PivCo-Huffman: Pivoted Coding for High-Speed Huffman Decoding with Wavelet-Tree Inspired Data Layout
Summary
This paper introduces PivCo-Huffman, a pivot-coded Huffman encoding approach inspired by wavelet trees to enable SIMD-friendly, high-throughput decoding. It details both naive top-down and bottom-up decoding variants, a suite of SIMD-optimized primitives, and selective ANS/FSE integration to boost compression where beneficial. Results show decoding speeds rivaling or surpassing state-of-the-art Huffman codecs on skewed datasets, with dataset-dependent gains and notable encoding considerations.