OpenMP GPU Acceleration and Portability of TRIMEG-C1 for Electromagnetic Gyrokinetic Simulations in Tokamak Plasmas

2026-06-23

AI Summary

This work addresses the challenges of computational efficiency and cross-GPU-platform portability in electromagnetic gyrokinetic simulations of tokamak plasmas by implementing, for the first time, a high-order C1 finite element method within the TRIMEG-C1 code using the OpenMP GPU offloading framework. The approach achieves efficient and portable acceleration on both NVIDIA and AMD GPUs through algorithmic optimizations in particle pushing and particle–mesh coupling, combined with an MPI–OpenMP hybrid parallelization strategy. On two AMD MI300A APUs, the particle pusher kernel demonstrates approximately 9× speedup over two AMD EPYC 9754 CPUs. Furthermore, the code successfully simulates ion temperature gradient (ITG) modes, accurately reproducing both the linear growth rates and two-dimensional mode structures, thereby validating the physical fidelity and cross-platform effectiveness of the proposed methodology.

Abstract

The Triangular mesh-based gyrokinetic code TRIMEG-C1 solves the gyrokinetic equations using the particle-in-cell scheme to simulate electromagnetic instabilities in tokamak plasmas. TRIMEG-C1 utilizes a high-order C1 finite element method, which captures the accurate physics with lower grid resolution than the C0 method. In this work, we focus on achieving a portable implementation on multiple graphics processing unit (GPU) architectures to accelerate the TRIMEG-C1 code for future physics studies. The OpenMP framework is chosen as the acceleration framework for GPU offloading on different hardware platforms, specifically, NVIDIA and AMD GPUs. The particle pushing procedure, as well as particle-to-grid operations have been adapted for GPU execution. A speedup of $\approx9$ for the particle pusher kernel is achieved on 2 AMD MI300A APUs (Accelerated Processing Unit) compared with 2 AMD 9754 CPUs. In addition, the efficiency of hybrid MPI-OpenMP offloading parallelization was assessed by oversubscribing GPU resources. The Ion Temperature Gradient (ITG) mode was simulated using the GPU implementation, and its correctness was verified by comparing the physics results in terms of the energy growth rate and the two-dimensional mode structures.

Problem

• GPU acceleration
• portability
• gyrokinetic simulation
• OpenMP
• tokamak plasmas

Innovation

• OpenMP GPU offloading
• C1 finite element method
• gyrokinetic simulation
• portable GPU acceleration
• particle-in-cell