Exploiting Vector Computers by Replication
Peter R. Benyon · The Computer Journal · 1985
Two broad classes of problem suitable for computers with vector architecture can be distinguished. The first have a systematic structure that makes them natural candidates for this type of machine. The second lack such structure but the algorithm as a whole has to be run many times. It may then be possible to vectorise across a large number of conceptual replicates of the same algorithm rather than trying to vectorise within the one algorithm. The conditions allowing this are stated. An advantage is that even scalars in the original become long vectors in the replicated version. A disadvantage is the amount of storage needed. The application of replication to a continuous simulation problem is discussed.