This new project, supported by EPSRC grant GR/L42889, will develop a parallel compiler for a pure functional subset of Standard ML. Exploitable parallelism at sites of higher order function use will be implemented through algorithmic skeletons written in C and MPI. The target architecture is the Fujitsu AP1000 in the first instance.
A functional prototype will be instrumented to determine communication and processing costs in higher order function use, through an implementation independent Structural Operation Semantics interpreter. Performance models for algorithmic skeletons equivalent to higher order functions will enable the identification of useful parallelism. Program transformation will be used to try to optimise parallelism.
The project will run for 3 years from September 1st 1997, and builds on earlier research. We are pleased to acknowledge the support of the Imperial College Fujitsu Parallel Computing Research Centre.Current group members working in this area are: