https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111280
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- Yeah, I think emitting .CLZ (t, 64) and compensating at RTL expansion would be best for this particular issue. I'm not sure what the constraints currently are - are we validating the 2nd argument at all when expanding? I don't think so (neither in verify_stmt). So we should at least validate against target behavior, maybe add compensation code.
