摘要
可重构计算以其优异的性能和高度的灵活性,在国际国内研究领域逐渐引起广泛的关注.然而,在研的可重构计算系统架构多种多样,编程模型多与体系结构相关,使用和移植都非常困难.本文为解决编程通用性问题,从可重构计算的基本特征出发,提出数据驱动的,支持异构任务并行计算的统一编程模型,并讨论其实现方法.该模型基于生产者.消费者通讯机制,支持多种类型的计算结点和通讯网络,具有高度的抽象性.实验结果显示,使用统一编程模型进行应用设计,在不同的架构上能够使用同样的用户程序,并且获得比纯硬件加速方式更高的加速比.
Due to the excellent performance and flexibility,reconfigurable computing has gained more and more attention throughout the world.But currently there're so many different platforms and programming frameworks which take you into the details of the specific hardware,and prevent their practical usage.In this paper,after deep study of the reconfiguration characteristic,a novel data-driven uniform programming model RECUPM is proposed,and its implementation is discussed.The model supports parallel hybrid-task computing.It's based on producer-consumer communication paradigm and can be adapted onto different types of networks and nodes.Experiments show that applications featuring RECUPM reuse the same source codes on different architectures, and outperform the pure hardware acceleration design.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2007年第11期2123-2128,共6页
Acta Electronica Sinica
基金
高等学校博士学科点专项科研基金(No.20050358040)
关键词
可重构计算
编程模型
生产者-消费者通讯模型
reconfigurable computing
programming model
producer-consumer communication model