A top popular approach for the automatic tuning of compiler optimizations

Due to the large number of optimization sequences available in the modern compilers, creating large search areas, a comprehensive exploration of all possible sequences of passes is not possible practically. Therefore, to realize optimal performance, the process of selecting a sequence of passes is a very difficult problem. In addition, finding the best optimization sequence arrangement can improve performance even for a simple program is not an easy task. However, it will additional another problem called phase order issue. In this paper, the proposed approach provides the potential to use techniques inspired by the field of Recommendation System (RS) as a solution to the compiler Autotuning problem. Our method is based on generating random optimization sequences and applying them to a group of different programs, and thus, recommending the ones that give the best execution time. Then, applying the unseen programs to the obtained optimization sequence according to the order of preferences predicted by the Top Popular (TP) algorithm. The results show that this method exceeds the standard -O3 level of the LLVM compiler in improving execution time using TP algorithm. The approach suggested using the standard evaluation of the three benchmark suites PolyBench, Shootout, and Stanford, including 50 different programs and using LLVM (Low Level Virtual Machine) compiler under Linux Ubuntu. The results obtained show that this method performs better than standard optimization level -O3 of LLVM compiler in improving the execution time by an average of 8 % for TP without rate, 20% for TP with rate for senario1 and finally 35% for TP with rate for scenario2.