一个通用的并行函数库

摘要

The design of parallel programs differs from sequential programs essentially in its conception and approach, and is quite difficult to develop. Therefore in the paper a universal parallel function library based on MPI was designed to reduce the difficulty of the development of parallel programs. The library encapsulates common functions used in the development of parallel programs, and has certain universality. The library can significantly simplify the development of MPI parallel program. In the paper the library was introduced from the aspects of domain decomposition,transform of array indices and loop variables, data exchange and global reduction, and data I/O. An example that solves Laplace's equation on a two-dimensional surface with fixed boundaries was introduced to show the way and the steps of translating sequential programs into parallel programs using the parallel library. The parallel library has been successfully applied in the parallelisation of a storm surge numerical model with fairly good effect.%并行程序的设计在思路和方法上与串行程序有着本质的差别,其开发难度也比较大.为此,设计开发了一个通用的函数库来降低并行程序的开发难度.该函数库以MPI为基础,对并行程序开发中的常用功能进行了封装,具有一定的通用性.该函数库可以大大简化MPI并行程序的开发.并从区域分解、数组下标和循环变量的转换、数据交换和全局规约、输入输出等几个方面对该函数库进行了介绍,然后以数值求解二维拉普拉斯方程为例介绍了使用该函数库将串行程序改为并行程序的方法和步骤.目前已经成功地将其应用到风暴潮数值模式的并行化中,取得了较好的效果.