One of the goals of the GPU Hackathons that we organize is to create a collaborative relationship with the participating teams to continue researching and creating success stories that demonstrate the capabilities of Parallware Trainer as an experiential learning tool of parallel programming.
The first edition of the CesgaHack, in the summer of 2017, hosted research teams from different universities in Spain and across a range of disciplines. The teams were working on a range of projects addressing topic such as the prediction of underwater acoustics, real-time prediction of near surface refractivity aiding the understanding of climate near the ground to a study simulation of ichthyoplankton dynamics.
EDANYA, from the University of Malaga, made the most progress during and after the event, learning how to address the parallelization of their C++ code for the prediction of tsunamis (Tsunami-HySEA). The Tsunami-HySEA code has been adopted as the official code for the Spanish and Italian tsunami early warning system as well as for the US National Tsunami Hazard Mitigation Program. As a result of the hackathon and the follow-up work the EDANYA team achieved a speedup of 7.57 using 8 OpenMP threads on a multicore processor and a 5.68 speedup using OpenACC on an accelerator (full details of the results are available in this paper). This work was presented at the Workshop on Education for High-Performance Computing (EduHPC-17) at the SC17 conference – the largest HPC conference in the world – in Denver, USA last November.
The paper demonstrates that Parallware Trainer has the potential to become an effective tool to enable experiential learning of parallel programming and that eases the discovery of the most popular parallel patterns used in scientific applications.
The experience of the EDANYA team at CesgaHack17 showed that the Parallware methodology, based on parallel patterns, is a simple and usable model to provide software parallelism to a team without previous knowledge on parallel programming. The EDANYA team were successful in parallelising a sparse code, by identifying the parallel patterns, and following the Parallware Trainer guidance on how to rewrite sequential code into efficient OpenMP/OpenACC-enabled parallel software. Starting from scratch and in only 3 days, the EDANYA team implemented a scalable OpenMP parallel version of their miniapp, and defined the roadmap for the development of an OpenACC version for NVIDIA GPUs.
Full details are available in the paper here: https://grid.cs.gsu.edu/~tcpp/curriculum/sites/default/files/paper%206_0.pdf
Get involved and try out Parallware Trainer
If you would like to use Parallware Trainer yourself we will be running the next GPU Hackathon in March: the CesgaHack18. The hackathon presents the opportunity tol build collaborations with the Appentra team similar to EDANYA’s, the opportunity to increase your software’s performance and improve your productivity enabling you to create more science and increase the impact of your work..
This year’s Hackathon will be truly international and presented in English. The deadline to register is February 11, 2018: do not miss the opportunity to participate and accelerate the execution of your simulation application.
Register here: www.appentra.com/hackathons/