Model for Evaluating Points-to Analysis in GraalVM Native Image Using Instrumentation-Based Profiling

Static analysis is an essential part of modern ahead-of-time compilers. The level of optimisation and the execution performance of a program directly depends on the analysis results. This paper presents a methodology for empirically evaluating the quality of static analysis implemented in the state-of-the-art Java AOT compiler – GraalVM Native Image. The model combines the data from the points-to analysis with the data from the instrumentation-based profiler to evaluate a generated binary. Finally, we present the initial results of our prototype on exemplary Java programs.