使用OpenMP+MPI的矩阵乘法并行实现

Parallel Implementation of Matrix Multiplication Using OpenMP + MPI

为了提升矩阵乘法的运算速度,优化运算性能,提出了一种基于并行计算的方法。采用OpenMP+MPI混合编程,选用华为鲲鹏处理器和PC机作为实验平台。其中MPI将分块后的小矩阵广播至每个进程,在每个进程中使用OpenMP进行矩阵相乘的运算,再使用MPI进行运算结果的聚集,最后显示时间性能等相关信息。分别选用PC机,单台和多台华为鲲鹏服务器,分配不同的进程数,对运算结果进行了相关性能的分析与总结。

In order to improve the operation speed of matrix multiplication and optimize the operation performance, a method based on parallel computing is proposed in this paper. OpenMP + MPI hybrid programming is adopted, and Huawei Kunpeng processor and PC are selected as the experimental platform. Among them, MPI broadcasts the partitioned small matrix to each process, using OpenMP to multiply the matrix in each process;then MPI is used to aggregate the operation results, and finally relevant information such as time performance is displayed. PC, single and multiple Huawei Kunpeng servers are selected to allocate different process numbers, the performance of the operation results is analyzed and summarized.