Re: [Algorithms] Comparing apples to apples : compression libraries
Brought to you by:
vexxed72
Menu
▾
▴
Re: [Algorithms] Comparing apples to apples : compression libraries
|
From: Jason H. <jas...@di...> - 2008-03-27 04:43:24
|
One can argue that anything optimized for SPU no longer resembles the original library. :-) I've looked over the Huffman format description for ZLib and a few details seemed unclear. LZ is going to perform pretty well regardless since your data is always still in the L2 cache, if not the L1. But you can implement Huffman in many ways, and in my own implementations just changing between a naive bit-shifting model and a cascaded table-driven approach can give you 300% the performance. On PC, where cache sizes are pretty large and OOE and long pipe stalls can hurt. I haven't had a chance to run the same tests on consoles (yet), but I expect cascaded tables with an 8-bit root table would still win. Maybe not on PS2. Not sure how many games are really bottlenecked by their decoders, but if yours is, it might be worth seeing how much you get back by switching Huffman off. I expect a bad implementation can waste a significant number of cycles while streaming. Thanks, JH Disclaimer: I took no part in the ZLib optimizations for Edge. I was just making a funny. gd...@er... wrote: > Hi John, > > Thanks for the very useful work again and a nice time saver. > > >>> My experiments with zlib on consoles was very disappointing. >>> > Which console(s)? > > Playstation 3 Edge libraries, available to PS3 licensees, have an SPU > optimized version for ZLib, that can decompress 40MB/sec using 25% of 1 SPU: > > From http://forum.beyond3d.com/showpost.php?p=956489&postcount=196 > > Thanks, > Erwin > > > > > Glen Miner writes: > > >> I'm curious what hardware this was tested on. My experiments with zlib on >> consoles was very disappointing. >> >> peace >> >> "John W. Ratcliff" <jo...@si...> wrote in message >> news:004f01c88ed0 >> >> Testing Compression rate and speed with various compressors. >> --------------------------------------------------------------- >> Compress:CT_CRYPTO :FROM: 10,436,335 TO: 2,498,433 23% 1,170 MS >> Compress:CT_MINILZO :FROM: 10,436,335 TO: 3,940,072 37% 97 MS >> Compress:CT_ZLIB :FROM: 10,436,335 TO: 3,299,771 31% 157 MS >> Compress:CT_BZIP :FROM: 10,436,335 TO: 2,270,695 21% 1,544 MS >> --------------------------------------------------------------- >> Testing Decompression speed with various decompressors. >> --------------------------------------------------------------- >> Decompress:CT_CRYPTO :FROM: 2,498,433 TO: 10,436,335 258 MS >> Decompress:CT_MINILZO :FROM: 3,940,072 TO: 10,436,335 42 MS >> Decompress:CT_ZLIB :FROM: 3,299,771 TO: 10,436,335 69 MS >> Decompress:CT_BZIP :FROM: 2,270,695 TO: 10,436,335 390 MS >> >> >> >> >> ------------------------------------------------------------------------- >> Check out the new SourceForge.net Marketplace. >> It's the best place to buy or sell services for >> just about anything Open Source. >> http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace >> _______________________________________________ >> GDAlgorithms-list mailing list >> GDA...@li... >> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list >> Archives: >> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list >> > > > > ------------------------------------------------------------------------- > Check out the new SourceForge.net Marketplace. > It's the best place to buy or sell services for > just about anything Open Source. > http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace > _______________________________________________ > GDAlgorithms-list mailing list > GDA...@li... > https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list > Archives: > http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list > > |
