NOTE! The NEON version of blake2s is currently NO FASTER than the
reference implementations. However, it is retained for reference
and in case it can be further improved.
The NEON version of blake2b is more than twice as fast as the
reference implementation on the Raspberry PI 2 Model B.
To simplify development we can ignore all output produced by running 'make' in
the b2sum, bench, ref and sse directories.
Someone should build the csharp implementation and add output files to
.gitignore...