CGAL users discussion list

[Samuel Hornus <>]

Hi all,

I'm using the CGAL surface mesher for meshing the surface of proteins.
The results are excellent (see image), but I have a speed issue:

- Linux, CGAL 3.7 : I get a surface with ~3600 vertices in ~2.1 seconds.
- Windows 7, CGAL 3.7, exact same source code : exact same surface
  computed in between 30 and 53 seconds !!

The code uses:

// kernel and cells
typedef
CGAL::Robust_circumcenter_traits_3<CGAL::Exact_predicates_inexact_constructions_kernel>
K; typedef CGAL::Surface_mesh_vertex_base_3<K> Vb; typedef
CGAL::Surface_mesh_cell_base_3<K> Cb; typedef
CGAL::Triangulation_cell_base_with_circumcenter_3<K, Cb>
Cb_with_circumcenter; typedef
CGAL::Triangulation_cell_base_with_info_3<int, K, Cb_with_circumcenter>
Cb_with_info;

// triangulation
typedef CGAL::Triangulation_data_structure_3<Vb, Cb_with_info> Tds;
typedef CGAL::Delaunay_triangulation_3<K, Tds> Tr;

// c2t3
typedef CGAL::Surface_mesh_complex_2_in_triangulation_3<Tr> C2t3;


Both are compiled with -DNDEBUG and -O2 (/O2).

I would like to ask you for suggestions on how this huge gap may be
shrunk ?

There are two things that I haven't checked yet and will check shortly:
        1. check that gmp/mpfr are compiled with /O2 as well
        2. test with a simple Cartesian<double> kernel

Thanks in advance for your feedback,
-- 
Samuel Hornus

Attachment: protein_surface.jpg
Description: JPEG image