AVX-accelerated 8-way parallel md5
This package contains the Intel AVX-specific code for a parallel md5 kernel.
The function which this package exports has the signature:
func WriteN(h []hash.Hash, p [][]byte) (n int, err error)
which provides an up-to-8-way parallel invocation of md5 rounds
on the given byte slices. (by analogy to hash.Write)
Note: all byte slices must be of the same length, and len(p) must be the
same as len(h). Furthermore, h must represent a collection of crypto/md5
hash digests, as created by md5.New(). Therefore, after calling
WriteN() any trailers may be written directly via hash.Write() and
hash.Sum() may be used to compute the final md5 checksum for each
digest.
An associated "server" will be published to provide transparent
parallelism via a drop-in replacement for the hash.Hash interface.