Automated data flow graph partitioning for a hierarchical approach to wordlength optimization

Abstract

Modern automatic analytical methods for studying range and accuracy in fixed-point systems are gradually replacing the traditional bit-true fixed-point simulations used in Word-Length Optimization (WLO) problems. But these models have several limitations that must be overcome if they are going to be used in real world applications. When targeting large systems, the mathematical expressions quickly become too large to be handled in reasonable times by numerical engines. This paper proposes adapting the classical Fiduccia-Mattheyses partitioning algorithm to the WLO domain to automatically generate hierarchical partitions of the systems to quantize. This is the first time this type of algorithms are used for this purpose. The algorithm has been successfully applied to large problems that could not be addressed before. It generates, in the order of minutes, maneuverable sub-problems where state-of-the-art models can be applied. Thus, scalability is achieved and the impact of the problem size as a constraint is minimized. © 2014 Springer International Publishing Switzerland.