A Collaborative Filtering Approach for the Automatic Tuning of Compiler Optimisations
Authors:
Stefano Cereda,
Gianluca Palermo,
Paolo Cremonesi,
Stefano Doni
Abstract:
Selecting the right compiler optimisations has a severe impact on programs' performance. Still, the available optimisations keep increasing, and their effect depends on the specific program, making the task human intractable. Researchers proposed several techniques to search in the space of compiler optimisations. Some approaches focus on finding better search algorithms, while others try to speed…
▽ More
Selecting the right compiler optimisations has a severe impact on programs' performance. Still, the available optimisations keep increasing, and their effect depends on the specific program, making the task human intractable. Researchers proposed several techniques to search in the space of compiler optimisations. Some approaches focus on finding better search algorithms, while others try to speed up the search by leveraging previously collected knowledge. The possibility to effectively reuse previous compilation results inspired us toward the investigation of techniques derived from the Recommender Systems field. The proposed approach exploits previously collected knowledge and improves its characterisation over time. Differently from current state-of-the-art solutions, our approach is not based on performance counters but relies on Reaction Matching, an algorithm able to characterise programs looking at how they react to different optimisation sets. The proposed approach has been validated using two widely used benchmark suites, cBench and PolyBench, including 54 different programs. Our solution, on average, extracted 90% of the available performance improvement 10 iterations before current state-of-the-art solutions, which corresponds to 40% fewer compilations and performance tests to perform.
△ Less
Submitted 11 May, 2020; v1 submitted 6 May, 2020;
originally announced May 2020.
