Dataflow in MATLAB: Algorithm Acceleration Through Concurrency

In this paper, we present a novel Data-Flow architecture for MATLAB. This architecture provides thread-level pipelining of MATLAB functions as well as general concurrency support. The proposed approach yields a significant speedup of current MATLAB implementations that rely on streaming data or employ data-dependent operations. Following the development of increased CPU core counts, this proposed framework will provide additional benefit only as this trend continues. A performance analysis of the proposed framework is performed, and we are able to demonstrate high-level throughput gains of specific applications. Discussions on implementation guidelines, as well as limitations of the framework, are proposed in this paper. Through the use of this tool, we have demonstrated a 802.11a receiver employing Software-Defined Radio hardware running in real time. From the user's perspective, this tool requires interaction only from the MATLAB language, handling all threading and data transfer without user intervention.