CGAL users discussion list

[Julian Panetta <>]

Dear Clément,

Thanks for the explanation! I'd searched the manual before posting and couldn't find anything about this. If this isn't already documented somewhere, maybe it's worth adding to the "Parallel Algorithms" section?

The extra points indeed cause no harm in my application, but worried me that something had broken.

-Julian

On Mon, Oct 3, 2016 at 10:47 AM, Clément Jamin <> wrote:

Hi Julian,

On Sat, Oct 1, 2016 at 11:07 PM, Julian Panetta <> wrote:

Running make_mesh_3 in parallel gives a similar output to the sequential version, but it's consistently surrounded by 20 disconnected vertices all well outside the domain's bounding sphere (marked with "plus" signs in the attached image).

Of course I can remove these, but it's disconcerting. Any ideas what's happening?

Yes, this is normal. Those points are added to get a better speed-up when parallelism is activated.

Basically, the Delaunay triangulation used by the mesher has an infinite vertex (see here http://doc.cgal.org/latest/Triangulation_3/index.html). A lot of cells are adjacent to this vertex, so there is contention on it when several threads work in parallel on the triangulation. Adding those "far points" reduces contention.

We could remove them after the meshing process, but it is costly. They usually do not harm since they are not adjacent to any cell of the mesh.

Clément.

For example, I see this in the implicit sphere example after pasting:

#define CGAL_LINKED_WITH_TBB

#define CGAL_CONCURRENT_MESH_3

before the CGAL includes (and changing CGAL::Default to K to build on CGAL 4.8.1).

Thanks,

Julian