A fast method for overflow effect analysis in fixed-point systems

Abstract

The refinement of fixed-point systems requires optimizing the word-length (WL) of functional units to prevent overflows and excessive quantization noises while minimizing implementation cost. Applications in image and signal processing domains are tolerant to overflows if their probability is small enough. In this context, determining the number of bits of the integer part of fixed-point data is a trade-off between the implementation cost and the degradation of application quality due to overflows. Classical methods that evaluate the overflow effects using complete simulation are time consuming. In this paper, we propose a new approach to accelerate the simulation of overflow effects and overcome the long execution time for analyzing the application quality criteria. The efficiency of our selective-simulation based approach is illustrated through experiments where simulations are accelerated up to 1000 times with respect to complete simulation based approach.