Jump to content

  • Log In with Google      Sign In   
  • Create Account


Partial template specialization


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
7 replies to this topic

#1 Misery   Members   -  Reputation: 305

Like
0Likes
Like

Posted 24 April 2012 - 07:22 AM

Hello,

I am trying to create a matrix template class. I would like Sum() function to act differently for bool types. What I need to make it elegant class is partial template specialization. How do I do it?


template <class DATA,class INT_TYPE>
class TMatrix
{
  private:
	 DATA *d;
	 INT_TYPE rows,cols;
	
  public:
	DATA Sum(); //count the sum of element
// for bool data type I would like to get:
   INT_TYPE Sum();  //counts the number of elements that are true, so it cannot return bool, but INT_TYPE instead
}


And how do I create definition of such function outside the class?

Thanks in advance,
Regards,
Misery

Sponsor:

#2 rip-off   Moderators   -  Reputation: 7641

Like
2Likes
Like

Posted 24 April 2012 - 08:23 AM

You could do something like this:
template<typename T>
struct sum_traits;

template<typename T>
class matrix {
public:
	typedef typename sum_traits<T>::sum_type sum_type;

	sum_type sum() const;
};
The nice thing about this is that it is general - the sum_type for char/short should probably be longer than the data type itself - at least a full int if not a long.

Example:

#include <vector>
#include <cstdlib>
#include <iostream>

template<typename T>
struct sum_traits;

template<typename T>
class example {
public:
	void add(T item) {
		data.push_back(item);
	}

	typedef typename sum_traits<T>::sum_type sum_type;

	sum_type sum() const {
		sum_type result = 0;
		for(int i = 0 ; i < data.size() ; ++i) {
			result += data[i];
		}
		return result;
	}

private:
    std::vector<T> data;
};

template<>
struct sum_traits<int> {
	typedef long sum_type;
};

template<>
struct sum_traits<bool> {
	typedef int sum_type;
};

int main() {
	example<int> a;
	for(int i = 0 ; i < 10 ; ++i ) {
		a.add(std::rand());
	}
	a.add(0xffffff00);

	std::cout << "Sum<int>: " << a.sum() << '\n';

	example<bool> b;
	for(int i = 0 ; i < 10000 ; ++i) {
		b.add(std::rand() > (RAND_MAX / 2));
	}
	std::cout << "Sum<bool>: " << b.sum() << '\n';
}


#3 Misery   Members   -  Reputation: 305

Like
0Likes
Like

Posted 25 April 2012 - 05:14 AM

It looks qute nice, but i would like to use usual template specialization in manner like:
	template <class INT_TYPE>
   INT_TYPE Matrix<bool,INT_TYPE>::Sum()
	{
	//sum all true elements
	}

How doI do that? My compiler always returnsan error saying that this function template requires two elements.

#4 SiCrane   Moderators   -  Reputation: 9386

Like
1Likes
Like

Posted 25 April 2012 - 06:11 AM

You can't specialize individual member functions of a template class without specializing the entire class.

#5 Misery   Members   -  Reputation: 305

Like
0Likes
Like

Posted 25 April 2012 - 06:12 AM

Thanks, so I'll find something else :]

#6 japro   Members   -  Reputation: 887

Like
1Likes
Like

Posted 25 April 2012 - 06:48 AM

A somewhat cheap trick to get around "specialize the whole class" thing is to create a helper class that only contains the sum functionality and inherit from that, that way you only have to specialize that one. The other way around works too obviously put the non special functionality into a base class and only specialize classes that inherit from it.

#7 rip-off   Moderators   -  Reputation: 7641

Like
1Likes
Like

Posted 25 April 2012 - 07:08 AM

Well, thinking on this again, the simpler solution that might suffice is default template arguments:

#include <vector>
#include <cstdlib>
#include <iostream>

template <typename type, typename size_type, typename sum_type = type>
class example
{
public:
    sum_type sum() const {
        sum_type result = 0;
        for(int i = 0 ; i < data.size() ; ++i) {
            result += data[i];
        }
        return result;
    }

    void add(type item) {
        data.push_back(item);
    }


private:
    std::vector<type> data;
};

int main() {
        example<int> a;
        for(int i = 0 ; i < 10 ; ++i ) {
                a.add(std::rand());
        }
        a.add(0xffffff00);

        std::cout << "Sum<int>: " << a.sum() << '\n';

        example<bool> b;
        for(int i = 0 ; i < 10000 ; ++i) {
                b.add(std::rand() > (RAND_MAX / 2));
        }
        std::cout << "Sum<bool>: " << b.sum() << '\n';
}
You can now create a typedef for bool_matrix as something like matrix<bool, size_t, size_t>.

#8 Misery   Members   -  Reputation: 305

Like
0Likes
Like

Posted 25 April 2012 - 11:59 PM

Actually, at last I decided to derive BoolMatrix class from Matrix class. The main problem I was trying to work around was that I wanted to use BoolMatrix argument in one of the Matrix methods. And the second problem was that I needed to make it possible to change the INT_TYPE used to matrix indexing, because in MSVC OpenMP works only on signed long int and I also use GCC and Intel Compiler.
Anyway, as always, thanks to You Guys I have tested a few very clever solutions and achieved the goal.
Thanks and regards Posted Image




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS