https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127166
--- Comment #2 from Jeffrey A. Law <law at gcc dot gnu.org> --- Yup. That's a real possibility (that the test is bogus). It's also worth noting those were all "vi" forms, so it may be the case that when we know one argument is an integer that we don't have to be as conservative. I haven't really looked at them to really know either way.
