fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 07:20:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

ac/nir: fix calculation of aligned_new_size

Browse source 
This should consider nir_round_up_components().

fossil-db (gfx1201):
Totals from 90 (0.11% of 79839) affected shaders:
MaxWaves: 1829 -> 1901 (+3.94%)
Instrs: 410780 -> 411825 (+0.25%); split: -0.02%, +0.27%
CodeSize: 2227956 -> 2234464 (+0.29%); split: -0.02%, +0.31%
VGPRs: 6952 -> 6760 (-2.76%); split: -3.11%, +0.35%
Latency: 3071765 -> 3073960 (+0.07%); split: -0.00%, +0.07%
InvThroughput: 766201 -> 767322 (+0.15%); split: -0.00%, +0.15%
VClause: 7887 -> 7898 (+0.14%); split: -0.08%, +0.22%
Copies: 48189 -> 48324 (+0.28%); split: -0.05%, +0.33%
PreVGPRs: 6605 -> 6595 (-0.15%); split: -0.18%, +0.03%
VALU: 237272 -> 238147 (+0.37%); split: -0.01%, +0.37%
SALU: 48987 -> 49003 (+0.03%)
VMEM: 15542 -> 15560 (+0.12%)
VOPD: 188 -> 200 (+6.38%)

fossil-db (navi31):
Totals from 89 (0.11% of 79825) affected shaders:
MaxWaves: 1811 -> 1883 (+3.98%)
Instrs: 403695 -> 404691 (+0.25%); split: -0.01%, +0.26%
CodeSize: 2150612 -> 2154860 (+0.20%); split: -0.03%, +0.23%
VGPRs: 6892 -> 6676 (-3.13%)
Latency: 3306107 -> 3310010 (+0.12%); split: -0.01%, +0.13%
InvThroughput: 813092 -> 814382 (+0.16%); split: -0.00%, +0.16%
VClause: 7999 -> 8010 (+0.14%); split: -0.06%, +0.20%
Copies: 50089 -> 50210 (+0.24%); split: -0.05%, +0.29%
PreVGPRs: 6596 -> 6586 (-0.15%); split: -0.18%, +0.03%
VALU: 239617 -> 240392 (+0.32%); split: -0.01%, +0.33%
SALU: 45349 -> 45363 (+0.03%)
VMEM: 15762 -> 15780 (+0.11%)
VOPD: 258 -> 262 (+1.55%)

fossil-db (navi21):
Totals from 89 (0.11% of 79825) affected shaders:
Instrs: 345634 -> 346426 (+0.23%); split: -0.00%, +0.23%
CodeSize: 1895616 -> 1900156 (+0.24%); split: -0.00%, +0.24%
Latency: 3043334 -> 3046859 (+0.12%); split: -0.01%, +0.13%
InvThroughput: 928236 -> 929626 (+0.15%); split: -0.01%, +0.16%
VClause: 7894 -> 7905 (+0.14%); split: -0.06%, +0.20%
Copies: 48694 -> 48785 (+0.19%); split: -0.03%, +0.22%
PreVGPRs: 6580 -> 6570 (-0.15%); split: -0.18%, +0.03%
VALU: 228323 -> 229072 (+0.33%); split: -0.01%, +0.33%
SALU: 47202 -> 47216 (+0.03%)
VMEM: 16546 -> 16564 (+0.11%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14458
Backport-to: 25.3
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38903>
This commit is contained in:
Rhys Perry 2025-12-11 09:58:29 +00:00 • committed by Marge Bot
parent c50474ac6f
commit 49d923078f
1 changed files with 3 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/amd/common/nir/ac_nir.c
View file

@ -551,8 +551,9 @@ ac_nir_mem_vectorize_callback(unsigned align_mul, unsigned align_offset, unsigne
/* Align the size to what the hw supports. */ /* Align the size to what the hw supports. */
unsigned unaligned_new_size = num_components * bit_size; unsigned unaligned_new_size = num_components * bit_size;
unsigned aligned_new_size = align_load_store_size(config->gfx_level, unaligned_new_size, unsigned aligned_new_size = nir_round_up_components(num_components) * bit_size;
uses_smem, is_shared); aligned_new_size = align_load_store_size(config->gfx_level, aligned_new_size,
uses_smem, is_shared);
if (uses_smem) { if (uses_smem) {
/* Maximize SMEM vectorization except for LLVM, which suffers from SGPR and VGPR spilling. /* Maximize SMEM vectorization except for LLVM, which suffers from SGPR and VGPR spilling.