evereq / lz4sharp

Port of LZ4 Compression algorithm to C#



LZ4Sharp

Port of Yann Collet’s LZ4 algorithm to C# by Clayton Stangeland.

Original LZ4 algorithm can be found at (http://code.google.com/p/lz4) it was created by Yann Collet. See license.txt for project license.

Build

Building in Debug or Release will use the already generated C# files

LZ4Compressor.c and LZ4Decompressor.c are updated by manually comparing to the original LZ4 code and fixing any C# differences from C but leaving some of the macros to make comparing easier.

Then building in the Generated configuration will only build the LZ4SharpGenerator project which has a prebuild event that uses the MinGW gcc from CodeBlocks to generate the C# files from LZ4Compressor.c and LZ4Decompressor.c.

Timings

Compression times do not include disk I/O.

P4 = Pentium 4 2GHz 32 bit Windows XP

i5 = Intel i5 2.67 GHZ 64 bit Windows 7 running LZ4Sharp in 64 bit

LZ4Sharp = Timings for ‘webster’ from silesia corpus (http://sun.aei.plsl.pl/~sdeor/index.php?page=silesia)

memcpy and LZ4 = Timing for file ‘webster’ in silesia corpus in m2mark.exe from http://sd-1.archive-host.com/membres/up/182754578/m2mark.zip (or from link to benchmark program at http://code.google.com/p/lz4)

gzip = GZip from the Intel compiler built as a dll and accessed from C# by PInvoke

Ratio = compressed size / decompressed size (lower is better)

Computer	Type	Compression (MB/s)	Decompression (MB/s)	Ratio (%)
P4	memcpy	217
P4	LZ4	82	304
P4	LZ4Sharp	43	150
P4	LZ4Sharp	49	181
i5	memcpy	1658
i5	LZ4	270	1184
i5	LZ4Sharp	207	758	49
i5	LZ4Sharp	267	838	47
i5	gzip	48	266	33

Conclusion:

LZ4 C# is about 2/3 the speed of the c version. (Also, LZ4C# is slightly faster on the whole silesia corpus than on just the ‘webster’ file) LZ4 C# Compressed silesia corpus is 47% of uncompressed LZ4 C# Compressed ‘webster’ file is 49% of uncompressed