On the Efficiency of MW-FDTD Methods Based on Parallel Computing Using OpenMP, OpenACC, and CUDA Python: Application to Lightning Electromagnetic Fields

In this paper, a Three-Dimensional (3D) Moving Window Finite-Difference Time-Domain (MW-FDTD) method is presented based on parallel computing to calculate lightning electromagnetic fields over large-scale terrains. According to the results, the proposed method requires only about × = % of the memory required for the calculation compared to a conventional FDTD method, where and are, respectively, the size of the entire domain in the traditional FDTD method, and the size of the divided blocks in the MW-FDTD method domain. Three different parallel approaches are used in this paper: 1) OpenMP (Open Multiprocessing) CPU implementation, 2) OpenACC (Open Accelerators) GPU implementation, and 3) CUDA (Compute Unified Device Architecture) Python GPU implementation. The efficiency of the utilized programming models is validated by comparing and verifying the obtained results using a serial CPU implementation. The speed-up factors achieved by OpenMP, OpenACC, and CUDA Python programming models, compared to single-threaded CPU series implementations, are respectively 21, 90, and 12.