Before pre-betting table: Name in parenthesis means, proposed by this person. After pre-betting table: Name in parenthesis means taking the lead in shaping. Please involve the right people or distribute the work. ## Blue line - [py2f] Continue granules in blue line work (**Sam**, **Abishek**, Christoph, Daniel) - Enabling multi node runs - Enable integration tests on CI that use GPU and ICON grid. - Testing other experiments, mch production run on Balfrin. - Integrate dycore granule - Verification and debugging approach ## Green line (Magdalena) - [metric fields](https://hackmd.io/zruuvWC1QRuECcy2pD7I5g) (Nikki, Andreas) - [vertical coordinate](https://hackmd.io/mCc0xFwsTdC0Lkh8Aoj7fA) (Magdalena) - JW test case on GPU, datatest on gpu backend (consistent constisten xp = cupy or numpy import throughout the model) (-> maybe Chia Rui with support of Abishek and Sam, they already started with it there) (Chia Rui) - There is a list of smaller things in [this clean up project](https://hackmd.io/@gridtools/ByImPbhJA) among others fixing the Field typing annotations in icon4py. I shortly talked with Nikki about that, if she gets bored she might do it. - grid manager, domain decomp ? (Mauro, Will) ## GT4Py - [GT4Py/DaCe] adhoc: Fast path to Combined IR with DaCe - (Edoardo) Continue work on translator GTIR to DaCe, taking `shift` and `neighbors` builtins as next step. - (Till) domain inference, GTIR passes (TODO: find out what dependencies are between the projects) - (Sara) PAST to GTIR (TODO: availability might not be sufficient) - Minimize the call overhead in `gt4py.next` programs (implement `FrozenStencil`-like feature). (Sam, Rico, ?) - Continue gt4py toolchain cleanup - (Ioannis) More structured-unstructured performance experiments? ## DaCe - [Halo Exchange Optimizations](https://hackmd.io/QtgHmxBAQNeuS06DEBBV2w): Initiated discussions to figure out how to retrieve information about the amount of neighbors accessed per stencil. This info will help us define suitable GHEX patterns (halo layers) and let DaCe decide which pattern is optimal per stencil. - [JaCe - JAX to DaCe translator] (Philip) ------------------------------- ## Available resources: + GridTools: * Christos: 1 week (away from 8th to 21st of May) * Edoardo: full cycle * Enrique: not available (PASC-related tasks) * Hannes: not available (PASC-related tasks) * Philip: full cycle * Rico: full cycle minus May 10th * Sara (ECMWF): 1 week (at the ENES HPC workshop from 21st to 25th of May, bank holidays/annual leave on 6th, 7th, 9th, 10th, 20th, 30th) * Till: 1.5 weeks (away from 6.-7., 20.-24.) + CSCS-EXCLAIM: * Abishek: 2.5 weeks (min) * Magdalena: 1 week * Nikki: all cycle except for 2 days * Sam: 2 weeks (public holidays, vacation) + CSCS: * Andreas: ? * Will: 1.5 weeks + C2SM: * Chia Rui: full cycle * Jonas: ? + MCH: * Christoph: 100% * Daniel (50%): ?