Arithmetic for vector and parallel computers

摘要

In recent years large scientific computations are more and more carried out on vector and parallel computers. The performance of these machines has made it possible that large simulation programs replace expensive testing of prototypes. Therefore it is important that the computation is not only fast but also reliable. This aspect is usually not covered by computer vendors. A definition of accurate computer arithmetic including vector and matrix arithmetic, which is exactly the field where vector and parallel computers are applied, has been developed in Karlsruhe in the last decades. We show that this kind of accurate arithmetic can also be implemented on vector or parallel computers efficiently.