possible bug in vtkRectilinearGrid


I have a simulation which runs on a simple, rectilinear grid (with 8 points in each Cartesian direction). When running in serial, Catalyst outputs the correct information, which can then be visualized in ParaView out of the box:

serial.pvsm (166.1 KB) is the correspondent ParaView state file, whose data is serial.zip (13.2 KB).

However, when running in parallel (with 2 ranks per Cartesian direction, 8 in total), Catalyst seems to output the correct grid data, but when attempting to open the pvtr file series in ParaView (at the moment I am only doing post-mortem analysis), only the part of the grid pertaining to rank 0 is shown:

parallel.pvsm (146.8 KB) is the correspondent ParaView state file, whose data is parallel.zip (61.0 KB). If I take an arbitrary time-step folder and open separately the vtr files, I can see the remaining parts of the grid are there:

In my simulation, all fields are point-based, so having an ‘empty’ cell layer between each subdomain should not be a problem.

Finally, trying to send to Catalyst also the coordinates of the halo points did not help: only the subdomain pertaining to rank 0 is shown (when opening the pvtr series), but the remaining ones are actually defined in the respective vtr files:

separated.pvsm (780.4 KB) is the state file for this separate visualization; the data is halos.zip (64.0 KB).

Does anyone spot an error on my implementation? If not, then there might be a bug within the vtkRectilinearGrid class.

Thank you very much,

Looking at your grid_2.pvtr file from your halos.zip your extents are incorrect. Check out Figure 3.6 in the Catalyst User’s Guide Version 5.8 for how to set it for each process and then also make sure to also set your vtkCPInputDataDescription::SetWholeExtent(). Additionally, even if you only have point data you still won’t want to have an ‘empty’ cell layer between each subdomain because certain filters will require that cell to exist.

1 Like

Thank you very much, Andy. You are right: I did as you mentioned and the issue has been solved.