As I understand it, where there is overlap between MKL and IPP functions, the IPP functions are optimized to smaller sizes than the MKL functions.

My question is, where is that line? (both for vector/buffer operations and for image/matrix operations)