Generalized methodology for array processor design of real-time systems

Abstract

Many techniques and design tools have been developed for mapping algorithms to array processors. Linear mapping is usually used for regular algorithms. Large and complex problems are not regular by nature and regularization may cause a computational overhead which prevents the ability to meet real-time deadlines. In this paper, a systematic design methodology for mapping partially-regular as well as regular Dependence Graphs is presented. In this approach the set of all optimal solutions is generated under the given constraints. Due to nature of the problem and the tight timing constraints of real-time systems the set of alternative solutions is limited. An image processing example is discussed.