Computer Engineering and Applications ›› 2007, Vol. 43 ›› Issue (35): 86-89.

• 学术探讨 • Previous Articles     Next Articles

Specification for sequential image processing algorithms based on SWP

TANG Wen-sheng,KUANG Wang-qiu   

  1. Department of Computer Teaching,Hunan Normal University,Changsha 410081,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-12-11 Published:2007-12-11
  • Contact: TANG Wen-sheng

基于子字并行的串行图像处理算法表示方法

唐文胜,匡旺秋   

  1. 湖南师范大学 计算机教学部,长沙 410081
  • 通讯作者: 唐文胜

Abstract: Porting sequential image processing algorithm to processors with multimedia extensions and multimedia-specific embedded microprocessors is a hot research field.There are intrinsic Sub-Word Parallelisms(SWP) in image processing algorithms.Current compilers can not locate the SWP within a basic block.This paper proposes an approach to generate explicit SWP specification from source code through program representation and pattern-matching approach.Our computation representation model is based on an extension of the MultiDimensional Synchronous DataFlow(MDSDF).We add a specialized pattern-library to compiler without modifying the compiler or program source code.The SWFG(Sub-Word Flow Graph) representation generated from pattern-matching data-flow and control flow analysis can be used as an intermediate representation for compiler’s next step processing.

Key words: image-processing, Sub-Word Parallelishs(SWP), pattern matching

摘要: 多媒体应用程序的一个最突出的特点就是其内在的并行性,要求同时对多个数据单元进行相同的规则操作,这种并行性被称为子字并行。如何让编译器产生有效的并行代码,如何从普通的串行C程序中识别出子字并行指令,是一个前沿性的研究课题。提出了一种用于实现对串行源代码进行显式表示的方法,该方法对数字信号处理领域常用的同步数据流图方法进行改进,结合模式匹配技术,着眼于内层循环,通过扩展规范的模式库,对带模式识别的数据流、控制流进行分析,能够从串行应用程序中自动地提取其固有的子字并行,得到显式的并行化中间表示,基于这种数据流图表示,采用改进的树模式匹配实现子字并行指令选择和代码生成。实验测试表明了该方法的有效性。

关键词: 图像处理, 子字并行, 模式匹配