Bulat-Ziganshin
in PAR2 experiments (see ECC_speed_comparison.zip) non-SIMD implementation is only 4x slower (405.0 vs 1559.2 sec). Probably you are used non-optimal solution. You may look into Plank/Par2 sources. My understanding is...
this looks like about 2 cpu cycles/byte, 1 cpb and 0.5 cpb, respectively first approach require 1 memory read per operation. According to [Intel optimization manual](http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf), Skylake L2$ can deliver...
Yeah, it's our internal machinery which is absolutely unintuitive for library users. I think that this issue may be considered as call for doc improvement. You can say that to...
http://sachingarg.com/compression/entropy_coding/64bit/sg_entropy_coders.zip looks like pretty standard codecs, at least for range coder part. more info at http://sachingarg.com/compression/entropy_coding/64bit/
One more idea: we can add pragmas to loops asking compiler to vectorize them. OpenMP pragmas in particular. Compilers lose opportunities to vectorize code mainly because 1) computation is too...
Насколько я понимаю, это начало спуска по кроличьей норе graph databases :))
I'm going to finally implement that: https://encode.ru/threads/2552-Lepton-image-compression?p=58615&viewfull=1#post58615
The work on shared library announced [here](https://github.com/dropbox/lepton/issues/35#issuecomment-434044234) would make creating 7-zip, FreeArc and similar plugins as easy as lzham/zstd plugins.
I live in Moscow and work on lossless data compression. can you contact me at [email protected] ? Look at https://encode.su/threads/3378-Learning-compression , it has a lot of links
> https://github.com/klauspost/reedsolomon ? it's 8-bit RS coder, you should compare speeds to ISA-L and CM256. Probably they are the same since arithmetic is well-established