https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70005
--- Comment #2 from Jeffrey A. Law <law at redhat dot com> --- Thanks. Something unexpected slipped through the IL confusing uncprop. So there's really two issues here. Why did an always-true/always-false test manage to survive until uncprop, a missed optimization, but probably not a regression. And uncprop should be tolerant of that IL state. I fixed the same thing in DOM a while back, but clearly forgot to update uncprop with the same fix. Testing the obvious fix to uncprop right now.
