gallium: basic and initial implementation of the stream output interface

aka transform feedback

4 files changed, 111 insertions, 0 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 61980c3e4f5..7c7702549e0 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -574,3 +574,25 @@ draw_get_rasterizer_no_cull( struct draw_context *draw,
    }
    return draw->rasterizer_no_cull[scissor][flatshade];
 }
+
+void
+draw_set_mapped_so_buffers(struct draw_context *draw,
+                           void *buffers[PIPE_MAX_SO_BUFFERS],
+                           unsigned num_buffers)
+{
+   int i;
+
+   for (i = 0; i < num_buffers; ++i) {
+      draw->so.buffers[i] = buffers[i];
+   }
+   draw->so.num_buffers = num_buffers;
+}
+
+void
+draw_set_so_state(struct draw_context *draw,
+                  struct pipe_stream_output_state *state)
+{
+   memcpy(&draw->so.state,
+          state,
+          sizeof(struct pipe_stream_output_state));
+}
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index b905c2f2da6..103d6538b81 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -162,6 +162,14 @@ draw_set_mapped_constant_buffer(struct draw_context *draw,
                                 const void *buffer,
                                 unsigned size);
 
+void
+draw_set_mapped_so_buffers(struct draw_context *draw,
+                           void *buffers[PIPE_MAX_SO_BUFFERS],
+                           unsigned num_buffers);
+void
+draw_set_so_state(struct draw_context *draw,
+                  struct pipe_stream_output_state *state);
+
 
 /***********************************************************************
  * draw_prim.c 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index a2bfb693c09..ca8f9cfab1e 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -235,6 +235,12 @@ struct draw_context
       struct tgsi_sampler **samplers;
    } gs;
 
+   struct {
+      struct pipe_stream_output_state state;
+      void *buffers[PIPE_MAX_SO_BUFFERS];
+      uint num_buffers;
+   } so;
+
    /* Clip derived state:
     */
    float plane[12][4];
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index f623c0743da..3b4237245e6 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -33,11 +33,13 @@
 #include "draw/draw_pt.h"
 #include "translate/translate.h"
 #include "translate/translate_cache.h"
+#include "util/u_format.h"
 
 struct pt_emit {
    struct draw_context *draw;
 
    struct translate *translate;
+   struct translate *so_translate;
 
    struct translate_cache *cache;
    unsigned prim;
@@ -45,6 +47,51 @@ struct pt_emit {
    const struct vertex_info *vinfo;
 };
 
+static void
+prepare_so_emit( struct pt_emit *emit,
+                 const struct vertex_info *vinfo )
+{
+   struct draw_context *draw = emit->draw;
+   unsigned i;
+   struct translate_key hw_key;
+   unsigned dst_offset = 0;
+   unsigned output_stride = 0;
+   boolean has_so = (draw->so.state.num_outputs > 0);
+
+   if (has_so) {
+
+      for (i = 0; i < draw->so.state.num_outputs; ++i) {
+         unsigned src_offset = (vinfo->attrib[i].src_index * 4 * sizeof(float) );
+         unsigned output_format = draw->so.state.format[i];
+         unsigned output_bytes = util_format_get_blocksize(output_format);
+
+         hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
+         hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+         hw_key.element[i].input_buffer = 0;
+         hw_key.element[i].input_offset = src_offset;
+         hw_key.element[i].instance_divisor = 0;
+         hw_key.element[i].output_format = output_format;
+         hw_key.element[i].output_offset = dst_offset;
+
+         dst_offset += output_bytes;
+         output_stride += output_bytes;
+      }
+      hw_key.nr_elements = draw->so.state.num_outputs;
+      hw_key.output_stride = output_stride;
+
+      if (!emit->so_translate ||
+          translate_key_compare(&emit->so_translate->key, &hw_key) != 0)
+      {
+         translate_key_sanitize(&hw_key);
+         emit->so_translate = translate_cache_find(emit->cache, &hw_key);
+      }
+   } else {
+      /* no stream output */
+      emit->so_translate = NULL;
+   }
+}
+
+
 void draw_pt_emit_prepare( struct pt_emit *emit,
 			   unsigned prim,
                            unsigned *max_vertices )
@@ -121,6 +168,8 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
    *max_vertices = (draw->render->max_vertex_buffer_bytes / 
                     (vinfo->size * 4));
 
+   prepare_so_emit( emit, vinfo );
+
    /* even number */
    *max_vertices = *max_vertices & ~1;
 }
@@ -135,6 +184,7 @@ void draw_pt_emit( struct pt_emit *emit,
 {
    struct draw_context *draw = emit->draw;
    struct translate *translate = emit->translate;
+   struct translate *so_translate = emit->so_translate;
    struct vbuf_render *render = draw->render;
    void *hw_verts;
 
@@ -186,6 +236,18 @@ void draw_pt_emit( struct pt_emit *emit,
                    draw->instance_id,
 		   hw_verts );
 
+   if (so_translate) {
+      void *so_buffer = draw->so.buffers[0];
+
+      /* XXX we only support single output buffer right now */
+      debug_assert(draw->so.num_buffers >= 0);
+
+      so_translate->set_buffer(translate, 0, vertex_data,
+                               stride, ~0);
+      so_translate->run(translate, 0, vertex_count,
+                        draw->instance_id, so_buffer);
+   }
+
    render->unmap_vertices( render, 
                            0, 
                            vertex_count - 1 );
@@ -205,6 +267,7 @@ void draw_pt_emit_linear(struct pt_emit *emit,
 {
    struct draw_context *draw = emit->draw;
    struct translate *translate = emit->translate;
+   struct translate *so_translate = emit->so_translate;
    struct vbuf_render *render = draw->render;
    void *hw_verts;
 
@@ -246,6 +309,18 @@ void draw_pt_emit_linear(struct pt_emit *emit,
                   draw->instance_id,
                   hw_verts);
 
+   if (so_translate) {
+      void *so_buffer = draw->so.buffers[0];
+
+      /* XXX we only support single output buffer right now */
+      debug_assert(draw->so.num_buffers >= 0);
+
+      so_translate->set_buffer(translate, 0,
+                               vertex_data, stride, count - 1);
+      so_translate->run(translate, 0, count,
+                        draw->instance_id, so_buffer);
+   }
+
    if (0) {
       unsigned i;
       for (i = 0; i < count; i++) {