On Fri, Apr 14, 2023 at 10:15:06AM +0000, Richard Biener wrote:
> Oops. Indeed target_avx checks whether it can compile sth with
> -O2 -mavx rather than verifying avx is present. I've seen scan
> failures with -m32/-march=cascadelake on a zen2 host. I'm not exactly
> sure why.
That is strange. Sure, -march=cascadelake implies -mavx (-mavx512f even),
but it would surprise me if on such a host avx_runtime wasn't true.
But we've been there before, I think cascadelake turns on the vector
epilogues.
In r13-6784 I've added --param vect-epilogues-nomask=0 to some testcases
that were affected at that point, but perhaps something is affected since
then. Will have a look.
Jakub
