查看文章

Ubiquitous multicore computers and affordable cloud computation services provides a great opportunity for increasing the speed or precision of program analysis algorithms. We present the first parallel algorithm for a context and field sensitive inclusion-based (Andersen-style) pointer analysis with on-the-fly call graph construction. We show how the pointer analysis and call graph construction can be naturally modeled as an actor system. We implement the algorithm as a drop-in replacement of WALA’s (TJ Watson Libraries for Analysis) pointer analysis and allow for all the flexibility of the original library, ie, highly configurable context sensitivity, pointer and instance abstraction, etc. We measure the parallel scalability of our algorithm and compare it to the the highly-optimized WALA implementation. We discuss the results, limitations, and avenues for improvement.