John Sargeant

Control of parallelism in the Manchester Dataflow machine

Fine-grain parallel machines, such as tagged-token dataflow machines, allow very high degrees of program parallelism to be exploited for many applications. In fact, so much parallelism can be generated that it is necessary to control parallelism in order to bound store usage.

Flagship: a parallel architecture for declarative programming

The Flagship project aims to produce a computing technology based on the declarative style of programming. A major component of that technology is the design for a parallel machine which can efficiently exploit the implicit parallelism in declarative programs. This paper describes the computational models which expose this implicit parallelism, and outlines an architecture designed to exploit it. The operational issues, such as dynamic load balancing, which arise in such a system are discussed, and the mechanisms being used to evaluate the architecture are described.

Stored data structures on the Manchester dataflow machine

Experience with the Manchester Dataflow Machine has highlighted the importance of efficient handling of stored data structures in a practical parallel machine. It has proved necessary to add a special-purpose structure store to the machine, and this paper describes the role of this structure store and the software which uses it. Some key issues in data structure handling for parallel machines are raised.

Uniting Functional and Object-Oriented Programming

United Functions and Objects (UFO) is a general-purpose, implicitly parallel language designed to allow a wide range of applications to be efficiently implemented on a wide range of parallel machines while minimising the conceptual difficulties for the programmer. To achieve this, it draws on the experience gained in the functional and object-oriented “worlds” and attempts to bring these worlds together in a harmonious fashion.

Exploiting Implicit Parallelism in Functional Programs with SLAM

An effective execution model is a vital component of any general-purpose implicitly parallel programming system. We introduce SLAM (Spreading Load with Active Messages), an execution model which overcomes many of the problems with previous approaches. SLAM is efficient enough to operate at low granularity without hardware support, andhas other necessary properties. Compiling for SLAM presents an unusual set of problems, and we describe how this is done from UFO-Lite, a simplifiedv ersion of the UnitedF unctions andOb jects programming language. Linear speedups are obtained for a program with irregular, fine-grain, parallelism on stock hardware.

Implicit Parallelism: The United Functions and Objects Approach

UFO is a general-purpose, implicitly parallel language designed to allow a wide range of applications to be implemented efficiently on a wide range of parallel machines while minimising the conceptual difficulties for the programmer. To achieve this, it draws on the experience gained in the functional and object-oriented “worlds” and attempts to bring these worlds together in a harmonious fashion.