Redoing the way we handle vertex shaders for the draw module.

5 files changed, 65 insertions, 18 deletions

diff --git a/src/mesa/pipe/draw/draw_context.c b/src/mesa/pipe/draw/draw_context.c
index 66c66ff698..5efb173228 100644
--- a/src/mesa/pipe/draw/draw_context.c
+++ b/src/mesa/pipe/draw/draw_context.c
@@ -155,14 +155,6 @@ void draw_set_viewport_state( struct draw_context *draw,
 }
 
 
-void
-draw_set_vertex_shader(struct draw_context *draw,
-                       const struct pipe_shader_state *shader)
-{
-   draw_flush( draw );
-   draw->vertex_shader = *shader;
-}
-
 
 void
 draw_set_vertex_buffer(struct draw_context *draw,
diff --git a/src/mesa/pipe/draw/draw_context.h b/src/mesa/pipe/draw/draw_context.h
index 0ccf5f6046..398e96d94d 100644
--- a/src/mesa/pipe/draw/draw_context.h
+++ b/src/mesa/pipe/draw/draw_context.h
@@ -92,9 +92,12 @@ void draw_set_rasterize_stage( struct draw_context *draw,
                                struct draw_stage *stage );
 
 
-void
-draw_set_vertex_shader(struct draw_context *draw,
-                       const struct pipe_shader_state *shader);
+void * draw_create_vertex_shader(struct draw_context *draw,
+                                 const struct pipe_shader_state *shader);
+void draw_bind_vertex_shader(struct draw_context *draw,
+                             void *vcso);
+void draw_delete_vertex_shader(struct draw_context *draw,
+                               void *vcso);
 
 
 void
diff --git a/src/mesa/pipe/draw/draw_private.h b/src/mesa/pipe/draw/draw_private.h
index ebef5347ab..2f52299f74 100644
--- a/src/mesa/pipe/draw/draw_private.h
+++ b/src/mesa/pipe/draw/draw_private.h
@@ -46,6 +46,7 @@
 
 #include "draw_vertex.h"
 
+#include "x86/rtasm/x86sse.h"
 
 /**
  * Basic vertex info.
@@ -104,7 +105,7 @@ struct draw_stage
 
    void (*tri)( struct draw_stage *,
 		struct prim_header * );
-   
+
    void (*end)( struct draw_stage * );
 
    void (*reset_stipple_counter)( struct draw_stage * );
@@ -116,6 +117,15 @@ struct draw_stage
 #define VCACHE_OVERFLOW        4
 #define VS_QUEUE_LENGTH        (VCACHE_SIZE + VCACHE_OVERFLOW + 1)	/* can never fill up */
 
+/**
+ * Private version of the compiled vertex_shader
+ */
+struct draw_vertex_shader {
+   const struct pipe_shader_state *state;
+#if defined(__i386__) || defined(__386__)
+   struct x86_function  sse2_program;
+#endif
+};
 
 /**
  * Private context for the drawing module.
@@ -145,7 +155,7 @@ struct draw_context
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
-   struct pipe_shader_state vertex_shader;
+   const struct draw_vertex_shader *vertex_shader;
    struct pipe_vertex_buffer feedback_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element feedback_element[PIPE_ATTRIB_MAX];
 
diff --git a/src/mesa/pipe/draw/draw_vertex_fetch.c b/src/mesa/pipe/draw/draw_vertex_fetch.c
index 2b839d641e..bc37e18c34 100644
--- a/src/mesa/pipe/draw/draw_vertex_fetch.c
+++ b/src/mesa/pipe/draw/draw_vertex_fetch.c
@@ -85,7 +85,7 @@ void draw_vertex_fetch( struct draw_context *draw,
       /*printf("fetch vertex %u: \n", j);*/
 
       /* loop over vertex attributes (vertex shader inputs) */
-      for (attr = 0; attr < draw->vertex_shader.num_inputs; attr++) {
+      for (attr = 0; attr < draw->vertex_shader->state->num_inputs; attr++) {
 
          unsigned buf = draw->vertex_element[attr].vertex_buffer_index;
          const void *src
diff --git a/src/mesa/pipe/draw/draw_vertex_shader.c b/src/mesa/pipe/draw/draw_vertex_shader.c
index a2e1cdc472..f7ddbd1909 100644
--- a/src/mesa/pipe/draw/draw_vertex_shader.c
+++ b/src/mesa/pipe/draw/draw_vertex_shader.c
@@ -36,6 +36,8 @@
 #include "draw_context.h"
 #include "draw_vertex.h"
 
+#include "x86/rtasm/x86sse.h"
+
 #include "pipe/tgsi/exec/tgsi_core.h"
 
 static INLINE unsigned
@@ -70,6 +72,7 @@ typedef void (XSTDCALL *codegen_function) (
    float (*constant)[4],
    struct tgsi_exec_vector *temporary );
 
+
 /**
  * Transform vertices with the current vertex program/shader
  * Up to four vertices can be shaded at a time.
@@ -92,7 +95,7 @@ run_vertex_program(struct draw_context *draw,
    const float *trans = draw->viewport.translate;
 
    assert(count <= 4);
-   assert(draw->vertex_shader.output_semantic_name[0]
+   assert(draw->vertex_shader->state->output_semantic_name[0]
           == TGSI_SEMANTIC_POSITION);
 
 #ifdef DEBUG
@@ -101,7 +104,7 @@ run_vertex_program(struct draw_context *draw,
 
    /* init machine state */
    tgsi_exec_machine_init(&machine,
-                          draw->vertex_shader.tokens,
+                          draw->vertex_shader->state->tokens,
                           PIPE_MAX_SAMPLERS,
                           NULL /*samplers*/ );
 
@@ -114,8 +117,8 @@ run_vertex_program(struct draw_context *draw,
    draw_vertex_fetch( draw, &machine, elts, count );
 
    /* run shader */
-   if( draw->vertex_shader.executable != NULL ) {
-      codegen_function func = (codegen_function) draw->vertex_shader.executable;
+   if( draw->vertex_shader->state->executable != NULL ) {
+      codegen_function func = (codegen_function) draw->vertex_shader->state->executable;
       func(
          machine.Inputs,
          machine.Outputs,
@@ -206,3 +209,42 @@ void draw_vertex_shader_queue_flush( struct draw_context *draw )
    draw->vs.queue_nr = 0;
 }
 
+
+void *
+draw_create_vertex_shader(struct draw_context *draw,
+                          const struct pipe_shader_state *shader)
+{
+   struct draw_vertex_shader *vs = calloc(1, sizeof(struct draw_vertex_shader));
+
+   vs->state = shader;
+#if defined(__i386__) || defined(__386__)
+   x86_init_func(&vs->sse2_program);
+
+   tgsi_emit_sse2(shader->tokens, &vs->sse2_program);
+
+   ((struct pipe_shader_state*)(vs->state))->executable =
+      x86_get_func(&vs->sse2_program);
+#endif
+
+   return vs;
+}
+
+void draw_bind_vertex_shader(struct draw_context *draw,
+                             void *vcso)
+{
+   draw_flush(draw);
+   draw->vertex_shader = (struct draw_vertex_shader*)(vcso);
+}
+
+void draw_delete_vertex_shader(struct draw_context *draw,
+                               void *vcso)
+{
+   struct draw_vertex_shader *vs = (struct draw_vertex_shader*)(vcso);
+#if defined(__i386__) || defined(__386__)
+   x86_release_func(&vs->sse2_program);
+#endif
+   free(vcso);
+}
+
+
+