Adaptive Compilation in the Merlin System for Parallel Machines

This paper describes the use of Adaptive Compilation, a system of optimizing code based on information gathered while running the code itself, to match the parallelism of the application to the characteristics of the underlying hardware. 1 Introduction Object oriented programming is not usually associated with high performance computing ( for very good reasons ), yet it is a software model that matches nicely the hardware of distributed memory parallel computers. The Merlin Project[AJ93] is an object oriented operating system designed around the Self language developed at Stanford University. It was designed to run on a variety of computers, including the MS8702[AJ92] ( a 64 node mesh distributed memory prototype built at the Laborat'orio de Sistemas Integr'aveis of the University of S~ao Paulo ) and clusters of networked workstations. The next section will trace de development of pure object oriented languages up to the development of Adaptive Compilation, the key technology for high..