From a6722285c2788a1d96d8196e2766b10d5257b805 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Fri, 5 Jul 2019 23:32:36 -0400 Subject: [PATCH] radeonsi/gfx10: simplify a streamout loop in gfx10_emit_ngg_epilogue Acked-by: Pierre-Eric Pelloux-Prayer Reviewed-by: Samuel Pitoiset --- src/gallium/drivers/radeonsi/gfx10_shader_ngg.c | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-) diff --git a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c index e074449064a..59f7f34c8a7 100644 --- a/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c +++ b/src/gallium/drivers/radeonsi/gfx10_shader_ngg.c @@ -538,19 +538,18 @@ void gfx10_emit_ngg_epilogue(struct ac_shader_abi *abi, outputs[i].semantic_name = info->output_semantic_name[i]; outputs[i].semantic_index = info->output_semantic_index[i]; - /* This is used only by streamout. */ for (unsigned j = 0; j < 4; j++) { - outputs[i].values[j] = - LLVMBuildLoad(builder, - addrs[4 * i + j], - ""); outputs[i].vertex_stream[j] = (info->output_streams[i] >> (2 * j)) & 3; - if (vertex_ptr) { + /* TODO: we may store more outputs than streamout needs, + * but streamout performance isn't that important. + */ + if (sel->so.num_outputs) { tmp = ac_build_gep0(&ctx->ac, vertex_ptr, LLVMConstInt(ctx->i32, 4 * i + j, false)); - tmp2 = ac_to_integer(&ctx->ac, outputs[i].values[j]); + tmp2 = LLVMBuildLoad(builder, addrs[4 * i + j], ""); + tmp2 = ac_to_integer(&ctx->ac, tmp2); LLVMBuildStore(builder, tmp2, tmp); } } -- 2.30.2