Graph grammars have been developed as an extension of the formal grammars on strings to grammars on graphs, and provide a mechanism in which transformations on graphs can be modeled in a mathematically precise way. In this paper, based on confluent graph grammars, the authors present a novel representation for data-flow graphs, control-flow graphs, combined control-data-graphs, bipartite graphs and hyperedge graphs. How to extract parallelism is specified automatically at different levels by graph rewriting, thus facilitating the design and implementation of parallel compilers and parallel languages.