https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102473
Bug ID: 102473
Summary: 521.wrf_r 5% slower at -Ofast and generic x86_64
tuning after r12-3426-g8f323c712ea76c
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jamborm at gcc dot gnu.org
CC: crazylht at gmail dot com
Blocks: 26163
Target Milestone: ---
Host: x86_64-linux
Target: x86_64-linux
All three x86_64 LNT machines have detected a 4.5-5.2% performance
regression of SPEC FPrate 2017 benchmarks 521.wrf_r when compiled with
-Ofast and the default (generic) march and mtune.
Zen2 based machine regressed by 5%:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=294.548.0
Zen1 based machine regressed by 5.2%:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=35.548.0
Kabylake based machine regressed by 4.5%:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=34.548.0
On an AMD zen2 based machine I have bisected the regression to commit
r12-3426-g8f323c712ea76c:
8f323c712ea76cc4506b03895e9b991e4e4b2baf is the first bad commit
commit 8f323c712ea76cc4506b03895e9b991e4e4b2baf
Author: liuhongt <[email protected]>
Date: Tue Sep 7 12:39:04 2021 +0800
Optimize v4sf reduction.
gcc/ChangeLog:
PR target/101059
* config/i386/sse.md (reduc_plus_scal_<mode>): Split to ..
(reduc_plus_scal_v4sf): .. this, New define_expand.
(reduc_plus_scal_v2df): .. and this, New define_expand.
I have confirmed that the commit causes a similar regression on
another Intel Skylake server.
On the Zen2 machine, this is the difference in samples collected by
perf for different symbols (before is commit 60eec23b5ed, after commit
8f323c712ea):
| Symbol | sys lib | Before | After |
diff | % |
|---------------------------------------------+---------+--------+-------+-------+-------|
| __logf_fma | yes | 68882 | 68940 |
+58 | +0.08 |
| __atanf | yes | 66664 | 66196 |
-468 | -0.70 |
| __module_advect_em_MOD_advect_scalar_pd | no | 62286 | 62348 |
+62 | +0.10 |
| __powf_fma | yes | 56213 | 56127 |
-86 | -0.15 |
| __module_mp_wsm5_MOD_nislfv_rain_plm | no | 46990 | 48340 |
+1350 | +2.87 |
| __module_mp_wsm5_MOD_wsm52d | no | 41031 | 40968 |
-63 | -0.15 |
| __module_small_step_em_MOD_advance_uv | no | 30908 | 30909 |
+1 | +0.00 |
| __module_small_step_em_MOD_advance_w | no | 28738 | 28600 |
-138 | -0.48 |
| __module_advect_em_MOD_advect_scalar | no | 28400 | 28429 |
+29 | +0.10 |
| __expf_fma | yes | 26702 | 26516 |
-186 | -0.70 |
| __module_big_step_utilities_em_MOD_phy_prep | no | 25878 | 25816 |
-62 | -0.24 |
| psim_unstable_ | no | 24994 | 25106 |
+112 | +0.45 |
| __module_bl_ysu_MOD_ysu2d | no | 24799 | 25251 |
+452 | +1.82 |
| psih_unstable_ | no | 22600 | 23139 |
+539 | +2.38 |
| __module_small_step_em_MOD_advance_mu_t | no | 22250 | 22232 |
-18 | -0.08 |
| __memset_avx2_unaligned_erms | yes | 21748 | 21613 |
-135 | -0.62 |
| _ZGVbN4vv_powf_sse4 | yes | 21206 | 21355 |
+149 | +0.70 |
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163
[Bug 26163] [meta-bug] missed optimization in SPEC (2k17, 2k and 2k6 and 95)