You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Filters that extract a grid with hidden or refined cells instead of skipping them in place #405
These filters call ghosts.visible() on a dataset with hidden or refined cells. That extracts the other cells into an unstructured copy, with explicit connectivity, stored positions and every field carried. On a structured grid it also loses its structured paths and the caches keyed on its topology and geometry. Every AMR box below the finest level has refined cells, so VTK's box-by-box filters (_by_box) pay this for nearly every box.
filter
what masked cells trigger
clip
extraction, then the table clip
surface_normals
extraction. As the copy is unstructured, consistent=True then runs the union-find rewinding a structured grid skips, and the mesh_kept cache is lost every call
glyphs
extraction of the whole box to drop the points only masked cells use
extract_surface
extraction (single datasets and pieces), then faces derived on the copy
extract_edges, feature_edges
extraction, then edges derived on the copy
triangulate
extraction (2D boxes)
threshold (curved or subdivided fields only)
extraction before split_by_range
Separately, extract_surface._fresh_boundary declines whenever data.ghosts is not None, even when the only marks are duplicates. So any dataset with ghost marks takes the full face derivation instead of the fast boundary.
Proposal. Leave masked cells out in place in each filter's count pass, as threshold's linear path and extract_cells already do through ghosts.without_hidden:
clip, triangulate: no pieces for masked cells;
extract_surface: a face is on the boundary when exactly one of its sides is unmasked;
extract_edges, feature_edges: drop edges whose incident cells are all masked;
glyphs: flag the points some unmasked cell uses;
surface_normals: masked cells add nothing to the point sums.
The output should stay the general path's (which extracts first), up to order. Let _fresh_boundary take duplicate marks.
Found while surveying ghost-driven slow paths for #316 (#401). Contour and slice are done there; the locating filters, tetrahedralize and integrate have issues of their own.
Activity
Filters leave hidden and refined cells out in place
surface_normals keeps masked cells, out of the normals
Merge pull request Kitware#410 from berkgeveci/vis/masked-filters