AdePT/Celeritas Monthly Standup

Tuesday 30 Jan 2024, 17:00 → 18:00 Europe/Zurich

30-jan-notes.md

1. Setting up timeslot, timetable for meetings
   1. Agreed to move timeslot to Wednesday at 4pm/5pm/11am GMT/CET/EST
   2. Will schedule meetings every two weeks, provisionally alternating between on-demand technical/hackthon and general standup. Will review as this progresses
2. Communication/Organisation
   1. Use HepGPU mailing list/egroup - Ben and Witek admins (Ben to review members, send round a test mail)
   2. Use Celer-AdePT Mattermost channel for other day-to-day communication (noted that there is also the Celeritas Slack channel)
   3. Minute meetings to capture action items, but where relevant these should become Issues on the relevant GitHub/GitLab/JIRA board
3. Standup
   1. Publication of report from GPU R&D panel expected soon as it is in final corrections stage.
   2. AdePT
      1. VecGeom is close to implementing all shapes in the surface model needed for CMS Run2 geometry. Final integration/testing loading of GDML underway, with further debugging and profiling to follow.
      2. Pull Request in progress reorganizing AdePT examples into a library to ease the integration in Celer-AdePT frontend. Need to understand and resolve some linking issue on CI machine, and package it for use in downstream packages.
   3. Celeritas
      1. Started work on implementing optical physics/transport, Cerenkov generator complete. Aim is not like Opticks etc, no use of Nvidia only features like RTX, so can work on AMD etc as well.
      2. Compile-time option to switch between CLHEP/CGS unit systems, e.g. always use CLHEP when using Geant4, with no changes to code so numbers/units are correct.
      3. Report submitted to GPU R&D panel being refined into a paper
   4. ExaTEPP
      1. Elliot plans to attend Hartree Exascale workshop in May
      2. Opportunity to hold hackathons in UK this year - cannot pay travel, but accomodation for extended periods possible.
4. Next Steps
   1. Common Celer-AdePT frontend remains first major milestone, so will review progress on AdePT library PR and maybe hold meeting in two weeks as hackthon on getting this into Celer-AdePT.
   2. Geometry conversion also felt to be key item, especially in-memory Geant4-VecGeom and vice versa conversion including GPU-enabled VecGeom. Code originally from G4VecGeomNav project, adapted and updated in Celeritas. Project to explore factoring this out of Celeritas so it can be shared/used by AdePT started
   3. Discussion on common parameters for host/device problem configuration such as field maps, etc. Felt this is best explored when Celer-AdePT has AdePT offload implemented as that will illuminate what each requires/uses and development of a common specification.

17:00 → 17:10 General Matters

Placeholder for general topics