> Am 29.03.2026 um 10:28 schrieb Alexandre Oliva <[email protected]>:
>
> On Mar 29, 2026, Richard Biener <[email protected]> wrote:
>
>> Moving the test is OK.
>
> I'll split out that big, then.
>
>> But the vect_double check looks wrong to me, esp as you are moving a
>> compile test.
>
> Call that unrelated and I'll go for it. I just happened to notice and
> wonder why we were even trying that testcase with vect_double given that
> the target didn't have any vectors whatsoever. On ppc, for example,
> vect_double checks for vsx hw. Why would it be wrong to check
> analogously for sse2 hw on x86? Conversely, why would it be correct for
> vect_double to return true for hardware that doesn't have any vectors
> whatsoever?
>
> I realize that the minimum standards set and expected by vect.exp on
> x86, of compile-time sse2 as the bare minimum, could support the notion
> of returning vect_double on x86 unconditionally, if vect_double is to be
> used only within vect.exp tests. But it's a deviation of the
> documentation and of practice on other targets that follow the
> documentation. The inconsistency seems undesirable and disturbing to
> me.
It is not supposed to be used outside of vect.exp and it relies on that passing
-msse2 unconditionally on x86.
It would be best if all of the vector targets would be in another .exp file so
they would not be visible unless explicitly included.
Richard
>
> That said, there's no regression in that regard. Please let me know
> whether to resubmit that bit at stage1 or not at all.
>
> --
> Alexandre Oliva, happy hacker https://blog.lx.oliva.nom.br/
> Free Software Activist FSFLA co-founder GNU Toolchain Engineer
> More tolerance and less prejudice are key for inclusion and diversity.
> Excluding neuro-others for not behaving ""normal"" is *not* inclusive!
