[Paraview] Structured Grid Extent trouble

[Biddiscombe, John A.]

Berk

>What is your pipeline?

H5Part->ParticlePartitionFilter->                   -> SPHProbeFilter(2 input)
                               ->SamplingGridFilter-^

The probefilter takes 2 inputs, one is the partitioned particles, the other is the sampling grid.
The partitionfilter replaces the extent translator with a PieceToBounds translator which maps piece number to 3D regions (from the spatial decomposition), the sampling grid filter creates a rectilinear grid, but divides its output by piece based on the 3D region which it gets from the ExtentTranslator supplied on the input, the grid is therefore created on the same process that the particles exist on. When the SPHprobeFilter does its probing, the particles(+ghosts particles) on process N are available and so are the probe sampling points - this differs from the standard probe filter because we don't duplicate the probe locations on each process or anything (if we had some arbitrary probe locations from another dataset, then I'd need to add a RepartitionFilter which redistributes the probe geometry based on the PieceToBounds Translator so that the probe locations were on the right process).
Anyway, I think I've located the problem. It was in the SPHProbeFilter which has to compute its extents correctly (from the 2nd input which ho,ds the probe geometry) and I had left in one SetWholeExtent call by accident from testing which had the piece extents and not the correct WholeExtent. I don't know how I missed it before, but things seem to be working now.
One new problem :
When I resample the data using the SPH kernel, I can generate a slice, or a volume: slices are great, but when I contour a volume, I am getting errors from vtkMath, "Unable to factor linear system" - now it's possible that there are slight discontinuities across process boundaries (if the ghost region is smaller than the kernel support radius) - do you know what reason might cause the error? (I wouldn't expect the contour filter to fail, just produce extra artefacts).

Thanks if you got this far!

JB


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20110720/b36b816a/attachment.htm>