Machine Learning Based Auto-tuning for Enhanced OpenCL Performance Portability

摘要

Heterogeneous computing, which combines devices with different architectures,is rising in popularity, and promises increased performance combined withreduced energy consumption. OpenCL has been proposed as a standard forprograming such systems, and offers functional portability. It does, however,suffer from poor performance portability, code tuned for one device must bere-tuned to achieve good performance on another device. In this paper, we usemachine learning-based auto-tuning to address this problem. Benchmarks are runon a random subset of the entire tuning parameter configuration space, and theresults are used to build an artificial neural network based model. The modelcan then be used to find interesting parts of the parameter space for furthersearch. We evaluate our method with different benchmarks, on several devices,including an Intel i7 3770 CPU, an Nvidia K40 GPU and an AMD Radeon HD 7970GPU. Our model achieves a mean relative error as low as 6.1%, and is able tofind configurations as little as 1.3% worse than the global minimum.