https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122608
--- Comment #1 from Jeffrey A. Law <law at gcc dot gnu.org> --- I would have expected that transformation as well. But I also happen to know that we're not that strong at creating COND_EXPRs in gimple -- there's multiple classes of this kind of problem we can and should fix. Daniel and I were looking at one recently involving returning +1, -1 that ought to be improved and shows up meaningfully in spec2017 and I'm hopefull it shows up generally in qsort comparison functions, but I haven't done that analysis.
