anycrc 0.7.3

Python CRC Computation Library

This is a Cython module with bindings to the crcany library. It supports calculating CRC hashes of arbitary sizes as well as updating a crc hash over time. It takes advantage of crcany's ability to efficiently combine multiple CRCs to parallelize the CRC's calculation.

Installation

pip install anycrc

Usage

Use an existing model:

>>> import anycrc
>>> crc32 = anycrc.Model('CRC32')
>>> crc32.calc(b'Hello World!')
472456355

Read the data in chunks:

>>> crc32.update(b'Hello ')
3928882368
>>> crc32.update(b'World!')
472456355

The update method changes the internally stored CRC value, while calc doesn't. You can use get to retrieve the CRC value stored within the object:

>>> crc32.get()
472456355

To specify the starting CRC value:

>>> crc32.set(3928882368)
>>> crc32.calc('World!')
472456355

To go back to the initial value, use:

>>> crc32.reset()

Specify your own CRC parameters:

>>> # width, poly, init, refin, refout, xorout
>>> my_crc = anycrc.CRC(10, 0b0101010101, 0x3ff, True, False, 0)
>>> my_crc.calc('Hello World!')
35

The CRC will be computed in parallel when the input is large enough. If you want to disable parallelism, use:

>>> anycrc.set_parallel(False)

For a list of pre-built models, check models.py. To get a list of the models at any time, use the following command:

python -m anycrc models

The maximum possible CRC width is 128 bits.

Benchmarks

Calculating the CRC32 for lorem ipsum 1 million times:

Module	Time Elapsed (s)	Speed (GiB/s)	Relative
anycrc	0.19	2.14	1.00
zlib	0.28	1.49	1.44
fastcrc	0.80	0.52	4.12
crcmod-plus	0.91	0.46	4.71

Calculating the CRC32 for the text of lorem ipsum repeated 1 million times in a single pass:

Module	Time Elapsed (s)	Speed (GiB/s)	Relative
anycrc (parallel)	0.02	24.51	1.00
anycrc (serial)	0.13	3.21	7.64
zlib	0.22	1.93	12.72
fastcrc	0.67	0.62	39.75
crcmod-plus	0.67	0.62	39.73

Tested on a 10th generation Intel i7 processor. Parallel performance will depend on your system.

Notes

Parallelism is disabled when the length of the input data is under 200k, as the serial method is faster in that case.

The input needs to be very large in order to notice the speed advantage of the parallel algorithm.