Hybrid Evolutionary Algorithm Based Solution for Register Allocation for Embedded Systems

Abstract

Embedded systems have an ever-increasing need for optimizing compilers to produce high quality codes with a limited general purpose register set. Either memory or registers are used to store the results of computation of a program. As compared to memory, accessing a register is much faster, but they are scarce resources and have to be utilized very efficiently. The optimization goal is to hold as many live variables as possible in registers in order to avoid expensive memory accesses. We present a hybrid evolutionary algorithm for graph coloring register allocation problem based on a new crossover operator called crossover by conflict-free sets(CCS) and a new local search function.