Methods, compute nodes, and computer program products are provided for DMA engine assisted local reduction. Embodiments include receiving, by a DMA engine, one or more data descriptors, each descriptor identifying a buffer containing an array for reduction; selecting, in dependence upon the arrays in the buffers and local hardware functional units available to the DMA engine, at least one local hardware functional unit; and reducing one or more arrays in the buffers identified by the data descriptors with the selected local hardware functional unit.
STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
 This invention was made with Government support under Contract No. B554331 awarded by the Department of Energy. The Government has certain rights in this invention.