Automated Mapping from a Domain Specific Language to a Commercial Embedded Multiprocessor

Application specific programmable systems are capable of high performance implementations while remaining flexible enough to support a range of applications. Architects of these systems achieve high performance through domain specific optimizations, often introduced at the expense of programming productivity. We examine one of the most performance critical and time consuming steps to arriving at efficient implementations on these platforms: the mapping of an application to a target architecture. We accelerate this step by constructing a model of the architecture which captures its key features while being amenable to automated mapping. With an analogous representation of the application and mapping formulated as an integer linear programming (ILP) problem, we demonstrate this approach finds solutions that are guaranteed to be close to optimal solutions with respect to this model. These solutions enable efficient implementations of representative network applications on a commercial network processor family. We show this approach can produce an implementation comparable to a hand crafted design.