https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55157
--- Comment #2 from Marc Glisse <glisse at gcc dot gnu.org> --- Instead of looking at the definition, it could also look at get_nonzero_bits. It would help when being a multiple of 16 is less obvious, but it wouldn't do anything useful for t*15. It might be easier to optimize g<=4 to g==0 and let existing optimizations do the rest.
