2026.01.22 Meeting Notes - parthenon-hpc-lab/parthenon GitHub Wiki

• Individual/group updates
• CAAR https://www.olcf.ornl.gov/caar/discovery-caar/
• bitwise identical sims/AMR segfaults
• Review non-WIP PRs

LR

• MG gid (so that internal coarse blocks also get an id)
  • previously gid were dense, now not any more as to include level offset
• splitting comm buffers in sets

CAAR

• solvers would be relevant
  • linking to trilinos?
  • would benefit rad diffusion (FLD) in riot
  • mixed precision in preconditioner?
  • linear solves per cell
• use tensor cores?
  • might be fine for smoothening
  • or local linear solves for initial cycles
• block local timestepping
  • yes, interesting
  • for deep AMR hierarchies
• load balancing
  • probably good combination with block local timestepping and solvers
• (parallel in time integration)

Bitwise identical sims/AMR issues

• follow up from last week: issue persists (in athenaPK)
• discovered that enabling sparse machinery (off by default) make the issue go away
• unclear if its' race condition or lib issue
• PG will provide reproducer

JM will schedule next sync tentatively on 19 Feb