# MultiXscale General/Common Meeting
## Present
* Alan O'Cais (UB)
* Matej
* Godehard
* Eli
* Rodrigo
* Anirudh
* Bob
* Caspar
* Hassane
* Jean-Noel
* Matthieu
* Lara
* Matteo
* Pedro
* Richard
* Satish
* Thomas
* Enrico

## AGENDA

* Work between pairs
* Lara – Tilen
- Test-suite
- [#131](https://github.com/EESSI/test-suite/pull/131) Add LAMMPS tests to EESSI test-suite
- Feadback on sanity checks
- Also explored extra test that accompanies the new OBMD package
- dev.eessi.io
- development of LAMMPS OBDM package send to lara
- public repo for development of OBDM package
- https://github.com/laraPPr/lammps
- Once everything for dev.eessi.io is set up Tilen can push the code that he wants to test there.
- Am testing the EESSI bot now from that repo to generate an easyconfig that will build lammps with a minimal number of packages and tests the new package
- [#1](https://github.com/laraPPr/lammps/pull/1) first dev of OBDM packages
- No meetings since the first general discussion we had
- communication via e-mail and slack
* Xin – Hassane
- Problem with OpenMPI integration in PyStencils
- Resolved
- Memory management
- Unit tests for this
- Some tools used to inspect this
- Need to check additional functionality
- GPU support
- Will then add documentation and upload online
* Satish – Jean-Noel
- Mainly working with test suite
- Milestone due next week
- Benchmark on two architectures
- Access to Deucalion (A64FX)
- No direct CVMFS right now
- Starting to do and deployments there
- Should get it up there within the next few days
- Memory very low per core (~700MB per core)
- Can perhaps scale down but may lead to issues with accuracy
- May need to reduce cores used per node
- Fallback to RISC-V at BSC
- Works at least on 1 node
- Julian at BSC may be able to help
- One PR already merged - P3M scenario
- Another in the pipeline (ready for review) - LJ case
- Some scalability issues, Jean-Noel working on these
- numpy issue on Genoa, all others pass
- Snellius, Hortense, Vega,...
-
* Pedro/Bob – Matteo
- Matteo working on theoretical side
- Code is now available at https://github.com/multixscale/RR-BGK_SISM_waLBerla
- Try to find the commit from their main repo
- Good time to start looking at CI now
- Anirudh would like to connect with waLBerla regarding integrating things in PyStencils
- Can reach out to CASTIEL2 to try to make a connection
* Other requests from each group?
- Jean-Noel, EESSI and getting specific builds of OpenMPI+CUDA available
- Can use `EESSI-extend` module to help here and then override the used MPI library
- Alan can help here, there is draft documentation on how to do the MPI override
* Additional deliverables
* Due July 5
* Co-design deliverable
- reviewed by Thomas, Kenneth is reviewing now
- Should add something specific on Kokkos for C++, Clang 19 and OpenMP offloading for Fortran (if anyone is using)
- Mathieu has someone starting in Q3 who may be interested in Clang/Fortran
* Technical concept for scalable coupling – draft
- Some graphical input to still include
- Another example for batteries required
- Should we separate out the electrostatics part
- Subsubsection might be enough, small reshuffle
- Missing a section for Hassan's work, need to have a roadmap (to multi-CPU, multi-GPU)
* Deliverable 2.4
* Ready for submission
* CECAM workshop – proposed date 8.4.2025 – 11.4.2025
- Matej will do it, deadline is 10th of July
- Will take place in Llublijana
- Walter and Godehard are also organisers
- Need to gather list of possible speakers
- No hackathon in workshop, this will take place in December instead connected to SLING
- Financial contributions - not strictly necessary, but welcome. Financial contribution will come from common SISSA/SLING node as part of CECAM Flagsip Program (but this node still does not currently exist)
- Could have co-organisation, Juelich node could contribute
- Can be mentioned in the text
* Other matters
* Move OneDrive to UB?
- Works for everyone as previously at USTUTT (online mode), but Rudolf cannot get offline access
- Will talk to him today to see if we can resolve this, if not we may need to consider another option
- Sciebo (via JSC) may be an option, but would need to work with someone at JSC to configure it.
* Next General meeting – end of September
- Need to send a doodle to find a date