# Cycle 13 brainstorming

###### tags: `functional cycle 13`

## GT4Py

- Implement for/while loop or scalar operator (bypassing IRs)
- Consolidate Iterator IR:
- Tuple
- Reduce/partial shifts
- move static offset information to IR
- (learning type inference pass)
- Merge functional to main
- setup sphinx documentation for declarative gt4py
- DaCe
- Optimizations:
- Temporary pass heuristic, domain
- Fuse more stencils
- Start using storages module for declarative
- Extend quick start guide with examples of all stable field view features
- Embedded field view
- Unstructured compiled backend integration
- Cleanup exceptions, improve error handling
- Contribute to CI/CD external
- testing for gt4py
- and icon-exclaim working
- **AD: Halo exchange?**

## ICON-exclaim/ICON4Py (need to identify gt4py related projects in shapeup)

- exclaim_gt4py_dycore (high priority)
- Performance analysis
- Probtest aquaplanet experiment starting from restart (time bound, can do without testing APE)
- Establish (non-spack) Jenkins CI for Daint/Tsa
- Assist Christian Zeman with applying [new method] (https://gmd.copernicus.org/articles/15/3183/2022/) to verify experiments (low priority)
- Make `exclaim_gt4py_dycore` work on Balfrin
- Continue working on spack build process
- run mch setup using gt4py dycore for benchmarking performance
- All stencils in GT4Py (high priority)
- remove dusk/dawn
- Merge icon-dsl to icon-exclaim/master
- ICON-liskov (middle to high priority; discuss during shapeup)
- integrate in ICON build system
- fused stencils
- cover all stencils (this would allow plain OpenACC or CPU)
- Performance measurement for fused diffusion (time bound)
- Possibility to trigger icon-exclaim from icon4py and gt4py
- Improvements to bindings generator (what was it about?)
- multiple field operators from program to allow manual splitting fusion
- Improve verification to work for bool fields (mid priority; finish it if bandwidth allows and is time limited)

## EXCLAIM green line

- Continue Diffusion granule (green line) (check list by Magdalena)
- bring together with Will's granule
- Domain science readability refactoring: first session for diffusion (high prio)
- add manual halo exchanges (low priority)
-
- simple driver (high prio)

- Full Dycore granule (mid priority)
- mo_solve_nonhydro
- mo_velocity_advection
- ?
- Tracer advection granule (low priority, don't need it for Q3 objective)