A first shot at an implementation for cgp_parent_data_write. It works in the testing I have done on my local project, but not sure if the "style" is correct. It currently uses an MPI_Allreduce call to determine the total size of the ParentElements and ParentElementsPosition nodes which are created during the function. It might be cleaner to have two functions – one to create the nodes and one to write the data to the nodes, but this implementation follows the behavior of the serial version.
This is in pull request #36