radeonsi/nir: don't lower fma, instead, fuse fma

We want fma. This decreases compile times by 4% for Borderlands 2.

48505 shaders in 30515 tests
Totals:
SGPRS: 2206584 -> 2204784 (-0.08 %)
VGPRS: 1647892 -> 1648964 (0.07 %)
Spilled SGPRs: 6256 -> 6078 (-2.85 %)
Spilled VGPRs: 72 -> 72 (0.00 %)
Private memory VGPRs: 2176 -> 2176 (0.00 %)
Scratch size: 2240 -> 2240 (0.00 %) dwords per thread
Code Size: 49680804 -> 49837988 (0.32 %) bytes
LDS: 74 -> 74 (0.00 %) blocks
Max Waves: 371387 -> 371352 (-0.01 %)

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>

diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index 9458f6b959a43afec96a80976d55cfe6f1413ea8..8dc13ee08b2ba52e72c23a3793d81be1f0a18673 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -511,7 +511,7 @@ static const struct nir_shader_compiler_options nir_options = {
        .lower_bitfield_insert_to_bitfield_select = true,
        .lower_bitfield_extract = true,
        .lower_sub = true,
-       .lower_ffma = true,
+       .fuse_ffma = true,
        .lower_fmod = true,
        .lower_pack_snorm_4x8 = true,
        .lower_pack_unorm_4x8 = true,