search

gandalfcode / gandalf

GANDALF (Graphical Astrophysics code for N-body Dynamics And Lagrangian Fluids)
GNU General Public License v2.0
44 stars 12 forks source link

SEREN format IO for MPI #26

Closed dhubber closed 7 years ago

dhubber commented 10 years ago

Need to write routines for parallel MPI IO for SEREN format

rbooth200 commented 8 years ago

This works properly for the unformatted format. Currently no other format is supported when MPI is used. I guess we can do without the column format when MPI is used, but are there any other formats that we should have?

We should probably also check that the specified output format is sensible at setup.

giovanni-rosotti commented 8 years ago

So I never added the support for the formatted output/column format because it's a pain (you don't know in advance the length in bytes of what the processors before you are going to write) and it doesn't have real use cases (if you are using mpi, it's because it's a big simulation, so why would you want to use formatted output?), so it's not worth the effort. If we do add HDF5 though, it obviously makes sense to make the MPI support it. It's possible that we don't even need to modify anything, as HDF5 is designed for parallel IO.

I do agree that the code should not let you run if the format is not sensible. Also, the userguide should report that with MPI you have a more limited choice of output formats.

giovanni-rosotti commented 8 years ago

See bd50019. I created a new branch for it to contain other error checking; feel free to commit there if you are aware of other clearly wrong combinations of parameters that should not be allowed. Btw, actually the column format does support MPI...

I'll leave this issue open until we have implemented HDF5

giovanni-rosotti commented 7 years ago

Contraddicting my previous statment, I will actually close this thread now and create a separate issue for HDF5