SHA-3 wrapper (keccak) for Python. The package is a wrapper around the optimized reference implementation from . 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 supported.


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 .

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 ="sha3_512")
>>> s = hashlib.sha3_512() # alternative
>>> s.digest_size
>>> s.update(b"data")
>>> s.hexdigest()

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 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)
lane complementing

64bit platforms

default settings of common options


Additional optimiation instructions (disabled by default):


use Stream SIMD extensions

limit to 64bit instructions, otherwise 128bit
w/o UseOnlySIMD64
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)

32bit platforms

default settings of common options

3 [default]
unknown, no UseBebigokimisa, Unrolling must be 2
use two 64k lookup tables for (de)interleaving (disabled by default)