CRC32 Performance

[damieng]

CRC32 Performance

[damieng]

You can't parallel process it but there are some tricks with slicing the table up if you don't mind the lookup table going from ~1KB to 8KB that Intel developed. I've put a new class up called Crc32Slice8 you could try - it should be drop-in compatible.

Let me know how you get on - I've not run extensive benchmarks on it yet.

[jayakrishnabh]

Hi Thanks for the Crc32Slice8.
For the same 55KB buffer size, performance improved by 0.3 ms per message processing.

[damieng]

A 16 slice version is possible but if the performance increase in your environment is as bad as that it probably isn't worth the effort.

[jayakrishnabh]

Hi Damien, It would be great to have 16 slice version if it improves the performance further better.
Please share the 16 slice version. Will try this so that our performance increases further.

[damieng]

Crc32Slice16 is now checked in.

[damieng]

I don't have any other ideas for improved performance so I'm going to close this down now.