Implementation Details of Density-Based Algorithms Using Dataflow
Advances in systems analysis, software engineering, and high performance computing book series · 2022
This chapter presents dataflow paradigm in general and different memory types and compiler's optimization constructs like tree reduction as key points for acceleration and discusses implementation details of KNN and k-means density based algorithms on the dataflow accelerators. On-chip and on-board memories allow data to be nearby computational units and thus can provide acceleration. Tree reduction enables higher acceleration by reducing resource allocation that could be used for memory management mechanisms. It is shown how part of an algorithm (calculating distances between neighbor elements or to cluster centers) can be migrated using advanced optimization constructs.