https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119108
Bug ID: 119108
Summary: [15 Regression] AArch64 Commit 'vect: Force alignment
peeling ...'
(https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=68326d5
d1a593d) causes regression in Snappy workload for
-mcpu=neoverse-v2.
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: matmal01 at gcc dot gnu.org
Target Milestone: ---
Created attachment 60650
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=60650&action=edit
Script to reproduce the observed slowdown.
Have observed a slowdown after the referenced commit.
Attaching script for reproduction. Results when `master` is commit 78380fd7f
inlined below (numbers are percentage change in time from "TOT with problem
commit reverted" to "TOT" -- positive numbers demonstrating the peeling has
caused a slowdown).
I ran the script with:
```
vshcmd: > cd $HOME
vshcmd: > rm -rf $HOME/testing-reproduction-script
vshcmd: > newdir $HOME/testing-reproduction-script
vshcmd: > git clone $HOME/gcc-source gcc_src
vshcmd: > parentdir=$HOME/testing-reproduction-script $HOME/Snappy/reproduce.sh
```
```
BM_UFlat/3/1 5.33333
BM_UFlat/3/2 7.14286
BM_UFlat/4/2 2.59319
BM_UFlat/5/1 2.86533
BM_UFlat/5/2 5.10708
BM_UValidate/3/2 2.08333
BM_UValidate/5/2 2.41758
BM_UIOVecSource/1/2 4.21903
BM_UIOVecSource/5/2 5.44218
BM_UIOVecSource/6/2 4.21348
BM_UIOVecSource/7/1 -3.6036
BM_UIOVecSource/7/2 6.84039
BM_UIOVecSource/8/2 3.86905
BM_UIOVecSource/9/2 2.90987
BM_UIOVecSource/11/2 5
BM_UIOVecSink/0 21.3523
BM_UFlatSink/3/1 9.58904
BM_UFlatSink/3/2 10.1449
BM_UFlatSink/5/1 3.17919
BM_ZFlat/1/2 4.54959
BM_ZFlat/5/1 2.73973
BM_ZFlat/5/2 6.31579
BM_ZFlat/6/1 -2.35294
BM_ZFlat/6/2 3.9548
BM_ZFlat/7/1 -3.15315
BM_ZFlat/7/2 5.51948
BM_ZFlat/8/2 3.99202
BM_ZFlat/9/2 3.25145
BM_ZFlat/11/2 5.83942
BM_ZFlatAll/2 3.57955
```
