This pull request adds a few modifications to fix scaling issues observed on 16k/32k MPI ranks on NERSC Cori system
Use MPI_AllReduce to exchange boxes instead of send/recv of a large number of boxes
Add option to determine unique nodes using global node ID instead of coincident coordinate search. Can be disabled using the CMake option -DTIOGA_HAS_NODEGID=OFF
This pull request adds a few modifications to fix scaling issues observed on 16k/32k MPI ranks on NERSC Cori system
MPI_AllReduce
to exchange boxes instead of send/recv of a large number of boxes-DTIOGA_HAS_NODEGID=OFF