On Mon, Nov 14, 2011 at 9:09 AM, Robert <libm...@robertweidlich.de> wrote: > Hello, > > I had some trouble with the VTK_IO eating up a lot of memory when I was > trying to output a lot of timesteps (> ~1000). I decided to get rid of > the raw pointers and replaced them with the vtkSmartPointer. Maybe the > libMesh AutoPointer would have worked as well, I haven't tested this. > > As the current code just outputs nodal data, I also removed the custom > solution assembly routine as there are at least problems with the > MONOMIAL basis functions I use for storing gauss-point data. > > VTK supports writing partioned meshes, so I implemented this as well. > VTK seems to need overlapping ghost cells between partitions, but as far > as I have figured it out libMesh just uses one level of ghost nodes. > In Paraviews this leads to internal surfaces, which can be removed by > using the "Group Dataset" and "Merge Blocks" filters. As I just started > to do parallel computations, this is not tested very well, but hopefully > useful anyway. > > Additionally I have added a flag and the correspoding api call for using > the compressed VTK output. > > I attached the patch and uploaded it to this no-paste: >> http://dpaste.de/4uEEA/ > because at least the libmesh-users-list is stripping of attachments.
Hi Robert, Thanks for sending in this patch. It might be a few days before any of the primary libmesh developers has a chance to review it... In the meantime, perhaps you could let us know some specific test cases you were using to test the VTK functionality. -- John ------------------------------------------------------------------------------ RSA(R) Conference 2012 Save $700 by Nov 18 Register now http://p.sf.net/sfu/rsa-sfdev2dev1 _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
