Fix #382 Phase Space Moving Window #391
Conversation
Due to ComputationalRadiationPhysics/libSplash#127 the interface used did not reflect the new position of the GPU in the output. The interfaced used now saves an MPI_Allgather and corrects that issue.
Adds the offset caused by the moving window for the y-direction. Cropping to the moving window would be better and will follow late on.
|
Please note that the used interface did not reflect the new position as this is its intended behavior. After all, your MPI position/rank did not change. |
|
But isn't the call to
|
|
Unfortunately not. For PDC, everything except the file access type if ignored from FileCreationAttr. This is probably not the best idea but noted in the usermanual.pdf. |
|
See cross post in ComputationalRadiationPhysics/libSplash#127 (comment) - I must have overlooked that this option has no effect for (yet) :) |
| DataCollector::FileCreationAttr fAttr; | ||
| Dimensions mpiPosition( gc.getPosition()[axis_element.first], 0, 0 ); |
There was a problem hiding this comment.
I can merge this but maybe you want to remove setting the mpiPosition beforehand?
There was a problem hiding this comment.
since I am explicitly setting the offsets it won't get wrong when the option gets available I guess.
so we could leave it in for now - makes the function a bit longer but maybe more clear "who is where".
Fix #382 Phase Space Moving Window
Due to ComputationalRadiationPhysics/libSplash#127 the used interface did not reflect the new position
of the GPU in the output. The interface proposed in this pull request saves an
MPI_Allgatherand corrects that issue (see #382).Internal namings have been updated to fit the new conventions from #128 (comment) .
The meta information for the moving window offset has been added (but this pull does not yet add cropping as planned in #89).