stack for at most 100 recursive calls and switch to a non-recursive function
that uses a local std::stack to emulate the call stack.
Benchmarks show the following results:
7 runs on 10,000,000 random in unit circle points
43.67074 : Old implementation
43.91474 : use stack for 100 first call then non-recursive with stack
44.07075 : use stack for 100 first call then non-recursive with vector
44.57079 : Local vector
45.0068 : Local stack
45.87886 : Local fixed size array
46.3829 : Local "home-made" small vector
53.41534 : Thread-local vector
-Wunused-local-typedefs is a new warning flag of gcc-4.7, and it will enabled
by -Wall since gcc-4.8 (not yet released).
The fix is a big set of removals of unused typedefs (or comments, or moves,
depending on the context).
The Constraint_hierarchy is now templated by T, an anonymous type (that is
a Vertex_handle when the constraint hierarchy is used in Ct_plus_2), and a
Compare operator, that have a comparison member function:
bool operator()(T, T);
In Ct_plus_2, the Compare operator passed to the Constraint_hierarchy
compares the vertices with a Less_xy_2, implemented by Less_x_2, and
Less_y_2.
The test test/Triangulation_2/include/CGAL/_test_cls_hierarchy_2.h has been
tweaked, to check that the type T passed in the hierarchy no longer needs
to have a point() member function, as it was before May 2010 (when GF has
determinized the Cdt_plus_2 using too simple patch).
The condition is now to be convertible to the point type, pair<point,info>,...
This was the behavior before the insert-by-range-with-info methods were
introduced.
There was a performance degradation between CGAL-3.7 and CGAL-3.8, when
Triangulation_2 is used with EPEC. This patch fixes the issue. Using a
functor that is specialized for EPEC, in inexact_orientation, to_double is
not called on p.x() but on p.approx().x().
I choose this solution since making the container global variable of
the class penalize classes that do not use remove functions.
The overhead introduced is small (less than 2-3% of the total remove time).
function (not with all compilers, but at least with g++-4.4).
I have a very nasty test case where list_edges.size is more than
700000. With a quadratic function, mmy cruncher computer was not even able
to insert *one* constraint in half a day.
- Remove also the unused variable vb, and the initialization of tempo
(which is re-initialized a second time later, without using its first
value).
| ------------------------------------------------------------------------
| r59165 | afabri | 2010-10-11 15:27:39 +0200 (Mon, 11 Oct 2010) | 1 line
| Changed paths:
| M /trunk/Triangulation_2/include/CGAL/Triangulation_euclidean_traits_xy_3.h
| M /trunk/Triangulation_2/include/CGAL/Triangulation_euclidean_traits_xz_3.h
| M /trunk/Triangulation_2/include/CGAL/Triangulation_euclidean_traits_yz_3.h
|
| Use the renamed class which now is in the Kernel package
| ------------------------------------------------------------------------
| r59159 | afabri | 2010-10-11 15:14:31 +0200 (Mon, 11 Oct 2010) | 1 line
| Changed paths:
| D /trunk/Triangulation_2/include/CGAL/internal/Triangulation_euclidean_traits_projected_3.h
|
| Move implementation of Projection_traits from Triangulation_2 to Kernel
| ------------------------------------------------------------------------
Now the work is in a branch.
Laurent Rineau
Sébastien Loriot
Marc Glisse
Andreas Fabri
Philipp Möller
Menelaos Karavelas
Olivier Devillers