An Object-Oriented Framewo rkfor Data Pa rallelism J.-M. J equel ez Irisa CNRS Distributed memory supercomputers are well kno wn for being di cult to program. We believe that the object-orien ted framew technologies can be used to drastically reduce this apparent ork complexity. As an example, we describe a framework where the data parallel codes can be encapsulated in object-oriented software components that can be reused, combined and customized with con dence by library designers to o er application programmers easy to use programming models. Additional Key Words and Phrases: data distribution, distributed systems, frameworks, object oriented design, parallel linear algebra library, reuse, softw are componen supercomputing ts, 1. INTRODUCTION Implementing e cient and reliable data parallel applications onto Distributed Memory P arallel ComputersDMPCs at the message passing level is a daunting endea vor. Instead of concentrating on the application design, the programmer has to deal with many tedious and error prone activities: parallelization, distribution, process handling, sync hronization and comm unication management Pancake and Bergmark 1990 . One of the main directions investigated so far consists in the design of semi-automatic parallelizing compilers. How ev building compilers that er, generate e cient code for non trivial
/lp/association-for-computing-machinery/an-object-oriented-framework-for-data-parallelism-FR2RaX0vHN