Skip to content

2022.03.03 Meeting Notes

Jump to bottom
Philipp Grete edited this page Mar 3, 2022 · 2 revisions

Agenda

Individual/group updates

BR

  • downstream work on MC rad. in Phoebus
  • two new PRs in Parthenon
    • for multidim. particles (e.g., for multiple species)
    • added PR to add infrastructure for user boundary conditions
  • working on third PR to allow for higher dim arrays (using nx5 and nx6 != 1 for halo exchanges)

JD

  • preparing talk to give tomorrow (audience some comp. phys, some CS, some hw)

FG

  • Will join LANL in Sep
  • Met with GS and Hui Li on plans for the summer
  • Focuses on thesis now, then look at coordinate systems next

PG

  • Submitted paper
  • Updated HIP/Clang CI
  • Discovered Catch2 bug with later GCC releases. Fiddling around with some details but WIP PR is open

Clang/Thrust sort incompat. (https://github.com/lanl/parthenon/issues/647)

  • Not going to fix corner case for a temp. fix (this should be handled by Kokkos::sort eventually)

  • Remove thrust for clang/cuda target

Next release

  • fix sparse restart bug first, then prepare next release
  • target 1.0 release once the paper is accepted

Next priorities

  • Wish list
    • update sparse comm pattern and remove the is isAlloc check and not always allocate comm buffers. LR working on this.
    • Additional solvers, e.g., CG. JD will continue to work on this.
    • Face centered variables.
      • Potentially start without AMR first (based on the cc_in_one)
      • Sparse face fields currently not required
    • Multigrid (port from Athena++?)
    • FFT interface (e.g., heFFTe)
  • Performance
    • make sure pack-in-one is used everywhere
    • also ensure MeshData objects are used everywhere
    • measure load balancing performance wrt current bottleneck
    • split comm. by distance
  • General discussion on allowing interfaces to larger libs (rather than reimpl. algorithms from scratch) without adding restriction in Parthenon itself

Review non-WIP PRs

TODO PG: check coverage test timeout (pot. reduce coverage).

Next meeting in three weeks

Clone this wiki locally