典型可重构架构的算法映射分析

Analysis of Mapping Algorithms on Typical Reconfigurable Architectures

可重构架构高效、灵活,能够满足嵌入式处理领域的高性能需求.通过对三种粗粒度与两种多粒度的典型可重构架构进行建模,将算法分类,并映射不同类型的算法到可重构架构上,从硬件利用率,计算时间,输入输出带宽,数据组织形式,数据复用率等五个方面进行分析,得到多粒度的可重构架构,配以灵活的互联能够更好地完成各类算法,但是总体的硬件利用率较低.而数据带宽仍然是可重构架构的瓶颈.为可重构架构设计提供理论依据.

Reconfigurable architectures,which are efficient and flexible,can meet the high performance needs of embedded processing field.This paper models three coarse-grained and two multi-grained typical reconfigurable architectures,then classifies algorithm and maps different types of algorithm on the reconfigurable architectures.Analysis has five parts： hardware utilization,computing time,input and output bandwidth,data arrangement and data reuse rate.Conclusion is that multi-grained reconfigurable architecture with flexible interconnect can perform different algorithms better,however,hardware utilization rate is a little lower.Data bandwidth is still the bottleneck of reconfigurable architectures.Provide theoretical basis for later design of reconfigurable architectures.