GpH is a minimal, conservative extension of Haskell'98 to support parallelism. Experience has shown that it is particularly good for constructing symbolic applications, especially those with irregular parallelism, e.g. where the size and number of tasks is dynamically determined. The project has been ongoing since 1994, initially at Glasgow, and now at Heriot-Watt and St Andrews Universities.
GpH extends Haskell'98 with parallel composition: par. Parallel and sequential composition are abstracted over as evaluation strategies to specify more elaborate parallel coordination, e.g. parList s applies strategy s to every element of a list in parallel. Evaluation strategies are lazy higher-order polymorphic functions that enable the programmer to separate the algorthmic and coordination parts of a program. A number of realistic programs have been parallelised using GpH, including a Haskell compiler and a natural language processor.
GpH is publicly available from the page below, and is implemented on GUM, a sophisticated runtime system that extends the standard GHC runtime system to manage dynamically much of the parallel execution, e.g. task and data placement. GUM is portable: using C and standard communication libraries (PVM or MPI), and hence GpH is available on a range of platforms, including shared-memory, distributed memory and workstation clusters, e.g. Beowulf. GpH shares implementation technology with the Eden and GdH languages.
Current work includes making GpH architecture independent, i.e. deliver good parallel performance on a range of platforms. Improved parallel profiling, parallel semantics and abstract machines, and performance comparison with other languages.