So i have a code that runs just fine in serial and except for a small part it runs fine (and much faster in parallel). Since the part that runs only in serial needs interpolate,SubMesh and a selfmade extracting function that needs the mesh on one process, i thought about running just that part in serial. And well i thought i could just save my mesh with
f = HDF5File(mesh.mpi_comm(),'mesh.h5','w')
f.write(mesh,'mesh')
then go to
if MPI.rank(mpi_comm_world()) == 0:
mesh = Mesh(mpi_comm_self())
f = HDF5File(mesh.mpi_comm(),'mesh.h5','r')
f.read(mesh, 'mesh',False)
#run the whole stuff in serial
#mesh.move is also used here, so the mesh is changed
and after that leave the if-clause and distribute the (locally saved and changed) mesh with
MeshPartitioning.build_distributed_mesh(mesh)
Well the first part, i.e. the writing, works, but the loading just doesn't happen, so i guess i can't do it the way i want to do it, but i don't understand why. And i don't understand how to make that work either.. Well the guess is, that it tries to load the mesh on every process, but since there is the if-clause, only process 0 gets there and the programm halts, but shouldn't mpi_comm_self() force it to load the mesh on just process 0?
Hope someone has an idea, i am thankful for every information and help :)
Best regards
