https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82795 --- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> --- Well, not really a "regression", it was broken before the original fix. The assert is simply bogus.