4 files changed, 36 insertions, 13 deletions

diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index 701ee4c72f..f472dd0ed2 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -39,11 +39,19 @@
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_parse.h"
 
-struct sp_exec_fragment_shader {
+struct sp_exec_fragment_shader
+{
    struct sp_fragment_shader base;
 };
 
 
+/** cast wrapper */
+static INLINE struct sp_exec_fragment_shader *
+sp_exec_fragment_shader(const struct sp_fragment_shader *base)
+{
+   return (struct sp_exec_fragment_shader *) base;
+}
+
 
 /**
  * Compute quad X,Y,Z,W for the four fragments in a quad.
@@ -86,10 +94,16 @@ exec_prepare( const struct sp_fragment_shader *base,
 	      struct tgsi_exec_machine *machine,
 	      struct tgsi_sampler *samplers )
 {
-   tgsi_exec_machine_bind_shader( machine,
-				  base->shader.tokens,
-				  PIPE_MAX_SAMPLERS,
-				  samplers );
+   /*
+    * Bind tokens/shader to the interpreter's machine state.
+    * Avoid redundant binding.
+    */
+   if (machine->Tokens != base->shader.tokens) {
+      tgsi_exec_machine_bind_shader( machine,
+                                     base->shader.tokens,
+                                     PIPE_MAX_SAMPLERS,
+                                     samplers );
+   }
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index 496ed43df2..31908a517b 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -40,7 +40,7 @@
 #include "tgsi/tgsi_sse2.h"
 
 
-#ifdef PIPE_ARCH_X86
+#if defined(PIPE_ARCH_X86)
 
 #include "rtasm/rtasm_x86sse.h"
 
@@ -92,7 +92,8 @@ fs_sse_run( const struct sp_fragment_shader *base,
 		       machine->Temps);
 
    /* init kill mask */
-   machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] = 0x0;
+   tgsi_set_kill_mask(machine, 0x0);
+   tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 
    shader->func( machine->Inputs,
 		 machine->Outputs,
diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c
index d05e12d1d9..b7aac7f84a 100644
--- a/src/gallium/drivers/softpipe/sp_quad_output.c
+++ b/src/gallium/drivers/softpipe/sp_quad_output.c
@@ -64,6 +64,14 @@ output_quad(struct quad_stage *qs, struct quad_header *quad)
             for (i = 0; i < 4; i++) { /* loop over color chans */
                tile->data.color[y][x][i] = quadColor[i][j];
             }
+            if (0) {
+               debug_printf("sp write pixel %d,%d: %g, %g, %g\n",
+                            quad->input.x0 + x,
+                            quad->input.y0 + y,
+                            quadColor[0][j],
+                            quadColor[1][j],
+                            quadColor[2][j]);
+            }
          }
       }
    }
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index bc8263c33e..13d8017393 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -773,10 +773,10 @@ static void setup_tri_coefficients( struct setup_context *setup )
    /* setup interpolation for all the remaining attributes:
     */
    for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
+      const uint vertSlot = vinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->interp_mode[fragSlot]) {
+      switch (vinfo->attrib[fragSlot].interp_mode) {
       case INTERP_CONSTANT:
          for (j = 0; j < NUM_CHANNELS; j++)
             const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
@@ -1084,10 +1084,10 @@ setup_line_coefficients(struct setup_context *setup,
    /* setup interpolation for all the remaining attributes:
     */
    for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
+      const uint vertSlot = vinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->interp_mode[fragSlot]) {
+      switch (vinfo->attrib[fragSlot].interp_mode) {
       case INTERP_CONSTANT:
          for (j = 0; j < NUM_CHANNELS; j++)
             const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
@@ -1331,10 +1331,10 @@ setup_point( struct setup_context *setup,
    const_coeff(setup, &setup->posCoef, 0, 3);
 
    for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) {
-      const uint vertSlot = vinfo->src_index[fragSlot];
+      const uint vertSlot = vinfo->attrib[fragSlot].src_index;
       uint j;
 
-      switch (vinfo->interp_mode[fragSlot]) {
+      switch (vinfo->attrib[fragSlot].interp_mode) {
       case INTERP_CONSTANT:
          /* fall-through */
       case INTERP_LINEAR: