Algorithm Selection: A Quantitative Optimization-Intensive Approach1

Hardware-software codesign has been widely accepted as a backbone of future CAD technology. Implementation platform selection is an important component of hardware-software codesign process which selects, for a given computation, the most suitable implementation platform. In this paper, we study the complementary component of hardware-software codesign, algorithm selection. Given a set of specifications for the targeted application, algorithm selection refers to choosing the most suitable algorithm for a given set of design goals and constraints, among several functionally equivalent alternatives. While implementation platform selection has been recently widely and vigorously studied, the algorithm selection problem has not been studied in CAD domain until now. We first introduce algorithm selection problem, and analyze and classify its degrees of freedom. Next, we demonstrate an extraordinary impact of algorithm selection for achieving high throughput, low cost, and low power implementations. We define the algorithm selection problem formally and prove that throughput and area optimization using algorithm selections are computationally intractable problems. We also introduce an efficient technique for low-bound evaluation of the throughput and cost during algorithm selection and propose a relaxation-based heuristic for throughput optimization. Finally, we present an algorithm for cost optimization using algorithm selection. For the first time, hardware-software codesign is treated using explicit quantitative, optimization intensive methods. The effectiveness of methodology and proposed algorithms is illustrated using real-life examples. 1. Preliminary versions of this work appeared in the Proc. IEEE/ACM ICCAD International Conference on Computer Aided Design, November 1994 and in Proc. 1996 IEEE ICASSP International Conference on Acoustic, Speech, and Signal Processing, June 1995.