https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124609
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target Milestone|--- |16.0
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=1101.60.0
improves compared to the
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.0=962.60.0 regresion
(same machine, just FDO vs non-FDO - a previous
regresion that was eventually fixed for one seems to be now fixed for the
other and re-surfaces?).
I don't remember mcf too much, in particular issues with FDO. But it does look
suspicious. Training with the ref set improves performance by 5% ...
Clearly there are code-gen differences because of the patch, but that was
expected. I did not try to analyze what changed.
