Skip to content
Vasileios Karakasis edited this page Nov 7, 2019 · 16 revisions

Welcome to the Eurohack19 wiki!

This page contains material for the EuroHack19 GPU Hackathon, held from Sep. 30 to Oct. 4, 2019, in Lugano, Switzerland. The goal of the GPU Hackathon is to provide application development teams a conducive environment to jump-start development of accelerator-capable versions of their applications by providing access to the best mentors from the GPU community from national labs, vendors and universities.

Slack workspace

Eurohack19 has a dedicated Slack workspace where we are going to share comments, material etc. during the hackathon. Each team will also have its own slack channel for coordinating. Please make sure to register to the workspace! If you don't have received an invitation, please let me know asap.

Presentations

The teams are expected to do an initial and a final short presentations during the hackathon. Here are more details:

Initial Presentation (Tue, Oct. 1)

  • Send (maximum) 1 paragraph abstract to SLACK general channel (we will use this to update the application descriptions here).
  • Presentation slot is 10 minutes total (make sure to allow for 1-2 minutes for Q&A).
  • 2-minute overview of the application (without mathematics!!)
  • GPU porting approach.
  • Current status: obstacles you encountered, lessons that you would like to share.
  • Plan for next 3 days

Final Presentation (Fri, Oct. 4)

Suggested content for final presentation:

  • Maximum 12 minutes total
  • 1-minute repeat summary of application
  • Explain the starting point (e.g., CPU-code compiled with PGI, CCE)
  • Steps you made porting the code to GPU (e.g., which components ported)
  • Speedup (or -down) if possible (1-2 slides...)
  • Candid feedback about your experiences, with the GPU paradigms, compilers, tools. Suggestions on how to improve the process (e.g., changes to paradigm "XXX would really help us", better documentation, etc.)
  • Any general comments about EuroHack....
  • We need a copy of your presentation to place in this page.

The Teams

BRAIMMU

The code simulates the immune system of the brain during the course of Alzheimer's disease. It uses an equation-based model for simulating the interaction among multiple aspects of the brain, including the neural activity, brain immune system, and the fibrillization and propagation of prion-like proteins. The method comprises a set of partial differential equations with the boundary and initial conditions taken from MRI data. The current version of the code is able to model immune cell chemotaxis, diffusion of macromolecules in the brain, and the neuronal death.

  • Final presentation [pdf]

deal-on-gpu

deal.II is a general purpose finite element library that helps solve a broad range of partial differential equations. In this hackathon, we are basically focussing on a mini-app solving the Poisson’s equation that arises in study of fluid mechanics and plasma physics. We hope to improve our GPU port here with the matrix-free linear systems solvers, more specifically, a custom implementation of preconditioned conjugate gradients.

  • Final presentation [pdf]

KTH-Nek5000

Nek5000 is an open-source code for the simulation of incompressible flow, including low-Mach number approximation. Nek5000 is widely used in a broad range of applications, including the study of thermal hydraulics in nuclear reactor cores, the modeling of ocean currents and the simulation of combustion in mechanical engines.

  • Final presentation [pdf]

Météo-France

Our application forecasts weather. Its code is quite old, very large and changes quickly. We are currently investigating the economic profit we can get by porting our application to GPUs, and the changes we need to make in order to have our code running on GPUs. We do that by extracting standalone test cases from our code, porting them to GPUs and try to compare the TCO of CPU based and GPU based solutions for this test cases. This approach also gives us the opportunity to understand how we should proceed for porting our whole application.

  • Final presentation [pdf]

Octo-Tiger

Octo-Tiger evolves the equations of self-gravitating hydrodynamic fluids using a fully three dimensional AMR grid. It is tailored to the modeling of interacting binary star systems. Octo-Tiger is able to provide much higher resolution than the SPH codes typically used to study binary star systems. We expect, once it is fully optimized and once we add additional physics, that it will become the finite volume code of choice for modeling interacting binary systems in the Newtonian limit.

  • Final presentation [pdf]

Octopus

Octopus is a pseudopotential real-space package aimed at the simulation of the electron-ion dynamics of one-, two-, and three-dimensional systems subject to time-dependent electromagnetic fields. The program is based on time-dependent density-functional theory (TDDFT) in the Kohn-Sham scheme. All quantities are expanded in a regular mesh in real space, and the simulations are performed in real time. The program has been successfully used to calculate linear and non-linear absorption spectra, harmonic spectra, laser induced fragmentation, etc. of a variety of systems, including solids.

  • Final presentation [pdf]

The ICONics

ICON-A is a general circulation model which includes a non-hydrostatic dynamical core (i.e., solver of the fully compressible 3D Euler equations) and numerous physical parameterizations (radiation, turbulence, graupel microphysics, and a land parameterization scheme).

  • Final presentation [pdf]

Utopia

Utopia is a C++ library for parallel non-linear multilevel solution strategies. Utopia combines advantages of high-level programming interfaces with the advantages of domain specific languages by means of expression-templates. On the one hand, it allows using high-level abstractions while providing access to the native low-level data-structures. On the other hand, it facilitates expressing complex numerical procedures by means of few lines of code. This is achieved by separating the model from the computation, thus allowing us to keep the implementation details hidden from the code of applications such as non-linear solution algorithms and finite element assembly. At Eurohack-2019 we aim at porting the Recursive-multilevel-trust-region method (RMTR) to GPGPU architectures using Kokkos and Tpetra.

  • Final presentation [pdf]

VeloxChem

VeloxChem encompasses response theory simulations of general linear(one-photon) electronic spectroscopies at the level of Hartree-Fock and Kohn-Sham DFTemploying standard localized Gaussian basis sets ranging up to angular momentum l=4 (g-functions). In the case of DFT, a set of standard exchange--correlation functionals are provided inthe categories of pure, GGA, hybrid, long-range corrected, and short-range corrected.

  • Final presentation [pdf]

Group Photo

Clone this wiki locally