摘要
Many compiler optimization techniques depend on which part code has been executed frequently. Profilingwill trace and record these information that a compiler needs automatically, which is useful to other phases duringcompile. Profiling will instrument some internal code that gather these information during a program executes. For ex-ample, edge profiling computes the frequency of basic block and the probability of edge in control flow graph. Valueprofiling determine which variables have invariant behavior.This paper explores the basic block frequency found from edge profiling. We use the edge profiling to performsome optimization for Spec2000 test cases, showing that the optimization base edge profiling can reduce a program'sexecution time up to 5 %.
Many compiler optimization techniques depend on which part code has been executed frequently. Profiling will trace and record these information that a compiler needs automatically, which is useful to other phases during compile. Profiling will instrument some internal code that gather these information during a program executes. For example , edge profiling computes the frequency of basic block and the probability of edge in control flow graph. Value profiling determine which variables have invariant behavior.
This paper explores the basic block frequency found from edge profiling. We use the edge profiling to perform some optimization for Spec2000 test cases, showing that the optimization base edge profiling can reduce a program's execution time up to 5%.
出处
《计算机科学》
CSCD
北大核心
2003年第1期46-48,66,共4页
Computer Science
基金
国家自然科学基金(69933020
60103006)