Added special case for RGB to RGB and RGBA to RGBA,
where the loops are rewritten, transforming pointer arithmetic to indexing in order
to allow auto-vectorization by clang.
For the RGBA to RGBA case there was a noticeable gain of 3.2% in performance.