oneAPI Open-Source Math Library Interface

Abstract

To HPC and AI analytics engineers, math primitives such as basic linear algebra subprograms or random number generators are key functionality that have highly optimized implementations for different CPUs, GPUs, and other accelerators. However, developers must deal with different programming models and interfaces provided by various hardware vendors due to a lack of industry-standard interfaces for math primitives. This paper introduces the oneMKL open-source interfaces project, a SYCL-based math primitives library, as a viable approach for bridging the cross-platform performance portability gap for math primitives across various HPC architectures. By exploiting SYCL interoperability, this project enables integration of optimized vendor-dependent libraries to maximize code reusability without compromising the performance. The cross-platform performance portability of the project is carried out on two major HPC hardware platforms, including Intel CPU, NVIDIA GPU, and also an integrated Intel GPU. Our results show competitive performance with native optimized vendor-dependent libraries.