• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2007, Vol. 29 ›› Issue (11): 131-133.

• 论文 • 上一篇    下一篇

矩阵向量乘在流处理器上的实现

杜静 敖富江 杨学军   

  • 出版日期:2007-11-01 发布日期:2010-05-30

  • Online:2007-11-01 Published:2010-05-30

摘要:

Imagine是一款致力于媒体处理的流体系结构芯片,而它是否适合科学计算应用是当前国际上的热点讨论问题。本文研究了典型科学计算核心——矩阵向量乘在Imagine处理器  上的有效映射,主要开发了四种不同的流程序版本,并且提出了优化方法。实验结果表明,经过优化的矩阵向量乘程序能够开发Imagine处理器的潜能。

关键词: 矩阵向量乘Imagine三级存储层次 计算密集性 stripmining

Abstract:

Imagine is a stream processor aiming at media applications. However, whether scientific computing applications are suited for Imagine is a hot topic.   In this paper, we propose an efficient stream program development technique on Imagine for matrix-vector multiplication, which is a representative scien tific computing kernel. Especially, we propose four different program versions of this kernel, and explore efficient programming optimizations. The experimental results show that matrix-vector multiplication can efficiently exploit the tremendous potential of the Imagine stream processor through program  ming optimization.

Key words: (matrix-vector multiplieation, Imagine, three level memory hierarehy, computational intesiveness;  , stripmining)