Efficient architectures for operators

Abstract

The paper discusses the possibility of high performance computing for restricted classes of algorithms. An analysis of signal processing algorithms has identified a number of common operators which can be implemented efficiently with VLSI technology. The integration of these operators into conventional workstation environments provides the potential for orders of magnitude performance improvements for large numbers of algorithms which are used in linear transform and linear algebra applications. A small set of preferred operators is defined and their implementation is discussed. Simulation results are used to predict the expected performance gains for some of the operators when used in a number of typical algorithms.<>