https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70300
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 38043
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38043&action=edit
gcc6-pr70300.patch
This is complete mess, I bet it would be easiest to disable these AMD tuning
flags when we have AVX512F enabled, otherwise we'll running into issues with
these hacks for a long time.
Anyway, in this case the problem is that vec_interleave_lowv4sf pattern doesn't
have =v, v alternative (that would be only possible for AVX512VL though), and
that for AVX512F but not VL we need to do something different anyway (either
unpack in destination only, disable the splitter altogether, ...).
