Analysis of Parallelism in Recursive Functions on Recursive Data Structures

In functional languages, iterative operations on data collections are naturally expressed using recursive functions on recursive data structures. In this paper, we present a method to extract data parallelism from recursive functions and generate data parallel programs. As the parallel model for object programs, we use polytypic parallel skeletons. This model can express data parallel operations on general re-cursive data structures. With this model, we can extract data parallelism from recursive functions on general recursive structures such as trees and lists. To extract appropriate computations for each parallel skeleton from an input function, we classify all subexpressions in the function body based on their conditions of parallel execution. Then, a data parallel program is generated from the classiied expressions of the function. Comparing our analytical approach with previous calculational approaches, our approach can parallelize more general recursive functions with complex data ow.