Page 1 of 1

Code parallelization

Posted: June 19th, 2020, 6:38 am
by MarcelMeyer
Does a GPU-based code version of Hysplit or some of its components exist?
What are main bottlenecks for code parallelization and how could these be addressed, e.g.: how to address the problem of iteratively loading meteorological data from disk into GPU memory quickly for conducting interpolation and particle advection calculations on the GPU?

Re: Code parallelization

Posted: June 22nd, 2020, 8:41 am
by christopher.loughner
We do not have a GPU version of HYSPLIT. A group in Australia has worked on porting HYSPLIT to a GPU based code. See:

F. Yu, P. Strazdins, J. Henrichs and T. Pugh, "Shared Memory and GPU Parallelization of an Operational Atmospheric Transport and Dispersion Application," 2019 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), Rio de Janeiro, Brazil, 2019, pp. 729-738, doi: 10.1109/IPDPSW.2019.00121.

Re: Code parallelization

Posted: June 22nd, 2020, 9:08 am
by MarcelMeyer
very interesting. I am thinking about trying to build a GPU based interactive particle simulator and this will be a good start. Thanks for the link.