fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-09 02:28:10 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

nir/opt_peephole_select: do not count fmul towards the limit when only used by fadd

Browse source 
Foz-DB Navi10:
Totals from 4077 (5.64% of 72319) affected shaders:
MaxWaves: 84057 -> 83325 (-0.87%); split: +0.07%, -0.94%
Instrs: 6019711 -> 6007338 (-0.21%); split: -0.27%, +0.07%
CodeSize: 32373984 -> 32356152 (-0.06%); split: -0.18%, +0.13%
VGPRs: 236588 -> 238172 (+0.67%); split: -0.05%, +0.72%
SpillSGPRs: 7341 -> 7367 (+0.35%); split: -0.65%, +1.01%
Latency: 61833147 -> 61386674 (-0.72%); split: -0.91%, +0.19%
InvThroughput: 22328993 -> 22364077 (+0.16%); split: -0.16%, +0.32%
VClause: 97803 -> 97832 (+0.03%); split: -0.29%, +0.32%
SClause: 147544 -> 146274 (-0.86%); split: -1.19%, +0.33%
Copies: 606083 -> 593887 (-2.01%); split: -2.27%, +0.26%
Branches: 171344 -> 164203 (-4.17%); split: -4.17%, +0.00%
PreSGPRs: 234116 -> 234922 (+0.34%); split: -0.17%, +0.52%
PreVGPRs: 211250 -> 211374 (+0.06%); split: -0.00%, +0.06%
VALU: 4130666 -> 4132669 (+0.05%); split: -0.11%, +0.16%
SALU: 854007 -> 852585 (-0.17%); split: -0.77%, +0.61%
VMEM: 162718 -> 162755 (+0.02%); split: -0.00%, +0.03%
SMEM: 237856 -> 236323 (-0.64%); split: -0.65%, +0.00%

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41348>
This commit is contained in:
Georg Lehmann 2026-05-04 16:13:03 +02:00 • committed by Marge Bot
parent 5f48aba9fa
commit 25add9cbd1
1 changed files with 13 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/compiler/nir/nir_opt_peephole_select.c
View file

@ -25,6 +25,7 @@
#include "nir.h"
#include "nir_builder.h"
#include "nir_control_flow.h"
#include "nir_search_helpers.h"
/*
* Implements a small peephole optimization that looks for
@ -230,10 +231,10 @@ block_check_for_allowed_instrs(nir_block *block, unsigned *count,
break;
case nir_instr_type_alu: {
nir_alu_instr *mov = nir_instr_as_alu(instr);
nir_alu_instr *alu = nir_instr_as_alu(instr);
bool movelike = false;
switch (mov->op) {
switch (alu->op) {
case nir_op_mov:
case nir_op_fneg:
case nir_op_ineg:
@ -280,11 +281,18 @@ block_check_for_allowed_instrs(nir_block *block, unsigned *count,
* merged as a destination modifier or source modifier on some
* other instruction.
*/
if (mov->op != nir_op_fsat && !movelike)
(*count)++;
if (alu->op != nir_op_fsat && !movelike) {
/* If this is a fmul that is only used by fadd, don't count it.
* It will likely be fused to fma/mad.
*/
if ((alu->op != nir_op_fmul && alu->op != nir_op_fmulz) ||
!is_only_used_by_fadd(alu)) {
(*count)++;
}
}
} else {
/* The only uses of this definition must be phis in the successor */
nir_foreach_use_including_if(use, &mov->def) {
nir_foreach_use_including_if(use, &alu->def) {
if (nir_src_is_if(use) ||
nir_src_use_instr(use)->type != nir_instr_type_phi ||
nir_src_use_instr(use)->block != block->successors[0])