https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90037
--- Comment #11 from Jeffrey A. Law <law at redhat dot com> --- That may be an interesting approach. I think we'd want the new blocks created by threading as well as the original blocks we threaded through since their in-degree gets reduced which in turn can expose new cprop opportunities. We'd have to somehow be able to mark or otherwise remember the blocks in question across a call to cfgcleanup so we'd know what SEME regions to optimize. But I guess missing one or more regions isn't fatal, it's "just" a missed optimization.