Merge commit 'origin/master' into glsl-pp-rework-2

92 files changed, 3841 insertions, 834 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 60ffa188df6..c79c56debd6 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -2756,19 +2756,32 @@ exec_instruction(
       if (mach->ExecMask) {
          /* do the call */
 
-         /* push the Cond, Loop, Cont stacks */
+         /* First, record the depths of the execution stacks.
+          * This is important for deeply nested/looped return statements.
+          * We have to unwind the stacks by the correct amount.  For a
+          * real code generator, we could determine the number of entries
+          * to pop off each stack with simple static analysis and avoid
+          * implementing this data structure at run time.
+          */
+         mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
+         mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
+         mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
+         /* note that PC was already incremented above */
+         mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
+
+         mach->CallStackTop++;
+
+         /* Second, push the Cond, Loop, Cont, Func stacks */
          assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
          mach->CondStack[mach->CondStackTop++] = mach->CondMask;
          assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
          assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
          mach->ContStack[mach->ContStackTop++] = mach->ContMask;
-
          assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
          mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
 
-         /* note that PC was already incremented above */
-         mach->CallStack[mach->CallStackTop++] = *pc;
+         /* Finally, jump to the subroutine */
          *pc = inst->InstructionExtLabel.Label;
       }
       break;
@@ -2785,18 +2798,24 @@ exec_instruction(
             *pc = -1;
             return;
          }
-         *pc = mach->CallStack[--mach->CallStackTop];
 
-         /* pop the Cond, Loop, Cont stacks */
-         assert(mach->CondStackTop > 0);
-         mach->CondMask = mach->CondStack[--mach->CondStackTop];
-         assert(mach->LoopStackTop > 0);
-         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
-         assert(mach->ContStackTop > 0);
-         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->CallStackTop > 0);
+         mach->CallStackTop--;
+
+         mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
+         mach->CondMask = mach->CondStack[mach->CondStackTop];
+
+         mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
+         mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
+
+         mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
+         mach->ContMask = mach->ContStack[mach->ContStackTop];
+
          assert(mach->FuncStackTop > 0);
          mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
 
+         *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
+
          UPDATE_EXEC_MASK(mach);
       }
       break;
@@ -3245,7 +3264,6 @@ tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
    mach->FuncMask = 0xf;
    mach->ExecMask = 0xf;
 
-   mach->CondStackTop = 0; /* temporarily subvert this assertion */
    assert(mach->CondStackTop == 0);
    assert(mach->LoopStackTop == 0);
    assert(mach->ContStackTop == 0);
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 3baa94dbdde..c72f76809d4 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -186,6 +186,17 @@ struct tgsi_exec_labels
  */
 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 
+
+/** function call/activation record */
+struct tgsi_call_record
+{
+   uint CondStackTop;
+   uint LoopStackTop;
+   uint ContStackTop;
+   uint ReturnAddr;
+};
+
+
 /**
  * Run-time virtual machine state for executing TGSI shader.
  */
@@ -249,7 +260,7 @@ struct tgsi_exec_machine
    int FuncStackTop;
 
    /** Function call stack for saving/restoring the program counter */
-   uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+   struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
    int CallStackTop;
 
    struct tgsi_full_instruction *Instructions;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
index 8a13885da9b..53e13b30e63 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -358,7 +358,7 @@ epilog(
 
 boolean
 tgsi_sanity_check(
-   struct tgsi_token *tokens )
+   const struct tgsi_token *tokens )
 {
    struct sanity_check_ctx ctx;
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.h b/src/gallium/auxiliary/tgsi/tgsi_sanity.h
index ca45e94c7ad..52263ff8832 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sanity.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.h
@@ -40,7 +40,7 @@ extern "C" {
  */
 boolean
 tgsi_sanity_check(
-   struct tgsi_token *tokens );
+   const struct tgsi_token *tokens );
 
 #if defined __cplusplus
 }
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 3cdf8b9f359..501fc05e729 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -39,8 +39,9 @@
 #include "tgsi/tgsi_info.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_util.h"
-#include "tgsi_exec.h"
-#include "tgsi_sse2.h"
+#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_sse2.h"
 
 #include "rtasm/rtasm_x86sse.h"
 
@@ -1360,6 +1361,32 @@ emit_store(
    const struct tgsi_full_instruction *inst,
    unsigned chan_index )
 {
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      sse_maxps(
+         func,
+         make_xmm( xmm ),
+         get_temp(
+            TGSI_EXEC_TEMP_00000000_I,
+            TGSI_EXEC_TEMP_00000000_C ) );
+
+      sse_minps(
+         func,
+         make_xmm( xmm ),
+         get_temp(
+            TGSI_EXEC_TEMP_ONE_I,
+            TGSI_EXEC_TEMP_ONE_C ) );
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+   }
+
+
    switch( reg->DstRegister.File ) {
    case TGSI_FILE_OUTPUT:
       emit_output(
@@ -1388,19 +1415,6 @@ emit_store(
    default:
       assert( 0 );
    }
-
-   switch( inst->Instruction.Saturate ) {
-   case TGSI_SAT_NONE:
-      break;
-
-   case TGSI_SAT_ZERO_ONE:
-      /* assert( 0 ); */
-      break;
-
-   case TGSI_SAT_MINUS_PLUS_ONE:
-      assert( 0 );
-      break;
-   }
 }
 
 #define STORE( FUNC, INST, XMM, INDEX, CHAN )\
@@ -1747,14 +1761,6 @@ emit_instruction(
    if (indirect_temp_reference(inst))
       return FALSE;
 
-   /* we don't handle saturation/clamping yet */
-   if (inst->Instruction.Saturate != TGSI_SAT_NONE)
-      return FALSE;
-
-   /* need to use extra temps to fix SOA dependencies : */
-   if (tgsi_check_soa_dependencies(inst))
-      return FALSE;
-
    switch (inst->Instruction.Opcode) {
    case TGSI_OPCODE_ARL:
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
@@ -1768,8 +1774,10 @@ emit_instruction(
    case TGSI_OPCODE_MOV:
    case TGSI_OPCODE_SWZ:
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
-         FETCH( func, *inst, 0, 0, chan_index );
-         STORE( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 4 + chan_index, 0, chan_index );
+      }
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 4 + chan_index, 0, chan_index );
       }
       break;
 
@@ -2929,6 +2937,22 @@ tgsi_emit_sse2(
                          parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ?
                          "vertex shader" : "fragment shader");
 	 }
+
+         if (tgsi_check_soa_dependencies(&parse.FullToken.FullInstruction)) {
+            uint opcode = parse.FullToken.FullInstruction.Instruction.Opcode;
+
+            /* XXX: we only handle src/dst aliasing in a few opcodes
+             * currently.  Need to use an additional temporay to hold
+             * the result in the cases where the code is too opaque to
+             * fix.
+             */
+            if (opcode != TGSI_OPCODE_MOV &&
+                opcode != TGSI_OPCODE_SWZ) {
+               debug_printf("Warning: src/dst aliasing in instruction"
+                            " is not handled:\n");
+               tgsi_dump_instruction(&parse.FullToken.FullInstruction, 1);
+            }
+         }
          break;
 
       case TGSI_TOKEN_TYPE_IMMEDIATE:
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index f7096bd8e2c..654426a903d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -31,6 +31,7 @@
 #include "tgsi/tgsi_ureg.h"
 #include "tgsi/tgsi_info.h"
 #include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_sanity.h"
 #include "util/u_memory.h"
 #include "util/u_math.h"
 
@@ -70,6 +71,7 @@ struct ureg_tokens {
 
 #define UREG_MAX_INPUT PIPE_MAX_ATTRIBS
 #define UREG_MAX_OUTPUT PIPE_MAX_ATTRIBS
+#define UREG_MAX_CONSTANT_RANGE 32
 #define UREG_MAX_IMMEDIATE 32
 #define UREG_MAX_TEMP 256
 #define UREG_MAX_ADDR 2
@@ -86,8 +88,10 @@ struct ureg_program
       unsigned semantic_name;
       unsigned semantic_index;
       unsigned interp;
-   } input[UREG_MAX_INPUT];
-   unsigned nr_inputs;
+   } fs_input[UREG_MAX_INPUT];
+   unsigned nr_fs_inputs;
+
+   unsigned vs_inputs[UREG_MAX_INPUT/32];
 
    struct {
       unsigned semantic_name;
@@ -107,9 +111,13 @@ struct ureg_program
    unsigned temps_active[UREG_MAX_TEMP / 32];
    unsigned nr_temps;
 
-   unsigned nr_addrs;
+   struct {
+      unsigned first;
+      unsigned last;
+   } constant_range[UREG_MAX_CONSTANT_RANGE];
+   unsigned nr_constant_ranges;
 
-   unsigned nr_constants;
+   unsigned nr_addrs;
    unsigned nr_instructions;
 
    struct ureg_tokens domain[2];
@@ -119,6 +127,9 @@ static union tgsi_any_token error_tokens[32];
 
 static void tokens_error( struct ureg_tokens *tokens )
 {
+   if (tokens->tokens && tokens->tokens != error_tokens)
+      FREE(tokens->tokens);
+
    tokens->tokens = error_tokens;
    tokens->size = Elements(error_tokens);
    tokens->count = 0;
@@ -228,25 +239,25 @@ ureg_src_register( unsigned file,
 
 
 
-static struct ureg_src 
-ureg_DECL_input( struct ureg_program *ureg,
-                 unsigned name,
-                 unsigned index,
-                 unsigned interp_mode )
+struct ureg_src 
+ureg_DECL_fs_input( struct ureg_program *ureg,
+                    unsigned name,
+                    unsigned index,
+                    unsigned interp_mode )
 {
    unsigned i;
 
-   for (i = 0; i < ureg->nr_inputs; i++) {
-      if (ureg->input[i].semantic_name == name &&
-          ureg->input[i].semantic_index == index) 
+   for (i = 0; i < ureg->nr_fs_inputs; i++) {
+      if (ureg->fs_input[i].semantic_name == name &&
+          ureg->fs_input[i].semantic_index == index) 
          goto out;
    }
 
-   if (ureg->nr_inputs < UREG_MAX_INPUT) {
-      ureg->input[i].semantic_name = name;
-      ureg->input[i].semantic_index = index;
-      ureg->input[i].interp = interp_mode;
-      ureg->nr_inputs++;
+   if (ureg->nr_fs_inputs < UREG_MAX_INPUT) {
+      ureg->fs_input[i].semantic_name = name;
+      ureg->fs_input[i].semantic_index = index;
+      ureg->fs_input[i].interp = interp_mode;
+      ureg->nr_fs_inputs++;
    }
    else {
       set_bad( ureg );
@@ -257,25 +268,14 @@ out:
 }
 
 
-
-struct ureg_src 
-ureg_DECL_fs_input( struct ureg_program *ureg,
-                    unsigned name,
-                    unsigned index,
-                    unsigned interp )
-{
-   assert(ureg->processor == TGSI_PROCESSOR_FRAGMENT);
-   return ureg_DECL_input( ureg, name, index, interp );
-}
-
-
 struct ureg_src 
 ureg_DECL_vs_input( struct ureg_program *ureg,
-                    unsigned name,
                     unsigned index )
 {
    assert(ureg->processor == TGSI_PROCESSOR_VERTEX);
-   return ureg_DECL_input( ureg, name, index, TGSI_INTERPOLATE_CONSTANT );
+   
+   ureg->vs_inputs[index/32] |= 1 << (index % 32);
+   return ureg_src_register( TGSI_FILE_INPUT, index );
 }
 
 
@@ -313,9 +313,57 @@ out:
  * value or manage any constant_buffer contents -- that's the
  * resposibility of the calling code.
  */
-struct ureg_src ureg_DECL_constant(struct ureg_program *ureg )
+struct ureg_src ureg_DECL_constant(struct ureg_program *ureg, 
+                                   unsigned index )
 {
-   return ureg_src_register( TGSI_FILE_CONSTANT, ureg->nr_constants++ );
+   unsigned minconst = index, maxconst = index;
+   unsigned i;
+
+   /* Inside existing range?
+    */
+   for (i = 0; i < ureg->nr_constant_ranges; i++) {
+      if (ureg->constant_range[i].first <= index &&
+          ureg->constant_range[i].last >= index)
+         goto out;
+   }
+
+   /* Extend existing range?
+    */
+   for (i = 0; i < ureg->nr_constant_ranges; i++) {
+      if (ureg->constant_range[i].last == index - 1) {
+         ureg->constant_range[i].last = index;
+         goto out;
+      }
+
+      if (ureg->constant_range[i].first == index + 1) {
+         ureg->constant_range[i].first = index;
+         goto out;
+      }
+
+      minconst = MIN2(minconst, ureg->constant_range[i].first);
+      maxconst = MAX2(maxconst, ureg->constant_range[i].last);
+   }
+
+   /* Create new range?
+    */
+   if (ureg->nr_constant_ranges < UREG_MAX_CONSTANT_RANGE) {
+      i = ureg->nr_constant_ranges++;
+      ureg->constant_range[i].first = index;
+      ureg->constant_range[i].last = index;
+   }
+
+   /* Collapse all ranges down to one:
+    */
+   i = 0;
+   ureg->constant_range[0].first = minconst;
+   ureg->constant_range[0].last = maxconst;
+   ureg->nr_constant_ranges = 1;
+
+out:
+   assert(i < ureg->nr_constant_ranges);
+   assert(ureg->constant_range[i].first <= index);
+   assert(ureg->constant_range[i].last >= index);
+   return ureg_src_register( TGSI_FILE_CONSTANT, index );
 }
 
 
@@ -566,6 +614,19 @@ ureg_emit_dst( struct ureg_program *ureg,
 }
 
 
+static void validate( unsigned opcode,
+                      unsigned nr_dst,
+                      unsigned nr_src )
+{
+#ifdef DEBUG
+   const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode );
+   assert(info);
+   if(info) {
+      assert(nr_dst == info->num_dst);
+      assert(nr_src == info->num_src);
+   }
+#endif
+}
 
 unsigned
 ureg_emit_insn(struct ureg_program *ureg,
@@ -576,6 +637,8 @@ ureg_emit_insn(struct ureg_program *ureg,
 {
    union tgsi_any_token *out;
 
+   validate( opcode, num_dst, num_src );
+   
    out = get_tokens( ureg, DOMAIN_INSN, 1 );
    out[0].value = 0;
    out[0].insn.Type = TGSI_TOKEN_TYPE_INSTRUCTION;
@@ -678,17 +741,6 @@ ureg_insn(struct ureg_program *ureg,
    unsigned insn, i;
    boolean saturate;
 
-#ifdef DEBUG
-   {
-      const struct tgsi_opcode_info *info = tgsi_get_opcode_info( opcode );
-      assert(info);
-      if(info) {
-         assert(nr_dst == info->num_dst);
-         assert(nr_src == info->num_src);
-      }
-   }
-#endif
-   
    saturate = nr_dst ? dst[0].Saturate : FALSE;
 
    insn = ureg_emit_insn( ureg, opcode, saturate, nr_dst, nr_src );
@@ -702,6 +754,53 @@ ureg_insn(struct ureg_program *ureg,
    ureg_fixup_insn_size( ureg, insn );
 }
 
+void
+ureg_tex_insn(struct ureg_program *ureg,
+              unsigned opcode,
+              const struct ureg_dst *dst,
+              unsigned nr_dst,
+              unsigned target,
+              const struct ureg_src *src,
+              unsigned nr_src )
+{
+   unsigned insn, i;
+   boolean saturate;
+
+   saturate = nr_dst ? dst[0].Saturate : FALSE;
+
+   insn = ureg_emit_insn( ureg, opcode, saturate, nr_dst, nr_src );
+
+   ureg_emit_texture( ureg, insn, target );                             \
+
+   for (i = 0; i < nr_dst; i++)
+      ureg_emit_dst( ureg, dst[i] );
+
+   for (i = 0; i < nr_src; i++)
+      ureg_emit_src( ureg, src[i] );
+
+   ureg_fixup_insn_size( ureg, insn );
+}
+
+
+void
+ureg_label_insn(struct ureg_program *ureg,
+                unsigned opcode,
+                const struct ureg_src *src,
+                unsigned nr_src,
+                unsigned *label_token )
+{
+   unsigned insn, i;
+
+   insn = ureg_emit_insn( ureg, opcode, FALSE, 0, nr_src );
+
+   ureg_emit_label( ureg, insn, label_token );                  \
+
+   for (i = 0; i < nr_src; i++)
+      ureg_emit_src( ureg, src[i] );
+
+   ureg_fixup_insn_size( ureg, insn );
+}
+
 
 
 static void emit_decl( struct ureg_program *ureg,
@@ -777,13 +876,22 @@ static void emit_decls( struct ureg_program *ureg )
 {
    unsigned i;
 
-   for (i = 0; i < ureg->nr_inputs; i++) {
-      emit_decl( ureg, 
-                 TGSI_FILE_INPUT, 
-                 i,
-                 ureg->input[i].semantic_name,
-                 ureg->input[i].semantic_index,
-                 ureg->input[i].interp );
+   if (ureg->processor == TGSI_PROCESSOR_VERTEX) {
+      for (i = 0; i < UREG_MAX_INPUT; i++) {
+         if (ureg->vs_inputs[i/32] & (1 << (i%32))) {
+            emit_decl_range( ureg, TGSI_FILE_INPUT, i, 1 );
+         }
+      }
+   }
+   else {
+      for (i = 0; i < ureg->nr_fs_inputs; i++) {
+         emit_decl( ureg, 
+                    TGSI_FILE_INPUT, 
+                    i,
+                    ureg->fs_input[i].semantic_name,
+                    ureg->fs_input[i].semantic_index,
+                    ureg->fs_input[i].interp );
+      }
    }
 
    for (i = 0; i < ureg->nr_outputs; i++) {
@@ -801,10 +909,13 @@ static void emit_decls( struct ureg_program *ureg )
                        ureg->sampler[i].Index, 1 );
    }
 
-   if (ureg->nr_constants) {
-      emit_decl_range( ureg,
-                       TGSI_FILE_CONSTANT,
-                       0, ureg->nr_constants );
+   if (ureg->nr_constant_ranges) {
+      for (i = 0; i < ureg->nr_constant_ranges; i++)
+         emit_decl_range( ureg,
+                          TGSI_FILE_CONSTANT,
+                          ureg->constant_range[i].first, 
+                          (ureg->constant_range[i].last + 1 -
+                           ureg->constant_range[i].first) );
    }
 
    if (ureg->nr_temps) {
@@ -890,6 +1001,15 @@ const struct tgsi_token *ureg_finalize( struct ureg_program *ureg )
                    ureg->domain[DOMAIN_DECL].count);
       tgsi_dump( tokens, 0 );
    }
+
+#if DEBUG
+   if (tokens && !tgsi_sanity_check(tokens)) {
+      debug_printf("tgsi_ureg.c, sanity check failed on generated tokens:\n");
+      tgsi_dump(tokens, 0);
+      assert(0);
+   }
+#endif
+
    
    return tokens;
 }
@@ -911,6 +1031,25 @@ void *ureg_create_shader( struct ureg_program *ureg,
 }
 
 
+const struct tgsi_token *ureg_get_tokens( struct ureg_program *ureg,
+                                          unsigned *nr_tokens )
+{
+   const struct tgsi_token *tokens;
+
+   ureg_finalize(ureg);
+
+   tokens = &ureg->domain[DOMAIN_DECL].tokens[0].token;
+
+   if (nr_tokens) 
+      *nr_tokens = ureg->domain[DOMAIN_DECL].size;
+
+   ureg->domain[DOMAIN_DECL].tokens = 0;
+   ureg->domain[DOMAIN_DECL].size = 0;
+   ureg->domain[DOMAIN_DECL].order = 0;
+   ureg->domain[DOMAIN_DECL].count = 0;
+
+   return tokens;
+}
 
 
 struct ureg_program *ureg_create( unsigned processor )
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index acbca59040c..f04f443b9e7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -82,10 +82,21 @@ ureg_create( unsigned processor );
 const struct tgsi_token *
 ureg_finalize( struct ureg_program * );
 
+/* Create and return a shader:
+ */
 void *
 ureg_create_shader( struct ureg_program *,
                     struct pipe_context *pipe );
 
+
+/* Alternately, return the built token stream and hand ownership of
+ * that memory to the caller:
+ */
+const struct tgsi_token *
+ureg_get_tokens( struct ureg_program *ureg,
+                 unsigned *nr_tokens );
+
+
 void 
 ureg_destroy( struct ureg_program * );
 
@@ -116,8 +127,7 @@ ureg_DECL_fs_input( struct ureg_program *,
 
 struct ureg_src
 ureg_DECL_vs_input( struct ureg_program *,
-                    unsigned semantic_name,
-                    unsigned semantic_index );
+                    unsigned index );
 
 struct ureg_dst
 ureg_DECL_output( struct ureg_program *,
@@ -130,7 +140,8 @@ ureg_DECL_immediate( struct ureg_program *,
                      unsigned nr );
 
 struct ureg_src
-ureg_DECL_constant( struct ureg_program * );
+ureg_DECL_constant( struct ureg_program *,
+                    unsigned index );
 
 struct ureg_dst
 ureg_DECL_temporary( struct ureg_program * );
@@ -233,6 +244,24 @@ ureg_insn(struct ureg_program *ureg,
           unsigned nr_src );
 
 
+void
+ureg_tex_insn(struct ureg_program *ureg,
+              unsigned opcode,
+              const struct ureg_dst *dst,
+              unsigned nr_dst,
+              unsigned target,
+              const struct ureg_src *src,
+              unsigned nr_src );
+
+
+void
+ureg_label_insn(struct ureg_program *ureg,
+                unsigned opcode,
+                const struct ureg_src *src,
+                unsigned nr_src,
+                unsigned *label);
+
+
 /***********************************************************************
  * Internal instruction helpers, don't call these directly:
  */
diff --git a/src/gallium/auxiliary/util/u_debug.h b/src/gallium/auxiliary/util/u_debug.h
index d42b65ce281..1380d98d7ee 100644
--- a/src/gallium/auxiliary/util/u_debug.h
+++ b/src/gallium/auxiliary/util/u_debug.h
@@ -88,6 +88,7 @@ _debug_printf(const char *format, ...)
  * - avoid outputing large strings (512 bytes is the current maximum length 
  * that is guaranteed to be printed in all platforms)
  */
+#if !defined(PIPE_OS_HAIKU)
 static INLINE void
 debug_printf(const char *format, ...)
 {
@@ -101,6 +102,7 @@ debug_printf(const char *format, ...)
 #endif
 }
 
+#endif /* !PIPE_OS_HAIKU */
 
 /*
  * ... isn't portable so we need to pass arguments in parentheses.
diff --git a/src/gallium/auxiliary/util/u_fifo.h b/src/gallium/auxiliary/util/u_fifo.h
new file mode 100644
index 00000000000..9e007de1ada
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_fifo.h
@@ -0,0 +1,94 @@
+/**************************************************************************
+ *
+ * Copyright © 2009 Jakob Bornecrantz
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef U_FIFO_H
+#define U_FIFO_H
+
+#include "util/u_memory.h"
+
+struct util_fifo
+{
+   size_t head;
+   size_t tail;
+   size_t num;
+   size_t size;
+};
+
+static INLINE struct util_fifo *
+u_fifo_create(size_t size)
+{
+   struct util_fifo *fifo;
+   fifo = MALLOC(sizeof(*fifo) + size * sizeof(void*));
+
+   fifo->head = 0;
+   fifo->tail = 0;
+   fifo->num = 0;
+   fifo->size = size;
+
+   return fifo;
+}
+
+static INLINE boolean
+u_fifo_add(struct util_fifo *fifo, void *ptr)
+{
+   void **array = (void**)&fifo[1];
+   if (fifo->num >= fifo->size)
+      return FALSE;
+
+   if (++fifo->head >= fifo->size)
+      fifo->head = 0;
+
+   array[fifo->head] = ptr;
+
+   ++fifo->num;
+
+   return TRUE;
+}
+
+static INLINE boolean
+u_fifo_pop(struct util_fifo *fifo, void **ptr)
+{
+   void **array = (void**)&fifo[1];
+
+   if (!fifo->num)
+      return FALSE;
+
+   if (++fifo->tail >= fifo->size)
+      fifo->tail = 0;
+
+   *ptr = array[fifo->tail];
+
+   ++fifo->num;
+
+   return TRUE;
+}
+
+static INLINE void
+u_fifo_destroy(struct util_fifo *fifo)
+{
+   FREE(fifo);
+}
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv
index 00a46d0cc48..6e82983e586 100644
--- a/src/gallium/auxiliary/util/u_format.csv
+++ b/src/gallium/auxiliary/util/u_format.csv
@@ -14,10 +14,10 @@ PIPE_FORMAT_L16_UNORM             , arith , 1, 1, un16,     ,     ,     , xxx1,
 PIPE_FORMAT_Z16_UNORM             , array , 1, 1, un16,     ,     ,     , x___, zs 
 PIPE_FORMAT_Z32_UNORM             , array , 1, 1, un32,     ,     ,     , x___, zs 
 PIPE_FORMAT_Z32_FLOAT             , array , 1, 1, f32 ,     ,     ,     , x___, zs 
-PIPE_FORMAT_S8Z24_UNORM           , arith , 1, 1, un8 , un24,     ,     , yx__, zs 
-PIPE_FORMAT_Z24S8_UNORM           , arith , 1, 1, un24, un8 ,     ,     , xy__, zs 
-PIPE_FORMAT_X8Z24_UNORM           , arith , 1, 1, un8 , un24,     ,     , y___, zs 
-PIPE_FORMAT_Z24X8_UNORM           , arith , 1, 1, un24, un8 ,     ,     , x___, zs 
+PIPE_FORMAT_S8Z24_UNORM           , arith , 1, 1, un24, un8 ,     ,     , xy__, zs 
+PIPE_FORMAT_Z24S8_UNORM           , arith , 1, 1, un8 , un24,     ,     , yx__, zs 
+PIPE_FORMAT_X8Z24_UNORM           , arith , 1, 1, un24, un8 ,     ,     , x___, zs 
+PIPE_FORMAT_Z24X8_UNORM           , arith , 1, 1, un8 , un24,     ,     , y___, zs 
 PIPE_FORMAT_S8_UNORM              , array , 1, 1, un8 ,     ,     ,     , _x__, zs 
 PIPE_FORMAT_R64_FLOAT             , array , 1, 1, f64 ,     ,     ,     , x001, rgb
 PIPE_FORMAT_R64G64_FLOAT          , array , 1, 1, f64 , f64 ,     ,     , xy01, rgb
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index 4c6c2bc00e1..b12c97dfb4d 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -341,6 +341,16 @@ util_is_inf_or_nan(float x)
 
 
 /**
+ * Test whether x is a power of two.
+ */
+static INLINE boolean
+util_is_pot(unsigned x)
+{
+   return (x & (x - 1)) == 0;
+}
+
+
+/**
  * Find first bit set in word.  Least significant bit is 1.
  * Return 0 if no bits set.
  */
diff --git a/src/gallium/auxiliary/util/u_network.c b/src/gallium/auxiliary/util/u_network.c
index 07d804ecdbf..bc4b7584067 100644
--- a/src/gallium/auxiliary/util/u_network.c
+++ b/src/gallium/auxiliary/util/u_network.c
@@ -6,7 +6,7 @@
 #if defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 #  include <winsock2.h>
 #  include <windows.h>
-#elif defined(PIPE_OS_LINUX)
+#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
 #  include <sys/socket.h>
 #  include <netinet/in.h>
 #  include <unistd.h>
@@ -54,7 +54,7 @@ u_socket_close(int s)
    if (s < 0)
       return;
 
-#if defined(PIPE_OS_LINUX)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
    shutdown(s, SHUT_RDWR);
    close(s);
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
@@ -169,7 +169,7 @@ u_socket_listen_on_port(uint16_t portnum)
 void
 u_socket_block(int s, boolean block)
 {
-#if defined(PIPE_OS_LINUX)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
    int old = fcntl(s, F_GETFL, 0);
    if (old == -1)
       return;
diff --git a/src/gallium/auxiliary/util/u_network.h b/src/gallium/auxiliary/util/u_network.h
index 14d3884427e..8c778f492ca 100644
--- a/src/gallium/auxiliary/util/u_network.h
+++ b/src/gallium/auxiliary/util/u_network.h
@@ -6,7 +6,7 @@
 
 #if defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 #  define PIPE_HAVE_SOCKETS
-#elif defined(PIPE_OS_LINUX)
+#elif defined(PIPE_OS_LINUX) || defined(PIPE_OS_HAIKU)
 #  define PIPE_HAVE_SOCKETS
 #endif
 
diff --git a/src/gallium/auxiliary/util/u_simple_screen.c b/src/gallium/auxiliary/util/u_simple_screen.c
index f01296b40fc..52382990155 100644
--- a/src/gallium/auxiliary/util/u_simple_screen.c
+++ b/src/gallium/auxiliary/util/u_simple_screen.c
@@ -52,8 +52,7 @@ pass_user_buffer_create(struct pipe_screen *screen,
                         unsigned bytes)
 {
    struct pipe_buffer *buffer =
-      screen->winsys->user_buffer_create(screen->winsys,
-                                             ptr, bytes);
+      screen->winsys->user_buffer_create(screen->winsys, ptr, bytes);
 
    buffer->screen = screen;
 
@@ -69,9 +68,8 @@ pass_surface_buffer_create(struct pipe_screen *screen,
                            unsigned *stride)
 {
    struct pipe_buffer *buffer =
-      screen->winsys->surface_buffer_create(screen->winsys,
-                                                width, height,
-                                                format, usage, tex_usage, stride);
+      screen->winsys->surface_buffer_create(screen->winsys, width, height,
+                                            format, usage, tex_usage, stride);
 
    buffer->screen = screen;
 
@@ -83,8 +81,7 @@ pass_buffer_map(struct pipe_screen *screen,
                 struct pipe_buffer *buf,
                 unsigned usage)
 {
-   return screen->winsys->buffer_map(screen->winsys,
-                                     buf, usage);
+   return screen->winsys->buffer_map(screen->winsys, buf, usage);
 }
 
 static void
@@ -106,8 +103,7 @@ pass_flush_frontbuffer(struct pipe_screen *screen,
                        struct pipe_surface *surf,
                        void *context_private)
 {
-   screen->winsys->flush_frontbuffer(screen->winsys,
-                                     surf, context_private);
+   screen->winsys->flush_frontbuffer(screen->winsys, surf, context_private);
 }
 
 static void
@@ -115,8 +111,7 @@ pass_fence_reference(struct pipe_screen *screen,
                      struct pipe_fence_handle **ptr,
                      struct pipe_fence_handle *fence)
 {
-   screen->winsys->fence_reference(screen->winsys,
-                                   ptr, fence);
+   screen->winsys->fence_reference(screen->winsys, ptr, fence);
 }
 
 static int
@@ -124,8 +119,7 @@ pass_fence_signalled(struct pipe_screen *screen,
                      struct pipe_fence_handle *fence,
                      unsigned flag)
 {
-   return screen->winsys->fence_signalled(screen->winsys,
-                                          fence, flag);
+   return screen->winsys->fence_signalled(screen->winsys, fence, flag);
 }
 
 static int
@@ -133,11 +127,11 @@ pass_fence_finish(struct pipe_screen *screen,
                   struct pipe_fence_handle *fence,
                   unsigned flag)
 {
-   return screen->winsys->fence_finish(screen->winsys,
-                                       fence, flag);
+   return screen->winsys->fence_finish(screen->winsys, fence, flag);
 }
 
-void u_simple_screen_init(struct pipe_screen *screen)
+void
+u_simple_screen_init(struct pipe_screen *screen)
 {
    screen->buffer_create = pass_buffer_create;
    screen->user_buffer_create = pass_user_buffer_create;
@@ -152,7 +146,8 @@ void u_simple_screen_init(struct pipe_screen *screen)
    screen->fence_finish = pass_fence_finish;
 }
 
-const char* u_simple_screen_winsys_name(struct pipe_screen *screen)
+const char *
+u_simple_screen_winsys_name(struct pipe_screen *screen)
 {
    return screen->winsys->get_name(screen->winsys);
 }
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c
index 1b8da9b6853..0d706f9449d 100644
--- a/src/gallium/auxiliary/util/u_simple_shaders.c
+++ b/src/gallium/auxiliary/util/u_simple_shaders.c
@@ -34,14 +34,8 @@
 
 
 #include "pipe/p_context.h"
-#include "util/u_debug.h"
-#include "pipe/p_defines.h"
-#include "pipe/p_screen.h"
 #include "pipe/p_shader_tokens.h"
-
-#include "util/u_memory.h"
 #include "util/u_simple_shaders.h"
-
 #include "tgsi/tgsi_ureg.h"
 
 
@@ -67,9 +61,7 @@ util_make_vertex_passthrough_shader(struct pipe_context *pipe,
       struct ureg_src src;
       struct ureg_dst dst;
 
-      src = ureg_DECL_vs_input( ureg,
-                                semantic_names[i],
-                                semantic_indexes[i]);
+      src = ureg_DECL_vs_input( ureg, i );
       
       dst = ureg_DECL_output( ureg,
                               semantic_names[i],
@@ -131,8 +123,6 @@ util_make_fragment_tex_shader(struct pipe_context *pipe )
 
 
 
-
-
 /**
  * Make simple fragment color pass-through shader.
  */
diff --git a/src/gallium/auxiliary/util/u_stream_stdc.c b/src/gallium/auxiliary/util/u_stream_stdc.c
index d8f648e5dd1..5cd05b29047 100644
--- a/src/gallium/auxiliary/util/u_stream_stdc.c
+++ b/src/gallium/auxiliary/util/u_stream_stdc.c
@@ -32,7 +32,7 @@
 
 #include "pipe/p_config.h"
 
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
 
 #include <stdio.h>
 
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
index 1235a67d264..0d6489c26e4 100644
--- a/src/gallium/auxiliary/util/u_tile.c
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -170,7 +170,7 @@ x8r8g8b8_get_tile_rgba(const unsigned *src,
          pRow[0] = ubyte_to_float((pixel >> 16) & 0xff);
          pRow[1] = ubyte_to_float((pixel >>  8) & 0xff);
          pRow[2] = ubyte_to_float((pixel >>  0) & 0xff);
-         pRow[3] = ubyte_to_float(0xff);
+         pRow[3] = 1.0F;
       }
       p += dst_stride;
    }
@@ -394,6 +394,52 @@ r5g6b5_put_tile_rgba(ushort *dst,
 
 
 
+/*** PIPE_FORMAT_R8G8B8_UNORM ***/
+
+static void
+r8g8b8_get_tile_rgba(const ubyte *src,
+                     unsigned w, unsigned h,
+                     float *p,
+                     unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] = ubyte_to_float(src[0]);
+         pRow[1] = ubyte_to_float(src[1]);
+         pRow[2] = ubyte_to_float(src[2]);
+         pRow[3] = 1.0f;
+         src += 3;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r8g8b8_put_tile_rgba(ubyte *dst,
+                     unsigned w, unsigned h,
+                     const float *p,
+                     unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         dst[0] = float_to_ubyte(pRow[0]);
+         dst[1] = float_to_ubyte(pRow[1]);
+         dst[2] = float_to_ubyte(pRow[2]);
+         dst += 3;
+      }
+      p += src_stride;
+   }
+}
+
+
+
 /*** PIPE_FORMAT_Z16_UNORM ***/
 
 /**
@@ -1106,6 +1152,9 @@ pipe_tile_raw_to_rgba(enum pipe_format format,
    case PIPE_FORMAT_R5G6B5_UNORM:
       r5g6b5_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
       break;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      r8g8b8_get_tile_rgba((ubyte *) src, w, h, dst, dst_stride);
+      break;
    case PIPE_FORMAT_L8_UNORM:
       l8_get_tile_rgba((ubyte *) src, w, h, dst, dst_stride);
       break;
@@ -1222,6 +1271,9 @@ pipe_put_tile_rgba(struct pipe_transfer *pt,
    case PIPE_FORMAT_R5G6B5_UNORM:
       r5g6b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
       break;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      r8g8b8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);
+      break;
    case PIPE_FORMAT_R8G8B8A8_UNORM:
       assert(0);
       break;
diff --git a/src/gallium/auxiliary/util/u_time.c b/src/gallium/auxiliary/util/u_time.c
index c16cdd0b226..b958a986353 100644
--- a/src/gallium/auxiliary/util/u_time.c
+++ b/src/gallium/auxiliary/util/u_time.c
@@ -35,7 +35,7 @@
 
 #include "pipe/p_config.h"
 
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
 #include <sys/time.h>
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
 #include <windows.h>
@@ -77,7 +77,7 @@ util_time_get_frequency(void)
 void 
 util_time_get(struct util_time *t)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    gettimeofday(&t->tv, NULL);
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
    LONGLONG temp;
@@ -102,7 +102,7 @@ util_time_add(const struct util_time *t1,
               int64_t usecs,
               struct util_time *t2)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    t2->tv.tv_sec = t1->tv.tv_sec + usecs / 1000000;
    t2->tv.tv_usec = t1->tv.tv_usec + usecs % 1000000;
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
@@ -124,7 +124,7 @@ int64_t
 util_time_diff(const struct util_time *t1, 
                const struct util_time *t2)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    return (t2->tv.tv_usec - t1->tv.tv_usec) + 
           (t2->tv.tv_sec - t1->tv.tv_sec)*1000000;
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
@@ -144,7 +144,7 @@ util_time_micros( void )
    
    util_time_get(&t1);
    
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    return t1.tv.tv_usec + t1.tv.tv_sec*1000000LL;
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
    util_time_get_frequency();
@@ -166,7 +166,7 @@ static INLINE int
 util_time_compare(const struct util_time *t1, 
                   const struct util_time *t2)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    if (t1->tv.tv_sec < t2->tv.tv_sec)
       return -1;
    else if(t1->tv.tv_sec > t2->tv.tv_sec)
diff --git a/src/gallium/auxiliary/util/u_time.h b/src/gallium/auxiliary/util/u_time.h
index 7a5c54d9b23..a6189a247bb 100644
--- a/src/gallium/auxiliary/util/u_time.h
+++ b/src/gallium/auxiliary/util/u_time.h
@@ -43,6 +43,11 @@
 #include <unistd.h> /* usleep */
 #endif
 
+#if defined(PIPE_OS_HAIKU)
+#include <sys/time.h> /* timeval */
+#include <unistd.h>
+#endif
+
 #include "pipe/p_compiler.h"
 
 
@@ -58,7 +63,7 @@ extern "C" {
  */
 struct util_time 
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
    struct timeval tv;
 #else
    int64_t counter;
@@ -89,7 +94,7 @@ util_time_timeout(const struct util_time *start,
                   const struct util_time *end,
                   const struct util_time *curr);
 
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_APPLE) || defined(PIPE_OS_HAIKU)
 #define util_time_sleep usleep
 #else
 void
diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
index 508f4560e48..b3a7774fd6a 100644
--- a/src/gallium/drivers/i915simple/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
@@ -44,6 +44,7 @@
 #include "pipe/p_inlines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_fifo.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
@@ -76,8 +77,13 @@ struct i915_vbuf_render {
    size_t vbo_size;
    size_t vbo_offset;
    void *vbo_ptr;
-   size_t vbo_alloc_size;
    size_t vbo_max_used;
+
+   /* stuff for the pool */
+   struct util_fifo *pool_fifo;
+   unsigned pool_used;
+   unsigned pool_buffer_size;
+   boolean pool_not_used;
 };
 
 
@@ -106,33 +112,72 @@ i915_vbuf_render_get_vertex_info(struct vbuf_render *render)
 }
 
 static boolean
+i915_vbuf_render_reserve(struct i915_vbuf_render *i915_render, size_t size)
+{
+   struct i915_context *i915 = i915_render->i915;
+
+   if (i915_render->vbo_size < size + i915_render->vbo_offset)
+      return FALSE;
+
+   if (i915->vbo_flushed)
+      return FALSE;
+
+   return TRUE;
+}
+
+static void
+i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
+{
+   struct i915_context *i915 = i915_render->i915;
+   struct intel_winsys *iws = i915->iws;
+
+   if (i915_render->vbo) {
+      if (i915_render->pool_not_used)
+         iws->buffer_destroy(iws, i915_render->vbo);
+      else
+         u_fifo_add(i915_render->pool_fifo, i915_render->vbo);
+      i915_render->vbo = NULL;
+   }
+
+   i915->vbo_flushed = 0;
+
+   i915_render->vbo_size = MAX2(size, i915_render->pool_buffer_size);
+   i915_render->vbo_offset = 0;
+
+   if (i915_render->vbo_size != i915_render->pool_buffer_size) {
+      i915_render->pool_not_used = TRUE;
+      i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64,
+            INTEL_NEW_VERTEX);
+   } else {
+      i915_render->pool_not_used = FALSE;
+
+      if (i915_render->pool_used >= 2) {
+         FLUSH_BATCH(NULL);
+         i915->vbo_flushed = 0;
+         i915_render->pool_used = 0;
+      }
+      u_fifo_pop(i915_render->pool_fifo, (void**)&i915_render->vbo);
+   }
+}
+
+static boolean
 i915_vbuf_render_allocate_vertices(struct vbuf_render *render,
                                    ushort vertex_size,
                                    ushort nr_vertices)
 {
    struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
    struct i915_context *i915 = i915_render->i915;
-   struct intel_winsys *iws = i915->iws;
    size_t size = (size_t)vertex_size * (size_t)nr_vertices;
 
    /* FIXME: handle failure */
    assert(!i915->vbo);
 
-   if (i915_render->vbo_size > size + i915_render->vbo_offset && !i915->vbo_flushed) {
-   } else {
-      i915->vbo_flushed = 0;
-      if (i915_render->vbo) {
-         iws->buffer_destroy(iws, i915_render->vbo);
-         i915_render->vbo = NULL;
-      }
-   }
+   if (!i915_vbuf_render_reserve(i915_render, size)) {
 
-   if (!i915_render->vbo) {
-      i915_render->vbo_size = MAX2(size, i915_render->vbo_alloc_size);
-      i915_render->vbo_offset = 0;
-      i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64,
-                                            INTEL_NEW_VERTEX);
+      if (i915->vbo_flushed)
+         i915_render->pool_used = 0;
 
+      i915_vbuf_render_new_buf(i915_render, size);
    }
 
    i915_render->vertex_size = vertex_size;
@@ -504,6 +549,7 @@ i915_vbuf_render_create(struct i915_context *i915)
 {
    struct i915_vbuf_render *i915_render = CALLOC_STRUCT(i915_vbuf_render);
    struct intel_winsys *iws = i915->iws;
+   int i;
 
    i915_render->i915 = i915;
    
@@ -524,14 +570,24 @@ i915_vbuf_render_create(struct i915_context *i915)
    i915_render->base.release_vertices = i915_vbuf_render_release_vertices;
    i915_render->base.destroy = i915_vbuf_render_destroy;
 
-   i915_render->vbo_alloc_size = 128 * 4096;
-   i915_render->vbo_size = i915_render->vbo_alloc_size;
+
+   i915_render->vbo = NULL;
+   i915_render->vbo_size = 0;
    i915_render->vbo_offset = 0;
-   i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size, 64,
-                                         INTEL_NEW_VERTEX);
+
+   i915_render->pool_used = FALSE;
+   i915_render->pool_buffer_size = 128 * 4096;
+   i915_render->pool_fifo = u_fifo_create(6);
+   for (i = 0; i < 6; i++)
+      u_fifo_add(i915_render->pool_fifo,
+                 iws->buffer_create(iws, i915_render->pool_buffer_size, 64,
+                                    INTEL_NEW_VERTEX));
+
+#if 0
    /* TODO JB: is this realy needed? */
    i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE);
    iws->buffer_unmap(iws, i915_render->vbo);
+#endif
 
    return &i915_render->base;
 }
diff --git a/src/gallium/drivers/i915simple/intel_winsys.h b/src/gallium/drivers/i915simple/intel_winsys.h
index f949f52a9ce..42c5e7470ec 100644
--- a/src/gallium/drivers/i915simple/intel_winsys.h
+++ b/src/gallium/drivers/i915simple/intel_winsys.h
@@ -150,6 +150,17 @@ struct intel_winsys {
    void (*buffer_unmap)(struct intel_winsys *iws,
                         struct intel_buffer *buffer);
 
+   /**
+    * Write to a buffer.
+    *
+    * Arguments follows pwrite(2)
+    */
+   int (*buffer_write)(struct intel_winsys *iws,
+                       struct intel_buffer *dst,
+                       const void *src,
+                       size_t size,
+                       size_t offset);
+
    void (*buffer_destroy)(struct intel_winsys *iws,
                           struct intel_buffer *buffer);
    /*@}*/
diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile
index 6e63a0c2b76..06c586e6bb9 100644
--- a/src/gallium/drivers/llvmpipe/Makefile
+++ b/src/gallium/drivers/llvmpipe/Makefile
@@ -15,9 +15,11 @@ C_SOURCES = \
 	lp_bld_depth.c \
 	lp_bld_flow.c \
 	lp_bld_format_aos.c \
+	lp_bld_format_soa.c \
 	lp_bld_interp.c \
 	lp_bld_intr.c \
 	lp_bld_logic.c \
+	lp_bld_sample_soa.c \
 	lp_bld_swizzle.c \
 	lp_bld_struct.c \
 	lp_bld_tgsi_soa.c \
@@ -44,7 +46,8 @@ C_SOURCES = \
 	lp_state_vs.c \
 	lp_surface.c \
 	lp_tex_cache.c \
-	lp_tex_sample.c \
+	lp_tex_sample_c.c \
+	lp_tex_sample_llvm.c \
 	lp_texture.c \
 	lp_tile_cache.c \
 	lp_tile_soa.c
diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README
index 498d21dea6c..89d08834a3c 100644
--- a/src/gallium/drivers/llvmpipe/README
+++ b/src/gallium/drivers/llvmpipe/README
@@ -8,13 +8,16 @@ Done so far is:
 
  - the whole fragment pipeline is code generated in a single function
  
+   - input interpolation
+   
    - depth testing
  
+   - texture sampling (not all state/formats are supported) 
+   
    - fragment shader TGSI translation
      - same level of support as the TGSI SSE2 exec machine, with the exception
        we don't fallback to TGSI interpretation when an unsupported opcode is
        found, but just ignore it
-     - texture sampling via an intrinsic call
      - done in SoA layout
      - input interpolation also code generated
  
@@ -28,16 +31,17 @@ Done so far is:
      any width and length
    - not all operations are implemented for these types yet though
 
-Most mesa/progs/demos/* work. Speed is on par with Keith's softpipe-opt branch,
-which includes hand written fast implementations for common cases.
+Most mesa/progs/demos/* work. 
 
 To do (probably by this order):
 
  - code generate stipple and stencil testing
 
- - code generate texture sampling
+ - translate the remaining bits of texture sampling state
 
  - translate TGSI control flow instructions, and all other remaining opcodes
+ 
+ - integrate with the draw module for VS code generation
 
  - code generate the triangle setup and rasterization
 
@@ -93,7 +97,7 @@ Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 
   make linux-llvm
 
-but the rest of these instructions assume scons is used.
+but the rest of these instructions assume that scons is used.
 
 
 Using
@@ -108,6 +112,9 @@ or
 
   export LD_LIBRARY_PATH=$PWD/build/linux-x86-debug/lib:$LD_LIBRARY_PATH
 
+For performance evaluation pass debug=no to scons, and use the corresponding
+lib directory without the "-debug" suffix.
+
 
 Unit testing
 ============
@@ -119,7 +126,7 @@ build/linux-???-debug/gallium/drivers/llvmpipe:
  - lp_test_conv: SIMD vector conversion
  - lp_test_format: pixel unpacking/packing
 
-Some of this tests can output results and benchmarks to a tab-seperated-file
+Some of this tests can output results and benchmarks to a tab-separated-file
 for posterior analysis, e.g.:
 
   build/linux-x86_64-debug/gallium/drivers/llvmpipe/lp_test_blend -o blend.tsv
@@ -133,10 +140,10 @@ Development Notes
   at the top of the lp_bld_*.c functions.  
 
 - All lp_bld_*.[ch] are isolated from the rest of the driver, and could/may be 
-  put in a standalone Gallium state -> LLVM IR translation module.
+  put in a stand-alone Gallium state -> LLVM IR translation module.
 
 - We use LLVM-C bindings for now. They are not documented, but follow the C++
   interfaces very closely, and appear to be complete enough for code
   generation. See 
   http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html
-  for a standalone example.
+  for a stand-alone example.
diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript
index 5c29bdac56e..dea4b703c45 100644
--- a/src/gallium/drivers/llvmpipe/SConscript
+++ b/src/gallium/drivers/llvmpipe/SConscript
@@ -3,7 +3,7 @@ Import('*')
 env = env.Clone()
 
 env.Tool('llvm')
-if 'LLVM_VERSION' not in env:
+if env.has_key('LLVM_VERSION') is False:
     print 'warning: LLVM not found: not building llvmpipe'
     Return()
 
@@ -23,8 +23,10 @@ llvmpipe = env.ConvenienceLibrary(
 		'lp_bld_depth.c',
 		'lp_bld_flow.c',
 		'lp_bld_format_aos.c',
+		'lp_bld_format_soa.c',
 		'lp_bld_interp.c',
 		'lp_bld_intr.c',
+		'lp_bld_sample_soa.c',
 		'lp_bld_struct.c',
 		'lp_bld_logic.c',
 		'lp_bld_swizzle.c',
@@ -52,7 +54,8 @@ llvmpipe = env.ConvenienceLibrary(
 		'lp_state_vs.c',
 		'lp_surface.c',
 		'lp_tex_cache.c',
-		'lp_tex_sample.c',
+		'lp_tex_sample_c.c',
+		'lp_tex_sample_llvm.c',
 		'lp_texture.c',
 		'lp_tile_cache.c',
 		'lp_tile_soa.c',
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
index 09a57ff33d5..ce3e5f91c07 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c
@@ -502,6 +502,31 @@ lp_build_div(struct lp_build_context *bld,
 }
 
 
+LLVMValueRef
+lp_build_lerp(struct lp_build_context *bld,
+              LLVMValueRef x,
+              LLVMValueRef v0,
+              LLVMValueRef v1)
+{
+   return lp_build_add(bld, v0, lp_build_mul(bld, x, lp_build_sub(bld, v1, v0)));
+}
+
+
+LLVMValueRef
+lp_build_lerp_2d(struct lp_build_context *bld,
+                 LLVMValueRef x,
+                 LLVMValueRef y,
+                 LLVMValueRef v00,
+                 LLVMValueRef v01,
+                 LLVMValueRef v10,
+                 LLVMValueRef v11)
+{
+   LLVMValueRef v0 = lp_build_lerp(bld, x, v00, v01);
+   LLVMValueRef v1 = lp_build_lerp(bld, x, v10, v11);
+   return lp_build_lerp(bld, y, v0, v1);
+}
+
+
 /**
  * Generate min(a, b)
  * Do checks for special cases.
@@ -566,20 +591,31 @@ lp_build_abs(struct lp_build_context *bld,
              LLVMValueRef a)
 {
    const union lp_type type = bld->type;
+   LLVMTypeRef vec_type = lp_build_vec_type(type);
 
    if(!type.sign)
       return a;
 
-   /* XXX: is this really necessary? */
+   if(type.floating) {
+      /* Mask out the sign bit */
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMValueRef mask = lp_build_int_const_scalar(type, ((unsigned long long)1 << type.width) - 1);
+      a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+      a = LLVMBuildAnd(bld->builder, a, mask, "");
+      a = LLVMBuildBitCast(bld->builder, a, vec_type, "");
+      return a;
+   }
+
 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
-   if(!type.floating && type.width*type.length == 128) {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      if(type.width == 8)
+   if(type.width*type.length == 128) {
+      switch(type.width) {
+      case 8:
          return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.b.128", vec_type, a);
-      if(type.width == 16)
+      case 16:
          return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.w.128", vec_type, a);
-      if(type.width == 32)
+      case 32:
          return lp_build_intrinsic_unary(bld->builder, "llvm.x86.ssse3.pabs.d.128", vec_type, a);
+      }
    }
 #endif
 
@@ -588,6 +624,184 @@ lp_build_abs(struct lp_build_context *bld,
 
 
 LLVMValueRef
+lp_build_sgn(struct lp_build_context *bld,
+             LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMValueRef cond;
+   LLVMValueRef res;
+
+   /* Handle non-zero case */
+   if(!type.sign) {
+      /* if not zero then sign must be positive */
+      res = bld->one;
+   }
+   else if(type.floating) {
+      /* Take the sign bit and add it to 1 constant */
+      LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+      LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+      LLVMValueRef sign;
+      LLVMValueRef one;
+      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+      one = LLVMConstBitCast(bld->one, int_vec_type);
+      res = LLVMBuildOr(bld->builder, sign, one, "");
+      res = LLVMBuildBitCast(bld->builder, res, vec_type, "");
+   }
+   else
+   {
+      LLVMValueRef minus_one = lp_build_const_scalar(type, -1.0);
+      cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero);
+      res = lp_build_select(bld, cond, bld->one, minus_one);
+   }
+
+   /* Handle zero */
+   cond = lp_build_cmp(bld, PIPE_FUNC_EQUAL, a, bld->zero);
+   res = lp_build_select(bld, cond, bld->zero, bld->one);
+
+   return res;
+}
+
+
+enum lp_build_round_sse41_mode
+{
+   LP_BUILD_ROUND_SSE41_NEAREST = 0,
+   LP_BUILD_ROUND_SSE41_FLOOR = 1,
+   LP_BUILD_ROUND_SSE41_CEIL = 2,
+   LP_BUILD_ROUND_SSE41_TRUNCATE = 3
+};
+
+
+static INLINE LLVMValueRef
+lp_build_round_sse41(struct lp_build_context *bld,
+                     LLVMValueRef a,
+                     enum lp_build_round_sse41_mode mode)
+{
+   const union lp_type type = bld->type;
+   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   const char *intrinsic;
+
+   assert(type.floating);
+   assert(type.width*type.length == 128);
+
+   switch(type.width) {
+   case 32:
+      intrinsic = "llvm.x86.sse41.round.ps";
+      break;
+   case 64:
+      intrinsic = "llvm.x86.sse41.round.pd";
+      break;
+   default:
+      assert(0);
+      return bld->undef;
+   }
+
+   return lp_build_intrinsic_binary(bld->builder, intrinsic, vec_type, a,
+                                    LLVMConstInt(LLVMInt32Type(), mode, 0));
+}
+
+
+LLVMValueRef
+lp_build_round(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
+LLVMValueRef
+lp_build_floor(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
+LLVMValueRef
+lp_build_ceil(struct lp_build_context *bld,
+              LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
+LLVMValueRef
+lp_build_trunc(struct lp_build_context *bld,
+               LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+
+   assert(type.floating);
+
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
+#endif
+
+   /* FIXME */
+   assert(0);
+   return bld->undef;
+}
+
+
+/**
+ * Convert to integer, through whichever rounding method that's fastest,
+ * typically truncating to zero.
+ */
+LLVMValueRef
+lp_build_int(struct lp_build_context *bld,
+             LLVMValueRef a)
+{
+   const union lp_type type = bld->type;
+   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+
+   assert(type.floating);
+
+   return LLVMBuildFPToSI(bld->builder, a, int_vec_type, "");
+}
+
+
+LLVMValueRef
+lp_build_ifloor(struct lp_build_context *bld,
+                LLVMValueRef a)
+{
+   a = lp_build_floor(bld, a);
+   a = lp_build_int(bld, a);
+   return a;
+}
+
+
+LLVMValueRef
 lp_build_sqrt(struct lp_build_context *bld,
               LLVMValueRef a)
 {
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.h b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
index fc8cb25966e..5e083b847fd 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_arit.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.h
@@ -72,6 +72,26 @@ lp_build_div(struct lp_build_context *bld,
              LLVMValueRef b);
 
 LLVMValueRef
+lp_build_lerp(struct lp_build_context *bld,
+              LLVMValueRef x,
+              LLVMValueRef v0,
+              LLVMValueRef v1);
+
+/**
+ * Bilinear interpolation.
+ *
+ * Values indices are in v_{yx}.
+ */
+LLVMValueRef
+lp_build_lerp_2d(struct lp_build_context *bld,
+                 LLVMValueRef x,
+                 LLVMValueRef y,
+                 LLVMValueRef v00,
+                 LLVMValueRef v01,
+                 LLVMValueRef v10,
+                 LLVMValueRef v11);
+
+LLVMValueRef
 lp_build_min(struct lp_build_context *bld,
              LLVMValueRef a,
              LLVMValueRef b);
@@ -86,6 +106,34 @@ lp_build_abs(struct lp_build_context *bld,
              LLVMValueRef a);
 
 LLVMValueRef
+lp_build_sgn(struct lp_build_context *bld,
+             LLVMValueRef a);
+
+LLVMValueRef
+lp_build_round(struct lp_build_context *bld,
+               LLVMValueRef a);
+
+LLVMValueRef
+lp_build_floor(struct lp_build_context *bld,
+               LLVMValueRef a);
+
+LLVMValueRef
+lp_build_ceil(struct lp_build_context *bld,
+              LLVMValueRef a);
+
+LLVMValueRef
+lp_build_trunc(struct lp_build_context *bld,
+               LLVMValueRef a);
+
+LLVMValueRef
+lp_build_int(struct lp_build_context *bld,
+             LLVMValueRef a);
+
+LLVMValueRef
+lp_build_ifloor(struct lp_build_context *bld,
+                LLVMValueRef a);
+
+LLVMValueRef
 lp_build_sqrt(struct lp_build_context *bld,
               LLVMValueRef a);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.c b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
index c8954c8a34f..c5a71d2c726 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.c
@@ -122,7 +122,7 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
       int shift = dst_width - n;
       res = LLVMBuildShl(builder, res, lp_build_int_const_scalar(src_type, shift), "");
 
-      /* Fill in the empty lower bits for added precision? */
+      /* TODO: Fill in the empty lower bits for additional precision? */
 #if 0
       {
          LLVMValueRef msb;
@@ -244,7 +244,7 @@ lp_build_const_pack_shuffle(unsigned n)
  * Expand the bit width.
  *
  * This will only change the number of bits the values are represented, not the
- * values themselved.
+ * values themselves.
  */
 static void
 lp_build_expand(LLVMBuilderRef builder,
@@ -391,11 +391,11 @@ lp_build_pack2(LLVMBuilderRef builder,
  * TODO: Handle saturation consistently.
  */
 static LLVMValueRef
-lp_build_trunc(LLVMBuilderRef builder,
-               union lp_type src_type,
-               union lp_type dst_type,
-               boolean clamped,
-               const LLVMValueRef *src, unsigned num_srcs)
+lp_build_pack(LLVMBuilderRef builder,
+              union lp_type src_type,
+              union lp_type dst_type,
+              boolean clamped,
+              const LLVMValueRef *src, unsigned num_srcs)
 {
    LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH];
    unsigned i;
@@ -565,7 +565,7 @@ lp_build_conv(LLVMBuilderRef builder,
 
    if(tmp_type.width > dst_type.width) {
       assert(num_dsts == 1);
-      tmp[0] = lp_build_trunc(builder, tmp_type, dst_type, TRUE, tmp, num_tmps);
+      tmp[0] = lp_build_pack(builder, tmp_type, dst_type, TRUE, tmp, num_tmps);
       tmp_type.width = dst_type.width;
       tmp_type.length = dst_type.length;
       num_tmps = 1;
@@ -689,7 +689,7 @@ lp_build_conv_mask(LLVMBuilderRef builder,
 
    if(src_type.width > dst_type.width) {
       assert(num_dsts == 1);
-      dst[0] = lp_build_trunc(builder, src_type, dst_type, TRUE, src, num_srcs);
+      dst[0] = lp_build_pack(builder, src_type, dst_type, TRUE, src, num_srcs);
    }
    else if(src_type.width < dst_type.width) {
       assert(num_srcs == 1);
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_debug.c b/src/gallium/drivers/llvmpipe/lp_bld_debug.c
index 30925b5f415..59d8f492e60 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_debug.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_debug.c
@@ -30,10 +30,27 @@
 #include <udis86.h>
 #endif
 
+#include "util/u_math.h"
 #include "util/u_debug.h"
 #include "lp_bld_debug.h"
 
 
+/**
+ * Check alignment.
+ *
+ * It is important that this check is not implemented as a macro or inlined
+ * function, as the compiler assumptions in respect to alignment of global
+ * and stack variables would often make the check a no op, defeating the
+ * whole purpose of the exercise.
+ */
+boolean
+lp_check_alignment(const void *ptr, unsigned alignment)
+{
+   assert(util_is_pot(alignment));
+   return ((uintptr_t)ptr & (alignment - 1)) == 0;
+}
+
+
 void
 lp_disassemble(const void* func)
 {
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_debug.h b/src/gallium/drivers/llvmpipe/lp_bld_debug.h
index ecdafef76d0..583e6132b4b 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_debug.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_debug.h
@@ -53,6 +53,10 @@ lp_build_name(LLVMValueRef val, const char *format, ...)
 }
 
 
+boolean
+lp_check_alignment(const void *ptr, unsigned alignment);
+
+
 void
 lp_disassemble(const void* func);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
index 2cd6e6b9217..e5fe81193fc 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c
@@ -179,12 +179,13 @@ lp_build_depth_test(LLVMBuilderRef builder,
       padding_right = 0;
       for(chan = 0; chan < z_swizzle; ++chan)
          padding_right += format_desc->channel[chan].size;
-      padding_left = format_desc->block.bits - format_desc->channel[z_swizzle].size;
+      padding_left = format_desc->block.bits -
+                     (padding_right + format_desc->channel[z_swizzle].size);
 
       if(padding_left || padding_right) {
-         const long long mask_left = ((long long)1 << (format_desc->block.bits - padding_left)) - 1;
-         const long long mask_right = ((long long)1 << (padding_right)) - 1;
-         z_bitmask = lp_build_int_const_scalar(type, mask_left & mask_right);
+         const unsigned long long mask_left = ((unsigned long long)1 << (format_desc->block.bits - padding_left)) - 1;
+         const unsigned long long mask_right = ((unsigned long long)1 << (padding_right)) - 1;
+         z_bitmask = lp_build_int_const_scalar(type, mask_left ^ mask_right);
       }
 
       if(padding_left)
@@ -210,5 +211,6 @@ lp_build_depth_test(LLVMBuilderRef builder,
       LLVMBuildStore(builder, dst, dst_ptr);
    }
 
+   /* FIXME */
    assert(!state->occlusion_count);
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.c b/src/gallium/drivers/llvmpipe/lp_bld_flow.c
index 9d99e1a9d9f..69ed014ff3d 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_flow.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.c
@@ -32,59 +32,261 @@
  */
 
 #include "util/u_debug.h"
+#include "util/u_memory.h"
 
 #include "lp_bld_type.h"
 #include "lp_bld_flow.h"
 
 
+#define LP_BUILD_FLOW_MAX_VARIABLES 32
+#define LP_BUILD_FLOW_MAX_DEPTH 32
+
+
+/**
+ * Enumeration of all possible flow constructs.
+ */
+enum lp_build_flow_construct_kind {
+   lP_BUILD_FLOW_SCOPE,
+   LP_BUILD_FLOW_SKIP,
+};
+
+
+/**
+ * Variable declaration scope.
+ */
+struct lp_build_flow_scope
+{
+   /** Number of variables declared in this scope */
+   unsigned num_variables;
+};
+
+
+/**
+ * Early exit. Useful to skip to the end of a function or block when
+ * the execution mask becomes zero or when there is an error condition.
+ */
+struct lp_build_flow_skip
+{
+   /** Block to skip to */
+   LLVMBasicBlockRef block;
+
+   /** Number of variables declared at the beginning */
+   unsigned num_variables;
+
+   LLVMValueRef *phi;
+};
+
+
+/**
+ * Union of all possible flow constructs' data
+ */
+union lp_build_flow_construct_data
+{
+   struct lp_build_flow_scope scope;
+   struct lp_build_flow_skip skip;
+};
+
+
+/**
+ * Element of the flow construct stack.
+ */
+struct lp_build_flow_construct
+{
+   enum lp_build_flow_construct_kind kind;
+   union lp_build_flow_construct_data data;
+};
+
+
+/**
+ * All necessary data to generate LLVM control flow constructs.
+ *
+ * Besides keeping track of the control flow construct themselves we also
+ * need to keep track of variables in order to generate SSA Phi values.
+ */
+struct lp_build_flow_context
+{
+   LLVMBuilderRef builder;
+
+   /**
+    * Control flow stack.
+    */
+   struct lp_build_flow_construct constructs[LP_BUILD_FLOW_MAX_DEPTH];
+   unsigned num_constructs;
+
+   /**
+    * Variable stack
+    */
+   LLVMValueRef *variables[LP_BUILD_FLOW_MAX_VARIABLES];
+   unsigned num_variables;
+};
+
+
+struct lp_build_flow_context *
+lp_build_flow_create(LLVMBuilderRef builder)
+{
+   struct lp_build_flow_context *flow;
+
+   flow = CALLOC_STRUCT(lp_build_flow_context);
+   if(!flow)
+      return NULL;
+
+   flow->builder = builder;
+
+   return flow;
+}
+
+
 void
-lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    LLVMBuilderRef builder,
-                    union lp_type type,
-                    LLVMValueRef value)
+lp_build_flow_destroy(struct lp_build_flow_context *flow)
 {
-   memset(mask, 0, sizeof *mask);
+   assert(flow->num_constructs == 0);
+   assert(flow->num_variables == 0);
+   FREE(flow);
+}
 
-   mask->builder = builder;
-   mask->reg_type = LLVMIntType(type.width * type.length);
-   mask->value = value;
+
+static union lp_build_flow_construct_data *
+lp_build_flow_push(struct lp_build_flow_context *flow,
+                   enum lp_build_flow_construct_kind kind)
+{
+   assert(flow->num_constructs < LP_BUILD_FLOW_MAX_DEPTH);
+   if(flow->num_constructs >= LP_BUILD_FLOW_MAX_DEPTH)
+      return NULL;
+
+   flow->constructs[flow->num_constructs].kind = kind;
+   return &flow->constructs[flow->num_constructs++].data;
+}
+
+
+static union lp_build_flow_construct_data *
+lp_build_flow_peek(struct lp_build_flow_context *flow,
+                   enum lp_build_flow_construct_kind kind)
+{
+   assert(flow->num_constructs);
+   if(!flow->num_constructs)
+      return NULL;
+
+   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
+   if(flow->constructs[flow->num_constructs - 1].kind != kind)
+      return NULL;
+
+   return &flow->constructs[flow->num_constructs - 1].data;
 }
 
 
+static union lp_build_flow_construct_data *
+lp_build_flow_pop(struct lp_build_flow_context *flow,
+                  enum lp_build_flow_construct_kind kind)
+{
+   assert(flow->num_constructs);
+   if(!flow->num_constructs)
+      return NULL;
+
+   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
+   if(flow->constructs[flow->num_constructs - 1].kind != kind)
+      return NULL;
+
+   return &flow->constructs[--flow->num_constructs].data;
+}
+
+
+/**
+ * Begin a variable scope.
+ *
+ *
+ */
 void
-lp_build_mask_update(struct lp_build_mask_context *mask,
-                     LLVMValueRef value)
+lp_build_flow_scope_begin(struct lp_build_flow_context *flow)
 {
+   struct lp_build_flow_scope *scope;
 
-   LLVMValueRef cond;
-   LLVMBasicBlockRef current_block;
-   LLVMBasicBlockRef next_block;
-   LLVMBasicBlockRef new_block;
+   scope = &lp_build_flow_push(flow, lP_BUILD_FLOW_SCOPE)->scope;
+   if(!scope)
+      return;
 
-   if(mask->value)
-      mask->value = LLVMBuildAnd(mask->builder, mask->value, value, "");
-   else
-      mask->value = value;
+   scope->num_variables = 0;
+}
 
-   /* FIXME: disabled until we have proper control flow helpers */
-#if 0
-   cond = LLVMBuildICmp(mask->builder,
-                        LLVMIntEQ,
-                        LLVMBuildBitCast(mask->builder, mask->value, mask->reg_type, ""),
-                        LLVMConstNull(mask->reg_type),
-                        "");
 
-   current_block = LLVMGetInsertBlock(mask->builder);
+/**
+ * Declare a variable.
+ *
+ * A variable is a named entity which can have different LLVMValueRef's at
+ * different points of the program. This is relevant for control flow because
+ * when there are mutiple branches to a same location we need to replace
+ * the variable's value with a Phi function as explained in
+ * http://en.wikipedia.org/wiki/Static_single_assignment_form .
+ *
+ * We keep track of variables by keeping around a pointer to where their
+ * current.
+ *
+ * There are a few cautions to observe:
+ *
+ * - Variable's value must not be NULL. If there is no initial value then
+ *   LLVMGetUndef() should be used.
+ *
+ * - Variable's value must be kept up-to-date. If the variable is going to be
+ *   modified by a function then a pointer should be passed so that its value
+ *   is accurate. Failure to do this will cause some of the variables'
+ *   transient values to be lost, leading to wrong results.
+ *
+ * - A program should be written from top to bottom, by always appending
+ *   instructions to the bottom with a single LLVMBuilderRef. Inserting and/or
+ *   modifying existing statements will most likely lead to wrong results.
+ *
+ */
+void
+lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
+                            LLVMValueRef *variable)
+{
+   struct lp_build_flow_scope *scope;
+
+   scope = &lp_build_flow_peek(flow, lP_BUILD_FLOW_SCOPE)->scope;
+   if(!scope)
+      return;
 
-   if(!mask->skip_block) {
-      LLVMValueRef function = LLVMGetBasicBlockParent(current_block);
-      mask->skip_block = LLVMAppendBasicBlock(function, "skip");
+   assert(*variable);
+   if(!*variable)
+      return;
+
+   assert(flow->num_variables < LP_BUILD_FLOW_MAX_VARIABLES);
+   if(flow->num_variables >= LP_BUILD_FLOW_MAX_VARIABLES)
+      return;
+
+   flow->variables[flow->num_variables++] = variable;
+   ++scope->num_variables;
+}
+
+
+void
+lp_build_flow_scope_end(struct lp_build_flow_context *flow)
+{
+   struct lp_build_flow_scope *scope;
+
+   scope = &lp_build_flow_pop(flow, lP_BUILD_FLOW_SCOPE)->scope;
+   if(!scope)
+      return;
 
-      mask->phi = LLVMBuildPhi(mask->builder, LLVMTypeOf(mask->value), "");
+   assert(flow->num_variables >= scope->num_variables);
+   if(flow->num_variables < scope->num_variables) {
+      flow->num_variables = 0;
+      return;
    }
 
+   flow->num_variables -= scope->num_variables;
+}
+
+
+static LLVMBasicBlockRef
+lp_build_flow_insert_block(struct lp_build_flow_context *flow)
+{
+   LLVMBasicBlockRef current_block;
+   LLVMBasicBlockRef next_block;
+   LLVMBasicBlockRef new_block;
+
+   current_block = LLVMGetInsertBlock(flow->builder);
+
    next_block = LLVMGetNextBasicBlock(current_block);
-   assert(next_block);
    if(next_block) {
       new_block = LLVMInsertBasicBlock(next_block, "");
    }
@@ -93,30 +295,148 @@ lp_build_mask_update(struct lp_build_mask_context *mask,
       new_block = LLVMAppendBasicBlock(function, "");
    }
 
-   LLVMAddIncoming(mask->phi, &mask->value, &current_block, 1);
-   LLVMBuildCondBr(mask->builder, cond, mask->skip_block, new_block);
+   return new_block;
+}
+
+void
+lp_build_flow_skip_begin(struct lp_build_flow_context *flow)
+{
+   struct lp_build_flow_skip *skip;
+   LLVMBuilderRef builder;
+   unsigned i;
+
+   skip = &lp_build_flow_push(flow, LP_BUILD_FLOW_SKIP)->skip;
+   if(!skip)
+      return;
+
+   skip->block = lp_build_flow_insert_block(flow);
+   skip->num_variables = flow->num_variables;
+   if(!skip->num_variables) {
+      skip->phi = NULL;
+      return;
+   }
 
-   LLVMPositionBuilderAtEnd(mask->builder, new_block);
-#endif
+   skip->phi = MALLOC(skip->num_variables * sizeof *skip->phi);
+   if(!skip->phi) {
+      skip->num_variables = 0;
+      return;
+   }
+
+   builder = LLVMCreateBuilder();
+   LLVMPositionBuilderAtEnd(builder, skip->block);
+
+   for(i = 0; i < skip->num_variables; ++i)
+      skip->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), "");
+
+   LLVMDisposeBuilder(builder);
 }
 
 
-LLVMValueRef
-lp_build_mask_end(struct lp_build_mask_context *mask)
+void
+lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
+                              LLVMValueRef cond)
 {
-   if(mask->skip_block) {
-      LLVMBasicBlockRef current_block = LLVMGetInsertBlock(mask->builder);
+   struct lp_build_flow_skip *skip;
+   LLVMBasicBlockRef current_block;
+   LLVMBasicBlockRef new_block;
+   unsigned i;
 
-      LLVMAddIncoming(mask->phi, &mask->value, &current_block, 1);
-      LLVMBuildBr(mask->builder, mask->skip_block);
+   skip = &lp_build_flow_peek(flow, LP_BUILD_FLOW_SKIP)->skip;
+   if(!skip)
+      return;
 
-      LLVMPositionBuilderAtEnd(mask->builder, mask->skip_block);
+   current_block = LLVMGetInsertBlock(flow->builder);
 
-      mask->value = mask->phi;
-      mask->phi = NULL;
-      mask->skip_block = NULL;
+   new_block = lp_build_flow_insert_block(flow);
+
+   for(i = 0; i < skip->num_variables; ++i) {
+      assert(*flow->variables[i]);
+      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
    }
 
+   LLVMBuildCondBr(flow->builder, cond, skip->block, new_block);
+
+   LLVMPositionBuilderAtEnd(flow->builder, new_block);
+ }
+
+
+void
+lp_build_flow_skip_end(struct lp_build_flow_context *flow)
+{
+   struct lp_build_flow_skip *skip;
+   LLVMBasicBlockRef current_block;
+   unsigned i;
+
+   skip = &lp_build_flow_pop(flow, LP_BUILD_FLOW_SKIP)->skip;
+   if(!skip)
+      return;
+
+   current_block = LLVMGetInsertBlock(flow->builder);
+
+   for(i = 0; i < skip->num_variables; ++i) {
+      assert(*flow->variables[i]);
+      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
+      *flow->variables[i] = skip->phi[i];
+   }
+
+   LLVMBuildBr(flow->builder, skip->block);
+   LLVMPositionBuilderAtEnd(flow->builder, skip->block);
+
+   FREE(skip->phi);
+}
+
+
+static void
+lp_build_mask_check(struct lp_build_mask_context *mask)
+{
+   LLVMBuilderRef builder = mask->flow->builder;
+   LLVMValueRef cond;
+
+   cond = LLVMBuildICmp(builder,
+                        LLVMIntEQ,
+                        LLVMBuildBitCast(builder, mask->value, mask->reg_type, ""),
+                        LLVMConstNull(mask->reg_type),
+                        "");
+
+   lp_build_flow_skip_cond_break(mask->flow, cond);
+}
+
+
+void
+lp_build_mask_begin(struct lp_build_mask_context *mask,
+                    struct lp_build_flow_context *flow,
+                    union lp_type type,
+                    LLVMValueRef value)
+{
+   memset(mask, 0, sizeof *mask);
+
+   mask->flow = flow;
+   mask->reg_type = LLVMIntType(type.width * type.length);
+   mask->value = value;
+
+   lp_build_flow_scope_begin(flow);
+   lp_build_flow_scope_declare(flow, &mask->value);
+   lp_build_flow_skip_begin(flow);
+
+   lp_build_mask_check(mask);
+}
+
+
+void
+lp_build_mask_update(struct lp_build_mask_context *mask,
+                     LLVMValueRef value)
+{
+   mask->value = LLVMBuildAnd( mask->flow->builder, mask->value, value, "");
+
+   lp_build_mask_check(mask);
+}
+
+
+LLVMValueRef
+lp_build_mask_end(struct lp_build_mask_context *mask)
+{
+   lp_build_flow_skip_end(mask->flow);
+   lp_build_flow_scope_end(mask->flow);
    return mask->value;
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.h b/src/gallium/drivers/llvmpipe/lp_bld_flow.h
index 1b634ff038d..9d76e3064dd 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_flow.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.h
@@ -41,23 +41,49 @@
 union lp_type;
 
 
+struct lp_build_flow_context;
+
+
+struct lp_build_flow_context *
+lp_build_flow_create(LLVMBuilderRef builder);
+
+void
+lp_build_flow_destroy(struct lp_build_flow_context *flow);
+
+void
+lp_build_flow_scope_begin(struct lp_build_flow_context *flow);
+
+void
+lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
+                            LLVMValueRef *variable);
+
+void
+lp_build_flow_scope_end(struct lp_build_flow_context *flow);
+
+void
+lp_build_flow_skip_begin(struct lp_build_flow_context *flow);
+
+void
+lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
+                              LLVMValueRef cond);
+
+void
+lp_build_flow_skip_end(struct lp_build_flow_context *flow);
+
+
 struct lp_build_mask_context
 {
-   LLVMBuilderRef builder;
+   struct lp_build_flow_context *flow;
 
    LLVMTypeRef reg_type;
 
    LLVMValueRef value;
-
-   LLVMValueRef phi;
-
-   LLVMBasicBlockRef skip_block;
 };
 
 
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    LLVMBuilderRef builder,
+                    struct lp_build_flow_context *flow,
                     union lp_type type,
                     LLVMValueRef value);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format.h b/src/gallium/drivers/llvmpipe/lp_bld_format.h
index 01c8a752d18..5ee06560932 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_format.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_format.h
@@ -31,20 +31,14 @@
 
 /**
  * @file
- * LLVM IR building helpers interfaces.
- *
- * We use LLVM-C bindings for now. They are not documented, but follow the C++
- * interfaces very closely, and appear to be complete enough for code
- * genration. See
- * http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html
- * for a standalone example.
+ * Pixel format helpers.
  */
 
 #include <llvm-c/Core.h>  
- 
-#include "pipe/p_format.h"
 
+#include "pipe/p_format.h"
 
+struct util_format_description;
 union lp_type;
 
 
@@ -56,9 +50,9 @@ union lp_type;
  * @return RGBA in a 4 floats vector.
  */
 LLVMValueRef
-lp_build_unpack_rgba(LLVMBuilderRef builder,
-                     enum pipe_format format, 
-                     LLVMValueRef packed);
+lp_build_unpack_rgba_aos(LLVMBuilderRef builder,
+                         enum pipe_format format,
+                         LLVMValueRef packed);
 
 
 /**
@@ -67,9 +61,9 @@ lp_build_unpack_rgba(LLVMBuilderRef builder,
  * @param rgba 4 float vector with the unpacked components.
  */
 LLVMValueRef
-lp_build_pack_rgba(LLVMBuilderRef builder,
-                   enum pipe_format format,
-                   LLVMValueRef rgba);
+lp_build_pack_rgba_aos(LLVMBuilderRef builder,
+                       enum pipe_format format,
+                       LLVMValueRef rgba);
 
 
 /**
@@ -81,9 +75,9 @@ lp_build_pack_rgba(LLVMBuilderRef builder,
  * @return RGBA in a 4 floats vector.
  */
 LLVMValueRef
-lp_build_load_rgba(LLVMBuilderRef builder,
-                   enum pipe_format format, 
-                   LLVMValueRef ptr);
+lp_build_load_rgba_aos(LLVMBuilderRef builder,
+                       enum pipe_format format,
+                       LLVMValueRef ptr);
 
 
 /**
@@ -92,10 +86,34 @@ lp_build_load_rgba(LLVMBuilderRef builder,
  * @param rgba 4 float vector with the unpacked components.
  */
 void 
-lp_build_store_rgba(LLVMBuilderRef builder,
-                    enum pipe_format format,
-                    LLVMValueRef ptr,
-                    LLVMValueRef rgba);
+lp_build_store_rgba_aos(LLVMBuilderRef builder,
+                        enum pipe_format format,
+                        LLVMValueRef ptr,
+                        LLVMValueRef rgba);
 
+LLVMValueRef
+lp_build_gather(LLVMBuilderRef builder,
+                unsigned length,
+                unsigned src_width,
+                unsigned dst_width,
+                LLVMValueRef base_ptr,
+                LLVMValueRef offsets);
+
+
+void
+lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
+                         const struct util_format_description *format_desc,
+                         union lp_type type,
+                         LLVMValueRef packed,
+                         LLVMValueRef *rgba);
+
+
+void
+lp_build_load_rgba_soa(LLVMBuilderRef builder,
+                       const struct util_format_description *format_desc,
+                       union lp_type type,
+                       LLVMValueRef base_ptr,
+                       LLVMValueRef offsets,
+                       LLVMValueRef *rgba);
 
 #endif /* !LP_BLD_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
index dcbc0076c7d..b9b5d84bed5 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_format_aos.c
@@ -32,9 +32,9 @@
 
 
 LLVMValueRef
-lp_build_unpack_rgba(LLVMBuilderRef builder,
-                     enum pipe_format format,
-                     LLVMValueRef packed)
+lp_build_unpack_rgba_aos(LLVMBuilderRef builder,
+                         enum pipe_format format,
+                         LLVMValueRef packed)
 {
    const struct util_format_description *desc;
    LLVMTypeRef type;
@@ -152,9 +152,9 @@ lp_build_unpack_rgba(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_pack_rgba(LLVMBuilderRef builder,
-                   enum pipe_format format,
-                   LLVMValueRef rgba)
+lp_build_pack_rgba_aos(LLVMBuilderRef builder,
+                       enum pipe_format format,
+                       LLVMValueRef rgba)
 {
    const struct util_format_description *desc;
    LLVMTypeRef type;
@@ -250,9 +250,9 @@ lp_build_pack_rgba(LLVMBuilderRef builder,
 
 
 LLVMValueRef
-lp_build_load_rgba(LLVMBuilderRef builder,
-                   enum pipe_format format,
-                   LLVMValueRef ptr)
+lp_build_load_rgba_aos(LLVMBuilderRef builder,
+                       enum pipe_format format,
+                       LLVMValueRef ptr)
 {
    const struct util_format_description *desc;
    LLVMTypeRef type;
@@ -272,15 +272,15 @@ lp_build_load_rgba(LLVMBuilderRef builder,
 
    packed = LLVMBuildLoad(builder, ptr, "");
 
-   return lp_build_unpack_rgba(builder, format, packed);
+   return lp_build_unpack_rgba_aos(builder, format, packed);
 }
 
 
 void
-lp_build_store_rgba(LLVMBuilderRef builder,
-                    enum pipe_format format,
-                    LLVMValueRef ptr,
-                    LLVMValueRef rgba)
+lp_build_store_rgba_aos(LLVMBuilderRef builder,
+                        enum pipe_format format,
+                        LLVMValueRef ptr,
+                        LLVMValueRef rgba)
 {
    const struct util_format_description *desc;
    LLVMTypeRef type;
@@ -294,7 +294,7 @@ lp_build_store_rgba(LLVMBuilderRef builder,
 
    type = LLVMIntType(desc->block.bits);
 
-   packed = lp_build_pack_rgba(builder, format, rgba);
+   packed = lp_build_pack_rgba_aos(builder, format, rgba);
 
    ptr = LLVMBuildBitCast(builder, ptr, LLVMPointerType(type, 0), "");
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c
new file mode 100644
index 00000000000..569e8d10a31
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c
@@ -0,0 +1,208 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include "util/u_format.h"
+
+#include "lp_bld_type.h"
+#include "lp_bld_const.h"
+#include "lp_bld_conv.h"
+#include "lp_bld_format.h"
+
+
+/**
+ * Gather elements from scatter positions in memory into a single vector.
+ *
+ * @param src_width src element width
+ * @param dst_width result element width (source will be expanded to fit)
+ * @param length length of the offsets,
+ * @param base_ptr base pointer, should be a i8 pointer type.
+ * @param offsets vector with offsets
+ */
+LLVMValueRef
+lp_build_gather(LLVMBuilderRef builder,
+                unsigned length,
+                unsigned src_width,
+                unsigned dst_width,
+                LLVMValueRef base_ptr,
+                LLVMValueRef offsets)
+{
+   LLVMTypeRef src_type = LLVMIntType(src_width);
+   LLVMTypeRef src_ptr_type = LLVMPointerType(src_type, 0);
+   LLVMTypeRef dst_elem_type = LLVMIntType(dst_width);
+   LLVMTypeRef dst_vec_type = LLVMVectorType(dst_elem_type, length);
+   LLVMValueRef res;
+   unsigned i;
+
+   res = LLVMGetUndef(dst_vec_type);
+   for(i = 0; i < length; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef elem_offset;
+      LLVMValueRef elem_ptr;
+      LLVMValueRef elem;
+
+      elem_offset = LLVMBuildExtractElement(builder, offsets, index, "");
+      elem_ptr = LLVMBuildGEP(builder, base_ptr, &elem_offset, 1, "");
+      elem_ptr = LLVMBuildBitCast(builder, elem_ptr, src_ptr_type, "");
+      elem = LLVMBuildLoad(builder, elem_ptr, "");
+
+      assert(src_width <= dst_width);
+      if(src_width > dst_width)
+         elem = LLVMBuildTrunc(builder, elem, dst_elem_type, "");
+      if(src_width < dst_width)
+         elem = LLVMBuildZExt(builder, elem, dst_elem_type, "");
+
+      res = LLVMBuildInsertElement(builder, res, elem, index, "");
+   }
+
+   return res;
+}
+
+
+static LLVMValueRef
+lp_build_format_swizzle(union lp_type type,
+                        const LLVMValueRef *inputs,
+                        enum util_format_swizzle swizzle)
+{
+   switch (swizzle) {
+   case UTIL_FORMAT_SWIZZLE_X:
+   case UTIL_FORMAT_SWIZZLE_Y:
+   case UTIL_FORMAT_SWIZZLE_Z:
+   case UTIL_FORMAT_SWIZZLE_W:
+      return inputs[swizzle];
+   case UTIL_FORMAT_SWIZZLE_0:
+      return lp_build_zero(type);
+   case UTIL_FORMAT_SWIZZLE_1:
+      return lp_build_one(type);
+   case UTIL_FORMAT_SWIZZLE_NONE:
+      return lp_build_undef(type);
+   default:
+      assert(0);
+      return lp_build_undef(type);
+   }
+}
+
+
+void
+lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
+                         const struct util_format_description *format_desc,
+                         union lp_type type,
+                         LLVMValueRef packed,
+                         LLVMValueRef *rgba)
+{
+   LLVMValueRef inputs[4];
+   unsigned start;
+   unsigned chan;
+
+   /* FIXME: Support more formats */
+   assert(format_desc->layout == UTIL_FORMAT_LAYOUT_ARITH);
+   assert(format_desc->block.width == 1);
+   assert(format_desc->block.height == 1);
+   assert(format_desc->block.bits <= 32);
+
+   /* Decode the input vector components */
+   start = 0;
+   for (chan = 0; chan < 4; ++chan) {
+      unsigned width = format_desc->channel[chan].size;
+      unsigned stop = start + width;
+      LLVMValueRef input;
+
+      input = packed;
+
+      switch(format_desc->channel[chan].type) {
+      case UTIL_FORMAT_TYPE_VOID:
+         input = NULL;
+         break;
+
+      case UTIL_FORMAT_TYPE_UNSIGNED:
+         if(type.floating) {
+            if(start)
+               input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(type, start), "");
+            if(stop < format_desc->block.bits) {
+               unsigned mask = ((unsigned long long)1 << width) - 1;
+               input = LLVMBuildAnd(builder, input, lp_build_int_const_scalar(type, mask), "");
+            }
+
+            if(format_desc->channel[chan].normalized)
+               input = lp_build_unsigned_norm_to_float(builder, width, type, input);
+            else
+               input = LLVMBuildFPToSI(builder, input, lp_build_vec_type(type), "");
+         }
+         else {
+            /* FIXME */
+            assert(0);
+            input = lp_build_undef(type);
+         }
+         break;
+
+      default:
+         /* fall through */
+         input = lp_build_undef(type);
+         break;
+      }
+
+      inputs[chan] = input;
+
+      start = stop;
+   }
+
+   if(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
+      enum util_format_swizzle swizzle = format_desc->swizzle[0];
+      LLVMValueRef depth = lp_build_format_swizzle(type, inputs, swizzle);
+      rgba[2] = rgba[1] = rgba[0] = depth;
+      rgba[3] = lp_build_one(type);
+   }
+   else {
+      for (chan = 0; chan < 4; ++chan) {
+         enum util_format_swizzle swizzle = format_desc->swizzle[chan];
+         rgba[chan] = lp_build_format_swizzle(type, inputs, swizzle);
+      }
+   }
+}
+
+
+void
+lp_build_load_rgba_soa(LLVMBuilderRef builder,
+                       const struct util_format_description *format_desc,
+                       union lp_type type,
+                       LLVMValueRef base_ptr,
+                       LLVMValueRef offsets,
+                       LLVMValueRef *rgba)
+{
+   LLVMValueRef packed;
+
+   assert(format_desc->layout == UTIL_FORMAT_LAYOUT_ARITH);
+   assert(format_desc->block.width == 1);
+   assert(format_desc->block.height == 1);
+   assert(format_desc->block.bits <= 32);
+
+   packed = lp_build_gather(builder,
+                            type.length, format_desc->block.bits, type.width,
+                            base_ptr, offsets);
+
+   lp_build_unpack_rgba_soa(builder, format_desc, type, packed, rgba);
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.c b/src/gallium/drivers/llvmpipe/lp_bld_logic.c
index 8631efd6c3e..995a69c0f4c 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_logic.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.c
@@ -313,8 +313,6 @@ lp_build_select(struct lp_build_context *bld,
       b = LLVMBuildBitCast(bld->builder, b, int_vec_type, "");
    }
 
-   /* TODO: On SSE4 we could do this with a single instruction -- PBLENDVB */
-
    a = LLVMBuildAnd(bld->builder, a, mask, "");
 
    /* This often gets translated to PANDN, but sometimes the NOT is
@@ -339,7 +337,7 @@ LLVMValueRef
 lp_build_select_aos(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef b,
-                    boolean cond[4])
+                    const boolean cond[4])
 {
    const union lp_type type = bld->type;
    const unsigned n = type.length;
@@ -376,9 +374,9 @@ lp_build_select_aos(struct lp_build_context *bld,
 
       return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), "");
    }
+   else {
 #if 0
-   else if(0) {
-      /* FIXME: Unfortunately select of vectors do not work */
+      /* XXX: Unfortunately select of vectors do not work */
       /* Use a select */
       LLVMTypeRef elem_type = LLVMInt1Type();
       LLVMValueRef cond[LP_MAX_VECTOR_LENGTH];
@@ -388,10 +386,9 @@ lp_build_select_aos(struct lp_build_context *bld,
             cond[j + i] = LLVMConstInt(elem_type, cond[i] ? 1 : 0, 0);
 
       return LLVMBuildSelect(bld->builder, LLVMConstVector(cond, n), a, b, "");
-   }
-#endif
-   else {
+#else
       LLVMValueRef mask = lp_build_const_mask_aos(type, cond);
       return lp_build_select(bld, mask, a, b);
+#endif
    }
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.h b/src/gallium/drivers/llvmpipe/lp_bld_logic.h
index 29b9e1c45b8..9099e0fb5bf 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_logic.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.h
@@ -66,7 +66,7 @@ LLVMValueRef
 lp_build_select_aos(struct lp_build_context *bld,
                     LLVMValueRef a,
                     LLVMValueRef b,
-                    boolean cond[4]);
+                    const boolean cond[4]);
 
 
 #endif /* !LP_BLD_LOGIC_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_sample.h b/src/gallium/drivers/llvmpipe/lp_bld_sample.h
new file mode 100644
index 00000000000..6f565af76d1
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_bld_sample.h
@@ -0,0 +1,135 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Texture sampling.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef LP_BLD_SAMPLE_H
+#define LP_BLD_SAMPLE_H
+
+
+#include <llvm-c/Core.h>
+
+struct pipe_texture;
+struct pipe_sampler_state;
+union lp_type;
+
+
+/**
+ * Sampler static state.
+ *
+ * These are the bits of state from pipe_texture and pipe_sampler_state that
+ * are embedded in the generated code.
+ */
+struct lp_sampler_static_state
+{
+   /* pipe_texture's state */
+   enum pipe_format format;
+   unsigned target:2;
+   unsigned pot_width:1;
+   unsigned pot_height:1;
+   unsigned pot_depth:1;
+
+   /* pipe_sampler_state's state */
+   unsigned wrap_s:3;
+   unsigned wrap_t:3;
+   unsigned wrap_r:3;
+   unsigned min_img_filter:2;
+   unsigned min_mip_filter:2;
+   unsigned mag_img_filter:2;
+   unsigned compare_mode:1;
+   unsigned compare_func:3;
+   unsigned normalized_coords:1;
+   unsigned prefilter:4;
+};
+
+
+/**
+ * Sampler dynamic state.
+ *
+ * These are the bits of state from pipe_texture and pipe_sampler_state that
+ * are computed in runtime.
+ *
+ * There are obtained through callbacks, as we don't want to tie the texture
+ * sampling code generation logic to any particular texture layout or pipe
+ * driver.
+ */
+struct lp_sampler_dynamic_state
+{
+
+   /** Obtain the base texture width. */
+   LLVMValueRef
+   (*width)( struct lp_sampler_dynamic_state *state,
+             LLVMBuilderRef builder,
+             unsigned unit);
+
+   /** Obtain the base texture height. */
+   LLVMValueRef
+   (*height)( struct lp_sampler_dynamic_state *state,
+              LLVMBuilderRef builder,
+              unsigned unit);
+
+   LLVMValueRef
+   (*stride)( struct lp_sampler_dynamic_state *state,
+              LLVMBuilderRef builder,
+              unsigned unit);
+
+   LLVMValueRef
+   (*data_ptr)( struct lp_sampler_dynamic_state *state,
+                LLVMBuilderRef builder,
+                unsigned unit);
+
+};
+
+
+/**
+ * Derive the sampler static state.
+ */
+void
+lp_sampler_static_state(struct lp_sampler_static_state *state,
+                        const struct pipe_texture *texture,
+                        const struct pipe_sampler_state *sampler);
+
+
+void
+lp_build_sample_soa(LLVMBuilderRef builder,
+                    const struct lp_sampler_static_state *static_state,
+                    struct lp_sampler_dynamic_state *dynamic_state,
+                    union lp_type fp_type,
+                    unsigned unit,
+                    unsigned num_coords,
+                    const LLVMValueRef *coords,
+                    LLVMValueRef lodbias,
+                    LLVMValueRef *texel);
+
+
+
+#endif /* LP_BLD_SAMPLE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c
new file mode 100644
index 00000000000..3ca25b0e76c
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c
@@ -0,0 +1,411 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Texture sampling.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+#include "util/u_debug.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "util/u_format.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_type.h"
+#include "lp_bld_const.h"
+#include "lp_bld_arit.h"
+#include "lp_bld_logic.h"
+#include "lp_bld_swizzle.h"
+#include "lp_bld_format.h"
+#include "lp_bld_sample.h"
+
+
+void
+lp_sampler_static_state(struct lp_sampler_static_state *state,
+                        const struct pipe_texture *texture,
+                        const struct pipe_sampler_state *sampler)
+{
+   memset(state, 0, sizeof *state);
+
+   if(!texture)
+      return;
+
+   if(!sampler)
+      return;
+
+   state->format            = texture->format;
+   state->target            = texture->target;
+   state->pot_width         = util_is_pot(texture->width[0]);
+   state->pot_height        = util_is_pot(texture->height[0]);
+   state->pot_depth         = util_is_pot(texture->depth[0]);
+
+   state->wrap_s            = sampler->wrap_s;
+   state->wrap_t            = sampler->wrap_t;
+   state->wrap_r            = sampler->wrap_r;
+   state->min_img_filter    = sampler->min_img_filter;
+   state->min_mip_filter    = sampler->min_mip_filter;
+   state->mag_img_filter    = sampler->mag_img_filter;
+   if(sampler->compare_mode) {
+      state->compare_mode      = sampler->compare_mode;
+      state->compare_func      = sampler->compare_func;
+   }
+   state->normalized_coords = sampler->normalized_coords;
+   state->prefilter         = sampler->prefilter;
+}
+
+
+
+/**
+ * Keep all information for sampling code generation in a single place.
+ */
+struct lp_build_sample_context
+{
+   LLVMBuilderRef builder;
+
+   const struct lp_sampler_static_state *static_state;
+
+   struct lp_sampler_dynamic_state *dynamic_state;
+
+   const struct util_format_description *format_desc;
+
+   /** Incoming coordinates type and build context */
+   union lp_type coord_type;
+   struct lp_build_context coord_bld;
+
+   /** Integer coordinates */
+   union lp_type int_coord_type;
+   struct lp_build_context int_coord_bld;
+
+   /** Output texels type and build context */
+   union lp_type texel_type;
+   struct lp_build_context texel_bld;
+};
+
+
+static void
+lp_build_sample_texel(struct lp_build_sample_context *bld,
+                      LLVMValueRef x,
+                      LLVMValueRef y,
+                      LLVMValueRef y_stride,
+                      LLVMValueRef data_ptr,
+                      LLVMValueRef *texel)
+{
+   struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMValueRef x_stride;
+   LLVMValueRef offset;
+
+   x_stride = lp_build_const_scalar(bld->int_coord_type, bld->format_desc->block.bits/8);
+
+   if(bld->format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
+      LLVMValueRef x_lo, x_hi;
+      LLVMValueRef y_lo, y_hi;
+      LLVMValueRef x_stride_lo, x_stride_hi;
+      LLVMValueRef y_stride_lo, y_stride_hi;
+      LLVMValueRef x_offset_lo, x_offset_hi;
+      LLVMValueRef y_offset_lo, y_offset_hi;
+      LLVMValueRef offset_lo, offset_hi;
+
+      x_lo = LLVMBuildAnd(bld->builder, x, int_coord_bld->one, "");
+      y_lo = LLVMBuildAnd(bld->builder, y, int_coord_bld->one, "");
+
+      x_hi = LLVMBuildLShr(bld->builder, x, int_coord_bld->one, "");
+      y_hi = LLVMBuildLShr(bld->builder, y, int_coord_bld->one, "");
+
+      x_stride_lo = x_stride;
+      y_stride_lo = lp_build_const_scalar(bld->int_coord_type, 2*bld->format_desc->block.bits/8);
+
+      x_stride_hi = lp_build_const_scalar(bld->int_coord_type, 4*bld->format_desc->block.bits/8);
+      y_stride_hi = LLVMBuildShl(bld->builder, y_stride, int_coord_bld->one, "");
+
+      x_offset_lo = lp_build_mul(int_coord_bld, x_lo, x_stride_lo);
+      y_offset_lo = lp_build_mul(int_coord_bld, y_lo, y_stride_lo);
+      offset_lo = lp_build_add(int_coord_bld, x_offset_lo, y_offset_lo);
+
+      x_offset_hi = lp_build_mul(int_coord_bld, x_hi, x_stride_hi);
+      y_offset_hi = lp_build_mul(int_coord_bld, y_hi, y_stride_hi);
+      offset_hi = lp_build_add(int_coord_bld, x_offset_hi, y_offset_hi);
+
+      offset = lp_build_add(int_coord_bld, offset_hi, offset_lo);
+   }
+   else {
+      LLVMValueRef x_offset;
+      LLVMValueRef y_offset;
+
+      x_offset = lp_build_mul(int_coord_bld, x, x_stride);
+      y_offset = lp_build_mul(int_coord_bld, y, y_stride);
+
+      offset = lp_build_add(int_coord_bld, x_offset, y_offset);
+   }
+
+   lp_build_load_rgba_soa(bld->builder,
+                          bld->format_desc,
+                          bld->texel_type,
+                          data_ptr,
+                          offset,
+                          texel);
+}
+
+
+static LLVMValueRef
+lp_build_sample_wrap(struct lp_build_sample_context *bld,
+                     LLVMValueRef coord,
+                     LLVMValueRef length,
+                     boolean is_pot,
+                     unsigned wrap_mode)
+{
+   struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+   LLVMValueRef length_minus_one;
+
+   length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
+
+   switch(wrap_mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      if(is_pot)
+         coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, "");
+      else
+         /* Signed remainder won't give the right results for negative
+          * dividends but unsigned remainder does.*/
+         coord = LLVMBuildURem(bld->builder, coord, length, "");
+      break;
+
+   case PIPE_TEX_WRAP_CLAMP:
+      coord = lp_build_max(int_coord_bld, coord, int_coord_bld->zero);
+      coord = lp_build_min(int_coord_bld, coord, length_minus_one);
+      break;
+
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      /* FIXME */
+   default:
+      assert(0);
+   }
+
+   return coord;
+}
+
+
+static void
+lp_build_sample_2d_nearest_soa(struct lp_build_sample_context *bld,
+                               LLVMValueRef s,
+                               LLVMValueRef t,
+                               LLVMValueRef width,
+                               LLVMValueRef height,
+                               LLVMValueRef stride,
+                               LLVMValueRef data_ptr,
+                               LLVMValueRef *texel)
+{
+   LLVMValueRef x;
+   LLVMValueRef y;
+
+   x = lp_build_ifloor(&bld->coord_bld, s);
+   y = lp_build_ifloor(&bld->coord_bld, t);
+
+   x = lp_build_sample_wrap(bld, x, width,  bld->static_state->pot_width,  bld->static_state->wrap_s);
+   y = lp_build_sample_wrap(bld, y, height, bld->static_state->pot_height, bld->static_state->wrap_t);
+
+   lp_build_sample_texel(bld, x, y, stride, data_ptr, texel);
+}
+
+
+static void
+lp_build_sample_2d_linear_soa(struct lp_build_sample_context *bld,
+                              LLVMValueRef s,
+                              LLVMValueRef t,
+                              LLVMValueRef width,
+                              LLVMValueRef height,
+                              LLVMValueRef stride,
+                              LLVMValueRef data_ptr,
+                              LLVMValueRef *texel)
+{
+   LLVMValueRef half;
+   LLVMValueRef s_ipart;
+   LLVMValueRef t_ipart;
+   LLVMValueRef s_fpart;
+   LLVMValueRef t_fpart;
+   LLVMValueRef x0, x1;
+   LLVMValueRef y0, y1;
+   LLVMValueRef neighbors[2][2][4];
+   unsigned chan;
+
+   half = lp_build_const_scalar(bld->coord_type, 0.5);
+   s = lp_build_sub(&bld->coord_bld, s, half);
+   t = lp_build_sub(&bld->coord_bld, t, half);
+
+   s_ipart = lp_build_floor(&bld->coord_bld, s);
+   t_ipart = lp_build_floor(&bld->coord_bld, t);
+
+   s_fpart = lp_build_sub(&bld->coord_bld, s, s_ipart);
+   t_fpart = lp_build_sub(&bld->coord_bld, t, t_ipart);
+
+   x0 = lp_build_int(&bld->coord_bld, s_ipart);
+   y0 = lp_build_int(&bld->coord_bld, t_ipart);
+
+   x0 = lp_build_sample_wrap(bld, x0, width,  bld->static_state->pot_width,  bld->static_state->wrap_s);
+   y0 = lp_build_sample_wrap(bld, y0, height, bld->static_state->pot_height, bld->static_state->wrap_t);
+
+   x1 = lp_build_add(&bld->int_coord_bld, x0, bld->int_coord_bld.one);
+   y1 = lp_build_add(&bld->int_coord_bld, y0, bld->int_coord_bld.one);
+
+   x1 = lp_build_sample_wrap(bld, x1, width,  bld->static_state->pot_width,  bld->static_state->wrap_s);
+   y1 = lp_build_sample_wrap(bld, y1, height, bld->static_state->pot_height, bld->static_state->wrap_t);
+
+   lp_build_sample_texel(bld, x0, y0, stride, data_ptr, neighbors[0][0]);
+   lp_build_sample_texel(bld, x1, y0, stride, data_ptr, neighbors[0][1]);
+   lp_build_sample_texel(bld, x0, y1, stride, data_ptr, neighbors[1][0]);
+   lp_build_sample_texel(bld, x1, y1, stride, data_ptr, neighbors[1][1]);
+
+   /* TODO: Don't interpolate missing channels */
+   for(chan = 0; chan < 4; ++chan) {
+      texel[chan] = lp_build_lerp_2d(&bld->texel_bld,
+                                     s_fpart, t_fpart,
+                                     neighbors[0][0][chan],
+                                     neighbors[0][1][chan],
+                                     neighbors[1][0][chan],
+                                     neighbors[1][1][chan]);
+   }
+}
+
+
+static void
+lp_build_sample_compare(struct lp_build_sample_context *bld,
+                        LLVMValueRef p,
+                        LLVMValueRef *texel)
+{
+   struct lp_build_context *texel_bld = &bld->texel_bld;
+   LLVMValueRef res;
+   unsigned chan;
+
+   if(!bld->static_state->compare_mode)
+      return;
+
+   /* TODO: Compare before swizzling, to avoid redundant computations */
+   res = NULL;
+   for(chan = 0; chan < 4; ++chan) {
+      LLVMValueRef cmp;
+      cmp = lp_build_cmp(texel_bld, bld->static_state->compare_func, p, texel[chan]);
+      cmp = lp_build_select(texel_bld, cmp, texel_bld->one, texel_bld->zero);
+
+      if(res)
+         res = lp_build_add(texel_bld, res, cmp);
+      else
+         res = cmp;
+   }
+
+   assert(res);
+   res = lp_build_mul(texel_bld, res, lp_build_const_scalar(texel_bld->type, 0.25));
+
+   /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
+   for(chan = 0; chan < 3; ++chan)
+      texel[chan] = res;
+   texel[3] = texel_bld->one;
+}
+
+
+void
+lp_build_sample_soa(LLVMBuilderRef builder,
+                    const struct lp_sampler_static_state *static_state,
+                    struct lp_sampler_dynamic_state *dynamic_state,
+                    union lp_type type,
+                    unsigned unit,
+                    unsigned num_coords,
+                    const LLVMValueRef *coords,
+                    LLVMValueRef lodbias,
+                    LLVMValueRef *texel)
+{
+   struct lp_build_sample_context bld;
+   LLVMValueRef width;
+   LLVMValueRef height;
+   LLVMValueRef stride;
+   LLVMValueRef data_ptr;
+   LLVMValueRef s;
+   LLVMValueRef t;
+   LLVMValueRef p;
+
+   /* Setup our build context */
+   memset(&bld, 0, sizeof bld);
+   bld.builder = builder;
+   bld.static_state = static_state;
+   bld.dynamic_state = dynamic_state;
+   bld.format_desc = util_format_description(static_state->format);
+   bld.coord_type = type;
+   bld.int_coord_type = lp_int_type(type);
+   bld.texel_type = type;
+   lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
+   lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
+   lp_build_context_init(&bld.texel_bld, builder, bld.texel_type);
+
+   /* Get the dynamic state */
+   width = dynamic_state->width(dynamic_state, builder, unit);
+   height = dynamic_state->height(dynamic_state, builder, unit);
+   stride = dynamic_state->stride(dynamic_state, builder, unit);
+   data_ptr = dynamic_state->data_ptr(dynamic_state, builder, unit);
+
+   s = coords[0];
+   t = coords[1];
+   p = coords[2];
+
+   width = lp_build_broadcast_scalar(&bld.int_coord_bld, width);
+   height = lp_build_broadcast_scalar(&bld.int_coord_bld, height);
+   stride = lp_build_broadcast_scalar(&bld.int_coord_bld, stride);
+
+   if(static_state->target == PIPE_TEXTURE_1D)
+      t = bld.coord_bld.zero;
+
+   if(static_state->normalized_coords) {
+      LLVMTypeRef coord_vec_type = lp_build_vec_type(bld.coord_type);
+      LLVMValueRef fp_width = LLVMBuildSIToFP(builder, width, coord_vec_type, "");
+      LLVMValueRef fp_height = LLVMBuildSIToFP(builder, height, coord_vec_type, "");
+      s = lp_build_mul(&bld.coord_bld, s, fp_width);
+      t = lp_build_mul(&bld.coord_bld, t, fp_height);
+   }
+
+   switch (static_state->min_img_filter) {
+   case PIPE_TEX_FILTER_NEAREST:
+      lp_build_sample_2d_nearest_soa(&bld, s, t, width, height, stride, data_ptr, texel);
+      break;
+   case PIPE_TEX_FILTER_LINEAR:
+   case PIPE_TEX_FILTER_ANISO:
+      lp_build_sample_2d_linear_soa(&bld, s, t, width, height, stride, data_ptr, texel);
+      break;
+   default:
+      assert(0);
+   }
+
+   /* FIXME: respect static_state->min_mip_filter */;
+   /* FIXME: respect static_state->mag_img_filter */;
+   /* FIXME: respect static_state->prefilter */;
+
+   lp_build_sample_compare(&bld, p, texel);
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_struct.c b/src/gallium/drivers/llvmpipe/lp_bld_struct.c
index 14d2b10df9c..3998ac374fe 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_struct.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_struct.c
@@ -42,17 +42,30 @@
 
 
 LLVMValueRef
+lp_build_struct_get_ptr(LLVMBuilderRef builder,
+                        LLVMValueRef ptr,
+                        unsigned member,
+                        const char *name)
+{
+   LLVMValueRef indices[2];
+   LLVMValueRef member_ptr;
+   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[1] = LLVMConstInt(LLVMInt32Type(), member, 0);
+   member_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), "");
+   lp_build_name(member_ptr, "%s.%s_ptr", LLVMGetValueName(ptr), name);
+   return member_ptr;
+}
+
+
+LLVMValueRef
 lp_build_struct_get(LLVMBuilderRef builder,
                     LLVMValueRef ptr,
                     unsigned member,
                     const char *name)
 {
-   LLVMValueRef indices[2];
    LLVMValueRef member_ptr;
    LLVMValueRef res;
-   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
-   indices[1] = LLVMConstInt(LLVMInt32Type(), member, 0);
-   member_ptr = LLVMBuildGEP(builder, ptr, indices, Elements(indices), "");
+   member_ptr = lp_build_struct_get_ptr(builder, ptr, member, name);
    res = LLVMBuildLoad(builder, member_ptr, "");
    lp_build_name(res, "%s.%s", LLVMGetValueName(ptr), name);
    return res;
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_struct.h b/src/gallium/drivers/llvmpipe/lp_bld_struct.h
index cbefdc9f815..740392f5611 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_struct.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_struct.h
@@ -53,6 +53,18 @@
              offsetof(_ctype, _cmember))
 
 
+/**
+ * Get value pointer to a structure member.
+ */
+LLVMValueRef
+lp_build_struct_get_ptr(LLVMBuilderRef builder,
+                        LLVMValueRef ptr,
+                        unsigned member,
+                        const char *name);
+
+/**
+ * Get the value of a structure member.
+ */
 LLVMValueRef
 lp_build_struct_get(LLVMBuilderRef builder,
                     LLVMValueRef ptr,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c
index ac7eed9379a..f35638be447 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c
@@ -161,7 +161,7 @@ lp_build_broadcast_aos(struct lp_build_context *bld,
 LLVMValueRef
 lp_build_swizzle1_aos(struct lp_build_context *bld,
                       LLVMValueRef a,
-                      unsigned char swizzle[4])
+                      const unsigned char swizzle[4])
 {
    const unsigned n = bld->type.length;
    unsigned i, j;
@@ -192,7 +192,7 @@ LLVMValueRef
 lp_build_swizzle2_aos(struct lp_build_context *bld,
                       LLVMValueRef a,
                       LLVMValueRef b,
-                      unsigned char swizzle[4])
+                      const unsigned char swizzle[4])
 {
    const unsigned n = bld->type.length;
    unsigned i, j;
@@ -201,11 +201,12 @@ lp_build_swizzle2_aos(struct lp_build_context *bld,
       return lp_build_swizzle1_aos(bld, a, swizzle);
 
    if(a == b) {
-      swizzle[0] %= 4;
-      swizzle[1] %= 4;
-      swizzle[2] %= 4;
-      swizzle[3] %= 4;
-      return lp_build_swizzle1_aos(bld, a, swizzle);
+      unsigned char swizzle1[4];
+      swizzle1[0] = swizzle[0] % 4;
+      swizzle1[1] = swizzle[1] % 4;
+      swizzle1[2] = swizzle[2] % 4;
+      swizzle1[3] = swizzle[3] % 4;
+      return lp_build_swizzle1_aos(bld, a, swizzle1);
    }
 
    if(swizzle[0] % 4 == 0 &&
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h
index d7dd6a8a604..cb0b6707ec8 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h
@@ -73,7 +73,7 @@ lp_build_broadcast_aos(struct lp_build_context *bld,
 LLVMValueRef
 lp_build_swizzle1_aos(struct lp_build_context *bld,
                       LLVMValueRef a,
-                      unsigned char swizzle[4]);
+                      const unsigned char swizzle[4]);
 
 
 /**
@@ -85,7 +85,7 @@ LLVMValueRef
 lp_build_swizzle2_aos(struct lp_build_context *bld,
                       LLVMValueRef a,
                       LLVMValueRef b,
-                      unsigned char swizzle[4]);
+                      const unsigned char swizzle[4]);
 
 
 #endif /* !LP_BLD_SWIZZLE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h
index 912db24aecb..10c251c4162 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h
@@ -44,14 +44,30 @@ struct lp_build_context;
 struct lp_build_mask_context;
 
 
-typedef void
-(*lp_emit_fetch_texel_soa_callback)( LLVMBuilderRef builder,
-                                void *context,
-                                unsigned unit,
-                                unsigned num_coords,
-                                const LLVMValueRef *coords,
-                                LLVMValueRef lodbias,
-                                LLVMValueRef *texel);
+/**
+ * Sampler code generation interface.
+ *
+ * Although texture sampling is a requirement for TGSI translation, it is
+ * a very different problem with several different approaches to it. This
+ * structure establishes an interface for texture sampling code generation, so
+ * that we can easily use different texture sampling strategies.
+ */
+struct lp_build_sampler_soa
+{
+   void
+   (*destroy)( struct lp_build_sampler_soa *sampler );
+
+   void
+   (*emit_fetch_texel)( struct lp_build_sampler_soa *sampler,
+                        LLVMBuilderRef builder,
+                        union lp_type type,
+                        unsigned unit,
+                        unsigned num_coords,
+                        const LLVMValueRef *coords,
+                        LLVMValueRef lodbias,
+                        LLVMValueRef *texel);
+};
+
 
 void
 lp_build_tgsi_soa(LLVMBuilderRef builder,
@@ -62,8 +78,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
                   const LLVMValueRef *pos,
                   const LLVMValueRef (*inputs)[4],
                   LLVMValueRef (*outputs)[4],
-                  lp_emit_fetch_texel_soa_callback emit_fetch_texel,
-                  void *emit_fetch_texel_context);
+                  struct lp_build_sampler_soa *sampler);
 
 
 #endif /* LP_BLD_TGSI_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
index d4d18febec7..b106ce2317f 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c
@@ -78,6 +78,11 @@
 #define CHAN_Z 2
 #define CHAN_W 3
 
+#define QUAD_TOP_LEFT     0
+#define QUAD_TOP_RIGHT    1
+#define QUAD_BOTTOM_LEFT  2
+#define QUAD_BOTTOM_RIGHT 3
+
 
 struct lp_build_tgsi_soa_context
 {
@@ -88,8 +93,7 @@ struct lp_build_tgsi_soa_context
    const LLVMValueRef (*inputs)[NUM_CHANNELS];
    LLVMValueRef (*outputs)[NUM_CHANNELS];
 
-   lp_emit_fetch_texel_soa_callback emit_fetch_texel;
-   void *emit_fetch_texel_context;
+   struct lp_build_sampler_soa *sampler;
 
    LLVMValueRef immediates[LP_MAX_IMMEDIATES][NUM_CHANNELS];
    LLVMValueRef temps[LP_MAX_TEMPS][NUM_CHANNELS];
@@ -98,6 +102,51 @@ struct lp_build_tgsi_soa_context
 };
 
 
+static const unsigned char
+swizzle_left[4] = {
+   QUAD_TOP_LEFT,     QUAD_TOP_LEFT,
+   QUAD_BOTTOM_LEFT,  QUAD_BOTTOM_LEFT
+};
+
+static const unsigned char
+swizzle_right[4] = {
+   QUAD_TOP_RIGHT,    QUAD_TOP_RIGHT,
+   QUAD_BOTTOM_RIGHT, QUAD_BOTTOM_RIGHT
+};
+
+static const unsigned char
+swizzle_top[4] = {
+   QUAD_TOP_LEFT,     QUAD_TOP_RIGHT,
+   QUAD_TOP_LEFT,     QUAD_TOP_RIGHT
+};
+
+static const unsigned char
+swizzle_bottom[4] = {
+   QUAD_BOTTOM_LEFT,  QUAD_BOTTOM_RIGHT,
+   QUAD_BOTTOM_LEFT,  QUAD_BOTTOM_RIGHT
+};
+
+
+static LLVMValueRef
+emit_ddx(struct lp_build_tgsi_soa_context *bld,
+         LLVMValueRef src)
+{
+   LLVMValueRef src_left  = lp_build_swizzle1_aos(&bld->base, src, swizzle_left);
+   LLVMValueRef src_right = lp_build_swizzle1_aos(&bld->base, src, swizzle_right);
+   return lp_build_sub(&bld->base, src_right, src_left);
+}
+
+
+static LLVMValueRef
+emit_ddy(struct lp_build_tgsi_soa_context *bld,
+         LLVMValueRef src)
+{
+   LLVMValueRef src_top    = lp_build_swizzle1_aos(&bld->base, src, swizzle_top);
+   LLVMValueRef src_bottom = lp_build_swizzle1_aos(&bld->base, src, swizzle_bottom);
+   return lp_build_sub(&bld->base, src_top, src_bottom);
+}
+
+
 /**
  * Register fetch.
  */
@@ -168,6 +217,7 @@ emit_fetch(
       break;
 
    case TGSI_UTIL_SIGN_SET:
+      /* TODO: Use bitwese OR for floating point */
       res = lp_build_abs( &bld->base, res );
       res = LLVMBuildNeg( bld->base.builder, res, "" );
       break;
@@ -185,6 +235,36 @@ emit_fetch(
 
 
 /**
+ * Register fetch with derivatives.
+ */
+static void
+emit_fetch_deriv(
+   struct lp_build_tgsi_soa_context *bld,
+   const struct tgsi_full_instruction *inst,
+   unsigned index,
+   const unsigned chan_index,
+   LLVMValueRef *res,
+   LLVMValueRef *ddx,
+   LLVMValueRef *ddy)
+{
+   LLVMValueRef src;
+
+   src = emit_fetch(bld, inst, index, chan_index);
+
+   if(res)
+      *res = src;
+
+   /* TODO: use interpolation coeffs for inputs */
+
+   if(ddx)
+      *ddx = emit_ddx(bld, src);
+
+   if(ddy)
+      *ddy = emit_ddy(bld, src);
+}
+
+
+/**
  * Register store.
  */
 static void
@@ -239,17 +319,18 @@ emit_store(
  * High-level instruction translators.
  */
 
+
 static void
 emit_tex( struct lp_build_tgsi_soa_context *bld,
           const struct tgsi_full_instruction *inst,
           boolean apply_lodbias,
-          boolean projected)
+          boolean projected,
+          LLVMValueRef *texel)
 {
    const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
    LLVMValueRef lodbias;
    LLVMValueRef oow;
    LLVMValueRef coords[3];
-   LLVMValueRef texel[4];
    unsigned num_coords;
    unsigned i;
 
@@ -289,12 +370,11 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
          coords[i] = lp_build_mul(&bld->base, coords[i], oow);
    }
 
-   bld->emit_fetch_texel(bld->base.builder, bld->emit_fetch_texel_context,
-                         unit, num_coords, coords, lodbias, texel);
-
-   FOR_EACH_DST0_ENABLED_CHANNEL( inst, i ) {
-      emit_store( bld, inst, 0, i, texel[i] );
-   }
+   bld->sampler->emit_fetch_texel(bld->sampler,
+                                  bld->base.builder,
+                                  bld->base.type,
+                                  unit, num_coords, coords, lodbias,
+                                  texel);
 }
 
 
@@ -347,14 +427,6 @@ emit_kil(
 }
 
 
-static void
-emit_kilp(
-   struct lp_build_tgsi_soa_context *bld )
-{
-   /* XXX todo / fix me */
-}
-
-
 /**
  * Check if inst src/dest regs use indirect addressing into temporary
  * register file.
@@ -382,25 +454,35 @@ indirect_temp_reference(const struct tgsi_full_instruction *inst)
 static int
 emit_instruction(
    struct lp_build_tgsi_soa_context *bld,
-   struct tgsi_full_instruction *inst )
+   const struct tgsi_full_instruction *inst,
+   const struct tgsi_opcode_info *info)
 {
    unsigned chan_index;
    LLVMValueRef src0, src1, src2;
    LLVMValueRef tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
-   LLVMValueRef dst0;
+   LLVMValueRef res;
+   LLVMValueRef dst0[NUM_CHANNELS];
 
    /* we can't handle indirect addressing into temp register file yet */
    if (indirect_temp_reference(inst))
       return FALSE;
 
+   assert(info->num_dst <= 1);
+   if(info->num_dst) {
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         dst0[chan_index] = bld->base.undef;
+      }
+   }
+
    switch (inst->Instruction.Opcode) {
 #if 0
    case TGSI_OPCODE_ARL:
+      /* FIXME */
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
          emit_flr(bld, 0, 0);
          emit_f2it( bld, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 #endif
@@ -408,19 +490,17 @@ emit_instruction(
    case TGSI_OPCODE_MOV:
    case TGSI_OPCODE_SWZ:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = emit_fetch( bld, inst, 0, chan_index );
       }
       break;
 
    case TGSI_OPCODE_LIT:
       if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ) {
-         emit_store( bld, inst, 0, CHAN_X, bld->base.one);
+         dst0[CHAN_X] = bld->base.one;
       }
       if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ) {
          src0 = emit_fetch( bld, inst, 0, CHAN_X );
-         dst0 = lp_build_max( &bld->base, src0, bld->base.zero);
-         emit_store( bld, inst, 0, CHAN_Y, dst0);
+         dst0[CHAN_Y] = lp_build_max( &bld->base, src0, bld->base.zero);
       }
       if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) {
          /* XMM[1] = SrcReg[0].yyyy */
@@ -432,20 +512,19 @@ emit_instruction(
          tmp1 = lp_build_pow( &bld->base, tmp1, tmp2);
          tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
          tmp2 = lp_build_cmp(&bld->base, PIPE_FUNC_GREATER, tmp0, bld->base.zero);
-         dst0 = lp_build_select(&bld->base, tmp2, tmp1, bld->base.zero);
-         emit_store( bld, inst, 0, CHAN_Z, dst0);
+         dst0[CHAN_Z] = lp_build_select(&bld->base, tmp2, tmp1, bld->base.zero);
       }
       if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) ) {
-         emit_store( bld, inst, 0, CHAN_W, bld->base.one);
+         dst0[CHAN_W] = bld->base.one;
       }
       break;
 
    case TGSI_OPCODE_RCP:
    /* TGSI_OPCODE_RECIP */
       src0 = emit_fetch( bld, inst, 0, CHAN_X );
-      dst0 = lp_build_rcp(&bld->base, src0);
+      res = lp_build_rcp(&bld->base, src0);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, dst0 );
+         dst0[chan_index] = res;
       }
       break;
 
@@ -453,9 +532,9 @@ emit_instruction(
    /* TGSI_OPCODE_RECIPSQRT */
       src0 = emit_fetch( bld, inst, 0, CHAN_X );
       src0 = lp_build_abs(&bld->base, src0);
-      dst0 = lp_build_rsqrt(&bld->base, src0);
+      res = lp_build_rsqrt(&bld->base, src0);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, dst0 );
+         dst0[chan_index] = res;
       }
       break;
 
@@ -479,16 +558,15 @@ emit_instruction(
          lp_build_exp2_approx(&bld->base, src0, p_exp2_int_part, p_frac_part, p_exp2);
 
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
-            emit_store( bld, inst, 0, CHAN_X, tmp0);
+            dst0[CHAN_X] = tmp0;
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ))
-            emit_store( bld, inst, 0, CHAN_Y, tmp1);
+            dst0[CHAN_Y] = tmp1;
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
-            emit_store( bld, inst, 0, CHAN_Z, tmp2);
+            dst0[CHAN_Z] = tmp2;
       }
       /* dst.w = 1.0 */
       if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) {
-         tmp0 = bld->base.one;
-         emit_store( bld, inst, 0, CHAN_W, tmp0);
+         dst0[CHAN_W] = bld->base.one;
       }
       break;
 
@@ -514,20 +592,18 @@ emit_instruction(
 
          /* dst.x = floor(lg2(abs(src.x))) */
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
-            emit_store( bld, inst, 0, CHAN_X, tmp0);
+            dst0[CHAN_X] = tmp0;
          /* dst.y = abs(src)/ex2(floor(lg2(abs(src.x)))) */
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) {
-            tmp1 = lp_build_div( &bld->base, src0, tmp1);
-            emit_store( bld, inst, 0, CHAN_Y, tmp1);
+            dst0[CHAN_Y] = lp_build_div( &bld->base, src0, tmp1);
          }
          /* dst.z = lg2(abs(src.x)) */
          if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
-            emit_store( bld, inst, 0, CHAN_Z, tmp2);
+            dst0[CHAN_Z] = tmp2;
       }
       /* dst.w = 1.0 */
       if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) {
-         tmp0 = bld->base.one;
-         emit_store( bld, inst, 0, CHAN_W, tmp0);
+         dst0[CHAN_W] = bld->base.one;
       }
       break;
 
@@ -535,8 +611,7 @@ emit_instruction(
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
-         dst0 = lp_build_mul(&bld->base, src0, src1);
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_mul(&bld->base, src0, src1);
       }
       break;
 
@@ -544,8 +619,7 @@ emit_instruction(
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
-         dst0 = lp_build_add(&bld->base, src0, src1);
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_add(&bld->base, src0, src1);
       }
       break;
 
@@ -563,7 +637,7 @@ emit_instruction(
       tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
       tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
@@ -585,28 +659,24 @@ emit_instruction(
       tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
       tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
    case TGSI_OPCODE_DST:
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) {
-         tmp0 = bld->base.one;
-         emit_store( bld, inst, 0, CHAN_X, tmp0);
+         dst0[CHAN_X] = bld->base.one;
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) {
          tmp0 = emit_fetch( bld, inst, 0, CHAN_Y );
          tmp1 = emit_fetch( bld, inst, 1, CHAN_Y );
-         tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
-         emit_store( bld, inst, 0, CHAN_Y, tmp0);
+         dst0[CHAN_Y] = lp_build_mul( &bld->base, tmp0, tmp1);
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
-         tmp0 = emit_fetch( bld, inst, 0, CHAN_Z );
-         emit_store( bld, inst, 0, CHAN_Z, tmp0);
+         dst0[CHAN_Z] = emit_fetch( bld, inst, 0, CHAN_Z );
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
-         tmp0 = emit_fetch( bld, inst, 1, CHAN_W );
-         emit_store( bld, inst, 0, CHAN_W, tmp0);
+         dst0[CHAN_W] = emit_fetch( bld, inst, 1, CHAN_W );
       }
       break;
 
@@ -614,8 +684,7 @@ emit_instruction(
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
-         dst0 = lp_build_min( &bld->base, src0, src1 );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_min( &bld->base, src0, src1 );
       }
       break;
 
@@ -623,8 +692,7 @@ emit_instruction(
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
-         dst0 = lp_build_max( &bld->base, src0, src1 );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_max( &bld->base, src0, src1 );
       }
       break;
 
@@ -634,8 +702,7 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
@@ -645,8 +712,7 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GEQUAL, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
@@ -658,7 +724,7 @@ emit_instruction(
          tmp2 = emit_fetch( bld, inst, 2, chan_index );
          tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
          tmp0 = lp_build_add( &bld->base, tmp0, tmp2);
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
@@ -666,8 +732,7 @@ emit_instruction(
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
          tmp1 = emit_fetch( bld, inst, 1, chan_index );
-         tmp0 = lp_build_sub( &bld->base, tmp0, tmp1);
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_sub( &bld->base, tmp0, tmp1);
       }
       break;
 
@@ -678,13 +743,19 @@ emit_instruction(
          src2 = emit_fetch( bld, inst, 2, chan_index );
          tmp0 = lp_build_sub( &bld->base, src1, src2 );
          tmp0 = lp_build_mul( &bld->base, src0, tmp0 );
-         dst0 = lp_build_add( &bld->base, tmp0, src2 );
-         emit_store( bld, inst, 0, chan_index, dst0 );
+         dst0[chan_index] = lp_build_add( &bld->base, tmp0, src2 );
       }
       break;
 
    case TGSI_OPCODE_CND:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         src0 = emit_fetch( bld, inst, 0, chan_index );
+         src1 = emit_fetch( bld, inst, 1, chan_index );
+         src2 = emit_fetch( bld, inst, 2, chan_index );
+         tmp1 = lp_build_const_scalar(bld->base.type, 0.5);
+         tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src2, tmp1);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, src0, src1 );
+      }
       break;
 
    case TGSI_OPCODE_DP2A:
@@ -698,45 +769,49 @@ emit_instruction(
       tmp1 = emit_fetch( bld, inst, 2, CHAN_X );  /* xmm1 = src[2].x */
       tmp0 = lp_build_add( &bld->base, tmp0, tmp1);              /* xmm0 = xmm0 + xmm1 */
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);  /* dest[ch] = xmm0 */
+         dst0[chan_index] = tmp0;  /* dest[ch] = xmm0 */
       }
       break;
 
-#if 0
    case TGSI_OPCODE_FRC:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_frc( bld, 0, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         src0 = emit_fetch( bld, inst, 0, chan_index );
+         tmp0 = lp_build_floor(&bld->base, src0);
+         tmp0 = lp_build_sub(&bld->base, tmp0, src0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
    case TGSI_OPCODE_CLAMP:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         tmp0 = emit_fetch( bld, inst, 0, chan_index );
+         src1 = emit_fetch( bld, inst, 1, chan_index );
+         src2 = emit_fetch( bld, inst, 2, chan_index );
+         tmp0 = lp_build_max(&bld->base, tmp0, src1);
+         tmp0 = lp_build_min(&bld->base, tmp0, src2);
+         dst0[chan_index] = tmp0;
+      }
       break;
 
    case TGSI_OPCODE_FLR:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_flr( bld, 0, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_floor(&bld->base, tmp0);
       }
       break;
 
    case TGSI_OPCODE_ROUND:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_rnd( bld, 0, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_round(&bld->base, tmp0);
       }
       break;
-#endif
 
    case TGSI_OPCODE_EX2: {
       tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
       tmp0 = lp_build_exp2( &bld->base, tmp0);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
    }
@@ -745,16 +820,16 @@ emit_instruction(
       tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
       tmp0 = lp_build_log2( &bld->base, tmp0);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
    case TGSI_OPCODE_POW:
       src0 = emit_fetch( bld, inst, 0, CHAN_X );
       src1 = emit_fetch( bld, inst, 1, CHAN_X );
-      dst0 = lp_build_pow( &bld->base, src0, src1 );
+      res = lp_build_pow( &bld->base, src0, src1 );
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, dst0 );
+         dst0[chan_index] = res;
       }
       break;
 
@@ -775,7 +850,7 @@ emit_instruction(
          tmp5 = tmp3;
          tmp5 = lp_build_mul( &bld->base, tmp5, tmp4);
          tmp2 = lp_build_sub( &bld->base, tmp2, tmp5);
-         emit_store( bld, inst, 0, CHAN_X, tmp2);
+         dst0[CHAN_X] = tmp2;
       }
       if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ||
           IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) {
@@ -786,31 +861,30 @@ emit_instruction(
          tmp3 = lp_build_mul( &bld->base, tmp3, tmp2);
          tmp1 = lp_build_mul( &bld->base, tmp1, tmp5);
          tmp3 = lp_build_sub( &bld->base, tmp3, tmp1);
-         emit_store( bld, inst, 0, CHAN_Y, tmp3);
+         dst0[CHAN_Y] = tmp3;
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
          tmp5 = lp_build_mul( &bld->base, tmp5, tmp4);
          tmp0 = lp_build_mul( &bld->base, tmp0, tmp2);
          tmp5 = lp_build_sub( &bld->base, tmp5, tmp0);
-         emit_store( bld, inst, 0, CHAN_Z, tmp5);
+         dst0[CHAN_Z] = tmp5;
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
-         tmp0 = bld->base.one;
-         emit_store( bld, inst, 0, CHAN_W, tmp0);
+         dst0[CHAN_W] = bld->base.one;
       }
       break;
 
    case TGSI_OPCODE_ABS:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         tmp0 = lp_build_abs( &bld->base, tmp0 ) ;
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_abs( &bld->base, tmp0 );
       }
       break;
 
    case TGSI_OPCODE_RCC:
+      /* deprecated? */
+      assert(0);
       return 0;
-      break;
 
    case TGSI_OPCODE_DPH:
       tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
@@ -827,7 +901,7 @@ emit_instruction(
       tmp1 = emit_fetch( bld, inst, 1, CHAN_W );
       tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
@@ -835,25 +909,27 @@ emit_instruction(
       tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
       tmp0 = lp_build_cos( &bld->base, tmp0 );
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
    case TGSI_OPCODE_DDX:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         emit_fetch_deriv( bld, inst, 0, chan_index, NULL, &dst0[chan_index], NULL);
+      }
       break;
 
    case TGSI_OPCODE_DDY:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         emit_fetch_deriv( bld, inst, 0, chan_index, NULL, NULL, &dst0[chan_index]);
+      }
       break;
 
-#if 0
    case TGSI_OPCODE_KILP:
       /* predicated kill */
-      emit_kilp( bld );
-      return 0; /* XXX fix me */
+      /* FIXME */
+      return 0;
       break;
-#endif
 
    case TGSI_OPCODE_KIL:
       /* conditional kill */
@@ -885,13 +961,14 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_EQUAL, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
    case TGSI_OPCODE_SFL:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         dst0[chan_index] = bld->base.zero;
+      }
       break;
 
    case TGSI_OPCODE_SGT:
@@ -899,8 +976,7 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
@@ -908,7 +984,7 @@ emit_instruction(
       tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
       tmp0 = lp_build_sin( &bld->base, tmp0 );
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 
@@ -917,8 +993,7 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LEQUAL, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
@@ -927,85 +1002,99 @@ emit_instruction(
          src0 = emit_fetch( bld, inst, 0, chan_index );
          src1 = emit_fetch( bld, inst, 1, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_NOTEQUAL, src0, src1 );
-         dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
       }
       break;
 
    case TGSI_OPCODE_STR:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         dst0[chan_index] = bld->base.one;
+      }
       break;
 
    case TGSI_OPCODE_TEX:
-      emit_tex( bld, inst, FALSE, FALSE );
+      emit_tex( bld, inst, FALSE, FALSE, dst0 );
       break;
 
    case TGSI_OPCODE_TXD:
+      /* FIXME */
       return 0;
       break;
 
    case TGSI_OPCODE_UP2H:
+      /* deprecated */
+      assert (0);
       return 0;
       break;
 
    case TGSI_OPCODE_UP2US:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_UP4B:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_UP4UB:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_X2D:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_ARA:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
 #if 0
    case TGSI_OPCODE_ARR:
+      /* FIXME */
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
          emit_rnd( bld, 0, 0 );
          emit_f2it( bld, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = tmp0;
       }
       break;
 #endif
 
    case TGSI_OPCODE_BRA:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_CAL:
+      /* FIXME */
       return 0;
       break;
 
-#if 0
    case TGSI_OPCODE_RET:
-      emit_ret( bld );
+      /* FIXME */
+      return 0;
       break;
-#endif
 
    case TGSI_OPCODE_END:
       break;
 
-#if 0
    case TGSI_OPCODE_SSG:
    /* TGSI_OPCODE_SGN */
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_sgn( bld, 0, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_sgn( &bld->base, tmp0 );
       }
       break;
-#endif
 
    case TGSI_OPCODE_CMP:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
@@ -1013,34 +1102,29 @@ emit_instruction(
          src1 = emit_fetch( bld, inst, 1, chan_index );
          src2 = emit_fetch( bld, inst, 2, chan_index );
          tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, bld->base.zero );
-         dst0 = lp_build_select( &bld->base, tmp0, src1, src2);
-         emit_store( bld, inst, 0, chan_index, dst0);
+         dst0[chan_index] = lp_build_select( &bld->base, tmp0, src1, src2);
       }
       break;
 
    case TGSI_OPCODE_SCS:
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) {
          tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
-         tmp0 = lp_build_cos( &bld->base, tmp0 );
-         emit_store( bld, inst, 0, CHAN_X, tmp0);
+         dst0[CHAN_X] = lp_build_cos( &bld->base, tmp0 );
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) {
          tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
-         tmp0 = lp_build_sin( &bld->base, tmp0 );
-         emit_store( bld, inst, 0, CHAN_Y, tmp0);
+         dst0[CHAN_Y] = lp_build_sin( &bld->base, tmp0 );
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
-         tmp0 = bld->base.zero;
-         emit_store( bld, inst, 0, CHAN_Z, tmp0);
+         dst0[CHAN_Z] = bld->base.zero;
       }
       IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
-         tmp0 = bld->base.one;
-         emit_store( bld, inst, 0, CHAN_W, tmp0);
+         dst0[CHAN_W] = bld->base.one;
       }
       break;
 
    case TGSI_OPCODE_TXB:
-      emit_tex( bld, inst, TRUE, FALSE );
+      emit_tex( bld, inst, TRUE, FALSE, dst0 );
       break;
 
    case TGSI_OPCODE_NRM:
@@ -1099,38 +1183,35 @@ emit_instruction(
 
             /* dst.x = xmm1 * src.x */
             if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X)) {
-               tmp4 = lp_build_mul( &bld->base, tmp4, tmp1);
-               emit_store(bld, inst, 0, CHAN_X, tmp4);
+               dst0[CHAN_X] = lp_build_mul( &bld->base, tmp4, tmp1);
             }
 
             /* dst.y = xmm1 * src.y */
             if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y)) {
-               tmp5 = lp_build_mul( &bld->base, tmp5, tmp1);
-               emit_store(bld, inst, 0, CHAN_Y, tmp5);
+               dst0[CHAN_Y] = lp_build_mul( &bld->base, tmp5, tmp1);
             }
 
             /* dst.z = xmm1 * src.z */
             if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z)) {
-               tmp6 = lp_build_mul( &bld->base, tmp6, tmp1);
-               emit_store(bld, inst, 0, CHAN_Z, tmp6);
+               dst0[CHAN_Z] = lp_build_mul( &bld->base, tmp6, tmp1);
             }
 
             /* dst.w = xmm1 * src.w */
             if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X) && dims == 4) {
-               tmp7 = lp_build_mul( &bld->base, tmp7, tmp1);
-               emit_store(bld, inst, 0, CHAN_W, tmp7);
+               dst0[CHAN_W] = lp_build_mul( &bld->base, tmp7, tmp1);
             }
          }
 
-         /* dst0.w = 1.0 */
+         /* dst.w = 1.0 */
          if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W) && dims == 3) {
-            tmp0 = bld->base.one;
-            emit_store(bld, inst, 0, CHAN_W, tmp0);
+            dst0[CHAN_W] = bld->base.one;
          }
       }
       break;
 
    case TGSI_OPCODE_DIV:
+      /* deprecated */
+      assert( 0 );
       return 0;
       break;
 
@@ -1143,118 +1224,157 @@ emit_instruction(
       tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);              /* xmm1 = xmm1 * xmm2 */
       tmp0 = lp_build_add( &bld->base, tmp0, tmp1);              /* xmm0 = xmm0 + xmm1 */
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
-         emit_store( bld, inst, 0, chan_index, tmp0);  /* dest[ch] = xmm0 */
+         dst0[chan_index] = tmp0;  /* dest[ch] = xmm0 */
       }
       break;
 
    case TGSI_OPCODE_TXL:
-      emit_tex( bld, inst, TRUE, FALSE );
+      emit_tex( bld, inst, TRUE, FALSE, dst0 );
       break;
 
    case TGSI_OPCODE_TXP:
-      emit_tex( bld, inst, FALSE, TRUE );
+      emit_tex( bld, inst, FALSE, TRUE, dst0 );
       break;
       
    case TGSI_OPCODE_BRK:
+      /* FIXME */
       return 0;
       break;
 
    case TGSI_OPCODE_IF:
+      /* FIXME */
       return 0;
       break;
 
    case TGSI_OPCODE_BGNFOR:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_REP:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_ELSE:
+      /* FIXME */
       return 0;
       break;
 
    case TGSI_OPCODE_ENDIF:
+      /* FIXME */
       return 0;
       break;
 
    case TGSI_OPCODE_ENDFOR:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_ENDREP:
+      /* deprecated */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_PUSHA:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_POPA:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_CEIL:
-      return 0;
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         tmp0 = emit_fetch( bld, inst, 0, chan_index );
+         dst0[chan_index] = lp_build_ceil(&bld->base, tmp0);
+      }
       break;
 
    case TGSI_OPCODE_I2F:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_NOT:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
-#if 0
    case TGSI_OPCODE_TRUNC:
       FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
          tmp0 = emit_fetch( bld, inst, 0, chan_index );
-         emit_f2it( bld, 0 );
-         emit_i2f( bld, 0 );
-         emit_store( bld, inst, 0, chan_index, tmp0);
+         dst0[chan_index] = lp_build_trunc(&bld->base, tmp0);
       }
       break;
-#endif
 
    case TGSI_OPCODE_SHL:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_SHR:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_AND:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_OR:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_MOD:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_XOR:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_SAD:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_TXF:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_TXQ:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
    case TGSI_OPCODE_CONT:
+      /* deprecated? */
+      assert(0);
       return 0;
       break;
 
@@ -1266,10 +1386,28 @@ emit_instruction(
       return 0;
       break;
 
+   case TGSI_OPCODE_NOISE1:
+   case TGSI_OPCODE_NOISE2:
+   case TGSI_OPCODE_NOISE3:
+   case TGSI_OPCODE_NOISE4:
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         dst0[chan_index] = bld->base.zero;
+      }
+      break;
+
+   case TGSI_OPCODE_NOP:
+      break;
+
    default:
       return 0;
    }
    
+   if(info->num_dst) {
+      FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+         emit_store( bld, inst, 0, chan_index, dst0[chan_index]);
+      }
+   }
+
    return 1;
 }
 
@@ -1283,8 +1421,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
                   const LLVMValueRef *pos,
                   const LLVMValueRef (*inputs)[NUM_CHANNELS],
                   LLVMValueRef (*outputs)[NUM_CHANNELS],
-                  lp_emit_fetch_texel_soa_callback emit_fetch_texel,
-                  void *emit_fetch_texel_context)
+                  struct lp_build_sampler_soa *sampler)
 {
    struct lp_build_tgsi_soa_context bld;
    struct tgsi_parse_context parse;
@@ -1299,8 +1436,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
    bld.inputs = inputs;
    bld.outputs = outputs;
    bld.consts_ptr = consts_ptr;
-   bld.emit_fetch_texel = emit_fetch_texel;
-   bld.emit_fetch_texel_context = emit_fetch_texel_context;
+   bld.sampler = sampler;
 
    tgsi_parse_init( &parse, tokens );
 
@@ -1309,16 +1445,18 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
 
       switch( parse.FullToken.Token.Type ) {
       case TGSI_TOKEN_TYPE_DECLARATION:
-         /* Input already interpolated */
+         /* Inputs already interpolated */
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
-         if (!emit_instruction( &bld, &parse.FullToken.FullInstruction )) {
+         {
             unsigned opcode = parse.FullToken.FullInstruction.Instruction.Opcode;
             const struct tgsi_opcode_info *info = tgsi_get_opcode_info(opcode);
-	    _debug_printf("warning: failed to translate tgsi opcode %s to LLVM\n",
-	                  info ? info->mnemonic : "<invalid>");
-	 }
+            if (!emit_instruction( &bld, &parse.FullToken.FullInstruction, info ))
+               _debug_printf("warning: failed to translate tgsi opcode %s to LLVM\n",
+                             info ? info->mnemonic : "<invalid>");
+         }
+
          break;
 
       case TGSI_TOKEN_TYPE_IMMEDIATE:
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.c b/src/gallium/drivers/llvmpipe/lp_bld_type.c
index 8e0026fd973..577644b7ab8 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_type.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_type.c
@@ -157,6 +157,17 @@ lp_build_int_vec_type(union lp_type type)
 }
 
 
+union lp_type
+lp_int_type(union lp_type type)
+{
+   union lp_type int_type;
+   int_type.value = 0;
+   int_type.width = type.width;
+   int_type.length = type.length;
+   return int_type;
+}
+
+
 void
 lp_build_context_init(struct lp_build_context *bld,
                       LLVMBuilderRef builder,
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.h b/src/gallium/drivers/llvmpipe/lp_bld_type.h
index 3ce566be641..9933e0b45c3 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_type.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_type.h
@@ -165,6 +165,10 @@ LLVMTypeRef
 lp_build_int_vec_type(union lp_type type);
 
 
+union lp_type
+lp_int_type(union lp_type type);
+
+
 void
 lp_build_context_init(struct lp_build_context *bld,
                       LLVMBuilderRef builder,
diff --git a/src/gallium/drivers/llvmpipe/lp_clear.c b/src/gallium/drivers/llvmpipe/lp_clear.c
index 580cca5b463..bdcff94b9bf 100644
--- a/src/gallium/drivers/llvmpipe/lp_clear.c
+++ b/src/gallium/drivers/llvmpipe/lp_clear.c
@@ -67,6 +67,7 @@ llvmpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
          util_pack_color(rgba, ps->format, &cv);
          lp_tile_cache_clear(llvmpipe->cbuf_cache[i], rgba, cv);
       }
+      llvmpipe->dirty_render_cache = TRUE;
    }
 
    if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 233d1df0e10..a4b2bd8c2ad 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -141,8 +141,6 @@ llvmpipe_is_texture_referenced( struct pipe_context *pipe,
          return PIPE_REFERENCED_FOR_WRITE;
    }
    
-   /* FIXME: we also need to do the same for the texture cache */
-   
    return PIPE_UNREFERENCED;
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index d288460a1b8..9465f763d50 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -44,15 +44,47 @@
 static void
 lp_jit_init_globals(struct llvmpipe_screen *screen)
 {
-   /* struct lp_jit_context */
+   LLVMTypeRef texture_type;
+
+   /* struct lp_jit_texture */
    {
       LLVMTypeRef elem_types[4];
+
+      elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_STRIDE] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_DATA]   = LLVMPointerType(LLVMInt8Type(), 0);
+
+      texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
+
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, width,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_WIDTH);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_HEIGHT);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, stride,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_STRIDE);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, data,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_DATA);
+      LP_CHECK_STRUCT_SIZE(struct lp_jit_texture,
+                           screen->target, texture_type);
+
+      LLVMAddTypeName(screen->module, "texture", texture_type);
+   }
+
+   /* struct lp_jit_context */
+   {
+      LLVMTypeRef elem_types[5];
       LLVMTypeRef context_type;
 
       elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* constants */
       elem_types[1] = LLVMPointerType(LLVMInt8Type(), 0);  /* samplers */
       elem_types[2] = LLVMFloatType();                     /* alpha_ref_value */
       elem_types[3] = LLVMPointerType(LLVMInt8Type(), 0);  /* blend_color */
+      elem_types[4] = LLVMArrayType(texture_type, PIPE_MAX_SAMPLERS); /* textures */
 
       context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
@@ -64,6 +96,9 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
                              screen->target, context_type, 2);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, blend_color,
                              screen->target, context_type, 3);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_context, textures,
+                             screen->target, context_type,
+                             LP_JIT_CONTEXT_TEXTURES_INDEX);
       LP_CHECK_STRUCT_SIZE(struct lp_jit_context,
                            screen->target, context_type);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index a7fb60f9f5c..58f716ede29 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -38,11 +38,31 @@
 
 #include "lp_bld_struct.h"
 
+#include "pipe/p_state.h"
+
 
 struct tgsi_sampler;
 struct llvmpipe_screen;
 
 
+struct lp_jit_texture
+{
+   uint32_t width;
+   uint32_t height;
+   uint32_t stride;
+   const void *data;
+};
+
+
+enum {
+   LP_JIT_TEXTURE_WIDTH = 0,
+   LP_JIT_TEXTURE_HEIGHT,
+   LP_JIT_TEXTURE_STRIDE,
+   LP_JIT_TEXTURE_DATA
+};
+
+
+
 /**
  * This structure is passed directly to the generated fragment shader.
  *
@@ -60,11 +80,12 @@ struct lp_jit_context
 
    struct tgsi_sampler **samplers;
 
-   /* TODO: alpha reference value */
    float alpha_ref_value;
 
-   /* TODO: blend constant color */
+   /* FIXME: store (also?) in floats */
    uint8_t *blend_color;
+
+   struct lp_jit_texture textures[PIPE_MAX_SAMPLERS];
 };
 
 
@@ -80,6 +101,11 @@ struct lp_jit_context
 #define lp_jit_context_blend_color(_builder, _ptr) \
    lp_build_struct_get(_builder, _ptr, 3, "blend_color")
 
+#define LP_JIT_CONTEXT_TEXTURES_INDEX 4
+
+#define lp_jit_context_textures(_builder, _ptr) \
+   lp_build_struct_get_ptr(_builder, _ptr, LP_JIT_CONTEXT_TEXTURES_INDEX, "textures")
+
 
 typedef void
 (*lp_jit_frag_func)(struct lp_jit_context *context,
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index 125035771e5..0ce1a37bd47 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -27,8 +27,6 @@
 
 
 #include "util/u_memory.h"
-#include "util/u_simple_screen.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
@@ -196,8 +194,7 @@ static void
 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-
-   struct pipe_winsys *winsys = _screen->winsys;
+   struct llvmpipe_winsys *winsys = screen->winsys;
 
    lp_jit_screen_cleanup(screen);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index d145f6d6bbc..2d2fc19a65f 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -44,6 +44,7 @@
 #include "pipe/p_thread.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "lp_bld_debug.h"
 #include "lp_tile_cache.h"
 #include "lp_tile_soa.h"
 
@@ -162,12 +163,12 @@ shade_quads(struct llvmpipe_context *llvmpipe,
    else
       depth = NULL;
 
-   /* TODO: blend color */
+   /* XXX: This will most likely fail on 32bit x86 without -mstackrealign */
+   assert(lp_check_alignment(mask, 16));
 
-   assert((((uintptr_t)mask) & 0xf) == 0);
-   assert((((uintptr_t)depth) & 0xf) == 0);
-   assert((((uintptr_t)color) & 0xf) == 0);
-   assert((((uintptr_t)llvmpipe->jit_context.blend_color) & 0xf) == 0);
+   assert(lp_check_alignment(depth, 16));
+   assert(lp_check_alignment(color, 16));
+   assert(lp_check_alignment(llvmpipe->jit_context.blend_color, 16));
 
    /* run shader */
    fs->current->jit_function( &llvmpipe->jit_context,
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index fb10329887d..7b26ce61a38 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -36,6 +36,7 @@
 #include "pipe/p_state.h"
 #include "tgsi/tgsi_scan.h"
 #include "lp_jit.h"
+#include "lp_bld_sample.h" /* for struct lp_sampler_static_state */
 
 
 #define LP_NEW_VIEWPORT      0x1
@@ -57,16 +58,20 @@
 
 struct tgsi_sampler;
 struct vertex_info;
-
+struct pipe_context;
+struct llvmpipe_context;
 
 struct lp_fragment_shader;
 
 
 struct lp_fragment_shader_variant_key
 {
+   enum pipe_format zsbuf_format;
    struct pipe_depth_state depth;
    struct pipe_alpha_state alpha;
    struct pipe_blend_state blend;
+
+   struct lp_sampler_static_state sampler[PIPE_MAX_SAMPLERS];
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c
index 6fbb057937e..e87976b9f36 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_derived.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c
@@ -250,7 +250,9 @@ void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe )
 
    if (llvmpipe->dirty & (LP_NEW_FS |
                           LP_NEW_BLEND |
-                          LP_NEW_DEPTH_STENCIL_ALPHA))
+                          LP_NEW_DEPTH_STENCIL_ALPHA |
+                          LP_NEW_SAMPLER |
+                          LP_NEW_TEXTURE))
       llvmpipe_update_fs( llvmpipe );
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 94170bd7161..618cf1ffb8d 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -85,6 +85,7 @@
 #include "lp_context.h"
 #include "lp_state.h"
 #include "lp_quad.h"
+#include "lp_tex_sample.h"
 
 
 static const unsigned char quad_offset_x[4] = {0, 1, 0, 1};
@@ -130,9 +131,8 @@ generate_pos0(LLVMBuilderRef builder,
  * Generate the depth test.
  */
 static void
-generate_depth(struct llvmpipe_context *lp,
-               LLVMBuilderRef builder,
-               const struct pipe_depth_state *state,
+generate_depth(LLVMBuilderRef builder,
+               const struct lp_fragment_shader_variant_key *key,
                union lp_type src_type,
                struct lp_build_mask_context *mask,
                LLVMValueRef src,
@@ -141,10 +141,10 @@ generate_depth(struct llvmpipe_context *lp,
    const struct util_format_description *format_desc;
    union lp_type dst_type;
 
-   if(!lp->framebuffer.zsbuf)
+   if(!key->depth.enabled)
       return;
 
-   format_desc = util_format_description(lp->framebuffer.zsbuf->format);
+   format_desc = util_format_description(key->zsbuf_format);
    assert(format_desc);
 
    /* Pick the depth type. */
@@ -164,7 +164,7 @@ generate_depth(struct llvmpipe_context *lp,
 #endif
 
    lp_build_depth_test(builder,
-                       state,
+                       &key->depth,
                        dst_type,
                        format_desc,
                        mask,
@@ -173,107 +173,6 @@ generate_depth(struct llvmpipe_context *lp,
 }
 
 
-struct build_fetch_texel_context
-{
-   LLVMValueRef context_ptr;
-
-   LLVMValueRef samplers_ptr;
-
-   /** Coords/texels store */
-   LLVMValueRef store_ptr;
-};
-
-
-void PIPE_CDECL
-lp_fetch_texel_soa( struct tgsi_sampler **samplers,
-                    uint32_t unit,
-                    float *store )
-{
-   struct tgsi_sampler *sampler = samplers[unit];
-
-#if 0
-   uint j;
-
-   debug_printf("%s sampler: %p (%p) store: %p\n",
-                __FUNCTION__,
-                sampler, *sampler,
-                store );
-
-   debug_printf("lodbias %f\n", store[12]);
-
-   for (j = 0; j < 4; j++)
-      debug_printf("sample %d texcoord %f %f\n",
-                   j,
-                   store[0+j],
-                   store[4+j]);
-#endif
-
-   {
-      float rgba[NUM_CHANNELS][QUAD_SIZE];
-      sampler->get_samples(sampler,
-                           &store[0],
-                           &store[4],
-                           &store[8],
-                           0.0f, /*store[12],  lodbias */
-                           rgba);
-      memcpy(store, rgba, sizeof rgba);
-   }
-
-#if 0
-   for (j = 0; j < 4; j++)
-      debug_printf("sample %d result %f %f %f %f\n",
-                   j,
-                   store[0+j],
-                   store[4+j],
-                   store[8+j],
-                   store[12+j]);
-#endif
-}
-
-
-static void
-emit_fetch_texel( LLVMBuilderRef builder,
-                  void *context,
-                  unsigned unit,
-                  unsigned num_coords,
-                  const LLVMValueRef *coords,
-                  LLVMValueRef lodbias,
-                  LLVMValueRef *texel)
-{
-   struct build_fetch_texel_context *bld = context;
-   LLVMTypeRef vec_type = LLVMTypeOf(coords[0]);
-   LLVMValueRef args[3];
-   unsigned i;
-
-   if(!bld->samplers_ptr)
-      bld->samplers_ptr = lp_jit_context_samplers(builder, bld->context_ptr);
-
-   if(!bld->store_ptr)
-      bld->store_ptr = LLVMBuildArrayAlloca(builder,
-                                            vec_type,
-                                            LLVMConstInt(LLVMInt32Type(), 4, 0),
-                                            "texel_store");
-
-   for (i = 0; i < num_coords; i++) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      LLVMValueRef coord_ptr = LLVMBuildGEP(builder, bld->store_ptr, &index, 1, "");
-      LLVMBuildStore(builder, coords[i], coord_ptr);
-   }
-
-   args[0] = bld->samplers_ptr;
-   args[1] = LLVMConstInt(LLVMInt32Type(), unit, 0);
-   args[2] = bld->store_ptr;
-
-   lp_build_intrinsic(builder, "fetch_texel", LLVMVoidType(), args, 3);
-
-   for (i = 0; i < NUM_CHANNELS; ++i) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
-      LLVMValueRef texel_ptr = LLVMBuildGEP(builder, bld->store_ptr, &index, 1, "");
-      texel[i] = LLVMBuildLoad(builder, texel_ptr, "");
-   }
-}
-
-
 /**
  * Generate the fragment shader, depth/stencil test, and alpha tests.
  */
@@ -286,7 +185,7 @@ generate_fs(struct llvmpipe_context *lp,
             LLVMValueRef context_ptr,
             unsigned i,
             const struct lp_build_interp_soa_context *interp,
-            struct build_fetch_texel_context *sampler,
+            struct lp_build_sampler_soa *sampler,
             LLVMValueRef *pmask,
             LLVMValueRef *color,
             LLVMValueRef depth_ptr)
@@ -298,6 +197,7 @@ generate_fs(struct llvmpipe_context *lp,
    LLVMValueRef consts_ptr;
    LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
    LLVMValueRef z = interp->pos[2];
+   struct lp_build_flow_context *flow;
    struct lp_build_mask_context mask;
    boolean early_depth_test;
    unsigned attrib;
@@ -309,25 +209,35 @@ generate_fs(struct llvmpipe_context *lp,
 
    consts_ptr = lp_jit_context_constants(builder, context_ptr);
 
-   lp_build_mask_begin(&mask, builder, type, *pmask);
+   flow = lp_build_flow_create(builder);
+
+   memset(outputs, 0, sizeof outputs);
+
+   lp_build_flow_scope_begin(flow);
+
+   /* Declare the color and z variables */
+   for(chan = 0; chan < NUM_CHANNELS; ++chan) {
+      color[chan] = LLVMGetUndef(vec_type);
+      lp_build_flow_scope_declare(flow, &color[chan]);
+   }
+   lp_build_flow_scope_declare(flow, &z);
+
+   lp_build_mask_begin(&mask, flow, type, *pmask);
 
    early_depth_test =
-      lp->depth_stencil->depth.enabled &&
-      lp->framebuffer.zsbuf &&
-      !lp->depth_stencil->alpha.enabled &&
-      !lp->fs->info.uses_kill &&
-      !lp->fs->info.writes_z;
+      key->depth.enabled &&
+      !key->alpha.enabled &&
+      !shader->info.uses_kill &&
+      !shader->info.writes_z;
 
    if(early_depth_test)
-      generate_depth(lp, builder, &key->depth,
+      generate_depth(builder, key,
                      type, &mask,
                      z, depth_ptr);
 
-   memset(outputs, 0, sizeof outputs);
-
    lp_build_tgsi_soa(builder, tokens, type, &mask,
                      consts_ptr, interp->pos, interp->inputs,
-                     outputs, emit_fetch_texel, sampler);
+                     outputs, sampler);
 
    for (attrib = 0; attrib < shader->info.num_outputs; ++attrib) {
       for(chan = 0; chan < NUM_CHANNELS; ++chan) {
@@ -368,12 +278,16 @@ generate_fs(struct llvmpipe_context *lp,
    }
 
    if(!early_depth_test)
-      generate_depth(lp, builder, &key->depth,
+      generate_depth(builder, key,
                      type, &mask,
                      z, depth_ptr);
 
    lp_build_mask_end(&mask);
 
+   lp_build_flow_scope_end(flow);
+
+   lp_build_flow_destroy(flow);
+
    *pmask = mask.value;
 
 }
@@ -392,6 +306,8 @@ generate_blend(const struct pipe_blend_state *blend,
                LLVMValueRef dst_ptr)
 {
    struct lp_build_context bld;
+   struct lp_build_flow_context *flow;
+   struct lp_build_mask_context mask_ctx;
    LLVMTypeRef vec_type;
    LLVMTypeRef int_vec_type;
    LLVMValueRef const_ptr;
@@ -400,11 +316,14 @@ generate_blend(const struct pipe_blend_state *blend,
    LLVMValueRef res[4];
    unsigned chan;
 
+   lp_build_context_init(&bld, builder, type);
+
+   flow = lp_build_flow_create(builder);
+   lp_build_mask_begin(&mask_ctx, flow, type, mask);
+
    vec_type = lp_build_vec_type(type);
    int_vec_type = lp_build_int_vec_type(type);
 
-   lp_build_context_init(&bld, builder, type);
-
    const_ptr = lp_jit_context_blend_color(builder, context_ptr);
    const_ptr = LLVMBuildBitCast(builder, const_ptr,
                                 LLVMPointerType(vec_type, 0), "");
@@ -422,11 +341,16 @@ generate_blend(const struct pipe_blend_state *blend,
    lp_build_blend_soa(builder, blend, type, src, dst, con, res);
 
    for(chan = 0; chan < 4; ++chan) {
-      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
-      lp_build_name(res[chan], "res.%c", "rgba"[chan]);
-      res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
-      LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, ""));
+      if(blend->colormask & (1 << chan)) {
+         LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), chan, 0);
+         lp_build_name(res[chan], "res.%c", "rgba"[chan]);
+         res[chan] = lp_build_select(&bld, mask, res[chan], dst[chan]);
+         LLVMBuildStore(builder, res[chan], LLVMBuildGEP(builder, dst_ptr, &index, 1, ""));
+      }
    }
+
+   lp_build_mask_end(&mask_ctx);
+   lp_build_flow_destroy(flow);
 }
 
 
@@ -462,7 +386,7 @@ generate_fragment(struct llvmpipe_context *lp,
    LLVMBuilderRef builder;
    LLVMValueRef x0;
    LLVMValueRef y0;
-   struct build_fetch_texel_context sampler;
+   struct lp_build_sampler_soa *sampler;
    struct lp_build_interp_soa_context interp;
    LLVMValueRef fs_mask[LP_MAX_VECTOR_LENGTH];
    LLVMValueRef fs_out_color[NUM_CHANNELS][LP_MAX_VECTOR_LENGTH];
@@ -586,8 +510,13 @@ generate_fragment(struct llvmpipe_context *lp,
                             a0_ptr, dadx_ptr, dady_ptr,
                             x0, y0, 2, 0);
 
-   memset(&sampler, 0, sizeof sampler);
-   sampler.context_ptr = context_ptr;
+#if 0
+   /* C texture sampling */
+   sampler = lp_c_sampler_soa_create(context_ptr);
+#else
+   /* code generated texture sampling */
+   sampler = lp_llvm_sampler_soa_create(key->sampler, context_ptr);
+#endif
 
    for(i = 0; i < num_fs; ++i) {
       LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
@@ -606,7 +535,7 @@ generate_fragment(struct llvmpipe_context *lp,
                   context_ptr,
                   i,
                   &interp,
-                  &sampler,
+                  sampler,
                   &fs_mask[i],
                   out_color,
                   depth_ptr_i);
@@ -615,6 +544,8 @@ generate_fragment(struct llvmpipe_context *lp,
          fs_out_color[chan][i] = out_color[chan];
    }
 
+   sampler->destroy(sampler);
+
    /* 
     * Convert the fs's output color and mask to fit to the blending type. 
     */
@@ -765,18 +696,45 @@ llvmpipe_set_constant_buffer(struct pipe_context *pipe,
  */
 static void
 make_variant_key(struct llvmpipe_context *lp,
+                 struct lp_fragment_shader *shader,
                  struct lp_fragment_shader_variant_key *key)
 {
+   unsigned i;
+
    memset(key, 0, sizeof *key);
 
-   memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
+   if(lp->framebuffer.zsbuf &&
+      lp->depth_stencil->depth.enabled) {
+      key->zsbuf_format = lp->framebuffer.zsbuf->format;
+      memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
+   }
 
    key->alpha.enabled = lp->depth_stencil->alpha.enabled;
    if(key->alpha.enabled)
       key->alpha.func = lp->depth_stencil->alpha.func;
    /* alpha.ref_value is passed in jit_context */
 
-   memcpy(&key->blend, lp->blend, sizeof key->blend);
+   if(lp->framebuffer.cbufs[0]) {
+      const struct util_format_description *format_desc;
+      unsigned chan;
+
+      memcpy(&key->blend, lp->blend, sizeof key->blend);
+
+      format_desc = util_format_description(lp->framebuffer.cbufs[0]->format);
+      assert(format_desc->layout == UTIL_FORMAT_COLORSPACE_RGB ||
+             format_desc->layout == UTIL_FORMAT_COLORSPACE_SRGB);
+
+      /* mask out color channels not present in the color buffer */
+      for(chan = 0; chan < 4; ++chan) {
+         enum util_format_swizzle swizzle = format_desc->swizzle[chan];
+         if(swizzle > 4)
+            key->blend.colormask &= ~(1 << chan);
+      }
+   }
+
+   for(i = 0; i < PIPE_MAX_SAMPLERS; ++i)
+      if(shader->info.file_mask[TGSI_FILE_SAMPLER] & (1 << i))
+         lp_sampler_static_state(&key->sampler[i], lp->texture[i], lp->sampler[i]);
 }
 
 
@@ -787,7 +745,7 @@ llvmpipe_update_fs(struct llvmpipe_context *lp)
    struct lp_fragment_shader_variant_key key;
    struct lp_fragment_shader_variant *variant;
 
-   make_variant_key(lp, &key);
+   make_variant_key(lp, shader, &key);
 
    variant = shader->variants;
    while(variant) {
diff --git a/src/gallium/drivers/llvmpipe/lp_state_sampler.c b/src/gallium/drivers/llvmpipe/lp_state_sampler.c
index 4fef541b1e3..c69d90c723a 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_sampler.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_sampler.c
@@ -98,6 +98,16 @@ llvmpipe_set_sampler_textures(struct pipe_context *pipe,
 
       pipe_texture_reference(&llvmpipe->texture[i], tex);
       lp_tex_tile_cache_set_texture(llvmpipe->tex_cache[i], tex);
+
+      if(tex) {
+         struct llvmpipe_texture *lp_tex = llvmpipe_texture(tex);
+         struct lp_jit_texture *jit_tex = &llvmpipe->jit_context.textures[i];
+         jit_tex->width = tex->width[0];
+         jit_tex->height = tex->height[0];
+         jit_tex->stride = lp_tex->stride[0];
+         if(!lp_tex->dt)
+            jit_tex->data = lp_tex->data;
+      }
    }
 
    llvmpipe->num_textures = num;
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 1d192355eed..d8455e56490 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -119,7 +119,7 @@ add_load_rgba_test(LLVMModuleRef module,
 
    lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 1, 0), &loop);
 
-   rgba = lp_build_load_rgba(builder, format, ptr);
+   rgba = lp_build_load_rgba_aos(builder, format, ptr);
    LLVMBuildStore(builder, rgba, rgba_ptr);
 
    lp_build_loop_end(builder, LLVMConstInt(LLVMInt32Type(), 4, 0), NULL, &loop);
@@ -160,7 +160,7 @@ add_store_rgba_test(LLVMModuleRef module,
 
    rgba = LLVMBuildLoad(builder, rgba_ptr, "");
 
-   lp_build_store_rgba(builder, format, ptr, rgba);
+   lp_build_store_rgba_aos(builder, format, ptr, rgba);
 
    LLVMBuildRetVoid(builder);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_cache.c b/src/gallium/drivers/llvmpipe/lp_tex_cache.c
index 23a94b5b0d5..773e8482425 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_cache.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_cache.c
@@ -154,7 +154,7 @@ lp_tex_tile_cache_validate_texture(struct llvmpipe_tex_tile_cache *tc)
       if (lpt->timestamp != tc->timestamp) {
          /* texture was modified, invalidate all cached tiles */
          uint i;
-         _debug_printf("INV %d %d\n", tc->timestamp, lpt->timestamp);
+         debug_printf("INV %d %d\n", tc->timestamp, lpt->timestamp);
          for (i = 0; i < NUM_ENTRIES; i++) {
             tc->entries[i].addr.bits.invalid = 1;
          }
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.h b/src/gallium/drivers/llvmpipe/lp_tex_sample.h
index 628ec3f1efd..9ad1bde9565 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample.h
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample.h
@@ -29,10 +29,13 @@
 #define LP_TEX_SAMPLE_H
 
 
+#include <llvm-c/Core.h>
+
 #include "tgsi/tgsi_exec.h"
 
 
 struct llvmpipe_tex_tile_cache;
+struct lp_sampler_static_state;
 
 
 /**
@@ -75,4 +78,24 @@ lp_get_samples(struct tgsi_sampler *tgsi_sampler,
                float rgba[NUM_CHANNELS][QUAD_SIZE]);
 
 
+/**
+ * Texture sampling code generator that just calls lp_get_samples C function
+ * for the actual sampling computation.
+ *
+ * @param context_ptr LLVM value with the pointer to the struct lp_jit_context.
+ */
+struct lp_build_sampler_soa *
+lp_c_sampler_soa_create(LLVMValueRef context_ptr);
+
+
+/**
+ * Pure-LLVM texture sampling code generator.
+ *
+ * @param context_ptr LLVM value with the pointer to the struct lp_jit_context.
+ */
+struct lp_build_sampler_soa *
+lp_llvm_sampler_soa_create(const struct lp_sampler_static_state *key,
+                           LLVMValueRef context_ptr);
+
+
 #endif /* LP_TEX_SAMPLE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c
index 94eb6dad5af..9a876f404df 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c
@@ -1578,3 +1578,136 @@ out:
    tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba );
 }
 
+
+void PIPE_CDECL
+lp_fetch_texel_soa( struct tgsi_sampler **samplers,
+                    uint32_t unit,
+                    float *store )
+{
+   struct tgsi_sampler *sampler = samplers[unit];
+
+#if 0
+   uint j;
+
+   debug_printf("%s sampler: %p (%p) store: %p\n",
+                __FUNCTION__,
+                sampler, *sampler,
+                store );
+
+   debug_printf("lodbias %f\n", store[12]);
+
+   for (j = 0; j < 4; j++)
+      debug_printf("sample %d texcoord %f %f\n",
+                   j,
+                   store[0+j],
+                   store[4+j]);
+#endif
+
+   {
+      float rgba[NUM_CHANNELS][QUAD_SIZE];
+      sampler->get_samples(sampler,
+                           &store[0],
+                           &store[4],
+                           &store[8],
+                           0.0f, /*store[12],  lodbias */
+                           rgba);
+      memcpy(store, rgba, sizeof rgba);
+   }
+
+#if 0
+   for (j = 0; j < 4; j++)
+      debug_printf("sample %d result %f %f %f %f\n",
+                   j,
+                   store[0+j],
+                   store[4+j],
+                   store[8+j],
+                   store[12+j]);
+#endif
+}
+
+
+#include "lp_bld_type.h"
+#include "lp_bld_intr.h"
+#include "lp_bld_tgsi.h"
+
+
+struct lp_c_sampler_soa
+{
+   struct lp_build_sampler_soa base;
+
+   LLVMValueRef context_ptr;
+
+   LLVMValueRef samplers_ptr;
+
+   /** Coords/texels store */
+   LLVMValueRef store_ptr;
+};
+
+
+static void
+lp_c_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
+{
+   FREE(sampler);
+}
+
+
+static void
+lp_c_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *_sampler,
+                                  LLVMBuilderRef builder,
+                                  union lp_type type,
+                                  unsigned unit,
+                                  unsigned num_coords,
+                                  const LLVMValueRef *coords,
+                                  LLVMValueRef lodbias,
+                                  LLVMValueRef *texel)
+{
+   struct lp_c_sampler_soa *sampler = (struct lp_c_sampler_soa *)_sampler;
+   LLVMTypeRef vec_type = LLVMTypeOf(coords[0]);
+   LLVMValueRef args[3];
+   unsigned i;
+
+   if(!sampler->samplers_ptr)
+      sampler->samplers_ptr = lp_jit_context_samplers(builder, sampler->context_ptr);
+
+   if(!sampler->store_ptr)
+      sampler->store_ptr = LLVMBuildArrayAlloca(builder,
+                                            vec_type,
+                                            LLVMConstInt(LLVMInt32Type(), 4, 0),
+                                            "texel_store");
+
+   for (i = 0; i < num_coords; i++) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef coord_ptr = LLVMBuildGEP(builder, sampler->store_ptr, &index, 1, "");
+      LLVMBuildStore(builder, coords[i], coord_ptr);
+   }
+
+   args[0] = sampler->samplers_ptr;
+   args[1] = LLVMConstInt(LLVMInt32Type(), unit, 0);
+   args[2] = sampler->store_ptr;
+
+   lp_build_intrinsic(builder, "fetch_texel", LLVMVoidType(), args, 3);
+
+   for (i = 0; i < NUM_CHANNELS; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      LLVMValueRef texel_ptr = LLVMBuildGEP(builder, sampler->store_ptr, &index, 1, "");
+      texel[i] = LLVMBuildLoad(builder, texel_ptr, "");
+   }
+}
+
+
+struct lp_build_sampler_soa *
+lp_c_sampler_soa_create(LLVMValueRef context_ptr)
+{
+   struct lp_c_sampler_soa *sampler;
+
+   sampler = CALLOC_STRUCT(lp_c_sampler_soa);
+   if(!sampler)
+      return NULL;
+
+   sampler->base.destroy = lp_c_sampler_soa_destroy;
+   sampler->base.emit_fetch_texel = lp_c_sampler_soa_emit_fetch_texel;
+   sampler->context_ptr = context_ptr;
+
+   return &sampler->base;
+}
+
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
new file mode 100644
index 00000000000..7d31705d014
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
@@ -0,0 +1,196 @@
+/**************************************************************************
+ * 
+ * Copyright 2009 VMware, Inc.
+ * All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Texture sampling code generation
+ *
+ * This file is nothing more than ugly glue between three largely independent
+ * entities:
+ * - TGSI -> LLVM translation (i.e., lp_build_tgsi_soa)
+ * - texture sampling code generation (i.e., lp_build_sample_soa)
+ * - LLVM pipe driver
+ *
+ * All interesting code is in the functions mentioned above. There is really
+ * nothing to see here.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_type.h"
+#include "lp_bld_intr.h"
+#include "lp_bld_sample.h"
+#include "lp_bld_tgsi.h"
+#include "lp_state.h"
+#include "lp_tex_sample.h"
+
+
+/**
+ * This provides the bridge between the sampler state store in lp_jit_context
+ * and lp_jit_texture and the sampler code generator. It provides the
+ * texture layout information required by the texture sampler code generator
+ * in terms of the state stored in lp_jit_context and lp_jit_texture in runtime.
+ */
+struct llvmpipe_sampler_dynamic_state
+{
+   struct lp_sampler_dynamic_state base;
+
+   const struct lp_sampler_static_state *static_state;
+
+   LLVMValueRef context_ptr;
+};
+
+
+/**
+ * This is the bridge between our sampler and the TGSI translator.
+ */
+struct lp_llvm_sampler_soa
+{
+   struct lp_build_sampler_soa base;
+
+   struct llvmpipe_sampler_dynamic_state dynamic_state;
+};
+
+
+/**
+ * Fetch the specified member of the lp_jit_texture structure.
+ *
+ * @sa http://llvm.org/docs/GetElementPtr.html
+ */
+static LLVMValueRef
+lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
+                       LLVMBuilderRef builder,
+                       unsigned unit,
+                       unsigned member_index,
+                       const char *member_name)
+{
+   struct llvmpipe_sampler_dynamic_state *state = (struct llvmpipe_sampler_dynamic_state *)base;
+   LLVMValueRef indices[4];
+   LLVMValueRef ptr;
+   LLVMValueRef res;
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+
+   /* context[0] */
+   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   /* context[0].textures */
+   indices[1] = LLVMConstInt(LLVMInt32Type(), LP_JIT_CONTEXT_TEXTURES_INDEX, 0);
+   /* context[0].textures[unit] */
+   indices[2] = LLVMConstInt(LLVMInt32Type(), unit, 0);
+   /* context[0].textures[unit].member */
+   indices[3] = LLVMConstInt(LLVMInt32Type(), member_index, 0);
+
+   ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
+
+   res = LLVMBuildLoad(builder, ptr, "");
+
+   lp_build_name(res, "context.texture%u.%s", unit, member_name);
+
+   return res;
+}
+
+
+/**
+ * Helper macro to instantiate the functions that generate the code to fetch
+ * the members of lp_jit_texture to fulfill the sampler code generator requests.
+ *
+ * This complexity is the price we have to pay to keep the texture sampler code
+ * generator a reusable module without dependencies to llvmpipe internals.
+ */
+#define LP_LLVM_TEXTURE_MEMBER(_name, _index) \
+   static LLVMValueRef \
+   lp_llvm_texture_##_name( struct lp_sampler_dynamic_state *base, \
+                            LLVMBuilderRef builder, \
+                            unsigned unit) \
+   { \
+      return lp_llvm_texture_member(base, builder, unit, _index, #_name ); \
+   }
+
+
+LP_LLVM_TEXTURE_MEMBER(width,    LP_JIT_TEXTURE_WIDTH)
+LP_LLVM_TEXTURE_MEMBER(height,   LP_JIT_TEXTURE_HEIGHT)
+LP_LLVM_TEXTURE_MEMBER(stride,   LP_JIT_TEXTURE_STRIDE)
+LP_LLVM_TEXTURE_MEMBER(data_ptr, LP_JIT_TEXTURE_DATA)
+
+
+static void
+lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
+{
+   FREE(sampler);
+}
+
+
+static void
+lp_llvm_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *base,
+                                     LLVMBuilderRef builder,
+                                     union lp_type type,
+                                     unsigned unit,
+                                     unsigned num_coords,
+                                     const LLVMValueRef *coords,
+                                     LLVMValueRef lodbias,
+                                     LLVMValueRef *texel)
+{
+   struct lp_llvm_sampler_soa *sampler = (struct lp_llvm_sampler_soa *)base;
+
+   assert(unit < PIPE_MAX_SAMPLERS);
+
+   lp_build_sample_soa(builder,
+                       &sampler->dynamic_state.static_state[unit],
+                       &sampler->dynamic_state.base,
+                       type,
+                       unit,
+                       num_coords,
+                       coords,
+                       lodbias,
+                       texel);
+}
+
+
+struct lp_build_sampler_soa *
+lp_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state,
+                           LLVMValueRef context_ptr)
+{
+   struct lp_llvm_sampler_soa *sampler;
+
+   sampler = CALLOC_STRUCT(lp_llvm_sampler_soa);
+   if(!sampler)
+      return NULL;
+
+   sampler->base.destroy = lp_llvm_sampler_soa_destroy;
+   sampler->base.emit_fetch_texel = lp_llvm_sampler_soa_emit_fetch_texel;
+   sampler->dynamic_state.base.width = lp_llvm_texture_width;
+   sampler->dynamic_state.base.height = lp_llvm_texture_height;
+   sampler->dynamic_state.base.stride = lp_llvm_texture_stride;
+   sampler->dynamic_state.base.data_ptr = lp_llvm_texture_data_ptr;
+   sampler->dynamic_state.static_state = static_state;
+   sampler->dynamic_state.context_ptr = context_ptr;
+
+   return &sampler->base;
+}
+
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 03b9243b828..93479a0314a 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -148,6 +148,7 @@ nv50_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
 	mt->image_nr = 1;
 	mt->level[0].pitch = *stride;
 	mt->level[0].image_offset = CALLOC(1, sizeof(unsigned));
+	mt->level[0].tile_mode = bo->tile_mode;
 
 	nouveau_bo_ref(bo, &mt->base.bo);
 	return &mt->base.base;
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 344c2cf6dde..d294356f75d 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -66,7 +66,8 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 			so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM);
 			break;
 		}
-		so_data(so, bo->tile_mode << 4);
+		so_data(so, nv50_miptree(pt)->
+				level[fb->cbufs[i]->level].tile_mode << 4);
 		so_data(so, 0x00000000);
 
 		so_method(so, tesla, 0x1224, 1);
@@ -110,7 +111,8 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 			so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM);
 			break;
 		}
-		so_data(so, bo->tile_mode << 4);
+		so_data(so, nv50_miptree(pt)->
+				level[fb->zsbuf->level].tile_mode << 4);
 		so_data(so, 0x00000000);
 
 		so_method(so, tesla, 0x1538, 1);
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index b266324f58d..6bf6f773b0c 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -60,13 +60,13 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
  	format = nv50_format(ps->format);
  	if (format < 0)
  		return 1;
-  
+
  	if (!bo->tile_flags) {
  		BEGIN_RING(chan, eng2d, mthd, 2);
  		OUT_RING  (chan, format);
  		OUT_RING  (chan, 1);
  		BEGIN_RING(chan, eng2d, mthd + 0x14, 5);
- 		OUT_RING  (chan, mt->level[0].pitch);
+		OUT_RING  (chan, mt->level[ps->level].pitch);
  		OUT_RING  (chan, ps->width);
  		OUT_RING  (chan, ps->height);
  		OUT_RELOCh(chan, bo, ps->offset, flags);
@@ -75,7 +75,7 @@ nv50_surface_set(struct nv50_screen *screen, struct pipe_surface *ps, int dst)
  		BEGIN_RING(chan, eng2d, mthd, 5);
  		OUT_RING  (chan, format);
  		OUT_RING  (chan, 0);
- 		OUT_RING  (chan, bo->tile_mode << 4);
+		OUT_RING  (chan, mt->level[ps->level].tile_mode << 4);
  		OUT_RING  (chan, 1);
  		OUT_RING  (chan, 0);
  		BEGIN_RING(chan, eng2d, mthd + 0x18, 4);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 6c5914baa35..52b1c9a6b26 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -184,8 +184,15 @@ struct r300_texture {
     /* Offsets into the buffer. */
     unsigned offset[PIPE_MAX_TEXTURE_LEVELS];
 
-    /* Stride (pitch?) of this texture in bytes */
-    unsigned stride;
+    /**
+     * If non-zero, override the natural texture layout with
+     * a custom stride (in bytes).
+     *
+     * \note Mipmapping fails for textures with a non-natural layout!
+     *
+     * \sa r300_texture_get_stride
+     */
+    unsigned stride_override;
 
     /* Total size of this texture, in bytes. */
     unsigned size;
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 1bc35c24867..a1b36ba2ed1 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -283,7 +283,7 @@ void r300_emit_fb_state(struct r300_context* r300,
     for (i = 0; i < fb->nr_cbufs; i++) {
         tex = (struct r300_texture*)fb->cbufs[i]->texture;
         assert(tex && tex->buffer && "cbuf is marked, but NULL!");
-        pixpitch = tex->stride / tex->tex.block.size;
+        pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
 
         OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1);
         OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
@@ -300,7 +300,7 @@ void r300_emit_fb_state(struct r300_context* r300,
     if (fb->zsbuf) {
         tex = (struct r300_texture*)fb->zsbuf->texture;
         assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
-        pixpitch = tex->stride / tex->tex.block.size;
+        pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
         OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index d05a736dd96..737396d8d97 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -31,6 +31,7 @@
 #include "r300_state_derived.h"
 
 /* r300_render: Vertex and index buffer primitive emission. */
+#define R300_MAX_VBO_SIZE  (1024 * 1024)
 
 struct r300_render {
     /* Parent class */
@@ -46,7 +47,10 @@ struct r300_render {
 
     /* VBO */
     struct pipe_buffer* vbo;
-    size_t vbo_alloc_size;
+    size_t vbo_size;
+    size_t vbo_offset;
+    size_t vbo_max_used;
+    void * vbo_ptr;
 };
 
 static INLINE struct r300_render*
@@ -75,19 +79,18 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render,
     struct pipe_screen* screen = r300->context.screen;
     size_t size = (size_t)vertex_size * (size_t)count;
 
-    if (r300render->vbo && (size > r300render->vbo_alloc_size)) {
-        pipe_buffer_reference(&r300render->vbo, NULL);
-    }
-
-    if (!r300render->vbo) {
+    if (size + r300render->vbo_offset > r300render->vbo_size) 
+    {
         r300render->vbo = pipe_buffer_create(screen,
                                              64,
                                              PIPE_BUFFER_USAGE_VERTEX,
-                                             size);
+                                             R300_MAX_VBO_SIZE);
+        r300render->vbo_size = R300_MAX_VBO_SIZE;
     }
 
-    r300render->vbo_alloc_size = MAX2(size, r300render->vbo_alloc_size);
     r300render->vertex_size = vertex_size;
+    r300->vbo = r300render->vbo;
+    r300->vbo_offset = r300render->vbo_offset;
 
     return (r300render->vbo) ? TRUE : FALSE;
 }
@@ -97,8 +100,10 @@ static void* r300_render_map_vertices(struct vbuf_render* render)
     struct r300_render* r300render = r300_render(render);
     struct pipe_screen* screen = r300render->r300->context.screen;
 
-    return (unsigned char*)pipe_buffer_map(screen, r300render->vbo,
-                                           PIPE_BUFFER_USAGE_CPU_WRITE);
+    r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
+                                          PIPE_BUFFER_USAGE_CPU_WRITE);
+
+    return (r300render->vbo_ptr + r300render->vbo_offset);
 }
 
 static void r300_render_unmap_vertices(struct vbuf_render* render,
@@ -107,15 +112,24 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
 {
     struct r300_render* r300render = r300_render(render);
     struct pipe_screen* screen = r300render->r300->context.screen;
+    CS_LOCALS(r300render->r300);
+    BEGIN_CS(2);
+    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
+    END_CS;
 
+    r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 
+                                    r300render->vertex_size * (max + 1));
     pipe_buffer_unmap(screen, r300render->vbo);
 }
 
 static void r300_render_release_vertices(struct vbuf_render* render)
 {
     struct r300_render* r300render = r300_render(render);
+    struct r300_context* r300 = r300render->r300;
 
-    pipe_buffer_reference(&r300render->vbo, NULL);
+    r300render->vbo_offset += r300render->vbo_max_used;
+    r300render->vbo_max_used = 0;
+    r300->vbo = NULL;
 }
 
 static boolean r300_render_set_primitive(struct vbuf_render* render,
@@ -163,14 +177,12 @@ static boolean r300_render_set_primitive(struct vbuf_render* render,
     return TRUE;
 }
 
-static void prepare_render(struct r300_render* render, unsigned count)
+static void r300_prepare_render(struct r300_render* render, unsigned count)
 {
     struct r300_context* r300 = render->r300;
 
     CS_LOCALS(r300);
 
-    r300->vbo = render->vbo;
-
     r300_emit_dirty_state(r300);
 }
 
@@ -183,7 +195,7 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
 
     CS_LOCALS(r300);
 
-    prepare_render(r300render, count);
+    r300_prepare_render(r300render, count);
 
     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 
@@ -208,7 +220,7 @@ static void r300_render_draw(struct vbuf_render* render,
 
     CS_LOCALS(r300);
 
-    prepare_render(r300render, count);
+    r300_prepare_render(r300render, count);
 
     /* Send our indices into an index buffer. */
     index_buffer = pipe_buffer_create(screen, 64, PIPE_BUFFER_USAGE_VERTEX,
@@ -217,25 +229,7 @@ static void r300_render_draw(struct vbuf_render* render,
         return;
     }
 
-/*
-    index_map = pipe_buffer_map(screen, index_buffer,
-                                PIPE_BUFFER_USAGE_CPU_WRITE);
-    memcpy(index_map, indices, count);
-    pipe_buffer_unmap(screen, index_buffer);
-
-    debug_printf("r300: Doing indexbuf render, count %d\n", count);
-
-    BEGIN_CS(8);
-    OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
-    OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
-           r300render->hwprim);
-    OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
-    OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2));
-    OUT_CS_INDEX_RELOC(index_buffer, 0, count, RADEON_GEM_DOMAIN_GTT, 0, 0);
-    END_CS; */
-
-    BEGIN_CS(4 + (count+1)/2);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count);
+    BEGIN_CS(2 + (count+1)/2);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
            r300render->hwprim);
@@ -273,6 +267,10 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300)
     r300render->base.release_vertices = r300_render_release_vertices;
     r300render->base.destroy = r300_render_destroy;
 
+    r300render->vbo = NULL;
+    r300render->vbo_size = 0;
+    r300render->vbo_offset = 0;
+
     return &r300render->base;
 }
 
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 15740f61252..593178c50be 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -323,13 +323,14 @@ r300_get_tex_transfer(struct pipe_screen *screen,
     if (trans) {
         pipe_texture_reference(&trans->transfer.texture, texture);
         trans->transfer.format = texture->format;
+        trans->transfer.x = x;
+        trans->transfer.y = y;
         trans->transfer.width = w;
         trans->transfer.height = h;
         trans->transfer.block = texture->block;
         trans->transfer.nblocksx = texture->nblocksx[level];
         trans->transfer.nblocksy = texture->nblocksy[level];
-        trans->transfer.stride = align(pf_get_stride(&trans->transfer.block,
-                                                     texture->width[level]), 32);
+        trans->transfer.stride = r300_texture_get_stride(tex, level);
         trans->transfer.usage = usage;
         trans->offset = offset;
     }
@@ -356,7 +357,7 @@ static void* r300_transfer_map(struct pipe_screen* screen,
     if (transfer->usage != PIPE_TRANSFER_READ) {
         flags |= PIPE_BUFFER_USAGE_CPU_WRITE;
     }
-    
+
     map = pipe_buffer_map(screen, tex->buffer, flags);
 
     if (!map) {
diff --git a/src/gallium/drivers/r300/r300_surface.c b/src/gallium/drivers/r300/r300_surface.c
index 96e6e4a77d4..cc6288cb519 100644
--- a/src/gallium/drivers/r300/r300_surface.c
+++ b/src/gallium/drivers/r300/r300_surface.c
@@ -29,7 +29,7 @@ static void r300_surface_setup(struct r300_context* r300,
                                unsigned w, unsigned h)
 {
     struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
-    unsigned pixpitch = dest->stride / dest->tex.block.size;
+    unsigned pixpitch = r300_texture_get_stride(dest, 0) / dest->tex.block.size;
     CS_LOCALS(r300);
 
     r300_emit_blend_state(r300, &blend_clear_state);
@@ -100,7 +100,7 @@ static void r300_surface_fill(struct pipe_context* pipe,
     struct r300_context* r300 = r300_context(pipe);
     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
     struct r300_texture* tex = (struct r300_texture*)dest->texture;
-    unsigned pixpitch = tex->stride / tex->tex.block.size;
+    unsigned pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
     boolean invalid = FALSE;
     CS_LOCALS(r300);
 
@@ -233,7 +233,7 @@ static void r300_surface_copy(struct pipe_context* pipe,
     struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
     struct r300_texture* srctex = (struct r300_texture*)src->texture;
     struct r300_texture* desttex = (struct r300_texture*)dest->texture;
-    unsigned pixpitch = srctex->stride / srctex->tex.block.size;
+    unsigned pixpitch = r300_texture_get_stride(srctex, 0) / srctex->tex.block.size;
     boolean invalid = FALSE;
     float fsrcx = srcx, fsrcy = srcy, fdestx = destx, fdesty = desty;
     CS_LOCALS(r300);
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 590052509cc..6e8c3683200 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -25,7 +25,6 @@
 static void r300_setup_texture_state(struct r300_texture* tex,
                                      unsigned width,
                                      unsigned height,
-                                     unsigned pitch,
                                      unsigned levels)
 {
     struct r300_texture_state* state = &tex->state;
@@ -38,7 +37,7 @@ static void r300_setup_texture_state(struct r300_texture* tex,
     /* XXX */
     state->format1 = r300_translate_texformat(tex->tex.format);
 
-    state->format2 = pitch - 1;
+    state->format2 = r300_texture_get_stride(tex, 0);
 
     /* Assume (somewhat foolishly) that oversized textures will
      * not be permitted by the state tracker. */
@@ -50,13 +49,30 @@ static void r300_setup_texture_state(struct r300_texture* tex,
     }
 
     debug_printf("r300: Set texture state (%dx%d, pitch %d, %d levels)\n",
-            width, height, pitch, levels);
+            width, height, levels);
+}
+
+/**
+ * Return the stride, in bytes, of the texture images of the given texture
+ * at the given level.
+ */
+unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
+{
+    if (tex->stride_override)
+        return tex->stride_override;
+
+    if (level > tex->tex.last_level) {
+        debug_printf("%s: level (%u) > last_level (%u)\n", level, tex->tex.last_level);
+        return 0;
+    }
+
+    return align(pf_get_stride(&tex->tex.block, tex->tex.width[level]), 32);
 }
 
 static void r300_setup_miptree(struct r300_texture* tex)
 {
     struct pipe_texture* base = &tex->tex;
-    int stride, size, offset;
+    int stride, size;
     int i;
 
     for (i = 0; i <= base->last_level; i++) {
@@ -74,7 +90,7 @@ static void r300_setup_miptree(struct r300_texture* tex)
          * XXX
          * POT, uncompressed, unmippmapped textures can be aligned to 32,
          * instead of 64. */
-        stride = align(pf_get_stride(&base->block, base->width[i]), 32);
+        stride = r300_texture_get_stride(tex, i);
         size = stride * base->nblocksy[i] * base->depth[i];
 
         tex->offset[i] = align(tex->size, 32);
@@ -84,10 +100,6 @@ static void r300_setup_miptree(struct r300_texture* tex)
                 "(%dx%dx%d px, pitch %d bytes)\n",
                 i, base->width[i], base->height[i], base->depth[i],
                 stride);
-        /* Save stride of first level to the texture. */
-        if (i == 0) {
-            tex->stride = stride;
-        }
     }
 }
 
@@ -109,7 +121,7 @@ static struct pipe_texture*
     r300_setup_miptree(tex);
 
     r300_setup_texture_state(tex, template->width[0], template->height[0],
-            template->width[0], template->last_level);
+                             template->last_level);
 
     tex->buffer = screen->buffer_create(screen, 1024,
                                         PIPE_BUFFER_USAGE_PIXEL,
@@ -189,11 +201,10 @@ static struct pipe_texture*
     pipe_reference_init(&tex->tex.reference, 1);
     tex->tex.screen = screen;
 
-    tex->stride = *stride;
+    tex->stride_override = *stride;
 
     /* XXX */
-    r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0],
-            tex->stride, 0);
+    r300_setup_texture_state(tex, tex->tex.width[0], tex->tex.height[0], 0);
 
     pipe_buffer_reference(&tex->buffer, buffer);
 
@@ -221,7 +232,7 @@ boolean r300_get_texture_buffer(struct pipe_texture* texture,
     pipe_buffer_reference(buffer, tex->buffer);
 
     if (stride) {
-        *stride = tex->stride;
+        *stride = r300_texture_get_stride(tex, 0);
     }
 
     return TRUE;
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 3b56f0307c0..3109af5baca 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -30,8 +30,12 @@
 #include "r300_context.h"
 #include "r300_reg.h"
 
+struct r300_texture;
+
 void r300_init_screen_texture_functions(struct pipe_screen* screen);
 
+unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level);
+
 /* Note the signature of R300_EASY_TX_FORMAT(A, R, G, B, FORMAT)... */
 static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
 {
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index d68a1041063..0913ca1bd5b 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -158,7 +158,6 @@ static unsigned translate_saturate(unsigned saturate)
     switch(saturate) {
         case TGSI_SAT_NONE: return SATURATE_OFF;
         case TGSI_SAT_ZERO_ONE: return SATURATE_ZERO_ONE;
-        case TGSI_SAT_MINUS_PLUS_ONE: return SATURATE_PLUS_MINUS_ONE;
     }
 
     fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c
index fa59277438c..8fac8e6e05f 100644
--- a/src/gallium/drivers/softpipe/sp_clear.c
+++ b/src/gallium/drivers/softpipe/sp_clear.c
@@ -36,8 +36,6 @@
 #include "util/u_pack_color.h"
 #include "sp_clear.h"
 #include "sp_context.h"
-#include "sp_surface.h"
-#include "sp_state.h"
 #include "sp_tile_cache.h"
 
 
@@ -85,5 +83,7 @@ softpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
       /* non-cached surface */
       pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, cv);
 #endif
-      }
+   }
+
+   softpipe->dirty_render_cache = TRUE;
 }
diff --git a/src/gallium/drivers/softpipe/sp_clear.h b/src/gallium/drivers/softpipe/sp_clear.h
index 2e450672f58..9be3b86fe9f 100644
--- a/src/gallium/drivers/softpipe/sp_clear.h
+++ b/src/gallium/drivers/softpipe/sp_clear.h
@@ -32,7 +32,6 @@
 #ifndef SP_CLEAR_H
 #define SP_CLEAR_H
 
-#include "pipe/p_state.h"
 struct pipe_context;
 
 extern void
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 86df320ea8a..c699c433e9c 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -83,7 +83,8 @@ softpipe_unmap_transfers(struct softpipe_context *sp)
 }
 
 
-static void softpipe_destroy( struct pipe_context *pipe )
+static void
+softpipe_destroy( struct pipe_context *pipe )
 {
    struct softpipe_context *softpipe = softpipe_context( pipe );
    uint i;
@@ -121,6 +122,15 @@ static void softpipe_destroy( struct pipe_context *pipe )
    FREE( softpipe );
 }
 
+
+/**
+ * if (the texture is being used as a framebuffer surface)
+ *    return PIPE_REFERENCED_FOR_WRITE
+ * else if (the texture is a bound texture source)
+ *    return PIPE_REFERENCED_FOR_READ  XXX not done yet
+ * else
+ *    return PIPE_UNREFERENCED
+ */
 static unsigned int
 softpipe_is_texture_referenced( struct pipe_context *pipe,
 				struct pipe_texture *texture,
@@ -129,15 +139,17 @@ softpipe_is_texture_referenced( struct pipe_context *pipe,
    struct softpipe_context *softpipe = softpipe_context( pipe );
    unsigned i;
 
-   if(softpipe->dirty_render_cache) {
+   if (softpipe->dirty_render_cache) {
       for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) {
-         if(softpipe->framebuffer.cbufs[i] && 
-            softpipe->framebuffer.cbufs[i]->texture == texture)
+         if (softpipe->framebuffer.cbufs[i] && 
+             softpipe->framebuffer.cbufs[i]->texture == texture) {
             return PIPE_REFERENCED_FOR_WRITE;
+         }
       }
-      if(softpipe->framebuffer.zsbuf && 
-         softpipe->framebuffer.zsbuf->texture == texture)
+      if (softpipe->framebuffer.zsbuf && 
+          softpipe->framebuffer.zsbuf->texture == texture) {
          return PIPE_REFERENCED_FOR_WRITE;
+      }
    }
    
    /* FIXME: we also need to do the same for the texture cache */
@@ -145,6 +157,7 @@ softpipe_is_texture_referenced( struct pipe_context *pipe,
    return PIPE_UNREFERENCED;
 }
 
+
 static unsigned int
 softpipe_is_buffer_referenced( struct pipe_context *pipe,
 			       struct pipe_buffer *buf)
@@ -152,6 +165,7 @@ softpipe_is_buffer_referenced( struct pipe_context *pipe,
    return PIPE_UNREFERENCED;
 }
 
+
 struct pipe_context *
 softpipe_create( struct pipe_screen *screen )
 {
@@ -222,7 +236,6 @@ softpipe_create( struct pipe_screen *screen )
    softpipe->pipe.is_buffer_referenced = softpipe_is_buffer_referenced;
 
    softpipe_init_query_funcs( softpipe );
-   softpipe_init_texture_funcs( softpipe );
 
    /*
     * Alloc caches for accessing drawing surfaces and textures.
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 70f09324311..548e40c4a8c 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -30,26 +30,21 @@
   *   Michel Dänzer <[email protected]>
   */
 
-#include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/internal/p_winsys_screen.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
 #include "sp_context.h"
 #include "sp_state.h"
 #include "sp_texture.h"
-#include "sp_tile_cache.h"
 #include "sp_screen.h"
 #include "sp_winsys.h"
 
 
-/* Simple, maximally packed layout.
- */
-
-
-/* Conventional allocation path for non-display textures:
+/**
+ * Conventional allocation path for non-display textures:
+ * Use a simple, maximally packed layout.
  */
 static boolean
 softpipe_texture_layout(struct pipe_screen *screen,
@@ -89,6 +84,10 @@ softpipe_texture_layout(struct pipe_screen *screen,
    return spt->buffer != NULL;
 }
 
+
+/**
+ * Texture layout for simple color buffers.
+ */
 static boolean
 softpipe_displaytarget_layout(struct pipe_screen *screen,
                               struct softpipe_texture * spt)
@@ -112,9 +111,6 @@ softpipe_displaytarget_layout(struct pipe_screen *screen,
 }
 
 
-
-
-
 static struct pipe_texture *
 softpipe_texture_create(struct pipe_screen *screen,
                         const struct pipe_texture *templat)
@@ -342,14 +338,13 @@ softpipe_transfer_map( struct pipe_screen *screen,
    /* May want to different things here depending on read/write nature
     * of the map:
     */
-   if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) 
-   {
+   if (transfer->texture && transfer->usage != PIPE_TRANSFER_READ) {
       /* Do something to notify sharing contexts of a texture change.
        * In softpipe, that would mean flushing the texture cache.
        */
       softpipe_screen(screen)->timestamp++;
    }
-   
+
    xfer_map = map + softpipe_transfer(transfer)->offset +
       transfer->y / transfer->block.height * transfer->stride +
       transfer->x / transfer->block.width * transfer->block.size;
@@ -360,7 +355,7 @@ softpipe_transfer_map( struct pipe_screen *screen,
 
 static void
 softpipe_transfer_unmap(struct pipe_screen *screen,
-                       struct pipe_transfer *transfer)
+                        struct pipe_transfer *transfer)
 {
    struct softpipe_texture *spt;
 
@@ -377,12 +372,6 @@ softpipe_transfer_unmap(struct pipe_screen *screen,
 
 
 void
-softpipe_init_texture_funcs(struct softpipe_context *sp)
-{
-}
-
-
-void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = softpipe_texture_create;
@@ -404,7 +393,7 @@ softpipe_get_texture_buffer( struct pipe_texture *texture,
                              struct pipe_buffer **buf,
                              unsigned *stride )
 {
-   struct softpipe_texture *tex = (struct softpipe_texture *)texture;
+   struct softpipe_texture *tex = (struct softpipe_texture *) texture;
 
    if (!tex)
       return FALSE;
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 893aa7d11d8..b1b6130b22c 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -74,9 +74,6 @@ softpipe_transfer(struct pipe_transfer *pt)
 
 
 extern void
-softpipe_init_texture_funcs( struct softpipe_context *softpipe );
-
-extern void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen);
 
 
diff --git a/src/gallium/include/pipe/p_compiler.h b/src/gallium/include/pipe/p_compiler.h
index e6a67f8c2fd..c13cffceb0a 100644
--- a/src/gallium/include/pipe/p_compiler.h
+++ b/src/gallium/include/pipe/p_compiler.h
@@ -93,9 +93,11 @@ typedef int     _Bool;
 #endif
 
 
+#ifndef __HAIKU__
 typedef unsigned int       uint;
-typedef unsigned char      ubyte;
 typedef unsigned short     ushort;
+#endif
+typedef unsigned char      ubyte;
 
 #if 0
 #define boolean bool
diff --git a/src/gallium/include/pipe/p_config.h b/src/gallium/include/pipe/p_config.h
index 4152d6ac36a..de99957d9d0 100644
--- a/src/gallium/include/pipe/p_config.h
+++ b/src/gallium/include/pipe/p_config.h
@@ -140,6 +140,9 @@
 #define PIPE_OS_WINDOWS
 #endif
 
+#if defined(__HAIKU__)
+#define PIPE_OS_HAIKU
+#endif
 
 /*
  * Subsystem.
diff --git a/src/gallium/include/pipe/p_thread.h b/src/gallium/include/pipe/p_thread.h
index 96e8e087447..b1606dc6526 100644
--- a/src/gallium/include/pipe/p_thread.h
+++ b/src/gallium/include/pipe/p_thread.h
@@ -39,7 +39,7 @@
 #include "util/u_debug.h" /* for assert */
 
 
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
 
 #include <pthread.h> /* POSIX threads headers */
 #include <stdio.h> /* for perror() */
@@ -213,7 +213,7 @@ typedef unsigned pipe_condvar;
  */
 
 typedef struct {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
    pthread_key_t key;
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
    DWORD key;
@@ -228,7 +228,7 @@ typedef struct {
 static INLINE void
 pipe_tsd_init(pipe_tsd *tsd)
 {
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
    if (pthread_key_create(&tsd->key, NULL/*free*/) != 0) {
       perror("pthread_key_create(): failed to allocate key for thread specific data");
       exit(-1);
@@ -245,7 +245,7 @@ pipe_tsd_get(pipe_tsd *tsd)
    if (tsd->initMagic != (int) PIPE_TSD_INIT_MAGIC) {
       pipe_tsd_init(tsd);
    }
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
    return pthread_getspecific(tsd->key);
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
    assert(0);
@@ -262,7 +262,7 @@ pipe_tsd_set(pipe_tsd *tsd, void *value)
    if (tsd->initMagic != (int) PIPE_TSD_INIT_MAGIC) {
       pipe_tsd_init(tsd);
    }
-#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
+#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS) || defined(PIPE_OS_HAIKU)
    if (pthread_setspecific(tsd->key, value) != 0) {
       perror("pthread_set_specific() failed");
       exit(-1);
diff --git a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
index 20d682de3fb..7c8e545824f 100644
--- a/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
+++ b/src/gallium/state_trackers/g3dvl/vl_basic_csc.c
@@ -425,7 +425,7 @@ static int vlCreateVertexShader
 	 */
 	for (i = 0; i < 2; ++i)
 	{
-		inst = vl_inst4(TGSI_OPCODE_MADD, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i, TGSI_FILE_CONSTANT, i * 2, TGSI_FILE_CONSTANT, i * 2 + 1);
+		inst = vl_inst4(TGSI_OPCODE_MAD, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i, TGSI_FILE_CONSTANT, i * 2, TGSI_FILE_CONSTANT, i * 2 + 1);
 		ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 	}
 
diff --git a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc
index ef4a4b2add9..34d93e1df0a 100644
--- a/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc
+++ b/src/gallium/state_trackers/g3dvl/vl_r16snorm_mc_buf_shaders.inc
@@ -615,7 +615,7 @@ static int vlCreateFragmentShaderFieldPMB
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* floor t3, t3			; Get rid of fractional part */
-	inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
+	inst = vl_inst2(TGSI_OPCODE_FLR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* mul t3, t3, c1.y		; Multiply by 2 */
@@ -632,7 +632,7 @@ static int vlCreateFragmentShaderFieldPMB
 
 	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
 	/* lerp t1, t3, t1, t2		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* add o0, t0, t1		; Add ref and differential to form final output */
@@ -969,7 +969,7 @@ static int vlCreateFragmentShaderFrameBMB
 	}
 
 	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
@@ -1116,7 +1116,7 @@ static int vlCreateFragmentShaderFieldBMB
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* floor t3, t3			; Get rid of fractional part */
-	inst = vl_inst2(TGSI_OPCODE_FLOOR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
+	inst = vl_inst2(TGSI_OPCODE_FLR, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 3);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* mul t3, t3, c1.y		; Multiply by 2 */
@@ -1143,7 +1143,7 @@ static int vlCreateFragmentShaderFieldBMB
 
 	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
 	/* lerp t1, t3, t1, t2		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/*
@@ -1158,11 +1158,11 @@ static int vlCreateFragmentShaderFieldBMB
 
 	/* TODO: Move to conditional tex fetch on t3 instead of lerp */
 	/* lerp t2, t3, t4, t5		; Choose between top and bottom fields based on Y % 2 */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5);
+	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 2, TGSI_FILE_TEMPORARY, 3, TGSI_FILE_TEMPORARY, 4, TGSI_FILE_TEMPORARY, 5);
 	ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
 
 	/* lerp t1, c1.x, t1, t2	; Blend past and future texels */
-	inst = vl_inst4(TGSI_OPCODE_LERP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
+	inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
 	inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.c b/src/gallium/state_trackers/xorg/xorg_composite.c
index c708ac31702..15c955450d0 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.c
+++ b/src/gallium/state_trackers/xorg/xorg_composite.c
@@ -4,6 +4,7 @@
 
 #include "cso_cache/cso_context.h"
 #include "util/u_draw_quad.h"
+#include "util/u_math.h"
 
 #include "pipe/p_inlines.h"
 
@@ -40,9 +41,25 @@ static const struct xorg_composite_blend xorg_blends[] = {
      PIPE_BLENDFACTOR_INV_SRC_ALPHA, PIPE_BLENDFACTOR_INV_SRC_ALPHA },
 };
 
+
 static INLINE void
-pixel_to_float4(PictFormatPtr format,
-                CARD32 pixel, float *color)
+pixel_to_float4(Pixel pixel, float *color)
+{
+   CARD32	    r, g, b, a;
+
+   a = (pixel >> 24) & 0xff;
+   r = (pixel >> 16) & 0xff;
+   g = (pixel >>  8) & 0xff;
+   b = (pixel >>  0) & 0xff;
+   color[0] = ((float)r) / 255.;
+   color[1] = ((float)g) / 255.;
+   color[2] = ((float)b) / 255.;
+   color[3] = ((float)a) / 255.;
+}
+
+static INLINE void
+render_pixel_to_float4(PictFormatPtr format,
+                       CARD32 pixel, float *color)
 {
    CARD32	    r, g, b, a;
 
@@ -148,7 +165,7 @@ setup_vertex_data0(struct exa_context *ctx,
    setup_vertex0(vertices[3], dstX, dstY + height,
                  ctx->solid_color);
 
-   return pipe_user_buffer_create(ctx->ctx->screen,
+   return pipe_user_buffer_create(ctx->pipe->screen,
                                   vertices,
                                   sizeof(vertices));
 }
@@ -194,12 +211,39 @@ setup_vertex_data1(struct exa_context *ctx,
    setup_vertex1(vertices[3], dstX, dstY + height,
                  s0, t1);
 
-   return pipe_user_buffer_create(ctx->ctx->screen,
+   return pipe_user_buffer_create(ctx->pipe->screen,
+                                  vertices,
+                                  sizeof(vertices));
+}
+
+static struct pipe_buffer *
+setup_vertex_data_tex(struct exa_context *ctx,
+                      float x0, float y0, float x1, float y1,
+                      float s0, float t0, float s1, float t1,
+                      float z)
+{
+   float vertices[4][2][4];
+
+   /* 1st vertex */
+   setup_vertex1(vertices[0], x0, y0,
+                 s0, t0);
+   /* 2nd vertex */
+   setup_vertex1(vertices[1], x1, y0,
+                 s1, t0);
+   /* 3rd vertex */
+   setup_vertex1(vertices[2], x1, y1,
+                 s1, t1);
+   /* 4th vertex */
+   setup_vertex1(vertices[3], x0, y1,
+                 s0, t1);
+
+   return pipe_user_buffer_create(ctx->pipe->screen,
                                   vertices,
                                   sizeof(vertices));
 }
 
 
+
 static INLINE void
 setup_vertex2(float vertex[3][4], float x, float y,
               float s0, float t0, float s1, float t1)
@@ -253,7 +297,7 @@ setup_vertex_data2(struct exa_context *ctx,
    setup_vertex2(vertices[3], dstX, dstY + height,
                  st0[0], st0[3], st1[0], st1[3]);
 
-   return pipe_user_buffer_create(ctx->ctx->screen,
+   return pipe_user_buffer_create(ctx->pipe->screen,
                                   vertices,
                                   sizeof(vertices));
 }
@@ -291,16 +335,20 @@ boolean xorg_composite_accelerated(int op,
 }
 
 static void
-bind_framebuffer_state(struct exa_context *exa, PicturePtr pDstPicture,
-                       struct exa_pixmap_priv *pDst)
+bind_clip_state(struct exa_context *exa)
+{
+}
+
+static void
+bind_framebuffer_state(struct exa_context *exa, struct exa_pixmap_priv *pDst)
 {
    unsigned i;
    struct pipe_framebuffer_state state;
    struct pipe_surface *surface = exa_gpu_surface(exa, pDst);
    memset(&state, 0, sizeof(struct pipe_framebuffer_state));
 
-   state.width  = pDstPicture->pDrawable->width;
-   state.height = pDstPicture->pDrawable->height;
+   state.width  = pDst->tex->width[0];
+   state.height = pDst->tex->height[0];
 
    state.nr_cbufs = 1;
    state.cbufs[0] = surface;
@@ -338,10 +386,12 @@ set_viewport(struct exa_context *exa, int width, int height,
 }
 
 static void
-bind_viewport_state(struct exa_context *exa, PicturePtr pDstPicture)
+bind_viewport_state(struct exa_context *exa, struct exa_pixmap_priv *pDst)
 {
-   int width = pDstPicture->pDrawable->width;
-   int height = pDstPicture->pDrawable->height;
+   int width = pDst->tex->width[0];
+   int height = pDst->tex->height[0];
+
+   debug_printf("Bind viewport (%d, %d)\n", width, height);
 
    set_viewport(exa, width, height, Y0_TOP);
 }
@@ -350,7 +400,8 @@ static void
 bind_blend_state(struct exa_context *exa, int op,
                  PicturePtr pSrcPicture, PicturePtr pMaskPicture)
 {
-   boolean component_alpha = pSrcPicture->componentAlpha;
+   boolean component_alpha = (pSrcPicture) ?
+                             pSrcPicture->componentAlpha : FALSE;
    struct xorg_composite_blend blend_opt;
    struct pipe_blend_state blend;
 
@@ -390,14 +441,17 @@ bind_shaders(struct exa_context *exa, int op,
    unsigned vs_traits = 0, fs_traits = 0;
    struct xorg_shader shader;
 
+   exa->has_solid_color = FALSE;
+
    if (pSrcPicture) {
       if (pSrcPicture->pSourcePict) {
          if (pSrcPicture->pSourcePict->type == SourcePictTypeSolidFill) {
             fs_traits |= FS_SOLID_FILL;
             vs_traits |= VS_SOLID_FILL;
-            pixel_to_float4(pSrcPicture->pFormat,
-                            pSrcPicture->pSourcePict->solidFill.color,
-                            exa->solid_color);
+            render_pixel_to_float4(pSrcPicture->pFormat,
+                                   pSrcPicture->pSourcePict->solidFill.color,
+                                   exa->solid_color);
+            exa->has_solid_color = TRUE;
          } else {
             debug_assert("!gradients not supported");
          }
@@ -478,15 +532,15 @@ setup_vs_constant_buffer(struct exa_context *exa,
    struct pipe_constant_buffer *cbuf = &exa->vs_const_buffer;
 
    pipe_buffer_reference(&cbuf->buffer, NULL);
-   cbuf->buffer = pipe_buffer_create(exa->ctx->screen, 16,
+   cbuf->buffer = pipe_buffer_create(exa->pipe->screen, 16,
                                      PIPE_BUFFER_USAGE_CONSTANT,
                                      param_bytes);
 
    if (cbuf->buffer) {
-      pipe_buffer_write(exa->ctx->screen, cbuf->buffer,
+      pipe_buffer_write(exa->pipe->screen, cbuf->buffer,
                         0, param_bytes, vs_consts);
    }
-   exa->ctx->set_constant_buffer(exa->ctx, PIPE_SHADER_VERTEX, 0, cbuf);
+   exa->pipe->set_constant_buffer(exa->pipe, PIPE_SHADER_VERTEX, 0, cbuf);
 }
 
 
@@ -500,22 +554,22 @@ setup_fs_constant_buffer(struct exa_context *exa)
    struct pipe_constant_buffer *cbuf = &exa->fs_const_buffer;
 
    pipe_buffer_reference(&cbuf->buffer, NULL);
-   cbuf->buffer = pipe_buffer_create(exa->ctx->screen, 16,
+   cbuf->buffer = pipe_buffer_create(exa->pipe->screen, 16,
                                      PIPE_BUFFER_USAGE_CONSTANT,
                                      param_bytes);
 
    if (cbuf->buffer) {
-      pipe_buffer_write(exa->ctx->screen, cbuf->buffer,
+      pipe_buffer_write(exa->pipe->screen, cbuf->buffer,
                         0, param_bytes, fs_consts);
    }
-   exa->ctx->set_constant_buffer(exa->ctx, PIPE_SHADER_FRAGMENT, 0, cbuf);
+   exa->pipe->set_constant_buffer(exa->pipe, PIPE_SHADER_FRAGMENT, 0, cbuf);
 }
 
 static void
-setup_constant_buffers(struct exa_context *exa, PicturePtr pDstPicture)
+setup_constant_buffers(struct exa_context *exa, struct exa_pixmap_priv *pDst)
 {
-   int width = pDstPicture->pDrawable->width;
-   int height = pDstPicture->pDrawable->height;
+   int width = pDst->tex->width[0];
+   int height = pDst->tex->height[0];
 
    setup_vs_constant_buffer(exa, width, height);
    setup_fs_constant_buffer(exa);
@@ -530,15 +584,15 @@ boolean xorg_composite_bind_state(struct exa_context *exa,
                                   struct exa_pixmap_priv *pMask,
                                   struct exa_pixmap_priv *pDst)
 {
-   bind_framebuffer_state(exa, pDstPicture, pDst);
-   bind_viewport_state(exa, pDstPicture);
+   bind_framebuffer_state(exa, pDst);
+   bind_viewport_state(exa, pDst);
    bind_blend_state(exa, op, pSrcPicture, pMaskPicture);
    bind_rasterizer_state(exa);
    bind_shaders(exa, op, pSrcPicture, pMaskPicture);
    bind_samplers(exa, op, pSrcPicture, pMaskPicture,
                  pDstPicture, pSrc, pMask, pDst);
-
-   setup_constant_buffers(exa, pDstPicture);
+   bind_clip_state(exa);
+   setup_constant_buffers(exa, pDst);
 
    return FALSE;
 }
@@ -548,7 +602,7 @@ void xorg_composite(struct exa_context *exa,
                     int srcX, int srcY, int maskX, int maskY,
                     int dstX, int dstY, int width, int height)
 {
-   struct pipe_context *pipe = exa->ctx;
+   struct pipe_context *pipe = exa->pipe;
    struct pipe_buffer *buf = 0;
 
    if (exa->num_bound_samplers == 0 ) { /* solid fill */
@@ -573,12 +627,393 @@ void xorg_composite(struct exa_context *exa,
    }
 
    if (buf) {
+      int num_attribs = 1; /*pos*/
+      num_attribs += exa->num_bound_samplers;
+      if (exa->has_solid_color)
+         ++num_attribs;
+
       util_draw_vertex_buffer(pipe, buf, 0,
                               PIPE_PRIM_TRIANGLE_FAN,
                               4,  /* verts */
-                              1 + exa->num_bound_samplers); /* attribs/vert */
+                              num_attribs); /* attribs/vert */
 
       pipe_buffer_reference(&buf, NULL);
    }
 }
 
+boolean xorg_solid_bind_state(struct exa_context *exa,
+                              struct exa_pixmap_priv *pixmap,
+                              Pixel fg)
+{
+   unsigned vs_traits, fs_traits;
+   struct xorg_shader shader;
+
+   pixel_to_float4(fg, exa->solid_color);
+   exa->has_solid_color = TRUE;
+
+   exa->solid_color[3] = 1.f;
+
+   debug_printf("Color Pixel=(%d, %d, %d, %d), RGBA=(%f, %f, %f, %f)\n",
+                (fg >> 24) & 0xff, (fg >> 16) & 0xff,
+                (fg >> 8) & 0xff,  (fg >> 0) & 0xff,
+                exa->solid_color[0], exa->solid_color[1],
+                exa->solid_color[2], exa->solid_color[3]);
+
+#if 0
+   exa->solid_color[0] = 1.f;
+   exa->solid_color[1] = 0.f;
+   exa->solid_color[2] = 0.f;
+   exa->solid_color[3] = 1.f;
+#endif
+
+   vs_traits = VS_SOLID_FILL;
+   fs_traits = FS_SOLID_FILL;
+
+   bind_framebuffer_state(exa, pixmap);
+   bind_viewport_state(exa, pixmap);
+   bind_rasterizer_state(exa);
+   bind_blend_state(exa, PictOpSrc, NULL, NULL);
+   setup_constant_buffers(exa, pixmap);
+   bind_clip_state(exa);
+
+   shader = xorg_shaders_get(exa->shaders, vs_traits, fs_traits);
+   cso_set_vertex_shader_handle(exa->cso, shader.vs);
+   cso_set_fragment_shader_handle(exa->cso, shader.fs);
+
+   return FALSE;
+}
+
+void xorg_solid(struct exa_context *exa,
+                struct exa_pixmap_priv *pixmap,
+                int x0, int y0, int x1, int y1)
+{
+   struct pipe_context *pipe = exa->pipe;
+   struct pipe_buffer *buf = 0;
+   float vertices[4][2][4];
+
+   x0 = 10; y0 = 10;
+   x1 = 300; y1 = 300;
+
+   /* 1st vertex */
+   setup_vertex0(vertices[0], x0, y0,
+                 exa->solid_color);
+   /* 2nd vertex */
+   setup_vertex0(vertices[1], x1, y0,
+                 exa->solid_color);
+   /* 3rd vertex */
+   setup_vertex0(vertices[2], x1, y1,
+                 exa->solid_color);
+   /* 4th vertex */
+   setup_vertex0(vertices[3], x0, y1,
+                 exa->solid_color);
+
+   buf = pipe_user_buffer_create(exa->pipe->screen,
+                                 vertices,
+                                 sizeof(vertices));
+
+
+   if (buf) {
+      debug_printf("Drawing buf is %p\n", buf);
+      util_draw_vertex_buffer(pipe, buf, 0,
+                              PIPE_PRIM_TRIANGLE_FAN,
+                              4,  /* verts */
+                              2); /* attribs/vert */
+
+      pipe_buffer_reference(&buf, NULL);
+   }
+}
+
+
+static INLINE void shift_rectx(float coords[4],
+                               const float *bounds,
+                               const float shift)
+{
+   coords[0] += shift;
+   coords[2] -= shift;
+   if (bounds) {
+      coords[2] = MIN2(coords[2], bounds[2]);
+      /* bound x/y + width/height */
+      if ((coords[0] + coords[2]) > (bounds[0] + bounds[2])) {
+         coords[2] = (bounds[0] + bounds[2]) - coords[0];
+      }
+   }
+}
+
+static INLINE void shift_recty(float coords[4],
+                               const float *bounds,
+                               const float shift)
+{
+   coords[1] += shift;
+   coords[3] -= shift;
+   if (bounds) {
+      coords[3] = MIN2(coords[3], bounds[3]);
+      if ((coords[1] + coords[3]) > (bounds[1] + bounds[3])) {
+         coords[3] = (bounds[1] + bounds[3]) - coords[1];
+      }
+   }
+}
+
+static INLINE void bound_rect(float coords[4],
+                              const float bounds[4],
+                              float shift[4])
+{
+   /* if outside the bounds */
+   if (coords[0] > (bounds[0] + bounds[2]) ||
+       coords[1] > (bounds[1] + bounds[3]) ||
+       (coords[0] + coords[2]) < bounds[0] ||
+       (coords[1] + coords[3]) < bounds[1]) {
+      coords[0] = 0.f;
+      coords[1] = 0.f;
+      coords[2] = 0.f;
+      coords[3] = 0.f;
+      shift[0] = 0.f;
+      shift[1] = 0.f;
+      return;
+   }
+
+   /* bound x */
+   if (coords[0] < bounds[0]) {
+      shift[0] = bounds[0] - coords[0];
+      coords[2] -= shift[0];
+      coords[0] = bounds[0];
+   } else
+      shift[0] = 0.f;
+
+   /* bound y */
+   if (coords[1] < bounds[1]) {
+      shift[1] = bounds[1] - coords[1];
+      coords[3] -= shift[1];
+      coords[1] = bounds[1];
+   } else
+      shift[1] = 0.f;
+
+   shift[2] = bounds[2] - coords[2];
+   shift[3] = bounds[3] - coords[3];
+   /* bound width/height */
+   coords[2] = MIN2(coords[2], bounds[2]);
+   coords[3] = MIN2(coords[3], bounds[3]);
+
+   /* bound x/y + width/height */
+   if ((coords[0] + coords[2]) > (bounds[0] + bounds[2])) {
+      coords[2] = (bounds[0] + bounds[2]) - coords[0];
+   }
+   if ((coords[1] + coords[3]) > (bounds[1] + bounds[3])) {
+      coords[3] = (bounds[1] + bounds[3]) - coords[1];
+   }
+
+   /* if outside the bounds */
+   if ((coords[0] + coords[2]) < bounds[0] ||
+       (coords[1] + coords[3]) < bounds[1]) {
+      coords[0] = 0.f;
+      coords[1] = 0.f;
+      coords[2] = 0.f;
+      coords[3] = 0.f;
+      return;
+   }
+}
+
+static INLINE void sync_size(float *src_loc, float *dst_loc)
+{
+   src_loc[2] = MIN2(src_loc[2], dst_loc[2]);
+   src_loc[3] = MIN2(src_loc[3], dst_loc[3]);
+   dst_loc[2] = src_loc[2];
+   dst_loc[3] = src_loc[3];
+}
+
+
+static void renderer_copy_texture(struct exa_context *exa,
+                                  struct pipe_texture *src,
+                                  float sx1, float sy1,
+                                  float sx2, float sy2,
+                                  struct pipe_texture *dst,
+                                  float dx1, float dy1,
+                                  float dx2, float dy2)
+{
+   struct pipe_context *pipe = exa->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_buffer *buf;
+   struct pipe_surface *dst_surf = screen->get_tex_surface(
+      screen, dst, 0, 0, 0,
+      PIPE_BUFFER_USAGE_GPU_WRITE);
+   struct pipe_framebuffer_state fb;
+   float s0, t0, s1, t1;
+   struct xorg_shader shader;
+
+   assert(src->width[0] != 0);
+   assert(src->height[0] != 0);
+   assert(dst->width[0] != 0);
+   assert(dst->height[0] != 0);
+
+#if 0
+   debug_printf("copy texture [%f, %f, %f, %f], [%f, %f, %f, %f]\n",
+                sx1, sy1, sx2, sy2, dx1, dy1, dx2, dy2);
+#endif
+
+#if 1
+   s0 = sx1 / src->width[0];
+   s1 = sx2 / src->width[0];
+   t0 = sy1 / src->height[0];
+   t1 = sy2 / src->height[0];
+#else
+   s0 = 0;
+   s1 = 1;
+   t0 = 0;
+   t1 = 1;
+#endif
+
+   assert(screen->is_format_supported(screen, dst_surf->format, PIPE_TEXTURE_2D,
+                                      PIPE_TEXTURE_USAGE_RENDER_TARGET, 0));
+
+   /* save state (restored below) */
+   cso_save_blend(exa->cso);
+   cso_save_samplers(exa->cso);
+   cso_save_sampler_textures(exa->cso);
+   cso_save_framebuffer(exa->cso);
+   cso_save_fragment_shader(exa->cso);
+   cso_save_vertex_shader(exa->cso);
+
+   cso_save_viewport(exa->cso);
+
+
+   /* set misc state we care about */
+   {
+      struct pipe_blend_state blend;
+      memset(&blend, 0, sizeof(blend));
+      blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+      blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+      blend.colormask = PIPE_MASK_RGBA;
+      cso_set_blend(exa->cso, &blend);
+   }
+
+   /* sampler */
+   {
+      struct pipe_sampler_state sampler;
+      memset(&sampler, 0, sizeof(sampler));
+      sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+      sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+      sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+      sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+      sampler.normalized_coords = 1;
+      cso_single_sampler(exa->cso, 0, &sampler);
+      cso_single_sampler_done(exa->cso);
+   }
+
+   set_viewport(exa, dst_surf->width, dst_surf->height, Y0_TOP);
+
+   /* texture */
+   cso_set_sampler_textures(exa->cso, 1, &src);
+
+   /* shaders */
+   shader = xorg_shaders_get(exa->shaders,
+                             VS_COMPOSITE,
+                             FS_COMPOSITE);
+   cso_set_vertex_shader_handle(exa->cso, shader.vs);
+   cso_set_fragment_shader_handle(exa->cso, shader.fs);
+
+   /* drawing dest */
+   memset(&fb, 0, sizeof(fb));
+   fb.width = dst_surf->width;
+   fb.height = dst_surf->height;
+   fb.nr_cbufs = 1;
+   fb.cbufs[0] = dst_surf;
+   {
+      int i;
+      for (i = 1; i < PIPE_MAX_COLOR_BUFS; ++i)
+         fb.cbufs[i] = 0;
+   }
+   cso_set_framebuffer(exa->cso, &fb);
+
+   /* draw quad */
+   buf = setup_vertex_data_tex(exa,
+                               dx1, dy1,
+                               dx2, dy2,
+                               s0, t0, s1, t1,
+                               0.0f);
+
+   if (buf) {
+      util_draw_vertex_buffer(exa->pipe, buf, 0,
+                              PIPE_PRIM_TRIANGLE_FAN,
+                              4,  /* verts */
+                              2); /* attribs/vert */
+
+      pipe_buffer_reference(&buf, NULL);
+   }
+
+   /* restore state we changed */
+   cso_restore_blend(exa->cso);
+   cso_restore_samplers(exa->cso);
+   cso_restore_sampler_textures(exa->cso);
+   cso_restore_framebuffer(exa->cso);
+   cso_restore_vertex_shader(exa->cso);
+   cso_restore_fragment_shader(exa->cso);
+   cso_restore_viewport(exa->cso);
+
+   pipe_surface_reference(&dst_surf, NULL);
+}
+
+void xorg_copy_pixmap(struct exa_context *ctx,
+                      struct exa_pixmap_priv *dst_priv, int dx, int dy,
+                      struct exa_pixmap_priv *src_priv, int sx, int sy,
+                      int width, int height)
+{
+   float dst_loc[4], src_loc[4];
+   float dst_bounds[4], src_bounds[4];
+   float src_shift[4], dst_shift[4], shift[4];
+   struct pipe_texture *dst = dst_priv->tex;
+   struct pipe_texture *src = src_priv->tex;
+
+   dst_loc[0] = dx;
+   dst_loc[1] = dy;
+   dst_loc[2] = width;
+   dst_loc[3] = height;
+   dst_bounds[0] = 0.f;
+   dst_bounds[1] = 0.f;
+   dst_bounds[2] = dst->width[0];
+   dst_bounds[3] = dst->height[0];
+
+   src_loc[0] = sx;
+   src_loc[1] = sy;
+   src_loc[2] = width;
+   src_loc[3] = height;
+   src_bounds[0] = 0.f;
+   src_bounds[1] = 0.f;
+   src_bounds[2] = src->width[0];
+   src_bounds[3] = src->height[0];
+
+   bound_rect(src_loc, src_bounds, src_shift);
+   bound_rect(dst_loc, dst_bounds, dst_shift);
+   shift[0] = src_shift[0] - dst_shift[0];
+   shift[1] = src_shift[1] - dst_shift[1];
+
+   if (shift[0] < 0)
+      shift_rectx(src_loc, src_bounds, -shift[0]);
+   else
+      shift_rectx(dst_loc, dst_bounds, shift[0]);
+
+   if (shift[1] < 0)
+      shift_recty(src_loc, src_bounds, -shift[1]);
+   else
+      shift_recty(dst_loc, dst_bounds, shift[1]);
+
+   sync_size(src_loc, dst_loc);
+
+   if (src_loc[2] >= 0 && src_loc[3] >= 0 &&
+       dst_loc[2] >= 0 && dst_loc[3] >= 0) {
+      renderer_copy_texture(ctx,
+                            src,
+                            src_loc[0],
+                            src_loc[1] + src_loc[3],
+                            src_loc[0] + src_loc[2],
+                            src_loc[1],
+                            dst,
+                            dst_loc[0],
+                            dst_loc[1] + dst_loc[3],
+                            dst_loc[0] + dst_loc[2],
+                            dst_loc[1]);
+   }
+}
+
diff --git a/src/gallium/state_trackers/xorg/xorg_composite.h b/src/gallium/state_trackers/xorg/xorg_composite.h
index 17dfcb199ea..e73f1c704a8 100644
--- a/src/gallium/state_trackers/xorg/xorg_composite.h
+++ b/src/gallium/state_trackers/xorg/xorg_composite.h
@@ -22,4 +22,16 @@ void xorg_composite(struct exa_context *exa,
                     int srcX, int srcY, int maskX, int maskY,
                     int dstX, int dstY, int width, int height);
 
+boolean xorg_solid_bind_state(struct exa_context *exa,
+                              struct exa_pixmap_priv *pixmap,
+                              Pixel fg);
+void xorg_solid(struct exa_context *exa,
+                struct exa_pixmap_priv *pixmap,
+                int x0, int y0, int x1, int y1);
+
+void xorg_copy_pixmap(struct exa_context *ctx,
+                      struct exa_pixmap_priv *dst, int dx, int dy,
+                      struct exa_pixmap_priv *src, int sx, int sy,
+                      int width, int height);
+
 #endif
diff --git a/src/gallium/state_trackers/xorg/xorg_dri2.c b/src/gallium/state_trackers/xorg/xorg_dri2.c
index 6431a0fe254..8a362596c75 100644
--- a/src/gallium/state_trackers/xorg/xorg_dri2.c
+++ b/src/gallium/state_trackers/xorg/xorg_dri2.c
@@ -118,6 +118,7 @@ driDoCreateBuffer(DrawablePtr pDraw, DRI2BufferPtr buffer, unsigned int format)
     }
 
     if (!tex) {
+	exaMoveInPixmap(private->pPixmap);
 	xorg_exa_set_shared_usage(private->pPixmap);
 	pScreen->ModifyPixmapHeader(private->pPixmap, 0, 0, 0, 0, 0, NULL);
 	tex = xorg_exa_get_texture(private->pPixmap);
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.c b/src/gallium/state_trackers/xorg/xorg_exa.c
index a17a71f23a8..312dab1544c 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa.c
@@ -82,6 +82,25 @@ exa_get_pipe_format(int depth, enum pipe_format *format, int *bbp)
     }
 }
 
+static void
+xorg_exa_init_state(struct exa_context *exa)
+{
+   struct pipe_depth_stencil_alpha_state dsa;
+
+   /* set common initial clip state */
+   memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
+   cso_set_depth_stencil_alpha(exa->cso, &dsa);
+}
+
+static void
+xorg_exa_common_done(struct exa_context *exa)
+{
+   exa->copy.src = NULL;
+   exa->copy.dst = NULL;
+   exa->has_solid_color = FALSE;
+   exa->num_bound_samplers = 0;
+}
+
 /*
  * Static exported EXA functions
  */
@@ -111,9 +130,9 @@ ExaDownloadFromScreen(PixmapPtr pPix, int x,  int y, int w,  int h, char *dst,
     if (!priv || !priv->tex)
 	return FALSE;
 
-    if (exa->ctx->is_texture_referenced(exa->ctx, priv->tex, 0, 0) &
+    if (exa->pipe->is_texture_referenced(exa->pipe, priv->tex, 0, 0) &
 	PIPE_REFERENCED_FOR_WRITE)
-	exa->ctx->flush(exa->ctx, 0, NULL);
+	exa->pipe->flush(exa->pipe, 0, NULL);
 
     transfer = exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0,
 					   PIPE_TRANSFER_READ, x, y, w, h);
@@ -178,9 +197,9 @@ ExaPrepareAccess(PixmapPtr pPix, int index)
 
     if (priv->map_count++ == 0)
     {
-	if (exa->ctx->is_texture_referenced(exa->ctx, priv->tex, 0, 0) &
+	if (exa->pipe->is_texture_referenced(exa->pipe, priv->tex, 0, 0) &
 	    PIPE_REFERENCED_FOR_WRITE)
-	    exa->ctx->flush(exa->ctx, 0, NULL);
+	    exa->pipe->flush(exa->pipe, 0, NULL);
 
 	priv->map_transfer =
 	    exa->scrn->get_tex_transfer(exa->scrn, priv->tex, 0, 0, 0,
@@ -231,15 +250,22 @@ ExaDone(PixmapPtr pPixmap)
     if (!priv)
 	return;
 
-    if (priv->src_surf)
-	exa->scrn->tex_surface_destroy(priv->src_surf);
-    priv->src_surf = NULL;
+#if 1
+    xorg_exa_flush(exa, PIPE_FLUSH_RENDER_CACHE, NULL);
+#else
+    xorg_finish(exa);
+#endif
+    xorg_exa_common_done(exa);
 }
 
 static void
 ExaDoneComposite(PixmapPtr pPixmap)
 {
+   ScrnInfoPtr pScrn = xf86Screens[pPixmap->drawable.pScreen->myNum];
+   modesettingPtr ms = modesettingPTR(pScrn);
+   struct exa_context *exa = ms->exa;
 
+   xorg_exa_common_done(exa);
 }
 
 static Bool
@@ -250,6 +276,7 @@ ExaPrepareSolid(PixmapPtr pPixmap, int alu, Pixel planeMask, Pixel fg)
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap);
     struct exa_context *exa = ms->exa;
 
+    debug_printf("ExaPrepareSolid - test\n");
     if (pPixmap->drawable.depth < 15)
 	return FALSE;
 
@@ -262,12 +289,11 @@ ExaPrepareSolid(PixmapPtr pPixmap, int alu, Pixel planeMask, Pixel fg)
     if (alu != GXcopy)
 	return FALSE;
 
-    if (!exa->ctx || !exa->ctx->surface_fill)
+    if (!exa->pipe)
 	return FALSE;
 
-    priv->color = fg;
-
-    return TRUE;
+    debug_printf("  ExaPrepareSolid(0x%x)\n", fg);
+    return xorg_solid_bind_state(exa, priv, fg);
 }
 
 static void
@@ -277,12 +303,18 @@ ExaSolid(PixmapPtr pPixmap, int x0, int y0, int x1, int y1)
     modesettingPtr ms = modesettingPTR(pScrn);
     struct exa_context *exa = ms->exa;
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pPixmap);
-    struct pipe_surface *surf = exa_gpu_surface(exa, priv);
 
-    exa->ctx->surface_fill(exa->ctx, surf, x0, y0, x1 - x0, y1 - y0,
-			   priv->color);
+    debug_printf("\tExaSolid(%d, %d, %d, %d)\n", x0, y0, x1, y1);
 
-    exa->scrn->tex_surface_destroy(surf);
+#if 0
+    if (x0 == 0 && y0 == 0 &&
+        x1 == priv->tex->width[0] &&
+        y1 == priv->tex->height[0]) {
+       exa->ctx->clear(exa->ctx, PIPE_CLEAR_COLOR | PIPE_CLEAR_DEPTHSTENCIL,
+                       exa->solid_color, 1., 0);
+    } else
+#endif
+       xorg_solid(exa, priv, x0, y0, x1, y1) ;
 }
 
 static Bool
@@ -295,6 +327,8 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
     struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap);
     struct exa_pixmap_priv *src_priv = exaGetPixmapDriverPrivate(pSrcPixmap);
 
+    debug_printf("ExaPrepareCopy\n");
+
     if (alu != GXcopy)
 	return FALSE;
 
@@ -310,27 +344,33 @@ ExaPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int xdir,
     if (!priv->tex || !src_priv->tex)
 	return FALSE;
 
-    if (!exa->ctx || !exa->ctx->surface_copy)
+    if (!exa->pipe)
 	return FALSE;
 
-    priv->src_surf = exa_gpu_surface(exa, src_priv);
+    exa->copy.src = src_priv;
+    exa->copy.dst = priv;
 
-    return TRUE;
+    /*XXX disabled until some issues with syncing are fixed */
+    return FALSE;
 }
 
 static void
 ExaCopy(PixmapPtr pDstPixmap, int srcX, int srcY, int dstX, int dstY,
 	int width, int height)
 {
-    ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum];
-    modesettingPtr ms = modesettingPTR(pScrn);
-    struct exa_context *exa = ms->exa;
-    struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap);
-    struct pipe_surface *surf = exa_gpu_surface(exa, priv);
+   ScrnInfoPtr pScrn = xf86Screens[pDstPixmap->drawable.pScreen->myNum];
+   modesettingPtr ms = modesettingPTR(pScrn);
+   struct exa_context *exa = ms->exa;
+   struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDstPixmap);
+
+   debug_printf("\tExaCopy(srcx=%d, srcy=%d, dstX=%d, dstY=%d, w=%d, h=%d)\n",
+                srcX, srcY, dstX, dstY, width, height);
 
-    exa->ctx->surface_copy(exa->ctx, surf, dstX, dstY, priv->src_surf,
-			   srcX, srcY, width, height);
-    exa->scrn->tex_surface_destroy(surf);
+   debug_assert(priv == exa->copy.dst);
+
+   xorg_copy_pixmap(exa, exa->copy.dst, dstX, dstY,
+                    exa->copy.src, srcX, srcY,
+                    width, height);
 }
 
 static Bool
@@ -342,6 +382,8 @@ ExaPrepareComposite(int op, PicturePtr pSrcPicture,
    modesettingPtr ms = modesettingPTR(pScrn);
    struct exa_context *exa = ms->exa;
 
+   debug_printf("ExaPrepareComposite\n");
+
    return xorg_composite_bind_state(exa, op, pSrcPicture, pMaskPicture,
                                     pDstPicture,
                                     exaGetPixmapDriverPrivate(pSrc),
@@ -358,6 +400,8 @@ ExaComposite(PixmapPtr pDst, int srcX, int srcY, int maskX, int maskY,
    struct exa_context *exa = ms->exa;
    struct exa_pixmap_priv *priv = exaGetPixmapDriverPrivate(pDst);
 
+   debug_printf("\tExaComposite\n");
+
    xorg_composite(exa, priv, srcX, srcY, maskX, maskY,
                   dstX, dstY, width, height);
 }
@@ -537,16 +581,16 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height,
 
 	    if (priv->tex) {
 		struct pipe_surface *dst_surf;
+                struct pipe_surface *src_surf;
 
 		dst_surf = exa->scrn->get_tex_surface(exa->scrn, texture, 0, 0, 0,
 						      PIPE_BUFFER_USAGE_GPU_WRITE);
-		priv->src_surf = exa_gpu_surface(exa, priv);
-		exa->ctx->surface_copy(exa->ctx, dst_surf, 0, 0, priv->src_surf,
-				       0, 0, min(width, texture->width[0]),
-				       min(height, texture->height[0]));
+		src_surf = exa_gpu_surface(exa, priv);
+		exa->pipe->surface_copy(exa->pipe, dst_surf, 0, 0, src_surf,
+                                        0, 0, min(width, texture->width[0]),
+                                        min(height, texture->height[0]));
 		exa->scrn->tex_surface_destroy(dst_surf);
-		exa->scrn->tex_surface_destroy(priv->src_surf);
-		priv->src_surf = NULL;
+		exa->scrn->tex_surface_destroy(src_surf);
 	    } else if (pPixmap->devPrivate.ptr) {
 		struct pipe_transfer *transfer;
 
@@ -563,6 +607,9 @@ ExaModifyPixmapHeader(PixmapPtr pPixmap, int width, int height,
 			       pPixmap->devKind, 0, 0);
 		exa->scrn->transfer_unmap(exa->scrn, transfer);
 		exa->scrn->tex_transfer_destroy(transfer);
+
+		xfree(pPixmap->devPrivate.ptr);
+		pPixmap->devPrivate.ptr = NULL;
 	    }
 	}
 #ifdef DRM_MODE_FEATURE_DIRTYFB
@@ -611,8 +658,8 @@ xorg_exa_close(ScrnInfoPtr pScrn)
       cso_destroy_context(exa->cso);
    }
 
-   if (exa->ctx)
-      exa->ctx->destroy(exa->ctx);
+   if (exa->pipe)
+      exa->pipe->destroy(exa->pipe);
 
    exaDriverFini(pScrn->pScreen);
    xfree(exa);
@@ -645,6 +692,12 @@ xorg_exa_init(ScrnInfoPtr pScrn)
    pExa->pixmapOffsetAlign = 0;
    pExa->pixmapPitchAlign  = 1;
    pExa->flags             = EXA_OFFSCREEN_PIXMAPS | EXA_HANDLES_PIXMAPS;
+#ifdef EXA_SUPPORTS_PREPARE_AUX
+   pExa->flags            |= EXA_SUPPORTS_PREPARE_AUX;
+#endif
+#ifdef EXA_MIXED_PIXMAPS
+   pExa->flags            |= EXA_MIXED_PIXMAPS;
+#endif
    pExa->maxX              = 8191; /* FIXME */
    pExa->maxY              = 8191; /* FIXME */
 
@@ -674,13 +727,15 @@ xorg_exa_init(ScrnInfoPtr pScrn)
    }
 
    exa->scrn = ms->screen;
-   exa->ctx = ms->api->create_context(ms->api, exa->scrn);
+   exa->pipe = ms->api->create_context(ms->api, exa->scrn);
    /* Share context with DRI */
-   ms->ctx = exa->ctx;
+   ms->ctx = exa->pipe;
 
-   exa->cso = cso_create_context(exa->ctx);
+   exa->cso = cso_create_context(exa->pipe);
    exa->shaders = xorg_shaders_create(exa);
 
+   xorg_exa_init_state(exa);
+
    return (void *)exa;
 
 out_err:
@@ -698,3 +753,19 @@ exa_gpu_surface(struct exa_context *exa, struct exa_pixmap_priv *priv)
 
 }
 
+void xorg_exa_flush(struct exa_context *exa, uint pipeFlushFlags,
+                    struct pipe_fence_handle **fence)
+{
+   exa->pipe->flush(exa->pipe, pipeFlushFlags, fence);
+}
+
+void xorg_exa_finish(struct exa_context *exa)
+{
+   struct pipe_fence_handle *fence = NULL;
+
+   xorg_exa_flush(exa, PIPE_FLUSH_RENDER_CACHE, &fence);
+
+   exa->pipe->screen->fence_finish(exa->pipe->screen, fence, 0);
+   exa->pipe->screen->fence_reference(exa->pipe->screen, &fence, NULL);
+}
+
diff --git a/src/gallium/state_trackers/xorg/xorg_exa.h b/src/gallium/state_trackers/xorg/xorg_exa.h
index 5b515be1397..43949b04a44 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa.h
+++ b/src/gallium/state_trackers/xorg/xorg_exa.h
@@ -14,7 +14,7 @@ struct xorg_shaders;
 struct exa_context
 {
    ExaDriverPtr pExa;
-   struct pipe_context *ctx;
+   struct pipe_context *pipe;
    struct pipe_screen *scrn;
    struct cso_context *cso;
    struct xorg_shaders *shaders;
@@ -26,6 +26,12 @@ struct exa_context
    int num_bound_samplers;
 
    float solid_color[4];
+   boolean has_solid_color;
+
+   struct {
+      struct exa_pixmap_priv *src;
+      struct exa_pixmap_priv *dst;
+   } copy;
 };
 
 
@@ -36,8 +42,6 @@ struct exa_pixmap_priv
 
    struct pipe_texture *tex;
    struct pipe_texture *depth_stencil_tex;
-   unsigned int color;
-   struct pipe_surface *src_surf; /* for copies */
 
    struct pipe_transfer *map_transfer;
    unsigned map_count;
@@ -46,5 +50,8 @@ struct exa_pixmap_priv
 struct pipe_surface *
 exa_gpu_surface(struct exa_context *exa, struct exa_pixmap_priv *priv);
 
+void xorg_exa_flush(struct exa_context *exa, uint pipeFlushFlags,
+                    struct pipe_fence_handle **fence);
+void xorg_exa_finish(struct exa_context *exa);
 
 #endif
diff --git a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
index cfee10c3b30..b30cbff4793 100644
--- a/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
+++ b/src/gallium/state_trackers/xorg/xorg_exa_tgsi.c
@@ -241,41 +241,38 @@ create_vs(struct pipe_context *pipe,
    boolean is_fill = vs_traits & VS_FILL;
    boolean is_composite = vs_traits & VS_COMPOSITE;
    boolean has_mask = vs_traits & VS_MASK;
+   unsigned input_slot = 0;
 
    ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
    if (ureg == NULL)
       return 0;
 
-   const0 = ureg_DECL_constant(ureg);
-   const1 = ureg_DECL_constant(ureg);
+   const0 = ureg_DECL_constant(ureg, 0);
+   const1 = ureg_DECL_constant(ureg, 1);
 
    /* it has to be either a fill or a composite op */
    debug_assert(is_fill ^ is_composite);
 
-   src = ureg_DECL_vs_input(ureg,
-                            TGSI_SEMANTIC_POSITION, 0);
+   src = ureg_DECL_vs_input(ureg, input_slot++);
    dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
    src = vs_normalize_coords(ureg, src,
                              const0, const1);
    ureg_MOV(ureg, dst, src);
 
-
    if (is_composite) {
-      src = ureg_DECL_vs_input(ureg,
-                               TGSI_SEMANTIC_GENERIC, 1);
+      src = ureg_DECL_vs_input(ureg, input_slot++);
       dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, 1);
       ureg_MOV(ureg, dst, src);
    }
+
    if (is_fill) {
-      src = ureg_DECL_vs_input(ureg,
-                               TGSI_SEMANTIC_COLOR, 1);
-      dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1);
+      src = ureg_DECL_vs_input(ureg, input_slot++);
+      dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
       ureg_MOV(ureg, dst, src);
    }
 
    if (has_mask) {
-      src = ureg_DECL_vs_input(ureg,
-                               TGSI_SEMANTIC_GENERIC, 2);
+      src = ureg_DECL_vs_input(ureg, input_slot++);
       dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 2);
       ureg_MOV(ureg, dst, src);
    }
@@ -318,8 +315,8 @@ create_fs(struct pipe_context *pipe,
                                      TGSI_SEMANTIC_POSITION,
                                      0,
                                      TGSI_INTERPOLATE_PERSPECTIVE);
-   }
-   if (is_fill) {
+   } else {
+      debug_assert(is_fill);
       if (is_solid)
          src_input = ureg_DECL_fs_input(ureg,
                                         TGSI_SEMANTIC_COLOR,
@@ -473,9 +470,9 @@ struct xorg_shader xorg_shaders_get(struct xorg_shaders *sc,
    struct xorg_shader shader = {0};
    void *vs, *fs;
 
-   vs = shader_from_cache(sc->exa->ctx, PIPE_SHADER_VERTEX,
+   vs = shader_from_cache(sc->exa->pipe, PIPE_SHADER_VERTEX,
                           sc->vs_hash, vs_traits);
-   fs = shader_from_cache(sc->exa->ctx, PIPE_SHADER_FRAGMENT,
+   fs = shader_from_cache(sc->exa->pipe, PIPE_SHADER_FRAGMENT,
                           sc->fs_hash, fs_traits);
 
    debug_assert(vs && fs);
diff --git a/src/gallium/winsys/drm/intel/dri/SConscript b/src/gallium/winsys/drm/intel/dri/SConscript
index 6c00861f517..f9738110723 100644
--- a/src/gallium/winsys/drm/intel/dri/SConscript
+++ b/src/gallium/winsys/drm/intel/dri/SConscript
@@ -12,8 +12,9 @@ drivers = [
     trace,
 ]
 
-env.SharedLibrary(
+env.LoadableModule(
     target ='i915_dri.so',
     source = COMMON_GALLIUM_SOURCES,
     LIBS = drivers + mesa + auxiliaries + env['LIBS'],
+    SHLIBPREFIX = '',
 )
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c
index 5ca3ad9762d..ebd1b607b78 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_batchbuffer.c
@@ -13,6 +13,7 @@
 #define INTEL_BATCH_CLIPRECTS    0x2
 
 #undef INTEL_RUN_SYNC
+#undef INTEL_MAP_BATCHBUFFER
 
 struct intel_drm_batchbuffer
 {
@@ -40,8 +41,11 @@ intel_drm_batchbuffer_reset(struct intel_drm_batchbuffer *batch)
                                   "gallium3d_batchbuffer",
                                   batch->actual_size,
                                   4096);
+
+#ifdef INTEL_MAP_BATCHBUFFER
    drm_intel_bo_map(batch->bo, TRUE);
    batch->base.map = batch->bo->virtual;
+#endif
 
    memset(batch->base.map, 0, batch->actual_size);
    batch->base.ptr = batch->base.map;
@@ -55,7 +59,13 @@ intel_drm_batchbuffer_create(struct intel_winsys *iws)
    struct intel_drm_winsys *idws = intel_drm_winsys(iws);
    struct intel_drm_batchbuffer *batch = CALLOC_STRUCT(intel_drm_batchbuffer);
 
+   batch->actual_size = idws->max_batch_size;
+
+#ifdef INTEL_MAP_BATCHBUFFER
    batch->base.map = NULL;
+#else
+   batch->base.map = MALLOC(batch->actual_size);
+#endif
    batch->base.ptr = NULL;
    batch->base.size = 0;
 
@@ -64,8 +74,6 @@ intel_drm_batchbuffer_create(struct intel_winsys *iws)
 
    batch->base.iws = iws;
 
-   batch->actual_size = idws->max_batch_size;
-
    intel_drm_batchbuffer_reset(batch);
 
    return &batch->base;
@@ -156,7 +164,11 @@ intel_drm_batchbuffer_flush(struct intel_batchbuffer *ibatch,
 
    used = batch->base.ptr - batch->base.map;
 
+#ifdef INTEL_MAP_BATCHBUFFER
    drm_intel_bo_unmap(batch->bo);
+#else
+   drm_intel_bo_subdata(batch->bo, 0, used, batch->base.map);
+#endif
 
    /* Do the sending to HW */
    ret = drm_intel_bo_exec(batch->bo, used, NULL, 0, 0);
@@ -202,7 +214,10 @@ intel_drm_batchbuffer_destroy(struct intel_batchbuffer *ibatch)
    if (batch->bo)
       drm_intel_bo_unreference(batch->bo);
 
-   free(batch);
+#ifndef INTEL_MAP_BATCHBUFFER
+   FREE(batch->base.map);
+#endif
+   FREE(batch);
 }
 
 void intel_drm_winsys_init_batchbuffer_functions(struct intel_drm_winsys *idws)
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
index e017cd2e982..0030f915a36 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_buffer.c
@@ -28,6 +28,7 @@ intel_drm_buffer_create(struct intel_winsys *iws,
    } else if (type == INTEL_NEW_VERTEX) {
       name = "gallium3d_vertex";
       pool = idws->pools.gem;
+      buf->map_gtt = TRUE;
    } else if (type == INTEL_NEW_SCANOUT) {
       name = "gallium3d_scanout";
       pool = idws->pools.gem;
@@ -109,6 +110,18 @@ intel_drm_buffer_unmap(struct intel_winsys *iws,
       drm_intel_bo_unmap(intel_bo(buffer));
 }
 
+static int
+intel_drm_buffer_write(struct intel_winsys *iws,
+                       struct intel_buffer *buffer,
+                       const void *data,
+                       size_t size,
+                       size_t offset)
+{
+   struct intel_drm_buffer *buf = intel_drm_buffer(buffer);
+
+   return drm_intel_bo_subdata(buf->bo, offset, size, (void*)data);
+}
+
 static void
 intel_drm_buffer_destroy(struct intel_winsys *iws,
                          struct intel_buffer *buffer)
@@ -130,5 +143,6 @@ intel_drm_winsys_init_buffer_functions(struct intel_drm_winsys *idws)
    idws->base.buffer_set_fence_reg = intel_drm_buffer_set_fence_reg;
    idws->base.buffer_map = intel_drm_buffer_map;
    idws->base.buffer_unmap = intel_drm_buffer_unmap;
+   idws->base.buffer_write = intel_drm_buffer_write;
    idws->base.buffer_destroy = intel_drm_buffer_destroy;
 }