[vtkusers] vtkImageAppend does not work correctly with multiple threads

[David Gobbi]

Hi Dominik,

It looks like an extent error around line 510 of vtkImageAppend.cxx,
here is the patch:

Change this:
     if (inExt[this->AppendAxis*2] <= inExt[this->AppendAxis*2 + 1])

To this:
     if (inExt[0] <= inExt[1] && inExt[2] <= inExt[3] && inExt[4] <= inExt[5])

Can you make this change to vtkImageAppend.cxx and let me know
if it fixes the issue?

 - David


On Fri, May 27, 2011 at 5:21 AM, Dominik Szczerba <dominik at itis.ethz.ch> wrote:
> Hi David,
>
> I submitted a bug with example data and steps to reproduce:
>
> http://www.vtk.org/Bug/view.php?id=12229
>
> Regards
> Dominik
>
> On Thu, May 26, 2011 at 8:33 PM, David Gobbi <david.gobbi at gmail.com> wrote:
>> Hi Dominik,
>>
>> Thanks, I just needed confirmation of what you had done to ensure that
>> the bug was specifically in vtkImageAppend and not in some other
>> filter in your pipeline.
>>
>> Thread-related problems in the imaging filters are often due to
>> incorrect assumptions by the execute methods about how the extents
>> will be broken up for different threads.  I notice that vtkImageAppend
>> does not have its own regression test, instead it is tested indirectly
>> by Imaging/Testing/Tcl/TestAllFlips.tcl and a few others, and only for
>> 2D images.
>>
>> If you can send a short test for it that crashes, one that can
>> eventually become part of the nightly testing, that would be the best
>> way to proceed.
>>
>>  - David
>>
>>
>> On Thu, May 26, 2011 at 12:01 PM, Dominik Szczerba <dominik at itis.ethz.ch> wrote:
>>> Yes, that was my point, it only works correctly with SetNumberOfThreads(1).
>>> So there seems to be a bug, I am happy to debug and patch,
>>> unfortunately, the code implementation is hardly documented...
>>> Any hints are appreciated.
>>> Dominik
>>>
>>> On Thu, May 26, 2011 at 5:58 PM, David Gobbi <david.gobbi at gmail.com> wrote:
>>>> Hi Dominik,
>>>>
>>>> If you call SetNumberOfThreads(1) does it make the problem go away?
>>>>
>>>>  - David
>>>>
>>>>
>>>> On Thu, May 26, 2011 at 9:37 AM, Dominik Szczerba <dominik at itis.ethz.ch> wrote:
>>>>>
>>>>> I have found that vtkImageAppend only works correctly for me
>>>>> (preserving extents, overlapping regions) with one thread. With more
>>>>> it crashes reported incorrect input/output pointers.
>>>>> Did anybody have similar problems?
>>>>>
>>>>> Thanks,
>>>>> Dominik
>>
>>
>