https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120747
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |WAITING
--- Comment #24 from Richard Biener <rguenth at gcc dot gnu.org> ---
So testing with r16-7431-ge55de74d50d236 435.gromacs does not miscompare for me
with -Ofast -march=znver4 on a Zen4 machine. The oldest trunk I have lying
around (r16-5041) also does not reproduce the miscompare, so I'm not sure it's
gone again or I can just not reproduce for one or another reason.
https://lnt.opensuse.org/db_default/v4/SPEC/graph?highlight_run=83231&plot.90=965.90.0
shows it's back, but IIRC you bumped tolerance? Can you see to turn that back
to the default?