Tuning Compiler Optimization Options via Simulated Annealing

To find a good set of compiler options for a particular CPU and software is actually a difficult task. Some tool based on genetic algorithm like AcovEA exists, and it requires to compile and run each program many times as to find optimized compiler options. If each program execution time is very long, then the process is rather time consuming. In this paper, a new tool, named Analysis of Compiler Options via Simulated Annealing (AcovSA), is introduced. AcovSA has the same functionality as AcovEA, meanwhile it has better control of execution time with Simulated Annealing. The experiment results, by comparing the benchmarks´ running time with the good optimization options set (OOS) obtain from AcovSA and AcovEA, show the efficiency and effectiveness of AcovSA.