作　　者： (王涛）; (赵映诚）; (刘鑫源）;

机构地区： 中国石油大学（华东）理学院,山东青岛266580

出　　处： 《计算机时代》 2017年第9期33-36,共4页

摘　　要： 在解决许多实际问题时,经常需要计算一些高阶矩阵。然而传统的串行计算方法往往效率比较低。因此,需将串行程序并行化来提高计算效率。文章分别研究了Windows API、Open MP、MPI、PPL这四种并行计算方法在矩阵乘法并行化中的应用。通过测试不同规模的矩阵,根据加速比衡量并行化的加速效果,对这四种并行化方法的加速效果进行了对比。结果表明,这四种方法都可以提高计算效率,其中MPI的加速效果最好。 In solving many practical problems, some high-order matrices often need to be calculated. However, the traditional serial computing methods are often inefficient. Therefore, serial programs need to be parallelized to improve computational efficiency. In this paper, four parallel computing methods, Windows, API, OpenMP, MPI and PPL, are studied on their application in matrix multiplication. By testing the matrices of different size, and measuring the acceleration effect of parallelization according to the acceleration ratio, the acceleration effect is compared. The results show that all the four methods can improve the computational efficiency, and the acceleration effect of MPI is the best.

关 键 词： 计算效率 串行计算 并行化 加速比