Automatically Generating Custom Instruction Set Extensions

General-purpose processors that are utilized as cores are often incapable of achieving the challenging cost, performance, and power demands of high-performance audio, video, and networking applications. To meet these demands, most systems employ a number of hardware accelerators to off-load the computationally demanding portions of the application. As an alternative to this strategy, we examine customizing the computation capabilities of a core processor for a particular application. Our goal is to enable some or all of the computation that is off-loaded to the accelerators to be taken over by the customized core. The computation capabilities of the core processor are extended with hardware in the form of a set custom function units and new instructions. The compiler is responsible for analyzing the target application and identifying a set of cost-effective custom function units. In this paper, we provide an overview of the system that we are developing to automatically identify instruction set extensions and report some preliminary analysis of four media benchmarks.