OP13(CND)
-OP13(CND0)
 OP13(DP2A)
 OP11(FRC)
 OP13(CLAMP)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index 4fe8553c423..8a13885da9b 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -199,10 +199,10 @@ iter_instruction(
    }
 
    if (info->num_dst != inst->Instruction.NumDstRegs) {
-      report_error( ctx, "Invalid number of destination operands, should be %u", info->num_dst );
+      report_error( ctx, "%s: Invalid number of destination operands, should be %u", info->mnemonic, info->num_dst );
    }
    if (info->num_src != inst->Instruction.NumSrcRegs) {
-      report_error( ctx, "Invalid number of source operands, should be %u", info->num_src );
+      report_error( ctx, "%s: Invalid number of source operands, should be %u", info->mnemonic, info->num_src );
    }
 
    /* Check destination and source registers' validity.
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 46f2387c158..3cdf8b9f359 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -2089,10 +2089,6 @@ emit_instruction(
       return 0;
       break;
 
-   case TGSI_OPCODE_CND0:
-      return 0;
-      break;
-
    case TGSI_OPCODE_DP2A:
       FETCH( func, *inst, 0, 0, CHAN_X );  /* xmm0 = src[0].x */
       FETCH( func, *inst, 1, 1, CHAN_X );  /* xmm1 = src[1].x */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index c0a0627e0b2..f7096bd8e2c 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -29,6 +29,7 @@
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
 #include "tgsi/tgsi_ureg.h"
+#include "tgsi/tgsi_info.h"
 #include "tgsi/tgsi_dump.h"
 #include "util/u_memory.h"
 #include "util/u_math.h"
@@ -71,6 +72,7 @@ struct ureg_tokens {
 #define UREG_MAX_OUTPUT PIPE_MAX_ATTRIBS
 #define UREG_MAX_IMMEDIATE 32
 #define UREG_MAX_TEMP 256
+#define UREG_MAX_ADDR 2
 
 #define DOMAIN_DECL 0
 #define DOMAIN_INSN 1
@@ -99,11 +101,15 @@ struct ureg_program
    } immediate[UREG_MAX_IMMEDIATE];
    unsigned nr_immediates;
 
+   struct ureg_src sampler[PIPE_MAX_SAMPLERS];
+   unsigned nr_samplers;
+
    unsigned temps_active[UREG_MAX_TEMP / 32];
    unsigned nr_temps;
 
+   unsigned nr_addrs;
+
    unsigned nr_constants;
-   unsigned nr_samplers;
    unsigned nr_instructions;
 
    struct ureg_tokens domain[2];
@@ -187,6 +193,8 @@ ureg_dst_register( unsigned file,
    dst.File      = file;
    dst.WriteMask = TGSI_WRITEMASK_XYZW;
    dst.Indirect  = 0;
+   dst.IndirectIndex = 0;
+   dst.IndirectSwizzle = 0;
    dst.Saturate  = 0;
    dst.Index     = index;
    dst.Pad1      = 0;
@@ -208,6 +216,8 @@ ureg_src_register( unsigned file,
    src.SwizzleW = TGSI_SWIZZLE_W;
    src.Pad      = 0;
    src.Indirect = 0;
+   src.IndirectIndex = 0;
+   src.IndirectSwizzle = 0;
    src.Absolute = 0;
    src.Index    = index;
    src.Negate   = 0;
@@ -254,6 +264,7 @@ ureg_DECL_fs_input( struct ureg_program *ureg,
                     unsigned index,
                     unsigned interp )
 {
+   assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
    return ureg_DECL_input( ureg, name, index, interp );
 }
 
@@ -263,6 +274,7 @@ ureg_DECL_vs_input( struct ureg_program *ureg,
                     unsigned name,
                     unsigned index )
 {
+   assert(ureg->processor == TGSI_PROCESSOR_VERTEX);
    return ureg_DECL_input( ureg, name, index, TGSI_INTERPOLATE_CONSTANT );
 }
 
@@ -346,11 +358,36 @@ void ureg_release_temporary( struct ureg_program *ureg,
 }
 
 
+/* Allocate a new address register.
+ */
+struct ureg_dst ureg_DECL_address( struct ureg_program *ureg )
+{
+   if (ureg->nr_addrs < UREG_MAX_ADDR)
+      return ureg_dst_register( TGSI_FILE_ADDRESS, ureg->nr_addrs++ );
+
+   assert( 0 );
+   return ureg_dst_register( TGSI_FILE_ADDRESS, 0 );
+}
+
 /* Allocate a new sampler.
  */
-struct ureg_src ureg_DECL_sampler( struct ureg_program *ureg )
+struct ureg_src ureg_DECL_sampler( struct ureg_program *ureg,
+                                   unsigned nr )
 {
-   return ureg_src_register( TGSI_FILE_SAMPLER, ureg->nr_samplers++ );
+   unsigned i;
+
+   for (i = 0; i < ureg->nr_samplers; i++)
+      if (ureg->sampler[i].Index == nr)
+         return ureg->sampler[i];
+   
+   if (i < PIPE_MAX_SAMPLERS) {
+      ureg->sampler[i] = ureg_src_register( TGSI_FILE_SAMPLER, nr );
+      ureg->nr_samplers++;
+      return ureg->sampler[i];
+   }
+
+   assert( 0 );
+   return ureg->sampler[0];
 }
 
 
@@ -363,6 +400,8 @@ static int match_or_expand_immediate( const float *v,
                                       unsigned *swizzle )
 {
    unsigned i, j;
+   
+   *swizzle = 0;
 
    for (i = 0; i < nr; i++) {
       boolean found = FALSE;
@@ -394,8 +433,8 @@ struct ureg_src ureg_DECL_immediate( struct ureg_program *ureg,
                                      const float *v,
                                      unsigned nr )
 {
-   unsigned i;
-   unsigned swizzle = 0;
+   unsigned i, j;
+   unsigned swizzle;
 
    /* Could do a first pass where we examine all existing immediates
     * without expanding.
@@ -423,6 +462,12 @@ struct ureg_src ureg_DECL_immediate( struct ureg_program *ureg,
    set_bad( ureg );
 
 out:
+   /* Make sure that all referenced elements are from this immediate.
+    * Has the effect of making size-one immediates into scalars.
+    */
+   for (j = nr; j < 4; j++)
+      swizzle |= (swizzle & 0x3) << (j * 2);
+
    return ureg_swizzle( ureg_src_register( TGSI_FILE_IMMEDIATE, i ),
                         (swizzle >> 0) & 0x3,
                         (swizzle >> 2) & 0x3,
@@ -442,31 +487,39 @@ ureg_emit_src( struct ureg_program *ureg,
    union tgsi_any_token *out = get_tokens( ureg, DOMAIN_INSN, size );
    unsigned n = 0;
 
+   assert(src.File != TGSI_FILE_NULL);
+   assert(src.File != TGSI_FILE_OUTPUT);
+   assert(src.File < TGSI_FILE_COUNT);
+   
    out[n].value = 0;
    out[n].src.File = src.File;
    out[n].src.SwizzleX = src.SwizzleX;
    out[n].src.SwizzleY = src.SwizzleY;
    out[n].src.SwizzleZ = src.SwizzleZ;
    out[n].src.SwizzleW = src.SwizzleW;
-   out[n].src.Indirect = src.Indirect;
    out[n].src.Index = src.Index;
+   out[n].src.Negate = src.Negate;
    n++;
    
    if (src.Absolute) {
+      out[0].src.Extended = 1;
+      out[0].src.Negate = 0;
       out[n].value = 0;
+      out[n].src_ext_mod.Type = TGSI_SRC_REGISTER_EXT_TYPE_MOD;
       out[n].src_ext_mod.Absolute = 1;
+      out[n].src_ext_mod.Negate = src.Negate;
       n++;
    }
 
    if (src.Indirect) {
+      out[0].src.Indirect = 1;
       out[n].value = 0;
       out[n].src.File = TGSI_FILE_ADDRESS;
-      out[n].src.SwizzleX = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleY = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleZ = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleW = TGSI_SWIZZLE_X;
-      out[n].src.Indirect = 0;
-      out[n].src.Index = 0;
+      out[n].src.SwizzleX = src.IndirectSwizzle;
+      out[n].src.SwizzleY = src.IndirectSwizzle;
+      out[n].src.SwizzleZ = src.IndirectSwizzle;
+      out[n].src.SwizzleW = src.IndirectSwizzle;
+      out[n].src.Index = src.IndirectIndex;
       n++;
    }
 
@@ -484,6 +537,13 @@ ureg_emit_dst( struct ureg_program *ureg,
    union tgsi_any_token *out = get_tokens( ureg, DOMAIN_INSN, size );
    unsigned n = 0;
 
+   assert(dst.File != TGSI_FILE_NULL);
+   assert(dst.File != TGSI_FILE_CONSTANT);
+   assert(dst.File != TGSI_FILE_INPUT);
+   assert(dst.File != TGSI_FILE_SAMPLER);
+   assert(dst.File != TGSI_FILE_IMMEDIATE);
+   assert(dst.File < TGSI_FILE_COUNT);
+   
    out[n].value = 0;
    out[n].dst.File = dst.File;
    out[n].dst.WriteMask = dst.WriteMask;
@@ -494,12 +554,11 @@ ureg_emit_dst( struct ureg_program *ureg,
    if (dst.Indirect) {
       out[n].value = 0;
       out[n].src.File = TGSI_FILE_ADDRESS;
-      out[n].src.SwizzleX = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleY = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleZ = TGSI_SWIZZLE_X;
-      out[n].src.SwizzleW = TGSI_SWIZZLE_X;
-      out[n].src.Indirect = 0;
-      out[n].src.Index = 0;
+      out[n].src.SwizzleX = dst.IndirectSwizzle;
+      out[n].src.SwizzleY = dst.IndirectSwizzle;
+      out[n].src.SwizzleZ = dst.IndirectSwizzle;
+      out[n].src.SwizzleW = dst.IndirectSwizzle;
+      out[n].src.Index = dst.IndirectIndex;
       n++;
    }
 
@@ -523,7 +582,6 @@ ureg_emit_insn(struct ureg_program *ureg,
    out[0].insn.NrTokens = 0;
    out[0].insn.Opcode = opcode;
    out[0].insn.Saturate = saturate;
-   out[0].insn.NrTokens = 0;
    out[0].insn.NumDstRegs = num_dst;
    out[0].insn.NumSrcRegs = num_src;
    out[0].insn.Padding = 0;
@@ -542,6 +600,9 @@ ureg_emit_label(struct ureg_program *ureg,
 {
    union tgsi_any_token *out, *insn;
 
+   if(!label_token)
+      return;
+
    out = get_tokens( ureg, DOMAIN_INSN, 1 );
    insn = retrieve_token( ureg, DOMAIN_INSN, insn_token );
 
@@ -617,6 +678,17 @@ ureg_insn(struct ureg_program *ureg,
    unsigned insn, i;
    boolean saturate;
 
+#ifdef DEBUG
+   {
+      const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode );
+      assert(info);
+      if(info) {
+         assert(nr_dst == info->num_dst);
+         assert(nr_src == info->num_src);
+      }
+   }
+#endif
+   
    saturate = nr_dst ? dst[0].Saturate : FALSE;
 
    insn = ureg_emit_insn( ureg, opcode, saturate, nr_dst, nr_src );
@@ -723,10 +795,10 @@ static void emit_decls( struct ureg_program *ureg )
                  TGSI_INTERPOLATE_CONSTANT );
    }
 
-   if (ureg->nr_samplers) {
+   for (i = 0; i < ureg->nr_samplers; i++) {
       emit_decl_range( ureg, 
                        TGSI_FILE_SAMPLER,
-                       0, ureg->nr_samplers );
+                       ureg->sampler[i].Index, 1 );
    }
 
    if (ureg->nr_constants) {
@@ -741,6 +813,12 @@ static void emit_decls( struct ureg_program *ureg )
                        0, ureg->nr_temps );
    }
 
+   if (ureg->nr_addrs) {
+      emit_decl_range( ureg,
+                       TGSI_FILE_ADDRESS,
+                       0, ureg->nr_addrs );
+   }
+
    for (i = 0; i < ureg->nr_immediates; i++) {
       emit_immediate( ureg,
                       ureg->immediate[i].v );
@@ -764,7 +842,7 @@ static void copy_instructions( struct ureg_program *ureg )
 
 
 static void
-fixup_header_size(struct ureg_program *ureg )
+fixup_header_size(struct ureg_program *ureg)
 {
    union tgsi_any_token *out = retrieve_token( ureg, DOMAIN_DECL, 1 );
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index 8836a1ea0eb..acbca59040c 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -31,6 +31,10 @@
 #include "pipe/p_compiler.h"
 #include "pipe/p_shader_tokens.h"
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+   
 struct ureg_program;
 
 /* Almost a tgsi_src_register, but we need to pull in the Absolute
@@ -48,6 +52,8 @@ struct ureg_src
    unsigned Absolute    : 1;  /* BOOL */
    int      Index       : 16; /* SINT */
    unsigned Negate      : 1;  /* BOOL */
+   int      IndirectIndex   : 16; /* SINT */
+   int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
 };
 
 /* Very similar to a tgsi_dst_register, removing unsupported fields
@@ -64,6 +70,8 @@ struct ureg_dst
    int      Index       : 16; /* SINT */
    unsigned Pad1        : 5;
    unsigned Pad2        : 1;  /* BOOL */
+   int      IndirectIndex   : 16; /* SINT */
+   int      IndirectSwizzle : 2;  /* TGSI_SWIZZLE_ */
 };
 
 struct pipe_context;
@@ -131,12 +139,21 @@ void
 ureg_release_temporary( struct ureg_program *ureg,
                         struct ureg_dst tmp );
 
+struct ureg_dst
+ureg_DECL_address( struct ureg_program * );
+
+/* Supply an index to the sampler declaration as this is the hook to
+ * the external pipe_sampler state.  Users of this function probably
+ * don't want just any sampler, but a specific one which they've set
+ * up state for in the context.
+ */
 struct ureg_src
-ureg_DECL_sampler( struct ureg_program * );
+ureg_DECL_sampler( struct ureg_program *,
+                   unsigned index );
 
 
 static INLINE struct ureg_src
-ureg_DECL_immediate4f( struct ureg_program *ureg,
+ureg_imm4f( struct ureg_program *ureg,
                        float a, float b,
                        float c, float d)
 {
@@ -149,7 +166,7 @@ ureg_DECL_immediate4f( struct ureg_program *ureg,
 }
 
 static INLINE struct ureg_src
-ureg_DECL_immediate3f( struct ureg_program *ureg,
+ureg_imm3f( struct ureg_program *ureg,
                        float a, float b,
                        float c)
 {
@@ -161,7 +178,7 @@ ureg_DECL_immediate3f( struct ureg_program *ureg,
 }
 
 static INLINE struct ureg_src
-ureg_DECL_immediate2f( struct ureg_program *ureg,
+ureg_imm2f( struct ureg_program *ureg,
                        float a, float b)
 {
    float v[2];
@@ -171,7 +188,7 @@ ureg_DECL_immediate2f( struct ureg_program *ureg,
 }
 
 static INLINE struct ureg_src
-ureg_DECL_immediate1f( struct ureg_program *ureg,
+ureg_imm1f( struct ureg_program *ureg,
                        float a)
 {
    float v[1];
@@ -392,6 +409,7 @@ static INLINE void ureg_##op( struct ureg_program *ureg,                \
 static INLINE struct ureg_src 
 ureg_negate( struct ureg_src reg )
 {
+   assert(reg.File != TGSI_FILE_NULL);
    reg.Negate ^= 1;
    return reg;
 }
@@ -399,6 +417,7 @@ ureg_negate( struct ureg_src reg )
 static INLINE struct ureg_src
 ureg_abs( struct ureg_src reg )
 {
+   assert(reg.File != TGSI_FILE_NULL);
    reg.Absolute = 1;
    reg.Negate = 0;
    return reg;
@@ -413,6 +432,12 @@ ureg_swizzle( struct ureg_src reg,
                     (reg.SwizzleZ << 4) |
                     (reg.SwizzleW << 6));
 
+   assert(reg.File != TGSI_FILE_NULL);
+   assert(x < 4);
+   assert(y < 4);
+   assert(z < 4);
+   assert(w < 4);
+
    reg.SwizzleX = (swz >> (x*2)) & 0x3;
    reg.SwizzleY = (swz >> (y*2)) & 0x3;
    reg.SwizzleZ = (swz >> (z*2)) & 0x3;
@@ -430,6 +455,7 @@ static INLINE struct ureg_dst
 ureg_writemask( struct ureg_dst reg,
                 unsigned writemask )
 {
+   assert(reg.File != TGSI_FILE_NULL);
    reg.WriteMask &= writemask;
    return reg;
 }
@@ -437,10 +463,33 @@ ureg_writemask( struct ureg_dst reg,
 static INLINE struct ureg_dst 
 ureg_saturate( struct ureg_dst reg )
 {
+   assert(reg.File != TGSI_FILE_NULL);
    reg.Saturate = 1;
    return reg;
 }
 
+static INLINE struct ureg_dst 
+ureg_dst_indirect( struct ureg_dst reg, struct ureg_src addr )
+{
+   assert(reg.File != TGSI_FILE_NULL);
+   assert(addr.File == TGSI_FILE_ADDRESS);
+   reg.Indirect = 1;
+   reg.IndirectIndex = addr.Index;
+   reg.IndirectSwizzle = addr.SwizzleX;
+   return reg;
+}
+
+static INLINE struct ureg_src 
+ureg_src_indirect( struct ureg_src reg, struct ureg_src addr )
+{
+   assert(reg.File != TGSI_FILE_NULL);
+   assert(addr.File == TGSI_FILE_ADDRESS);
+   reg.Indirect = 1;
+   reg.IndirectIndex = addr.Index;
+   reg.IndirectSwizzle = addr.SwizzleX;
+   return reg;
+}
+
 static INLINE struct ureg_dst
 ureg_dst( struct ureg_src src )
 {
@@ -449,6 +498,8 @@ ureg_dst( struct ureg_src src )
    dst.File      = src.File;
    dst.WriteMask = TGSI_WRITEMASK_XYZW;
    dst.Indirect  = src.Indirect;
+   dst.IndirectIndex = src.IndirectIndex;
+   dst.IndirectSwizzle = src.IndirectSwizzle;
    dst.Saturate  = 0;
    dst.Index     = src.Index;
    dst.Pad1      = 0;
@@ -469,6 +520,8 @@ ureg_src( struct ureg_dst dst )
    src.SwizzleW  = TGSI_SWIZZLE_W;
    src.Pad       = 0;
    src.Indirect  = dst.Indirect;
+   src.IndirectIndex = dst.IndirectIndex;
+   src.IndirectSwizzle = dst.IndirectSwizzle;
    src.Absolute  = 0;
    src.Index     = dst.Index;
    src.Negate    = 0;
@@ -478,4 +531,60 @@ ureg_src( struct ureg_dst dst )
 
 
 
+static INLINE struct ureg_dst
+ureg_dst_undef( void )
+{
+   struct ureg_dst dst;
+
+   dst.File      = TGSI_FILE_NULL;
+   dst.WriteMask = 0;
+   dst.Indirect  = 0;
+   dst.IndirectIndex = 0;
+   dst.IndirectSwizzle = 0;
+   dst.Saturate  = 0;
+   dst.Index     = 0;
+   dst.Pad1      = 0;
+   dst.Pad2      = 0;
+
+   return dst;
+}
+
+static INLINE struct ureg_src
+ureg_src_undef( void )
+{
+   struct ureg_src src;
+
+   src.File      = TGSI_FILE_NULL;
+   src.SwizzleX  = 0;
+   src.SwizzleY  = 0;
+   src.SwizzleZ  = 0;
+   src.SwizzleW  = 0;
+   src.Pad       = 0;
+   src.Indirect  = 0;
+   src.IndirectIndex = 0;
+   src.IndirectSwizzle = 0;
+   src.Absolute  = 0;
+   src.Index     = 0;
+   src.Negate    = 0;
+   
+   return src;
+}
+
+static INLINE boolean
+ureg_src_is_undef( struct ureg_src src )
+{
+   return src.File == TGSI_FILE_NULL;
+}
+
+static INLINE boolean
+ureg_dst_is_undef( struct ureg_dst dst )
+{
+   return dst.File == TGSI_FILE_NULL;
+}
+
+
+#ifdef __cplusplus
+}
+#endif
+
 #endif
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.h b/src/gallium/auxiliary/tgsi/tgsi_util.h
index 7877f345587..21eb656327e 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_util.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.h
@@ -32,6 +32,10 @@
 extern "C" {
 #endif
 
+struct tgsi_src_register;
+struct tgsi_src_register_ext_swz;
+struct tgsi_full_src_register;
+
 void *
 tgsi_align_128bit(
    void *unaligned );