CGAL users discussion list

[<>]

I am new to CGAL. I have previously used the GTS (GNU Triangular Surfaces
Library) to compute union/difference/intersection on triangulated surfaces. 
GTS
turned out to be very fast indeed but produced instable results as it
computations are not robust or have bugs. It uses doubles with exact
predicates. 
For that reason I am investigating the switch to CGAL to replace our GTS based
code.  I have tested the Join/Difference and Intersection operations on
Nef_Polyhedron_3 and are impressed with the accuracy and correctness of the
algorithms. But unfortunetly there is a major problem with respect to
performance. The binary operations with CGAL are 30 to 60 times SLOWER than 
the
same operations on the same objects with GTS.  My burning question is did I
overlook anything to get the best performance from CGAL?

I am using the following configuration:

I get the best results with the following Kernel definition:

typedef CGAL::Lazy_exact_nt<CGAL::Quotient<CGAL::MP_Float>> NT;
typedef CGAL::Simple_cartesian<NT> Kernel;

I have defined the CGAL_NDEBUG preprocessor flag to turn off any pre/post
conditions or assertions. 

I am using a Release build (Visual Studio 2005) with maximum optimization.

The performance figures I get with intersections/unions/differences on simple
objects like Cubes are around 100mS on a modern 2.4GHz P4 dual core processor.
This seems extremely slow given that in total only 16 vertices are given to 
the
algorithm.

Any suggestions would be very welcome!

Martein Bakker