2026 04 23 meeting minutes

Attendees:

General Topics

• 0.1 release
  • Milestone overdue by 1 month
  • A few remaining items, but we should do a release now
  • Some key additions:
    • MPI-backend exposed in the core API
    • Refactored Requests: shared API between backends, easier to write CommSpace-generic code (doc up-to-date)
    • Refactored Communicators: shared API between backends, new capabilities: split, duplicate (doc up-to-date)
    • DeepCopy packing strategy can now handle Views of arbitrary rank
    • Collectives don't have limitations on View ranks anymore (doc specifies semantics)
    • Cleaned up Kokkos Comm traits (doc up-to-date)
  • TODOs
    • Test some stuff on El Dorado, just to know what works and what doesn't
    • Merge #229
    • Tag a commit (Gabriel)
      • Make a GitHub release out of it (Gabriel)

Changelog (post 0.1)

• changelog.md, modified when we merge a PR
  • When you open a PR, you get a number -> add a commit to your PR that adds the PR to the changelog

• Items postponed to 0.2 (end of June, PASC):
  • static_assert on GPU buffers in MPICH / fix El Dorado MPICH detection
  • Non-GPU-aware MPI host staging (Gabriel)
    • critical to have before PASC
  • Spack package recipe (Cedric)
    • better to have 0.1 released anyway before writing the recipe
  • NCCL and MPI results from the same code (Gabriel)
    • Nicole has the Heat3D benchmark in her fork -> open a PR
  • Blog post (Carl)
  • Fixing documentation (Carl)
    • Cedric opened #201
    • Things have been partially addressed in multiple PRs that have been merged recently
  • miniFE port (Carl)
  • Lulesh port (Gabriel + minions)

Carried forward from last time

• Updates on CI dev container? (Cedric)

  • Waiting for build system rework from Gabriel

• MPICH (via Spack) + CudaSpace views, exclusive scan fails

  • SNL CI only tests Open MPI + CUDA
    • Carl will make a container
  • Not sure if it's an MPICH problem or something else we're doing wrong
  • Gabriel is splitting up the unit test

Round Table

• Carl

  • low-level API port of MiniFE is done, think about high-level next

• Evan

  • stream triggered MPI + Kokkos Comm @ SNL this summer
    • compiles on Tuolumne, at least one test not working

• Nicole

  • no specific plans for the internship yet

• Gabriel

  • Planning to engage with RCCL / NCCL generic layer external contribution
  • Host staging for PASC