Eliminating two kinds of data flow inaccuracy in the presence of pointer aliasing

Abstract

Program languages with sophisticated usage of pointers as C are hard to analyze. Recent researches on pointer analysis focus on tracking the possible values of pointers, when a program point is reached, and great progress has been achieved. However, how to apply the result of pointer analysis to dataflow analysis and other program optimization/parallelization is not well studied. This paper presents an efficient interprocedural framework based on two insights in real C program and its use in deriving an context-sensitive pointer analysis algorithm and an accurate interprocedural modification side effects (MOD) computation. Based on the result of the pointer analysis, the inaccuracy induced by merging aliasing information is also studied.