On the impact of naming methods for heap-oriented pointers in C programs

Abstract

Many applications written in C allocate memory blocks for their major data structures from the heap space at run time. The analysis of heap-oriented pointers in such programs is critical for compilers to generate high-performance code. However, most previous research on pointer analysis mostly focuses on pointers pointing to global or local variables. In this paper, we study points-to analysis of heap-oriented pointers using profiling information. An instrumentation tool and a set of library routines are developed to measure points-to sets of memory references at run time. Different naming methods for heap-oriented pointers are studied. We found that it is very important to adopt appropriate naming methods to recognize wrapper functions for memory allocation and memory management functions defined by users. Based on these naming methods, the approaches in pointer analysis, such as flow sensitivity and context sensitivity, are examined with the run-time tool. The program characteristics are observed at run time to evaluate what kind of compiler analysis is needed. Experiments are conducted on SPEC CPU2000 integer benchmarks. We found that flow sensitivity and context sensitivity have little impact on the analysis of heap-oriented pointers.