Abstract

The era of exascale supercomputing promises great advances in high-performance research computing, from increased simulation resolution and larger ensemble sizes to more realistic and complex models. To make the most of the next generation of HPC resources, however, we must re-engineer our applications to run efficiently on a variety of hardware architectures. Simultaneously the growth of software complexity and the adoption of techniques like machine-learning and autodifferentiation present a challenge to the current paradigm of scientific simulation development.

This talk details our approach to building one solution to both problems: NDSL , the NOAA /NASA Domain Specific Language. NDSL allows weather and climate modelers to write a model in Python, which is then translated into highly-optimized CUDA , C++, or HIP by the DSL compiler to run at scale on flagship supercomputers. Beyond performance gains, NDSL ’s separation of concerns between scientific modelers and performance engineers improves the development experience and productivity for both groups. We will discuss what this means in practice, the challenges of the DSL paradigm, and how to build a programming framework domain scientists are excited to adopt.