[Insight-developers] CastImageFilterTest failures

[Cory Quammen]

Luis and Bill,

I've posted a patch that disables optimization for the itkCastImageFilterTest:

http://review.source.kitware.com/#change,1831

Cory

On Mon, Jun 6, 2011 at 12:38 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
> On Sun, Jun 5, 2011 at 7:16 PM, Luis Ibanez <luis.ibanez at kitware.com> wrote:
>> Cory, Bill,
>>
>> One of the systems where this tests is failing is
>> zion.kitware. This is a 32bits system with gcc 4.4.
>>
>> I just spent some quality time with the test, and it
>> seems that the failure only occurs when compiling
>> for Release.
>
> Thanks for taking the time to do this, Luis.
>
>> The build for Debug, and the build without optimization
>> let the test pass.
>>
>> --
>>
>> Under Release the failure is quite curious because
>> if you repeat the check comparing the two float values
>> it returns alternatively true and false (as shown in the
>> attached code).
>>
>> It looks like a bug of the compiler optimization.
>>
>>
>> I suggest that we use CMake property flags for
>> excluding this .cxx from optimization at compilation
>> time.
>>
>
> That sounds reasonable to me.
>
> I'll work on a patch.
>
> Cory
>
>>
>>    Luis
>>
>>
>> -----------------------------------
>> On Fri, Jun 3, 2011 at 9:48 AM, Cory Quammen <cquammen at cs.unc.edu> wrote:
>>> Bill,
>>>
>>> This is a very strange problem on some systems. Somehow, the
>>> static_cast in the Cast functor in itkCastImageFilter.h produces
>>> different results than the static_cast in my test when casting from
>>> float to int.
>>>
>>> Basically, if you have
>>>
>>> int a = ...;
>>> float b = ...;
>>>
>>> b = static_cast< float >( a );
>>> b == static_cast< float >( a );
>>>
>>> the equality test in the last line of code can be false! This
>>> frightens and confuses me.
>>>
>>> My last change to the test fixed the error on MSVC by storing the
>>> static_cast result in a temporary variable and then doing an equality
>>> test between the temporary to the output of the CastImageFilter, but
>>> it obviously didn't fix it everywhere.
>>>
>>> I have no idea why the test is failing. Maybe one of the static_casts
>>> is casting from a higher-precision register than the other. I've
>>> Googled for similar problems reported elsewhere, but haven't found
>>> anything yet.
>>>
>>> For now, should I disable the float-to-int cast test to make the
>>> dashboard green?
>>>
>>> I have no idea why the test seems to be completely failing on SunOS.
>>>
>>> Gaetan,
>>>
>>> I'd be grateful if you could run the CastImageFilterTest in a debugger
>>> and send me the stack trace when it segfaults. To minimize the time
>>> you spend on this, the commands should be:
>>>
>>> cd $ITK_BUILD
>>> gdb bin/ITK-ImageFilterBaseTestDriver
>>> r itkCastImageFilterTest
>>>
>>> Thanks,
>>> Cory
>>>
>>>
>>> On Fri, Jun 3, 2011 at 8:35 AM, Bill Lorensen <bill.lorensen at gmail.com> wrote:
>>>> Cory,
>>>>
>>>> CastImageFilterTest is failing on 3 platforms:
>>>> http://www.cdash.org/CDash/testSummary.php?project=2&name=itkCastImageFilterTest&date=2011-06-03
>>>>
>>>> The two linux platforms are failing on
>>>> Casting from int to float ... [FAILED]
>>>>
>>>> The Sun system is seg faulting.
>>>>
>>>> Bill
>>>>
>>>
>>>
>>>
>>> --
>>> Cory Quammen
>>> Computer Integrated Systems for Microscopy and Manipulation (CISMM)
>>> Department of Computer Science
>>> University of North Carolina at Chapel Hill
>>> http://www.cs.unc.edu/~cquammen
>>> _______________________________________________
>>> Powered by www.kitware.com
>>>
>>> Visit other Kitware open-source projects at
>>> http://www.kitware.com/opensource/opensource.html
>>>
>>> Kitware offers ITK Training Courses, for more information visit:
>>> http://kitware.com/products/protraining.html
>>>
>>> Please keep messages on-topic and check the ITK FAQ at:
>>> http://www.itk.org/Wiki/ITK_FAQ
>>>
>>> Follow this link to subscribe/unsubscribe:
>>> http://www.itk.org/mailman/listinfo/insight-developers
>>>
>>
>
>
>
> --
> Cory Quammen
> Computer Integrated Systems for Microscopy and Manipulation (CISMM)
> Department of Computer Science
> University of North Carolina at Chapel Hill
> http://www.cs.unc.edu/~cquammen
>



-- 
Cory Quammen
Computer Integrated Systems for Microscopy and Manipulation (CISMM)
Department of Computer Science
University of North Carolina at Chapel Hill
http://www.cs.unc.edu/~cquammen