On Wed, Oct 21, 2015 at 5:55 PM alex
What do you mean by not evaluating decreased? Then we use decreased to check which visitor function we should call, so I am not sure if one could get rid of that without having two versions of the code (one for the "classical" version, and one for an arbitrary distance map).
Sorry I am not clear. Yes, I do think that you should get rid of the redundant check. And, yes that would mean that if you want the arbitrary distance map, you should have two versions.
I would only agree with that if we could demonstrate that there is a statistically significant performance advantage of doing so. Looking at the code, I am skeptical that this check has much or any impact on performance on most modern platforms. Do you have any evidence that shows there would be a benefit to having two versions of the code?