On Sun, Mar 30, 2014 at 4:10 PM, Bob Friesenhahn <bfriesen@simple.dallas.tx.us> wrote:
On Sun, 30 Mar 2014, Glenn Randers-Pehrson wrote:
> Because of the limited payoff and because of potential problems with alignment and consequently
> possible security issues, I recommend not taking this.


This operation is highly unlikely to be one which shows up in a
profiler as taking significant time.

On the other hand, it might be worth optimizing png_do_bgr() and png_do_swap()
which get called for every row and loop over every pixel.

I'll ask the original poster to take a look at those.   I believe we can guarantee
that the row is 16-byte aligned and has at least 16 bytes of padding.