A shape matching approach for scheduling fine-grained parallelism