Proposal of Incremental Software Simulation for Reduction of Evaluation Time

Abstract

Software optimization techniques are necessary to fully utilize modern high-performance computer architectures. In development of the optimization techniques, repetitive simulations are needed to evaluate multiple candidates of optimization methods on partial code of program. Simulation of entire program codes generally takes huge amount of time. Therefore, a problem occurs that repeated simulations make the evaluation time enormously long. In this paper, we propose an incremental software simulation, a method for reduction of evaluation time for the development of software optimization technique by using check pointing and code substitution. Check pointing technique allows us to save and restore simulation process image. We can limit execution of the simulation only to necessary parts of the target programs by using code substitution. Our method can reduce the evaluation time by check pointing the simulation process and by resuming the execution from the checkpoint after modification of the check pointed process image. Evaluation results show that the total simulation time is reduced 34 percents on average in SPEC CPU2000 benchmark programs.