Hybrid OpenCL-MPI parallelization of the FDTD method

This paper presents preliminary evaluation of the hybrid parallelization of the Finite-Difference Time-Domain (FDTD) method based on Open Computing Language (OpenCL) and the Message Passing Interface (MPI). Due to the portability of OpenCL, developed code targets not only distributed shared memory computer clusters based on multi-core central processing units (CPUs), but also clusters accelerated by graphics processing units (GPUs). The computational domain is decomposed along the slowest direction, and electromagnetic field boundary data is shared between neighboring subdomains using either OpenCL or MPI communication. The communication overhead between GPUs is proportional to the area of the boundary and represents the rate-limiting step of the method. This paper subsequently shows results of numerical tests aimed at evaluation of the hybrid OpenCL-MPI FDTD solver in electromagnetic simulations.