material interface filter counts fragments incorrectly on large non overlapping AMR data
From Bucky:
We have been very successful using the Material Interface Filter, on grid-based datasets, for counting fragments of a fractured shell, in all versions including v5.5.0. A very peculiar thing happens when the non-overlapping AMR dataset is large: the filter completely fails to count the fragments. That is, the result of applying the filter is four (4) fragments, when there should by of order thousands. The filtering result on two datasets is given in the table below. The first, labeled MR is for a medium-resolution case, and the second labeled HR is a high-resolution case (both are the same calculation, with different grid resolution). These datasets are available on the LANL yellow net (but cannot be transferred away, at least right now). We remain hopeful that paraview can correctly count the fragments in the HR case. There will be future cases that are even larger. If there are any preference switches that can be tried, or other build configurations, we can test them very quickly. Thanks much, Bucky K.
++++++++++++++++++++++++++++++++++++++++++++++++++++++
MR HR
14,315,904 cells 107,868,672 cells 15,228,928 points 114,216,960 points 6.3e+2 MB 4.7e+3 MB 128 files 768 files 2096 fragments 4 (four) fragments
(the datafiles are created one each per PE, using the paraview XMLHierarchicalBoxDataWriter; observing isosurfaces on both datasets suggests that the HR case should have >3000 fragments.)