Threading doesn't make much sense here. The runtime of hash computation is usually dominated by I/O performance. On a typical consumer computer SSDs have a sequential read performance of 200 to 500 MiB/sec. SHA-512 performance between 100 to 150 Mib/sec. Threading could make parallel computation a bit faster, but at the cost of a much more complex implementation. Let's keep it simple.

Both Python's hashlib implementation and OpenSSL aren't the best foundation if you are aiming for maximum performance. All libraries do couple of malloc()s, memcpy() and have additional overheads, too.

And please don't use MD5 in your examples. Go for sha256.