[bitm0de]

@bitm0de: Can you explain what you mean with "overlapping buffers" ?
                      I think your talking about something totally different ..

Nope. I'm talking specifically about copying memory from src to dst when those memory regions are overlapping. 'rep movsb' was not designed for this. A better implementation would check the offsets to make sure that this isn't the case, and use some other looping method perhaps to copy bytes from src to dst instead in such a case.