https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124114
Bug ID: 124114
Summary: 7% slowdown of pelbench on Zen{1,4} since
r16-7431-ge55de74d50d236
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: jmelcr at gcc dot gnu.org
CC: pheeck at gcc dot gnu.org, rguenth at gcc dot gnu.org
Blocks: 26163
Target Milestone: ---
Host: x86_64-suse-linux
Target: x86_64-suse-linux
As seen here
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=1101.10.0
there was a 7% exec time slowdown of the perlbench SPEC 2006
benchmark when compiled with -O2 -march=x86-64-v3 -g -flto=128 on a Zen4
machine. I bisected it to r16-7431-ge55de74d50d236.
e55de74d50d236db0472431fe242f3244e41b884 is the first bad commit
commit e55de74d50d236db0472431fe242f3244e41b884
Author: Richard Biener <[email protected]>
Date: Tue Feb 10 09:46:08 2026 +0100
tree-optimization/107690 - avoid creating un-analyzable loop exits
The following adds a heuristic to ifcombine that avoids turning
analyzable loop exits into unanalyzable ones. This allows vectorizing
the testcase in the PR again. I've refrained from actually
analyzing niters but instead used a cheaper heuristic. I believe
we'll only ever attempt to combine two ifs if they are in the same
loop and if either both exit the loop or stay within.
PR tree-optimization/107690
* tree-ssa-ifcombine.cc (ifcombine_ifandif): Do not merge
possibly analyzable exit conditions.
* g++.dg/vect/vect-pr107690.cc: New testcase.
gcc/testsuite/g++.dg/vect/vect-pr107690.cc | 16 ++++++++++++++++
gcc/tree-ssa-ifcombine.cc | 16 ++++++++++++++++
2 files changed, 32 insertions(+)
create mode 100644 gcc/testsuite/g++.dg/vect/vect-pr107690.cc
bisect found first bad commit
This is a regression against GCC 13/14. See the comparison
here:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=1110.10.0&plot.1=1128.10.0&plot.2=1122.10.0&plot.3=1104.10.0&plot.4=1144.10.0&plot.5=1101.10.0&;
There is also a different perlbench slowdown in the same timeframe (so probably
caused by the same commit):
4% Zen1 -O2 -march=native -g
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=291.10.0
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)
