Research On Code Compression Technique For Variable Length Instruction Encodings

With the development of embedded technology, embedded system will be applied more widely in every aspects of human life. In the field of embedded system, software development faces new challenges and how to shorten the codes and improve the efficiency has become problems of concern in recent years, as being an active research field. Thus, code compression technology will be one of the key technologies for future embedded development.First, we analysis the features of embedded system and the embedded software in order to raise the compression problem. Based on the comparison and analysis between data compression and code compression, we summarize some criterions for code compression. Then we discuss the Challenges posed by the variable length machines. After this, we present a compression scheme based on dictionary which utilizes unused encoding space of an instruction set architecture to encode code-words, so as to solve the problem due to variable length instructions. Also the pipeline architecture for decompression and Branch Address Table(BAT) have also be presented at large.In order to verify the scheme, we use a cycle accurate simulator for the CR16C pipeline and test our scheme on 4 benchmarks in MiBench and one open source application software named gzip, and gain 85.4% compression ratio. We restrict the size of the BAT to 512, 1024 and 2048 to calculate the time overhead incurred for searching the BAT. Finally we make research on the result of compression experiment and summarize it.