行划分的矩阵相乘并行改进及其DSP实现  

The improvememnt realization of matrix multiplication that based on rows partition and in DSP

作　　者：张新菊[1] 刘羽[1] 韩枭[1] 

机构地区：[1]桂林工学院电子与计算机系,541004

出　　处：《微计算机信息》2008年第20期216-218,共3页Control & Automation

基　　金：天然场源大地电磁数据处理并行计算研究项目编号:0339070颁发部门:广西科技厅

摘　　要：在阵列信号处理中需要大量的矩阵运算,而其中最基本的就是矩阵相乘运算。本文就矩阵相乘的行划分并行实现进行了改进,将A矩阵的一行和整个B矩阵传输到每个工作进程,其中第一个工作进程指定在某台机器,其余工作进程由PVM选择在合适的机器上产生。该并行实现基于PVM环境,采用主从编程模式,然后给出了改进的行划分矩阵相乘在DSP上实现的方案。Owing to needing a great deal of matrix operation in the array signal processing, but among them the most essential is the matrix multiplication.This text studies the improvement that based on line partition of matrix multiplication, delivers one line of matrix A and the whole matrix B to each progress, and the first work progress is appointed in the certain machine, the other work pro- gresses are produced in the appropriated machine by PVM.This scheme is based on the PVM environment and Master/Slave mode, then achieves the matrix multiplication of improved line partition in the DSP.

关 键 词：工作站机群 PVM 并行算法 DSP 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术]