Merge branch 'master' of git+ssh://agd5f@git.freedesktop.org/git/mesa/mesa into r6xx-rewrite

16 files changed, 543 insertions, 340 deletions

diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 81e4eae401..41fcb16a0a 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -44,7 +44,6 @@
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
 
-#include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_scan.h"
 
 
@@ -55,6 +54,8 @@ struct draw_vertex_shader;
 struct draw_context;
 struct draw_stage;
 struct vbuf_render;
+struct tgsi_exec_machine;
+struct tgsi_sampler;
 
 
 /**
@@ -185,7 +186,7 @@ struct draw_context
       uint position_output;
 
       /** TGSI program interpreter runtime state */
-      struct tgsi_exec_machine machine;
+      struct tgsi_exec_machine *machine;
 
       uint num_samplers;
       struct tgsi_sampler **samplers;
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index c057cd67fd..790e89ed82 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -43,6 +43,8 @@
 #include "translate/translate.h"
 #include "translate/translate_cache.h"
 
+#include "tgsi/tgsi_exec.h"
+
 
 
 
@@ -146,16 +148,8 @@ draw_delete_vertex_shader(struct draw_context *draw,
 boolean 
 draw_vs_init( struct draw_context *draw )
 {
-   tgsi_exec_machine_init(&draw->vs.machine);
-
-   /* FIXME: give this machine thing a proper constructor:
-    */
-   draw->vs.machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16);
-   if (!draw->vs.machine.Inputs)
-      return FALSE;
-
-   draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16);
-   if (!draw->vs.machine.Outputs)
+   draw->vs.machine = tgsi_exec_machine_create();
+   if (!draw->vs.machine)
       return FALSE;
 
    draw->vs.emit_cache = translate_cache_create();
@@ -178,12 +172,6 @@ draw_vs_init( struct draw_context *draw )
 void
 draw_vs_destroy( struct draw_context *draw )
 {
-   if (draw->vs.machine.Inputs)
-      align_free(draw->vs.machine.Inputs);
-
-   if (draw->vs.machine.Outputs)
-      align_free(draw->vs.machine.Outputs);
-
    if (draw->vs.fetch_cache)
       translate_cache_destroy(draw->vs.fetch_cache);
 
@@ -196,8 +184,7 @@ draw_vs_destroy( struct draw_context *draw )
    if (draw->vs.aligned_constant_storage)
       align_free((void*)draw->vs.aligned_constant_storage);
 
-   tgsi_exec_machine_free_data(&draw->vs.machine);
-
+   tgsi_exec_machine_destroy(draw->vs.machine);
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index f2368dde5c..41cc802613 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -41,6 +41,7 @@
 
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_scan.h"
+#include "tgsi/tgsi_exec.h"
 
 
 struct exec_vertex_shader {
@@ -201,7 +202,7 @@ draw_create_vs_exec(struct draw_context *draw,
    vs->base.run_linear = vs_exec_run_linear;
    vs->base.delete = vs_exec_delete;
    vs->base.create_varient = draw_vs_varient_generic;
-   vs->machine = &draw->vs.machine;
+   vs->machine = draw->vs.machine;
 
    return &vs->base;
 }
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
index d35db57d57..ad184bd696 100644
--- a/src/gallium/auxiliary/draw/draw_vs_ppc.c
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -48,6 +48,7 @@
 #include "rtasm/rtasm_ppc.h"
 #include "tgsi/tgsi_ppc.h"
 #include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_exec.h"
 
 
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 77ba5152f9..fb58983e01 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -48,27 +48,16 @@
 #include "rtasm/rtasm_x86sse.h"
 #include "tgsi/tgsi_sse2.h"
 #include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_exec.h"
 
 #define SSE_MAX_VERTICES 4
 
-typedef void (PIPE_CDECL *codegen_function) (
-   const struct tgsi_exec_vector *input, /* 1 */
-   struct tgsi_exec_vector *output, /* 2 */
-   float (*constant)[4],        /* 3 */
-   struct tgsi_exec_vector *temporary, /* 4 */
-   float (*immediates)[4],      /* 5 */
-   const float (*aos_input)[4], /* 6 */
-   uint num_inputs,             /* 7 */
-   uint input_stride,           /* 8 */
-   float (*aos_output)[4],      /* 9 */
-   uint num_outputs,            /* 10 */
-   uint output_stride );        /* 11 */
 
 struct draw_sse_vertex_shader {
    struct draw_vertex_shader base;
    struct x86_function sse2_program;
 
-   codegen_function func;
+   tgsi_sse2_vs_func func;
    
    struct tgsi_exec_machine *machine;
 };
@@ -118,11 +107,9 @@ vs_sse_run_linear( struct draw_vertex_shader *base,
 
       /* run compiled shader
        */
-      shader->func(machine->Inputs,
-		   machine->Outputs,
-		   (float (*)[4])constants,
-		   machine->Temps,
-		   (float (*)[4])shader->base.immediates,
+      shader->func(machine,
+		   constants,
+		   shader->base.immediates,
                    input,
                    base->info.num_inputs,
                    input_stride,
@@ -184,7 +171,7 @@ draw_create_vs_sse(struct draw_context *draw,
    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
                                       sizeof(float), 16);
 
-   vs->machine = &draw->vs.machine;
+   vs->machine = draw->vs.machine;
    
    x86_init_func( &vs->sse2_program );
 
@@ -194,7 +181,7 @@ draw_create_vs_sse(struct draw_context *draw,
                         TRUE )) 
       goto fail;
       
-   vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
+   vs->func = (tgsi_sse2_vs_func) x86_get_func( &vs->sse2_program );
    if (!vs->func) {
       goto fail;
    }
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index 57fcf6de2a..1acf3c373e 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -993,6 +993,15 @@ void sse_pmovmskb( struct x86_function *p,
    emit_modrm(p, dst, src);
 }
 
+void sse_movmskps( struct x86_function *p,
+                   struct x86_reg dst,
+                   struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x50);
+   emit_modrm(p, dst, src);
+}
+
 /***********************************************************************
  * SSE2 instructions
  */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
index 1b5eaaca85..731a651796 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -223,6 +223,7 @@ void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg sr
 void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
 void sse2_punpcklbw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movmskps( struct x86_function *p, struct x86_reg dst, struct x86_reg src);
 
 void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 5cb322a5fa..fe571a86bc 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -365,15 +365,26 @@ tgsi_exec_machine_bind_shader(
 }
 
 
-void
-tgsi_exec_machine_init(
-   struct tgsi_exec_machine *mach )
+struct tgsi_exec_machine *
+tgsi_exec_machine_create( void )
 {
+   struct tgsi_exec_machine *mach;
    uint i;
 
-   mach->Temps = (struct tgsi_exec_vector *) tgsi_align_128bit( mach->_Temps);
+   mach = align_malloc( sizeof *mach, 16 );
+   if (!mach)
+      goto fail;
+
    mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR];
 
+   mach->Samplers = NULL;
+   mach->Consts = NULL;
+   mach->Tokens = NULL;
+   mach->Primitives = NULL;
+   mach->InterpCoefs = NULL;
+   mach->Instructions = NULL;
+   mach->Declarations = NULL;
+
    /* Setup constants. */
    for( i = 0; i < 4; i++ ) {
       mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].u[i] = 0x00000000;
@@ -393,22 +404,24 @@ tgsi_exec_machine_init(
    (void) print_chan;
    (void) print_temp;
 #endif
+
+   return mach;
+
+fail:
+   align_free(mach);
+   return NULL;
 }
 
 
 void
-tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach)
+tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
 {
-   if (mach->Instructions) {
+   if (mach) {
       FREE(mach->Instructions);
-      mach->Instructions = NULL;
-      mach->NumInstructions = 0;
-   }
-   if (mach->Declarations) {
       FREE(mach->Declarations);
-      mach->Declarations = NULL;
-      mach->NumDeclarations = 0;
    }
+
+   align_free(mach);
 }
 
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index da22baad3e..8a9100f4c3 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -29,6 +29,7 @@
 #define TGSI_EXEC_H
 
 #include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
 
 #if defined __cplusplus
 extern "C" {
@@ -94,7 +95,6 @@ struct tgsi_exec_labels
 
 
 #define TGSI_EXEC_NUM_TEMPS       128
-#define TGSI_EXEC_NUM_TEMP_EXTRAS   6
 #define TGSI_EXEC_NUM_IMMEDIATES  256
 
 /*
@@ -162,9 +162,14 @@ struct tgsi_exec_labels
 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 #define TGSI_EXEC_MASK_C            2
 
+/* 4 register buffer for various purposes */
 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
+#define TGSI_EXEC_NUM_TEMP_R        4
+
+#define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
+#define TGSI_EXEC_NUM_ADDRS         1
+#define TGSI_EXEC_NUM_TEMP_EXTRAS   9
 
-#define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 5)
 
 
 #define TGSI_EXEC_MAX_COND_NESTING  20
@@ -187,24 +192,21 @@ struct tgsi_exec_labels
 struct tgsi_exec_machine
 {
    /* Total = program temporaries + internal temporaries
-    *         + 1 padding to align to 16 bytes
     */
-   struct tgsi_exec_vector       _Temps[TGSI_EXEC_NUM_TEMPS +
-                                        TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
+   struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
+                                       TGSI_EXEC_NUM_TEMP_EXTRAS];
+
+   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
+
+   struct tgsi_exec_vector       Inputs[PIPE_MAX_ATTRIBS];
+   struct tgsi_exec_vector       Outputs[PIPE_MAX_ATTRIBS];
 
-   /*
-    * This will point to _Temps after aligning to 16B boundary.
-    */
-   struct tgsi_exec_vector       *Temps;
    struct tgsi_exec_vector       *Addrs;
 
    struct tgsi_sampler           **Samplers;
 
-   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
    unsigned                      ImmLimit;
    const float                   (*Consts)[4];
-   struct tgsi_exec_vector       *Inputs;
-   struct tgsi_exec_vector       *Outputs;
    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 
@@ -251,9 +253,11 @@ struct tgsi_exec_machine
    struct tgsi_exec_labels Labels;
 };
 
+struct tgsi_exec_machine *
+tgsi_exec_machine_create( void );
+
 void
-tgsi_exec_machine_init(
-   struct tgsi_exec_machine *mach );
+tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 
 
 void 
@@ -268,10 +272,6 @@ tgsi_exec_machine_run(
    struct tgsi_exec_machine *mach );
 
 
-void
-tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
-
-
 static INLINE void
 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index ba2bfdef06..df49638d51 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -32,6 +32,7 @@
 #include "util/u_debug.h"
 #include "pipe/p_shader_tokens.h"
 #include "util/u_math.h"
+#include "util/u_memory.h"
 #if defined(PIPE_ARCH_SSE)
 #include "util/u_sse.h"
 #endif
@@ -100,37 +101,55 @@ get_const_base( void )
 {
    return x86_make_reg(
       file_REG32,
-      reg_CX );
+      reg_AX );
 }
 
 static struct x86_reg
-get_input_base( void )
+get_machine_base( void )
 {
    return x86_make_reg(
       file_REG32,
-      reg_AX );
+      reg_CX );
+}
+
+static struct x86_reg
+get_input_base( void )
+{
+   return x86_make_disp(
+      get_machine_base(),
+      Offset(struct tgsi_exec_machine, Inputs) );
 }
 
 static struct x86_reg
 get_output_base( void )
 {
-   return x86_make_reg(
-      file_REG32,
-      reg_DX );
+   return x86_make_disp(
+      get_machine_base(),
+      Offset(struct tgsi_exec_machine, Outputs) );
 }
 
 static struct x86_reg
 get_temp_base( void )
 {
+   return x86_make_disp(
+      get_machine_base(),
+      Offset(struct tgsi_exec_machine, Temps) );
+}
+
+static struct x86_reg
+get_coef_base( void )
+{
    return x86_make_reg(
       file_REG32,
       reg_BX );
 }
 
 static struct x86_reg
-get_coef_base( void )
+get_sampler_base( void )
 {
-   return get_output_base();
+   return x86_make_reg(
+      file_REG32,
+      reg_DI );
 }
 
 static struct x86_reg
@@ -138,7 +157,7 @@ get_immediate_base( void )
 {
    return x86_make_reg(
       file_REG32,
-      reg_DI );
+      reg_DX );
 }
 
 
@@ -168,6 +187,15 @@ get_const(
 }
 
 static struct x86_reg
+get_sampler_ptr(
+   unsigned unit )
+{
+   return x86_make_disp(
+      get_sampler_base(),
+      unit * sizeof( struct tgsi_sampler * ) );
+}
+
+static struct x86_reg
 get_input(
    unsigned vec,
    unsigned chan )
@@ -520,24 +548,15 @@ emit_coef_dady(
  * that the stack pointer is 16 byte aligned, as expected.
  */
 static void
-emit_func_call_dst(
+emit_func_call(
    struct x86_function *func,
-   unsigned xmm_save,
-   unsigned xmm_dst,
+   unsigned xmm_save_mask,
+   const struct x86_reg *arg,
+   unsigned nr_args,
    void (PIPE_CDECL *code)() )
 {
    struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
    unsigned i, n;
-   unsigned xmm_mask;
-   
-   /* Bitmask of the xmm registers to save */
-   xmm_mask = (1 << xmm_save) - 1;
-   xmm_mask &= ~(1 << xmm_dst);
-
-   sse_movaps(
-      func,
-      get_temp( TEMP_R0, 0 ),
-      make_xmm( xmm_dst ) );
 
    x86_push(
       func,
@@ -549,8 +568,10 @@ emit_func_call_dst(
       func,
       x86_make_reg( file_REG32, reg_DX) );
    
+   /* Store XMM regs to the stack
+    */
    for(i = 0, n = 0; i < 8; ++i)
-      if(xmm_mask & (1 << i))
+      if(xmm_save_mask & (1 << i))
          ++n;
    
    x86_sub_imm(
@@ -559,26 +580,42 @@ emit_func_call_dst(
       n*16);
 
    for(i = 0, n = 0; i < 8; ++i)
-      if(xmm_mask & (1 << i)) {
+      if(xmm_save_mask & (1 << i)) {
          sse_movups(
             func,
             x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ),
             make_xmm( i ) );
          ++n;
       }
+
+   for (i = 0; i < nr_args; i++) {
+      /* Load the address of the buffer we use for passing arguments and
+       * receiving results:
+       */
+      x86_lea(
+	 func,
+	 ecx,
+	 arg[i] );
    
-   x86_lea(
-      func,
-      ecx,
-      get_temp( TEMP_R0, 0 ) );
-   
-   x86_push( func, ecx );
+      /* Push actual function arguments (currently just the pointer to
+       * the buffer above), and call the function:
+       */
+      x86_push( func, ecx );
+   }
+
    x86_mov_reg_imm( func, ecx, (unsigned long) code );
    x86_call( func, ecx );
-   x86_pop(func, ecx );
-   
+
+   /* Pop the arguments (or just add an immediate to esp)
+    */
+   for (i = 0; i < nr_args; i++) {
+      x86_pop(func, ecx );
+   }
+
+   /* Pop the saved XMM regs:
+    */
    for(i = 0, n = 0; i < 8; ++i)
-      if(xmm_mask & (1 << i)) {
+      if(xmm_save_mask & (1 << i)) {
          sse_movups(
             func,
             make_xmm( i ),
@@ -602,34 +639,86 @@ emit_func_call_dst(
    x86_pop(
       func,
       x86_make_reg( file_REG32, reg_AX) );
+}
+
+static void
+emit_func_call_dst_src1(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst,
+   unsigned xmm_src0,
+   void (PIPE_CDECL *code)() )
+{
+   struct x86_reg store = get_temp( TEMP_R0, 0 );
+   unsigned xmm_mask = ((1 << xmm_save) - 1) & ~(1 << xmm_dst);
+   
+   /* Store our input parameters (in xmm regs) to the buffer we use
+    * for passing arguments.  We will pass a pointer to this buffer as
+    * the actual function argument.
+    */
+   sse_movaps(
+      func,
+      store,
+      make_xmm( xmm_src0 ) );
+
+   emit_func_call( func,
+                   xmm_mask,
+                   &store,
+                   1,
+                   code );
 
    sse_movaps(
       func,
       make_xmm( xmm_dst ),
-      get_temp( TEMP_R0, 0 ) );
+      store );
 }
 
+
 static void
-emit_func_call_dst_src(
+emit_func_call_dst_src2(
    struct x86_function *func,
    unsigned xmm_save, 
    unsigned xmm_dst,
-   unsigned xmm_src,
+   unsigned xmm_src0,
+   unsigned xmm_src1,
    void (PIPE_CDECL *code)() )
 {
+   struct x86_reg store = get_temp( TEMP_R0, 0 );
+   unsigned xmm_mask = ((1 << xmm_save) - 1) & ~(1 << xmm_dst);
+
+   /* Store two inputs to parameter buffer.
+    */
    sse_movaps(
       func,
-      get_temp( TEMP_R0, 1 ),
-      make_xmm( xmm_src ) );
+      store,
+      make_xmm( xmm_src0 ) );
 
-   emit_func_call_dst(
+   sse_movaps(
       func,
-      xmm_save,
-      xmm_dst,
-      code );
+      x86_make_disp( store, 4 * sizeof(float) ),
+      make_xmm( xmm_src1 ) );
+
+
+   /* Emit the call
+    */
+   emit_func_call( func,
+                   xmm_mask,
+                   &store,
+                   1,
+                   code );
+
+   /* Retrieve the results:
+    */
+   sse_movaps(
+      func,
+      make_xmm( xmm_dst ),
+      store );
 }
 
 
+
+
+
 #if defined(PIPE_ARCH_SSE)
 
 /*
@@ -782,10 +871,11 @@ emit_cos(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save, 
       xmm_dst,
+      xmm_dst,
       cos4f );
 }
 
@@ -812,10 +902,11 @@ emit_ex2(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       ex24f );
 }
 
@@ -857,10 +948,11 @@ emit_flr(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       flr4f );
 }
 
@@ -880,10 +972,11 @@ emit_frc(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       frc4f );
 }
 
@@ -910,10 +1003,11 @@ emit_lg2(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       lg24f );
 }
 
@@ -975,13 +1069,15 @@ emit_pow(
    struct x86_function *func,
    unsigned xmm_save, 
    unsigned xmm_dst,
-   unsigned xmm_src )
+   unsigned xmm_src0,
+   unsigned xmm_src1 )
 {
-   emit_func_call_dst_src(
+   emit_func_call_dst_src2(
       func,
       xmm_save,
       xmm_dst,
-      xmm_src,
+      xmm_src0,
+      xmm_src1,
       pow4f );
 }
 
@@ -1017,10 +1113,11 @@ emit_rnd(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       rnd4f );
 }
 
@@ -1099,10 +1196,11 @@ emit_sgn(
    unsigned xmm_save, 
    unsigned xmm_dst )
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       sgn4f );
 }
 
@@ -1121,10 +1219,11 @@ emit_sin (struct x86_function *func,
           unsigned xmm_save, 
           unsigned xmm_dst)
 {
-   emit_func_call_dst(
+   emit_func_call_dst_src1(
       func,
       xmm_save,
       xmm_dst,
+      xmm_dst,
       sin4f );
 }
 
@@ -1140,6 +1239,12 @@ emit_sub(
       make_xmm( xmm_src ) );
 }
 
+
+
+
+
+
+
 /**
  * Register fetch.
  */
@@ -1298,20 +1403,164 @@ emit_store(
 #define STORE( FUNC, INST, XMM, INDEX, CHAN )\
    emit_store( FUNC, XMM, &(INST).FullDstRegisters[INDEX], &(INST), CHAN )
 
+
+static void PIPE_CDECL
+fetch_texel( struct tgsi_sampler **sampler,
+             float *store )
+{
+#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, 16 * sizeof(float));
+   }
+
+#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
+}
+
 /**
  * High-level instruction translators.
  */
 
 static void
+emit_tex( struct x86_function *func,
+          const struct tgsi_full_instruction *inst,
+          boolean lodbias,
+          boolean projected)
+{
+   const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
+   struct x86_reg args[2];
+   unsigned count;
+   unsigned i;
+
+   switch (inst->InstructionExtTexture.Texture) {
+   case TGSI_TEXTURE_1D:
+   case TGSI_TEXTURE_SHADOW1D:
+      count = 1;
+      break;
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+   case TGSI_TEXTURE_SHADOW2D:
+   case TGSI_TEXTURE_SHADOWRECT:
+      count = 2;
+      break;
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+      count = 3;
+      break;
+   default:
+      assert(0);
+      return;
+   }
+
+   if (lodbias) {
+      FETCH( func, *inst, 3, 0, 3 );
+   }
+   else {
+      emit_tempf(
+         func,
+         3,
+         TGSI_EXEC_TEMP_00000000_I,
+         TGSI_EXEC_TEMP_00000000_C );
+
+   }
+
+   /* store lodbias whether enabled or not -- fetch_texel currently
+    * respects it always.
+    */
+   sse_movaps( func,
+               get_temp( TEMP_R0, 3 ),
+               make_xmm( 3 ) );
+
+   
+   if (projected) {
+      FETCH( func, *inst, 3, 0, 3 );
+
+      emit_rcp( func, 3, 3 );
+   }
+
+   for (i = 0; i < count; i++) {
+      FETCH( func, *inst, i, 0, i );
+
+      if (projected) {
+         sse_mulps(
+            func,
+            make_xmm( i ),
+            make_xmm( 3 ) );
+      }
+      
+      /* Store in the argument buffer:
+       */
+      sse_movaps(
+         func,
+         get_temp( TEMP_R0, i ),
+         make_xmm( i ) );
+   }
+
+   args[0] = get_temp( TEMP_R0, 0 );
+   args[1] = get_sampler_ptr( unit );
+
+
+   emit_func_call( func,
+                   0,
+                   args,
+                   Elements(args),
+                   fetch_texel );
+
+   /* If all four channels are enabled, could use a pointer to
+    * dst[0].x instead of TEMP_R0 for store?
+    */
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, i ) {
+
+      sse_movaps(
+         func,
+         make_xmm( 0 ),
+         get_temp( TEMP_R0, i ) );
+
+      STORE( func, *inst, 0, 0, i );
+   }
+}
+
+
+static void
 emit_kil(
    struct x86_function *func,
    const struct tgsi_full_src_register *reg )
 {
    unsigned uniquemask;
-   unsigned registers[4];
-   unsigned nextregister = 0;
-   unsigned firstchan = ~0;
+   unsigned unique_count = 0;
    unsigned chan_index;
+   unsigned i;
 
    /* This mask stores component bits that were already tested. Note that
     * we test if the value is less than zero, so 1.0 and 0.0 need not to be
@@ -1331,18 +1580,11 @@ emit_kil(
          uniquemask |= 1 << swizzle;
 
          /* allocate register */
-         registers[chan_index] = nextregister;
          emit_fetch(
             func,
-            nextregister,
+            unique_count++,
             reg,
             chan_index );
-         nextregister++;
-
-         /* mark the first channel used */
-         if( firstchan == ~0 ) {
-            firstchan = chan_index;
-         }
       }
    }
 
@@ -1353,32 +1595,32 @@ emit_kil(
       func,
       x86_make_reg( file_REG32, reg_DX ) );
 
-   FOR_EACH_CHANNEL( chan_index ) {
-      if( uniquemask & (1 << chan_index) ) {
-         sse_cmpps(
+   for (i = 0 ; i < unique_count; i++ ) {
+      struct x86_reg dataXMM = make_xmm(i);
+
+      sse_cmpps(
+         func,
+         dataXMM,
+         get_temp(
+            TGSI_EXEC_TEMP_00000000_I,
+            TGSI_EXEC_TEMP_00000000_C ),
+         cc_LessThan );
+      
+      if( i == 0 ) {
+         sse_movmskps(
             func,
-            make_xmm( registers[chan_index] ),
-            get_temp(
-               TGSI_EXEC_TEMP_00000000_I,
-               TGSI_EXEC_TEMP_00000000_C ),
-            cc_LessThan );
-
-         if( chan_index == firstchan ) {
-            sse_pmovmskb(
-               func,
-               x86_make_reg( file_REG32, reg_AX ),
-               make_xmm( registers[chan_index] ) );
-         }
-         else {
-            sse_pmovmskb(
-               func,
-               x86_make_reg( file_REG32, reg_DX ),
-               make_xmm( registers[chan_index] ) );
-            x86_or(
-               func,
-               x86_make_reg( file_REG32, reg_AX ),
-               x86_make_reg( file_REG32, reg_DX ) );
-         }
+            x86_make_reg( file_REG32, reg_AX ),
+            dataXMM );
+      }
+      else {
+         sse_movmskps(
+            func,
+            x86_make_reg( file_REG32, reg_DX ),
+            dataXMM );
+         x86_or(
+            func,
+            x86_make_reg( file_REG32, reg_AX ),
+            x86_make_reg( file_REG32, reg_DX ) );
       }
    }
 
@@ -1573,7 +1815,7 @@ emit_instruction(
                get_temp(
                   TGSI_EXEC_TEMP_MINUS_128_I,
                   TGSI_EXEC_TEMP_MINUS_128_C ) );
-            emit_pow( func, 3, 1, 2 );
+            emit_pow( func, 3, 1, 1, 2 );
             FETCH( func, *inst, 0, 0, CHAN_X );
             sse_xorps(
                func,
@@ -1917,7 +2159,7 @@ emit_instruction(
    /* TGSI_OPCODE_POW */
       FETCH( func, *inst, 0, 0, CHAN_X );
       FETCH( func, *inst, 1, 1, CHAN_X );
-      emit_pow( func, 0, 0, 1 );
+      emit_pow( func, 0, 0, 0, 1 );
       FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
          STORE( func, *inst, 0, 0, chan_index );
       }
@@ -2086,21 +2328,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TEX:
-      if (0) {
-	 /* Disable dummy texture code: 
-	  */
-	 emit_tempf(
-	    func,
-	    0,
-	    TEMP_ONE_I,
-	    TEMP_ONE_C );
-	 FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
-	    STORE( func, *inst, 0, 0, chan_index );
-	 }
-      }
-      else {
-	 return 0;
-      }
+      emit_tex( func, inst, FALSE, FALSE );
       break;
 
    case TGSI_OPCODE_TXD:
@@ -2198,7 +2426,7 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TXB:
-      return 0;
+      emit_tex( func, inst, TRUE, FALSE );
       break;
 
    case TGSI_OPCODE_NRM:
@@ -2306,9 +2534,13 @@ emit_instruction(
       break;
 
    case TGSI_OPCODE_TXL:
-      return 0;
+      emit_tex( func, inst, TRUE, FALSE );
       break;
 
+   case TGSI_OPCODE_TXP:
+      emit_tex( func, inst, FALSE, TRUE );
+      break;
+      
    case TGSI_OPCODE_BRK:
       return 0;
       break;
@@ -2488,7 +2720,7 @@ emit_declaration(
 
 static void aos_to_soa( struct x86_function *func, 
                         uint arg_aos,
-                        uint arg_soa, 
+                        uint arg_machine, 
                         uint arg_num, 
                         uint arg_stride )
 {
@@ -2503,7 +2735,10 @@ static void aos_to_soa( struct x86_function *func,
    x86_push( func, x86_make_reg( file_REG32, reg_BX ) );
 
    x86_mov( func, aos_input,  x86_fn_arg( func, arg_aos ) );
-   x86_mov( func, soa_input,  x86_fn_arg( func, arg_soa ) );
+   x86_mov( func, soa_input,  x86_fn_arg( func, arg_machine ) );
+   x86_lea( func, soa_input,  
+	    x86_make_disp( soa_input, 
+			   Offset(struct tgsi_exec_machine, Inputs) ) );
    x86_mov( func, num_inputs, x86_fn_arg( func, arg_num ) );
    x86_mov( func, stride,     x86_fn_arg( func, arg_stride ) );
 
@@ -2545,28 +2780,30 @@ static void aos_to_soa( struct x86_function *func,
    x86_jcc( func, cc_NE, inner_loop );
 
    /* Restore EBX */
-   x86_pop( func, aos_input );
+   x86_pop( func, x86_make_reg( file_REG32, reg_BX ) );
 }
 
-static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num, uint stride )
+static void soa_to_aos( struct x86_function *func, 
+			uint arg_aos, 
+			uint arg_machine, 
+			uint arg_num, 
+			uint arg_stride )
 {
-   struct x86_reg soa_output;
-   struct x86_reg aos_output;
-   struct x86_reg num_outputs;
-   struct x86_reg temp;
+   struct x86_reg soa_output = x86_make_reg( file_REG32, reg_AX );
+   struct x86_reg aos_output = x86_make_reg( file_REG32, reg_BX );
+   struct x86_reg num_outputs = x86_make_reg( file_REG32, reg_CX );
+   struct x86_reg temp = x86_make_reg( file_REG32, reg_DX );
    int inner_loop;
 
-   soa_output = x86_make_reg( file_REG32, reg_AX );
-   aos_output = x86_make_reg( file_REG32, reg_BX );
-   num_outputs = x86_make_reg( file_REG32, reg_CX );
-   temp = x86_make_reg( file_REG32, reg_DX );
-
    /* Save EBX */
-   x86_push( func, aos_output );
+   x86_push( func, x86_make_reg( file_REG32, reg_BX ) );
 
-   x86_mov( func, soa_output, x86_fn_arg( func, soa ) );
-   x86_mov( func, aos_output, x86_fn_arg( func, aos ) );
-   x86_mov( func, num_outputs, x86_fn_arg( func, num ) );
+   x86_mov( func, aos_output, x86_fn_arg( func, arg_aos ) );
+   x86_mov( func, soa_output, x86_fn_arg( func, arg_machine ) );
+   x86_lea( func, soa_output, 
+	    x86_make_disp( soa_output, 
+			   Offset(struct tgsi_exec_machine, Outputs) ) );
+   x86_mov( func, num_outputs, x86_fn_arg( func, arg_num ) );
 
    /* do */
    inner_loop = x86_get_label( func );
@@ -2583,7 +2820,7 @@ static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num,
       sse_unpcklps( func, make_xmm( 3 ), make_xmm( 4 ) );
       sse_unpckhps( func, make_xmm( 5 ), make_xmm( 4 ) );
 
-      x86_mov( func, temp, x86_fn_arg( func, stride ) );
+      x86_mov( func, temp, x86_fn_arg( func, arg_stride ) );
       x86_push( func, aos_output );
       sse_movlps( func, x86_make_disp( aos_output, 0 ), make_xmm( 0 ) );
       sse_movlps( func, x86_make_disp( aos_output, 8 ), make_xmm( 3 ) );
@@ -2607,20 +2844,13 @@ static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num,
    x86_jcc( func, cc_NE, inner_loop );
 
    /* Restore EBX */
-   x86_pop( func, aos_output );
+   x86_pop( func, x86_make_reg( file_REG32, reg_BX ) );
 }
 
 /**
  * Translate a TGSI vertex/fragment shader to SSE2 code.
  * Slightly different things are done for vertex vs. fragment shaders.
  *
- * Note that fragment shaders are responsible for interpolating shader
- * inputs. Because on x86 we have only 4 GP registers, and here we
- * have 5 shader arguments (input, output, const, temp and coef), the
- * code is split into two phases -- DECLARATION and INSTRUCTION phase.
- * GP register holding the output argument is aliased with the coeff
- * argument, as outputs are not needed in the DECLARATION phase.
- *
  * \param tokens  the TGSI input shader
  * \param func  the output SSE code/function
  * \param immediates  buffer to place immediates, later passed to SSE func
@@ -2634,7 +2864,6 @@ tgsi_emit_sse2(
    boolean do_swizzles )
 {
    struct tgsi_parse_context parse;
-   boolean instruction_phase = FALSE;
    unsigned ok = 1;
    uint num_immediates = 0;
 
@@ -2646,74 +2875,48 @@ tgsi_emit_sse2(
 
    /* Can't just use EDI, EBX without save/restoring them:
     */
-   x86_push(
-      func,
-      get_immediate_base() );
-
-   x86_push(
-      func,
-      get_temp_base() );
-
+   x86_push( func, x86_make_reg( file_REG32, reg_BX ) );
+   x86_push( func, x86_make_reg( file_REG32, reg_DI ) );
 
    /*
     * Different function args for vertex/fragment shaders:
     */
-   if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
-      /* DECLARATION phase, do not load output argument. */
-      x86_mov(
-         func,
-         get_input_base(),
-         x86_fn_arg( func, 1 ) );
-      /* skipping outputs argument here */
-      x86_mov(
-         func,
-         get_const_base(),
-         x86_fn_arg( func, 3 ) );
-      x86_mov(
-         func,
-         get_temp_base(),
-         x86_fn_arg( func, 4 ) );
-      x86_mov(
-         func,
-         get_coef_base(),
-         x86_fn_arg( func, 5 ) );
-      x86_mov(
-         func,
-         get_immediate_base(),
-         x86_fn_arg( func, 6 ) );
-   }
-   else {
-      assert(parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX);
-
+   if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX) {
       if (do_swizzles)
          aos_to_soa( func, 
-                     6,         /* aos_input */
-                     1,         /* machine->input */
-                     7,         /* num_inputs */
-                     8 );       /* input_stride */
+                     4,         /* aos_input */
+                     1,         /* machine */
+                     5,         /* num_inputs */
+                     6 );       /* input_stride */
+   }
 
+   x86_mov(
+      func,
+      get_machine_base(),
+      x86_fn_arg( func, 1 ) );
+   x86_mov(
+      func,
+      get_const_base(),
+      x86_fn_arg( func, 2 ) );
+   x86_mov(
+      func,
+      get_immediate_base(),
+      x86_fn_arg( func, 3 ) );
+
+   if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
       x86_mov(
-         func,
-         get_input_base(),
-         x86_fn_arg( func, 1 ) );
-      x86_mov(
-         func,
-         get_output_base(),
-         x86_fn_arg( func, 2 ) );
-      x86_mov(
-         func,
-         get_const_base(),
-         x86_fn_arg( func, 3 ) );
-      x86_mov(
-         func,
-         get_temp_base(),
-         x86_fn_arg( func, 4 ) );
+	 func,
+	 get_coef_base(),
+	 x86_fn_arg( func, 4 ) );
+
       x86_mov(
-         func,
-         get_immediate_base(),
-         x86_fn_arg( func, 5 ) );
+	 func,
+	 get_sampler_base(),
+	 x86_make_disp( get_machine_base(),
+                        Offset( struct tgsi_exec_machine, Samplers ) ) );
    }
 
+
    while( !tgsi_parse_end_of_tokens( &parse ) && ok ) {
       tgsi_parse_token( &parse );
 
@@ -2727,17 +2930,6 @@ tgsi_emit_sse2(
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
-         if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
-            if( !instruction_phase ) {
-               /* INSTRUCTION phase, overwrite coeff with output. */
-               instruction_phase = TRUE;
-               x86_mov(
-                  func,
-                  get_output_base(),
-                  x86_fn_arg( func, 2 ) );
-            }
-         }
-
          ok = emit_instruction(
             func,
             &parse.FullToken.FullInstruction );
@@ -2781,18 +2973,17 @@ tgsi_emit_sse2(
 
    if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX) {
       if (do_swizzles)
-         soa_to_aos( func, 9, 2, 10, 11 );
+         soa_to_aos( func, 
+		     7, 	/* aos_output */
+		     1, 	/* machine */
+		     8, 	/* num_outputs */
+		     9 );	/* output_stride */
    }
 
    /* Can't just use EBX, EDI without save/restoring them:
     */
-   x86_pop(
-      func,
-      get_temp_base() );
-
-   x86_pop(
-      func,
-      get_immediate_base() );
+   x86_pop( func, x86_make_reg( file_REG32, reg_DI ) );
+   x86_pop( func, x86_make_reg( file_REG32, reg_BX ) );
 
    emit_ret( func );
 
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.h b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
index af838b2a25..d81ee3d00e 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
@@ -34,6 +34,7 @@ extern "C" {
 
 struct tgsi_token;
 struct x86_function;
+struct tgsi_interp_coef;
 
 unsigned
 tgsi_emit_sse2(
@@ -42,6 +43,33 @@ tgsi_emit_sse2(
    float (*immediates)[4],
    boolean do_swizzles );
 
+
+/* This is the function prototype generated when do_swizzles is false
+ * -- effectively for fragment shaders.
+ */
+typedef void (PIPE_CDECL *tgsi_sse2_fs_function) (
+   struct tgsi_exec_machine *machine, /* 1 */
+   const float (*constant)[4],		    /* 2 */
+   const float (*immediate)[4],		    /* 3 */
+   const struct tgsi_interp_coef *coef	    /* 4 */
+   );
+
+
+/* This is the function prototype generated when do_swizzles is true
+ * -- effectively for vertex shaders.
+ */
+typedef void (PIPE_CDECL *tgsi_sse2_vs_func) (
+   struct tgsi_exec_machine *machine, /* 1 */
+   const float (*constant)[4],        /* 2 */
+   const float (*immediate)[4],       /* 3 */
+   const float (*aos_input)[4], /* 4 */
+   uint num_inputs,             /* 5 */
+   uint input_stride,           /* 6 */
+   float (*aos_output)[4],      /* 7 */
+   uint num_outputs,            /* 8 */
+   uint output_stride );        /* 9 */
+
+
 #if defined __cplusplus
 }
 #endif
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index 31c3ca21c5..f4fa0905d7 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -45,17 +45,6 @@
 
 #include "rtasm/rtasm_x86sse.h"
 
-/* Surely this should be defined somewhere in a tgsi header:
- */
-typedef void (PIPE_CDECL *codegen_function)(
-   const struct tgsi_exec_vector *input,
-   struct tgsi_exec_vector *output,
-   const float (*constant)[4],
-   struct tgsi_exec_vector *temporary,
-   const struct tgsi_interp_coef *coef,
-   float (*immediates)[4]
-   //, const struct tgsi_exec_vector *quadPos
- );
 
 
 /**
@@ -65,7 +54,7 @@ struct sp_sse_fragment_shader
 {
    struct sp_fragment_shader base;
    struct x86_function sse2_program;
-   codegen_function func;
+   tgsi_sse2_fs_function func;
    float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
 };
 
@@ -83,6 +72,7 @@ fs_sse_prepare( const struct sp_fragment_shader *base,
 		struct tgsi_exec_machine *machine,
 		struct tgsi_sampler **samplers )
 {
+   machine->Samplers = samplers;
 }
 
 
@@ -107,12 +97,10 @@ fs_sse_run( const struct sp_fragment_shader *base,
    tgsi_set_kill_mask(machine, 0x0);
    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 
-   shader->func( machine->Inputs,
-		 machine->Outputs,
+   shader->func( machine,
 		 machine->Consts,
-		 machine->Temps,
-		 machine->InterpCoefs,
-                 shader->immediates
+                 (const float (*)[4])shader->immediates,
+		 machine->InterpCoefs
 		 //	 , &machine->QuadPos
       );
 
@@ -151,7 +139,7 @@ softpipe_create_fs_sse(struct softpipe_context *softpipe,
       return NULL;
    }
 
-   shader->func = (codegen_function) x86_get_func( &shader->sse2_program );
+   shader->func = (tgsi_sse2_fs_function) x86_get_func( &shader->sse2_program );
    if (!shader->func) {
       x86_release_func( &shader->sse2_program );
       FREE(shader);
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index ca637a1d6a..28f8d1a60e 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -52,7 +52,7 @@
 struct quad_shade_stage
 {
    struct quad_stage stage;  /**< base class */
-   struct tgsi_exec_machine machine;
+   struct tgsi_exec_machine *machine;
    struct tgsi_exec_vector *inputs, *outputs;
 };
 
@@ -73,7 +73,7 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    struct quad_shade_stage *qss = quad_shade_stage( qs );
    struct softpipe_context *softpipe = qs->softpipe;
-   struct tgsi_exec_machine *machine = &qss->machine;
+   struct tgsi_exec_machine *machine = qss->machine;
    boolean z_written;
    
    /* Consts do not require 16 byte alignment. */
@@ -146,7 +146,7 @@ shade_begin(struct quad_stage *qs)
    struct softpipe_context *softpipe = qs->softpipe;
 
    softpipe->fs->prepare( softpipe->fs, 
-			  &qss->machine,
+			  qss->machine,
 			  (struct tgsi_sampler **)
                              softpipe->tgsi.frag_samplers_list );
 
@@ -159,9 +159,8 @@ shade_destroy(struct quad_stage *qs)
 {
    struct quad_shade_stage *qss = (struct quad_shade_stage *) qs;
 
-   tgsi_exec_machine_free_data(&qss->machine);
-   FREE( qss->inputs );
-   FREE( qss->outputs );
+   tgsi_exec_machine_destroy(qss->machine);
+
    FREE( qs );
 }
 
@@ -170,19 +169,24 @@ struct quad_stage *
 sp_quad_shade_stage( struct softpipe_context *softpipe )
 {
    struct quad_shade_stage *qss = CALLOC_STRUCT(quad_shade_stage);
-
-   /* allocate storage for program inputs/outputs, aligned to 16 bytes */
-   qss->inputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->inputs) + 16);
-   qss->outputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->outputs) + 16);
-   qss->machine.Inputs = align16(qss->inputs);
-   qss->machine.Outputs = align16(qss->outputs);
+   if (!qss)
+      goto fail;
 
    qss->stage.softpipe = softpipe;
    qss->stage.begin = shade_begin;
    qss->stage.run = shade_quad;
    qss->stage.destroy = shade_destroy;
 
-   tgsi_exec_machine_init( &qss->machine );
+   qss->machine = tgsi_exec_machine_create();
+   if (!qss->machine)
+      goto fail;
 
    return &qss->stage;
+
+fail:
+   if (qss && qss->machine)
+      tgsi_exec_machine_destroy(qss->machine);
+
+   FREE(qss);
+   return NULL;
 }
diff --git a/src/gallium/state_trackers/egl/egl_context.c b/src/gallium/state_trackers/egl/egl_context.c
index edd49486e5..f03a29582a 100644
--- a/src/gallium/state_trackers/egl/egl_context.c
+++ b/src/gallium/state_trackers/egl/egl_context.c
@@ -113,7 +113,7 @@ drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext
 	if (!ctx)
 		goto err_c;
 
-	_eglInitContext(drv, dpy, &ctx->base, config, attrib_list);
+	_eglInitContext(drv, dpy, &ctx->base, conf, attrib_list);
 
 	ctx->pipe = dev->api->create_context(dev->api, dev->screen);
 	if (!ctx->pipe)
@@ -129,8 +129,8 @@ drm_create_context(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config, EGLContext
 	if (!ctx->st)
 		goto err_gl;
 
-	/* generate handle and insert into hash table */
-	_eglSaveContext(&ctx->base);
+        /* link to display */
+        _eglLinkContext(&ctx->base, _eglLookupDisplay(dpy));
 	assert(_eglGetContextHandle(&ctx->base));
 
 	return _eglGetContextHandle(&ctx->base);
@@ -147,10 +147,8 @@ EGLBoolean
 drm_destroy_context(_EGLDriver *drv, EGLDisplay dpy, EGLContext context)
 {
 	struct drm_context *c = lookup_drm_context(context);
-	_eglRemoveContext(&c->base);
-	if (c->base.IsBound) {
-		c->base.DeletePending = EGL_TRUE;
-	} else {
+        _eglUnlinkContext(&c->base);
+	if (!c->base.IsBound) {
 		st_destroy_context(c->st);
 		c->pipe->destroy(c->pipe);
 		free(c);
diff --git a/src/gallium/state_trackers/egl/egl_surface.c b/src/gallium/state_trackers/egl/egl_surface.c
index de8194a46a..86f2ea97e5 100644
--- a/src/gallium/state_trackers/egl/egl_surface.c
+++ b/src/gallium/state_trackers/egl/egl_surface.c
@@ -232,7 +232,7 @@ drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
 	if (!surf)
 		goto err;
 
-	if (!_eglInitSurface(drv, dpy, &surf->base, EGL_PBUFFER_BIT, config, attrib_list))
+	if (!_eglInitSurface(drv, &surf->base, EGL_PBUFFER_BIT, conf, attrib_list))
 		goto err_surf;
 
 	surf->w = width;
@@ -245,7 +245,7 @@ drm_create_pbuffer_surface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
 	                                    (void*)surf);
 	drm_visual_modes_destroy(visual);
 
-	_eglSaveSurface(&surf->base);
+        _eglLinkSurface(&surf->base, _eglLookupDisplay(dpy));
 	return surf->base.Handle;
 
 err_surf:
@@ -364,10 +364,9 @@ EGLBoolean
 drm_destroy_surface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface)
 {
 	struct drm_surface *surf = lookup_drm_surface(surface);
-	_eglRemoveSurface(&surf->base);
-	if (surf->base.IsBound) {
-		surf->base.DeletePending = EGL_TRUE;
-	} else {
+	_eglUnlinkSurface(&surf->base);
+
+	if (!surf->base.IsBound) {
 		if (surf->screen)
 			drm_takedown_shown_screen(drv, surf->screen);
 		st_unreference_framebuffer(surf->stfb);
diff --git a/src/gallium/winsys/egl_xlib/egl_xlib.c b/src/gallium/winsys/egl_xlib/egl_xlib.c
index c10e3c00ff..e1ddcae97b 100644
--- a/src/gallium/winsys/egl_xlib/egl_xlib.c
+++ b/src/gallium/winsys/egl_xlib/egl_xlib.c
@@ -345,7 +345,7 @@ xlib_eglCreateContext(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
       return EGL_NO_CONTEXT;
 
    /* let EGL lib init the common stuff */
-   if (!_eglInitContext(drv, dpy, &ctx->Base, config, attrib_list)) {
+   if (!_eglInitContext(drv, &ctx->Base, conf, attrib_list)) {
       free(ctx);
       return EGL_NO_CONTEXT;
    }
@@ -370,7 +370,7 @@ xlib_eglCreateContext(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
       return EGL_NO_CONTEXT;
    }
 
-   _eglSaveContext(&ctx->Base);
+   _eglLinkContext(&ctx->Base, _eglLookupDisplay(dpy));
 
    return _eglGetContextHandle(&ctx->Base);
 }
@@ -381,10 +381,8 @@ xlib_eglDestroyContext(_EGLDriver *drv, EGLDisplay dpy, EGLContext ctx)
 {
    struct xlib_egl_context *context = lookup_context(ctx);
    if (context) {
-      if (context->Base.IsBound) {
-         context->Base.DeletePending = EGL_TRUE;
-      }
-      else {
+      _eglUnlinkContext(&context->Base);
+      if (!context->Base.IsBound) {
          /* API-dependent clean-up */
          switch (context->Base.ClientAPI) {
          case EGL_OPENGL_ES_API:
@@ -491,13 +489,13 @@ xlib_eglCreateWindowSurface(_EGLDriver *drv, EGLDisplay dpy, EGLConfig config,
       return EGL_NO_SURFACE;
 
    /* Let EGL lib init the common stuff */
-   if (!_eglInitSurface(drv, dpy, &surf->Base, EGL_WINDOW_BIT,
-                        config, attrib_list)) {
+   if (!_eglInitSurface(drv, &surf->Base, EGL_WINDOW_BIT,
+                        conf, attrib_list)) {
       free(surf);
       return EGL_NO_SURFACE;
    }
 
-   _eglSaveSurface(&surf->Base);
+   _eglLinkSurface(&surf->Base, disp);
 
    /*
     * Now init the Xlib and gallium stuff
@@ -534,11 +532,8 @@ xlib_eglDestroySurface(_EGLDriver *drv, EGLDisplay dpy, EGLSurface surface)
 {
    struct xlib_egl_surface *surf = lookup_surface(surface);
    if (surf) {
-      _eglHashRemove(_eglGlobal.Surfaces, (EGLuint) surface);
-      if (surf->Base.IsBound) {
-         surf->Base.DeletePending = EGL_TRUE;
-      }
-      else {
+      _eglUnlinkSurface(&surf->Base);
+      if (!surf->Base.IsBound) {
          XFreeGC(surf->Dpy, surf->Gc);
          st_unreference_framebuffer(surf->Framebuffer);
          free(surf);