[vtk-developers] perf improvements to vtkDijkstra* classes - Mac test failure

[David Cole]

In VTK/Graphics/Testing/CMakeLists.txt, the list of header files occurring
after the VTK_GRAPHICS_EXPORT symbol is a list of "exceptions". Add your new
header file to that list since it does not declare a class that derives from
vtkObject.
HTH,
David


On Thu, Aug 28, 2008 at 3:57 PM, Dean Inglis <dean.inglis at camris.ca> wrote:

>
> Hi,
>
> In doing the PIMPL'ng I wrote vtkDijkstraGraphInternals.h
> to contain stl containers used be classes vtkDijkstraGraphGeodesicPath
> and vtkDijkstraImageGeodesicPath.  Im getting HeaderTesting-Graphics
> is failing for continuous build Darwin-ppc-cocoa-java:
>
> Use export macro: VTK_GRAPHICS_EXPORT
>
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkDijkstraGraphInternals.h does not define any
> classes
>
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkDijkstraGraphInternals.h has 2 includes:
>   24: #include <vtkstd/vector>
>   25: #include <vtkstd/map>
>
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkDijkstraGraphInternals.h has non-portable
> include(s):
>   24: #include <vtkstd/vector>
>   25: #include <vtkstd/map>
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkDijkstraGraphInternals.h does not have type
> macro
> Should be:
>  vtkTypeRevisionMacro(, );
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkDiscreteMarchingCubes.h does not define
> PrintSelf
> method:
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkHyperOctreeDepth.h does not define PrintSelf
> method:
> File: /Users/kitware/Dashboards/My
> Tests/VTK-cont/Graphics/vtkHyperOctreeLimiter.h does not define PrintSelf
> method:
>
> There were warnings:
> * No PrintSelf method
>
> There were errors:
> * Multiple includes
> * Non-portable includes
> * No type macro
> * No class defined
>
>
> Any ideas on how to fix this????
>
> thanks,
> Dean
>
>
>
> -----Original Message-----
> From: Amy Squillacote [mailto:ahs at cfdrc.com]
> Sent: August-28-08 8:49 AM
> To: Karthik Krishnan
> Cc: Dean Inglis; vtk-developers at vtk.org; 'Will Schroeder'
> Subject: Re: [vtk-developers] perf improvements to vtkDijkstra* classes
>
> Yes, the input type can absolutely be overwritten using
> FillInputPortInformation. For an example, see vtkContourFilter.
>
> - Amy
>
> Karthik Krishnan wrote:
> > Hi Dean:
> >
> > Thanks a lot for the contrib.
> >
> > I took a look at vtkDijkstraImageGeodesicPath and the first input
> > polydata is required but isn't used. The second input (cost function
> > image) is the one used.
> >
> > I think you can have a class deriving from vtkPolyDataAlgorithm that
> > has its first input as something other than vtkpolydata. This can be
> > done by overriding the method FillInputPortInformation and having the
> > first input be a vtkImageData. Can't it ?
> >
> > If not, maybe we should change vtkGeodesicPath to derive from
> > vtkAlgorithm.
> >
> > The other changes to vtkDijkstraGraphGeodesicPath look great.
> >
> > Thanks
> > --
> > karthik
> >
> > Dean Inglis wrote:
> >>
> >> Hi Will,
> >>
> >> thanks! Been having some fun here.
> >>
> >> Hi Karthik,
> >>
> >> I made a separate header file: vtkDijkstraGraphInternals that contains
> >>
> >> all the replacement stl containers so that
> >> vtkDijkstraGraphGeodesicPath and vtkDijkstraImageGeodesicPath
> >>
> >> have access. I left the implementation of the binary heap as is in
> >> vtkDijkstraGraphGeodesicPath
> >>
> >> with the exception that it now uses stl containers. I broke the cost
> >> function into "static"
> >>
> >> and "dynamic" parts so that edge costs can be calculated more
> >> efficiently. I have also
> >>
> >> taken the liberty to rename internal containers from things like
> >> this->f to this->OpenVertices
> >>
> >> as given in vtkDijkstraGraphInternals (see below). I will wait till
> >> tomorrow to commit in
> >>
> >> case you have any reservations or would like to discuss.
> >>
> >> thanks,
> >>
> >> Dean
> >>
> >> ------------------------------------------------------------------------
> >>
> >> *From:* Will Schroeder [mailto:will.schroeder at kitware.com]
> >> *Sent:* August-26-08 5:14 PM
> >> *To:* dean.inglis at camris.ca
> >> *Cc:* Karthik Krishnan
> >> *Subject:* Re: [vtk-developers] perf improvements to vtkDijkstra*
> >> classes
> >>
> >> I have no objections and have been doing similar things to other VTK
> >> classes. Great work!
> >> W
> >>
> >> 2008/8/26 Dean Inglis <dean.inglis at sympatico.ca
> >> <mailto:dean.inglis at sympatico.ca>>
> >>
> >> Is there any interest in (or objections to) PIMPL'ing some (or all) of
> >>
> >> the storage containers in vtkDijkstraGraphGeodesicPath?
> >>
> >> In a test class, I have reimplemented vtkDijkstraGraphGeodesicPath
> >>
> >> using the following substitutions and noticed substantial (qualitative)
> >>
> >> speed gains in interactivity:
> >>
> >> 1) - change containers f and s to vector<bool>
> >>
> >> 2) - change containers pre, Heap, and p to vector<int>
> >>
> >> 3) - change container for cost; d, to vector<float>
> >>
> >> 4) - change container for Adjacency from an array of vtkIdList* to
> >>
> >> vector<map<int,float>> for unique insertion of vertex id's and
> >>
> >> associated edge costs
> >>
> >> Probably most of the improvement comes from item 4) since
> >>
> >> you can pre-calculate all edge costs at the time of constructing
> >>
> >> the adjacency array, which only needs to be done when the input
> >>
> >> changes (not when start and end vertices change).
> >>
> >> Dean
> >>
> >>
> >> _______________________________________________
> >> vtk-developers mailing list
> >> vtk-developers at vtk.org <mailto:vtk-developers at vtk.org>
> >> http://www.vtk.org/mailman/listinfo/vtk-developers
> >>
> >>
> >>
> >>
> >> --
> >> William J. Schroeder, PhD
> >> Kitware, Inc.
> >> 28 Corporate Drive
> >> Clifton Park, NY 12065
> >> will.schroeder at kitware.com <mailto:will.schroeder at kitware.com>
> >> http://www.kitware.com
> >> 518-371-3971 (phone and fax)
> >>
> >
>
> --
> Amy Squillacote                    Phone: (256) 726-4839
> Computer Scientist                 Fax: (256) 726-4806
> CFD Research Corporation           Web: http://www.cfdrc.com
> 215 Wynn Drive, Suite 501
> Huntsville, AL  35805
>
>
> _______________________________________________
> vtk-developers mailing list
> vtk-developers at vtk.org
> http://www.vtk.org/mailman/listinfo/vtk-developers
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20080828/11f65c54/attachment.html>