Functional Program Transformation for Parallelisation Using Skeletons

It can be challenging to use algorithmic skeletons in parallel program development as it is tedious to manually identify parallel computations in an algorithm and there may be mismatches between the algorithm and skeletons. Also, parallel programs defined using skeletons often employ inefficient intermediate data structures. In this paper, we present a program transformation method to address these issues by using an existing technique called distillation to reduce the use of intermediate data structures and an encoding technique to combine the inputs of a program into a single input whose structure matches that of the program. This facilitates automatic identification of skeletons that suit the algorithmic structure of the transformed program.