Parallel Flow Graph Matching for Automated Program Recognition

Abstract

A flow graph matching algorithm has been implemented on the Connection Machine which employs parallel techniques to allow efficient subgraph matching. By constructing many different matchings in parallel, the algorithm is able to perform subgraph matching in polynomial time in the size of the graphs. The automated program recognition system can use this algorithm to help make a more efficient flow graph parser. The process of automated program recognition involves recognizing familiar data structures and algorithmic fragments (called clichés) in a program so that a hierarchical description of the program can be constructed. The recognition is done by representing the program as a flow graph and parsing it with a graph grammar which encodes the clichés. In order to find clichés in the midst of unfamiliar code, it is necessary to run the parser on all possible subgraphs of the graph, thus starting the parser an exponential number of times. This is too inefficient for practical use on large programs, so this algorithm has been implemented to allow the matchings to be performed in polynomial time.