Experiences with Cross-platform Portability of NWChemEx using SYCL on Exascale Supercomputers
The advent of exascale computing presents significant opportunities for advancing computational chemistry, particularly in the accurate modeling of molecular systems. We explore the integration of a gold standard quantum chemistry algorithm—Coupled Cluster with Single, Double, and Perturbative Triple excitations [CCSD(T)] method developed within the NWChemEx library using SYCL and the oneMKL library to enable portability and performance across diverse exascale hardware architectures.
Our work specifically addresses the challenges of porting, debugging, and performance engineering of algorithms to various vendors—NVIDIA, AMD, and Intel—utilized in Department of Energy (DOE) exascale facilities. Leveraging SYCL and oneAPI’s portability, we achieve efficient parallelization and execution across these diverse GPU architectures, ensuring that the code remains performant and scalable.
Benchmarks conducted on these DOE exascale systems demonstrate portability, efficiency, and scalability.
Speaker(s)
Abhishek Bagusetty - Assistant Computational Scientist, Argonne National Laboratory
Abhishek (Abhi) Bagusetty is an Assistant Computational Scientist at the Leadership Computing Facility (LCF) of Argonne National Laboratory. He holds a PhD in Computational Modeling and Simulation (Chemical Engineering) from the University of Pittsburgh. Later, he joined the ALCF in 2020, and his research interests are high-performance computing, GP-GPU programming models (SYCL, CUDA, OpenCL), and molecular simulation of materials.
