An Evaluation of Directive-Based Parallelization on the GPU Using a Parboil Benchmark