high performance computing on graphics processing units: hgpu.org

The heterogeneous parallel computing era has been accompanied by an ever-increasing number of disparate programming models. As a result, improving performance via heterogeneous computing is currently very challenging for application programmers. Domain-specific languages (DSLs) are a potential solution to this problem, as they can provide productivity, performance, and portability within the confines of a specific domain. However, making the DSL approach useful on a large scale requires lowering the barrier for DSL development. We describe a reusable compiler infrastructure called the Delite Compiler Framework that drastically simplifies the process of building embedded parallel DSLs. DSL developers can easily implement domain-specific operations by extending this framework, which provides static optimizations and code generation for heterogeneous hardware. We also describe the Delite Runtime, which automatically schedules and executes DSL operations on heterogeneous hardware. We demonstrate the potential of the DSL approach by showing the performance of applications written in OptiML, a machine learning DSL developed with the framework, on a system with multi-core CPUs and GPU.