Code compression in embedded processors using multi-dictionary and pattern blocks

A novel method using multi-dictionary and pattern blocks is proposed for code compression to effectively reduce the memory size which program code stored. According to the repeat times of different pattern blocks and non-block codes, the code set is clustered into several clusters. Pattern blocks cluster is compressed using dictionary-based compression method and each dictionary entry points to a pattern block. The other clusters are for non-block codes and the codeword for each cluster is different. We use shorter codeword to point to the more frequently occurred cluster. By finding the best clusters for both blocked and non-block codes, the entropy of our encode strategy is high and the compression efficiency is the best. Theoretical proof and experimental results show that the compression effect of this method is superior to the traditional multi-dictionary based code compression. Code of MiBench benchmark compiled under the ARM and MIPS instruction set architecture are compressed with this method, and the average compression ratio are between 50% ~ 55%,64%~70%, separately.