[vtkusers] Why is the image red?

[Allan James]

David,

Unfortunately I am on Windows and my trial version for the Intel compiler suite has recently run out.. But would be good to see the results of this.

Is there any standard code that people use for speed/performance testing of VTK?

Anyone else using the Intel compilers (with profiling tools) and able to test the effect of pointer arithmetic on auto-vectorization?

Cheers,

Allan James
High Performance Computing & Research Support
Queensland University of Technology
(07) 3138 9264
ar.james at qut.edu.au
http://www.qut.edu.au/its/hpc
 
><(((º>  ._.·´¯`·..  >++(((º>  ._.·´¯`·..  >++(((º>

-----Original Message-----
From: David Gobbi [mailto:david.gobbi at gmail.com] 
Sent: Wednesday, 16 December 2009 10:59 AM
To: Allan James
Cc: vtkusers at vtk.org
Subject: Re: [vtkusers] Why is the image red?

I'd have to say that I don't know.  Are you willing to do some VTK
code profiling with the intel compiler?

   David


On Tue, Dec 15, 2009 at 5:42 PM, Allan James <ar.james at qut.edu.au> wrote:
>
> Hi David,
>
> Doesn't the use of pointer arithmetic within loops like this prevent certain compilers (for example - Intel) from performing automatic vectorization?
>
> I'm pretty sure the Intel compilers will ignore such loops - which is a shame because the performance increase can be significant..
>
> Cheers,
>
> Allan James
> High Performance Computing & Research Support
> Queensland University of Technology
> (07) 3138 9264
> ar.james at qut.edu.au
> http://www.qut.edu.au/its/hpc
>
>><(((º>  ._.·´¯`·..  >++(((º>  ._.·´¯`·..  >++(((º>
>
>>       On Mon, Dec 14, 2009 at 6:02 PM, David Doria <daviddoria+vtk at gmail.com> wrote:
>>       >
>>       > ? ? ?ptr = static_cast<T *>(image->GetScalarPointer(p0, p1, z));
>>       >
>>       > ? ? ?pf = drawColor;
>>       > ? ? ?// Assign color to pixel.
>>       > ? ? ?for (idxV = 0; idxV <= maxV; ++idxV)
>>       > ? ? ? ?{
>>       > ? ? ? ?*ptr = static_cast<T>(*pf++);
>>       > ? ? ? ?ptr++;
>>       > ? ? ? ?}
>>       >
>>       > Couldn't this be simply:
>>       >
>>       > // Assign color to pixel.
>>       > for (component = 0; component <= maxV; ++component)
>>       > ?{
>>       > ?image->SetScalarComponentFromDouble(p0, p1, z, component, drawColor[i]);
>>       > ?}
>>       >
>>       > This function is a template function which is called with
>>       > vtkTemplateMacro. I don't understand what that is doing - isn't the
>>       > whole point of templates that you don't have to declare all of these
>>       > different types explicitly?
>>       >
>>       > Thanks,
>>       >
>>       > David
>>
>>       All of the VTK imaging filters work by doing pointer arithmetic and
>>       direct access to memory.  It's the VTK convention.  It's possibly not
>>       the best way for things to be done, but after you're familiar with the
>>       common pointer arithmetic coding patters it's easy to read this code
>>       and see what it's doing.  Really, if the main problem that I see is
>>       that the variables weren't named very well.
>>
>>       The SetScalarComponentFromDouble() method isn't 100% inlined and it
>>       involves a virtual function call, so it's far too inefficient to use
>>       inside an image filter.  Not to mention that it does a few
>>       multiplications and additions internally.  Yikes.
>>
>>       The big "switch" in the vtkTemplateMacro allows the execute methods to
>>       be templated while the classes themselves aren't.  Because the filter
>>       classes themselves aren't templated, an instance of a vtk filter class
>>       can actually operate on any data type that you feed it.  In contrast
>>       with e.g. ITK where filter instances are type specific.
>>
>>          David
>
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the VTK FAQ at: http://www.vtk.org/Wiki/VTK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.vtk.org/mailman/listinfo/vtkusers
>