https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116743
Bug ID: 116743
Summary: Commit `3d9e6767939e` causes ~10% perf regression
Product: gcc
Version: 12.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: gcov-profile
Assignee: unassigned at gcc dot gnu.org
Reporter: rvmallad at amazon dot com
Target Milestone: ---
In our tests of MySQL HAMMERDB tests, we found that the commit `3d9e6767939e`
had a regression of ~10% using compiler release `12.3.0` vs. `11.5.0`. I see
this commit tries to improve the AutoFDO count propagation algorithm. But in
our case, we saw a regression. Here is some raw perf data:
Throughput (higher better) Baseline AutoFDO
11.5.0 59,973 65,978
12.3.0 58,504 58,452
12.3.0 w revert `3d9e6767939e` 59,287 66,914
I am trying to prepare a test-case/ reproducer for the same. Interim, can we
revisit this optimization algorithm? Thanks.
