GPU support for averaging #48
Conversation
mean pressure variable named as `mean-E' instead of `mean-p'
src/M2ulPhyS.cpp
Outdated
| @@ -306,7 +306,7 @@ void M2ulPhyS::initVariables() | |||
| if(dim == 3) | |||
| { | |||
| ioData.registerIOVar("/meanSolution","mean-w",3); | |||
| ioData.registerIOVar("/meanSolution","mean-E",4); | |||
| ioData.registerIOVar("/meanSolution","mean-p",4); | |||
There was a problem hiding this comment.
will this impact restart capability for runs we have ongoing that are saving a variable named /meanSolution/mean-E presently?
There was a problem hiding this comment.
I think it will but only if you restart with the option CONTINUE_MEAN_CALC. If that is the case, I guess, one possibility would be to change the name of the variable on the .h5 file directly with a script. If not possible I can work out something else.
There was a problem hiding this comment.
I'm assuming our current runs would want to continue using CONTINUE_MEAN_CALC. So, holding off on landing this - I will need to create a companion script where we can update the name in existing files prior to restarting.
Is this just a naming error, ie, it has been pressure since the beginning?
There was a problem hiding this comment.
Yes, it's just the name. The variable stored is pressure
|
Closing because this is too stale. It is unclear to me exactly what change was intended here. The title and first comment imply that it contains an extention of the averaging on the gpu, but from the commits referenced I don't see it. But... the diff is also very weird, since somehow github thinks 0 files have been changed. Regardless, we are doing averaging on the gpu (as of #67), so at least that part has been handled. The only non-merge, non-style commit here appears to be 4bbb6c7, which fixes a variable name in the mean. Since this PR is too stale to be useful otherwise, I'm closing it and opening an issue about the name (#157). |
With these modifications averaging happens now on GPU with no data movement CPU-GPU.
The tests uses the same reference solutions as those used for the CPU tests. Tests performed with multiple MPI processes are skipped when not enough GPUs are found