GPU Parallelization of Wave Equation Based Discontinuous Galerkin Time Domain Method

Abstract

In this paper, a GPU-accelerated improved discontinuous Galerkin time domain scheme based on Helmholtz vector wave equation (GPU DGTD-WE) has been proposed. By mapping FEM meshes to CUDA multi-layer parallel architectures and implementing some optimization operations, good acceleration performance can be obtained. By using matrix projection between the matrices in the global coordinates to universal matrices in local ones, high memory compression rate can be achieved for highly efficient solution of large-scale EM problems. Numerical examples including a substrate integrated waveguide filter and a missile-loaded microstrip patch antenna array are given to demonstrate the good performance of the proposed method.