One of the key requirements for the Lattice QCD Application Development as part of the US Exascale Computing Project is performance portability across multiple architectures. Using the Grid C++ expression template as a starting point, we report on the progress made with regards to the Grid GPU offloading strategies. We present both the successes and issues encountered in using CUDA, OpenACC and Just-In-Time compilation. Experimentation and performance on GPUs with a SU(3)×SU(3) streaming test will be reported. We will also report on the challenges of using current OpenMP 4.x for GPU offloading in the same code.
@article{arxiv.1710.09409,
title = {Performance Portability Strategies for Grid C++ Expression Templates},
author = {Peter A. Boyle and M. A. Clark and Carleton DeTar and Meifeng Lin and Verinder Rana and Alejandro Vaquero Avilés-Casco},
journal= {arXiv preprint arXiv:1710.09409},
year = {2018}
}
Comments
8 pages, 4 figures. Talk presented at the 35th International Symposium on Lattice Field Theory, 18-24 June 2017, Granada, Spain