EGSnrc
EGSnrc copied to clipboard
nrc-cnrc
Using IPSS Only in Windows Causes egs_chamber to crash
When running simulations that make use of IPSS, egs_chamber (the latest EGSnrc-feature-dynamic-source in this case but also observed in EGS_2016 and EGS_2017 releases), the simulation will crash if correlated sampling is also not enabled.
This issue is specific to Windows in my experience since it doesn't affect our Linux parallel computing cluster.
I've seen this in my chamber model input files but also created a very simple situation to reproduce the issue. A scoring sphere inside of a slightly larger sphere with a point source of electrons hitting the geometry. This simple input still causes egs_chamber to crash if ONLY IPSS is enabled. Once again, enabling correlated sampling (CS) will cause the simulation to run as it should but in situations where only IPSS is desired with no CS, the simulations seem to crash on windows.
Remaking both egs++ and egs_chamber with optimization OFF and disabling the ffast-math flag in the Windows .conf file did not resolve the issue.
Similar behaviour was reported here by one of my collaborators: https://plus.google.com/112946223991473184526/posts/YeanRdGkSYA but we never really found any solution (or the solution for him was to start using Linux only :) )
Thank you for the link! I knew I saw something similar somewhere. I guess my only contribution was to mention that enabling correlated sampling also seems to fix the issue.
Let me check in the next week or so as I'm traveling but I'll definitely let you know!