A New Huffman Code for Accelerating GPU Decompression (preliminary version)

Abstract

A Huffman code is a variable-length entropy code used in data compression. The compression operation of a Huffman code can be done very efficiently on the GPU. However, the decompression is very hard to parallelize, because the compressed data have no separator to identify each variable-length codeword. We present a Huffman Code with Gap Array (GHCA), which accelerates the GPU decompression. The experimental results using Geforce RTX2080Ti GPU shows that, the GHCA may have at most 1.5% size overhead, but the decompression is 1.536-4713 times faster than the previously published GPU decoding for Huffman codes.