Instruction Set Encoding Optimization for Code Size Reduction

Abstract

In an embedded system, the cost of storing a program on-chip can be as high as the cost of the microprocessor itself. We examine how much a given application's program size can be reduced when an instruction set is tailored to the application. We provide different algorithms for calculating an optimized instruction set and evaluate their impact on the size of several benchmark programs. Our results show that an average reduction of 11% is possible, and further improvement can be achieved by changing the instruction length of the given architecture. However compiling other applications with such an optimized instruction set might produce larger code sizes.