Project

General

Profile

Geant4 EM Physics Code Review Weekly Phone Meeting

Meeting

Date: December 20, 2013 at 10.00 AM US CDT (Fermilab local time)
Place: 1-866-740-1260 (ReadyTalk line) (Host : Krzysztof)

Participants:

Krzysztof Genser
Soon Yung Jun
Boyana Norris

Discussion

(Summary by Krzysztof)

Boyana updated TAU instructions:
https://cdcvs.fnal.gov/redmine/projects/g4emreview/wiki/Performance_tools_notes

We discussed last meeting summary:
https://cdcvs.fnal.gov/redmine/projects/g4emreview/wiki/Meeting_20131213

Soon mentioned that while analyzing G4VMultipleScattering::AlongStepDoIt and
AlongStepGetPhysicalInteractionLength he noted many invocations of the "model"
which in this case is predominantly G4UrbanMscModel::SampleScatering

It is hard to say (as in many cases) where most of the time is spent in those
functions. It would be nice to have a tool to tell which lines of the code are
most compute intensive.

Plans:

Boyana will expand the TAU results tables to include exclusive
function data and more results including cache misses.
She will also instrument TAU to look at the "Sample" functions
as well as G4UrbanMscModel and G4UniversalFluctuation::SampleFluctuations
functions and functions which do "nothing", i.e. do not perform any calculations.
The last item may support Soon's finding from
https://cdcvs.fnal.gov/redmine/projects/g4emreview/wiki/Meeting_20131115
where he found that all classes derived from G4VEnergyLossProcess except
G4eIonization and G4IonIonization should be "Discrete" processes and not
"ContinuousDiscrete"

Krzysztof will continue to look at G4UrbanMscModel
(should we inline some of the smaller functions?)

Andrea will look at G4UniversalFluctuation::SampleFluctuations

We will also revisit G4xyd helper class impact.

We should start writing our report at the end of January

Next meetings: 1/10