2024.03.14 Meeting Notes

Tentative agenda

• Individual/group updates
• Next release
• Hackathon
• Swarm communication
• review non-WIP PRs

Individual/group updates

LR

• Forrest of Octrees PR is ready for review (currently supporting the same things we can currently support)
  • includes all necessary internal changes
  • one key change is an update to logical location, but there's an interface to get to the legacy ones
  • additional key change is that almost all boundary val infrastructure is gone (apart for particles)
  • removed MeshBlockTree (it's a hash map now)
  • no MG yet
  • should be basis for additional downstream tests
  • ready for review now, other features follow in separate PRs
• also MG performance updates are ready for review
  • should get merged before the one above

PM

• gid in VariableState needs to be updated as the gid may change for AMR
• working on swarm packs based on sparse packs
  • likely biggest roadblock is packing different types
  • may be simplified by type based indices

JM

• digging through IO mine
• revising output for different topological elements
  • also removed legacy output
  • already looking for feedback now
  • should not impact existing postprocessing tools

SJ

• new to the codebase
• working on RIOT downstream

PG

• implemented tracers in AthenaPK ("inspired" by Phoebus' implementation)
• separated CIC interpolation routine and open PR
• progress on OpenPMD output (3D cell based data works)
• yt frontend waiting for final review from yt community

Next release

• MG performance update should go in first
• then release afterwards
• the all the PRs that break downstream/backwards compatibility

Hackathon

• Application submitted
• Attendance is hybrid
• Will discuss more details once accepted and closer to the start

Swarm communication

• Downstream codes currently need to update neighbor block info data manually, see https://github.com/parthenon-hpc-lab/parthenon/issues/1013
• This might be surprising behavior to downstream users (having comm related pieces in non-comm kernels)
• PM and BR are already working on updating the infrastructure more generally and will likely update this along the way

Other comments

• Using a POSIX threads for the tasking currently interferes with using OMPTarget offloading. Should be looked into at some point (kokkos Partition manager as alternative?)
• std::result_of is deprecated in c++17 and we should replace it

Tentative next meeting 28 Mar