BWT post coder

Question

BWT post coder

encode84 opened this issue 2 years ago · comments

Hello,
The model in BWT post coder (cm.c) is the exact copy from the BCM compressor by Ilya Muravyov, with some code restructure. Yet no reference, even not in the Thanks section. Any ideas?

Kamila Szewczyk · Answer 1 · Mon May 16 2022 22:52:46 GMT+0800 (China Standard Time)

Hi! cm.c is a modified version of https://github.com/r-lyeh/stdpack.c/blob/master/src/bcm.c (which itself seems to use Matt Mahoney's arithmetic coder/existing fpaq0 code?) and as the comment header states, is put in the public domain (a.k.a. no attribution required). That said, if you insist, I'll add a mention later. Thanks.

Ilya Muravyov · Answer 2 · Tue May 17 2022 17:44:29 GMT+0800 (China Standard Time)

FYI, the original BCM is here:
https://github.com/FS-make-simple/bcm

Kamila Szewczyk · Answer 3 · Tue May 17 2022 17:52:59 GMT+0800 (China Standard Time)

I see. I've been meaning to replace this particular postcoder with some rank coding model, which could be quite a bit faster. I also tried a lot of tweaks to this postcoder (porting over Kanzi modifications, etc...) - none of them were particularly successful. From my observations, decreasing run threshold (swap out run > 2 to run > 1) seems to give better results on some files, just like ORing the run flag as the 9th bit into the context for C2 accesses.

I have already identified the entropy coder part as a modified version of fpaq0's one. I have also experimented with rANS/FSE, but I never got far enough with it.