Arrays in Julia are column-major, so by iterating through them row by
row the program is constantly refilling the cache and leaves a lot of
performance on the table.

The results are mostly identical with the output from the C reference,
but there are some floating point errors that prevent the hashes from
matching.

This option is
 - redundant, the OpenMP threads can be set with an environment variable
 - not portable to other threading frameworks (e.g. Julia)
 - not necessary, because we want all threads to be used