Add lame support for instanceID to draw module.

It's all screaming for integer support -- fake it with float for now.

7 files changed, 46 insertions, 13 deletions

diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index bb8a8ff491..d40c035240 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -271,7 +271,8 @@ vbuf_start_prim( struct vbuf_stage *vbuf, uint prim )
 	 emit_sz = 0;
 	 break;
       }
-      
+
+      hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       hw_key.element[i].input_buffer = src_buffer;
       hw_key.element[i].input_offset = src_offset;
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
index 20edf7a227..d5e0d92a60 100644
--- a/src/gallium/auxiliary/draw/draw_pt.h
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -183,7 +183,8 @@ struct pt_emit *draw_pt_emit_create( struct draw_context *draw );
 struct pt_fetch;
 void draw_pt_fetch_prepare( struct pt_fetch *fetch,
                             unsigned vertex_input_count,
-			    unsigned vertex_size );
+                            unsigned vertex_size,
+                            unsigned instance_id_index );
 
 void draw_pt_fetch_run( struct pt_fetch *fetch,
 			const unsigned *elts,
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index d0abeb9336..4fb53276bb 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -121,7 +121,8 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
 	 emit_sz = 0;
 	 break;
       }
-      
+
+      hw_key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       hw_key.element[i].input_buffer = src_buffer;
       hw_key.element[i].input_offset = src_offset;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index f88a839f61..36c27e22ff 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -58,12 +58,14 @@ struct pt_fetch {
  */
 void draw_pt_fetch_prepare( struct pt_fetch *fetch,
                             unsigned vs_input_count,
-			    unsigned vertex_size )
+                            unsigned vertex_size,
+                            unsigned instance_id_index )
 {
    struct draw_context *draw = fetch->draw;
    unsigned nr_inputs;
-   unsigned i, nr = 0;
+   unsigned i, nr = 0, ei = 0;
    unsigned dst_offset = 0;
+   unsigned num_extra_inputs = 0;
    struct translate_key key;
 
    fetch->vertex_size = vertex_size;
@@ -78,6 +80,7 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
    {
       /* Need to set header->vertex_id = 0xffff somehow.
        */
+      key.element[nr].type = TRANSLATE_ELEMENT_NORMAL;
       key.element[nr].input_format = PIPE_FORMAT_R32_FLOAT;
       key.element[nr].input_buffer = draw->pt.nr_vertex_buffers;
       key.element[nr].input_offset = 0;
@@ -92,16 +95,27 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
        */
       dst_offset += 4 * sizeof(float);
    }
-      
-   assert( draw->pt.nr_vertex_elements >= vs_input_count );
 
-   nr_inputs = MIN2( vs_input_count, draw->pt.nr_vertex_elements );
+   if (instance_id_index != ~0) {
+      num_extra_inputs++;
+   }
+
+   assert(draw->pt.nr_vertex_elements + num_extra_inputs >= vs_input_count);
+
+   nr_inputs = MIN2(vs_input_count, draw->pt.nr_vertex_elements + num_extra_inputs);
 
    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;
-      key.element[nr].instance_divisor = draw->pt.vertex_element[i].instance_divisor;
+      if (i == instance_id_index) {
+         key.element[nr].type = TRANSLATE_ELEMENT_INSTANCE_ID;
+         key.element[nr].input_format = PIPE_FORMAT_R32_FLOAT;    /* XXX: Make it UINT. */
+      } else {
+         key.element[nr].type = TRANSLATE_ELEMENT_NORMAL;
+         key.element[nr].input_format = draw->pt.vertex_element[ei].src_format;
+         key.element[nr].input_buffer = draw->pt.vertex_element[ei].vertex_buffer_index;
+         key.element[nr].input_offset = draw->pt.vertex_element[ei].src_offset;
+         key.element[nr].instance_divisor = draw->pt.vertex_element[ei].instance_divisor;
+         ei++;
+      }
       key.element[nr].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       key.element[nr].output_offset = dst_offset;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 771d94b973..2a604470e9 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -166,6 +166,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
 	 continue;
       }
 
+      key.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       key.element[i].input_format = input_format;
       key.element[i].input_buffer = input_buffer;
       key.element[i].input_offset = input_offset;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
index 932113783d..0238f2e234 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -58,6 +58,8 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
    struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
    struct draw_context *draw = fpme->draw;
    struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+   unsigned i;
+   boolean instance_id_index = ~0;
 
    /* Add one to num_outputs because the pipeline occasionally tags on
     * an additional texcoord, eg for AA lines.
@@ -65,6 +67,15 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
    unsigned nr = MAX2( vs->info.num_inputs,
 		       vs->info.num_outputs + 1 );
 
+   /* Scan for instanceID system value.
+    */
+   for (i = 0; i < vs->info.num_inputs; i++) {
+      if (vs->info.input_semantic_name[i] == TGSI_SEMANTIC_INSTANCEID) {
+         instance_id_index = i;
+         break;
+      }
+   }
+
    fpme->prim = prim;
    fpme->opt = opt;
 
@@ -78,7 +89,8 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
 
    draw_pt_fetch_prepare( fpme->fetch, 
                           vs->info.num_inputs,
-			  fpme->vertex_size );
+                          fpme->vertex_size,
+                          instance_id_index );
    /* XXX: it's not really gl rasterization rules we care about here,
     * but gl vs dx9 clip spaces.
     */
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 8e14bdd6bd..60b7a3ea36 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -284,6 +284,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    fetch.nr_elements = key->nr_inputs;
    fetch.output_stride = vsvg->temp_vertex_stride;
    for (i = 0; i < key->nr_inputs; i++) {
+      fetch.element[i].type = TRANSLATE_ELEMENT_NORMAL;
       fetch.element[i].input_format = key->element[i].in.format;
       fetch.element[i].input_buffer = key->element[i].in.buffer;
       fetch.element[i].input_offset = key->element[i].in.offset;
@@ -299,6 +300,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    for (i = 0; i < key->nr_outputs; i++) {
       if (key->element[i].out.format != EMIT_1F_PSIZE)
       {      
+         emit.element[i].type = TRANSLATE_ELEMENT_NORMAL;
          emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
          emit.element[i].input_buffer = 0;
          emit.element[i].input_offset = key->element[i].out.vs_output * 4 * sizeof(float);
@@ -308,6 +310,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
          assert(emit.element[i].input_offset <= fetch.output_stride);
       }
       else {
+         emit.element[i].type = TRANSLATE_ELEMENT_NORMAL;
          emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
          emit.element[i].input_buffer = 1;
          emit.element[i].input_offset = 0;