TechRxiv
2022-TAP[J]-WeijiaHe-GPU-PMLFMA.pdf (1.46 MB)
Download file

Solving Electromagnetic Scattering Problems with Tens of Billions of Unknowns Using GPU Accelerated Massively Parallel MLFMA

Download (1.46 MB)
preprint
posted on 26.05.2022, 17:35 authored by Wei-Jia HeWei-Jia He, Zeng Yang, Xiao-Wei Huang, Wu Wang, Minglin Yang, Xin-Qing Sheng
In this paper, a massively parallel approach of the multilevel fast multipole algorithm (PMLFMA) on graphics processing unit (GPU) heterogeneous platform, noted as GPU-PMLFMA, is presented for solving extremely large electromagnetic scattering problems involving tens of billions of unknowns, In this approach, the flexible and efficient ternary partitioning scheme is employed at first to partition the MLFMA octree among message passing interface (MPI) processes. Then the computationally intensive parts of the PMLFMA on each MPI process, matrix filling, aggregation and disaggregation, etc., are accelerated by using the GPU. Different parallelization strategies in coincidence with the ternary parallel MLFMA approach are designed for GPU to ensures a high computational throughput. Special memory usage strategy is designed to improve the computational efficiency and benefit data re-using. The CPU/GPU asynchronous computing pattern is designed with the OpenMP and CUDA respectively for accelerating the CPU and GPU execution parts and computation time overlapped. GPU architecture-based optimization strategies are implemented to further improve the computational efficiency. Numerical results demonstrate that the proposed GPU-PMLFMA can achieve over 3 times speed-up, compared with the 8-threaded conventional PMLFMA. Solutions of scattering by electrically large and complicated objects with about 24000 wavelengths and over 41.8 billion unknowns, are presented.

Funding

NSFC under Grant No. 61971034

National Key R&D Program of China under Grant 2017YFB0202500

History

Email Address of Submitting Author

heweijia@bit.edu.cn

ORCID of Submitting Author

0000-0002-4614-6545

Submitting Author's Institution

Beijing Institute of Technology

Submitting Author's Country

China