Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE FIFTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GRID AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi, B.H.V. Topping and G. Várady
Comparison of Intel Xeon Phi Offload Runtimes
L. Maly1,2, J. Zapletal1,2, M. Merta1, L. Riha1 and V. Vondrak1,2
1IT4Innovations, VSB-Technical University of Ostrava, Czech Republic
L. Maly, J. Zapletal, M. Merta, L. Riha, V. Vondrak, "Comparison of Intel Xeon Phi Offload Runtimes", in P. Iványi, B.H.V. Topping, G. Várady, (Editors), "Proceedings of the Fifth International Conference on Parallel, Distributed, Grid and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 1, 2017. doi:10.4203/ccp.111.1
Keywords: Intel Xeon Phi coprocessor, offload, Hetero Stream Library, domain decomposition, HTFETI.
The Intel® Xeon Phi™ coprocessors are representatives of Intel’s Many Integrated Core (MIC) architecture and represent one of the possibilities how to accelerate scientific or engineering codes. There are several runtimes enabling a computation to be offloaded to the coprocessor. In this paper we compare the Intel Language Extension for Offload (LEO) and several approaches utilizing the Hetero Streams Library (hStreams). As a benchmark we use the dense matrix-vector multiplication since it is a crucial part of our method for acceleration of the HTFETI domain decomposition solver Espreso. We provide an overview of the approaches and present results of numerical experiments.
purchase the full-text of this paper (price £22)