SPIR-V Compression(aras-p.info)
aras-p.info
SPIR-V Compression
http://aras-p.info/blog/2016/09/01/SPIR-V-Compression/
6 comments
Interesting 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.
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.
The 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.
Ah thanks, for some reason I misread the table. Yes indeed, very good results.
Somewhat mis-leading graphs since higher-quality zlib compression levels weren't evaluated. Zstd is DOA thanks to Facebook's patent poison pill.
This was a fun read for sure. It's always inspiring to see someone take on a new domain, learn something and produce something functional.
http://mattmahoney.net/dc/dce.html#Section_571