[Insight-developers] itk performance numbers

Rupert Brooks rupert.brooks at gmail.com
Wed Jul 25 11:53:45 EDT 2012


Brad,

Agreed, VS2005 is not my favorite either, which is why im doing tests also
with a recent gcc.  If the performance degradation was limited to VS2005,
that would not be a big deal in my view.   I'm stuck with it at the moment,
but I can hardly expect you folks to do your best work for it.

If and when i give other compilers a shot, i will keep you posted.

Rupert
--------------------------------------------------------------
Rupert Brooks
rupert.brooks at gmail.com



On Wed, Jul 25, 2012 at 10:52 AM, Bradley Lowekamp
<blowekamp at mail.nih.gov>wrote:

> Hello Rupert,
>
> I see you are still using Visual Studio 8 2005, I am curious how VS10
> would compare. I know that there are a couple special case optimization
> added to ITKv4 which VS8 is not able to take advantage off.
>
> As others have reported we are begining to address some of these issues.
>
> Thanks for sharing!
> Brad
>
> On Jul 25, 2012, at 10:12 AM, Rupert Brooks wrote:
>
> Hi,
>
> I was commenting the other day that ITK4 seems to have lower performance
> than ITK3.  I put together a little benchmark to test this objectively.
>  Results for a couple of systems are below.  The benchmark is neither
> particularly sophisticated nor exact, but i have tried to make it typical
> of what i use ITK for.  I have not tried it with Hans' patch as that got
> abandoned.  Will follow up separately on that.
>
> In the numbers below, you will see that performance seems to have degraded
> across the board.  One exception is the 2D image test.  In this test i was
> lazy, and i left the image axis aligned with unit spacing.  I am suspecting
> a special case optimization in ITK4.  At the moment, this blocks my ability
> to upgrade my projects to ITK4, as they are performance critical.  Any
> hints and tips to get those milliseconds back will be much appreciated.
>
> The code is at https://github.com/rupertbrooks/itkbench, i'd be curious
> if others get the same results, and of course, criticism and improvements
> are welcome.
>
> Note that the output about what processor / cores is on the system is
> bogus. The windows system is a core i7 with 4 cores, the linux one is a
> Core2 Quad.  I suspect a bug in itksys::SystemInformation, but i will
> follow up on that separately also.
>
> In all itk builds, ITK_USE_REVIEW is on.  In the itk3 builds,
> USE_OPTIMZIZED_REGISTRATION, and the new statistics framework are turned on.
>
> Cheers,
> Rupert
>
> Data follows.....
>
> itk3 Windows XP 32bit Visual Studio Pro 2005  Build RelWithDebInfo
> System: CAMD5C5PMN1
> Processor: Pentium III (0.18 micron) With 1 Or 2 MB On-Die L2 Cache
>  Serial #:
>     Cache: -1
>     Clock: 2800
>     Cores: 4 cpus x 1 Cores = 4
> OSName:     Windows
>   Release:  XP Professional
>   Version:  Service Pack 3 (Build 2600)
>   Platform: x86
>   Operating System is 32 bit
> ITK Version: 3.20.0
> Virtual Memory: Total: 2047 Available: 2021
> Physical Memory: Total:3581 Available: 1457
>            Probe Name:        Count          Min           Mean
> Stdev            Max        Total
>  FilterChain_1_threads            10     0.0345688     0.0361805
> 0.0381523     0.0375099      0.361805
>  FilterChain_2_threads            10     0.0257301     0.0265144
> 0.0279574     0.0282478      0.265144
>  FilterChain_3_threads            10      0.022007     0.0232452
> 0.0245125     0.0240822      0.232452
>  FilterChain_4_threads            10     0.0199661     0.0221756
> 0.0235205     0.0288773      0.221756
>                Image2D            10      0.144909       0.14665
>  0.154586      0.148136        1.4665
>                Image3D            10       0.19125      0.193378
>  0.203846      0.196495       1.93378
>  MeanSquares_1_threads            10      0.612728      0.616159
>  0.649503      0.627789       6.16159
>  MeanSquares_2_threads            10      0.359703      0.368051
>  0.388106      0.392696       3.68051
>  MeanSquares_3_threads            10      0.268024      0.280672
>  0.296142      0.312557       2.80672
>  MeanSquares_4_threads            10      0.199249      0.213209
>  0.224918        0.2243       2.13209
>
>
> itk4 Windows XP 32bit Visual Studio Pro 2005  Build RelWithDebInfo
>
> System: CAMD5C5PMN1
> Processor: Pentium III (0.18 micron) With 1 Or 2 MB On-Die L2 Cache
>  Serial #:
>     Cache: -1
>     Clock: 2800
>     Cores: 4 cpus x 1 Cores = 4
> OSName:     Windows
>   Release:  XP Professional
>   Version:  Service Pack 3 (Build 2600)
>   Platform: x86
>   Operating System is 32 bit
> ITK Version: 4.3.0
> Virtual Memory: Total: 2047 Available: 2008
> Physical Memory: Total:3581 Available: 1574
>            Probe Name:        Count          Min           Mean
> Stdev            Max        Total
>  FilterChain_1_threads            10     0.0797005     0.0808155
> 0.0851898     0.0820313      0.808155
>  FilterChain_2_threads            10     0.0670509     0.0921925
> 0.0991914      0.109177      0.921925
>  FilterChain_3_threads            10     0.0648499      0.113566
>  0.120975      0.125893       1.13566
>  FilterChain_4_threads            10     0.0536919      0.104662
>  0.114558      0.132114       1.04662
>                Image2D            10     0.0953979     0.0965282
>  0.101753     0.0978088      0.965282
>                Image3D            10      0.196468      0.198288
>  0.209017      0.200562       1.98288
>  MeanSquares_1_threads            10      0.982071      0.986089
> 1.03944      0.996445       9.86089
>  MeanSquares_2_threads            10      0.686028      0.728621
>  0.768821      0.799675       7.28621
>  MeanSquares_3_threads            10      0.534885      0.554387
>  0.5846      0.578133       5.54387
>  MeanSquares_4_threads            10      0.441551      0.463463
>  0.488903      0.502502       4.63463
>
>
> itk3 Linux 64 bit gcc 4.6.1 Build Release
> System: morrigan
> Processor: Intel(R) Core(TM)2 Duo CPU     T9500  @ 2.60GHz
>  Serial #:
>     Cache: 2048
>     Clock: 2003
>     Cores: 4 cpus x 4 Cores = 16
> OSName:     Linux
>   Release:  3.0.0-21-generic
>   Version:  #35-Ubuntu SMP Fri May 25 17:57:41 UTC 2012
>   Platform: x86_64
>   Operating System is 64 bit
> ITK Version: 3.20.1
> Virtual Memory: Total: 20795 Available: 20795
> Physical Memory: Total:8001 Available: 3416
>            Probe Name:        Count          Min           Mean
> Stdev
>          Max        Total
>  FilterChain_1_threads            10     0.0309269      0.031097
> 0.0327796    0.031528       0.31097
>  FilterChain_2_threads            10      0.021672     0.0248219
> 0.0261976    0.026264      0.248219
>  FilterChain_3_threads            10     0.0213931     0.0217796
> 0.0229634   0.0228851      0.217796
>  FilterChain_4_threads            10     0.0198419     0.0207421
> 0.0218725   0.0216701      0.207421
>                Image2D            10     0.0263629     0.0264159
> 0.0278451   0.0267591      0.264159
>                Image3D            10     0.0263779     0.0263906
> 0.0278181    0.026463      0.263906
>  MeanSquares_1_threads            10      0.463909      0.465901
>  0.491113    0.473644       4.65901
>  MeanSquares_2_threads            10      0.349138       0.40529
>  0.429438    0.487785        4.0529
>  MeanSquares_3_threads            10      0.331908      0.357685
>  0.377834    0.400941       3.57685
>  MeanSquares_4_threads            10      0.299428      0.336914
>  0.356878    0.395144       3.36914
>
> itk4 Linux 64-bit gcc 4.6.1 Build Release
> System: morrigan
> Processor: Intel(R) Core(TM)2 Duo CPU     T9500  @ 2.60GHz
>  Serial #:
>     Cache: 2048
>     Clock: 2003
>     Cores: 4 cpus x 4 Cores = 16
> OSName:     Linux
>   Release:  3.0.0-21-generic
>   Version:  #35-Ubuntu SMP Fri May 25 17:57:41 UTC 2012
>   Platform: x86_64
>   Operating System is 64 bit
> ITK Version: 4.3.0
> Virtual Memory: Total: 20795 Available: 20795
> Physical Memory: Total:8001 Available: 3400
>            Probe Name:        Count          Min           Mean
> Stdev
>          Max        Total
>  FilterChain_1_threads            10     0.0663331     0.0668247
> 0.0704416   0.0681288      0.668247
>  FilterChain_2_threads            10        0.0353     0.0393441
>  0.042048   0.0534279      0.393441
>  FilterChain_3_threads            10      0.025275     0.0265869
> 0.0282408   0.0364931      0.265869
>  FilterChain_4_threads            10     0.0208249     0.0222288
> 0.0235333    0.028336      0.222288
>                Image2D            10      0.026361      0.026421
> 0.0278503   0.0266371       0.26421
>                Image3D            10     0.0263991     0.0266509
> 0.0280954   0.0274949      0.266509
>  MeanSquares_1_threads            10       0.69005       0.69459
>  0.732169    0.701676        6.9459
>  MeanSquares_2_threads            10      0.590396      0.699826
>  0.739994    0.766702       6.99826
>  MeanSquares_3_threads            10      0.527146      0.538844
>  0.568047    0.548791       5.38844
>  MeanSquares_4_threads            10      0.417008      0.454278
>  0.480139    0.533593       4.54278
> --------------------------------------------------------------
> Rupert Brooks
> rupert.brooks at gmail.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://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-developers
>
>
> ========================================================
>
> Bradley Lowekamp
>
> Medical Science and Computing for
>
> Office of High Performance Computing and Communications
>
> National Library of Medicine
>
> blowekamp at mail.nih.gov
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-developers/attachments/20120725/d1a52056/attachment.htm>


More information about the Insight-developers mailing list