2020.07.01 Meeting Notes

Agenda

• Individual/group updates
• Discussion of variable packing
• Updated development model (@pgrete)
• Compiler warnings (@pgrete)
• Usage of FORCEINLINE macro (@pgrete: nothing conclusive yet)
• Performance (@pgrete)
• Code Coverage Collection for Regression Tests
• Quick run-through of outstanding non-WIP pull request reviews

Goals

LANL CS Team

• All public Athena names have been removed from interface
• Linting runs in CMake now.
• Debugging Mesh modernization
• Code coverage is working for unit tests now - wants to get some collection

LANL Physics Team

• Streamlined the task types
• Getting LANL code up to date with Parthenon
• Investigation into solvers for physics like Gravity
  • Experimenting with coupling to a typical task list
  • MPI and GPU are working just fine

Athena Team

• CI pipeline fixed
• WIP on Volta runs on IAS system
• Streams/threads/MPI PR - good results! 8 streams across 4 GPUs across 4 processes each use 4 threads
• Jim's been doing some self-education on Kokkos with Athena

Discussion

Variable Packing Performance

Tons of kernel launch overhead on doing kernel launches on a per-variable basis. Really want to launch one kernel for many variables at once - related to buffer packing, etc.

Meeting

Will send meeting to everybody on Monday + Carola.

Updated Development Model

Two branches:

• master
• develop

Release tags

@pgrete will do a write-up on this development model.

Compiler Warnings

General support for requiring -Wall in CI on specific versions of compilers. But not block it in the build.

Kokkos clang-tidy tools are available now: https://github.com/kokkos/llvm-project

FORCEINLINE Investigation

Nothing conclusive yet - a combination of FORCEINLINE and ipo are causing out-of-memory issues when building with Intel compilers

Code Coverage Collection for Regression Tests

Discussion on PR for this: https://github.com/lanl/parthenon/pull/187

MeshBlock Lookup Performance Issue

Andrew will submit targeted PR to improve lookup times