Project

General

Profile

Geant4 EM Physics Code Review Weekly Phone Meeting

Meeting

Date: May 2, 2014 at 9.15 AM US CST (Fermilab local time)
Place: 1-866-740-1260 (ReadyTalk line) (Host : Krzysztof)

Participants:

John Apostolakis
Andrea Dotti
Krzysztof Genser
Soon Yung Jun
Boyana Norris

Discussion

(Summary by Krzysztof)

We discussed Boyana's recently uploaded TAU results:

https://cdcvs.fnal.gov/redmine/projects/g4emreview/wiki/SimpleCalo_performance_analysis

We made the following suggestions regarding the presentation and
further analysis:

  • We felt that most of the data from that page should be presented in
    the report appendix
  • To verify the hypothesis that some of the G4VEnergyLossProcess
    derived classes do little work in the AlongStepDoIt and
    AlongStepGetPhysicalInteractionLength member functions, combine the
    analysis done in the PostStepDoIt case with the one done for
    lightweight functions to find the type of the object of the called
    AlongStepDoIt functions and the number of instructions per call in
    those cases
  • To mitigate TAU features related to the length of the profiling run,
    trigger the profiling in G4RunManager::DoEventLoop
  • To investigate the cache misses in G4VPhysicsVector try profiling
    the G4xydForv10 branch and comparing it to v10
  • Likewise, try changing the type of the G4VPhysicsVector (and
    G4VPhysics2DVector) main containers from double to float (for
    v10.p01)

John will send further suggestions/instruction related to the track/stack handling.

Boyana is going to upload/edit more data today and will send e-mail
when we should take another look at the above page.

Next meeting May 9, 2014