https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120067
Robin Dapp <rdapp at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rdapp at gcc dot gnu.org
--- Comment #4 from Robin Dapp <rdapp at gcc dot gnu.org> ---
(In reply to Paul-Antoine Arras from comment #3)
> (In reply to rdapp.gcc from comment #2)
> > There is indeed a problem with =zvl that we have been discussing in the
> > patchwork sync call for a while already. The issue is that we're handling
> > VLS
> > modes as VLA modes in certain situations resulting in unfortunate codegen.
> > The
> > vector extracts (slidedowns) are one example of it.
>
> I guess this is related to
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119361#c2 then.
> Is there anything I could do to help with that?
Yes, that's it. I don't think right now there is a lot to do for this
particular issue. I have a local patch that still needs to be thoroughly
tested. Going to pick it up again in a few weeks.
But it will surely be helpful if you continue to weed through SPEC looking for
cases where our vectorization seems dubious - either vs aarch64 or vs LLVM.
SPECfp in particular hasn't seen much attention AFAICT. Same with SPEC 2006.
I have looked at some of its integer benchmarks (or rather the open source
applications they are derived from) but didn't spend a lot of time on it.
Maybe just wait a bit before drawing definite conclusion from pure icount
numbers :)
