https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121165
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #4)
> (In reply to Andrew Pinski from comment #3)
> > # PT = { D.371302 D.371303 }
> > # ALIGN = 8, MISALIGN = 0
> > # _129 = PHI <&D.371302(3), &D.371303(5)>
> > # .MEM_461 = VDEF <.MEM_458>
> > # USE = nonlocal escaped { D.371302 D.371303 }
> > # CLB = nonlocal escaped
> > .UBSAN_NULL (_129, 0B, 8);
> > # VUSE <.MEM_461>
> > avail_79 = *_129;
> >
> > So basically phiprop no longer does it work. so we missed out a lot.
>
> And I'm not sure we can do anything about this besides trying to have
> phiprop elide the .UBSAN_NULL when the PHI def of _129 is a an address
> of a decl on an edge and hoist it as well for the case of an SSA name?
>
> But would that be appreciated and correct?
I filed PR 121168 to track the phiprop issue because I am not 100% convinced it
is the only issue with the original testcase.
