Skip to content

2023.05.25 Meeting Notes

Jump to bottom
Philipp Grete edited this page Jun 14, 2023 · 3 revisions

Agenda

  • Individual/group updates
  • Report Athena User/Dev Meeting (PG/JD)
  • Next release/versioning (PG) (postponed)
  • Documentation (PG) (postponed)
  • Path forward for packs (LR/JM) (postponed)
  • review non-WIP PRs

Individual/group updates

JM

BP

  • small bugfix for ghost restart
  • testing downstream CT based on local face centered fields
  • KHARMA took part in ALCC hackathon
    • got things working with nvhpc (some new/modern version)
    • identified mostly issues (e.g., around allocation) in KHARMA, less so in Parthenon
    • division are expensive (even in register space)
    • didn't see any machine error
  • KHARMA has been ported to Frontier
    • had to clean up some host function that were marked host-device
    • similarly passing around std::functions

LR

  • working on face centered fields step by step
    • done with CellVar -> Var, ready to merge
    • adding machinery to define shapes for non-cell fields and associated packing
      • one more index for topological elements (e.g., x-face, x1x2 edge etc) based on structs
    • adding bound comm and prolong/restrict (still WIP but almost done and untested)
  • noticed that MeshBlockPack(?) constructor is in a header file, which resulted in massive binary (700+M...)
  • additional PR for AMR/load balancing without too much extra memory allocation

JD

  • looking at unified pack interface in combination with type-variables approach

FG

  • fighting with setup on Frontier
  • looking at long initialization times and reading sbcasted files

PM

  • working on AMR per comparison between Parthenon and AthenaK

TG

  • excited to test face fields in a sandbox

JS

  • focused on results for INCITE (and other) runs

PG

  • "invested" way too much time on trying to get things compile on Frontier with Ascent
    • with latest Ascent develop things are working now with Kokkos 4.0.01
    • Kokkos CMake logic in Parthenon needs update
  • reviewed PRs
  • looked at adding flow chart infrastructure for tasks

Report Athena User/Dev Meeting

  • lots of exciting talks (both astrophysics and development)
  • from a future direction point of iew:
    • in the past there has not been too much interested by "module" developers to push things upstream
    • incentives for dev to push module upstream might not be there yet (even to Athena++)
    • Athena++ is very mature and widely used now. Will also be continued to be used in the future.
    • Parthenon/AthenaPK/AthenaK will likely remain separate efforts (for now) but code sharing is still easy
    • AthenaK focus on results (as the code is mature and working) -- will revisit (for new features) later this year

Non WIP PRs

Clone this wiki locally