摘要
针对多约束下粗粒度可重构计算阵列数据流图的映射问题,进行了时间Petri网的分析与建模,并提出了宽度映射和宽度贪心映射两种算法;基于一组实验基准和随机选取的处理单元阵列PEA4×4和PEA5×5对这两种算法进行测试。结果表明,相比于宽度映射算法,宽度贪心映射算法在映射块数、执行时延、输入输出访存成本等方面均获得了优化,平均总时延降低了24.5%。
Aiming at the mapping problem of data flow graph of coarse-grained reconfigurable computing array under multiple con⁃straints,the time Petri net is analyzed and modeled,and two algorithms,width mapping and width greedy mapping,are proposed.These two algorithms are tested based on a set of experimental benchmarks and randomly selected processing unit arrays PEA4×4 and PEA5×5.The results show that compared with the width mapping algorithm,the width greedy mapping algorithm is opti⁃mized in the number of mapping blocks,execution delay,input and output memory access costs,and the average total delay is re⁃duced by 24.5%.
作者
方冉
陈乃金
FANG Ran;CHEN Naijin(School of Application Engineering,Anhui Business and Technology College,Hefei Anhui 231131,China;School of Computer and Information,Anhui Polytechnic University,Wuhu Anhui 241000,China)
出处
《盐城工学院学报(自然科学版)》
2024年第3期51-58,共8页
Journal of Yancheng Institute of Technology:Natural Science Edition
基金
安徽省高校自然科学研究项目重点项目(KJ2021A1510)
关键词
PETRI网
时延特性分析
粗粒度可重构体系结构
映射
Petri net
time delay characteristics analysis
coarse grained reconfigurable architecture
mapping
