• Sources: primary, discussion
  • Summary: The native LZ engine ships starting in v0.2.0, and the published benchmark tables were run with v0.3.0, reporting decompression over 2x faster than Zstandard across the compression level range and up to 50% faster than LZ4. Freedom from the Zstandard wire format lets offsets be bucketed by a V-optimal histogram and bitpacked with no entropy codec, decoding 2x to 3.5x faster. PivCo Huffman, a new Huffman layout by Marcin Zukowski, replaces the codec that accounted for over half of LZ decompression CPU and is enabled by default in v0.3.0. The stated tradeoff is that OpenZL compresses slightly worse than Zstandard at higher levels, because it does not yet support repeat offsets.
  • Why it matters: The gains come from two independent sources, an unconstrained wire format and a new Huffman layout, so only the first is limited to pipelines that control both ends.