Predicting Performance Potential of Modern DSPs

Naji Ghazal, Richard Newton, and Jan Rabaey

Proceedings Design Automation Conference 2000, Los Angeles, June 2000

High-level development tools for digital signal processors (DSPs) remain unable to extract optimal performance from them without the designer's in-depth knowledge of the architecture. In this paper we describe our approach to Retargetable Estimation and show how and why it can be effective in quickly predicting and guiding toward hand-optimized performance of moderns DSPs for a given application described in a high-level language. We also contrast the advantages of this scheme with those of a full-featured optimizing compiler.