A FEATURE-BASED APPROACH FOR MINING PARALLELISABLE POINTS OF PROGRAM

Guo Shen,Li Peifeng,Zhu Qiaoming
DOI: https://doi.org/10.3969/j.issn.1000-386X.2011.04.007
2011-01-01
Abstract:Exploring parallelisable points of codes in program is the first step of parallel compiling technology.In this paper,we propose a novel method based on SVM to mine the parallelisable points in running program according to the features it uses to represent the parallelity of program,including program executing time,circulation part of program,data dependency analysis,and ratio of execution time and the times of circulation,etc.Experimental results show that the parallelisable points mined by our method meet the needs of practical application much better,and the mined points can get considerable performance speedup ratio after to be parallelised.
What problem does this paper attempt to address?