GUM is a portable, parallel implementation of the Haskell functional language. Despite sustained research interest in parallel functional programming, GUM is one of the first such systems to be made publicly available. GUM is message-based, and portability is facilitated by using the PVM communications harness that is available on many multi-processors. As a result, GUM is available for both shared-memory (Sun SPARCserver multiprocessors) and distributed-memory (networks of workstations) architectures. The high message-latency of distributed machines is ameliorated by sending messages asynchronously, and by sending large packets of related data in each message. Initial performance figures demonstrate absolute speedups relative to the best sequential compiler technology. To improve the performance of a parallel Haskell program GUM provides tools for monitoring and visualising the behaviour of threads and of processors during execution.

This paper describes equals, a fast parallel implementation of a lazy functional language on a commercially available shared-memory parallel machine, the Sequent Symmetry. In contrast to previous implementations, we detect parallelism automatically using strictness analysis. Another important difference between equals and previous implementations is the use of reference counting for memory management, instead of garbage collection. Our implementation shows that reference counting leads to very good scalability, low memory requirements and improved locality. We compare our results with other sequential (SML/NJ) and parallel (h; Gi-machine and GAML) implementations of functional languages.

In ML style pattern matching, pattern size is not constrained and ambiguous patterns are allowed. This generality leads to a clear and concise programming style but is challenging in the context of lazy evaluation. A first challenge concerns language designers: in lazy ML, the evaluation order of expressions follows actual data dependencies. That is, only the computations that are needed to produce the final result are performed. Once given a proper (that is, non-ambiguous) semantics, pattern matching should be compiled in a similar spirit: any value matching a given pattern should be recognized by performing only the minimal number of elementary tests needed to do so. This challenge was first met by A. Laville. A second challenge concerns compiler designers. As it stands, Laville's compilation algorithm cannot be incorporated in an actual lazy ML compiler for efficiency and completeness reasons. As a matter of fact, Laville's original algorithm did not fully treat the case of intege...

this paper we take a more direct approach: we compile pattern matching on overlapping patterns. We first recall the semantics of lazy pattern matching, as given by A. Laville [5]. Then, we explain our compilation technique as a source to source transformation. Given a set of patterns, several compilations are possible, we prove that they all satisfy a partial correctness

The research presented in this thesis is about the design and implementation of Naira, a parallel, parallelising compiler for a rich, purely functional programming language. The source language of the compiler is a subset of Haskell 1.2. The front end of Naira is written entirely in the Haskell subset being compiled. Naira has been successfully parallelised and it is the largest successfully parallelised Haskell program having achieved good absolute speedups on a network of SUN workstations. Having the same basic structure as other production compilers of functional languages, Naira's parallelisation technology should carry forward to other functional language compilers. The back end of Naira is written in C and generates parallel code in the C language which is envisioned to be run on distributed-memory machines. The code generator is based on a novel compilation scheme specified using a restricted form of Milner's ß-calculus which achieves asynchronous communication. We present the f...

: Graph copying is used in parallel implementations of functional languages on architectures with distributed memory. This paper will explore the costs of graph copying in detail. It will make clear that these costs can form a bottleneck for a class of serious parallel programs. This not only comprises practical divide and conquer style programs - of which parallel matrix multiplication is an example -, but also programs that use pipelines, such as the sieve of Erathostenes. One can observe that copying costs vary widely for different data structures. We will show how arrays can be used to reduce copying costs considerably. This resulted in significant speed-ups for the examples above. Keywords: parallel functional programming, distributed memory, graph copying, graph conversions, delays 1. Introduction Graph rewriting systems have been studied as a computational model for functional languages and their implementation. The lazy functional language Concurrent Clean, for instance, is bas...

: is a system for parallel evaluation of lazy functional programs implemented on a Sequent Symmetry. The distinguishing features of over previous implementations are propagation of Normal Form demand and memory reclamation via reference counting. In an earlier paper, the validity of these ideas was established based on a preliminary implementation. Our experience, based on extensive experimentation with this implementation, led to the design and implementation of a new high-performance runtime system for described in this paper. We discuss the rationale behind the various design choices and provide quantitative evidence for the validity of these choices. Keywords: Lazy Functional Languages, Implementation, Parallelism, Performance Evaluation 1 Introduction Functional languages offer a conceptually simple approach to programming parallel computers. Detection and exploitation of parallelism in functional programs is simpler than in imperative programs, due to referential transparency. T...

A clustered architecture has been designed to exploit divide and conquer parallelism in functional programs. The programming methodology developed for the machine is based on explicit annotations and program transformations. It has been successfully applied to a number of algorithms resulting in a benchmark of small and medium size parallel functional programs. Sophisticated compilation techniques are used such as strictness analysis on non-flat domains and RISC and VLIW code generation. Parallel jobs are distributed by an efficient hierarchical scheduler. A special processor for graph reduction has been designed as a basic building block for the machine. A prototype of a single cluster machine has been constructed with stock hardware. This paper describes the experience with the project and its current state. 1 Introduction Functional programming is founded on the lambda calculus, which is a mathematical theory that provides a sound basis for work on reduction machines [5]. This is p...

Contents Contents 1 Table of Contents 2 2 Executive Summary 4 3 Infrastructure Description 7 4 Resource Allocation 9 5 Budget 14 5.1 Management structure : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 21 6 Research 22 6.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 22 6.2 Parallel XSB-Prolog - Warren : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 24 6.3 Parallel Contraction-Based Theorem Proving - Hsiang, Bachmair : : : : : : : : : : 26 6.4 Parallel SYNCHEM - Gelernter : : : : : : : : : : : : : : : : : : : : : : : : : : : : 28 6.5 EQUALS: A Parallel Equational Language - Ramakrishnan : : : : : : : : : : : : : 31 6.5.1 Overview : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 31 6.5.2 Planned Enhancements : : : : : : : : : : : : : : : : : : : : : :

Equals is a system for parallel evaluation of lazy functional programs implemented on a Sequent Symmetry. A preliminary implementation of equals [4] was used to establish the validity of Normal Form (NF) demand propagation and memory reclamation via reference counting. However, that implementation did not exploit vertical parallelism, where the arguments to a function can be evaluated in parallel with the function itself. The overheads of vertical parallelism can be as high as the overheads in systems that do not propagate NF demand. Hence careful integration of vertical parallelism with horizontal parallelism (parallelism among the arguments to a function) is needed to fully benefit from NF-demand propagation. In this paper we describe schemes to harness vertical parallelism within the context of the current equals system. We identify the aspects of the runtime system (task management) that affect the overall efficiency of the combined system. We also provide preliminary performance figures indicating the effectiveness of the selected schemes.