A parallelizing method for implementing image processing tasks on SIMD linear processor arrays

Proceedings Fourth IEEE International Workshop on Computer Architecture for Machine Perception. CAMP'97 Pub Date : 1997-10-20 DOI:10.1109/CAMP.1997.631944

S. Kyo, S. Okazaki, Y. Fujita, N. Yamashita

引用次数: 10

Abstract

This paper describes Stack-Based Method (SBM), a general parallelizing method for implementing region parallel operations on Linear Processor Arrays (LPAs). By taking full advantage of the addressing autonomy of each PE. SBM proceeds a pixel-updating-wave (PUW) in a image content directed way across the the image which is separately located on local memories of PEs. Based on SBM, each PE can detect and operate only on pixel data which are ready to be operated in a parallel and autonomous way under the SIMD constraint. Enhancements of existing LPA architectures for full support of SBM is also discussed.

查看原文本刊更多论文

一种在SIMD线性处理器阵列上实现图像处理任务的并行化方法

本文介绍了一种在线性处理器阵列(lpa)上实现区域并行运算的通用并行化方法——基于堆栈的方法(Stack-Based Method, SBM)。通过充分利用每个PE的寻址自主权。SBM以图像内容定向的方式在单独位于pe本地存储器上的图像上进行像素更新波(PUW)。基于SBM，每个PE只能在SIMD约束下对准备并行自主操作的像素数据进行检测和操作。本文还讨论了对现有LPA体系结构的增强，以充分支持SBM。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Proceedings Fourth IEEE International Workshop on Computer Architecture for Machine Perception. CAMP'97

自引率

0.00%

发文量