Automatic parallelization of lazy functional programs

Abstract

We present a parallelizing compiler for lazy functional programs that uses strictness analysis to detect the implicit parallelism within programs. It generates an intermediate functional program, where a special syntactic construct ‘letpar’, which is semantically equivalent to the well-known let-construct, is used to indicate subexpressions for which a parallel execution is allowed. Only for sufficiently complex expressions a parallelization will be worthwhile. For small expressions the communication overhead may outweigh the benefits of the parallel execution. Therefore, the parallelizing compiler uses some heuristics to estimate the complexity of expressions. The distributed implementation of parallelized functional programs described in [Loogen et al. 89] enabled us to investigate the impact of various parallelization strategies on the runtimes and speedups. The strategy, which only allows the parallel execution of non-predefined function calls in strict positions, shows the best runtimes and reasonable speedup results.