Some notes...

This gist contains a bunch of optimized tableless CRC implementations. So far I've found CRC-16 and CRC-32 versions which work, and seem
faster than the typical double-looped variants. I'd like to modify the code to work with CRC-8 and other polynomimals, but so far it is
quite cryptic how they work.

As far as performance goes.. they are about 2x or 3x faster than the double-looped variants. But that's not saying much, because
it's still 35% slower than HalfSipHash. And it cannot even compare to MurmurHash3 (~80% slower).
CRC may be able to be optimized with multi-byte read versions.

However for CRC-32, it seems that you can unroll the inner-loop and performance is still comparable to the hand-optimized version
(within 10%). So take that with a grain of salt.