摘要
粗粒度可重构阵列结构具有计算效率高的特点,但不能完美支持控制较复杂的算法.本文基于数据流驱动原理,提出了一种可重配置的混合粒度阵列架构,将细粒度可重构技术和粗粒度可重构技术相结合,实现了对控制密集型算法的支持,整个结构支持多发射循环迭代技术和空间展开循环迭代技术.通过对算法采取不同的映射方法和优化技术,相对于静态流执行情况至少可以获得15.6%的性能提升.
The coarse-grained reconfigurable array has the characteristics of high computational efficiency,but it does not perfectly support the control of complex algorithms.Based on theprinciple of data stream driven,this paper proposes a hybrid-grained reconfigurable architecture,which support the algorithms of control-intensive by combining fine-grained reconfigurable technology and coarse-grained reconfigurable technology.The architecture supports techniques of loop iteration fast issue and loop iteration spatial expansion.By adopting different mapping methods,we can get at least 20 percentage performance improvement.
作者
俞磊
景乃锋
王琴
YU Lei;JING Nan-feng;WANG Qin(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《微电子学与计算机》
北大核心
2019年第5期19-22,28,共5页
Microelectronics & Computer
关键词
数据流驱动
可重构阵列
混合粒度
多发射
空间展开
data stream drive
reconfigurablearray
hybrid-grained
fastissue
spatial expansion
