AMA: asynchronous management of accelerators for task-based programming models

Cita com:
hdl:2117/104266
Document typeConference report
Defense date2015
PublisherElsevier
Rights accessOpen Access
Except where otherwise noted, content on this work
is licensed under a Creative Commons license
:
Attribution-NonCommercial-NoDerivs 3.0 Spain
Abstract
Computational science has benefited in the last years from emerging accelerators that increase the performance of scientific simulations, but using these devices hinders the programming task. This paper presents AMA: a set of optimization techniques to efficiently manage multi-accelerator systems. AMA maximizes the overlap of computation and communication in a blocking-free way. Then, we can use such spare time to do other work while waiting for device operations. Implemented on top of a task-based framework, the experimental evaluation of AMA on a quad-GPU node shows that we reach the performance of a hand-tuned native CUDA code, with the advantage of fully hiding the device management. In addition, we obtain up to more than 2x performance speed-up with respect to the original framework implementation.
CitationPlanas, J., Badia, R.M., Ayguadé, E., Labarta, J. AMA: asynchronous management of accelerators for task-based programming models. A: International Conference on Computational Science. "Procedia Computer Science (Vol. 51, 2015)". Reykjavík: Elsevier, 2015, p. 130-139.
ISBN1877-0509
Publisher versionhttp://www.sciencedirect.com/science/article/pii/S1877050915010200
Files | Description | Size | Format | View |
---|---|---|---|---|
1-s2.0-S1877050915010200-main.pdf | 1,275Mb | View/Open |