MPI-supported IO for dumping fields for visualizations about sopht-mpi

[fankiat]

Some note for my future reference: A better way of doing this is probably to derive a MPIO (multi-process I/O) class from the IO class which is already available in sopht-simulator (previously sopht-backend). This way we retain xdmf generation features, and achieve the same metadata structure of the generated hdf5 file, thus assuring consistency with the current working code style in sopht-simulator. I have a rough idea how to do this and will look into this once the MPI-supported 3D kernels and/or simulator are in place.

[fankiat]

Some note for my future reference: A better way of doing this is probably to derive a MPIO (multi-process I/O) class from the IO class which is already available in sopht-simulator (previously sopht-backend). This way we retain xdmf generation features, and achieve the same metadata structure of the generated hdf5 file, thus assuring consistency with the current working code style in sopht-simulator. I have a rough idea how to do this and will look into this once the MPI-supported 3D kernels and/or simulator are in place.

After a more in-depth and careful look, I realize that while a lot of the code in sopht.utils.IO can be reused (with regards to the metadata structure and data organization), there are more subtle changes when it comes to actually performing the write operation in an MPI context. One of the few things is for example, each MPI process carries local domain data, which is a chunk of the entire global array. This means that when writing, the operation needs to be offset accordingly. There are several more issues that are somewhat minor and could be fixed with minor tweaking like generating xdmf file should probably be done by a master_rank and so on. All of these minor changes, while do not affect or change the workflow or pipeline of the IO module, push to me think that we need to instead just rewrite a new module called MPIIO which addresses accordingly all these minor issues, including the ones mentioned above.

In the next few days, I will be writing this new module as well as tests to make sure they work as expected.