Cartesian Biweekly 2021 03 30

2021/03/30 GT4Py bi-weekly

To Dos

Review To Dos from previous sync

[ ] Will ensure CSCS can compile microphysics code — working on it

Agenda

Problems found with horizontal regions

• Main problem: extents cannot be generalized to the entire IJ plane for horizontal regions in all cases
• +The regions problem
• Differences between GridTools execution model and GT4Py parallel model
  • Tobias presented the summary (link)
  • Enrique: Open discussion with Felix — present in sync meeting with MeteoSwiss?
  • Multiple GT computations was always intended to be implemented in GT4Py.
  • These are frontend-independent changes.
• Open PRs:
  • #367, multi-stage merging changes

Progress on horizontal regions

• Show transportdelp (link)

Status update: FV3core [@Eddie D]

• Show performance plot (link)

CSCS: Status of sprint

• @Johann D - do we need higher dimensional fields PR finished?
  • Yes, needed for for loops.
  • Enrique has been working on stencil call_run.
  • Hopefully finished this week.
• Anton and Felix working on parallel model proposal — perhaps tomorrow
• Rico working on wrapper generator for GTC toolchain.
• Linus working on DaCe backend in GT4Py — naive version.
• Til is developing Python bindings for GHex, also unstructured.
• Hannes looking at performance, and also unstructured backend.
• Langwen has been working a bit with Felix on the GTC CUDA backend.

Status update on GT C++ v1 compilation times [@Tobias W]

• Investigation with Anton
  • Discrepancies between what students and Anton report
  • Need for another study?
• +Compile time issues