Compiling Stream-Language Applications to a Reconfigurable Array Processor

New parallel architectures are emerging to meet the increased computational demands of streaming applications. This creates a need for high-level, architecture-independent languages. One such language is StreamIt, designed around the notions of streams and stream transformers, which allows efficient mapping to a variety of architectures. This paper presents our approach of compiling StreamIt applications to the XPP reconfigurable array architecture. We focus mainly on the compiler back end. Although StreamIt exposes the parallelism in the stream program, still a thorough analysis is needed to adapt it to the target architecture. A code generator has been designed for the XPP. It has been demonstrated that by applying optimizations, performance comparable to the low level NML implementation can be achieved. Moreover, the construction of the compiler makes it possible to port StreamIt applications to multiprocessor architectures by doing some architecture specific modifications in the back end.