shared_ptr<T[]> example

// Initially this:
unique_ptr<BYTE[]> paddedInput(add_pad_border(pImg, nWidth, nHeight, nPatch));
// and in the main loop several calls to
inline void ExtractPatch(T const * const from, int const x, int const y, int const yStep, int const length, T * const to) {/* ... */}
// like this:
ExtractPatch(paddedInput.get(), x, y * extendedWidth, extendedWidth, patchLength, w1);
////////////////////////////// 10.7s

// Then changed to:
BYTE const * const paddedInput_ = add_pad_border(pImg, nWidth, nHeight, nPatch);
vector<BYTE> paddedInput(paddedInput_, paddedInput_ + (nWidth + 2 * nPatch) * (nHeight + 2 * nPatch));
delete [] paddedInput_;
// With the same inline function called like this:
ExtractPatch(&paddedInput[0], x, y * extendedWidth, extendedWidth, patchLength, w1);
////////////////////////////// 9.4s

// Lastly with vector parameter
inline void ExtractPatch(std::vector<T> const & from, int const x, int const y, int const yStep, int const length, T * const to) {/* ... */}
// and calls like:
ExtractPatch(paddedInput, wx - nPatch, windowRow , extendedWidth, patchLength, w2);
////////////////////////////// 14.9s

// All in release mode with 8 cores fully utilized.