https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112548
Filip Kastl <pheeck at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|[14/15 regression] 5% exec |429.mcf sometimes speeds up
|time regression in 429.mcf |by 5% and then slows down
|on AMD zen4 CPU (since |again (AMD zen4 CPU, LTO,
|r14-5076-g01c18f58d37865) |PGO)
--- Comment #32 from Filip Kastl <pheeck at gcc dot gnu.org> ---
This actually isn't a 14/15 Regression. My bad. Comparison of trunk and GCC
13 here:
https://lnt.opensuse.org/db_default/v4/SPEC/graph?plot.2=974.60.0&plot.4=991.60.0&;
Btw this is the status of this bug: Together with Robin we've given some more
effort into sharing the GCDA data (we did that through mail so this isn't
recorded here). After some time I've decided to give up because we didn't make
any progress.
If you look at the graph, you can see that the benchmark sometimes speeds up by
~5% and then returns to its original speed. I guess it could be interesting to
find out what that means. It doesn't seem like a big priority though. I think
that there wasn't anything wrong with Robin's patch and that the patch just
randomly triggered this phenomenon (the slowdown after the speedup).
I'm going to leave this bug open and rename it. If someone thinks it would be
better to close it, I won't object.
