Compact DUMP TAS #934
Compact DUMP TAS #934
Conversation
There was a problem hiding this comment.
Very nice! Mostly just comments that I think are needed to be added/updated, plus one potential logic change (the "overwrite" behaviour in case there are multiple tas for one element).
Regarding the inversion routine, I just wanted to remind that this is one of the duplicate routines that would be nice to clean up at some point.
Didn't want to alter the logic at this point. Just replace the storage arrays. I think this should be noted for further updates to the DUMP module. Also, there is no longer a need to restrict the dump to single elements since we're now using compact arrays. But again, logic change should be done in connection with a full review of the DUMP module.
Make an issue? |
|
Regarding invert_tas, I don't know exactly where everything is anymore, so it's better if you do it (or not). I saw it was mentioned (as in moved around) in PR #616 also. |
|
I was having a chat with @amereghe earlier about cleaning up the utilities routines a bit and move them into a set of tools modules. I think we can include this one then. |
|
Thanks! |
This PR replaces the
dumptas,dumptasinv, anddumpcloarrays inDUMPwith an array of structs that is expanded only when there is an element with a dump assigned to it. An integer map points between the element index and the storage array index.This saves quite a bit of memory. For n FCC test with 60 particles it saves 33%, and for a HL-LHC test with 2000 particles it save 24%.