Implementations of Signal Processing Kernels using Stream Virtual Machine for Raw Processor

Stream processing exploits the properties of the stream applications such as parallelism and regularity. DARPA’s Polymorphous Computing Architectures (PCA) program is developing both hardware and software that support stream (and thread) processing with a two-level compiler infrastructure. The Morphware Forum was formed to develop standard software interfaces to promote common interfaces and software reuse [2]. The Stream Virtual Machine (SVM) framework [1] is the intermediate language between the high-level compiler and low-level compiler for streaming applications. We implemented a library-based version of the SVM for the Raw architecture. The Raw SVM library in conjunction with Raw C compiler comprises a functional low-level SVM compiler. We implemented a matrix multiplication and FIR bank kernel using the library and optimization manual optimization to identify performance issues surrounding the SVM.