Automatic type inference for resynthesis on hardware description languages

Abstract

Reconfigurable architectures are object of many studies that try to find tools for a more efficient way of programming, and friendly for software engineers. In spite of the recent advances, there are some applications in which the generated circuits are quite far from optimal. In this paper, we describe a resynthesis methodology, which is suitable for usual multimedia applications. This method transforms the whole algorithm, written in Small Talk, in a graph of LUTs that implements the required operations without the use of library components. The quality of the obtained circuitry is guaranteed by the use of "type inference", a context dependent technique that is the focus of current research on the design of high level circuit generators. Type inference technique comprises two stages: forward and backward data propagation, which allows a high level resynthesis of the algorithm. Thus, our environment automatically optimizes the word length and size of operators, and reduces their internal data paths and switching activity. © 2011 IEEE.