GPU-accelerated computations of FDTD-compatible Green's function

We present results of an acceleration of the discrete Green's function (DGF) computations on a graphics processing unit (GPU). Recently, closed-form expression for the DGF was derived, which facilitates applications of this function in the finite-difference time-domain (FDTD) simulations. However, the new DGF expression may cause long runtimes and accuracy problems. Therefore, we have proposed the DGF implementation on the GPU employing a multiple precision arithmetic library and the CUDA parallel programming interface. We have obtained six-fold speedup of the GPU implementation over a code executed on a multicore central processing unit. Achieved results indicate that GPUs represent an inexpensive source of computational power for the DGF computations.