visit: Expressions created using Exodus datasets yield wrong results

I met with a user who was trying to define expressions using data read in via the Exodus reader. When they define a simple scalar expression using one of the variables in their dataset, the results are off by several orders of magnitude. I am able to reproduce this on the open side using testdata/exodus_test_data/aircraft.exoII, although their example just had the .e file extension. See the example below:

Define a new scalar mesh variable as follows:

ElementBlock*1.0

where ElementBlock is some variable in the database.

Plotting ElementBlock along with our expression yields the following:

Image

I observed this behavior in VisIt 3.4.2, 3.4.1, and 3.3.3, all on LC toss4 machines.

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 19 (19 by maintainers)

Most upvoted comments

Happy to report that the fix in #20422 works for the data on the closed side.

In looking at .vlog and .vtk files (from running with -debug 5 -dump), the VTK files indicate VisIt thinks the ElementBlock data is double. However, the .vlog files indicate that TransformManager converted from int to float.

I do believe the transform manager is converting from int to float but I think it is somehow getting mislabled as double.

BTW…ElementBlock*ElementBlock/ElementBlock (which is mathematically identical too ElementBlock*1, does consistently work fine. So does sqrt(ElementBlock*ElementBlock)

What I think is happening is that when an integer variable appears as a primal in an expression, it is somehow not getting handled properly by TransformManager (either it is NOT getting transformed or it is not finding the transformed result in the cache correctly).