## Cycle 1: availability

Enrique: null

Chia Rui: 1 week

Nikki: - 1 week

Till: - 1 week

Philip: ~70%

Edoardo: -0.5 week

Christos: 80%

Sara: 80%

## Cycle 27 Tasks (28th of January until 25th of February)

* CI/CD (inc. performance monitoring) up and running on Santis and Balfrin with CSCS-CI
* Cycle 1: Icon-exclaim build with spack + uenv & icon4py with uv (pip) (Will, Annika, Jonas, Matthieu)
* Cycle 1: Performance monitoring in icon4py or GT4Py (Ioannis & Christos, Shaping: Till)

* Dycore (+diffusion) granule verification and testing
* Cycle 1: Verification Single and multi node with ITIR + gtfn (Christoph, Daniel, Hannes)
* Cycle 1 or 2: nullptr for optional arguments (Christoph, Daniel, Hannes)
* Cycle 1 or 2: Evaluate reworking granule interface (Christoph, Daniel, Hannes)
* Switch to GTIR + ? as soon as icon4py is updated
* Optimization (with DaCe) (Philip, Christos)
* Cycle 1&2:
* Writing and verifying combined stencil (Magdalena, Yilu)
* Substitute stencils by combined stencils (velocity advection) and test them with DaCe (Magdalena, Nikki)
* One depends on concat where

* GT4Py
* Bring icon4py up to gt4py-main, switch on GTIR (Till)
* Cycle 1: concat_where & DaCe lowering (Till, Edoardo)
* Cycle 1: bring GT4Py to DaCe main (Edoardo)
* Cycle 1: Uncoupling domain and type inference, not expand domain if unnecessary (Sara)

-------------------------------------------

## Quarter tasks backup

* CI/CD (inc. performance monitoring) up and running on Santis and Balfrin with CSCS-CI
* Cycle 1: Icon-exclaim build with spack + uenv & icon4py with uv (pip)
* Cycle 1: Performance monitoring in icon4py
* Cycle 2: Performance monitoring in icon-exclaim
* Cycle 2: cuda-memcheck + bit-identity + memory consumption (define test list) in daily CI for icon-exclaim to robust against nlevp1 and others

* Dycore (+diffusion) granule verification and testing
* Cycle 1: Verification Single and multi node with ITIR + gtfn
* Switch to GTIR + ? as soon as icon4py is updated
* Cycle 1: Writing and verifying combined stencil
* Cycle X: Add serialization to CI and automate
* Optimization (with DaCe)
* Cycle 1&2: Substitute stencils by combined stencils (velocity advection) and test them with DaCe
* One depends on concat where
* Cycle 2: Interface overhead

* Overall performance
* Cycle 2: Evaluate MCH production memory consumption with SEND
* Cycle X > 2: Possible memory optimization
* Cycle X: Passing cuda stream from icon-exclaim to DaCe (or alternative)

* GT4Py

* Bring icon4py up to gt4py-main, switch on GTIR
* Cycle 1: concat_where
* Cycle 1: bring GT4Py to DaCe main
* Cycle 2: Unstructured extend analysis
* Cycle X: Constant folding

* Readability and reliability
* Cycle X: Avoiding usage of field_operators within a program
* Cycle X: Using khalfdim
* Cycle X: rename offset_provider
* Cycle 1 or 2: nullptr for optional arguments ?
* Cycle 1 or 2: Evaluate reworking granule interface
* Cleanup
* Decommission Liskov (CI running with granule + vel. adv. verifying)
* Cleanups in GT4Py and icon4py
*