yade-dev team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

> Very good. On my laptop TW.triangulate() + TW.computeVolumes() need 
> 2.7sec for 100k spheres. May be to add this script to a examples?
A script probably needs a few imports and other things, and I didn't 
take time to tune one.
The example in docstring is quite clear I think, but If you feel like 
putting those lines in a script, feel free to commit.


> Does TW.triangulate() need to do on each moment when I need to update 
> a volumes?
I'd say yes, since any movement can invalidate the triangulation (and 
simultaneously, the dual tesselation). Defining Voronoi vertices in an 
invalid triangulation would be inconsistent.

It is possible to move(vertice) in CGAL::triangulation, i.e. update the 
triangulation locally, accounting for the displacement of one vertex 
(here vertex=sphere). In my experience, moving all spheres one by one 
will be way slower than simply erasing the current triangulation and 
building a new one. This is because move(vertex) is just a smart version 
of remove(vertex)+insert(vertex,some_info_from_remove). Even if 
insert(vertex,some_info_from_remove) is faster than insert(vertex), this 
is wasting some time for keeping the triangulation valid after the 
removal operation.
Maybe if you update only 1 or 2 positions in 100k spheres, it can be 
worth a try. The function is actually already wrapped in TW, but not for 
python. You could add a python wrapper like I did for other functions.
move (double x, double y, double z, double rad, unsigned int id);

You still need to recompute voronoi vertices and volumes. For now they 
will be recomputed everywhere, not only in updated cells, which is not 
exactly optimal... It would not be very difficult to implement local 
update but it needs time.

Bruno