https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110717
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Thanks. Shouldn't INTVAL (operands[2]) < <MODE_SIZE> * BITS_PER_UNIT be UINTVAL (operands[2]) < <MODE_SIZE> * BITS_PER_UNIT just to make sure it doesn't trigger for negative?