[Mesa-dev] [PATCH v2 13/25] radeonsi: load the streamout buf descriptors closer to their use

[Nicolai Hähnle]

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

LLVM can still decide to hoist the loads since they're marked invariant.
---
 src/gallium/drivers/radeonsi/si_shader.c | 25 +++++++++++--------------
 1 file changed, 11 insertions(+), 14 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 7e222c0..f1a4946 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2178,33 +2178,20 @@ static void emit_streamout_output(struct si_shader_context *ctx,
  * to buffers. */
 static void si_llvm_emit_streamout(struct si_shader_context *ctx,
 				   struct si_shader_output_values *outputs,
 				   unsigned noutput)
 {
 	struct pipe_stream_output_info *so = &ctx->shader->selector->so;
 	struct gallivm_state *gallivm = &ctx->gallivm;
 	LLVMBuilderRef builder = gallivm->builder;
 	int i;
 	struct lp_build_if_state if_ctx;
-	LLVMValueRef so_buffers[4];
-	LLVMValueRef buf_ptr = LLVMGetParam(ctx->main_fn,
-					    SI_PARAM_RW_BUFFERS);
-
-	/* Load the descriptors. */
-	for (i = 0; i < 4; ++i) {
-		if (ctx->shader->selector->so.stride[i]) {
-			LLVMValueRef offset = lp_build_const_int32(gallivm,
-								   SI_VS_STREAMOUT_BUF0 + i);
-
-			so_buffers[i] = build_indexed_load_const(ctx, buf_ptr, offset);
-		}
-	}
 
 	/* Get bits [22:16], i.e. (so_param >> 16) & 127; */
 	LLVMValueRef so_vtx_count =
 		unpack_param(ctx, ctx->param_streamout_config, 16, 7);
 
 	LLVMValueRef tid = get_thread_id(ctx);
 
 	/* can_emit = tid < so_vtx_count; */
 	LLVMValueRef can_emit =
 		LLVMBuildICmp(builder, LLVMIntULT, tid, so_vtx_count, "");
@@ -2223,26 +2210,36 @@ static void si_llvm_emit_streamout(struct si_shader_context *ctx,
 		 *                attrib_offset
                  */
 
 		LLVMValueRef so_write_index =
 			LLVMGetParam(ctx->main_fn,
 				     ctx->param_streamout_write_index);
 
 		/* Compute (streamout_write_index + thread_id). */
 		so_write_index = LLVMBuildAdd(builder, so_write_index, tid, "");
 
-		/* Compute the write offset for each enabled buffer. */
+		/* Load the descriptor and compute the write offset for each
+		 * enabled buffer. */
 		LLVMValueRef so_write_offset[4] = {};
+		LLVMValueRef so_buffers[4];
+		LLVMValueRef buf_ptr = LLVMGetParam(ctx->main_fn,
+						    SI_PARAM_RW_BUFFERS);
+
 		for (i = 0; i < 4; i++) {
 			if (!so->stride[i])
 				continue;
 
+			LLVMValueRef offset = lp_build_const_int32(gallivm,
+								   SI_VS_STREAMOUT_BUF0 + i);
+
+			so_buffers[i] = build_indexed_load_const(ctx, buf_ptr, offset);
+
 			LLVMValueRef so_offset = LLVMGetParam(ctx->main_fn,
 							      ctx->param_streamout_offset[i]);
 			so_offset = LLVMBuildMul(builder, so_offset, LLVMConstInt(ctx->i32, 4, 0), "");
 
 			so_write_offset[i] = LLVMBuildMul(builder, so_write_index,
 							  LLVMConstInt(ctx->i32, so->stride[i]*4, 0), "");
 			so_write_offset[i] = LLVMBuildAdd(builder, so_write_offset[i], so_offset, "");
 		}
 
 		/* Write streamout data. */
-- 
2.7.4