On Thu, Oct 31, 2024 at 08:21:09AM +0000, Kugan Vivekanandarajah wrote:
>
>
> > On 31 Oct 2024, at 6:18 pm, Jakub Jelinek <ja...@redhat.com> wrote:
> >
> > External email: Use caution opening links or attachments
> >
> >
> > On Tue, Oct 29, 2024 at 05:01:40AM +0000, Kugan Vivekanandarajah wrote:
> >> For param_vect_max_version_for_alias_checks of 15, the average code size
> >> increase is 0.058%
> >
> > The test FAILs e.g. on i686-linux, using vect_double etc. outside of
> > */vect/ directories is wrong, as the effective target checks rely on the
> > extra flags added by vect.exp.
> > Can you please move the alias-checks.C test to g++.dg/vect/
>
> Sorry for the breakage. Testing in i686-linux as you suggested. Is it OK to
> if that passes regression.
No.
The ChangeLog entry should look more like
* g++.dg/alias-checks.C: Move to ...
* g++.dg/vect/alias-checks.C: ... here.
or so, but more importantly, in */vect/ one shouldn't use dg-options, that
overrides the options carefully set up by vect.exp.
The -O3 option should be supplied in dg-additional-options,
-fdump-tree-vect-all should be likely dropped, vect.exp already adds
-fdump-tree-vect-details and that should be sufficient for the test IMHO.
Jakub
