[NEWS][ZLIB] Intel Makes Major Zlib Performance Improvements

[AlexeyKnyshev]

Intel's Open-Source Technology Center has prepared a set of thirteen patches against Zlib to sharply improve the deflate performance.

Jim Kukunas of Intel OTC published the set of 13 patches on Monday that include medium and quick deflate strategies, a faster hash function with SSE 4.2 support, PCLMULQDQ-optimized CRC folding, SSE2 hash shifting, and other changes/tuning.

In terms of the deflate performance changes for Zlib with these patches, Jim noted that for an Intel Core i5 2530M (Sandy Bridge) system, level one compression is 71% faster with a sacrifice of about 30% compression (which more exciting for Ogre), level 6 is about 50% faster with no real change in compression, and level 9 is about 22% faster with no change at all in the compression.

More details on the Zlib DEFLATE performance improvements that will hopefully be merged soon into the data compression library can be found via the Zlib mailing list.

News from http://www.phoronix.com/ (Original)

[masterfalcon]

That's pretty exciting. It'll be great for those changes to get merged in.

[oiram]

Great news!!!

[bstone]

Very interesting but how would increased compression performance benefit Ogre exactly?

[Kojack]

In the top post it said decompression was 71% faster (with a compression ratio loss) or 50% with almost no compression loss.

[bstone]

In the top post it said decompression was 71% faster (with a compression ratio loss) or 50% with almost no compression loss.

Didn't it say "level one compression is 71% faster with a sacrifice of about 30% compression"? Which makes sense to me. While gaining decompression speed while sacrificing compression level does not (for deflate)!