CUDA kernel source code

[amosyou]

hi! I noticed that PTX is loaded for the decompression kernel. I was curious on why PTX is used here instead of creating CUDA bindings. Will the CUDA kernel source code will be released as well?

[tonyzhang617]

We use PTX because it offers better efficiency (with -O3 optimizations) compared to raw CUDA kernels compiled through CuPy. We do plan to open source the kernel soon. Stay tuned!

[amosyou]

awesome thanks!