[Insight-developers] DigitallyReconstructedRadiograph1Test fails on 32-bit linux, OK on 64?
Matt McCormick
matt.mccormick at kitware.com
Sat Oct 27 12:26:04 EDT 2012
On Sat, Oct 27, 2012 at 9:03 AM, Steve M. Robbins <steve at sumost.ca> wrote:
> On October 26, 2012 01:34:13 AM Matt McCormick wrote:
>> Hi Steve,
>>
>> Please review these patches. The test now passes on my 32-bit MinGW build.
>>
>> http://review.source.kitware.com/#/c/8177/
>> http://review.source.kitware.com/#/c/8178/
>
> Yes, this fixes the problem. And I learned about a new feature of ITK:
> CompensatedSummation!
>
Outstanding! That bug has been around for a long time...
> Regarding the latter, I noticed that itk::NumericTraits<T>::AccumulateType
> generally has more bits than T, e.g. int uses long as AccumulateType. However
> unsigned int uses unsigned int as AccumulateType.
> Why not (unsigned) long?
Pre-CompensatedSummation, AccumulateType was a little better than the
real type, but not by much. However, precision accumulation is not a
problem with integers.
> Similarly, float's accumulate type is double, but std::complex<float> uses
> complex<float> rather than complex<double>.
Good catch. I think that could be considered a bug.
Thanks,
Matt
> Thanks again,
> -Steve
>
>>
>> Thanks,
>> Matt
>>
>> On Mon, Oct 15, 2012 at 12:45 PM, Matt McCormick
>>
>> <matt.mccormick at kitware.com> wrote:
>> > Hi Steve,
>> >
>> > I can reproduce this on the 32-bit MinGW build. Thanks for bring the
>> > issue up. I will take a look and see if I can fix the problem.
>> >
>> > Thanks,
>> > Matt
>> >
>> > On Mon, Oct 15, 2012 at 4:41 AM, Steve M. Robbins <steve at sumost.ca> wrote:
>> >> Hi,
>> >>
>> >> The DigitallyReconstructedRadiograph1Test has been failing for me for
>> >> months across a number of 32-bit linux machines. The difference image
>> >> has a small stripe of errors; see
>> >> http://open.cdash.org/viewTest.php?onlyfailed&buildid=2606109
>> >>
>> >> I see "amber12" test machine is 32-bit mingw windows build and it
>> >> passes the test. Is anyone testing ITK on 32 bit linux these days?
>> >> Does the test pass for you?
>
More information about the Insight-developers
mailing list