Workshop: High Performance Python for Science at Scale
Authors: Cody Balos, David Gardner, and Steven Roberts (Lawrence Livermore National Laboratory); Daniel Reynolds (University of Maryland, Baltimore County); and Carol Woodward (Lawrence Livermore National Laboratory)
Abstract: We discuss sundials4py, a Python interface to the SUNDIALS suite of high-performance numerical time integrators, on AMD and NVIDIA GPU-based supercomputers. We show how sundials4py provides portable access to SUNDIALS’ accelerator-enabled solvers and zero-copy interoperability with numpy, PyTorch, and JAX. We also demonstrate how nanobind and litgen are used to generate much of the Python interface binding code, reducing the effort required to expose a large, evolving HPC library while integrating it naturally with the Python GPU ecosystem. Prepared by LLNL under Contract DE-AC52-07NA27344. LLNL-ABS-2023376.
Back to High Performance Python for Science at Scale Archive Listing
Back to Full Workshop Archive Listing