2023.02.02 Meeting Notes

Agenda

Individual/group updates
Status of face centered fields
Parthenon Enhancement Proposal 1 -- Support Subclassing StateDescriptor, please read/skim through https://github.com/parthenon-hpc-lab/parthenon/issues/816 ahead of the sync
Upcoming conferences (e.g., https://www.simonsfoundation.org/event/athena-workshop/ and https://www.icnsmeetings.com/conference/astronum2023/index.html)
Next release
non-WIP PRs

BP

Opening issues
- Many incidental, to be returned to later
- PEP1 is current focus
Moving kharma back to parthenon/develop

JM

Minor bugfixes
HPC system issues
- Chicoma lost cuda-aware MPI. Initial results switching to host-pinned buffers suggest a ~25% performance degradation for phoebus.
- Will collect issues on parthenon wiki

PM

Created a test problem that helped identify a bug in communication across refinement boundaries
- Filled ghost cells with NANs, noticed they weren't being written for beyond-first stages

LR

Identified bug in communication across refinement boundaries
- Boundary prolongation task gets meshblock data, that provides a meshblock pointer, that calls a refinement object, that calls a prolongation operation that itself calls base stage data. Hence higher order stages weren't being updated in ghost cells properly.
- Looks like a first-order-in-time error but may actually be worse than that, especially if one is only calling Update in active zones
Packing meshblocks only works for 1D fields
Added a feature in the riot branch that allows for fewer base mesh blocks than processors (useful for AMR-intensive calculations)

BR

Ensure that XDMF fix PR #796 eventually gets in to code
Several immediate things to do for particles infrastructure
- prefix sums for array indexing operations
- remove inheritance from particle boundaries
- particle I/O (with JM)

PG

Contributing to Ascent support
- Appears to be a convenient way to e.g. extract slices from data at scale
- Usable via runtime YAML files
- Optional external dependency
ORNL increasing use of parthenon because it is easy to get to work
- parthenon-hydro for student winter cluster competition
- adding parthenon to HPC crash course

FG

JS

Working on AMR load balancing in AthenaK
- Would be useful to compare with load balancing in parthenon because the approaches are somewhat different

Talk to Josh due to PR that is a current blocker
- other block-AMR work is currently a timesink

General agreement that this would be a good thing
- BP will create a PR in a week or so if there aren't objections
- Task for BR: Think about how particles would work in this system

Generally a good idea to coordinate parthenon-related talks to minimize repetition if multiple collaborators are at the same conference