4 ms·
SPIR-V Compression
- giovannibajo1 10y agoInteresting experiment, I'm always curious about data-class-dependent filters that affect compression. In this case, I must say that it's not really exciting to see that, after all this work, zstd level 20 is comparable. If I were to choose, I wouldn't add a complex filter (which might even bring bugs) for a 1% final gain, but YMMV.
- fulafel 10y agoThe table says Zstd 20 gets down to 339.4 kB without the filter and 226.7 kB with the filter. This seems like a very good result for the filter.
- giovannibajo1 10y agoAh thanks, for some reason I misread the table. Yes indeed, very good results.
- ajosh 10y agoThis was a fun read for sure. It's always inspiring to see someone take on a new domain, learn something and produce something functional.
- joseraul 10y agoSuch specialized filters can improve compression a lot. For instance, the e8e9 transform converts relative jumps to absolute jumps in x86 EXEs, increasing compression because lots of jumps actually point to the same address. http://mattmahoney.net/dc/dce.html#Section_571 http://mattmahoney.net/dc/dce.html#Section_571
- erichocean 10y agoSomewhat mis-leading graphs since higher-quality zlib compression levels weren't evaluated. Zstd is DOA thanks to Facebook's patent poison pill.