2024.03.28 Meeting Notes

Agenda

• Individual/group updates
• Future of pmb->par-for https://github.com/parthenon-hpc-lab/parthenon/issues/1034
• On the use of "one-sided MPI"
• review non-WIP PRs

Individual updates

JM

• node/edge/face centered hdf5 IO now almost done (for real!)
• PR ready for testing. Edge and face centered fields work in parthenon-mhd
• Still sth off with xdmf
• Final works on phdf (including cleanup)
• Upcoming new example app for vector advection (to also allow regression test of non-cell centered fields)

LR

• Fixed bug in ownership model for AMR for non-cell centered fields. Already merged.
• Finished Forrest of Tree PR. Ready for final review.
• Additional cleanup PR in the pipeline including MG fixes
• Merge Forrest first, then smaller one to follow
• Up next: more complex base grids

BR

• working on KHARMA release release (still based on a frankenbranch of Parthenon with some backported features)
• working on yt based viz
  • need support for transformed coords
  • support for vectors
  • will open PR once the main parthenon-frontend is merged. PG will follow up.
• looking into TOML parsers for input files

MG

• working on downstream on Phoebus

PB

• wrote Spack recipe -- will upstream

PM

• PR for new version of SwarmPacks (packing particles over blocks)
• New complexity: packing of different types
  • mixed packs currently not supported
• Request: people please look at the interface and comment
  • currently passing in vector of string and type based indices
• Under the hood some cleanup/code dedup could happen (as some code is derived from SparsePacks)
• Otherwise code ready for testing

PG

• Refactor RestartReader (now abstract base class with no hdf5 dependency)
• Looking into Cuda IPC bug

Future of pmb->par-for

• https://github.com/parthenon-hpc-lab/parthenon/issues/1034
• Generic functionality is preferable (as it causes less confusion)
• For now, add shortened global interface
• Eventually deprecate the pmb->par_for
• Potentially introduce a md version (again as generic function that takes an md object) in the future

On the use of "one-sided MPI"

• we've been called out twice now for mentioning "one-sided MPI", which is technically not 100% accurate from the MPI point of view
• JM will update the README anyway and fix the wording along the way

Review non-WIP PRs

• will get reviewed/merged without additional tests (as already used in downstream codes) but opened an issue to keep track
  • interpolation https://github.com/parthenon-hpc-lab/parthenon/pull/1020
  • modification from a new input file on restarts https://github.com/parthenon-hpc-lab/parthenon/pull/1004
  • analysis restarts https://github.com/parthenon-hpc-lab/parthenon/pull/963
• PG will review Forrest of Tree PR and non-cell centered output with emphasis on downstream support
• Hdf5<->Params link will potentially fixed in separate PR so https://github.com/parthenon-hpc-lab/parthenon/pull/1027 can go in now