https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119393
Alex Coplan <acoplan at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |acoplan at gcc dot
gnu.org
--- Comment #5 from Alex Coplan <acoplan at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> >I've attached a reduced reproducer which (unfortunately) still requires LTO,
> >but it is at least fairly well reduced (only two small TUs).
>
> One thing you could do is combine the 2 TUs (since they are small) and then
> use -fwhole-program or add static on some of the variables. that might be
> enough to get one TU.
Yeah, I tried that (no such luck, in that case both before/after compilers seem
to give the "bad" codegen). I don't fancy using -fgimple since the GIMPLE
dumps always need significant massaging to get accepted by the GIMPLE FE.
By all means have a go at getting a non-LTO reproducer, for now I will try to
debug where things go wrong in the vectorizer.
