[vtkusers] vtkXMLPUnstructuredGridWriter() howto?

[Nico Schlömer]

> PxReader to directly read each file on a given process with MPI so
> your data are already partitioned.

Oh, and how about you work with this data on N processes, and then
want to write it to a file without having to transfer all the data to
one core? I thought that was the general idea: One file per process,
in/out.
My situation is exactly like that: I've got this huge std::vector the
entries of which are distributed over the N processes, and like to do
some I/O with it.

Cheers,
Nico



On Thu, Aug 26, 2010 at 11:10 PM, Sebastien Jourdain
<sebastien.jourdain at kitware.com> wrote:
> I didn't use any writer recently, but PxWriter usually write 1 file
> per piece and a master file somehow. So you can later on, use a
> PxReader to directly read each file on a given process with MPI so
> your data are already partitioned.
>
> Seb
>
> On Thu, Aug 26, 2010 at 11:04 PM, David Doria <daviddoria+vtk at gmail.com> wrote:
>> On Thu, Aug 26, 2010 at 10:52 PM, Sebastien Jourdain
>> <sebastien.jourdain at kitware.com> wrote:
>>>
>>> I think you just need a vtkUnstructuredGrid and set it to the input of
>>> the writer. Specify the number of pieces you want and call Update() on
>>> the writer.
>>>
>>> Seb
>>
>>
>> There was a question about SetNumberOfPieces the other day that never got
>> resolved - Sebastien, could you take a
>> look: http://www.vtk.org/Wiki/VTK/Examples/Cxx/Broken/IO/XMLPImageDataWriter
>> Nico, once we get this issue resolved I'll make the same type of example
>> for vtkXMLPUnstructuredGridWriter.
>> Someone please correct me if I'm wrong, but these P*Writer's don't write in
>> parallel, they simple write data as multiple files (pieces) that can later
>> be used for parallel processing.
>> David
>