fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-24 10:38:11 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

ac: handle new float multadd opcodes

Browse source 
This advertizes both ffma and fmad on a couple of chipsets to support
VK_KHR_shader_fma and to improve OpenCL fma performance, where fma is not
optional and the emulation is more expensive than slow fma.

Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41165>
This commit is contained in:
Karol Herbst 2026-04-24 01:06:59 +02:00 • committed by Marge Bot
parent 0082745b0a
commit 570bfe1ee0
1 changed files with 14 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/amd/common/nir/ac_nir.c
View file

@ -14,6 +14,8 @@
void ac_nir_set_options(const struct ac_compiler_info *info, bool use_llvm,
nir_shader_compiler_options *options)
{
memset(options, 0, sizeof(*options));
/* |---------------------------------- Performance & Availability --------------------------------|
* |MAD/MAC/MADAK/MADMK|MAD_LEGACY|MAC_LEGACY| FMA |FMAC/FMAAK/FMAMK|FMA_LEGACY|PK_FMA_F16,|Best choice
* Arch | F32,F16,F64 | F32,F16 | F32,F16 |F32,F16,F64 | F32,F16 | F32 |PK_FMAC_F16|F16,F32,F64
@ -34,7 +36,17 @@ void ac_nir_set_options(const struct ac_compiler_info *info, bool use_llvm,
* gfx10 and older prefer MAD for F32 because of the legacy instruction.
*/
memset(options, 0, sizeof(*options));
options->float_mul_add32 = nir_float_muladd_support_has_ffma;
if (info->has_mad32)
options->float_mul_add32 |= nir_float_muladd_support_prefers_split;
if (info->gfx_level >= GFX8) {
options->float_mul_add16 = nir_float_muladd_support_has_ffma;
if (info->gfx_level == GFX8)
options->float_mul_add16 |= nir_float_muladd_support_prefers_split;
}
options->float_mul_add64 = nir_float_muladd_support_has_ffma;
options->vertex_id_zero_based = true;
options->lower_scmp = true;
options->lower_flrp16 = true;
@ -972,6 +984,7 @@ ac_nir_op_supports_packed_math_16bit(const nir_alu_instr* alu)
case nir_op_fadd:
case nir_op_fsub:
case nir_op_fmul:
case nir_op_ffma:
case nir_op_ffma_old:
case nir_op_fdiv:
case nir_op_flrp: