draw/translate: fix instancing

We were incorrectly computing the buffer offset when using the
instances. The buffer offset is always equal to:
start_instance * stride + (instance_num / instance_divisor) *
stride
We were completely ignoring the start instance quite
often producing instances that completely wrong, e.g. if
start instance = 5, instance divisor = 2, then on the first
iteration it should be:
5 * stride, not (5/2) * stride as we'd have currently, and if
start instance = 1, instance divisor = 3, then on the first
iteration it should be:
1 * stride, not 0 as we'd have.
This fixes it and adjusts all the code to the changes.

Signed-off-by: Zack Rusin <[email protected]>

9 files changed, 47 insertions, 9 deletions

diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 97b463f4ff8..f9bcadc6cfa 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -674,6 +674,7 @@ generate_vs(struct draw_llvm_variant *variant,
 
 static void
 generate_fetch(struct gallivm_state *gallivm,
+               struct draw_context *draw,
                LLVMValueRef vbuffers_ptr,
                LLVMValueRef *res,
                struct pipe_vertex_element *velem,
@@ -704,10 +705,17 @@ generate_fetch(struct gallivm_state *gallivm,
    struct lp_build_if_state if_ctx;
 
    if (velem->instance_divisor) {
-      /* array index = instance_id / instance_divisor */
-      index = LLVMBuildUDiv(builder, instance_id,
-                            lp_build_const_int32(gallivm, velem->instance_divisor),
-                            "instance_divisor");
+      /* Index is equal to the start instance plus the number of current 
+       * instance divided by the divisor. In this case we compute it as:
+       * index = start_instance + ((instance_id - start_instance) / divisor)
+       */
+      LLVMValueRef current_instance;
+      index = lp_build_const_int32(gallivm, draw->start_instance);
+      current_instance = LLVMBuildSub(builder, instance_id, index, "");
+      current_instance = LLVMBuildUDiv(builder, current_instance,
+                                       lp_build_const_int32(gallivm, velem->instance_divisor),
+                                       "instance_divisor");
+      index = LLVMBuildAdd(builder, index, current_instance, "instance");
    }
 
    stride = lp_build_umul_overflow(gallivm, vb_stride, index, &ofbit);
@@ -1697,7 +1705,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant,
             LLVMValueRef vb_index =
                lp_build_const_int32(gallivm, velem->vertex_buffer_index);
             LLVMValueRef vb = LLVMBuildGEP(builder, vb_ptr, &vb_index, 1, "");
-            generate_fetch(gallivm, vbuffers_ptr,
+            generate_fetch(gallivm, draw, vbuffers_ptr,
                            &aos_attribs[j][i], velem, vb, true_index,
                            system_values.instance_id);
          }
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 578433c2006..d3b38eb2df6 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, ~0);
-      vbuf->translate->run(vbuf->translate, 0, 1, 0, vbuf->vertex_ptr);
+      vbuf->translate->run(vbuf->translate, 0, 1, 0, 0, vbuf->vertex_ptr);
 
       if (0) draw_dump_emitted_vertex(vbuf->vinfo, (uint8_t *)vbuf->vertex_ptr);
       
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index fd52c2d6b4c..f42cded118a 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -306,6 +306,7 @@ struct draw_context
    } extra_shader_outputs;
 
    unsigned instance_id;
+   unsigned start_instance;
 
 #ifdef HAVE_LLVM
    struct draw_llvm *llvm;
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index c4d06de84bb..e89ccd25401 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -533,6 +533,7 @@ draw_vbo(struct draw_context *draw,
 
    for (instance = 0; instance < info->instance_count; instance++) {
       draw->instance_id = instance + info->start_instance;
+      draw->start_instance = info->start_instance;
       /* check for overflow */
       if (draw->instance_id < instance ||
           draw->instance_id < info->start_instance) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index 4c96d7414ee..fc64048be1b 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -171,6 +171,7 @@ draw_pt_emit(struct pt_emit *emit,
    translate->run(translate,
 		  0,
 		  vertex_count,
+                  draw->start_instance,
                   draw->instance_id,
 		  hw_verts );
 
@@ -234,6 +235,7 @@ draw_pt_emit_linear(struct pt_emit *emit,
    translate->run(translate,
                   0,
                   count,
+                  draw->start_instance,
                   draw->instance_id,
                   hw_verts);
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
index 3740deab5a8..8716f657777 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -168,6 +168,7 @@ draw_pt_fetch_run(struct pt_fetch *fetch,
    translate->run_elts( translate,
 			elts,
 			count,
+                        draw->start_instance,
                         draw->instance_id,
 			verts );
 }
@@ -195,6 +196,7 @@ draw_pt_fetch_run_linear(struct pt_fetch *fetch,
    translate->run( translate,
                    start,
                    count,
+                   draw->start_instance,
                    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 dc6decba70e..22ec8d6b2b5 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -210,6 +210,7 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle,
    feme->translate->run_elts( feme->translate,
 			      fetch_elts,
 			      fetch_count,
+                              draw->start_instance,
                               draw->instance_id,
 			      hw_verts );
 
@@ -267,6 +268,7 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->start_instance,
                          draw->instance_id,
                          hw_verts );
 
@@ -326,6 +328,7 @@ static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
    feme->translate->run( feme->translate,
                          start,
                          count,
+                         draw->start_instance,
                          draw->instance_id,
                          hw_verts );
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_so_emit.c b/src/gallium/auxiliary/draw/draw_pt_so_emit.c
index d624a990bc6..a6d1da4eb2e 100644
--- a/src/gallium/auxiliary/draw/draw_pt_so_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_so_emit.c
@@ -182,12 +182,29 @@ static void so_emit_prim(struct pt_so_emit *so,
 
          buffer = (float *)((char *)draw->so.targets[ob]->mapping +
                             draw->so.targets[ob]->target.buffer_offset +
-                            draw->so.targets[ob]->internal_offset) + state->output[slot].dst_offset;
+                            draw->so.targets[ob]->internal_offset) +
+            state->output[slot].dst_offset;
          
          if (idx == so->pos_idx && pcp_ptr)
-            memcpy(buffer, &pre_clip_pos[start_comp], num_comps * sizeof(float));
+            memcpy(buffer, &pre_clip_pos[start_comp],
+                   num_comps * sizeof(float));
          else
-            memcpy(buffer, &input[idx][start_comp], num_comps * sizeof(float));
+            memcpy(buffer, &input[idx][start_comp],
+                   num_comps * sizeof(float));
+#if 0
+         {
+            int j;
+            debug_printf("VERT[%d], offset = %d, slot[%d] sc = %d, num_c = %d, idx = %d = [",
+                         i + draw->so.targets[ob]->emitted_vertices,
+                         draw->so.targets[ob]->internal_offset,
+                         slot, start_comp, num_comps, idx);
+            for (j = 0; j < num_comps; ++j) {
+               unsigned *ubuffer = (unsigned*)buffer;
+               debug_printf("%d (0x%x), ", ubuffer[j], ubuffer[j]);
+            }
+            debug_printf("]\n");
+         }
+#endif
       }
       for (ob = 0; ob < draw->so.num_targets; ++ob) {
          struct draw_so_target *target = draw->so.targets[ob];
diff --git a/src/gallium/auxiliary/draw/draw_vs_variant.c b/src/gallium/auxiliary/draw/draw_vs_variant.c
index 152c1303183..37500c7db8d 100644
--- a/src/gallium/auxiliary/draw/draw_vs_variant.c
+++ b/src/gallium/auxiliary/draw/draw_vs_variant.c
@@ -168,6 +168,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_variant *variant,
    vsvg->fetch->run_elts( vsvg->fetch, 
                           elts,
                           count,
+                          vsvg->draw->start_instance,
                           vsvg->draw->instance_id,
                           temp_buffer );
 
@@ -211,6 +212,7 @@ static void PIPE_CDECL vsvg_run_elts( struct draw_vs_variant *variant,
 
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->start_instance,
                     vsvg->draw->instance_id,
                     output_buffer );
 
@@ -234,6 +236,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_variant *variant,
    vsvg->fetch->run( vsvg->fetch, 
                      start,
                      count,
+                     vsvg->draw->start_instance,
                      vsvg->draw->instance_id,
                      temp_buffer );
 
@@ -274,6 +277,7 @@ static void PIPE_CDECL vsvg_run_linear( struct draw_vs_variant *variant,
    
    vsvg->emit->run( vsvg->emit,
                     0, count,
+                    vsvg->draw->start_instance,
                     vsvg->draw->instance_id,
                     output_buffer );