Question

I am kind of startled, especially after reading this.

I use

template <class T>
int GetPosition(vector<T> mVec, T element)
{
    return find(mVec.begin(), mVec.end(), element) - mVec.begin();
}

and

template <class T>
int GetPosition(map<T, int> mMap, T element)
{
    return mMap.find(element)->second;
}

as template functions to get the index of a specific element in my vector respectivly list.

Elements are unique pointers to objects, of which i want to retrieve the index off.

I then use this template in a for-loop like

for(int i = 0; i < myCount; i++)
{
  index = GetPosition(myVector, elements[i]) //or GetPosition(myMap, elements[i])
}

While all bits of information i gathered suggested to use a map, the map implementation is several orders of magnitude slower: 57 ms using the vector variant in comparision 70000ms using the map.

Something is severly borked here, but i do not know what. Do you?

Development plattform is MS VS 2008 Standard sp1, on windows XP

Answer 1

Since you are passing them by value, you are coping the vector and map in every call you are making. I’d sat that this makes the results pretty much meaningless.

Pass them as reference or const reference and run the test again.

template <class T>
int GetPosition(const vector<T>& mVec, T element)
{
    return find(mVec.begin(), mVec.end(), element) - mVec.begin();
}

template <class T>
int GetPosition(const map<T, int>& mMap, T element)
{
    return mMap.find(element)->second;
}