主催: 一般社団法人 日本機械学会
会議名: 第36回 計算力学講演会
開催日: 2023/10/25 - 2023/10/27
In this paper we report an efficient implementation method of the preconditioners for the full-wave electromagnetic finite element method using general purpose calculation libraries. Preconditioning methods suitable for parallel processing like an extended node patch preconditioner, which we have reported, require suitable implementation such that independent subsets are effectively selected and parallelized. The extended node patch preconditioner is sufficiently efficient when properly parallelized under sophisticated partitioning and processing scheduling. To optimize partitioning subsets into independent groups allotted to CPU cores and parallelization schedule we propose to use general purpose numerical analysis module which plays optimal assignment of each stencil calculation and ordering. We implemented this approach and verified the effectiveness.