# Brainstorming Cycle 31 08/25

## Blueline: deliver production ready icon-exclaim

- _Fast startup times, deployment_
continue uenv deployment (Rico, Daniel)
- measure startup time once ready
- ~~_feature completness_ (Hannes)~~
- ~~pass CFL back to Fortran~~
- Debug mch experiments OpenACC crash (Christoph)
- _Performance_ (Christoph)
- Blueline performance report
- add more timers in OpenACC (to match the combined programs)
- show the numbers in bencher
- understand DaCe performance
- look at profile including halo exchanges
- look at at-scale runs and check for start-up and other overheads

## Greenline

- halos (Magdalena)
- torus (Mikael)
- _Warm bubble experiment_
- investigate status of microphysics (Will, Chia Rui, is shaping needed?)
- do both versions work with a compiled backend?
- is performance good enough to run it?
- start working on the driver (Magdalena)

## ICON4Py

- merge remaining combined programs and cleanups (Christoph)
- Program customization: (Hannes)
- allow specifying backend options
- allow switching between compile-time and run-time domains
- Benchmarking only in ICON4Py (Magdalena/Philip)
- finish previous benchmarking project
- add necessary utilities to do DaCe measurments in ICON4Py instead of standalone repo

## PMAP

- _Better weak scaling of PMAP-G_
PMAP-GO Cleanup and merge distributed version into main
- enable CI
- _Using GT4Py main_
- Enable compiled programs on field_operators: Merge [#2069](https://github.com/GridTools/gt4py/pull/2069)

## GT4Py / DaCe

- [x] field_operator with multiple output domains (Hannes)
- [x] Containers (Till/Hannes)
- [ ] CI on beverin (Till)
- [x] finish compile-time domains (Till/Sara)
- [ ] ~~_Enable multi-node runs on hundreds of nodes_ (Till)
improving gt4py compilation/caching for this case~~
- [x] DaCe specific: (Philip)
- (First we have to find out why it is slower than GTFN, probably because of `15_to_28`.)
- `concat_where`, especially nested, expressions are lowered in a [not fully optimal way](https://hackmd.io/tZ3BKzwNTlWwv81fW2H2ww#concat_where-Expressions).
- ~~We should do a comparison between the DaCe and GT4Py optimizer, what is also interesting is run some optimization on GTIR level and then go to DaCe.~~
- Inlining
- Improve optimizations to cope with different domain sizes that will appear when we introduce compile time domains.