Skip to content

2020.05.20 Meeting Notes

Jump to bottom
Andrew Gaspar edited this page May 20, 2020 · 3 revisions

Agenda

  • Individual/group updates
  • Profiling/Performance
  • AMR -> Kokkos
  • Quick run-through of outstanding non-WIP pull request reviews

Goals

From two weeks ago for this meeting:

All of these in service of hydro on a static grid, single GPU

  • Container iterator work will complete
  • Josh will take a crack at coordinate systems
  • Phil will use that as a baseline for the kokkos-ification of parthenon

Next Two Week Goals

  • Discuss offline the best strategy for 1) doing performance tests and 2) automating them
  • Start Profiling Code w/o AMR
    • F. Glines
    • J. Miller (~a little bit)
  • Start Kokkosifying AMR
    • Assignment of execution spaces to mesh blocks (necessary for performance)
      • P. Grete
    • Porting for functionality (just get things not crashing and working correctly)
      • J. Dolence
      • A. Gaspar

Notes

Galen is asking if there are any administrative tasks to take care of currently - nothing comes to mind. Mentions that program managers are very interested in the Parthenon project and are in support of its goals.

Group Updates

LANL CS

J. Brown got the IndexShape change merged. He's investigating adding code coverage, and is looking to add the ability to disable examples.

LANL Physics

Coordinate Systems and Variable Pack are now in master.

Sriram and Joshua fixed a bug in HDF5 output.

Jonah helped with variable packing stuff and been active in reviewing code.

Athena Physics

Last weeks goals were completed. Static grid simulations now run fully on the GPU with MPI. 🎉

Implemented additional regression and convergence tests.

Implemented variable packing and nested parallel-for wrapper to take advantage of scratch space. Might have to change how inline Reconstruction functions work to take advantage of scratch space.

Topics

Profiling/Performance

Related to factor 2x slowdown compared to K-Athena.

First see where we are when using AMR. Tasking is currently not using Cuda streams - we need that for AMR.

Haven't touched the prolongation tasks - communication between levels. Can't simply do index offsets to calculate targets: have to carefully compute indices in the GPU kernel.

Phil will do developer guide on what needs to be done for performance on GPU - there are now sufficient examples in the code to write up a guide.

J. Dolence expressed interest in AMR. Forrest says he'd be interested in helping both.

Would be nice to have performance regression tests.

Clone this wiki locally