https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72785
--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> --- I believe the testcase is simply abusing b_c_p. Can you elaborate on the "as-if" rule here? Note we drop all b_c_p to true/false in the first VRP pass (after inlining), so a less agressive fix would be to simply make sure this happens before any path isolation is done (of course we now have early threading...) The BB duplication CFG hook is certainly too aggressive for this "fix" as it doesn't affect only path isolation.
