Comparing Points-to Static Analysis with Runtime Recorded Profiling Data
Codrut Stancu, Christian Wimmer, Stefan Brunthaler, Per Larsen, Michael Franz: Comparing Points-to Static Analysis with Runtime Recorded Profiling Data. In Proceedings of the International Conference on Principles and Practices of Programming on the Java Platform: Virtual Machines, Languages, and Tools, pages 157–168. ACM Press, 2014. doi:10.1145/2647508.2647524Abstract
We present an empirical study that sheds new light on static analysis results precision by comparing them with runtime collected data. Our motivation is finding additional sources of information that can guide static analysis for increased application performance.
This is the first step in formulating an adaptive approach to static analysis that uses dynamic information to increase results precision of frequently executed code. The adaptive approach allows static analysis to (i) scale to real world applications (ii) identify important optimization opportunities. Our preliminary results show that runtime profiling is 10% more accurate in optimizing frequently executed virtual calls and 73% more accurate in optimizing frequently executed type checks.