摘要
在可配置处理器的定制指令设计过程中，需要提取热点代码数据流图的凸连通子图．为实现子图的快速枚举，对有向无环图内的凸子图特性进行了研究．根据凸子图特性和节点邻接关系，提出了一种AS（adjacent search）算法用于枚举有向无环图内满足I／O端口约束的凸连通子图．实验数据显示，AS算法比现有算法具有更高的效率，加速比可达10～1000X．当现有算法因数据流图规模较大而失效时，应用AS算法仍能成功完成子图枚举．
Enumerating convex connected subgraphs from data flow graphs of application hot-spots is required when designing instructions for a configurable processor. To achieve fast enumeration, properties of convex subgraphs of directed acyclic graph are studied. Based on the properties of convex subgraphs and adjacency of vertices, AS （adjacent search）, a novel algorithm for enumerating convex connected subgraphs satisfying I/O constraints is presented. Results of experiments show that AS algorithm is more efficient than the existing algorithm, and rate is 10~1000X as fast. While the existing algorithm fails on large scale data-flow graphs, the AS algorithm is still able to accomplish enumeration successfully.
出处
《软件学报》
EI
CSCD
北大核心
2010年第12期3106-3115,共10页
Journal of Software
基金
国家高技术研究发展计划(863)No.2007AA01Z2b3
国家重点基础研究发展计划(973)No.2007CB310608~~
关键词
凸连通子图
有向无环图
数据流图
枚举
可配置处理器
定制指令
convex connected subgraph
directed acyclic graph
dataflow graph
enumeration
configurable processor
custom instruction
