Hybrid-parallel Algorithms for 2D Green's Functions

Green's functions are used to solve non homogeneous integral equations with boundary conditions. In electromagnetism they are used for a large class of problems, including the analysis of microwave circuits, antennas, and the investigation of breakdown phenomena in microwave transmission lines. In many cases it is necessary to compute a large number of Green's functions between pairs of source and observation points. Today, basic computational systems are composed of multiple CPUs (multicores) and GPUs, and it is necessary to adapt the software used for the solution of scientific problems to these systems. This paper studies the optimization of the computation of 2D rectangular waveguide Green's functions in clusters of multicores with GPUs. Different basic algorithms are considered, and the most efficient implementations are obtained by adequately combining basic algorithms in hybrid parallelism programs.