CGAL users discussion list

[Sebastien Loriot <>]

I think it will depend on the package/algorithm that is used. For 
example I know that some effort has been made to make Mesh_3
deterministic even when parallelism is used.

For spatial sorting, there is an initial random sampling so no
seeding is used it is not deterministic, even in sequential.

So unfortunately, except without a careful look to how the
parallelism is used in the packages you are using I can't
give you an answer.

Best,

Sebastien.

On 7/23/22 05:23, Niklas Hambüchen ( via cgal-discuss Mailing 
List) wrote:
> Hi,
>
> I'm learning about CGAL's TBB-based parallelism features, such as:
>
> * parallel spatial sorting
> * parallel KD tree creation
> * parallel triangulation creation
>
> What I can't conclude from the docs:
> Do these functions guarantee determinism (scheduling will not affect the 
> results), or even guarantee to give the exact same results as their 
> sequential variants?
>
> I imagine that divide-and-conquer methods like KD trees and sorting 
> should make determinism easy, while for triangulations I can imagine 
> that the order of point insertions will make it hard.
>
> Beyond those specific cases:
> What is CGAL's general stance on parallelism determinism -- is the 
> library's default goal to provide determinism, or can it only be 
> expected if explicitly documented?
>
> I would like to make my use cases faster with parallelism, but only if I 
> do not lose determinism for reproducibility.
>
> Thanks for the info!
> Niklas