[ITK-dev] efficiency of vnl_matrix

Bradley Lowekamp blowekamp at mail.nih.gov
Thu Mar 12 10:20:53 EDT 2015


Chuck,

Thank you for giving us that important conclusion, under a quite difficult situations.

I wonder if there is any distinction in the usage of vnl_matrix vs vnl_matrix_fixed. I would expect that operations done for pixel transforms should have there dimension know at run-time and should be able to use the vnl_matrix_fixed.

I also have considered methods to transform a whole array of points at a time. I wonder if for 3x3*3*256 sized operations ( scan-line size ) if there would be benefit with the library based operations.

Brad


On Mar 12, 2015, at 10:02 AM, Chuck Atkins <chuck.atkins at kitware.com> wrote:

> I worked with Julie Langou, maintainer of LAPACK, on this project a few years ago.  The funding situation ended up very strange and messy and we basically had to cram 3 months worth of effort into 3 weeks, so needless to say, we were not able to really achieve our goals.  However, we spent a fair amount of time profiling ITK and analyzing it's hot spots from vnl to determine where to best spend the small ammount of time we had.  The results were not as straight forward as we expected.  It turns out that most of the use for vnl_matrix and vnl_vector was actually for an enourmous number of operations on very small sized vectors and matricies (dimensions of 2, 3, or 4), often for coordinate and geometry calculations or for small per-pixel operations that were not easily vectorized in the implementation at the time.  In these cases, the overhead of calling out to a BLAS or LAPACK library was much too expensive and the existing use of VNL was far more optimal.  This falls apart, however when trying to use vnl for more complex algorithms since the larger matrix operations will be where the benefit can be seen.  So just re-implementing the vnl vector and matrix classes and operators with underlying BLAS and LAPACK routines turned out to not be the best solution for ITK as a whole.
> 
> - Chuck
> 
> On Thu, Mar 12, 2015 at 8:32 AM, Bradley Lowekamp <blowekamp at mail.nih.gov> wrote:
> Hello,
> 
> If I was writing my own ITK classes, and needed a fast matrix library I would likely pursue an additional dependency on an efficient numeric library for that project, such as eigen.
> 
> However for the broad appeal of ITK I would think a flexible back end would be best. As I think there are a variety of BLAS and LAPACK libraries available ( commercial, open source, vender free ). It would be nice to pick one what has been optimized for the current architecture. I would think it would be most flexible to use this interface in the back end of a chosen numeric interface ( currently VNL ). Unfortunately, I don't have as much experience with these libraries as I'd like.
> 
> Brad
> 
> On Mar 12, 2015, at 5:15 AM, m.staring at lumc.nl wrote:
> 
> > Hi,
> >
> > I think the eigen library is a mature and very fast library for these kind of things:
> > http://eigen.tuxfamily.org/index.php?title=Main_Page
> >
> > You may want to check it out, to see if it offers what you need.
> >
> > It would be great to be able to use these within the itk.
> >
> > 2c
> > Marius
> >
> > -----Original Message-----
> > From: Insight-developers [mailto:insight-developers-bounces at itk.org] On Behalf Of Jian Cheng
> > Sent: Wednesday, March 11, 2015 23:17
> > To: Matt McCormick
> > Cc: Chuck Atkins; ITK
> > Subject: Re: [ITK-dev] efficiency of vnl_matrix
> >
> > Hi Matt,
> >
> > Thanks for your help, and also for the ITK workshop in UNC last time.
> >
> > It is very unfortunate. The efficiency of these numerical math operators are very important for many applications.
> >
> > I recently released an ITK based toolbox, called dmritool, for diffusion MRI data processing.
> > It has some files to add some supports of blas, lapack, mkl to vnl_matrix and vnl_vector.
> > http://diffusionmritool.github.io/dmritool_doxygen/utlBlas_8h_source.html
> > http://diffusionmritool.github.io/dmritool_doxygen/utlVNLBlas_8h_source.html
> >
> > Those functions are not internally for vnl_matrix class. They are operators for the data pointer stored in vnl_matrix object.
> > Thus, later I made a N-dimensional array library which internally includes those functions, and also supports expression template to avoid temporary copies.
> > http://diffusionmritool.github.io/dmritool_doxygen/utlMatrix_8h_source.html
> > http://diffusionmritool.github.io/dmritool_doxygen/utlVector_8h_source.html
> >
> > The efficiency comparison between vnl_vector/vnl_matrix and the vector/matrix using openblas, lapack, or mkl can be found by running those two tests https://github.com/DiffusionMRITool/dmritool/blob/master/Modules/HelperFunctions/test/utlVNLBlasGTest.cxx
> > https://github.com/DiffusionMRITool/dmritool/blob/master/Modules/HelperFunctions/test/utlVNLLapackGTest.cxx
> >
> > Maybe some codes can be used as patches in somewhere in ITK. I am not sure. Maybe we need more discussion on it.
> > With your help and discussion, I will be very glad to make my first patch to ITK.
> > Thanks.
> >
> > best,
> > Jian Cheng
> >
> >
> > On 03/11/2015 04:39 PM, Matt McCormick wrote:
> >> Hi Jian,
> >>
> >> Yes, it would be wonderful to improve the efficiency of these basic
> >> numerical operations.
> >>
> >> Funding for the Refactor Numerical Libraries has currently ended, and
> >> the effort is currently frozen.  However, you are more than welcome to
> >> pick it up and we can help you get it into ITK.  More information on
> >> the patch submission process can be found here [1] and in the ITK
> >> Software Guide.
> >>
> >> Thanks,
> >> Matt
> >>
> >> [1]
> >> https://insightsoftwareconsortium.github.io/ITKBarCamp-doc/CommunitySo
> >> ftwareProcess/SubmitAPatchToGerrit/index.html
> >>
> >> On Wed, Mar 11, 2015 at 4:07 PM, Jian Cheng <jian.cheng.1983 at gmail.com> wrote:
> >>> Hi,
> >>>
> >>> My task using ITK has intensive matrix-matrix product, pseudo-inverse, etc.
> >>> Thus the performance is actually mainly determined by the matrix
> >>> library I used.
> >>> Firstly I use vnl_matrix and vnl_vector in ITK. Then I found it is
> >>> very inefficient because vnl matrix lib does not use blas and lapack.
> >>> After I wrote my own matrix class which uses openblas and lapack, I
> >>> got a hug gain of performance.
> >>>
> >>> I found there is a proposal to improve the efficiency of numerical
> >>> libraries in ITK.
> >>> http://www.itk.org/Wiki/ITK/Release_4/Refactor_Numerical_Libraries
> >>> I am not sure how is the progress of the proposal.
> >>> I wonder when the vnl matrix lib can internally support blas and
> >>> lapack, or mkl, so that we can just use it without lose of the efficiency.
> >>> Thanks.
> >>>
> >>> best,
> >>> Jian Cheng
> >>> _______________________________________________
> >>> 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.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://public.kitware.com/mailman/listinfo/insight-developers
> >
> > _______________________________________________
> > 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.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://public.kitware.com/mailman/listinfo/insight-developers
> > _______________________________________________
> > 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.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://public.kitware.com/mailman/listinfo/insight-developers
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-developers/attachments/20150312/8a7351a2/attachment.html>


More information about the Insight-developers mailing list