Evaluating automatic parallelization for efficient execution on shared-memory multiprocessors