Hello,
I am trying to generate .xdmf to permit ParaView to directly read .h5 output from a parallel CFD simulation. The simulation generates one .h5 file per rank per saved step, including overlap cells.
Writing a postprocessing program that merges a single gigantic output file per step is simple, however the problems with this approach are obvious: The need for large-memory nodes capable of storing an entire step in memory, and even with _4_ ranks the merged output files are approaching 100GB a piece.
After days of searching and experimentation I was eventually able to find an example for how to do this with one file per step (hence the fallback approach of merging), however I have had no luck finding any explanation or example of how to partition a grid across multiple files in xdmf.
This is where I am at, trying to make it read a spatial collection of grids per step. It is marvellously effective at causing segfaults.
[edit: okay hopefully it’ll display this now]
> ```
>
>
>
>
>
>
>
>
> sim_geometry1.h5:/geometry_mesh
>
>
>
> 3D_XYZ_rank0_00.h5:/fluid1/mass
>
>
>
>
>
> sim_geometry2.h5:/geometry_mesh
>
>
>
> 3D_XYZ_rank1_00.h5:/fluid1/mass
>
>
>
>
>
>
>
> ```
So I give up. To posit a simpler scenario: If I have a 32x1x1 grid, and geometryA/massA h5 files contain elements [0 1 2 … 13 14 15 16 17 18 19] and geometryB/massB contain elements [12 13 14 15 16 17 18 … 30 31] - i.e. a partitioned rectilinear grid with ghost cells - what do I write?
It seems like it’s got to be either a spatial collection or a subset. I’ve seen examples of collections that are groups of _different objects_ but this is a single object whose data is stored in multiple files.