Speeding up program execution using reconfigurable hardware and a hardware function library

Abstract

This paper describes a co-design environment which follows a new approach for speeding up compute intensive applications. The environment consists of three major components. First, a target architecture consisting of a uniprocessor host and a board with dynamically reconfigurable FPGAs and memory modules; second, a library of functions pre-synthesized for hardware or software implementation; and third, a tool which takes as input an application described in C and partitions it into hardware and software parts at functional granularity using information obtained by profiling the application. An important feature of the partitioning tool is a new efficient heuristic specifically suited for the architecture with reconfigurable hardware.