Before, we were using the Mesher_level.h file of the Mesh_2 package,
but the parallel version introduced some changes and a
dependency from Mesh_2 to Mesh_3 that is not wished.
I restored the original Mesher_level.h file in Mesh_2.
Now, we add points on a far spheres by default in the parallel version, but not in the sequential version. So, two macros: CGAL_SEQUENTIAL_MESH_3_ADD_OUTSIDE_POINTS_ON_A_FAR_SPHERE and CGAL_PARALLEL_MESH_3_DO_NOT_ADD_OUTSIDE_POINTS_ON_A_FAR_SPHERE.
There is just one grid lock data structure class now, with several specialization.
Moreover, now the type of the lock data structure is only defined in the triangulation.
Added a better support for parallelism in Triangulation_3.
Simplified how to enable/disable concurrency in Mesh_3.
Moved the Compact_container stategies to a new file.
+ some bugfixes (most of them are related to the parallel version)
This version works (sequential and parallel) as long as the following macros are NOT defined:
* CGAL_COMPACT_MESH_VERTEX_CELL
* CGAL_INTRUSIVE_LIST
* FORCE_STRUCTURAL_FILTERING
That feature can be enabled with g++, clang++, and icpc (intel). It could
be implemented for MSVC, with the flag /Zs, but that is not yet done (and
probably will not).
The option is controlled by a CMake option, CGAL_ENABLE_CHECK_HEADERS, that
is disabled by default. If that option is enabled by the user, then CMake
will check if the compiler $CXX understand the syntax:
$CXX -x c++ -fsyntax-only CGAL/header.h
and send an error otherwise.
Then phony targets will be created:
- a target check_CGAL__header_h for each header <CGAL/header.h>,
- a target check_pkg_<pkg> for each package <pgk>,
- and a target check_headers for the whole CGAL.
Those new targets currently give a lot of compilation errors if
CGAL_ENABLE_CHECK_HEADERS is enabled!
+ fix several missing includes in Mesh_2.
- New worksharing strategy, which groups tasks by batches and let TBB manage the load balancing => the results are slightly improved, compared to parallel_do
- New params in config file
- Function get_number_of_bad_elements in the mesher levels (for debugging)
- option CGAL_MESH_3_ADD_OUTSIDE_POINTS_ON_A_FAR_SPHERE to reduce contention on the infinite vertex
Anyway, now the refinement of the facets is parallel. The speedup is around 2-3 on a 12-core CPU. Can do better, but at least it shows it's not sequential.
The refinement of the cells is not parallel, yet.
One last thing: the program needs to be restarted between two refinements (should be fixed soon).
Clement Jamin
Laurent Rineau