Moving Heterogeneous GPU Computing into the Mainstream with Directive - Based , High - Level Programming Models ( Position Paper ) DOE

Graphics Processing Units (GPUs)-based heterogeneous systems have emerged as promising alternatives for highperformance computing. However, their programming complexity poses significant challenges for developers. Recently, several directive-based, GPU programming models have been proposed by both academia and industry ( [1], [2], [3], [4], [5], [6]) to provide better productivity than existing ones, such as CUDA and OpenCL. Directive-based models provide different levels of abstraction, and programming efforts required to conform to their models and optimize the performance also vary. Understanding the differences in these new models will give valuable insights on the general applicability and performance of the directive-based GPU programming models. This position paper evaluates the existing high-level GPU programming models to identify current issues and future directions that stimulate further discussion and research to push the state of the art in productive GPU programming models.