2024.06.06 Meeting Notes

Agenda

JD

PM

particles with SMR work (AMR not yet as redistributing on remeshing not implemented yet)
discovered that boundary conditions were not applied on all faces for <3D problems
- got fixed
discovered that particles can be initialized off mesh(blocks) in pgens which first fails on first boundary exchange
- basic assumption is that particles are mesh tied so should not live outside and we should catch
- PM will implement

LR

Finished "fine" fields implementation including IO and AMR. Ready for final review.
Added new advection example with sparse packs and new update machinery (which should eventually move into interface folder)
- also includes restart regression test
update machinery now includes more general update like stokes (rather than flux divergence)
- include wave vector regression test in a CT test case
Next: updating wave vector example to use sparse packs (to eventually replace sparse advection)
- need to make sure that new test case has enough coverage
- potentially split example in actual examples and regression tests
thinking about formalizing block partitioning (also with regard to multigrid)
- will open issue to keep track of thoughts

BP

PG

Spent quite a bit of time tracking down regressions while reviewing the mesh refactor
- regressions turned out to be independent of that PR
- derefinement counter for restarts got fixed by LR
- still tracking down in asymmetric mesh construction with SMR at INCITE scale
AthenaPK/Parthenon will take place in JUREAP (early user program including benchmarking for JUPITER, which is going to be the first exascale system in Europe)
Got invited for a AthenaPK/Parthenon(-hydro) tutorial at the "Towards exascale-ready astrophysics" meeting in Sep
- accepted (in order to get a deadline also wrt to the OLCF tutorial session mentioned a couple of weeks ago)
Workstation for CI with Nvidia and AMD GPUs finally arrived.
- Software seems to be running
- will transition CI soon
Still working on OpenPMD output (though the POC is already running including restarts at 6000^3)

Our custom machine files machinery could probably be replaced by using CMAKE presets (starting 3.19)
Sticking to standard probably good idea
but no strong opinions
people seem to not use machine files that broadly
alternative, bash scripts similar to warpx or KHARMA
currently not high priority, so will open an (good first) issue to keep track

z curve seems to work well for hydro et al
timer based load balancing still WIP
for other application might not be the best choice like "transport sweeps"
question:
- should "specific" changes go "upstream"?
- should space filling curve be a downstream callback
Take away: go for an interface that allows callbacks and provide "standard" options upstream (similar to the prolongation/restriction machinery)

Add a vector wave equation example #1091
- got approvals, but needs fine field
Updates to DataCollection #1092 (also depends on fine field): PG will review (after fine fields)
Refactor Mesh Constructors #1055 PG will review finalize review by end of the week.
Add fine field option #991 PG will review new changes
Implement general 2D forest meshes #1047 : LR will find second reviewer from LANL