https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118460
--- Comment #6 from Richard Earnshaw <rearnsha at gcc dot gnu.org> --- Yes, that would work as a work-around. It's a bit of a hack though. The expectation is that we would use vsel for pretty much everything when it is available (particularly for fast-math), but we fail to do that because we don't check properly when expanding the mov?fcc patterns. I suspect there may be additional problems if -mrestrict-it is enforced (v8-a, mostly) because the conditional move patterns are disabled in that case.
