Accelerating the shooting and bouncing ray based electromagnetic scattering calculations via CPU and GPU Implementations: Application to PREDICS tool

Abstract

In this paper, we present the parallel computation of physical electromagnetic (EM) solvers, namely physical optics (PO), shooting and bouncing rays (SBR), and the physical theory of diffraction (PTD), which are used for EM scattering/diffraction calculations. The accelerated solvers have been tested on CPU and GPU kernels for significantly faster calculation times. As an application, the developed implementation is employed to recently developed tool of PREDICS that uses combined techniques PO, SBR, and PTD for the calculation of EM scattering/diffraction and radar cross section (RCS) from electrically large, complex objects. First, the CPU parallelization of these three combined techniques on the PREDICS code is explained. The parallelization implementation is specially tailored for the nature of SBR technique such that the CPU iteration loop is accomplished over triangular facets of the CAD file by employing a read/write spin lock task dispatching technique. Next, a lock-free implementation of the PO, SBR, and PTD techniques on the GPU threads is presented. Thanks to availability of many GPU cores, the parallelization procedure is widened such that triangular facets, observation angles of elevation and azimuth are used for a much faster calculation set-up. Numerical examples for the EM/RCS calculation of simple canonical and electrically big, complex objects are shared to demonstrate the effectiveness, accuracy and the computational acceleration based on the multi-CPU and multi-GPU implementations. The key metric of the contribution of the study is the speed-up value of the implementations. The acceleration gained via multi-CPU realization yielded directly in parallel with the number of available CPU cores. On the other hand, the acceleration factor can reach to hundreds when multi-GPUs are used. In fact, the acceleration factor has reached to 4.7 for a machine with 4 Core 8 Thread CPU processor, and 141.3 for a 2560 CUDA GPU Core system.