Dataflow processing of matrices and vectors: Experimental analysis

In this paper we focus on algorithms that extend classical control-flow computation with a dataflow computing paradigm. In particular, our goal is to experimentally explore various dataflow techniques and features, which enable the acceleration of algorithms. One of the most important challenges in designing a dataflow algorithm is to determine the data choreography resulting the running-time performance as good as possible. In the paper, our subject of interest are the algorithms that use matrices and vectors as an underlaying data structure. We discuss several variants of data choreographies for streaming of matrices. We also present the results of experimental evaluation of using these choreographies on several matrix-related problems.