Minimally Synchronous Parallel ML with Parallel Composition
Résumé
Parallel divide-and-conquer parallel algorithms are very common in the litterature on parallel algorithms. In order to ease their implementation in Minimally Synchronous Parallel ML (MSPML), we add a new primitive to this asynchronous parallel functional language. MSPML is based on a parallel data structure called parallel vectors and operations to manipulate them. This paper presents semantics and an implementation of MSPML with parallel composition.