Dear all,

I successfully compiled FEniCS using intel compilers+mkl using a customized
dorsal. The solver and other parts work fine. But there is a strange
problem with the CSG mesh generator. It cannot mesh CSG with circles. In
the python interface, it results in segfault. In C++ interface, it simply
hangs.
The following is a test example:
------------------------------------------------------------------------------------------

#include <dolfin.h>
using namespace dolfin;
int main()
{
  Rectangle r(0.5, 0.5, 1.5, 1.5);
  Circle c(1, 1, 1);
  Mesh mesh2d(r,1); // This is
fine.

  Mesh mesh2d(c,1); // This hangs for at least
1.5h.
  return 0;
}
-------------------------------------------------------------------------------------------
I explored this a bit with gdb.
The last dolfin call is dolfin/generation/CSGMeshGenerator.cpp:44
   generator.generate(mesh);
It seems to me that the last non-cyclically called line is
include/CGAL/Delaunay_mesher_2.h:255
   clusters_.create_clusters();
which it was never able to pass.
Inside that function, there is a loop which seems to be inifinite.

I have a hunch that this might be caused by intel compilers' handling of
float rounding (hence the rects are ok but circs are not). I am wondering
if anyone has some experience with this or some suggestions. Thanks~

Relevant info:
intel compiler/IMP: version 2013.update5
mkl: 11.0.5.192
cgal: 4.1 (from FEniCS website)
dolfin: 1.2

CGAL is compiled with:
-D CMAKE_BUILD_TYPE:STRING=Debug
-D BUILD_SHARED_LIBS:BOOL=ON
-D CGAL_CXX_FLAGS:STRING="-lstdc++"
-D WITH_CGAL_Core:BOOL=OFF
-D WITH_CGAL_ImageIO:BOOL=OFF
-D WITH_CGAL_Qt3:BOOL=OFF
-D WITH_CGAL_Qt4:BOOL=OFF
-D CMAKE_CXX_COMPILER:FILEPATH='mpicxx'
-D CMAKE_C_COMPILER:FILEPATH='mpicc
-D Boost_USE_MULTITHREADED:BOOL=${BOOST_USE_MULTITHREADED}

Best regards,
Larry.
------------------------------------
Lizao Li
Graduate Assistant
School of Mathematics
University of Minnesota, Twin Cities
_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support

Reply via email to