Roundtable
Anton
Meet up with supervisor from Helsinki. Maybe schedule a meeting with more people?
Continue with configuration of neural network.
Doga
Dataset is not looking good because too many things we don't understand. But at least some things we understand. Generated a larger dataset to investigate further.
tt-di-higgs had unblinding request.
Plan to wrap up summer student report corrections.
Problems with building ACTS-Athena, because core does not get recognised.
Benjamin
Surprised by the comments on the reports, they seem quite high level only.
Some truth in the comments, but feels like it misses the point. Discussion now is which seems feasible and not of the solution is good.
GSF PRs picked up again.
Preparing the ACAT poster.
Noemi
Feedback collection on tracking cc. Starting to figure out what to ask for. Then we need to prevent a document for cds.
Carlo
Started working on additional c-tuner, but still early stages. Changes in both ACTS in Athena. Quite slow since compiling takes a lot of time.
PR on clustering.
Joana
Mass tolerance did not work because Jacobians were wrong, but does now work. Probably intersectors could be simpler, tried but unit-tests fail.
Alex
Discussions with Johannes on straw fitter.
Disappointed about the deadline extension on the RCS AI proposals.
Andreas Salzburger
Surprised by the deadline extension on the RCS AI proposals.
Please fill Paul's survey on the ACTS workshop.
Stephen
Annoyed by the Saturday morning message for the comments on the report.
Worked on detray RK-stepper. Running out on registers in nvidia GPU spills over. The free Jacobian can trigger this with an 8x8 matrix. Big for the stack not the memory.
Andreas Stefl: Hopefully half of the matrix can be thrown away. Sometimes the matrices are also not symmetric which makes problem.
Alex: Could we do an error-propagation to estimate this effect? -> Could be a reasonable thing to see where the largest numerical errors come in.
Neza
Seeding optimisation over the weekend. Quick look on G50. Now going to debugging traccc.
How to merge projects?
Andreas Salzburger: Start with supporting newest and then go ahead synchronos.
PF
Follow up on strip propagation.
Look at chi2-bumps. We cannot reduce without losing efficiency.
Andreas Stefl
Work on strip seeding and checked counts. Not efficiency tested yet. Start with MR request in Athena and just accept the changes.
Steps for the future with virtual function calls, but need to test first. Expect a speedup of 15%. Currently also depending on auto-vectorisation but it is fragile. Probably compiler dependent.
Andreas Salzburger: Can we add a metric to detect non-vectorisation by compiler in code? -> probably not, since it is not predictable.
Grid based seeding: probably not much more to do to improve the performance.
Eta-bins would be only for filling the grid, since z-binning is meaningless in forward region.
ACAT slides until Thursday.
