radeonsi: plumb the output vertex_stream through to si_shader_output_values
authorNicolai Hähnle <nicolai.haehnle@amd.com>
Tue, 29 Nov 2016 14:43:44 +0000 (15:43 +0100)
committerNicolai Hähnle <nicolai.haehnle@amd.com>
Mon, 12 Dec 2016 08:04:19 +0000 (09:04 +0100)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
src/gallium/drivers/radeonsi/si_shader.c

index e496c50c97c63f7f4cdefd45ebfdfa4e686b9e96..abcdeddbf56102b6816c6e3475f99fc23d155176 100644 (file)
@@ -57,6 +57,7 @@ struct si_shader_output_values
        LLVMValueRef values[4];
        unsigned semantic_name;
        unsigned semantic_index;
+       ubyte vertex_stream[4];
 };
 
 static void si_init_shader_ctx(struct si_shader_context *ctx,
@@ -2893,11 +2894,15 @@ static void si_llvm_emit_vs_epilogue(struct lp_build_tgsi_context *bld_base)
                outputs[i].semantic_name = info->output_semantic_name[i];
                outputs[i].semantic_index = info->output_semantic_index[i];
 
-               for (j = 0; j < 4; j++)
+               for (j = 0; j < 4; j++) {
                        outputs[i].values[j] =
                                LLVMBuildLoad(gallivm->builder,
                                              ctx->soa.outputs[i][j],
                                              "");
+                       outputs[i].vertex_stream[j] =
+                               (info->output_streams[i] >> (2 * j)) & 3;
+               }
+
        }
 
        /* Return the primitive ID from the LLVM function. */
@@ -6378,6 +6383,9 @@ si_generate_gs_copy_shader(struct si_screen *sscreen,
                outputs[i].semantic_index = gsinfo->output_semantic_index[i];
 
                for (chan = 0; chan < 4; chan++) {
+                       outputs[i].vertex_stream[chan] =
+                               (gsinfo->output_streams[i] >> (2 * chan)) & 3;
+
                        args[2] = lp_build_const_int32(gallivm,
                                                       (i * 4 + chan) *
                                                       gs_selector->gs_max_out_vertices * 16 * 4);