An Improved BitMask Based Code Compression Algorithm for Embedded Systems

Abstract

Engineers must consider performance, power consumption and cost when they design digital systems. Embedded systems are more constrained in all these considerations. Memory is one of the key factors that affect all of them. Code compression is a technique for embedded systems to reduce the memory usage. Bit Mask based code compression is a modified version of dictionary based code compression. The basic of Bit Mask is to record mismatch values and their positions to compress more instructions and use exclusive or operation with the reference instruction to decode the codeword. In this paper, we applied separated dictionary and variable mask numbers to the Bit Mask algorithm to reduce the codeword length of high frequency instructions. A novel dictionary selection algorithm is also proposed to increase the instruction match rates. According to our experimental results, our method can improve in average 3% compression ratio for smaller benchmarks, and over 6% improvement for bigger benchmarks.