CGAL users discussion list

[Sylvain Pion <>]

Alejandro Aragon a écrit :
> By the way, talking about points in CGAL. I decided to create my own 
> point class, templated by the dimension of the point to see the 
> difference in creation time with that of the CGAL types. I introduced an 
> implicit conversion so that I can use my point type where a CGAL 
> algorithm needs it. Then my test was just creating points. It was 
> surprising the time that it took the creation of 10e9 Point_2 objects 
> compared to my simple templated point class:
>
> int main(int argc, char *argv[]) {
>
>     typedef CGAL::Point_2 PointType;
>
>     Timer t1;
>     cout<<"CGAL point"<<endl;
>     t1.Reset();
>     for (long i=0; i<1.0e9; ++i) {
>         PointType p(3,5);
>     }
>     t1.Tac();
>     cout<<"my point"<<endl;
>     t1.Reset();
>     for (long i=0; i<1.0e9; ++i) {
>         geometry::Point<2> p(3,5);
>     }
>     t1.Tac();
>
>     exit(0);
> }
>
> $./a.out
> CGAL point
> Time: 74.2738sec
> Measurement granularity: 1000000 of a second
> my point
> Time: 1.57362sec
> Measurement granularity: 1000000 of a second
>
> so can someone tell me why is this? I would never expect an almost 
> factor of 50 in execution times! and this was just for creating a point.

Cartesian<> uses reference-counting, which obviously shows-up
in such a synthetic benchmark.
Try to compare with Simple_cartesian maybe?

--
Sylvain Pion
INRIA Sophia-Antipolis
Geometrica Project-Team
CGAL, http://cgal.org/