[Insight-users] ITK FFTW vs system FFTW

Sun Apr 14 10:28:04 EDT 2013

Brad,
I don't know why they don't automatically enable it. I understand what you
mean but I would at least default to system FFTW if it is available in the
system. One can expect this one to be optimized. The other thing is that
FFTW_OPTIMIZATION_CONFIGURATION could be made more visible with an option.
Simon

On Sun, Apr 14, 2013 at 4:20 PM, Bradley Lowekamp <blowekamp at mail.nih.gov>wrote:

> Simon,
>
> Thanks for trying out the patch. Sorry for the parse error blunder.
>
> Regarding the sse2 not being enable by default. I would view that as an
> issue with FFTW. Why do they not enable it by default? Every x86_64
> processor has sse2 instructions. In general, determining optimal
> compilation flags for different platforms, and architectures in not done by
> defaults in ITK. I think it is just too big of a rabbit hole to open up.
>
> Brad
>
> On Apr 14, 2013, at 8:15 AM, Simon Rit <simon.rit at creatis.insa-lyon.fr>
> wrote:
>
> Hi,
> Thanks! A few things. First, going from the latest release to HEAD, I had
> a compilation issue:
> /home/srit/src/itk4/itk/Modules/Filtering/FFT/include/
> itkFFTWHalfHermitianToRealInverseFFTImageFilter.hxx:103:5:   required
> from ‘void itk::FFTWHalfHermitianToRealInverseFFTImageFilter<TInputImage,
> TOutputImage>::BeforeThreadedGenerateData() [with TInputImage = itk::Image<std::complex<double>,
> 3u>; TOutputImage = itk::Image<double, 3u>]’
> /home/srit/src/rtk/rtk/testing/rtkdisplaceddetectortest.cxx:203:1:
> required from here
> /usr/lib/gcc/x86_64-redhat-linux/4.7.2/../../../../include/c++/4.7.2/bits/
> stl_algobase.h:329:8: erreur: incompatible types in assignment of ‘const
> std::complex<double>’ to ‘double [2]’
>
> I fixed it with a pointer cast
> diff --git a/Modules/Filtering/FFT/include/itkFFTWHalfHermitianToRealInve
> rseFFTImageFilter.hxx b/Modules/Filtering/FFT/include/
> itkFFTWHalfHermitianToRealInverseFFTImageFilter.hxx
> index 9338cb1..806d803 100644
> --- a/Modules/Filtering/FFT/include/itkFFTWHalfHermitianToRealInve
> rseFFTImageFilter.hxx
> +++ b/Modules/Filtering/FFT/include/itkFFTWHalfHermitianToRealInve
> rseFFTImageFilter.hxx
> @@ -100,7 +100,7 @@ FFTWHalfHermitianToRealInverseFFTImageFilter<
> TInputImage, TOutputImage >
>                                        !m_CanUseDestructiveAlgorithm );
>    if( !m_CanUseDestructiveAlgorithm )
>      {
> -    itk::algorithm::copy_n(inputPtr->GetBufferPointer(), totalInputSize,
>  in);
> +    itk::algorithm::copy_n((typename FFTWProxyType::ComplexType*)inputPtr
> ->GetBufferPointer(), totalInputSize,  in);
>      }
>    FFTWProxyType::Execute( plan );
>
> Second, there is a missing parenthesis in the patch:
> /home/srit/src/itk4/lin64-dg/fftwd_configure_step.cmake:36:
> Parse error.  Function missing ending ")".  End of file reached.
>
> Finally, the patch propagates -O3 but does not fully fix the speed issue.
> After a bit more digging, it turns out that sse2 instructions are not
> automatically activated. One must set FFTW_OPTIMIZATION_CONFIGURATION to
> "--enable-sse2". Shouldn't this be done automatically? Setting it fixed
> the compilation time.
> Thanks again for your help,
>  Simon
>
>
> On Fri, Apr 12, 2013 at 8:27 PM, Bradley Lowekamp <blowekamp at mail.nih.gov>wrote:
>
>> Hello,
>>
>> I have placed a patch on gerrit for review:
>>
>> http://review.source.kitware.com/#/c/10848/
>>
>> Please let us know if works for you.
>>
>> Brad
>>
>> On Apr 12, 2013, at 12:15 PM, Bradley Lowekamp <blowekamp at mail.nih.gov>
>> wrote:
>>
>> > Looking at the cmake build scripts:
>> >
>> >
>> https://github.com/Kitware/ITK/blob/master/CMake/itkExternal_FFTW.cmake#L44
>> >
>> > It looks like CMAKE_C_FLAGS_RELEASE is not getting passed.
>> >
>> > I'll see if I can take this on after lunch.
>> >
>> > Brad
>> >
>> > On Apr 12, 2013, at 11:30 AM, Simon Rit <simon.rit at creatis.insa-lyon.fr>
>> wrote:
>> >
>> >> I'm not sure it's the reason: I don't expect a two fold improvement
>> >> and the system has the same version installed, 3.3.2. Are we sure that
>> >> the release flags are passed to FFTW? When I looked at the compilation
>> >> log, I don't see -O3, here is an excerpt:
>> >>
>> >> /bin/sh ../libtool --tag=CC   --mode=compile /usr/bin/cc -std=gnu99
>> >> -DHAVE_CONFIG_H -I.
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel -I..
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/simd
>> >> -Wno-uninitialized -Wno-unused-parameter -Wall -Wcast-align
>> >> -Wdisabled-optimization -Wextra -Wformat=2 -Winvalid-pch
>> >> -Wno-format-nonliteral -Wpointer-arith -Wshadow -Wunused
>> >> -Wwrite-strings -funit-at-a-time -Wno-strict-overflow  -MT
>> >> cpy2d-pair.lo -MD -MP -MF .deps/cpy2d-pair.Tpo -c -o cpy2d-pair.lo
>> >> /home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel/cpy2d-pair.c
>> >> libtool: compile:  /usr/bin/cc -std=gnu99 -DHAVE_CONFIG_H -I.
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel -I..
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/simd -Wno-uninitialized
>> >> -Wno-unused-parameter -Wall -Wcast-align -Wdisabled-optimization
>> >> -Wextra -Wformat=2 -Winvalid-pch -Wno-format-nonliteral
>> >> -Wpointer-arith -Wshadow -Wunused -Wwrite-strings -funit-at-a-time
>> >> -Wno-strict-overflow -MT cpy2d-pair.lo -MD -MP -MF
>> >> .deps/cpy2d-pair.Tpo -c
>> >> /home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel/cpy2d-pair.c -o
>> >> cpy2d-pair.o
>> >> mv -f .deps/cpy2d-pair.Tpo .deps/cpy2d-pair.Plo
>> >> /bin/sh ../libtool --tag=CC   --mode=compile /usr/bin/cc -std=gnu99
>> >> -DHAVE_CONFIG_H -I.
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel -I..
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/simd
>> >> -Wno-uninitialized -Wno-unused-parameter -Wall -Wcast-align
>> >> -Wdisabled-optimization -Wextra -Wformat=2 -Winvalid-pch
>> >> -Wno-format-nonliteral -Wpointer-arith -Wshadow -Wunused
>> >> -Wwrite-strings -funit-at-a-time -Wno-strict-overflow  -MT cpy2d.lo
>> >> -MD -MP -MF .deps/cpy2d.Tpo -c -o cpy2d.lo
>> >> /home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel/cpy2d.c
>> >> libtool: compile:  /usr/bin/cc -std=gnu99 -DHAVE_CONFIG_H -I.
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel -I..
>> >> -I/home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/simd -Wno-uninitialized
>> >> -Wno-unused-parameter -Wall -Wcast-align -Wdisabled-optimization
>> >> -Wextra -Wformat=2 -Winvalid-pch -Wno-format-nonliteral
>> >> -Wpointer-arith -Wshadow -Wunused -Wwrite-strings -funit-at-a-time
>> >> -Wno-strict-overflow -MT cpy2d.lo -MD -MP -MF .deps/cpy2d.Tpo -c
>> >> /home/srit/src/itk4/lin64-dg/fftwd/src/fftwd/kernel/cpy2d.c -o cpy2d.o
>> >> mv -f .deps/cpy2d.Tpo .deps/cpy2d.Plo
>> >>
>> >>
>> >> On Fri, Apr 12, 2013 at 5:21 PM, Bill Lorensen <
>> bill.lorensen at gmail.com> wrote:
>> >>> Looks like the fftw folks have made some efficiency improvements.
>> >>>
>> >>>
>> >>>
>> >>> On Fri, Apr 12, 2013 at 11:18 AM, Simon Rit <
>> simon.rit at creatis.insa-lyon.fr>
>> >>> wrote:
>> >>>>
>> >>>> I should have mentioned it: yes, it is!
>> >>>>
>> >>>> On Fri, Apr 12, 2013 at 5:16 PM, Bill Lorensen <
>> bill.lorensen at gmail.com>
>> >>>> wrote:
>> >>>>> Is your ITK built Release?
>> >>>>>
>> >>>>>
>> >>>>>
>> >>>>> On Fri, Apr 12, 2013 at 11:13 AM, Simon Rit
>> >>>>> <simon.rit at creatis.insa-lyon.fr>
>> >>>>> wrote:
>> >>>>>>
>> >>>>>> Hi,
>> >>>>>> We have recently observed on a fedora distribution that FFTW is two
>> >>>>>> times slower when one uses the ITK compilation
>> >>>>>> (USE_SYSTEM_FFTW:BOOL=OFF, default) than when one uses the system
>> fftw
>> >>>>>> (USE_SYSTEM_FFTW:BOOL=ON).
>> >>>>>>
>> >>>>>> We have looked for an explanation in the flags (the Fedora flags
>> are
>> >>>>>> available here
>> >>>>>> http://pkgs.fedoraproject.org/cgit/fftw.git/tree/fftw.spec)
>> >>>>>> but we have not managed to figure it out. Would someone happen to
>> have
>> >>>>>> an explanation and solution?
>> >>>>>>
>> >>>>>> Thanks in advance,
>> >>>>>> Simon
>> >>>>>> _____________________________________
>> >>>>>> 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
>> >>>>>
>> >>>>>
>> >>>>>
>> >>>>>
>> >>>>> --
>> >>>>> Unpaid intern in BillsBasement at noware dot com
>> >>>
>> >>>
>> >>>
>> >>>
>> >>> --
>> >>> Unpaid intern in BillsBasement at noware dot com
>> >> _____________________________________
>> >> 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
>> >
>> > _____________________________________
>> > 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
>>
>> _____________________________________
>> 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
>>
>
>
>
> _____________________________________
> 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
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130414/8fc89f44/attachment.htm>