Performance Portability for Fortran CFD Software with GALÆXI

With GPU acceleration now ubiquitous in HPC, there is more pressure than ever to port existing simulation software to leverage the new hardware. There are many GPU programming models available, but they often have limited, or completely lack, support for Fortran. This leads owners of existing Fortran scientific codes to struggle with porting their software to multiple computing architectures. One option is to turn to C++ to achieve the greatest performance portability.
This webinar is a guide for the process of taking a Fortran CFD code and porting it to GPUs using CUDA/HIP C++. The basics of GPU hardware and programming will be covered, to serve as a rough guide for adapting algorithms and data structures. The high-order discontinuous Galerkin spectral element CFD code GALÆXI is then presented. The important design questions and porting requirements in GALÆXI are discussed, which researchers must answer and understand to effectively port their own Fortran codes with CUDA/HIP C++.