A New Architecture for FUN3D on Modern HPC Systems

The goal of this project is to refactor NASA's FUN3D computation fluid dynamic (CFD) simulation code to enable it to take full advantage of accelerator hardware available on modern hybrid computing environments. We will rearchitect FUN3D to take advantage of HPC tools we are currently building for NASA Goddard, using a dynamically-scheduled, task-based approach, with the goals of improving scalability, performance on the CPU, and the amenability to accelerators current and future. In Phase I, the focus will be on proving the validity and applicability of our approach for a given subset of FUN3D code. To meet this goal, we have identified two primary objectives for Phase I: rearchitect FUN3D and develop a prototype demonstrating feasibility. This will decouple the science being performed from the intricacies of implementations on a variety of platforms. In Phase II, we will build off this work to create a full accelerated solver for ongoing use. Once complete, the full solver will be efficient on today's hardware and easily adaptable to future systems.