2024.01.11 Meeting Notes

Agenda

• Individual/group updates
• Potential Gordon Bell submission
• review non-WIP PRs

Updates

PM

• Fixed bug for using custom prolong/restriction ops for SparsePools (missed metadata for custom ops). Already merged.
• AMR with curvilinear grids using custom ops in new downstream codes seems to work fine.
• Found issue wrt to which variables are prolongated restricted (we talked about this in the past: cons vs prim) and added a workaround
• FG and PM will touch base and the discuss next time how to go forward (wrt to adding a clean solution or providing a workaround/documentation or ...)

LR

• working downstream (alongside with performance improvements for MG)

JD

• addressed comments in refactored interface to performance instrumentation
• refactored/cleaned up tasking (particularly wrt to iterative tasks)
  • builds a graph (rather than while loops with checking for dependencies)
  • graph is now persistent rather than being destroyed and recreate every single timestep
  • support for nested lists (allows for clean separation of outer/inner solvers/lists)
  • RegionalDependencies are gone (used for global syncs) and replaced by "task qualifiers" with support for reductions
  • Task execution is threaded (ThreadPool)
    • BUT GOTCHA rest of Parthenon is not necessarily thread safe (particularly wrt to pack caching)
    • might eventually allow for supporting streams again JM
• started initial refactor of IO interface (to eventually allow for faces and edges)
  • abstracted/separated non-hdf5 specific, generic ops incl flattening of data
• refactoring/debugging global boundary conditions parsing/enrollment
  • but particles boundaries broke (though they'd have to be rewritten soon anyway)
• JM, JD, LR, SM are working on proposal for new downstream code

FG

• curvilinear coordinates for disks works with PLM on static grid
• will sync with PD wrt AMR/prolong/restrict
• working through process to get the code out

JS

• mostly working on physics around numerical relativity
• prolong/restrict for numerical relativity needed special care in AthenaK, e.g., coarse and fine buffers would both have to be communicated (because of higher dimensional operators at higher >4th order)

PG

• actively working on the OpenPMD outputs
• needs some additional interfaces for downstream codes to provide the necessary metadata (such as units)

Potential Gordon Bell submission

• currently don't have all the physics implemented that would support a compelling science case that highlights Parthenon capabilities (e.g., inline reaction with sparse variables)
• will keep this in mind (and revisit in ~2 month) so that we have the capabilities in place for 2025

Next meeting tentative 25 Jan