https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127402
Drea Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|tree-optimization |target
--- Comment #2 from Drea Pinski <pinskia at gcc dot gnu.org> ---
So looking at what fixed the issue back in March, that would be
r16-8128-g5f1024922d3f67 . BUT that revision made PGO worse. So I am 99% sure
there is either code alignment difference or a loop alignment which is causing
the jump. Likewise of PR 124609 . Basically any small code (seemily
unrelated) generation changes performance a lot one way or another on x86.
Most likely due to branch prediction and/or cmov difference.
