https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65752
--- Comment #51 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Jeffrey A. Law from comment #50) > Richi, > I haven't followed this BZ at all, but I absolutely trust you on issues WRT > alias analysis. If we can't propagate these conditional equivalences for > pointers, I'll happily tweak DOM to avoid that. Unfortunately it isn't that easy - even propagating equivalences for integers may cause the same issue. And even if we fix all issues on GIMPLE we're still left with the fundamental brokeness of RTL alias analysis (PR49330).
