矩阵相乘算法优化的研究  

Research of Matrix Multiplication Algorithm Optimization

作　　者：钱晓捷[1] 杨镇江[1] 杜志刚[1] 李秀芳[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450001

出　　处：《微计算机信息》2009年第27期182-183,208,共3页Control & Automation

摘　　要：本文对经典矩阵相乘A*B算法提出多种优化方法:根据局部性原理,提出对矩阵B进行转置;根据计算机缓存的大小与矩阵A与矩阵B的规模进行嵌套循环分块,通过对分块大小的调整比较获得最优的分块;利用循环展开技术以提高程序的并行性。实验结果表明,优化后的算法缩短了运行时间,获得了较优的运行效率。Many kinds of optimization methods to the classical matrix multiplication A＊B algorithm are proposed in this paper： transpose the matrix B according to the principle of locality; circulate and block the program according to the size of computer cache and the scale of matrix A and matrix B, compare and adjust the size of block to get the better blocking; enhance the program parallelism using loop unrolling. The result of the experiment shows that the optimized algorithm can not only reduce the time of operation, but also gain better operation efficiency.

关 键 词：矩阵相乘算法 矩阵转置 循环分块 循环展开 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术]