A Multi-GPU Accelerated Parallel Domain Decomposition One-Step Leapfrog ADI-FDTD
Shuo Liu, Bin Zou, Haoyang Zhang, Shulei Ren
Abstract
In this letter, a multi-GPU accelerated one-step leapfrog alternative-direction-implicit finite-difference time-domain (ADI-FDTD) based on parallel SPIKE tridiagonal systems solver is presented. Through our algorithm simplification and execution optimization, online computation and data exchange between GPUs of the SPIKE solver is dramatically reduced. In the simulation experiment, the bistatic RCS of the benchmark model calculated by the proposed method is consistent with the solutions given by one-step leapfrog ADI-FDTD based on Thomas tridiagonal systems solver and Yee's FDTD. Compared with the single-GPU-based implementation, the multi-GPU accelerated one-step leapfrog ADI-FDTD solver can achieve 3.94× speedups at most when executed at a workstation with 4 GPUs.