This paper presents efforts to improve the hierarchical parallelism of a two scale simulation code. Two methods to improve the GPU parallel performance were developed and compared. The first used the NVIDIA Multi-Process Service and the second moved the entire sub-problem loop into a single kernel using Kokkos hierarchical parallelism and a PackedView data structure. Both approaches improved parallel performance with the second method providing the greatest improvements.
@article{arxiv.2305.07030,
title = {Using Hierarchical Parallelism to Accelerate the Solution of Many Small Partial Differential Equations},
author = {Jacob Merson and Mark S. Shephard},
journal= {arXiv preprint arXiv:2305.07030},
year = {2023}
}
Comments
Originally presented at the HiPar workshop at The International Conference for High Performance Computing, Networking, Storage, and Analysis (2020)