[Soot-list] Parallel Spark

[919 at sjtu.edu.cn]

    I am writing to you to for advice on parallelism of soot.jimple.spark. Since spark is one of the most efficient frameworks of points-to analysis for Java, the algorithm becomes critical in performance. However, whole program analysis may lead to performance crash because of the propagation algorithm; sensitivity aggravates the performance; Java library can produce too many points-to relations and long points-to paths. 

    I tried to inspect the source code of soot.jimple.spark, and explore the parallelism in the construction of on-fly-call-graph. Many loops were found, and I think two of those loops can be scaled. The first loop is in the method OnFlyCallGraphBuilder.processReachables(). The second loop is in the method MethodPAG.buildNormal(). The first loop can be scaled to a parallel analysis among different methods. The second loop can be scaled to a parallel analysis among different statements.