Load-Balanced Fortran-Based Out-of-GPU Memory Implementation of the Method of Moments

In this letter, an efficient Fortran-based out-ofgraphics processing unit (GPU) memory implementation of the method of moments (MoM) on heterogeneous low-cost single CPU/GPU computing platform is presented. The MoM solution process is based on the frequency-domain integral equation formulation and employs a wire-grid approach to evaluate the radiation pattern of the glide path antenna system. To handle processor heterogeneity and balance the workload between the host and the device, a novel task scheduling strategy, both for the matrix assembly phase and the solution of the MoM matrix equation, is proposed. Using the CUDA-capable GTX 680 GPU from NVIDIA, a speedup ratio of about 1.44× is reported when compared to the classical hybrid CPU/GPU approach, and about 2.21× when compared to the multicore CPU implementation.