BareRose/nibrans

[JamesB]

I haven't tried it, but stumbled across this this morning and thought I'd share:

https://github.com/BareRose/nibrans

It's an adaptive nibble-bases rANS entropy encoder. It looks to have interfaces sufficient to simply accept whole buffers, but also via eg https://github.com/BareRose/nibrans/...nibrans.h#L214 shows how to call lower level primitives nbraEncPut. It appears to be based on Fabian's / Charles' blog posts on how to do 16-bit CDF updates for adaptive rANS.

It claims around 75-95MB/s enc/dec performance.

More here: https://www.reddit.com/r/gamedev/com...ased_adaptive/

[dnd]

The compression ratio is too weak for the entropy coder benchmark.
Each nibble is encoded separately. Normally, for an alphabet size of 256 the high nibble is used as context in the low nibble encoding.
In this case the speed can't be compared.

      C Size  ratio%   Name            File              
   621450120    62.1   TurboANXN o0    enwik9
   744770627    74.5   nibrans         enwik9

      C Size  ratio%   Name            File             
   194794442    19.5   TurboANXN o0    enwik9bwt
   236249093    23.6   nibrans         enwik9bwt

[JamesB]

Thanks for the testing. That sounds like a glaring failure in the code, although presumably it's trivial to fix this and use the context correctly.