https://hgpu.org/?p=12980
CUVLE: Variable-Length Encoding on CUDA