General
Mind Connecting the dots deadlines.
Meeting in 2 weeks will be on a Tuesday.
Roundtable
Paul
hepMC conversion. Looking at the output now. Improvig the testbed. SLight chances are expected, since we are also changeing the input.
Event recording also changed to make hepMC more of an EDM since event recording isn't deisgned for that.
Generic detector can now built partially (pixels) with Gen3. Strips should be straight forward. Passive surfaces are not fully supported yet. Idea already there but not sure how navigation policy would work together.
Plan: Take Gen1 description and pipe it in the new construction.
Problem with passive surfaces: Gen1 had them defined at layers and looked up like that. Now without layers it needs to be considered in volume. Just needs some thinking.
Spack: Maybe we can merge this week, since everything already green in the CI.
Difficult to have a properly tested built documentation. Spack might break on any update.
Benjamin: Can we test the chain of shell commands in the documentation? We already ran into issues.
Alex: Pull the script into docs? -> Difficult because the built script makes some jumps to figure out compiler and stuff
We might need to get rid of the documentation as we have it right now, since it is not supported anymore. So there will be major changes in the future.
Andi updated svg with neighbouring bins for surface array. Seems to work out of the box for Gen3.
Andreas Salzburger
Proto-axis needs a followup.
Working on clean-up
Anton
Finished installing all ACTS dependencies.
Neza
Midcycle review lat week: Not many comments.
Can we integrate traccc in ACTS? Mainly people want to call it from inside ACTS. Probably enough, to have it as plugin, since namespace does not seem crucial.
Discussions with Ben on the report. Signature plots are removed.
Submitted abstract for connecting the dots.
Dimitra
Analysis: presented new results. Worked fine.
Gen3 integration with Johannes: Good to have a mockup in ACTS. Helps to test the navigation policy. (New navigation policy is necessary).
Alex
GX2F: Fixed a Athena issue, now compatible again.
Now efficiency is comparable to KF.
https://github.com/acts-project/acts/pull/3889
https://acts-herald.app.cern.ch/view/acts-project/acts/2876608037/html/trackfitting_gx2f_vs_kf/performance_trackfitting.html
Benjamin
Mid cycle review: Found reasons for in-efficiencies in single muons. Setup with Gen1 Itk with Athena csv dump. Also inefficiencies in transition regions. Still trying to figure out which to tackle first for the largest benefit.
Trying to merge GNN for ONNX. We can remove pytorch from the workflow.
Big issue to get memory to ONNX modules with out having them on the GPU before. Documentation is quite nice on how to set tensor with cuda pointer.
Latency of network inference seems to be an issue. Always GPU executor and arbitrary number of device executors. (These need to be linked.) CUDA executor has fewer dependencies, but is less performant than more specific approaches.
Outcome from a workshop: most people are happy with CPU execution.
There is no principle blocker to bring this chain to Athena.
Doga
DPG in Göttingen: Interesting talks in cosmology.
Got a summer student and will arrive 9.6.
Plans:
- HTTD truth association. Already a list of problems.
- Tackle comments on QP int note
PF
Midcycle review: Polishing the plots.
Tried to bring some vectorisation to ACTS. Specifically the duplets in the seeding. Did not yield real improvement.
Many early exits in duplets due to radius incompatibilities.
If there is time, interested to investigate how it works on the triplets.
Andreas Stefl
Thesis writing.
