GPU Porting of Plasma Physics Codes
DescriptionWe will present the global strategy used for the GPU porting of multiple fusion research codes. The chosen strategy aims to keep a single version of each code using OpenMP/OpenACC directives. Performance portability on different architectures (NVIDIA and AMD GPUs) and with various compilers will be investigated. We will present how algorithms have been improved to get better performance and our selected strategies to improve performance portability. We will focus only on 3 representative codes to illustrate this global strategy: (1) CAS3D, a magnetohydrodynamic code used to study the properties of fusion plasmas in non-axisymmetric configurations such as stellarators. We will show how the introduction of generic pragmas allows the use of either OpenMP or OpenACC to get the best performance according to the available compilers. (2) ASCOT5, an orbit-following Monte Carlo code, dedicated to stellarator configuration. We will present how we have exploited the independent time evolution of each particle and used the multi-level parallelism features of OpenMP/OpenACC to improve load balancing. (3) GYSELA-X, simulating electrostatic plasma turbulence and transport in Tokamak devices. 5D features of GYSELA-X lead to the exploitation OpenMP multi-level parallelism, results on the recent HPE-Cray ADASTRA machine equipped with AMD MI250X GPUs will be presented.
TimeTuesday, June 2711:00 - 11:30 CEST