Improved sv8 huffman decoding using canonical property of the huffman tables. It reduces memory usage for the tables and improves speed (mainly with small LUT size).