Searching for the Ideal Number of Threads on Asymmetric Multiprocessors

Asymmetric multicore processors (AMP) combine high-performance cores with more energy-efficient ones, capitalizing on the diverse performance demands of modern devices (e.g., smartphones and tablets). Although hardware players have been designing more powerful AMPs for desktop and server computers, such as the Apple M1 and Intel Alder Lake family, these processors impose new challenges on parallel computing researchers on how to properly use them to their fullest. As we show in this paper, the best number of threads and which combination of the kind of cores will very likely change with the applications and user requirements, i.e., performance, energy, or energy-delay product (EDP). Based on that, we propose HyTuning, a heuristic algorithm that improves the above-mentioned requirements (as a user choice) by automatically adjusting the number of threads and selecting the most appropriate cores to execute a given application on an Intel Core i9-12900K Alder Lake processor. Our experimental results considering eleven parallel benchmarks show that HyTuning outperforms the other three compared strategies on average by 1.40x, 2.08x, and 2.05x optimizing for execution time, energy, and EDP, respectively.