https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127182
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
It fails also on trunk, but from what I can see, only when using -mcpu=power8,
when using -mcpu=power9 or -mcpu=power10 it passes.
https://gcc.gnu.org/pipermail/gcc-testresults/2026-September/885931.html
which is --with-cpu=power8 has the FAILs, while
https://gcc.gnu.org/pipermail/gcc-testresults/2026-September/885930.html
which is --with-cpu=power9 doesn't, nor
https://gcc.gnu.org/pipermail/gcc-testresults/2026-September/885932.html
which is --with-cpu=power10.
Both the usadv16qi and usadv8hi named patterns have TARGET_P9_VECTOR condition,
so clearly for powerpc* we need at least -mcpu=power9 to vectorize those.
