Open BenWibking opened 2 years ago
Possibly related to: https://github.com/visit-dav/visit/issues/17602?
@BenWibking thanks for the report.
I created a simple mesh example and was about to reproduce your issue.
I did find we were using float32
the coords where we should not be when we hand data to VTK-m, but that change alone isn't sufficient to resolve - we need to trace types though VTK-m to make sure there aren't other cases.
I think there might be code in VTK-m that has the same issue as this VTK code does: https://gitlab.kitware.com/vtk/vtk/-/merge_requests/11235
There is a scaling or numerical precision issue somewhere in the rendering pipeline. Although well within the range of double precision values, rendering an AMR mesh with sides of length 1.5e20 in code units fails completely:
Rescaling the box to have sides of length 1 before passing the mesh to Conduit fixes the problem:
This rescaling should not be necessary if doubles are used throughout the pipeline.