[Mesa-dev] [PATCH v2 18/25] radeonsi: fetch only outputs of current vertex stream from the GSVS ring

[Nicolai Hähnle]

From: Nicolai Hähnle <nicolai.haehnle at amd.com>

---
 src/gallium/drivers/radeonsi/si_shader.c | 41 +++++++++++++++++++-------------
 1 file changed, 25 insertions(+), 16 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 5b1acc6..4200da5 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -6262,61 +6262,70 @@ si_generate_gs_copy_shader(struct si_screen *sscreen,
 	args[8] = uint->zero; /* TFE */
 
 	/* Fetch the vertex stream ID.*/
 	LLVMValueRef stream_id;
 
 	if (gs_selector->so.num_outputs)
 		stream_id = unpack_param(&ctx, ctx.param_streamout_config, 24, 2);
 	else
 		stream_id = uint->zero;
 
-	/* Fetch vertex data from GSVS ring */
+	/* Fill in output information. */
 	for (i = 0; i < gsinfo->num_outputs; ++i) {
-		unsigned chan;
-
 		outputs[i].semantic_name = gsinfo->output_semantic_name[i];
 		outputs[i].semantic_index = gsinfo->output_semantic_index[i];
 
-		for (chan = 0; chan < 4; chan++) {
+		for (int 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);
-
-			outputs[i].values[chan] =
-				LLVMBuildBitCast(gallivm->builder,
-						 lp_build_intrinsic(gallivm->builder,
-								 "llvm.SI.buffer.load.dword.i32.i32",
-								 ctx.i32, args, 9,
-								 LP_FUNC_ATTR_READONLY),
-						 ctx.f32, "");
 		}
 	}
 
 	for (int stream = 0; stream < 4; stream++) {
 		struct lp_build_if_state if_ctx_stream;
 
 		if (!gsinfo->num_stream_output_components[stream])
 			continue;
 
 		if (stream > 0 && !gs_selector->so.num_outputs)
 			continue;
 
 		LLVMValueRef is_stream =
 			LLVMBuildICmp(builder, LLVMIntEQ,
 				      stream_id,
 				      lp_build_const_int32(gallivm, stream), "");
 
 		lp_build_if(&if_ctx_stream, gallivm, is_stream);
 
+		/* Fetch vertex data from GSVS ring */
+		for (i = 0; i < gsinfo->num_outputs; ++i) {
+			for (unsigned chan = 0; chan < 4; chan++) {
+				if (outputs[i].vertex_stream[chan] != stream) {
+					outputs[i].values[chan] = ctx.soa.bld_base.base.undef;
+					continue;
+				}
+
+				args[2] = lp_build_const_int32(
+					gallivm,
+					(i * 4 + chan) * gs_selector->gs_max_out_vertices * 16 * 4);
+
+				outputs[i].values[chan] =
+					LLVMBuildBitCast(gallivm->builder,
+						 lp_build_intrinsic(gallivm->builder,
+								 "llvm.SI.buffer.load.dword.i32.i32",
+								 ctx.i32, args, 9,
+								 LP_FUNC_ATTR_READONLY),
+						 ctx.f32, "");
+			}
+		}
+
+		/* Streamout and exports. */
 		if (gs_selector->so.num_outputs) {
 			si_llvm_emit_streamout(&ctx, outputs,
 					       gsinfo->num_outputs,
 					       stream);
 		}
 
 		if (stream == 0)
 			si_llvm_export_vs(bld_base, outputs, gsinfo->num_outputs);
 
 		lp_build_endif(&if_ctx_stream);
-- 
2.7.4