>> There is a third possibility, which is a compromise:
>> In the case of higher order variables, one could write each component
>> separately as scalar variables, perhaps with a suffix in the name
>> denoting the component. This way, all information is written, while
>> Paraview will still be able to recognize the first order solution as
>> scalars and display it as it does currently.
>
> [...] This actually sounds like a decent improvement, if it
> works for you. [...]
I have implemented this, and it works fine with Paraview. Would you
like me to supply the patch so you (guys) can see what it actually
looks like?
Getting back to the original motivation for this discussion:
Parallelism in VTK output. I have attached a patch which fixes the
issue for me. There are certainly many ways of doing it, mine attempts
to somewhat minimize the amount of modifications of the current code.
Essentially, the solution vectors are localized to one.
Roman
Index: src/mesh/vtk_io.C
===================================================================
--- src/mesh/vtk_io.C (revision 4205)
+++ src/mesh/vtk_io.C (working copy)
@@ -300,17 +300,20 @@
}*/
void VTKIO::solution_to_vtk(const EquationSystems& es, vtkUnstructuredGrid*& grid){
- // write only single processor
- if(libMesh::processor_id()==0){
- const MeshBase& mesh = (MeshBase&)es.get_mesh();
+ const MeshBase& mesh = es.get_mesh();
+ const unsigned int n_nodes = mesh.n_nodes();
- const unsigned int n_nodes = es.get_mesh().n_nodes();
- // loop over the systems
- for(unsigned int i=0;i<es.n_systems();++i){ // for all systems, regardless of whether they are active or not
-
- const System& sys = es.get_system(i);
+ // loop over the systems
+ for(unsigned int i=0;i<es.n_systems();++i){ // for all systems, regardless of whether they are active or not
+ const System& sys = es.get_system(i);
+ std::vector<Number> soln;
+ sys.solution->localize_to_one(soln);
+
+ // write only on one processor
+ if(libMesh::processor_id()==0){
+
const unsigned int n_vars = sys.n_vars();
// loop over variables
for(unsigned int j=0;j<n_vars;++j){
@@ -328,9 +331,9 @@
const unsigned int dof_nr = mesh.node(k).dof_number(i,j,0);
#ifdef LIBMESH_USE_COMPLEX_NUMBERS
- data->SetValue(k,sys.current_solution(dof_nr).real());
+ data->SetValue(k,soln[dof_nr].real());
#else
- data->SetValue(k,sys.current_solution(dof_nr));
+ data->SetValue(k,soln[dof_nr]);
#endif
}
@@ -592,26 +595,34 @@
libmesh_error();
#else
- if (libMesh::processor_id() == 0)
- {
+ vtkXMLPUnstructuredGridWriter* writer = 0;
+ vtkPoints* pnts = 0;
+ vtkCellArray* cells = 0;
+ int * types = 0;
+
+ if (libMesh::processor_id() == 0){
+
// check if the filename extension is pvtu
libmesh_assert(fname.substr(fname.rfind("."),fname.size())==".pvtu");
+
+ const MeshBase& mesh = es.get_mesh();
+
/*
* we only use Unstructured grids
*/
_vtk_grid = vtkUnstructuredGrid::New();
- vtkXMLPUnstructuredGridWriter* writer= vtkXMLPUnstructuredGridWriter::New();
+ writer = vtkXMLPUnstructuredGridWriter::New();
// libMesh::out<<"get points "<<std::endl;
- vtkPoints* pnts = nodes_to_vtk((const MeshBase&)es.get_mesh());
+ pnts = nodes_to_vtk(mesh);
_vtk_grid->SetPoints(pnts);
- int * types = new int[es.get_mesh().n_active_elem()];
+ types = new int[mesh.n_active_elem()];
// libMesh::out<<"get cells"<<std::endl;
- vtkCellArray* cells = cells_to_vtk((const MeshBase&)es.get_mesh(), types);
-
+ cells = cells_to_vtk(mesh, types);
// libMesh::out<<"set cells"<<std::endl;
_vtk_grid->SetCells(types,cells);
+ }
// I'd like to write out meshdata, but this requires some coding, in
// particular, non_const meshdata iterators
@@ -620,12 +631,16 @@
// meshdata_to_vtk(md,_vtk_grid);
// libmesh_assert (soln.size() ==mesh.n_nodes()*names.size());
// libMesh::out<<"write solution"<<std::endl;
+
+ // call solution_to_vtk on all processors to allow localization of the solution in it
solution_to_vtk(es,_vtk_grid);
//#ifdef DEBUG
// if(true) // add some condition here, although maybe it is more sensible to give each vector a flag on whether it is to be written out or not
// system_vectors_to_vtk(es,_vtk_grid);
//#endif
+
+ if (libMesh::processor_id() == 0){
writer->SetInput(_vtk_grid);
writer->SetFileName(fname.c_str());
writer->SetDataModeToAscii();
@@ -636,7 +651,7 @@
cells->Delete();
_vtk_grid->Delete();
writer->Delete();
- }
+ }
#endif
}
------------------------------------------------------------------------------
Free Software Download: Index, Search & Analyze Logs and other IT data in
Real-Time with Splunk. Collect, index and harness all the fast moving IT data
generated by your applications, servers and devices whether physical, virtual
or in the cloud. Deliver compliance at lower cost and gain new business
insights. http://p.sf.net/sfu/splunk-dev2dev
_______________________________________________
Libmesh-devel mailing list
Libmesh-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-devel
