some cleanup and switching on energy fluctuations

[WitekPokorski]

This PR enables energy loss fluctuations which are needed to properly reproduce the cell energy distribution. It also cleans one leftover printf. It removes one delete which causes a crash in G4.10.7 (not sure why it works in G4.11, but I guess, there is no problem removing it anyway).

[phsft-bot]

Can one of the admins verify this patch?

[hahnjo]

Note that energy loss fluctuations were previously measured to be significantly slower on the GPU due to added branch divergence, that's why I turned them off by default.

[WitekPokorski]

Indeed, I observe ~25% slow down with them on. It would probably make sense to have the possibility to switch them on/off at run time like in Geant4.

[JuanGonzalezCaminero]

If that's the case, I would make it a compile-time option for now, rather than leaving it always enabled.

I'd do the same for delete fTrackingManager;, it could stay behind some "Legacy compatibility mode", even if it would switch on just this line for now

[agheata]

Yes, please use a configuration option, default OFF for this. Why do you want to protect the deletion of the tracking manager? Do you understand why it crashes with Geant4.10.7?

[WitekPokorski]

Configuration option is not ideal either, because users will need to recompile to compare the two options. We need to be able to switch it on/off at run time, but I guess that requires a change in G4HepEm. I need to reinvestigate the problem with the delete in 10.7. Can't remember anymore where it was coming from. Some difference in the way vector of processes was deleted, or similar.

[WitekPokorski]

OK, I have added the configuration option to switch ON energy loss fluctuations (OFF by default). As for the delete fTrackingManager I haven't found out why it's crashing in 10.7 and not in 11, but I looked at g4hepem/apps/examples/TestEm3/PhysListG4EmTracking.hh and there, there is no deletion of the custom tracking manager. It seems that after doing

G4Electron::Definition()->SetTrackingManager(trackingManager);
G4Positron::Definition()->SetTrackingManager(trackingManager);
G4Gamma::Definition()->SetTrackingManager(trackingManager);

the ownership is transferred to Geant4. Again, I don 't know why it does not crash in G4.11, but it seems that not calling the delete in our physics constructor is the right approach.

[WitekPokorski]

It seems to be that the original design was that the custom tracking manager were deleted by G4VUserPhysicsList::RemoveTrackingManager() method. This works fine in 10.7 and can be seen in the following back trace:

#0  AdePTTrackingManager::~AdePTTrackingManager (this=0x7f3ed027c510, __in_chrg=<optimized out>) at /home/witoldp/lhcb-sim11-g4+cuda/AdePT/src/AdePTTrackingManager.cc:22
#1  0x00007f3f447e3155 in G4VUserPhysicsList::RemoveTrackingManager() () from /home/witoldp/lhcb-sim11-g4+cuda/Geant4/InstallArea/x86_64_v3-el9-gcc12+cuda12_1-opt/lib/Geant4-10.7.3.1/../libG4run.so
#2  0x00007f3f447e3336 in G4VUserPhysicsList::TerminateWorker() () from /home/witoldp/lhcb-sim11-g4+cuda/Geant4/InstallArea/x86_64_v3-el9-gcc12+cuda12_1-opt/lib/Geant4-10.7.3.1/../libG4run.so
#3  0x00007f3f447c2d78 in G4WorkerRunManager::~G4WorkerRunManager() () from /home/witoldp/lhcb-sim11-g4+cuda/Geant4/InstallArea/x86_64_v3-el9-gcc12+cuda12_1-opt/lib/Geant4-10.7.3.1/../libG4run.so

This is why when we try to delete our tracking manager (and we should not) it crashes. However, in G4.11 this line:

physicsList->TerminateWorker();

has disappeared from G4WorkerRunManager::~G4WorkerRunManager(), so G4VUserPhysicsList::RemoveTrackingManager() is not called anymore and when we call the delete it does not crash. I believe, however, that the behaviour in 10.7 was the correct one (hence, we should not call the delete of the tracking manager ourselves), so I am not sure why it was changed in G4.11. @hahnjo any comment on this?

[hahnjo]

It seems to be that the original design was that the custom tracking manager were deleted by G4VUserPhysicsList::RemoveTrackingManager() method. [...]

This is why when we try to delete our tracking manager (and we should not) it crashes. However, in G4.11 this line:

physicsList->TerminateWorker();

has disappeared from G4WorkerRunManager::~G4WorkerRunManager(), so G4VUserPhysicsList::RemoveTrackingManager() is not called anymore and when we call the delete it does not crash. I believe, however, that the behaviour in 10.7 was the correct one (hence, we should not call the delete of the tracking manager ourselves), so I am not sure why it was changed in G4.11. @hahnjo any comment on this?

Thanks for pinging and sorry for the delay in coming back to this with a proper fix after discussing offline. The call to TerminateWorker() was indeed removed in Geant4 11.0 to fix "termination issues" when making the tasking mode the default. The fix was a bit more involved because it required first solving that, but there's now a merge request that will hopefully make it into the December release.