Exploring Parameter-Type Compiler Options for Accelerating Program Execution

This paper proposes an extensive use of parameter-type compiler optimization options to speed up program execution. Compilers usually provide a lot of options to control their optimizers. However, existing methods of exploring the best option sets for given programs dealt only with switch-type options which specifies activation/deactivation of individual optimizers. This paper extends the search domain to parameter-type options which control optimizers by numerical values. A genetic algorithm is employed to explore the best combination of the parameters as well as the optimization switches for given programs. An experiment on GCC 4.8.1 with four benchmark programs shows that the proposed method accelerates the program execution by 10.4%, at the cost of 38.0% increase in the exploration time.