Why C++ template code runs faster in my lambdas?

Question

I've been doing some simple vector data accumulate tests and wondering why the following code:

template<typename T>
T accum(typename vector<T>::const_iterator first, typename vector<T>::const_iterator last, T init) {
  for(; first != last; ++first)
    init += *first;
  return init;
}

runs faster than the following code (both in lamda) ?

test("Iterator/direct", table, [](auto& values) {
  auto sum {0.0};
  for(auto i = values.begin(); i != values.end(); ++i)
    sum += *i;
  return(sum);
});

In my eyes the algorithm itself should be quite much the same.

The whole source code (53 lines) can be found in this Gist .

Answer 1

Compile with -O2 option and the execution times will be practically negligible. Thank you very much for all commenters.