Exploiting parallelism by customizing evaluation

Abstract

NiMo is a totally graphic language from the family of Higher Order Typed languages with a strong Data flow inspiration. The interpreter is a specialized graph transformation system, and therefore the language operational semantics is given in terms of graph transformations. In NiMo parallelization is implicit and the evaluation policy is customizable following a process-centered approach. Here we explore some of the methodological possibilities that it opens. Some classical examples illustrate how combining modes greatly increases processor usage, decreases channel population, and achieves subnet synchronization in a very easy and intuitive way. We also present a stream programming technique and a real case application for generative and multistage-programming.