http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60851
--- Comment #3 from Richard Henderson <rth at gcc dot gnu.org> --- Oh poo. This is essentially the same as PR60704. There I chose to fix it in the backend, in a very hacky sort of way, because I thought these insns were passing constrain_operands and it was just the use of the enabled attribute that was different. I think accepting an "m" with !strict should be ok. Or if that's not something we want to do on the 4.9 branch, the additional constraint letter should work as well. This ought to mean we could revert the patch for PR60704 too.