Performance improvement: use @mp_sqr@ and @mpmont_reduce@ instead of