An Evaluative Comparison of Performance Portability across GPU Programming Models