Dataflow Frequency Analysis CS701

This lecture continues the presentation of how to perform Dataflow Frequency Analysis [5] on the folded DAG structure generated via the Larus algorithm for collecting a whole-program path. The frequency-analysis technique can be applied to the class of bi-distributive dataflowanalysis problems. This lecture formalizes the technique of generating dataflow-frequency facts by traversing the DAG representation of a grammar that represents a whole-program path (without unfolding the DAG). 1 Whole Program Paths The previous lecture explained Larus’s algorithm to collect a “whole-program path” (WPP) [2]. The algorithm uses the Sequitur algorithm [3] to build a somewhat degenerate context-free grammar that accepts exactly one word, which corresponds to the path that was executed by an instrumented program. The terminal symbols of the context-free grammar are the acyclic Ball-Larus path fragments [1] of the program’s control flow graph. The whole-program path is returned in the form of a DAG, each of whose leaves is a path fragment. Consider the control flow graph shown in Fig. 1. There are three acyclic path fragments in the CFG: [