基于域划分的嵌入式处理器代码压缩方法

Field-partition based code compression approach for embedded processors

为了提高基于域划分的代码压缩方法的压缩率，在传统的位置模型的基础上，挖掘了不同类型指令内部符号间的相关性，提出了一种新的相关模型——类型模型．综合位置模型和类型模型，进一步得到另一种相关模型——混合模型．提出了一种使用混合模型，采用霍夫曼编码的代码压缩方法以及相应的低开销硬件解压缩实现方案．以自主研发的16位嵌入式数字信号处理器（DSP）SPOCK指令集为例，对OggVorbis测试压缩性能．实验结果表明，使用混合模型在少量硬件代价下对SPOCK指令集的压缩率达到53．16％，比传统的位置模型方法提高了12％～14％．

In order to improve the compression ratio of field-partition based code compression, a novel correlation model -- type model was proposed, which captures the correlations between the symbols of different instruction type. A code compression approach combining the type model and the traditional position model and a corresponding low-overhead decompression hardware architecture using Huffman coding were also presented. Applied to a self-developed 16 bit embedded digital signal processor （DSP） -- SPOCK with OggVorbis as benchmark program, this approach achieved the best compression ratio of 53. 16%, which was 12%-14% better than that of the traditional methods with small hardware cost.