https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68398
--- Comment #5 from Jeffrey A. Law <law at redhat dot com> --- Doing it after the loop optimizer doesn't help. I haven't really looked into why. The concerns around not creating new subloops or multiple latches pre-date a lot of the loop infrastructure changes Richi has made, so it may make sense to relax those -- I'll have to look into that more deeply.