#666: have CarpetIOScalar output grid scalar directly rather than applying reductions to them --------------------------+------------------------------------------------- Reporter: rhaas | Owner: eschnett Type: enhancement | Status: review Priority: minor | Milestone: Component: Carpet | Version: Resolution: | Keywords: --------------------------+-------------------------------------------------
Comment (by rhaas):
I aggree with all statemnets. I don't think that I can remove the MPI communication without changing more lines in CarpetReduce (essentially making the "origin" reduction a special case). This is one of the reasons why I did not propose it initially :-) Right now one can simply abuse eg. the 'maximum' reduction to get timeseries output for scalars that are identical on all processors (also without saving on MPI communication, but since this happens at IO time, I was less concerned about and MPI call).
No, I don't really want to add a fake reduction. Hence the originally proposed the change to how CarpetIOScalar works which admittedly was not that much better. :-)
Yes, the best thing is likely to modify CarpetIOASCII. If you have time to look into it that would be highly appreciated. Please let me know if you need help to test/debug it (I could also give it a try but that will take a while).