基于Epiphany计算并行矩阵乘法的研究  

The research based on the Epiphany of computing parallel matrix multiplication

作　　者：曾弘扬 

机构地区：[1]北京工业大学软件学院,北京100124

出　　处：《电子设计工程》2015年第24期52-55,共4页Electronic Design Engineering

摘　　要：为了提高矩阵乘法的计算效率,首先通过分析串行实现的矩阵乘法当中的不足之处,找出在单核上计算所存在的问题,提出将操作矩阵进行二维块状划分并行计算的解决方案。基于二维网状结构的多核处理器,先从一般矩阵与特殊矩阵向量间的乘法开始分析,再给出计算一般矩阵乘法的Cannon算法。结合多核并行运算的天然优势,通过对所划分子块的合理调度,来克服核间通信传输速率缓慢的弊端,以实现高效计算矩阵乘法的目的。通过模拟实验与单核上的串行计算对比,证实并行计算的矩阵乘法在耗时上有了显著的提高。In order to improve the computational efficiency of matrix multiplication, first, through the analysis of the deficiency of the matrix multiplication of serial implementation, find out the problems of calculation on mononuclear, puts forward the operation matrix for two-dimensional massive parallel computing solutions. Based on two-dimensional mesh structure of multi-core processors, start with the general matrix and special matrix vector multiplication between began to analysis, and then the calculation of general matrix multiplication Cannon algorithm. Natural advantages, combined with multi-core parallel computing based on the delimit the reasonable scheduling of molecular block, to overcome the disadvantages of nuclear communication transmission rate is slow, in order to realize the purpose of efficient calculation of matrix multiplication. Through the simulation experiment and mononuclear serial calculation contrast, confirmed that the parallel computing of matrix multiplication has improved significantly in the time-consuming.

关 键 词：矩阵乘法 多核 并行计算 Cannon 

分 类 号：TN918.1[电子电信—通信与信息系统]