https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124726
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Torbjorn SVENSSON from comment #2)
> (In reply to Richard Biener from comment #1)
> > They also pass for -march=znver4, I think the simple
> >
> > /* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" { xfail { { !
> > aarch64_sve } && { ! riscv_v } } } } } */
> >
> > are simply bad for maintainance. I'm not sure what we need here,
> > it likely needs to be evaluated on a case by case basis. A sample
> > of two reveals that vect_condition and/or vect_cond_mixed might be
> > needed.
>
> I do not think it's as simple as to use `vect_condition`.
>
> If I change the xfail expression in the test listed in comment 0, it will
> work for the targets I test on arm-none-eabi, but will start to fail for
> x86_64-linux-gnu.
>
> There are also a bunch of other tests in the same directory that does test
> for "vectorized 1 loops" that are either xfail or pass.
>
> For the targets that I test on arm-none-eabi, most of them are unsupported.
> Only Cortex-A7 with -mfloat-abi=hard appears to actually run these tests.
>
> Based on this, I see no other option than adding arm*-*-* or arm-none-eabi,
> even if it's ugly.
can we at least have something more ISA specific, like an effective target for
neon or mve or whatever it causes to XPASS?
