An automatic thread decomposition approach for pipelined multithreading. (1st January 2013)

• Record Type: Journal Article
• Authors: Zhang, Yuanming; Ootsu, Kanemitsu; Yokota, Takashi; Baba, Takanobu
• Is Part Of: International journal of high performance computing and networking. Volume 7:Number 3(2013)
• Page Start: 227
• Page End: 237
• Publication Date: 2013-01-01
• Subjects: high performance -- pipelined multithreading -- PMT -- multi-core processors -- thread decomposition -- optimised pipeline; High performance computing -- Periodicals; Computer networks -- Periodicals; High performance computing; Periodicals; 004.05
• Languages: English
• ISSNs: 1740-0562

Abstract

Thread decomposition is critical for pipelined multithreading (PMT) to gain higher performance on target multi-core processors. This paper presents an automatic thread decomposition approach, which maps the decomposition problem onto a graph-theoretic framework to construct an optimised directed acyclic graph (DAG) with minimal bottleneck node size and balanced node size. In this approach, control dependence is treated as special data dependence and then an effective approach is proposed to remove redundant control dependences. A weighted DAG is constructed by assigning appropriate weights to all nodes and all dependences according to profile information. An automatic thread decomposition algorithm is given to generate an optimised pipeline based on the weighted DAG. The algorithm has been evaluated on a commodity multi-core processor, and experimental results show that it has achieved speedup ranging from 113% to 174% on some SPEC CPU 2000 benchmark programs.