From 2c61594d84911f486aa2edb4b8e561e780139d20 Mon Sep 17 00:00:00 2001 From: Dave Airlie Date: Wed, 4 Oct 2017 06:33:02 +1000 Subject: radv: lower ffma in nir. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit So it appears the Vulkan SPIR-V fma opcode can be equivalent to a mad operation, and the fma hw opcode on AMD hw is issued like a double opcode so is slower. Also the radeonsi stack does this. This appears to improve performance on a number of games from Feral, and thanks to Feral for noticing the problem. I'm reposting this one as Marek indicated he thinks this is what we should be doing on AMD hw. Reviewed-by: Nicolai Hähnle Reviewed-by: Marek Olšák Cc: "17.2" Signed-off-by: Dave Airlie --- src/amd/vulkan/radv_shader.c | 1 + 1 file changed, 1 insertion(+) (limited to 'src/amd') diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c index e0944a478e1..535d0f4f58b 100644 --- a/src/amd/vulkan/radv_shader.c +++ b/src/amd/vulkan/radv_shader.c @@ -63,6 +63,7 @@ static const struct nir_shader_compiler_options nir_options = { .lower_unpack_unorm_4x8 = true, .lower_extract_byte = true, .lower_extract_word = true, + .lower_ffma = true, .max_unroll_iterations = 32 }; -- cgit v1.2.3