Massively Parallel Approach of Multilevel Fast Multipole Algorithm on DCU Clusters for Large Electromagnetic Scattering problems

A massively parallel approach of the multilevel fast multipole algorithm (MLFMA) on Deep Computing Unit (DCU) clusters using the Heterogeneous Interface for Portability (HIP) programming model and the Message Passing Interface (MPI) is presented (DCU-MLFMA). The proposed MPI-based multi-DCU parallel implementation is capable of solving challenging problems involving billions of unknowns with a significant speedup, compared with the conventional CPU-based ternary parallelization approach of MLFMA (CPU-MLFMA). Numerical results, including a ship model with length about 6000 wavelengths modeled with 2.6 billion unknowns, are given to demonstrate the capability, as well as the efficiency of the presented method.