Hybrid Programming Implementation of BiCG Method on SMP Cluster Architecture

Nowadays, the rapid development of the SMP clusters provide a meaningful solution to enormously improve the ability for solving large scale problems in which MPI and Open MP mixed programming mode is used. In this paper, we present a hybrid programming implementation in Multilevel Fast Multipole Algorithm (MLFMA) to compute electro-magnetic scattering from extremely large targets. Message passing communication is employed by MPI to achieve coarse-grained parallelism between nodes, while shared memory communication is implemented by Open MP within node to achieve fine-grained parallelism. It has been proved by experiments that this hybrid programming method could reduce not only the running time of the parallel MLFMA, but also the occupancy of memory space.