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

In this article, 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 billions unknowns, approach, flexible and efficient ternary partitioning scheme employed at first to partition MLFMA octree among message-passing interface (MPI) processes. Then, computationally intensive parts PMLFMA each MPI process, matrix filling, aggregation disaggregation, so are accelerated by using GPU. Different parallelization strategies in coincidence with designed GPU ensure high computational throughput. Special memory usage strategy improve efficiency benefit data reusing. The CPU/GPU asynchronous computing pattern OpenMP compute unified device architecture (CUDA), respectively, accelerating CPU execution computation time overlapped. architecture-based optimization implemented further efficiency. Numerical results demonstrate that proposed GPU-PMLFMA can achieve over three times speedup, compared eight-threaded conventional PMLFMA. Solutions electrically complicated objects about 24 000 wavelengths 41.8 billion unknowns presented.