SHA-3 wrapper (keccak) for Python. The package is a wrapper around the
optimized reference implementation from http://keccak.noekeon.org/ . Only
the optimizations for 32 and 64bit platforms are used. The optimized SSE and
ARM assembly variants are ignored for now.
The module is a standalone version of my SHA-3 module from Python 3.4
(currently under development). The code in sha3module.c has been modified to
be compatible with Python 2.6 to 3.4. Python 2.5 and earlier are not
pysha3 has been successfully tested on at least 8 CPU architectures, 7 compiler
families and 10 operating systems families:
- Linux (GCC 4.3 and 4.6, clang 3.0) on X86, X86_64 and ARMv6 (little endian)
- Windows (VS 2008, VS 2010) on X86 and X86_64
- FreeBSD 7.4 - 10 (GCC, clang) on X86 and X86_64
- NetBSD 5.1 (GCC) on X86 and X86_64
- Mac OS X 10.8 (clang4) on X86_64
- HP-UX (HP C/aC++) on IA64 (little endian) and PA-RISC (big endian)
- Solaris 10 (Oracle Solaris Studio 12.3) on SPARC (big endian)
- Solaris 11 (Oracle Solaris Studio 12.3) on X86_64.
- AIX (XLC 12.1) on PowerPC (big endian)
- Tru64 (Compaq C) on Alpha (little endian)
Thank you very much to Trent Nelson for http://www.snakebite.net/ .
Donation for ARMv7 support
The Keccak reference implementation contains optimized assembly code for ARM.
However neither Trent nor I have access to ARM platforms except for my 1st
generation Raspberry Pi. We gladly accept hardware donations of ARM boards
and periphery, either to the Snakebite network or to me personally. Please
contact me if you like to help.
The `sha3` module contains several constructors for hash objects with a
PEP 247 compatible interface. The module provides `sha3_228()`, `sha3_256()`,
`sha3_384()`, and `sha3_512()`.
The `sha3` module monkey patches the `hashlib` module . The monkey patch is
automatically activated with the first import of the `sha3` module. The
`hashlib` module of Python 3.4 will support the four SHA-3 algorithms
on all platforms. Therefore you shouldn't use the sha3 module directly
and rather go through the `hashlib` interface::
>>> import sys
>>> import hashlib
>>> if sys.version_info < (3, 4):
... import sha3
>>> s = hashlib.new("sha3_512")
>>> s = hashlib.sha3_512() # alternative
**Don't use SHA-3 for HMAC!** HMAC hasn't been specified for SHA-3 yet and no
test vectors are available, too.
Comments from sha3module header
The code is based on KeccakReferenceAndOptimized-3.2.zip from 29 May 2012.
The reference implementation is altered in this points:
- C++ comments are converted to ANSI C comments.
- All functions and globals are declared static.
- The typedef for UINT64 is commented out.
- brg_endian.h is removed.
- KeccakF-1600-opt[32|64]-settings.h are commented out
- Some unused functions are commented out to silence compiler warnings.
In order to avoid name clashes with other software I have to declare all
Keccak functions and global data as static. The C code is directly
included into this file in order to access the static functions.
Keccak can be tuned with several paramenters. I try to explain all options
as far as I understand them. The reference implementation also contains
assembler code for ARM platforms (NEON instructions).
loop unrolling (24, 12, 8, 6, 4, 3, 2, 1)
default settings of common options
Additional optimiation instructions (disabled by default):
use Stream SIMD extensions
limit to 64bit instructions, otherwise 128bit
requires compiler argument `-mssse3` or `-mtune=core2` or better
use 64bit MMX instructions
use AMD's eXtended Operations (128bit SSE extension)
When neither `UseSSE`, `UseMMX` nor `UseXOP` is configured, `ROL64`
(rotate left 64) is implemented as:
use shld (shift left) asm optimization
shift and xor
`UseBebigokimisa` can't be used in combination with `UseSSE`, `UseMMX` or
`UseXOP`. `UseOnlySIMD64` has no effect unless UseSSE is specified.
Tests have shown that `UseSSE` + `UseOnlySIMD64` is about three to four
times SLOWER than `UseBebigokimisa`. `UseSSE` and `UseMMX` are about two
times slower. (tested by CH and AP)
default settings of common options
unknown, no `UseBebigokimisa`, `Unrolling` must be 2
use two 64k lookup tables for (de)interleaving (disabled by default)