An Evaluation of Autotuning Techniques for the Compiler Optimization Problems

Diversity of today’s architectures have forced programmers and compiler researchers to port their application across many different platforms. Compiler auto-tuning itself plays a major role within that process as it has certain levels of complexities that the standard optimization levels fail to bring the best results due to their average performance output. To address the problem, different optimization techniques has been used for traversing, pruning the huge space, adaptability and portability. In this paper, we evaluate our different autotuning approaches including the use of Design Space Exploration (DSE) techniques and machine learning to further tackle the both problems of selecting and the phase-ordering of the compiler optimizations. It has been experimentally demonstrated that using these techniques have positive effects on the performance metrics of the applications under analysis and can bring up to 60% performance improvement with respect to standard optimization levels (e.g. -O2 and -O3) on the selection problem and up to 4% w.r.t. to LLVM’s standard optimization on the phase-ordering problem.