On Tue, 16 Jul 2024, Jakub Jelinek wrote: > On Tue, Jul 16, 2024 at 10:55:30AM +0200, Richard Biener wrote: > > On Tue, 16 Jul 2024, Jakub Jelinek wrote: > > > > > On Tue, Jul 16, 2024 at 10:43:27AM +0200, Richard Biener wrote: > > > > I've pushed it to trunk now and am running local CPU 2017 to check for > > > > obvious fallout on Zen4 so we can make 14.2 RC early next week. There's > > > > still the question of GCC 11.5 which got the backport of zen4 support > > > > with this "wrong" costs but RC1 was already last week and we're set > > > > to release on Friday. > > > > > > > > I'd like to hear your opinion on that (13.3 and 12.4 also got the bogus > > > > value so eventually 11.5 getting the bogus value isn't too bad). > > > > > > > > Btw, I just see that znver5 tables have the same issue, I'll push the > > > > obvious change there as well. > > > > > > I'd just change 11.5 and do a RC2. > > > > I'm undecided on how important it is or whether it would need a RC, I > > don't expect anybody to use -march=znver4 with a GCC 11 code-base since > > it wasn't available there with 11.4. > > -march=native is something people will actually use even with GCC 11 code > base and I've just tested that on znver4 box > ~/xgcc -B ~/ -S -march=native -xc /dev/null -o /dev/null -v 2>&1 | sed -n > 's/^.* \(-march=[0-9a-z]*\) .*$/\1/p' > -march=znver4 > ~/xgcc --version > xgcc (GCC) 11.4.1 20240620 > Copyright (C) 2021 Free Software Foundation, Inc. > This is free software; see the source for copying conditions. There is NO > warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. > > So, I think we really should fix this for 11.5.
OK, note we have the broken state in 12.4 and 13.3 so that got some testing there. I actually expect less problems with the fixed values as the bogus ones will cause alignment peeling that's not seen before with -march=native, possibly triggering latent issues like PR115841 (not that in particular since GCC 11 isn't affected). I'll push the cost adjustment to 11 later but will wait with the other branches to watch SPEC results as Honza requested. Richard.
