TwAS

Abstract

A major drawback of shape analysis techniques is the trade-off between speed and precision. We present TwAS: a novel method to combine two shape analysis techniques-a fastbutless precise technique with a slow but more precise one-to get the best of both (speed as well as precision). The novelty of our approach is the use of fast analysis to filter heap variables for which the shape information is already precise and can not be refined further This allows us to run the slow analysis on only a small portion of the program, thus improving its performance. We implemented TwAS in GCC as a dynamic plugin as an inter-procedural data-flow analysis and evaluated it on standard benchmarks against the component analyses. TwAS is able to achieve the same precision as the slow analysis at a marginal slowdown compared to the fast analysis. TwAS is able to improve precision for 5 out of 8 Olden benchmarks compared to the fast-but-imprecise analysis, while retaining the precision for the others. At the same time, TwAS is 4X--2500X faster than the precise-but-slow analysis for these benchmarks, without any loss in precision.