Implement draw_arrays_instanced() in softpipe.

Modify the translate module to respect instance divisors and accept
instance id as a parameter to calculate input vertex offset.

9 files changed, 51 insertions, 1 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index 465b8f10c6..c0f6a61411 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -151,6 +151,14 @@ void draw_set_mapped_constant_buffer(struct draw_context *draw,
 void draw_arrays(struct draw_context *draw, unsigned prim,
 		 unsigned start, unsigned count);
 
+void
+draw_arrays_instanced(struct draw_context *draw,
+                      unsigned mode,
+                      unsigned start,
+                      unsigned count,
+                      unsigned startInstance,
+                      unsigned instanceCount);
+
 void draw_flush(struct draw_context *draw);
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 1a5269c0de..bb8a8ff491 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -138,7 +138,7 @@ emit_vertex( struct vbuf_stage *vbuf,
       /* Note: we really do want data[0] here, not data[pos]: 
        */
       vbuf->translate->set_buffer(vbuf->translate, 0, vertex->data[0], 0);
-      vbuf->translate->run(vbuf->translate, 0, 1, vbuf->vertex_ptr);
+      vbuf->translate->run(vbuf->translate, 0, 1, 0, vbuf->vertex_ptr);
 
       if (0) draw_dump_emitted_vertex(vbuf->vinfo, (uint8_t *)vbuf->vertex_ptr);
       
@@ -275,6 +275,7 @@ vbuf_start_prim( struct vbuf_stage *vbuf, uint prim )
       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;
+      hw_key.element[i].instance_divisor = 0;
       hw_key.element[i].output_format = output_format;
       hw_key.element[i].output_offset = dst_offset;
 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 3850cede1e..129d919a84 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -226,6 +226,8 @@ struct draw_context
 
    unsigned reduced_prim;
 
+   unsigned instance_id;
+
    void *driver_private;
 };
 
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 2801dbafe4..1217b9e5d7 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -312,5 +312,28 @@ draw_arrays(struct draw_context *draw, unsigned prim,
 #endif
 
    /* drawing done here: */
+   draw->instance_id = 0;
    draw_pt_arrays(draw, prim, start, count);
 }
+
+void
+draw_arrays_instanced(struct draw_context *draw,
+                      unsigned mode,
+                      unsigned start,
+                      unsigned count,
+                      unsigned startInstance,
+                      unsigned instanceCount)
+{
+   unsigned reduced_prim = u_reduced_prim(mode);
+   unsigned instance;
+
+   if (reduced_prim != draw->reduced_prim) {
+      draw_do_flush(draw, DRAW_FLUSH_STATE_CHANGE);
+      draw->reduced_prim = reduced_prim;
+   }
+
+   for (instance = 0; instance < instanceCount; instance++) {
+      draw->instance_id = instance + startInstance;
+      draw_pt_arrays(draw, mode, start, count);
+   }
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index 064e16c295..d0abeb9336 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -125,6 +125,7 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
       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;
+      hw_key.element[i].instance_divisor = 0;
       hw_key.element[i].output_format = output_format;
       hw_key.element[i].output_offset = dst_offset;
 
@@ -204,6 +205,7 @@ void draw_pt_emit( struct pt_emit *emit,
    translate->run( translate,
 		   0, 
 		   vertex_count,
+                   draw->instance_id,
 		   hw_verts );
 
    render->unmap_vertices( render, 
@@ -263,6 +265,7 @@ void draw_pt_emit_linear(struct pt_emit *emit,
    translate->run(translate,
                   0,
                   count,
+                  draw->instance_id,
                   hw_verts);
 
    if (0) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index 305bfef435..e8174a2971 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -81,6 +81,7 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
       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;
+      key.element[nr].instance_divisor = 0;
       key.element[nr].output_format = PIPE_FORMAT_R32_FLOAT;
       key.element[nr].output_offset = dst_offset;
       dst_offset += 1 * sizeof(float);
@@ -100,6 +101,7 @@ void draw_pt_fetch_prepare( struct pt_fetch *fetch,
       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;
       key.element[nr].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       key.element[nr].output_offset = dst_offset;
 
@@ -183,6 +185,7 @@ void draw_pt_fetch_run_linear( struct pt_fetch *fetch,
    translate->run( translate,
                    start,
                    count,
+                   draw->instance_id,
                    verts );
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index e7fe6b3b76..40bfc0fbb2 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -169,6 +169,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
       key.element[i].input_format = input_format;
       key.element[i].input_buffer = input_buffer;
       key.element[i].input_offset = input_offset;
+      key.element[i].instance_divisor = src->instance_divisor;
       key.element[i].output_format = output_format;
       key.element[i].output_offset = dst_offset;
       
@@ -314,6 +315,7 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->instance_id,
                          hw_verts );
 
    if (0) {
@@ -374,6 +376,7 @@ static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->instance_id,
                          hw_verts );
 
    draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) );
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index e3b807ebd0..00036cfe68 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -43,6 +43,7 @@ struct draw_varient_input
    enum pipe_format format;
    unsigned buffer;
    unsigned offset; 
+   unsigned instance_divisor;
 };
 
 struct draw_varient_output
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 7ee567d478..4cc080f803 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -180,6 +180,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
 
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->instance_id,
                     output_buffer );
 
    FREE(temp_buffer);
@@ -202,6 +203,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
    vsvg->fetch->run( vsvg->fetch, 
                      start,
                      count,
+                     vsvg->draw->instance_id,
                      temp_buffer );
 
    vsvg->base.vs->run_linear( vsvg->base.vs, 
@@ -238,6 +240,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
    
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->instance_id,
                     output_buffer );
 
    FREE(temp_buffer);
@@ -283,6 +286,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
       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;
+      fetch.element[i].instance_divisor = 0;
       fetch.element[i].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
       fetch.element[i].output_offset = i * 4 * sizeof(float);
       assert(fetch.element[i].output_offset < fetch.output_stride);
@@ -297,6 +301,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
          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);
+         emit.element[i].instance_divisor = 0;
          emit.element[i].output_format = draw_translate_vinfo_format(key->element[i].out.format);
          emit.element[i].output_offset = key->element[i].out.offset;
          assert(emit.element[i].input_offset <= fetch.output_stride);
@@ -305,6 +310,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
          emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
          emit.element[i].input_buffer = 1;
          emit.element[i].input_offset = 0;
+         emit.element[i].instance_divisor = 0;
          emit.element[i].output_format = PIPE_FORMAT_R32_FLOAT;
          emit.element[i].output_offset = key->element[i].out.offset;
       }