one call for the fastest version?

[bobjen...@gmail.com]

Going from memory here ... I timed CityHash128 and CityHash256, and 128 was faster for strings under 1000 bytes, but 256 was faster for longer strings.  Could you add an interface that looks at the length of the string, and calls the fastest variant of CityHash for that length?