A Framework for Generating Group-Parallel Programs

We consider the generation of mixed task and data parallel programs and discuss how a clear separation of task and data parallelism can support the development of efficient programs. The separation between the two levels is preserved during the design and translation phases while allowing specific information to flow between the two levels. Thus, the model is able to combine well-known results in the detection of data parallelism with the design of task parallelism for a specific application. We show how parallel program parts can be generated from a data-parallel and a task-parallel specification and how the interaction between the two levels can be established