gallium: a lot more complete implementation of stream output

[mesa.git] / src / gallium / auxiliary / draw / draw_vs_sse.c

diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 0e2036f12a3c37d26a1fea29337b00ba4da3a072..14c95082a9df467634e4fcf517c2c3db3b7b0066 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -37,7 +37,7 @@
 
 #include "draw_vs.h"
 
-#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE)
+#if defined(PIPE_ARCH_X86)
 
 #include "pipe/p_shader_tokens.h"
 
@@ -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;
 };
@@ -78,6 +67,10 @@ static void
 vs_sse_prepare( struct draw_vertex_shader *base,
                struct draw_context *draw )
 {
+   struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
+   struct tgsi_exec_machine *machine = shader->machine;
+
+   machine->Samplers = draw->vs.samplers;
 }
 
 
@@ -90,7 +83,7 @@ static void
 vs_sse_run_linear( struct draw_vertex_shader *base,
                   const float (*input)[4],
                   float (*output)[4],
-                  const float (*constants)[4],
+                  const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
                   unsigned count,
                   unsigned input_stride,
                   unsigned output_stride )
@@ -118,11 +111,9 @@ 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,
+                   (const float (*)[4])constants[0],
+                  shader->base.immediates,
                    input,
                    base->info.num_inputs,
                    input_stride,
@@ -174,9 +165,9 @@ draw_create_vs_sse(struct draw_context *draw,
 
    vs->base.draw = draw;
    if (1)
-      vs->base.create_varient = draw_vs_varient_aos_sse;
+      vs->base.create_varient = draw_vs_create_varient_aos_sse;
    else
-      vs->base.create_varient = draw_vs_varient_generic;
+      vs->base.create_varient = draw_vs_create_varient_generic;
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
@@ -184,7 +175,7 @@ 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 +185,7 @@ 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;
    }