https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126923
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |testsuite-fail
Last reconfirmed| |2026-08-18
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
That's expected ... the first step would be to clean up x86-64-v3 (or even
-v2), Tamar has posted RFC patches to integrate such into regular testing of
vect.exp at least.
In the majority of cases the issue is extra vector epilogs that alter the
behavior. For testcases that are not about epilog vectorization I'd suggest
to add --param vect-epilogues-nomask=0 to disable epilog vectorization (but
you'd also have to double-check results on aarch64/riscv).
I'd not recommend to cater for different behavior due to -mtune (znverN turn
alter some vector tune-ctrls for example) but stick to -mtune=generic at first.
