DOPPLER · Domain-optimised parallelisation by polymorphic language embeddings and rewritings

Concurrent and parallel programming are becoming indispensable forexploiting modern hardware. Because possible speed increases of singleprocessors have reached their limit, increasing transistor count willyield more, but not necessarily faster cores, and this for theforeseeable future. This means that, from now on, parallelism insoftware will have to double every 18 months to keep up withhardware. This problem has been identified as the ``Popular ParallelProgramming'' grand challenge by the computer architecture community.The proposed project will research new ways to solve this challenge.We start with a set of domain-specific languages which naturally admita high degree of parallelism. The domain specific languages areintegrated in a common host language using polymorphic languageembeddings. Such embeddings provide a high degree of abstraction,which gives considerable freedom in their actual representation andimplementation. The new direction taken by this proposal is to combinepolymorphic embeddings with optimizing rewritings in a stagedcompilation process. This can lead to highly parallel and efficientimplementations on a variety of heterogeneous hardware.