摘要： 主要研究卷积计算的数据并行方法在MCC-SIMD计算机的设计和实现问题。在参考国内外现有资料的基础上，从图论的角度用卷积路径的方法对卷积计算的有关概念进行了严格的形式化定义；分析了4-连接卷积路径下数据并行卷积计算时处理元阵列的移动规律;在此基础上给出了数据并行卷积计算算法；最后对算法的计算复杂性和通信复杂性进行了简要分析。文中还对有关结论进行了证明。

关键词: 数据并行, 卷积计算, 卷积路径

Abstract: This paper mainly discusses the design and implementation of convolution computation in image processing on MCC-SIMD computer. By a detailed study on the current literature, in terms of graph, this paper proposes formal definitions on convolution calculation by the method of convolution path. It analyzes the regular pattern of the shift of PE array while calculating on data parallel convolution on 4-connected convolution path. On the basis of the above, a data parallel algorithm of convolution calculation is proposed. In the end, computation complexity and communication complexity are analyzed. Also ,some conclusions concerned have been proved.

中图分类号: 

• TP301.6