大规模矩阵乘积的几个并行算法

Several Parallel Algorithms for the Multiplication of Large Matrix

给出了微机网络并行计算环境和拥有共享内存的并行计算环境下大规模矩阵相乘的几种并行算法,分析了这些算法的计算和通讯时间复杂性,在多台微机上安装并行消息传递机制MPI(message passing interface),构成微机网络并行计算环境,在该环境下对所给算法作了数值试验,结果表明所给出的算法非常有效,同时在程序没有改动的条件下,将微机网络并行环境下的程序移植到具有共享内存的并行计算机IBMP650上,结果表明,所给算法具有更高的加速比和并行效率。

Several parallel algorithms of matrix multiplication under the environment of PC net- work and the shared memory parallel environment are given with the complexity of computation and message passing analyzed. The message passing interface （MPI） is installed on several microcomputers to construct the PC network parallel computing environment. Results of numerical experiment show that the parallel algorithm is highly effective. Besides, numerical experiments on the shared memory computer IBMP650 also show that the efficiency is very high.