high performance computing on graphics processing units: hgpu.org

Heterogeneous many-core architectures offer a way to cope with energy consumption limitations of various computing systems from small mobile devices to large data-centers. However, programmers typically must consider a large diversity of architectural information to develop efficient software. In this paper we present our ongoing work towards a Platform Description Language (PDL) that enables to capture key architectural patterns of commonly used heterogeneous computing systems. PDL architecture patterns support programmers and tool chains by providing platform information in a well-defined and explicit manner. We have developed a source-to-source compiler that utilizes PDL descriptors to transform sequential task-based programs to a form that is convenient for execution on heterogeneous many-core computing systems. We show various usage scenarios of our PDL and demonstrate our approach for a commonly used scientific kernel.