Thanks for catching the issue! It's clearly not as "obvious" as I initially
thought. :)

I have split up the test as you suggested. Please let me know if I missed
anything.
https://gcc.gnu.org/pipermail/gcc-patches/2026-January/705975.html

Thanks,
Pengxuan

On Thu, Jan 15, 2026 at 12:23 PM Jakub Jelinek <[email protected]> wrote:

> On Thu, Jan 15, 2026 at 12:02:49PM -0800, Pengxuan Zheng wrote:
> > This fixes failures on i686 and arm.
> >
> > Pushed to trunk as obvious.
>
> That is not obvious, it is actually wrong.
> x86_64-*-* can mean both 64-bit and 32-bit (e.g. when testing with
> --target_board=unix\{-m32,-m64\} or even
> --target_board=unix\{-m32,-m32/-mno-sse,-m64\}
> ).  Furthermore, some x86_64-*-* targets like Solaris default to -m32.
>
> In effective targets when using target triplets, you always want to use
> both x86_64-*-* i?86-*-* next to each other.  Now, either you also add
> /* { dg-additional-options "-msse2" { target i?86-*-* x86_64-*-* } } */
> or need to use something like
> { { { x86_64-*-* i?86-*-* } && lp64 } || aarch64-*-* }
> as effective target (or { ! ia32 } instead of lp64).
> But I think the dg-additional-options is better.
>
> Anyway, as I wrote in my earlier mail, this means that for scalars
> it will also be tested just on the two arches, perhaps that is
> undesirable and splitting the test is still worthwhile.
>
> Or as Jeff said, scan earlier dump instead of optimized and add -Wno-psabi
> and test on all targets.
>
> > gcc/testsuite/ChangeLog:
> >
> >       * gcc.dg/pr123109.c: Require target x86_64 and aarch64.
> >
> > Signed-off-by: Pengxuan Zheng <[email protected]>
> > ---
> >  gcc/testsuite/gcc.dg/pr123109.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/gcc/testsuite/gcc.dg/pr123109.c
> b/gcc/testsuite/gcc.dg/pr123109.c
> > index 9b83d578713..0d51863ca40 100644
> > --- a/gcc/testsuite/gcc.dg/pr123109.c
> > +++ b/gcc/testsuite/gcc.dg/pr123109.c
> > @@ -1,4 +1,4 @@
> > -/* { dg-do compile } */
> > +/* { dg-do compile { target { x86_64-*-* aarch64-*-* } } } */
> >  /* { dg-options "-O2 -fdump-tree-optimized" } */
> >
> >  typedef int v4si __attribute__((vector_size(4 * sizeof(int))));
>
>         Jakub
>
>

Reply via email to