传统的基于过程语言的算法实现很难应用于大规模并行计算，OpenMP和MPI等现有并行框架存在着并行实现困难、开发成本大和灵活度差等诸多问题。通过应用基于函数语言的并行新方法，有效简化并行代码的设计，提升并行算法的开发自由度，并可支持动态分区等复杂并行需求。通过将其应用于具有天然并行属性的FDTD剖分及仿真算法，发现可实现高达50%加速比的高效并行，并在26 h内成功求解高达6.9亿未知量的电大尺寸航空母舰甲板模型电磁全波仿真问题。%The traditional procedure languages are not natively parallelizable, while the present parallelization frameworks such as OpenMP and MPI are difficult to apply and malntaln. In this paper, a new parallel imple⁃mentation method is proposed using the functional language, which rapidly reduces the developing cost, thus enhancing the freedom of parallel algorithm implementation, and could satisfy advanced parallel require⁃ments such as dynamic partitioning, etc. For validation, the proposed new algorithm is applied on the FDTD meshing and simulation algorithm successfully, where an acceleration ratio up to 50% is achieved, and an electro-large motherboard problem with over 690 million unknown numbers is successfully solved in 26 hours.
展开▼
