[Insight-developers] Overloading methods

[M Stauffer -V-]

Yes, thanks. It works when I explicitly pass the template parameter. But
in my case I don't want to have to do that because I'm modifying
CentralDifferenceImageFunction::Evaluate* methods, which doesn't seem
like a good method to change if not absolutely necessary.
 
-M


  _____  

From: Bradley Lowekamp [mailto:blowekamp at mail.nih.gov] 
Sent: Thursday, August 23, 2012 8:22 AM
To: M Stauffer -V-
Cc: Insight-developers at itk.org
Subject: Re: [Insight-developers] Overloading methods


Hello, 

The reason you methods is not working is because C++ can't deduce the
template parameters from the argument types passed to the function. That
is to say since T is only used for the return type, and not as an
argument to the function, so you should explicitly specify it as a
template parameters. Something like this should work:

r = this->EnableIfTest<OutputType>( 3 );

Brad


On Aug 22, 2012, at 4:08 PM, M Stauffer -V- wrote:


Great, thanks Brad.
 
I can reproduce what you explain below, but it's not working when I rely
on only the return parameter type for the template parameter deduction.
That is, the following won't compile
 
template< typename T >
typename EnableIfC< IsSame< T, ScalarDerivativeType >::Value, T >::Type
EnableIfTest(int val)
{
  std::cout << "Output == ScalarDerivativeType. val " << val << "\n" <<
std::endl;
  T o;
  o.Fill(1);
  return o;
}

(and similarly with DisableIfC...)
 
In implementation:
 
  OutputType r;
  r = this->EnableIfTest(3);

I think I'll skip this for now because I have a working implementation
and the extra overhead it has, because it requires a nested function
call, is only a few %.
 
I'll put up a patch soon and hopefully you could take a look at it,
Brad. I'll let you know.
 
-M
 
 
 


  _____  

From: Bradley Lowekamp [mailto:blowekamp at mail.nih.gov] 
Sent: Tuesday, August 21, 2012 3:37 PM
To: Michael Stauffer
Cc: Insight-developers at itk.org
Subject: Re: [Insight-developers] Overloading methods


Hello Michael, 

Here is the declaration of one of the GetComponent methods:

template< typename T>
    typename EnableIfC<
      IsSame<T, typename NumericTraits<T>::ValueType>::Value,
      T >::Type
  GetComponent(const T pix,
               unsigned int itkNotUsed( idx ) ) const;

This uses the EnableIf idom to answer lets look at a simplification of
function:

template< typename T>  T GetComponent(const T pix,  unsigned int idx )
const;

So I just simplified it some. So for function C++ has the ability to
implicitly deduce the template parameters from the arguments passed to a
function. So if you did the following:

int i;
GetComponent( i, 10 )

C++ can deduce the template parameter T, based on the function argument
i being of type int.


There are a couple  C++ tricks to overloading functions that should be
considered before the EnableIf idiom should be considered. Using the
EnableIf idom is really a last resort when simpler more understandable
techniques can not be used.


Brad

On Aug 17, 2012, at 6:17 PM, Michael Stauffer wrote:

I'm using the new EnableIfC and DisableIfC routines, to try and optimize
itkCentralDifferenceImageFunction::Evaluate* methods by specializing for
scalar and vector pixel types. I have it working using my own SFINAE
method I pulled off the web. But this, requires calling templated
subfunctions, which cost about 2% in overhead.




Looking at itkEnableIf.h and its usage in
PatchBasedDenoisingImageFilter, I was hoping to be able to call the
specialized Evalute* methods directly.


The method PatchBasedDenoisingImageFilter::GetComponent is are
template-specialized to use one of two versions depending on whether the
pixel is scalar or not. The method is templated.
But in
PatchBasedDenoisingImageFilter::ComputeSignedEuclideanDifferenceAndWeigh
tedSquaredNorm(), GetComponent is called without any template
parameters. How does this work? I must be missing something.




template 
void
PatchBasedDenoisingImageFilter
::ComputeSignedEuclideanDifferenceAndWeightedSquaredNorm(const
PixelType& a, const PixelType& b,
const RealArrayType& weight,
bool itkNotUsed(useCachedComputations),
SizeValueType itkNotUsed(cacheIndex),
EigenValuesCacheType& itkNotUsed(eigenValsCache),
EigenVectorsCacheType& itkNotUsed(eigenVecsCache),
RealType& diff, RealArrayType& norm)
{
for (unsigned int pc = 0; pc < m_NumPixelComponents; ++pc)
{
RealValueType tmpDiff = GetComponent(b, pc) - GetComponent(a, pc);
RealValueType tmpWeight = weight[pc];
SetComponent(diff, pc, tmpDiff);
norm[pc] = tmpWeight * tmpWeight * tmpDiff * tmpDiff;
}
}





========================================================
Bradley Lowekamp  
Medical Science and Computing for
Office of High Performance Computing and Communications
National Library of Medicine 
blowekamp at mail.nih.gov






========================================================

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/20120823/56dbc5dc/attachment.htm>