[Insight-users] FFTW thread safety

[Cory Quammen]

Also, I meant to ask you why adding a Lock/Unlock pair in
BeforeThreadedGenerateData was necessary. The lock appears to be
initialized at instantiation, so perhaps it isn't necessary for you to
do that?

Thanks,
Cory

On Wed, Nov 14, 2012 at 12:09 PM, Cory Quammen <cquammen at cs.unc.edu> wrote:
> Simon,
>
> I posted a patch to gerrit. You can review it here:
>
> http://review.source.kitware.com/#/c/8506/
>
> Thanks,
> Cory
>
> On Wed, Nov 14, 2012 at 9:53 AM, Cory Quammen <cquammen at cs.unc.edu> wrote:
>> Simon,
>>
>> Thank you for the patch. I will post a patch based on your patch to
>> gerrit when I have a moment today, unless someone beats me to it.
>>
>> Cory
>>
>> On Wed, Nov 14, 2012 at 9:48 AM, Simon Rit
>> <simon.rit at creatis.insa-lyon.fr> wrote:
>>> Hi,
>>> I have experienced problems with FFTW in ITK 4. I am doing many
>>> parallel FFT filters, each parallel thread having the piece of the
>>> image defined by ThreadedGenerateData. This is a filter part of
>>> www.openrtk.org which code is here:
>>> https://github.com/SimonRit/RTK/blob/master/code/rtkFFTRampImageFilter.txx
>>>
>>> I have originally done that with the code of Gaëtan Lehmann
>>> (http://www.insight-journal.org/browse/publication/717) but I had to
>>> fix it then for thread safety. I have now moved to ITK4 but I think
>>> that the thread safety issue is still here since I have random
>>> crashes. The problem is that the destroy functions are not protected
>>> by the mutex although they are not thread safe
>>> (http://www.fftw.org/doc/Thread-safety.html). So a lock / unlock must
>>> be added around it in itkFFTWCommon.h.
>>> The patch is below, let me know what you think.
>>> Simon
>>>
>>> PS: It is also worth of note that I have added a Lock/Unlock in the
>>> BeforeThreadedGenerateData function of my filter to ensure that the
>>> mutex is initialized.
>>>
>>> The patch:
>>>
>>> @@ -351,7 +350,9 @@ public:
>>>    }
>>>    static void DestroyPlan(PlanType p)
>>>    {
>>> +    FFTWGlobalConfiguration::Lock();
>>>      fftwf_destroy_plan(p);
>>> +    FFTWGlobalConfiguration::Unlock();
>>>    }
>>>  };
>>>
>>> @@ -653,7 +654,9 @@ public:
>>>    }
>>>    static void DestroyPlan(PlanType p)
>>>    {
>>> +    FFTWGlobalConfiguration::Lock();
>>>      fftw_destroy_plan(p);
>>> +    FFTWGlobalConfiguration::Unlock();
>>>    }
>>>  };
>>> _____________________________________
>>> 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://www.kitware.com/products/protraining.php
>>>
>>> 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-users
>>
>>
>>
>> --
>> Cory Quammen
>> Research Associate
>> Department of Computer Science
>> The University of North Carolina at Chapel Hill
>
>
>
> --
> Cory Quammen
> Research Associate
> Department of Computer Science
> The University of North Carolina at Chapel Hill



-- 
Cory Quammen
Research Associate
Department of Computer Science
The University of North Carolina at Chapel Hill