Hierarchical parallelization of imperative programs

We develop a general and efficient hierarchical parallelization algorithm for /spl I.bold/w/spl I.bold/h/spl I.bold/i/spl I.bold/l/spl I.bold/e programs which, in contrast to previous attempts, does not include run-time checks, preserves determinism and maintains the size of the argument program. The point of our algorithm is the hierarchical decomposition of the argument program into sequences of possibly complex statements, like, for example, nested loops, which are parallelized independently. Parallel execution of non-conflicting complex statements is achieved by hierarchical parallelization. The algorithm is going to be implemented for the generation of efficient code for a Sequent Symmetry, a parallel machine with shared memory.<>