For large-scaled and high precision electromagnetic field simulations with the FDTD (Finite-Difference Time-Domain) method, it is essential to utilize effectively the vector and parallel computation structure of computers. In the FDTD method, Berenger's PML (Perfectly Matched Layer) absorbing boundaries have a quite large weight, which may affect directly the performance of the whole FDTD program. In this paper, a basic consideration of the PML programming which is adequate to the vector and parallel computation was given, and the consequent programming was shown as an example. From the evaluation results on a super-computer (NEC SX7), it was found that the speed was raised 4~5 times for 8 parallel computations and 8 times for 16 parallel computations, respectively. Compared to the case without parallelism, a speed-up of 26 times was achieved for 16 parallel computations.
展开▼