| The age of parallel computing has brought with it the need for compilers which explore sequential code, discover the inherent parallelism and generate optimized code which fully utilizes the existing hardware. The analysis techniques employed by such compilers are of considerable interest to researchers. This thesis reviews our enhancements and generalizations of many of these techniques.; Of particular importance is the representation of applications by data-flow graphs (DFGs), whose nodes represent tasks and whose edges represent data dependencies among tasks. Much research has been done attempting to optimize such applications by applying graph transformation techniques to the corresponding DFG. The work discussed in this thesis has two major components. First is the generalization of the DFG model to describe a broader group of situations. We then develop the transformation techniques which may manipulate this more general model and achieve optimality. Finally, we discuss the application of DFGs to new problems heretofore unexplored, including nested loop fusion and data dependence analysis. |
