The implemenation is supposed to be reasonably portable, incl. little endian and big endian systems, and reasonably fast.

• MD5 / RFC 1321
• SHA1 / RFC 6234
• SHA224 / RFC 6234
• SHA256 / RFC 6234
• SHA384 / RFC 6234
• SHA512 / RFC 6234

• RFC 2104 / 6234

• HKDF / RFC 5869 / 6234
• PBKDF2 / RFC 2898 / 6070

The code is C++11 code without any external dependencies.

All algorithms will be tested by issueing make check, which happens to be the default target. The test will by default hash all hash* files (aka. the source and binaries) plus somme vectors in test-vectorsand compare the output against what the Python hashlib produces.

Passing tests:

• Debian lenny x86_64 le, gcc-4.7
• Debian wheezy mips/manta be, gcc-4.7
• OSX 10.9/Darwin x86_64 le, clang-apple-5.1
• OSX 10.9/Darwin i686 le, clang-apple-5.1
• OSX 10.9/Darwin x86_64 le, gcc-4.8
• OSX 10.9/Darwin i686 le, gcc-4.8
• Windows x86_64-w64-mingw32 gcc-4.8.3
• Windows WIN32 MSVC12u2 (create your own .sln)
• Windows WIN64 MSVC12u2 (create your own .sln)

Well...

Any copyright is dedicated to the Public Domain.
http://creativecommons.org/publicdomain/zero/1.0/
Written in 2014 by Nils Maier

This stuff is probably OK, and it has tests. Then again, it was not really peer reviewed...

As for performance: It is pretty much on par with other implementation, incl. the ones found in OpenSSL and/or Apple Common Crypto. It is not meant to be the most performant implementation on earth, though.