Merge branch 'gallium-polygon-stipple'

[mesa.git] / src / gallium / auxiliary / tgsi / tgsi_exec.c

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 2bcb33392a8cec174029c44b9b5474e7b739e54d..712e8aca794c4252804f8bb76f799bde68b3e60f 100644 (file)
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -109,6 +109,42 @@ micro_ceil(union tgsi_exec_channel *dst,
    dst->f[3] = ceilf(src->f[3]);
 }
 
+static void
+micro_clamp(union tgsi_exec_channel *dst,
+            const union tgsi_exec_channel *src0,
+            const union tgsi_exec_channel *src1,
+            const union tgsi_exec_channel *src2)
+{
+   dst->f[0] = src0->f[0] < src1->f[0] ? src1->f[0] : src0->f[0] > src2->f[0] ? src2->f[0] : src0->f[0];
+   dst->f[1] = src0->f[1] < src1->f[1] ? src1->f[1] : src0->f[1] > src2->f[1] ? src2->f[1] : src0->f[1];
+   dst->f[2] = src0->f[2] < src1->f[2] ? src1->f[2] : src0->f[2] > src2->f[2] ? src2->f[2] : src0->f[2];
+   dst->f[3] = src0->f[3] < src1->f[3] ? src1->f[3] : src0->f[3] > src2->f[3] ? src2->f[3] : src0->f[3];
+}
+
+static void
+micro_cmp(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1,
+          const union tgsi_exec_channel *src2)
+{
+   dst->f[0] = src0->f[0] < 0.0f ? src1->f[0] : src2->f[0];
+   dst->f[1] = src0->f[1] < 0.0f ? src1->f[1] : src2->f[1];
+   dst->f[2] = src0->f[2] < 0.0f ? src1->f[2] : src2->f[2];
+   dst->f[3] = src0->f[3] < 0.0f ? src1->f[3] : src2->f[3];
+}
+
+static void
+micro_cnd(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1,
+          const union tgsi_exec_channel *src2)
+{
+   dst->f[0] = src2->f[0] > 0.5f ? src0->f[0] : src1->f[0];
+   dst->f[1] = src2->f[1] > 0.5f ? src0->f[1] : src1->f[1];
+   dst->f[2] = src2->f[2] > 0.5f ? src0->f[2] : src1->f[2];
+   dst->f[3] = src2->f[3] > 0.5f ? src0->f[3] : src1->f[3];
+}
+
 static void
 micro_cos(union tgsi_exec_channel *dst,
           const union tgsi_exec_channel *src)
@@ -232,22 +268,26 @@ micro_lg2(union tgsi_exec_channel *dst,
 
 static void
 micro_lrp(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1,
+          const union tgsi_exec_channel *src2)
 {
-   dst->f[0] = src[0].f[0] * (src[1].f[0] - src[2].f[0]) + src[2].f[0];
-   dst->f[1] = src[0].f[1] * (src[1].f[1] - src[2].f[1]) + src[2].f[1];
-   dst->f[2] = src[0].f[2] * (src[1].f[2] - src[2].f[2]) + src[2].f[2];
-   dst->f[3] = src[0].f[3] * (src[1].f[3] - src[2].f[3]) + src[2].f[3];
+   dst->f[0] = src0->f[0] * (src1->f[0] - src2->f[0]) + src2->f[0];
+   dst->f[1] = src0->f[1] * (src1->f[1] - src2->f[1]) + src2->f[1];
+   dst->f[2] = src0->f[2] * (src1->f[2] - src2->f[2]) + src2->f[2];
+   dst->f[3] = src0->f[3] * (src1->f[3] - src2->f[3]) + src2->f[3];
 }
 
 static void
 micro_mad(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1,
+          const union tgsi_exec_channel *src2)
 {
-   dst->f[0] = src[0].f[0] * src[1].f[0] + src[2].f[0];
-   dst->f[1] = src[0].f[1] * src[1].f[1] + src[2].f[1];
-   dst->f[2] = src[0].f[2] * src[1].f[2] + src[2].f[2];
-   dst->f[3] = src[0].f[3] * src[1].f[3] + src[2].f[3];
+   dst->f[0] = src0->f[0] * src1->f[0] + src2->f[0];
+   dst->f[1] = src0->f[1] * src1->f[1] + src2->f[1];
+   dst->f[2] = src0->f[2] * src1->f[2] + src2->f[2];
+   dst->f[3] = src0->f[3] * src1->f[3] + src2->f[3];
 }
 
 static void
@@ -264,6 +304,12 @@ static void
 micro_rcp(union tgsi_exec_channel *dst,
           const union tgsi_exec_channel *src)
 {
+#if 0 /* for debugging */
+   assert(src->f[0] != 0.0f);
+   assert(src->f[1] != 0.0f);
+   assert(src->f[2] != 0.0f);
+   assert(src->f[3] != 0.0f);
+#endif
    dst->f[0] = 1.0f / src->f[0];
    dst->f[1] = 1.0f / src->f[1];
    dst->f[2] = 1.0f / src->f[2];
@@ -284,6 +330,12 @@ static void
 micro_rsq(union tgsi_exec_channel *dst,
           const union tgsi_exec_channel *src)
 {
+#if 0 /* for debugging */
+   assert(src->f[0] != 0.0f);
+   assert(src->f[1] != 0.0f);
+   assert(src->f[2] != 0.0f);
+   assert(src->f[3] != 0.0f);
+#endif
    dst->f[0] = 1.0f / sqrtf(fabsf(src->f[0]));
    dst->f[1] = 1.0f / sqrtf(fabsf(src->f[1]));
    dst->f[2] = 1.0f / sqrtf(fabsf(src->f[2]));
@@ -292,22 +344,24 @@ micro_rsq(union tgsi_exec_channel *dst,
 
 static void
 micro_seq(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->f[0] = src[0].f[0] == src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] == src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] == src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] == src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = src0->f[0] == src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] == src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] == src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] == src1->f[3] ? 1.0f : 0.0f;
 }
 
 static void
 micro_sge(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->f[0] = src[0].f[0] >= src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] >= src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] >= src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] >= src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = src0->f[0] >= src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] >= src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] >= src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] >= src1->f[3] ? 1.0f : 0.0f;
 }
 
 static void
@@ -322,12 +376,13 @@ micro_sgn(union tgsi_exec_channel *dst,
 
 static void
 micro_sgt(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->f[0] = src[0].f[0] > src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] > src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] > src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] > src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = src0->f[0] > src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] > src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] > src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] > src1->f[3] ? 1.0f : 0.0f;
 }
 
 static void
@@ -342,32 +397,53 @@ micro_sin(union tgsi_exec_channel *dst,
 
 static void
 micro_sle(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->f[0] = src[0].f[0] <= src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] <= src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] <= src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] <= src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = src0->f[0] <= src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] <= src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] <= src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] <= src1->f[3] ? 1.0f : 0.0f;
 }
 
 static void
 micro_slt(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->f[0] = src[0].f[0] < src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] < src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] < src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] < src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = src0->f[0] < src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] < src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] < src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] < src1->f[3] ? 1.0f : 0.0f;
 }
 
 static void
 micro_sne(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
+{
+   dst->f[0] = src0->f[0] != src1->f[0] ? 1.0f : 0.0f;
+   dst->f[1] = src0->f[1] != src1->f[1] ? 1.0f : 0.0f;
+   dst->f[2] = src0->f[2] != src1->f[2] ? 1.0f : 0.0f;
+   dst->f[3] = src0->f[3] != src1->f[3] ? 1.0f : 0.0f;
+}
+
+static void
+micro_sfl(union tgsi_exec_channel *dst)
+{
+   dst->f[0] = 0.0f;
+   dst->f[1] = 0.0f;
+   dst->f[2] = 0.0f;
+   dst->f[3] = 0.0f;
+}
+
+static void
+micro_str(union tgsi_exec_channel *dst)
 {
-   dst->f[0] = src[0].f[0] != src[1].f[0] ? 1.0f : 0.0f;
-   dst->f[1] = src[0].f[1] != src[1].f[1] ? 1.0f : 0.0f;
-   dst->f[2] = src[0].f[2] != src[1].f[2] ? 1.0f : 0.0f;
-   dst->f[3] = src[0].f[3] != src[1].f[3] ? 1.0f : 0.0f;
+   dst->f[0] = 1.0f;
+   dst->f[1] = 1.0f;
+   dst->f[2] = 1.0f;
+   dst->f[3] = 1.0f;
 }
 
 static void
@@ -395,50 +471,12 @@ enum tgsi_exec_datatype {
 /*
  * Shorthand locations of various utility registers (_I = Index, _C = Channel)
  */
-#define TEMP_0_I           TGSI_EXEC_TEMP_00000000_I
-#define TEMP_0_C           TGSI_EXEC_TEMP_00000000_C
-#define TEMP_7F_I          TGSI_EXEC_TEMP_7FFFFFFF_I
-#define TEMP_7F_C          TGSI_EXEC_TEMP_7FFFFFFF_C
-#define TEMP_80_I          TGSI_EXEC_TEMP_80000000_I
-#define TEMP_80_C          TGSI_EXEC_TEMP_80000000_C
-#define TEMP_FF_I          TGSI_EXEC_TEMP_FFFFFFFF_I
-#define TEMP_FF_C          TGSI_EXEC_TEMP_FFFFFFFF_C
-#define TEMP_1_I           TGSI_EXEC_TEMP_ONE_I
-#define TEMP_1_C           TGSI_EXEC_TEMP_ONE_C
-#define TEMP_2_I           TGSI_EXEC_TEMP_TWO_I
-#define TEMP_2_C           TGSI_EXEC_TEMP_TWO_C
-#define TEMP_128_I         TGSI_EXEC_TEMP_128_I
-#define TEMP_128_C         TGSI_EXEC_TEMP_128_C
-#define TEMP_M128_I        TGSI_EXEC_TEMP_MINUS_128_I
-#define TEMP_M128_C        TGSI_EXEC_TEMP_MINUS_128_C
 #define TEMP_KILMASK_I     TGSI_EXEC_TEMP_KILMASK_I
 #define TEMP_KILMASK_C     TGSI_EXEC_TEMP_KILMASK_C
 #define TEMP_OUTPUT_I      TGSI_EXEC_TEMP_OUTPUT_I
 #define TEMP_OUTPUT_C      TGSI_EXEC_TEMP_OUTPUT_C
 #define TEMP_PRIMITIVE_I   TGSI_EXEC_TEMP_PRIMITIVE_I
 #define TEMP_PRIMITIVE_C   TGSI_EXEC_TEMP_PRIMITIVE_C
-#define TEMP_CC_I          TGSI_EXEC_TEMP_CC_I
-#define TEMP_CC_C          TGSI_EXEC_TEMP_CC_C
-#define TEMP_3_I           TGSI_EXEC_TEMP_THREE_I
-#define TEMP_3_C           TGSI_EXEC_TEMP_THREE_C
-#define TEMP_HALF_I        TGSI_EXEC_TEMP_HALF_I
-#define TEMP_HALF_C        TGSI_EXEC_TEMP_HALF_C
-#define TEMP_R0            TGSI_EXEC_TEMP_R0
-#define TEMP_P0            TGSI_EXEC_TEMP_P0
-
-#define IS_CHANNEL_ENABLED(INST, CHAN)\
-   ((INST).Dst[0].Register.WriteMask & (1 << (CHAN)))
-
-#define IS_CHANNEL_ENABLED2(INST, CHAN)\
-   ((INST).Dst[1].Register.WriteMask & (1 << (CHAN)))
-
-#define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
-   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
-      if (IS_CHANNEL_ENABLED( INST, CHAN ))
-
-#define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
-   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
-      if (IS_CHANNEL_ENABLED2( INST, CHAN ))
 
 
 /** The execution mask depends on the conditional mask and the loop mask */
@@ -449,13 +487,33 @@ enum tgsi_exec_datatype {
 static const union tgsi_exec_channel ZeroVec =
    { { 0.0, 0.0, 0.0, 0.0 } };
 
+static const union tgsi_exec_channel OneVec = {
+   {1.0f, 1.0f, 1.0f, 1.0f}
+};
+
+static const union tgsi_exec_channel P128Vec = {
+   {128.0f, 128.0f, 128.0f, 128.0f}
+};
+
+static const union tgsi_exec_channel M128Vec = {
+   {-128.0f, -128.0f, -128.0f, -128.0f}
+};
+
 
-#define CHECK_INF_OR_NAN(chan) do {\
-      assert(!util_is_inf_or_nan((chan)->f[0]));\
-      assert(!util_is_inf_or_nan((chan)->f[1]));\
-      assert(!util_is_inf_or_nan((chan)->f[2]));\
-      assert(!util_is_inf_or_nan((chan)->f[3]));\
-   } while (0)
+/**
+ * Assert that none of the float values in 'chan' are infinite or NaN.
+ * NaN and Inf may occur normally during program execution and should
+ * not lead to crashes, etc.  But when debugging, it's helpful to catch
+ * them.
+ */
+static INLINE void
+check_inf_or_nan(const union tgsi_exec_channel *chan)
+{
+   assert(!util_is_inf_or_nan((chan)->f[0]));
+   assert(!util_is_inf_or_nan((chan)->f[1]));
+   assert(!util_is_inf_or_nan((chan)->f[2]));
+   assert(!util_is_inf_or_nan((chan)->f[3]));
+}
 
 
 #ifdef DEBUG
@@ -487,6 +545,21 @@ print_temp(const struct tgsi_exec_machine *mach, uint index)
 #endif
 
 
+void
+tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
+                               unsigned num_bufs,
+                               const void **bufs,
+                               const unsigned *buf_sizes)
+{
+   unsigned i;
+
+   for (i = 0; i < num_bufs; i++) {
+      mach->Consts[i] = bufs[i];
+      mach->ConstsSize[i] = buf_sizes[i];
+   }
+}
+
+
 /**
  * Check if there's a potential src/dst register data dependency when
  * using SOA execution.
@@ -518,18 +591,22 @@ tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst)
    for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
       if ((inst->Src[i].Register.File ==
            inst->Dst[0].Register.File) &&
-          (inst->Src[i].Register.Index ==
-           inst->Dst[0].Register.Index)) {
+          ((inst->Src[i].Register.Index ==
+            inst->Dst[0].Register.Index) ||
+           inst->Src[i].Register.Indirect ||
+           inst->Dst[0].Register.Indirect)) {
          /* loop over dest channels */
          uint channelsWritten = 0x0;
-         FOR_EACH_ENABLED_CHANNEL(*inst, chan) {
-            /* check if we're reading a channel that's been written */
-            uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan);
-            if (channelsWritten & (1 << swizzle)) {
-               return TRUE;
-            }
+         for (chan = 0; chan < NUM_CHANNELS; chan++) {
+            if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+               /* check if we're reading a channel that's been written */
+               uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan);
+               if (channelsWritten & (1 << swizzle)) {
+                  return TRUE;
+               }
 
-            channelsWritten |= (1 << chan);
+               channelsWritten |= (1 << chan);
+            }
          }
       }
    }
@@ -551,12 +628,10 @@ tgsi_exec_machine_bind_shader(
 {
    uint k;
    struct tgsi_parse_context parse;
-   struct tgsi_exec_labels *labels = &mach->Labels;
    struct tgsi_full_instruction *instructions;
    struct tgsi_full_declaration *declarations;
    uint maxInstructions = 10, numInstructions = 0;
    uint maxDeclarations = 10, numDeclarations = 0;
-   uint instno = 0;
 
 #if 0
    tgsi_dump(tokens, 0);
@@ -564,9 +639,30 @@ tgsi_exec_machine_bind_shader(
 
    util_init_math();
 
+   if (numSamplers) {
+      assert(samplers);
+   }
+
    mach->Tokens = tokens;
    mach->Samplers = samplers;
 
+   if (!tokens) {
+      /* unbind and free all */
+      if (mach->Declarations) {
+         FREE( mach->Declarations );
+      }
+      mach->Declarations = NULL;
+      mach->NumDeclarations = 0;
+
+      if (mach->Instructions) {
+         FREE( mach->Instructions );
+      }
+      mach->Instructions = NULL;
+      mach->NumInstructions = 0;
+
+      return;
+   }
+
    k = tgsi_parse_init (&parse, mach->Tokens);
    if (k != TGSI_PARSE_OK) {
       debug_printf( "Problem parsing!\n" );
@@ -575,7 +671,31 @@ tgsi_exec_machine_bind_shader(
 
    mach->Processor = parse.FullHeader.Processor.Processor;
    mach->ImmLimit = 0;
-   labels->count = 0;
+
+   if (mach->Processor == TGSI_PROCESSOR_GEOMETRY &&
+       !mach->UsedGeometryShader) {
+      struct tgsi_exec_vector *inputs =
+         align_malloc(sizeof(struct tgsi_exec_vector) *
+                      TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS,
+                      16);
+      struct tgsi_exec_vector *outputs =
+         align_malloc(sizeof(struct tgsi_exec_vector) *
+                      TGSI_MAX_TOTAL_VERTICES, 16);
+
+      if (!inputs)
+         return;
+      if (!outputs) {
+         align_free(inputs);
+         return;
+      }
+
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
+
+      mach->Inputs = inputs;
+      mach->Outputs = outputs;
+      mach->UsedGeometryShader = TRUE;
+   }
 
    declarations = (struct tgsi_full_declaration *)
       MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
@@ -593,7 +713,6 @@ tgsi_exec_machine_bind_shader(
    }
 
    while( !tgsi_parse_end_of_tokens( &parse ) ) {
-      uint pointer = parse.Position;
       uint i;
 
       tgsi_parse_token( &parse );
@@ -616,6 +735,19 @@ tgsi_exec_machine_bind_shader(
                ++mach->NumOutputs;
             }
          }
+         if (parse.FullToken.FullDeclaration.Declaration.File ==
+             TGSI_FILE_IMMEDIATE_ARRAY) {
+            unsigned reg;
+            struct tgsi_full_declaration *decl =
+               &parse.FullToken.FullDeclaration;
+            debug_assert(decl->Range.Last < TGSI_EXEC_NUM_IMMEDIATES);
+            for (reg = decl->Range.First; reg <= decl->Range.Last; ++reg) {
+               for( i = 0; i < 4; i++ ) {
+                  int idx = reg * 4 + i;
+                  mach->ImmArray[reg][i] = decl->ImmediateData.u[idx].Float;
+               }
+            }
+         }
          memcpy(declarations + numDeclarations,
                 &parse.FullToken.FullDeclaration,
                 sizeof(declarations[0]));
@@ -637,11 +769,6 @@ tgsi_exec_machine_bind_shader(
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
-         assert( labels->count < MAX_LABELS );
-
-         labels->labels[labels->count][0] = instno;
-         labels->labels[labels->count][1] = pointer;
-         labels->count++;
 
          /* save expanded instruction */
          if (numInstructions == maxInstructions) {
@@ -699,18 +826,23 @@ tgsi_exec_machine_create( void )
    mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES;
    mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0];
 
-   /* Setup constants. */
+   mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
+   mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
+   if (!mach->Inputs || !mach->Outputs)
+      goto fail;
+
+   /* Setup constants needed by the SSE2 executor. */
    for( i = 0; i < 4; i++ ) {
-      mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].u[i] = 0x00000000;
-      mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].u[i] = 0x7FFFFFFF;
-      mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].u[i] = 0x80000000;
-      mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].u[i] = 0xFFFFFFFF;
-      mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].f[i] = 1.0f;
-      mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].f[i] = 2.0f;
-      mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].f[i] = 128.0f;
-      mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].f[i] = -128.0f;
-      mach->Temps[TEMP_3_I].xyzw[TEMP_3_C].f[i] = 3.0f;
-      mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C].f[i] = 0.5f;
+      mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000;
+      mach->Temps[TGSI_EXEC_TEMP_7FFFFFFF_I].xyzw[TGSI_EXEC_TEMP_7FFFFFFF_C].u[i] = 0x7FFFFFFF;
+      mach->Temps[TGSI_EXEC_TEMP_80000000_I].xyzw[TGSI_EXEC_TEMP_80000000_C].u[i] = 0x80000000;
+      mach->Temps[TGSI_EXEC_TEMP_FFFFFFFF_I].xyzw[TGSI_EXEC_TEMP_FFFFFFFF_C].u[i] = 0xFFFFFFFF;    /* not used */
+      mach->Temps[TGSI_EXEC_TEMP_ONE_I].xyzw[TGSI_EXEC_TEMP_ONE_C].f[i] = 1.0f;
+      mach->Temps[TGSI_EXEC_TEMP_TWO_I].xyzw[TGSI_EXEC_TEMP_TWO_C].f[i] = 2.0f;    /* not used */
+      mach->Temps[TGSI_EXEC_TEMP_128_I].xyzw[TGSI_EXEC_TEMP_128_C].f[i] = 128.0f;
+      mach->Temps[TGSI_EXEC_TEMP_MINUS_128_I].xyzw[TGSI_EXEC_TEMP_MINUS_128_C].f[i] = -128.0f;
+      mach->Temps[TGSI_EXEC_TEMP_THREE_I].xyzw[TGSI_EXEC_TEMP_THREE_C].f[i] = 3.0f;
+      mach->Temps[TGSI_EXEC_TEMP_HALF_I].xyzw[TGSI_EXEC_TEMP_HALF_C].f[i] = 0.5f;
    }
 
 #ifdef DEBUG
@@ -722,7 +854,11 @@ tgsi_exec_machine_create( void )
    return mach;
 
 fail:
-   align_free(mach);
+   if (mach) {
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
+      align_free(mach);
+   }
    return NULL;
 }
 
@@ -731,18 +867,22 @@ void
 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
 {
    if (mach) {
-      FREE(mach->Instructions);
-      FREE(mach->Declarations);
-   }
+      if (mach->Instructions)
+         FREE(mach->Instructions);
+      if (mach->Declarations)
+         FREE(mach->Declarations);
+
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
 
-   align_free(mach);
+      align_free(mach);
+   }
 }
 
 static void
-micro_add(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
+micro_add(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
    dst->f[0] = src0->f[0] + src1->f[0];
    dst->f[1] = src0->f[1] + src1->f[1];
@@ -771,27 +911,35 @@ micro_div(
 }
 
 static void
-micro_float_clamp(union tgsi_exec_channel *dst,
-                  const union tgsi_exec_channel *src)
+micro_rcc(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src)
 {
    uint i;
 
    for (i = 0; i < 4; i++) {
-      if (src->f[i] > 0.0f) {
-         if (src->f[i] > 1.884467e+019f)
+      float recip = 1.0f / src->f[i];
+
+      if (recip > 0.0f) {
+         if (recip > 1.884467e+019f) {
             dst->f[i] = 1.884467e+019f;
-         else if (src->f[i] < 5.42101e-020f)
+         }
+         else if (recip < 5.42101e-020f) {
             dst->f[i] = 5.42101e-020f;
-         else
-            dst->f[i] = src->f[i];
+         }
+         else {
+            dst->f[i] = recip;
+         }
       }
       else {
-         if (src->f[i] < -1.884467e+019f)
+         if (recip < -1.884467e+019f) {
             dst->f[i] = -1.884467e+019f;
-         else if (src->f[i] > -5.42101e-020f)
+         }
+         else if (recip > -5.42101e-020f) {
             dst->f[i] = -5.42101e-020f;
-         else
-            dst->f[i] = src->f[i];
+         }
+         else {
+            dst->f[i] = recip;
+         }
       }
    }
 }
@@ -811,10 +959,9 @@ micro_lt(
 }
 
 static void
-micro_max(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
+micro_max(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
    dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0];
    dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1];
@@ -823,10 +970,9 @@ micro_max(
 }
 
 static void
-micro_min(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
+micro_min(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
    dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0];
    dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1];
@@ -835,10 +981,9 @@ micro_min(
 }
 
 static void
-micro_mul(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
+micro_mul(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
    dst->f[0] = src0->f[0] * src1->f[0];
    dst->f[1] = src0->f[1] * src1->f[1];
@@ -846,60 +991,6 @@ micro_mul(
    dst->f[3] = src0->f[3] * src1->f[3];
 }
 
-#if 0
-static void
-micro_imul64(
-   union tgsi_exec_channel *dst0,
-   union tgsi_exec_channel *dst1,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
-{
-   dst1->i[0] = src0->i[0] * src1->i[0];
-   dst1->i[1] = src0->i[1] * src1->i[1];
-   dst1->i[2] = src0->i[2] * src1->i[2];
-   dst1->i[3] = src0->i[3] * src1->i[3];
-   dst0->i[0] = 0;
-   dst0->i[1] = 0;
-   dst0->i[2] = 0;
-   dst0->i[3] = 0;
-}
-#endif
-
-#if 0
-static void
-micro_umul64(
-   union tgsi_exec_channel *dst0,
-   union tgsi_exec_channel *dst1,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
-{
-   dst1->u[0] = src0->u[0] * src1->u[0];
-   dst1->u[1] = src0->u[1] * src1->u[1];
-   dst1->u[2] = src0->u[2] * src1->u[2];
-   dst1->u[3] = src0->u[3] * src1->u[3];
-   dst0->u[0] = 0;
-   dst0->u[1] = 0;
-   dst0->u[2] = 0;
-   dst0->u[3] = 0;
-}
-#endif
-
-
-#if 0
-static void
-micro_movc(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1,
-   const union tgsi_exec_channel *src2 )
-{
-   dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0];
-   dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1];
-   dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2];
-   dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3];
-}
-#endif
-
 static void
 micro_neg(
    union tgsi_exec_channel *dst,
@@ -931,20 +1022,9 @@ micro_pow(
 }
 
 static void
-micro_sqrt( union tgsi_exec_channel *dst,
-            const union tgsi_exec_channel *src )
-{
-   dst->f[0] = sqrtf( src->f[0] );
-   dst->f[1] = sqrtf( src->f[1] );
-   dst->f[2] = sqrtf( src->f[2] );
-   dst->f[3] = sqrtf( src->f[3] );
-}
-
-static void
-micro_sub(
-   union tgsi_exec_channel *dst,
-   const union tgsi_exec_channel *src0,
-   const union tgsi_exec_channel *src1 )
+micro_sub(union tgsi_exec_channel *dst,
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
    dst->f[0] = src0->f[0] - src1->f[0];
    dst->f[1] = src0->f[1] - src1->f[1];
@@ -953,99 +1033,140 @@ micro_sub(
 }
 
 static void
-fetch_src_file_channel(
-   const struct tgsi_exec_machine *mach,
-   const uint file,
-   const uint swizzle,
-   const union tgsi_exec_channel *index,
-   union tgsi_exec_channel *chan )
-{
-   switch( swizzle ) {
-   case TGSI_SWIZZLE_X:
-   case TGSI_SWIZZLE_Y:
-   case TGSI_SWIZZLE_Z:
-   case TGSI_SWIZZLE_W:
-      switch( file ) {
-      case TGSI_FILE_CONSTANT:
-         assert(mach->Consts);
-         if (index->i[0] < 0)
-            chan->f[0] = 0.0f;
-         else
-            chan->f[0] = mach->Consts[index->i[0]][swizzle];
-         if (index->i[1] < 0)
-            chan->f[1] = 0.0f;
-         else
-            chan->f[1] = mach->Consts[index->i[1]][swizzle];
-         if (index->i[2] < 0)
-            chan->f[2] = 0.0f;
-         else
-            chan->f[2] = mach->Consts[index->i[2]][swizzle];
-         if (index->i[3] < 0)
-            chan->f[3] = 0.0f;
-         else
-            chan->f[3] = mach->Consts[index->i[3]][swizzle];
-         break;
+fetch_src_file_channel(const struct tgsi_exec_machine *mach,
+                       const uint file,
+                       const uint swizzle,
+                       const union tgsi_exec_channel *index,
+                       const union tgsi_exec_channel *index2D,
+                       union tgsi_exec_channel *chan)
+{
+   uint i;
 
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   assert(swizzle < 4);
 
-      case TGSI_FILE_TEMPORARY:
-         assert(index->i[0] < TGSI_EXEC_NUM_TEMPS);
-         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   switch (file) {
+   case TGSI_FILE_CONSTANT:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS);
+         assert(mach->Consts[index2D->i[i]]);
 
-      case TGSI_FILE_IMMEDIATE:
-         assert( index->i[0] < (int) mach->ImmLimit );
-         chan->f[0] = mach->Imms[index->i[0]][swizzle];
-         assert( index->i[1] < (int) mach->ImmLimit );
-         chan->f[1] = mach->Imms[index->i[1]][swizzle];
-         assert( index->i[2] < (int) mach->ImmLimit );
-         chan->f[2] = mach->Imms[index->i[2]][swizzle];
-         assert( index->i[3] < (int) mach->ImmLimit );
-         chan->f[3] = mach->Imms[index->i[3]][swizzle];
-         break;
+         if (index->i[i] < 0) {
+            chan->u[i] = 0;
+         } else {
+            /* NOTE: copying the const value as a uint instead of float */
+            const uint constbuf = index2D->i[i];
+            const uint *buf = (const uint *)mach->Consts[constbuf];
+            const int pos = index->i[i] * 4 + swizzle;
+            /* const buffer bounds check */
+            if (pos < 0 || pos >= mach->ConstsSize[constbuf]) {
+               if (0) {
+                  /* Debug: print warning */
+                  static int count = 0;
+                  if (count++ < 100)
+                     debug_printf("TGSI Exec: const buffer index %d"
+                                  " out of bounds\n", pos);
+               }
+               chan->u[i] = 0;
+            }
+            else
+               chan->u[i] = buf[pos];
+         }
+      }
+      break;
 
-      case TGSI_FILE_ADDRESS:
-         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_INPUT:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         /*
+         if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
+            debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
+                         index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
+                         index2D->i[i], index->i[i]);
+                         }*/
+         int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i];
+         assert(pos >= 0);
+         assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS);
+         chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i];
+      }
+      break;
 
-      case TGSI_FILE_PREDICATE:
-         assert(index->i[0] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[1] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[2] < TGSI_EXEC_NUM_PREDS);
-         assert(index->i[3] < TGSI_EXEC_NUM_PREDS);
-         chan->u[0] = mach->Predicates[0].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Predicates[0].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Predicates[0].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Predicates[0].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_SYSTEM_VALUE:
+      /* XXX no swizzling at this point.  Will be needed if we put
+       * gl_FragCoord, for example, in a sys value register.
+       */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         chan->f[i] = mach->SystemValue[index->i[i]][0];
+      }
+      break;
 
-      case TGSI_FILE_OUTPUT:
-         /* vertex/fragment output vars can be read too */
-         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
-         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
-         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
-         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
-         break;
+   case TGSI_FILE_TEMPORARY:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
+         assert(index2D->i[i] == 0);
 
-      default:
-         assert( 0 );
+         chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_TEMPORARY_ARRAY:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
+         assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS);
+
+         chan->u[i] =
+            mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_IMMEDIATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit);
+         assert(index2D->i[i] == 0);
+
+         chan->f[i] = mach->Imms[index->i[i]][swizzle];
+      }
+      break;
+
+   case TGSI_FILE_IMMEDIATE_ARRAY:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index2D->i[i] == 0);
+
+         chan->f[i] = mach->ImmArray[index->i[i]][swizzle];
+      }
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_PREDICATE:
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i];
+      }
+      break;
+
+   case TGSI_FILE_OUTPUT:
+      /* vertex/fragment output vars can be read too */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         assert(index->i[i] >= 0);
+         assert(index2D->i[i] == 0);
+
+         chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i];
       }
       break;
 
    default:
-      assert( 0 );
+      assert(0);
+      for (i = 0; i < QUAD_SIZE; i++) {
+         chan->u[i] = 0;
+      }
    }
 }
 
@@ -1057,6 +1178,7 @@ fetch_source(const struct tgsi_exec_machine *mach,
              enum tgsi_exec_datatype src_datatype)
 {
    union tgsi_exec_channel index;
+   union tgsi_exec_channel index2D;
    uint swizzle;
 
    /* We start with a direct index into a register file.
@@ -1092,15 +1214,15 @@ fetch_source(const struct tgsi_exec_machine *mach,
       index2.i[1] =
       index2.i[2] =
       index2.i[3] = reg->Indirect.Index;
-
+      assert(reg->Indirect.File == TGSI_FILE_ADDRESS);
       /* get current value of address register[swizzle] */
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index2,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index2,
+                             &ZeroVec,
+                             &indir_index);
 
       /* add value of address register to the offset */
       index.i[0] += indir_index.i[0];
@@ -1121,44 +1243,22 @@ fetch_source(const struct tgsi_exec_machine *mach,
     * subscript to a register file. Effectively it means that
     * the register file is actually a 2D array of registers.
     *
-    *    file[1][3] == file[1*sizeof(file[1])+3],
+    *    file[3][1],
     *    where:
     *       [3] = Dimension.Index
     */
    if (reg->Register.Dimension) {
-      /* The size of the first-order array depends on the register file type.
-       * We need to multiply the index to the first array to get an effective,
-       * "flat" index that points to the beginning of the second-order array.
-       */
-      switch (reg->Register.File) {
-      case TGSI_FILE_INPUT:
-      case TGSI_FILE_SYSTEM_VALUE:
-         index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
-         break;
-      case TGSI_FILE_CONSTANT:
-         index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER;
-         break;
-      default:
-         assert( 0 );
-      }
-
-      index.i[0] += reg->Dimension.Index;
-      index.i[1] += reg->Dimension.Index;
-      index.i[2] += reg->Dimension.Index;
-      index.i[3] += reg->Dimension.Index;
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = reg->Dimension.Index;
 
       /* Again, the second subscript index can be addressed indirectly
        * identically to the first one.
        * Nothing stops us from indirectly addressing the indirect register,
        * but there is no need for that, so we won't exercise it.
        *
-       *    file[1][ind[4].y+3],
+       *    file[ind[4].y+3][1],
        *    where:
        *       ind = DimIndirect.File
        *       [4] = DimIndirect.Index
@@ -1176,24 +1276,25 @@ fetch_source(const struct tgsi_exec_machine *mach,
          index2.i[3] = reg->DimIndirect.Index;
 
          swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
-         fetch_src_file_channel(
-            mach,
-            reg->DimIndirect.File,
-            swizzle,
-            &index2,
-            &indir_index );
-
-         index.i[0] += indir_index.i[0];
-         index.i[1] += indir_index.i[1];
-         index.i[2] += indir_index.i[2];
-         index.i[3] += indir_index.i[3];
+         fetch_src_file_channel(mach,
+                                reg->DimIndirect.File,
+                                swizzle,
+                                &index2,
+                                &ZeroVec,
+                                &indir_index);
+
+         index2D.i[0] += indir_index.i[0];
+         index2D.i[1] += indir_index.i[1];
+         index2D.i[2] += indir_index.i[2];
+         index2D.i[3] += indir_index.i[3];
 
          /* for disabled execution channels, zero-out the index to
           * avoid using a potential garbage value.
           */
          for (i = 0; i < QUAD_SIZE; i++) {
-            if ((execmask & (1 << i)) == 0)
-               index.i[i] = 0;
+            if ((execmask & (1 << i)) == 0) {
+               index2D.i[i] = 0;
+            }
          }
       }
 
@@ -1201,15 +1302,20 @@ fetch_source(const struct tgsi_exec_machine *mach,
        * files, we would have to check whether Dimension is followed
        * by a dimension register and continue the saga.
        */
+   } else {
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = 0;
    }
 
    swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
-   fetch_src_file_channel(
-      mach,
-      reg->Register.File,
-      swizzle,
-      &index,
-      chan );
+   fetch_src_file_channel(mach,
+                          reg->Register.File,
+                          swizzle,
+                          &index,
+                          &index2D,
+                          chan);
 
    if (reg->Register.Absolute) {
       if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
@@ -1239,12 +1345,14 @@ store_dest(struct tgsi_exec_machine *mach,
    uint i;
    union tgsi_exec_channel null;
    union tgsi_exec_channel *dst;
+   union tgsi_exec_channel index2D;
    uint execmask = mach->ExecMask;
    int offset = 0;  /* indirection offset */
    int index;
 
-   if (dst_datatype == TGSI_EXEC_DATA_FLOAT) {
-      CHECK_INF_OR_NAN(chan);
+   /* for debugging */
+   if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) {
+      check_inf_or_nan(chan);
    }
 
    /* There is an extra source register that indirectly subscripts
@@ -1272,53 +1380,127 @@ store_dest(struct tgsi_exec_machine *mach,
       swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
 
       /* fetch values from the address/indirection register */
-      fetch_src_file_channel(
-         mach,
-         reg->Indirect.File,
-         swizzle,
-         &index,
-         &indir_index );
+      fetch_src_file_channel(mach,
+                             reg->Indirect.File,
+                             swizzle,
+                             &index,
+                             &ZeroVec,
+                             &indir_index);
 
       /* save indirection offset */
       offset = indir_index.i[0];
    }
 
-   switch (reg->Register.File) {
-   case TGSI_FILE_NULL:
-      dst = &null;
-      break;
+   /* There is an extra source register that is a second
+    * subscript to a register file. Effectively it means that
+    * the register file is actually a 2D array of registers.
+    *
+    *    file[3][1],
+    *    where:
+    *       [3] = Dimension.Index
+    */
+   if (reg->Register.Dimension) {
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = reg->Dimension.Index;
 
-   case TGSI_FILE_OUTPUT:
-      index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
-         + reg->Register.Index;
-      dst = &mach->Outputs[offset + index].xyzw[chan_index];
-#if 0
-      if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
-         fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask);
-         for (i = 0; i < QUAD_SIZE; i++)
-            if (execmask & (1 << i))
-               fprintf(stderr, "%f, ", chan->f[i]);
-         fprintf(stderr, ")\n");
-      }
-#endif
-      break;
+      /* Again, the second subscript index can be addressed indirectly
+       * identically to the first one.
+       * Nothing stops us from indirectly addressing the indirect register,
+       * but there is no need for that, so we won't exercise it.
+       *
+       *    file[ind[4].y+3][1],
+       *    where:
+       *       ind = DimIndirect.File
+       *       [4] = DimIndirect.Index
+       *       .y = DimIndirect.SwizzleX
+       */
+      if (reg->Dimension.Indirect) {
+         union tgsi_exec_channel index2;
+         union tgsi_exec_channel indir_index;
+         const uint execmask = mach->ExecMask;
+         unsigned swizzle;
+         uint i;
 
-   case TGSI_FILE_TEMPORARY:
-      index = reg->Register.Index;
-      assert( index < TGSI_EXEC_NUM_TEMPS );
-      dst = &mach->Temps[offset + index].xyzw[chan_index];
+         index2.i[0] =
+         index2.i[1] =
+         index2.i[2] =
+         index2.i[3] = reg->DimIndirect.Index;
+
+         swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
+         fetch_src_file_channel(mach,
+                                reg->DimIndirect.File,
+                                swizzle,
+                                &index2,
+                                &ZeroVec,
+                                &indir_index);
+
+         index2D.i[0] += indir_index.i[0];
+         index2D.i[1] += indir_index.i[1];
+         index2D.i[2] += indir_index.i[2];
+         index2D.i[3] += indir_index.i[3];
+
+         /* for disabled execution channels, zero-out the index to
+          * avoid using a potential garbage value.
+          */
+         for (i = 0; i < QUAD_SIZE; i++) {
+            if ((execmask & (1 << i)) == 0) {
+               index2D.i[i] = 0;
+            }
+         }
+      }
+
+      /* If by any chance there was a need for a 3D array of register
+       * files, we would have to check whether Dimension is followed
+       * by a dimension register and continue the saga.
+       */
+   } else {
+      index2D.i[0] =
+      index2D.i[1] =
+      index2D.i[2] =
+      index2D.i[3] = 0;
+   }
+
+   switch (reg->Register.File) {
+   case TGSI_FILE_NULL:
+      dst = &null;
       break;
 
-   case TGSI_FILE_ADDRESS:
+   case TGSI_FILE_OUTPUT:
+      index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
+         + reg->Register.Index;
+      dst = &mach->Outputs[offset + index].xyzw[chan_index];
+#if 0
+      if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
+         fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask);
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (execmask & (1 << i))
+               fprintf(stderr, "%f, ", chan->f[i]);
+         fprintf(stderr, ")\n");
+      }
+#endif
+      break;
+
+   case TGSI_FILE_TEMPORARY:
       index = reg->Register.Index;
-      dst = &mach->Addrs[index].xyzw[chan_index];
+      assert( index < TGSI_EXEC_NUM_TEMPS );
+      dst = &mach->Temps[offset + index].xyzw[chan_index];
       break;
 
-   case TGSI_FILE_LOOP:
-      assert(reg->Register.Index == 0);
-      assert(mach->LoopCounterStackTop > 0);
-      assert(chan_index == CHAN_X);
-      dst = &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[chan_index];
+   case TGSI_FILE_TEMPORARY_ARRAY:
+      index = reg->Register.Index;
+      assert( index < TGSI_EXEC_NUM_TEMPS );
+      assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS );
+      /* XXX we use index2D.i[0] here but somehow we might
+       * end up with someone trying to store indirectly in
+       * different buffers */
+      dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      index = reg->Register.Index;
+      dst = &mach->Addrs[index].xyzw[chan_index];
       break;
 
    case TGSI_FILE_PREDICATE:
@@ -1412,9 +1594,6 @@ store_dest(struct tgsi_exec_machine *mach,
 #define FETCH(VAL,INDEX,CHAN)\
     fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
 
-#define STORE(VAL,INDEX,CHAN)\
-   store_dest(mach, VAL, &inst->Dst[INDEX], inst, CHAN, TGSI_EXEC_DATA_FLOAT)
-
 
 /**
  * Execute ARB-style KIL which is predicated by a src register.
@@ -1501,8 +1680,21 @@ emit_primitive(struct tgsi_exec_machine *mach)
    }
 }
 
+static void
+conditional_emit_primitive(struct tgsi_exec_machine *mach)
+{
+   if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
+      int emitted_verts =
+         mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]];
+      if (emitted_verts) {
+         emit_primitive(mach);
+      }
+   }
+}
+
+
 /*
- * Fetch a four texture samples using STR texture coordinates.
+ * Fetch four texture samples using STR texture coordinates.
  */
 static void
 fetch_texel( struct tgsi_sampler *sampler,
@@ -1545,7 +1737,7 @@ exec_tex(struct tgsi_exec_machine *mach,
    union tgsi_exec_channel r[4];
    const union tgsi_exec_channel *lod = &ZeroVec;
    enum tgsi_sampler_control control;
-   uint chan_index;
+   uint chan;
 
    if (modifier != TEX_MODIFIER_NONE) {
       FETCH(&r[3], 0, CHAN_W);
@@ -1595,6 +1787,36 @@ exec_tex(struct tgsi_exec_machine *mach,
                   &r[0], &r[1], &r[2], &r[3]);  /* outputs */
       break;
 
+   case TGSI_TEXTURE_1D_ARRAY:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+
+      if (modifier == TEX_MODIFIER_PROJECTED) {
+         micro_div(&r[0], &r[0], &r[3]);
+      }
+
+      fetch_texel(mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lod,     /* S, T, P, LOD */
+                  control,
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
+   case TGSI_TEXTURE_2D_ARRAY:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      if (modifier == TEX_MODIFIER_PROJECTED) {
+         micro_div(&r[0], &r[0], &r[3]);
+         micro_div(&r[1], &r[1], &r[3]);
+      }
+
+      fetch_texel(mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lod,     /* S, T, P, LOD */
+                  control,
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
    case TGSI_TEXTURE_3D:
    case TGSI_TEXTURE_CUBE:
       FETCH(&r[0], 0, CHAN_X);
@@ -1617,8 +1839,21 @@ exec_tex(struct tgsi_exec_machine *mach,
       assert(0);
    }
 
-   FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-      STORE(&r[chan_index], 0, chan_index);
+#if 0
+   debug_printf("fetch r: %g %g %g %g\n",
+         r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]);
+   debug_printf("fetch g: %g %g %g %g\n",
+         r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]);
+   debug_printf("fetch b: %g %g %g %g\n",
+         r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]);
+   debug_printf("fetch a: %g %g %g %g\n",
+         r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]);
+#endif
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
    }
 }
 
@@ -1628,7 +1863,7 @@ exec_txd(struct tgsi_exec_machine *mach,
 {
    const uint unit = inst->Src[3].Register.Index;
    union tgsi_exec_channel r[4];
-   uint chan_index;
+   uint chan;
 
    /*
     * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
@@ -1678,8 +1913,168 @@ exec_txd(struct tgsi_exec_machine *mach,
       assert(0);
    }
 
-   FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-      STORE(&r[chan_index], 0, chan_index);
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+
+
+static void
+exec_sample(struct tgsi_exec_machine *mach,
+            const struct tgsi_full_instruction *inst,
+            uint modifier)
+{
+   const uint resource_unit = inst->Src[1].Register.Index;
+   const uint sampler_unit = inst->Src[2].Register.Index;
+   union tgsi_exec_channel r[4];
+   const union tgsi_exec_channel *lod = &ZeroVec;
+   enum tgsi_sampler_control control;
+   uint chan;
+
+   if (modifier != TEX_MODIFIER_NONE) {
+      if (modifier == TEX_MODIFIER_LOD_BIAS)
+         FETCH(&r[3], 3, CHAN_X);
+      else /*TEX_MODIFIER_LOD*/
+         FETCH(&r[3], 0, CHAN_W);
+
+      if (modifier != TEX_MODIFIER_PROJECTED) {
+         lod = &r[3];
+      }
+   }
+
+   if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
+      control = tgsi_sampler_lod_explicit;
+   } else {
+      control = tgsi_sampler_lod_bias;
+   }
+
+   switch (mach->Resources[resource_unit].Resource) {
+   case TGSI_TEXTURE_1D:
+   case TGSI_TEXTURE_SHADOW1D:
+      FETCH(&r[0], 0, CHAN_X);
+
+      if (modifier == TEX_MODIFIER_PROJECTED) {
+         micro_div(&r[0], &r[0], &r[3]);
+      }
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &ZeroVec, &ZeroVec, lod,  /* S, T, P, LOD */
+                  control,
+                  &r[0], &r[1], &r[2], &r[3]);     /* R, G, B, A */
+      break;
+
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+   case TGSI_TEXTURE_SHADOW2D:
+   case TGSI_TEXTURE_SHADOWRECT:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      if (modifier == TEX_MODIFIER_PROJECTED) {
+         micro_div(&r[0], &r[0], &r[3]);
+         micro_div(&r[1], &r[1], &r[3]);
+         micro_div(&r[2], &r[2], &r[3]);
+      }
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &r[1], &r[2], lod,     /* S, T, P, LOD */
+                  control,
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      if (modifier == TEX_MODIFIER_PROJECTED) {
+         micro_div(&r[0], &r[0], &r[3]);
+         micro_div(&r[1], &r[1], &r[3]);
+         micro_div(&r[2], &r[2], &r[3]);
+      }
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &r[1], &r[2], lod,
+                  control,
+                  &r[0], &r[1], &r[2], &r[3]);
+      break;
+
+   default:
+      assert(0);
+   }
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+static void
+exec_sample_d(struct tgsi_exec_machine *mach,
+              const struct tgsi_full_instruction *inst)
+{
+   const uint resource_unit = inst->Src[1].Register.Index;
+   const uint sampler_unit = inst->Src[2].Register.Index;
+   union tgsi_exec_channel r[4];
+   uint chan;
+   /*
+    * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet.
+    */
+
+   switch (mach->Resources[resource_unit].Resource) {
+   case TGSI_TEXTURE_1D:
+   case TGSI_TEXTURE_SHADOW1D:
+
+      FETCH(&r[0], 0, CHAN_X);
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &ZeroVec, &ZeroVec, &ZeroVec,   /* S, T, P, BIAS */
+                  tgsi_sampler_lod_bias,
+                  &r[0], &r[1], &r[2], &r[3]);           /* R, G, B, A */
+      break;
+
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+   case TGSI_TEXTURE_SHADOW2D:
+   case TGSI_TEXTURE_SHADOWRECT:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &r[1], &r[2], &ZeroVec,   /* inputs */
+                  tgsi_sampler_lod_bias,
+                  &r[0], &r[1], &r[2], &r[3]);     /* outputs */
+      break;
+
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      fetch_texel(mach->Samplers[sampler_unit],
+                  &r[0], &r[1], &r[2], &ZeroVec,
+                  tgsi_sampler_lod_bias,
+                  &r[0], &r[1], &r[2], &r[3]);
+      break;
+
+   default:
+      assert(0);
+   }
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
    }
 }
 
@@ -1755,22 +2150,27 @@ static void
 exec_declaration(struct tgsi_exec_machine *mach,
                  const struct tgsi_full_declaration *decl)
 {
+   if (decl->Declaration.File == TGSI_FILE_RESOURCE) {
+      mach->Resources[decl->Range.First] = decl->Resource;
+      return;
+   }
+
    if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
-      if (decl->Declaration.File == TGSI_FILE_INPUT ||
-          decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
+      if (decl->Declaration.File == TGSI_FILE_INPUT) {
          uint first, last, mask;
 
          first = decl->Range.First;
          last = decl->Range.Last;
          mask = decl->Declaration.UsageMask;
 
-         if (decl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
-            assert(decl->Semantic.Index == 0);
-            assert(first == last);
-            assert(mask == TGSI_WRITEMASK_XYZW);
-
-            mach->Inputs[first] = mach->QuadPos;
-         } else if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
+         /* XXX we could remove this special-case code since
+          * mach->InterpCoefs[first].a0 should already have the
+          * front/back-face value.  But we should first update the
+          * ureg code to emit the right UsageMask value (WRITEMASK_X).
+          * Then, we could remove the tgsi_exec_machine::Face field.
+          */
+         /* XXX make FACE a system value */
+         if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
             uint i;
 
             assert(decl->Semantic.Index == 0);
@@ -1811,15 +2211,40 @@ exec_declaration(struct tgsi_exec_machine *mach,
          }
       }
    }
+
+   if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
+      mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First;
+   }
+}
+
+
+typedef void (* micro_op)(union tgsi_exec_channel *dst);
+
+static void
+exec_vector(struct tgsi_exec_machine *mach,
+            const struct tgsi_full_instruction *inst,
+            micro_op op,
+            enum tgsi_exec_datatype dst_datatype)
+{
+   unsigned int chan;
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         union tgsi_exec_channel dst;
+
+         op(&dst);
+         store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
+      }
+   }
 }
 
-typedef void (* micro_op)(union tgsi_exec_channel *dst,
-                          const union tgsi_exec_channel *src);
+typedef void (* micro_unary_op)(union tgsi_exec_channel *dst,
+                                const union tgsi_exec_channel *src);
 
 static void
 exec_scalar_unary(struct tgsi_exec_machine *mach,
                   const struct tgsi_full_instruction *inst,
-                  micro_op op,
+                  micro_unary_op op,
                   enum tgsi_exec_datatype dst_datatype,
                   enum tgsi_exec_datatype src_datatype)
 {
@@ -1839,7 +2264,7 @@ exec_scalar_unary(struct tgsi_exec_machine *mach,
 static void
 exec_vector_unary(struct tgsi_exec_machine *mach,
                   const struct tgsi_full_instruction *inst,
-                  micro_op op,
+                  micro_unary_op op,
                   enum tgsi_exec_datatype dst_datatype,
                   enum tgsi_exec_datatype src_datatype)
 {
@@ -1861,10 +2286,35 @@ exec_vector_unary(struct tgsi_exec_machine *mach,
    }
 }
 
+typedef void (* micro_binary_op)(union tgsi_exec_channel *dst,
+                                 const union tgsi_exec_channel *src0,
+                                 const union tgsi_exec_channel *src1);
+
+static void
+exec_scalar_binary(struct tgsi_exec_machine *mach,
+                   const struct tgsi_full_instruction *inst,
+                   micro_binary_op op,
+                   enum tgsi_exec_datatype dst_datatype,
+                   enum tgsi_exec_datatype src_datatype)
+{
+   unsigned int chan;
+   union tgsi_exec_channel src[2];
+   union tgsi_exec_channel dst;
+
+   fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype);
+   fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype);
+   op(&dst, &src[0], &src[1]);
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
+      }
+   }
+}
+
 static void
 exec_vector_binary(struct tgsi_exec_machine *mach,
                    const struct tgsi_full_instruction *inst,
-                   micro_op op,
+                   micro_binary_op op,
                    enum tgsi_exec_datatype dst_datatype,
                    enum tgsi_exec_datatype src_datatype)
 {
@@ -1877,7 +2327,7 @@ exec_vector_binary(struct tgsi_exec_machine *mach,
 
          fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
          fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
-         op(&dst.xyzw[chan], src);
+         op(&dst.xyzw[chan], &src[0], &src[1]);
       }
    }
    for (chan = 0; chan < NUM_CHANNELS; chan++) {
@@ -1887,10 +2337,15 @@ exec_vector_binary(struct tgsi_exec_machine *mach,
    }
 }
 
+typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst,
+                                  const union tgsi_exec_channel *src0,
+                                  const union tgsi_exec_channel *src1,
+                                  const union tgsi_exec_channel *src2);
+
 static void
 exec_vector_trinary(struct tgsi_exec_machine *mach,
                     const struct tgsi_full_instruction *inst,
-                    micro_op op,
+                    micro_trinary_op op,
                     enum tgsi_exec_datatype dst_datatype,
                     enum tgsi_exec_datatype src_datatype)
 {
@@ -1904,7 +2359,7 @@ exec_vector_trinary(struct tgsi_exec_machine *mach,
          fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
          fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
          fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype);
-         op(&dst.xyzw[chan], src);
+         op(&dst.xyzw[chan], &src[0], &src[1], &src[2]);
       }
    }
    for (chan = 0; chan < NUM_CHANNELS; chan++) {
@@ -1928,7 +2383,7 @@ exec_dp3(struct tgsi_exec_machine *mach,
    for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
       fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
       fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
-      micro_mad(&arg[2], arg);
+      micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
    }
 
    for (chan = 0; chan < NUM_CHANNELS; chan++) {
@@ -1952,7 +2407,7 @@ exec_dp4(struct tgsi_exec_machine *mach,
    for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
       fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
       fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
-      micro_mad(&arg[2], arg);
+      micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
    }
 
    for (chan = 0; chan < NUM_CHANNELS; chan++) {
@@ -1973,69 +2428,416 @@ exec_dp2a(struct tgsi_exec_machine *mach,
    fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
    micro_mul(&arg[2], &arg[0], &arg[1]);
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[0], arg);
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
+
+   fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_add(&arg[0], &arg[0], &arg[1]);
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+static void
+exec_dph(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   unsigned int chan;
+   union tgsi_exec_channel arg[3];
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_mul(&arg[2], &arg[0], &arg[1]);
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
+
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   micro_add(&arg[0], &arg[0], &arg[1]);
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+static void
+exec_dp2(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   unsigned int chan;
+   union tgsi_exec_channel arg[3];
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_mul(&arg[2], &arg[0], &arg[1]);
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
+
+   for (chan = 0; chan < NUM_CHANNELS; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+static void
+exec_nrm4(struct tgsi_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   unsigned int chan;
+   union tgsi_exec_channel arg[4];
+   union tgsi_exec_channel scale;
+
+   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_mul(&scale, &arg[0], &arg[0]);
+
+   for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
+      union tgsi_exec_channel product;
+
+      fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&product, &arg[chan], &arg[chan]);
+      micro_add(&scale, &scale, &product);
+   }
+
+   micro_rsq(&scale, &scale);
+
+   for (chan = CHAN_X; chan <= CHAN_W; chan++) {
+      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+         micro_mul(&arg[chan], &arg[chan], &scale);
+         store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+}
+
+static void
+exec_nrm3(struct tgsi_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
+      unsigned int chan;
+      union tgsi_exec_channel arg[3];
+      union tgsi_exec_channel scale;
+
+      fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&scale, &arg[0], &arg[0]);
+
+      for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
+         union tgsi_exec_channel product;
+
+         fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
+         micro_mul(&product, &arg[chan], &arg[chan]);
+         micro_add(&scale, &scale, &product);
+      }
+
+      micro_rsq(&scale, &scale);
+
+      for (chan = CHAN_X; chan <= CHAN_Z; chan++) {
+         if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+            micro_mul(&arg[chan], &arg[chan], &scale);
+            store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+         }
+      }
+   }
+
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
+
+static void
+exec_scs(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) {
+      union tgsi_exec_channel arg;
+      union tgsi_exec_channel result;
+
+      fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+         micro_cos(&result, &arg);
+         store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      }
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+         micro_sin(&result, &arg);
+         store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
+
+static void
+exec_x2d(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   union tgsi_exec_channel r[4];
+   union tgsi_exec_channel d[2];
+
+   fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) {
+      fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[2], &r[2], &r[0]);
+      fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[3], &r[3], &r[1]);
+      micro_add(&r[2], &r[2], &r[3]);
+      fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      micro_add(&d[0], &r[2], &r[3]);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) {
+      fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[2], &r[2], &r[0]);
+      fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[3], &r[3], &r[1]);
+      micro_add(&r[2], &r[2], &r[3]);
+      fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      micro_add(&d[1], &r[2], &r[3]);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
+
+static void
+exec_rfl(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   union tgsi_exec_channel r[9];
+
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
+      /* r0 = dp3(src0, src0) */
+      fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[0], &r[2], &r[2]);
+      fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[8], &r[4], &r[4]);
+      micro_add(&r[0], &r[0], &r[8]);
+      fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[8], &r[6], &r[6]);
+      micro_add(&r[0], &r[0], &r[8]);
+
+      /* r1 = dp3(src0, src1) */
+      fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[1], &r[2], &r[3]);
+      fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[8], &r[4], &r[5]);
+      micro_add(&r[1], &r[1], &r[8]);
+      fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&r[8], &r[6], &r[7]);
+      micro_add(&r[1], &r[1], &r[8]);
+
+      /* r1 = 2 * r1 / r0 */
+      micro_add(&r[1], &r[1], &r[1]);
+      micro_div(&r[1], &r[1], &r[0]);
+
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+         micro_mul(&r[2], &r[2], &r[1]);
+         micro_sub(&r[2], &r[2], &r[3]);
+         store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      }
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+         micro_mul(&r[4], &r[4], &r[1]);
+         micro_sub(&r[4], &r[4], &r[5]);
+         store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      }
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+         micro_mul(&r[6], &r[6], &r[1]);
+         micro_sub(&r[6], &r[6], &r[7]);
+         store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+      }
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
+
+static void
+exec_xpd(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   union tgsi_exec_channel r[6];
+   union tgsi_exec_channel d[3];
+
+   fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+
+   micro_mul(&r[2], &r[0], &r[1]);
+
+   fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+
+   micro_mul(&r[5], &r[3], &r[4] );
+   micro_sub(&d[CHAN_X], &r[2], &r[5]);
+
+   fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+
+   micro_mul(&r[3], &r[3], &r[2]);
+
+   fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+
+   micro_mul(&r[1], &r[1], &r[5]);
+   micro_sub(&d[CHAN_Y], &r[3], &r[1]);
 
-   fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   micro_add(&arg[0], &arg[0], &arg[1]);
+   micro_mul(&r[5], &r[5], &r[4]);
+   micro_mul(&r[0], &r[0], &r[2]);
+   micro_sub(&d[CHAN_Z], &r[5], &r[0]);
 
-   for (chan = 0; chan < NUM_CHANNELS; chan++) {
-      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
-         store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
-      }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
    }
 }
 
 static void
-exec_dph(struct tgsi_exec_machine *mach,
+exec_dst(struct tgsi_exec_machine *mach,
          const struct tgsi_full_instruction *inst)
 {
-   unsigned int chan;
-   union tgsi_exec_channel arg[3];
+   union tgsi_exec_channel r[2];
+   union tgsi_exec_channel d[4];
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   micro_mul(&arg[2], &arg[0], &arg[1]);
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      micro_mul(&d[CHAN_Y], &r[0], &r[1]);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[2], arg);
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[0], arg);
+static void
+exec_log(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   union tgsi_exec_channel r[3];
+
+   fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_abs(&r[2], &r[0]);  /* r2 = abs(r0) */
+   micro_lg2(&r[1], &r[2]);  /* r1 = lg2(r2) */
+   micro_flr(&r[0], &r[1]);  /* r0 = floor(r1) */
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      micro_exp2(&r[0], &r[0]);       /* r0 = 2 ^ r0 */
+      micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */
+      store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
+}
 
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
-   micro_add(&arg[0], &arg[0], &arg[1]);
+static void
+exec_exp(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   union tgsi_exec_channel r[3];
 
-   for (chan = 0; chan < NUM_CHANNELS; chan++) {
-      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
-         store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
-      }
+   fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   micro_flr(&r[1], &r[0]);  /* r1 = floor(r0) */
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      micro_exp2(&r[2], &r[1]);       /* r2 = 2 ^ r1 */
+      store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+      micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */
+      store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+      micro_exp2(&r[2], &r[0]);       /* r2 = 2 ^ r0 */
+      store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
    }
 }
 
 static void
-exec_dp2(struct tgsi_exec_machine *mach,
+exec_lit(struct tgsi_exec_machine *mach,
          const struct tgsi_full_instruction *inst)
 {
-   unsigned int chan;
-   union tgsi_exec_channel arg[3];
+   union tgsi_exec_channel r[3];
+   union tgsi_exec_channel d[3];
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
-   micro_mul(&arg[2], &arg[0], &arg[1]);
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
+   }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) {
+      fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+         micro_max(&d[CHAN_Y], &r[0], &ZeroVec);
+         store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+      }
 
-   fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
-   micro_mad(&arg[2], arg);
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
+         fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
+         micro_max(&r[1], &r[1], &ZeroVec);
 
-   for (chan = 0; chan < NUM_CHANNELS; chan++) {
-      if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
-         store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+         fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT);
+         micro_min(&r[2], &r[2], &P128Vec);
+         micro_max(&r[2], &r[2], &M128Vec);
+         micro_pow(&r[1], &r[1], &r[2]);
+         micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec);
+         store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
       }
    }
+   if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
+      store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
+   }
 }
 
 static void
@@ -2144,42 +2946,46 @@ micro_not(union tgsi_exec_channel *dst,
 
 static void
 micro_shl(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] << src[1].u[0];
-   dst->u[1] = src[0].u[1] << src[1].u[1];
-   dst->u[2] = src[0].u[2] << src[1].u[2];
-   dst->u[3] = src[0].u[3] << src[1].u[3];
+   dst->u[0] = src0->u[0] << src1->u[0];
+   dst->u[1] = src0->u[1] << src1->u[1];
+   dst->u[2] = src0->u[2] << src1->u[2];
+   dst->u[3] = src0->u[3] << src1->u[3];
 }
 
 static void
 micro_and(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] & src[1].u[0];
-   dst->u[1] = src[0].u[1] & src[1].u[1];
-   dst->u[2] = src[0].u[2] & src[1].u[2];
-   dst->u[3] = src[0].u[3] & src[1].u[3];
+   dst->u[0] = src0->u[0] & src1->u[0];
+   dst->u[1] = src0->u[1] & src1->u[1];
+   dst->u[2] = src0->u[2] & src1->u[2];
+   dst->u[3] = src0->u[3] & src1->u[3];
 }
 
 static void
 micro_or(union tgsi_exec_channel *dst,
-         const union tgsi_exec_channel *src)
+         const union tgsi_exec_channel *src0,
+         const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] | src[1].u[0];
-   dst->u[1] = src[0].u[1] | src[1].u[1];
-   dst->u[2] = src[0].u[2] | src[1].u[2];
-   dst->u[3] = src[0].u[3] | src[1].u[3];
+   dst->u[0] = src0->u[0] | src1->u[0];
+   dst->u[1] = src0->u[1] | src1->u[1];
+   dst->u[2] = src0->u[2] | src1->u[2];
+   dst->u[3] = src0->u[3] | src1->u[3];
 }
 
 static void
 micro_xor(union tgsi_exec_channel *dst,
-          const union tgsi_exec_channel *src)
+          const union tgsi_exec_channel *src0,
+          const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] ^ src[1].u[0];
-   dst->u[1] = src[0].u[1] ^ src[1].u[1];
-   dst->u[2] = src[0].u[2] ^ src[1].u[2];
-   dst->u[3] = src[0].u[3] ^ src[1].u[3];
+   dst->u[0] = src0->u[0] ^ src1->u[0];
+   dst->u[1] = src0->u[1] ^ src1->u[1];
+   dst->u[2] = src0->u[2] ^ src1->u[2];
+   dst->u[3] = src0->u[3] ^ src1->u[3];
 }
 
 static void
@@ -2194,62 +3000,68 @@ micro_f2i(union tgsi_exec_channel *dst,
 
 static void
 micro_idiv(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] / src[1].i[0];
-   dst->i[1] = src[0].i[1] / src[1].i[1];
-   dst->i[2] = src[0].i[2] / src[1].i[2];
-   dst->i[3] = src[0].i[3] / src[1].i[3];
+   dst->i[0] = src0->i[0] / src1->i[0];
+   dst->i[1] = src0->i[1] / src1->i[1];
+   dst->i[2] = src0->i[2] / src1->i[2];
+   dst->i[3] = src0->i[3] / src1->i[3];
 }
 
 static void
 micro_imax(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] > src[1].i[0] ? src[0].i[0] : src[1].i[0];
-   dst->i[1] = src[0].i[1] > src[1].i[1] ? src[0].i[1] : src[1].i[1];
-   dst->i[2] = src[0].i[2] > src[1].i[2] ? src[0].i[2] : src[1].i[2];
-   dst->i[3] = src[0].i[3] > src[1].i[3] ? src[0].i[3] : src[1].i[3];
+   dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3];
 }
 
 static void
 micro_imin(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] < src[1].i[0] ? src[0].i[0] : src[1].i[0];
-   dst->i[1] = src[0].i[1] < src[1].i[1] ? src[0].i[1] : src[1].i[1];
-   dst->i[2] = src[0].i[2] < src[1].i[2] ? src[0].i[2] : src[1].i[2];
-   dst->i[3] = src[0].i[3] < src[1].i[3] ? src[0].i[3] : src[1].i[3];
+   dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3];
 }
 
 static void
 micro_isge(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] >= src[1].i[0] ? -1 : 0;
-   dst->i[1] = src[0].i[1] >= src[1].i[1] ? -1 : 0;
-   dst->i[2] = src[0].i[2] >= src[1].i[2] ? -1 : 0;
-   dst->i[3] = src[0].i[3] >= src[1].i[3] ? -1 : 0;
+   dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0;
+   dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0;
+   dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0;
+   dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0;
 }
 
 static void
 micro_ishr(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] >> src[1].i[0];
-   dst->i[1] = src[0].i[1] >> src[1].i[1];
-   dst->i[2] = src[0].i[2] >> src[1].i[2];
-   dst->i[3] = src[0].i[3] >> src[1].i[3];
+   dst->i[0] = src0->i[0] >> src1->i[0];
+   dst->i[1] = src0->i[1] >> src1->i[1];
+   dst->i[2] = src0->i[2] >> src1->i[2];
+   dst->i[3] = src0->i[3] >> src1->i[3];
 }
 
 static void
 micro_islt(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->i[0] = src[0].i[0] < src[1].i[0] ? -1 : 0;
-   dst->i[1] = src[0].i[1] < src[1].i[1] ? -1 : 0;
-   dst->i[2] = src[0].i[2] < src[1].i[2] ? -1 : 0;
-   dst->i[3] = src[0].i[3] < src[1].i[3] ? -1 : 0;
+   dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0;
+   dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0;
+   dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0;
+   dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0;
 }
 
 static void
@@ -2274,122 +3086,135 @@ micro_u2f(union tgsi_exec_channel *dst,
 
 static void
 micro_uadd(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] + src[1].u[0];
-   dst->u[1] = src[0].u[1] + src[1].u[1];
-   dst->u[2] = src[0].u[2] + src[1].u[2];
-   dst->u[3] = src[0].u[3] + src[1].u[3];
+   dst->u[0] = src0->u[0] + src1->u[0];
+   dst->u[1] = src0->u[1] + src1->u[1];
+   dst->u[2] = src0->u[2] + src1->u[2];
+   dst->u[3] = src0->u[3] + src1->u[3];
 }
 
 static void
 micro_udiv(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] / src[1].u[0];
-   dst->u[1] = src[0].u[1] / src[1].u[1];
-   dst->u[2] = src[0].u[2] / src[1].u[2];
-   dst->u[3] = src[0].u[3] / src[1].u[3];
+   dst->u[0] = src0->u[0] / src1->u[0];
+   dst->u[1] = src0->u[1] / src1->u[1];
+   dst->u[2] = src0->u[2] / src1->u[2];
+   dst->u[3] = src0->u[3] / src1->u[3];
 }
 
 static void
 micro_umad(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1,
+           const union tgsi_exec_channel *src2)
 {
-   dst->u[0] = src[0].u[0] * src[1].u[0] + src[2].u[0];
-   dst->u[1] = src[0].u[1] * src[1].u[1] + src[2].u[1];
-   dst->u[2] = src[0].u[2] * src[1].u[2] + src[2].u[2];
-   dst->u[3] = src[0].u[3] * src[1].u[3] + src[2].u[3];
+   dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0];
+   dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1];
+   dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2];
+   dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3];
 }
 
 static void
 micro_umax(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] > src[1].u[0] ? src[0].u[0] : src[1].u[0];
-   dst->u[1] = src[0].u[1] > src[1].u[1] ? src[0].u[1] : src[1].u[1];
-   dst->u[2] = src[0].u[2] > src[1].u[2] ? src[0].u[2] : src[1].u[2];
-   dst->u[3] = src[0].u[3] > src[1].u[3] ? src[0].u[3] : src[1].u[3];
+   dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3];
 }
 
 static void
 micro_umin(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] < src[1].u[0] ? src[0].u[0] : src[1].u[0];
-   dst->u[1] = src[0].u[1] < src[1].u[1] ? src[0].u[1] : src[1].u[1];
-   dst->u[2] = src[0].u[2] < src[1].u[2] ? src[0].u[2] : src[1].u[2];
-   dst->u[3] = src[0].u[3] < src[1].u[3] ? src[0].u[3] : src[1].u[3];
+   dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3];
 }
 
 static void
 micro_umod(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] % src[1].u[0];
-   dst->u[1] = src[0].u[1] % src[1].u[1];
-   dst->u[2] = src[0].u[2] % src[1].u[2];
-   dst->u[3] = src[0].u[3] % src[1].u[3];
+   dst->u[0] = src0->u[0] % src1->u[0];
+   dst->u[1] = src0->u[1] % src1->u[1];
+   dst->u[2] = src0->u[2] % src1->u[2];
+   dst->u[3] = src0->u[3] % src1->u[3];
 }
 
 static void
 micro_umul(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] * src[1].u[0];
-   dst->u[1] = src[0].u[1] * src[1].u[1];
-   dst->u[2] = src[0].u[2] * src[1].u[2];
-   dst->u[3] = src[0].u[3] * src[1].u[3];
+   dst->u[0] = src0->u[0] * src1->u[0];
+   dst->u[1] = src0->u[1] * src1->u[1];
+   dst->u[2] = src0->u[2] * src1->u[2];
+   dst->u[3] = src0->u[3] * src1->u[3];
 }
 
 static void
 micro_useq(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] == src[1].u[0] ? ~0 : 0;
-   dst->u[1] = src[0].u[1] == src[1].u[1] ? ~0 : 0;
-   dst->u[2] = src[0].u[2] == src[1].u[2] ? ~0 : 0;
-   dst->u[3] = src[0].u[3] == src[1].u[3] ? ~0 : 0;
+   dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0;
+   dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0;
+   dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0;
+   dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0;
 }
 
 static void
 micro_usge(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] >= src[1].u[0] ? ~0 : 0;
-   dst->u[1] = src[0].u[1] >= src[1].u[1] ? ~0 : 0;
-   dst->u[2] = src[0].u[2] >= src[1].u[2] ? ~0 : 0;
-   dst->u[3] = src[0].u[3] >= src[1].u[3] ? ~0 : 0;
+   dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0;
+   dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0;
+   dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0;
+   dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0;
 }
 
 static void
 micro_ushr(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] >> src[1].u[0];
-   dst->u[1] = src[0].u[1] >> src[1].u[1];
-   dst->u[2] = src[0].u[2] >> src[1].u[2];
-   dst->u[3] = src[0].u[3] >> src[1].u[3];
+   dst->u[0] = src0->u[0] >> src1->u[0];
+   dst->u[1] = src0->u[1] >> src1->u[1];
+   dst->u[2] = src0->u[2] >> src1->u[2];
+   dst->u[3] = src0->u[3] >> src1->u[3];
 }
 
 static void
 micro_uslt(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] < src[1].u[0] ? ~0 : 0;
-   dst->u[1] = src[0].u[1] < src[1].u[1] ? ~0 : 0;
-   dst->u[2] = src[0].u[2] < src[1].u[2] ? ~0 : 0;
-   dst->u[3] = src[0].u[3] < src[1].u[3] ? ~0 : 0;
+   dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0;
+   dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0;
+   dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0;
+   dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0;
 }
 
 static void
 micro_usne(union tgsi_exec_channel *dst,
-           const union tgsi_exec_channel *src)
+           const union tgsi_exec_channel *src0,
+           const union tgsi_exec_channel *src1)
 {
-   dst->u[0] = src[0].u[0] != src[1].u[0] ? ~0 : 0;
-   dst->u[1] = src[0].u[1] != src[1].u[1] ? ~0 : 0;
-   dst->u[2] = src[0].u[2] != src[1].u[2] ? ~0 : 0;
-   dst->u[3] = src[0].u[3] != src[1].u[3] ? ~0 : 0;
+   dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0;
+   dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0;
+   dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0;
+   dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0;
 }
 
 static void
@@ -2398,9 +3223,7 @@ exec_instruction(
    const struct tgsi_full_instruction *inst,
    int *pc )
 {
-   uint chan_index;
    union tgsi_exec_channel r[10];
-   union tgsi_exec_channel d[8];
 
    (*pc)++;
 
@@ -2414,36 +3237,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_LIT:
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
-         FETCH( &r[0], 0, CHAN_X );
-         if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
-            micro_max(&d[CHAN_Y], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
-         }
-
-         if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
-            FETCH( &r[1], 0, CHAN_Y );
-            micro_max( &r[1], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
-
-            FETCH( &r[2], 0, CHAN_W );
-            micro_min( &r[2], &r[2], &mach->Temps[TEMP_128_I].xyzw[TEMP_128_C] );
-            micro_max( &r[2], &r[2], &mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C] );
-            micro_pow( &r[1], &r[1], &r[2] );
-            micro_lt(&d[CHAN_Z], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
-         }
-
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-            STORE(&d[CHAN_Y], 0, CHAN_Y);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-            STORE(&d[CHAN_Z], 0, CHAN_Z);
-         }
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
-      }
+      exec_lit(mach, inst);
       break;
 
    case TGSI_OPCODE_RCP:
@@ -2455,66 +3249,19 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_EXP:
-      FETCH( &r[0], 0, CHAN_X );
-      micro_flr( &r[1], &r[0] );  /* r1 = floor(r0) */
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
-         micro_exp2( &r[2], &r[1] );       /* r2 = 2 ^ r1 */
-         STORE( &r[2], 0, CHAN_X );        /* store r2 */
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
-         micro_sub( &r[2], &r[0], &r[1] ); /* r2 = r0 - r1 */
-         STORE( &r[2], 0, CHAN_Y );        /* store r2 */
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
-         micro_exp2( &r[2], &r[0] );       /* r2 = 2 ^ r0 */
-         STORE( &r[2], 0, CHAN_Z );        /* store r2 */
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
-      }
+      exec_exp(mach, inst);
       break;
 
    case TGSI_OPCODE_LOG:
-      FETCH( &r[0], 0, CHAN_X );
-      micro_abs( &r[2], &r[0] );  /* r2 = abs(r0) */
-      micro_lg2( &r[1], &r[2] );  /* r1 = lg2(r2) */
-      micro_flr( &r[0], &r[1] );  /* r0 = floor(r1) */
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
-         STORE( &r[0], 0, CHAN_X );
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
-         micro_exp2( &r[0], &r[0] );       /* r0 = 2 ^ r0 */
-         micro_div( &r[0], &r[2], &r[0] ); /* r0 = r2 / r0 */
-         STORE( &r[0], 0, CHAN_Y );
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
-         STORE( &r[1], 0, CHAN_Z );
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
-      }
+      exec_log(mach, inst);
       break;
 
    case TGSI_OPCODE_MUL:
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-         micro_mul(&d[chan_index], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_ADD:
-      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-         FETCH( &r[0], 0, chan_index );
-         FETCH( &r[1], 1, chan_index );
-         micro_add(&d[chan_index], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_DP3:
@@ -2526,56 +3273,15 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_DST:
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
-         FETCH( &r[0], 0, CHAN_Y );
-         FETCH( &r[1], 1, CHAN_Y);
-         micro_mul(&d[CHAN_Y], &r[0], &r[1]);
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
-         FETCH(&d[CHAN_Z], 0, CHAN_Z);
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
-         FETCH(&d[CHAN_W], 1, CHAN_W);
-      }
-
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-         STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-         STORE(&d[CHAN_Y], 0, CHAN_Y);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         STORE(&d[CHAN_Z], 0, CHAN_Z);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
-         STORE(&d[CHAN_W], 0, CHAN_W);
-      }
+      exec_dst(mach, inst);
       break;
 
    case TGSI_OPCODE_MIN:
-      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-
-         /* XXX use micro_min()?? */
-         micro_lt(&d[chan_index], &r[0], &r[1], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_MAX:
-      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-
-         /* XXX use micro_max()?? */
-         micro_lt(&d[chan_index], &r[0], &r[1], &r[1], &r[0] );
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_SLT:
@@ -2591,14 +3297,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SUB:
-      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-         micro_sub(&d[chan_index], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_LRP:
@@ -2606,15 +3305,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_CND:
-      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_lt(&d[chan_index], &mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C], &r[2], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_DP2A:
@@ -2626,16 +3317,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_CLAMP:
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         FETCH(&r[0], 0, chan_index);
-         FETCH(&r[1], 1, chan_index);
-         micro_max(&r[0], &r[0], &r[1]);
-         FETCH(&r[1], 2, chan_index);
-         micro_min(&d[chan_index], &r[0], &r[1]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_FLR:
@@ -2655,53 +3337,11 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_POW:
-      FETCH(&r[0], 0, CHAN_X);
-      FETCH(&r[1], 1, CHAN_X);
-
-      micro_pow( &r[0], &r[0], &r[1] );
-
-      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-         STORE( &r[0], 0, chan_index );
-      }
+      exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_XPD:
-      FETCH(&r[0], 0, CHAN_Y);
-      FETCH(&r[1], 1, CHAN_Z);
-
-      micro_mul( &r[2], &r[0], &r[1] );
-
-      FETCH(&r[3], 0, CHAN_Z);
-      FETCH(&r[4], 1, CHAN_Y);
-
-      micro_mul( &r[5], &r[3], &r[4] );
-      micro_sub(&d[CHAN_X], &r[2], &r[5]);
-
-      FETCH(&r[2], 1, CHAN_X);
-
-      micro_mul( &r[3], &r[3], &r[2] );
-
-      FETCH(&r[5], 0, CHAN_X);
-
-      micro_mul( &r[1], &r[1], &r[5] );
-      micro_sub(&d[CHAN_Y], &r[3], &r[1]);
-
-      micro_mul( &r[5], &r[5], &r[4] );
-      micro_mul( &r[0], &r[0], &r[2] );
-      micro_sub(&d[CHAN_Z], &r[5], &r[0]);
-
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-         STORE(&d[CHAN_X], 0, CHAN_X);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-         STORE(&d[CHAN_Y], 0, CHAN_Y);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         STORE(&d[CHAN_Z], 0, CHAN_Z);
-      }
-      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
-      }
+      exec_xpd(mach, inst);
       break;
 
    case TGSI_OPCODE_ABS:
@@ -2709,12 +3349,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_RCC:
-      FETCH(&r[0], 0, CHAN_X);
-      micro_div(&r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0]);
-      micro_float_clamp(&r[0], &r[0]);
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&r[0], 0, chan_index);
-      }
+      exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_DPH:
@@ -2758,52 +3393,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_RFL:
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
-          IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
-          IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         /* r0 = dp3(src0, src0) */
-         FETCH(&r[2], 0, CHAN_X);
-         micro_mul(&r[0], &r[2], &r[2]);
-         FETCH(&r[4], 0, CHAN_Y);
-         micro_mul(&r[8], &r[4], &r[4]);
-         micro_add(&r[0], &r[0], &r[8]);
-         FETCH(&r[6], 0, CHAN_Z);
-         micro_mul(&r[8], &r[6], &r[6]);
-         micro_add(&r[0], &r[0], &r[8]);
-
-         /* r1 = dp3(src0, src1) */
-         FETCH(&r[3], 1, CHAN_X);
-         micro_mul(&r[1], &r[2], &r[3]);
-         FETCH(&r[5], 1, CHAN_Y);
-         micro_mul(&r[8], &r[4], &r[5]);
-         micro_add(&r[1], &r[1], &r[8]);
-         FETCH(&r[7], 1, CHAN_Z);
-         micro_mul(&r[8], &r[6], &r[7]);
-         micro_add(&r[1], &r[1], &r[8]);
-
-         /* r1 = 2 * r1 / r0 */
-         micro_add(&r[1], &r[1], &r[1]);
-         micro_div(&r[1], &r[1], &r[0]);
-
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-            micro_mul(&r[2], &r[2], &r[1]);
-            micro_sub(&r[2], &r[2], &r[3]);
-            STORE(&r[2], 0, CHAN_X);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-            micro_mul(&r[4], &r[4], &r[1]);
-            micro_sub(&r[4], &r[4], &r[5]);
-            STORE(&r[4], 0, CHAN_Y);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-            micro_mul(&r[6], &r[6], &r[1]);
-            micro_sub(&r[6], &r[6], &r[7]);
-            STORE(&r[6], 0, CHAN_Z);
-         }
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
-         STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W);
-      }
+      exec_rfl(mach, inst);
       break;
 
    case TGSI_OPCODE_SEQ:
@@ -2811,9 +3401,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_SFL:
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, chan_index);
-      }
+      exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_SGT:
@@ -2833,9 +3421,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_STR:
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, chan_index);
-      }
+      exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_TEX:
@@ -2892,42 +3478,7 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_X2D:
-      FETCH(&r[0], 1, CHAN_X);
-      FETCH(&r[1], 1, CHAN_Y);
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
-          IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         FETCH(&r[2], 2, CHAN_X);
-         micro_mul(&r[2], &r[2], &r[0]);
-         FETCH(&r[3], 2, CHAN_Y);
-         micro_mul(&r[3], &r[3], &r[1]);
-         micro_add(&r[2], &r[2], &r[3]);
-         FETCH(&r[3], 0, CHAN_X);
-         micro_add(&d[CHAN_X], &r[2], &r[3]);
-         
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
-          IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
-         FETCH(&r[2], 2, CHAN_Z);
-         micro_mul(&r[2], &r[2], &r[0]);
-         FETCH(&r[3], 2, CHAN_W);
-         micro_mul(&r[3], &r[3], &r[1]);
-         micro_add(&r[2], &r[2], &r[3]);
-         FETCH(&r[3], 0, CHAN_Y);
-         micro_add(&d[CHAN_Y], &r[2], &r[3]);
-         
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-         STORE(&d[CHAN_X], 0, CHAN_X);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-         STORE(&d[CHAN_Y], 0, CHAN_Y);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         STORE(&d[CHAN_X], 0, CHAN_Z);
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
-         STORE(&d[CHAN_Y], 0, CHAN_W);
-      }
+      exec_x2d(mach, inst);
       break;
 
    case TGSI_OPCODE_ARA:
@@ -2993,6 +3544,8 @@ exec_instruction(
 
          if (mach->CallStackTop == 0) {
             /* returning from main() */
+            mach->CondStackTop = 0;
+            mach->LoopStackTop = 0;
             *pc = -1;
             return;
          }
@@ -3029,106 +3582,23 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_CMP:
-      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_lt(&d[chan_index], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2]);
-      }
-      FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
-         STORE(&d[chan_index], 0, chan_index);
-      }
+      exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_SCS:
-      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
-         FETCH( &r[0], 0, CHAN_X );
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-            micro_cos(&r[1], &r[0]);
-            STORE(&r[1], 0, CHAN_X);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-            micro_sin(&r[1], &r[0]);
-            STORE(&r[1], 0, CHAN_Y);
-         }
-      }
-      if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
-         STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z );
-      }
-      if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
-         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
-      }
+      exec_scs(mach, inst);
       break;
 
    case TGSI_OPCODE_NRM:
-      /* 3-component vector normalize */
-      if(IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
-         IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
-         IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-         /* r3 = sqrt(dp3(src0, src0)) */
-         FETCH(&r[0], 0, CHAN_X);
-         micro_mul(&r[3], &r[0], &r[0]);
-         FETCH(&r[1], 0, CHAN_Y);
-         micro_mul(&r[4], &r[1], &r[1]);
-         micro_add(&r[3], &r[3], &r[4]);
-         FETCH(&r[2], 0, CHAN_Z);
-         micro_mul(&r[4], &r[2], &r[2]);
-         micro_add(&r[3], &r[3], &r[4]);
-         micro_sqrt(&r[3], &r[3]);
-
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
-            micro_div(&r[0], &r[0], &r[3]);
-            STORE(&r[0], 0, CHAN_X);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
-            micro_div(&r[1], &r[1], &r[3]);
-            STORE(&r[1], 0, CHAN_Y);
-         }
-         if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
-            micro_div(&r[2], &r[2], &r[3]);
-            STORE(&r[2], 0, CHAN_Z);
-         }
-      }
-      if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
-         STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W);
-      }
+      exec_nrm3(mach, inst);
       break;
 
    case TGSI_OPCODE_NRM4:
-      /* 4-component vector normalize */
-      {
-         union tgsi_exec_channel tmp, dot;
-
-         /* tmp = dp4(src0, src0): */
-         FETCH( &r[0], 0, CHAN_X );
-         micro_mul( &tmp, &r[0], &r[0] );
-
-         FETCH( &r[1], 0, CHAN_Y );
-         micro_mul( &dot, &r[1], &r[1] );
-         micro_add( &tmp, &tmp, &dot );
-
-         FETCH( &r[2], 0, CHAN_Z );
-         micro_mul( &dot, &r[2], &r[2] );
-         micro_add( &tmp, &tmp, &dot );
-
-         FETCH( &r[3], 0, CHAN_W );
-         micro_mul( &dot, &r[3], &r[3] );
-         micro_add( &tmp, &tmp, &dot );
-
-         /* tmp = 1 / sqrt(tmp) */
-         micro_sqrt( &tmp, &tmp );
-         micro_div( &tmp, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &tmp );
-
-         FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
-            /* chan = chan * tmp */
-            micro_mul( &r[chan_index], &tmp, &r[chan_index] );
-            STORE( &r[chan_index], 0, chan_index );
-         }
-      }
+      exec_nrm4(mach, inst);
       break;
 
    case TGSI_OPCODE_DIV:
-      assert( 0 );
+      exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
       break;
 
    case TGSI_OPCODE_DP2:
@@ -3177,18 +3647,13 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_END:
+      /* make sure we end primitives which haven't
+       * been explicitly emitted */
+      conditional_emit_primitive(mach);
       /* halt execution */
       *pc = -1;
       break;
 
-   case TGSI_OPCODE_REP:
-      assert (0);
-      break;
-
-   case TGSI_OPCODE_ENDREP:
-       assert (0);
-       break;
-
    case TGSI_OPCODE_PUSHA:
       assert (0);
       break;
@@ -3253,29 +3718,6 @@ exec_instruction(
       emit_primitive(mach);
       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 );
-      }
-      ++mach->LoopCounterStackTop;
-      STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 0, CHAN_X);
-      /* update LoopMask */
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
-         mach->LoopMask &= ~0x1;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
-         mach->LoopMask &= ~0x2;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
-         mach->LoopMask &= ~0x4;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
-         mach->LoopMask &= ~0x8;
-      }
-      /* TODO: if mach->LoopMask == 0, jump to end of loop */
-      UPDATE_EXEC_MASK(mach);
-      /* fall-through (for now) */
    case TGSI_OPCODE_BGNLOOP:
       /* push LoopMask and ContMasks */
       assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
@@ -3290,56 +3732,6 @@ exec_instruction(
       mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
       break;
 
-   case TGSI_OPCODE_ENDFOR:
-      assert(mach->LoopCounterStackTop > 0);
-      micro_sub(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y], 
-                &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y],
-                &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
-      /* update LoopMask */
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
-         mach->LoopMask &= ~0x1;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
-         mach->LoopMask &= ~0x2;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
-         mach->LoopMask &= ~0x4;
-      }
-      if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
-         mach->LoopMask &= ~0x8;
-      }
-      micro_add(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 
-                &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 
-                &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_X], 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;
-
-         mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
-      }
-      UPDATE_EXEC_MASK(mach);
-      break;
-      
    case TGSI_OPCODE_ENDLOOP:
       /* Restore ContMask, but don't pop */
       assert(mach->ContStackTop > 0);
@@ -3537,6 +3929,54 @@ exec_instruction(
       exec_endswitch(mach);
       break;
 
+   case TGSI_OPCODE_LOAD:
+      assert(0);
+      break;
+
+   case TGSI_OPCODE_LOAD_MS:
+      assert(0);
+      break;
+
+   case TGSI_OPCODE_SAMPLE:
+      exec_sample(mach, inst, TEX_MODIFIER_NONE);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_B:
+      exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_C:
+      exec_sample(mach, inst, TEX_MODIFIER_NONE);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_C_LZ:
+      exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_D:
+      exec_sample_d(mach, inst);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_L:
+      exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD);
+      break;
+
+   case TGSI_OPCODE_GATHER4:
+      assert(0);
+      break;
+
+   case TGSI_OPCODE_RESINFO:
+      assert(0);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_POS:
+      assert(0);
+      break;
+
+   case TGSI_OPCODE_SAMPLE_INFO:
+      assert(0);
+      break;
+
    default:
       assert( 0 );
    }
@@ -3579,14 +4019,6 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
       mach->Primitives[0] = 0;
    }
 
-   for (i = 0; i < QUAD_SIZE; i++) {
-      mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C].u[i] =
-         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_X_SHIFT) |
-         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Y_SHIFT) |
-         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Z_SHIFT) |
-         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_W_SHIFT);
-   }
-
    /* execute declarations (interpolants) */
    for (i = 0; i < mach->NumDeclarations; i++) {
       exec_declaration( mach, mach->Declarations+i );
@@ -3666,6 +4098,9 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    }
 #endif
 
+   /* Strictly speaking, these assertions aren't really needed but they
+    * can potentially catch some bugs in the control flow code.
+    */
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);