Comparing Rapid Type Analysis with Points-To Analysis in GraalVM Native Image

Whole-program analysis is an essential technique that enables advanced compileroptimizations. An important example of such a method is points-to analysis usedby ahead-of-time (AOT) compilers to discover program elements (classes, methods,fields) that may be used on at least one program path during the run of theprogram and hence need to be compiled. GraalVM Native Image uses a points-toanalysis to optimize Java applications, which is a time-consuming step of thebuild. We explore how much the analysis time can be improved by replacing thepoints-to analysis with a rapid type analysis (RTA), which computes reachableelements faster by allowing more imprecision. We propose several extensions ofprevious approaches to RTA, making it parallel, incremental, and supporting heapsnapshotting. We present an extensive experimental evaluation of the effects ofusing RTA instead of points-to analysis, showing, e.g., that the analysis timeof Spring PetClinic can be reduced by 64 \% and the overall build timeby 35 \% at the cost of increasing the image size by 15 \%.