https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122079
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- And comment #1 shows I think why it isn't enough to handle MAX on the expression side. If we'd track a "generation count" for the ANTIC_IN set we combine from successors you'd see that we happily mix generations at the points we initially consider it to be MAX.
