Comment by zenogais

Like the author said this is completely unoptimized. The natural next step in optimization might be to profile and then SIMD optimize the slow bits in compression and decompression. This would likely produce a significant speedup and may even bridge the gap with lz4.