-
Notifications
You must be signed in to change notification settings - Fork 146
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
egs_chamber either produces zeros or exceeds MXSTACK #314
Comments
@ojalaj in the latest version |
Thank you @mainegra for the info/tips. So far I've done the tests in interactive mode, since all the nodes on our cluster are reserved for one long parallel simulation (and practically all the HW resources are thus in use). This means that it might be that the egs_chamber simulation is running out of memory, so I need to wait for until the previous parallel simulation ends until I can really test what you suggested. The simulations I've tested so far crashed in the beginning of the first chunk, so practically no particles were simulated. If the simulation runs out of memory maybe there could be some clear(er) message to the user of that? |
I've now done testing on both clusters ('develop' branch from March and June), setting $EADL_RELAX either .true of .false, but the problem persists, i.e., with MXSTACK below certain threshold two test files without VRTs run OK and the third test file with VRTs fails and when MXSTACK is increased, all test files produce zero results. I've also monitored memory usage with top, but there has always been memory available (at least 0.5 GB). I'll send you the test input files for a try. |
Thanks @ojalaj ! This info is very useful! |
1. Arrays extrafloat_type, extralong_types, extrainttemp, extrafloattemp, could be initialized with zero dimension if the IAEA phsp did not store any addition int (e.g. latch) or float (e.g. zlast) variables for each particle. These arrays are now given fixed dimension MAXEXTRAS. 2. Source had the temerity to actually assign any latch value read from the phsp file to the incident particle, not realizing that, at least in the case of egs_chamber, this variable is used for setting splitting no. Now, latch is set to 0, similar to egs_phsp_source. This partially fixes the problem that @ojalaj had in Issue #314.
Initialize "extra" arrays with a fixed dimension MAXEXTRAS. Arrays extrafloat_type, extralong_types, extrainttemp, and extrafloattemp were initialized with zero dimension when the IAEA phsp did not store additional integers (e.g., latch) or floats (e.g., zlast) for each particle. Set latch to 0, similar to what is done in egs_phsp_source. Source had the temerity to actually assign any latch value read from the phsp file to the incident particle, not realizing that, at least in the case of egs_chamber, this variable is used for setting the splitting number. This partially fixes Issue #314 reported by @ojalaj.
Initialize "extra" arrays with a fixed dimension MAXEXTRAS. Arrays extrafloat_type, extralong_types, extrainttemp, and extrafloattemp were initialized with zero dimension when the IAEA phsp did not store additional integers (e.g., latch) or floats (e.g., zlast) for each particle. Set latch to 0, similar to what is done in egs_phsp_source. Source had the temerity to actually assign any latch value read from the phsp file to the incident particle, not realizing that, at least in the case of egs_chamber, this variable is used for setting the splitting number. This partially fixes Issue #314 reported by @ojalaj.
I've been testing/learning to use egs_chamber. I've used a simple test file having a cylinder water phantom with a thin disc of air, where a collimated beam from a linac (IAEA phsp file) hits. I've also edited the example files for testing purposes - the one with simple Farmer chamber using either no VRTs at all or then using those ones that are utilized in test files (RR/XCSE/IPSS).
I've run simulations on two Linux clusters - one using the
develop
branch from March and the other one using the latestdevelop
branch. In the former cluster I'm able to run the two test files with no VRTs with expected results, but when running the Farmer chamber example with VRTs, I getIn subroutine ANNIH_AT_REST stack size exceeded!
error. I've already previously increased theMXSTACK
to 900000 to get other test files running, but now when I increaseMXSTACK
to 950000, all three different test files run without errors, but produce zero results with 100% uncertainty.What is interesting, is that on the other cluster I needed to decrease
MXSTACK
to 10000 in order to get expected results from first two test files with no VRTs, but with the third test case, I getIn subroutine ANNIH_AT_REST stack size exceeded!
. When I increaseMXSTACK
to 100000, all three different test files run without errors, but produce zero results with 100% uncertainty.So basically there are two problems:
Why there is such behaviour, i.e., with
MXSTACK
values below certain level I get expected results, but after applying VRTs, it needs to be increased, but all test files start producing zero results.Why there is so large difference in the
MXSTACK
"threshold" values between two clusters/develop branch versions?This behaviour has already been reported in G+ forum, but no solutions/explanations were provided.
The text was updated successfully, but these errors were encountered: