Recently, driven by the need of compressing the blueprint files from DysonSphereProject, I opened the source of paq8l and compiled it again.

I remember that, there could be an error which cause paq8l failed to compress with option -9, and I remember that this simple issue could be solved by modified 2 lines of the original code

$ diff paq8l.cpp paq
3096c3096
<   const int BLOCK=MEM*64;
---
>   const unsigned int BLOCK=MEM*64;
3098c3098
<     int size=BLOCK;
---
>     unsigned int size=BLOCK;

actually, only the latter change is enough, since the root cause is not here.

when compress level is set to 9, const int BLOCK=MEM*64; becomes const int BLOCK=(0x10000<<9)*64;, the RHS, (0x10000<<9)*64, is int, thus -2147483647 is the result, which could not pass the coming check.

for a modern computer, using such modification could be enough for -9.

What’s more, I modified the original code to add avx-2 and avx-512 implementations, to compile that file, use

$ # Linux:
$ g++ paq8l-avx*.cpp -O3 -o paq8l-avx512 -s -fomit-frame-pointer -DNOASM -march=native -DUNIX
> # Windows:
> g++ paq8l-avx*.cpp -O3 -o paq8l-avx512 -s -fomit-frame-pointer -DNOASM -march=native -DWINDOWS

Have fun:)