llvmpipe: Texture cache in 4 ubytes instead of 4 floats.

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

diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index b96335b7895bfb8285a68e33c51b86f2682f0dd7..65c3a34c3474a664fbac65042117800174eda9db 100644 (file)
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -25,7 +25,8 @@
  *
  **************************************************************************/
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
 #include "draw/draw_context.h"
 #include "draw/draw_private.h"
 #include "draw/draw_vbuf.h"
@@ -56,9 +57,11 @@ struct pt_fetch {
  *
  */
 void draw_pt_fetch_prepare( struct pt_fetch *fetch,
+                            unsigned vs_input_count,
                            unsigned vertex_size )
 {
    struct draw_context *draw = fetch->draw;
+   unsigned nr_inputs;
    unsigned i, nr = 0;
    unsigned dst_offset = 0;
    struct translate_key key;
@@ -89,8 +92,11 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
       dst_offset += 4 * sizeof(float);
    }
       
+   assert( draw->pt.nr_vertex_elements >= vs_input_count );
 
-   for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
+   nr_inputs = MIN2( vs_input_count, draw->pt.nr_vertex_elements );
+
+   for (i = 0; i < nr_inputs; i++) {
       key.element[nr].input_format = draw->pt.vertex_element[i].src_format;
       key.element[nr].input_buffer = draw->pt.vertex_element[i].vertex_buffer_index;
       key.element[nr].input_offset = draw->pt.vertex_element[i].src_offset;
@@ -114,7 +120,7 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
       fetch->translate = translate_cache_find(fetch->cache, &key);
 
       {
-        static struct vertex_header vh = { 0, 1, 0, 0xffff };
+         static struct vertex_header vh = { 0, 1, 0, UNDEFINED_VERTEX_ID, { .0f, .0f, .0f, .0f } };
         fetch->translate->set_buffer(fetch->translate,
                                      draw->pt.nr_vertex_buffers,
                                      &vh,
@@ -144,7 +150,7 @@ void draw_pt_fetch_run( struct pt_fetch *fetch,
                            i, 
                            ((char *)draw->pt.user.vbuffer[i] + 
                             draw->pt.vertex_buffer[i].buffer_offset),
-                           draw->pt.vertex_buffer[i].pitch );
+                           draw->pt.vertex_buffer[i].stride );
    }
 
    translate->run_elts( translate,
@@ -166,6 +172,42 @@ void draw_pt_fetch_run( struct pt_fetch *fetch,
 }
 
 
+void draw_pt_fetch_run_linear( struct pt_fetch *fetch,
+                               unsigned start,
+                               unsigned count,
+                               char *verts )
+{
+   struct draw_context *draw = fetch->draw;
+   struct translate *translate = fetch->translate;
+   unsigned i;
+
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      translate->set_buffer(translate,
+                           i,
+                           ((char *)draw->pt.user.vbuffer[i] +
+                            draw->pt.vertex_buffer[i].buffer_offset),
+                           draw->pt.vertex_buffer[i].stride );
+   }
+
+   translate->run( translate,
+                   start,
+                   count,
+                   verts );
+
+   /* Edgeflags are hard to fit into a translate program, populate
+    * them separately if required.  In the setup above they are
+    * defaulted to one, so only need this if there is reason to change
+    * that default:
+    */
+   if (fetch->need_edgeflags) {
+      for (i = 0; i < count; i++) {
+         struct vertex_header *vh = (struct vertex_header *)(verts + i * fetch->vertex_size);
+         vh->edgeflag = draw_pt_get_edgeflag( draw, start + i );
+      }
+   }
+}
+
+
 struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw )
 {
    struct pt_fetch *fetch = CALLOC_STRUCT(pt_fetch);