http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57660
--- Comment #2 from Jeffrey A. Law <law at redhat dot com> --- We've got two choices here. First is to somehow make the number of replacements we look for be conditional on the target. For targets with an appropriate branch cost (such as m68K) we'll see 2X the number of replacements as x86_64. The other approach is to disable the test on those targets, much like we do with vrp87.c. I'm investigating the former, if I can find a clean way to do that it'd be the preferred solution. Otherwise I'll do the latter.
