This is about 25% faster than its predecessor for large numbers.

--HG--
extra : amend_source : 9917efd5e13a7284857236e9162614a0731e0636