Abstract:

Iterative compilation is a compiler optimization technique and it has been proved that it can significantly improve program performance. However, iterative compilation needs to compile and run a program multiple times, so the process is time-consuming. In order to make full use of the existing multi-core computing resources and speed up the iterative process, we parallelize the evaluation phase of iterative compilation based on a new iterative framework named “OpenTuner”. Meanwhile, we explore the impact of the parallelization of evaluation phase in iterative compilation on the optimization results, and analyze the experimental results.

Key words: compiler optimization, iterative compilation, parallelization, program tuning