e-ago/hpgmg-cuda-async

GPUDirect Async implementation of HPGMG-FV CUDA

HPGMG-FV CUDA Async

Build

In build.sh script you need to set the GPUDirect Async path. You can download all GPUDirect Async libraries and headers from here GDAsync suite

HPGMG is an HPC benchmarking effort and supercomputing ranking metric based on geometric multigrid methods. NVIDIA reworked the original implementation moving from a CPU implementation to an hybrid CPU-GPU solution. In this repository, starting from the NVIDIA solution, we leverage the communications with GPUDirect Async, recently released by NVIDIA.

For further information about Async and benchmarks, please refer to:

• "GPUDirect Async: exploring GPU synchronous communication techniques for InfiniBand clusters", E. Agostini, D. Rossetti, S. Potluri. Journal of Parallel and Distributed Computing, Vol. 114, Pages 28-45, April 2018
• "Offloading communication control logic in GPU accelerated applications", E. Agostini, D. Rossetti, S. Potluri. Proceedings of the 17th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid’ 17), IEEE Conference Publications, Pages 248-257, Nov 2016