Optimize unionfind

[bouk]

I figured out that a big part of the union find time spent was initializing the data array. I've optimized it by using two memsets instead of an array, which saves ~4% for an image sized 4896×3264

Before

After

[christian-rauch]

Is this essentially saving you 5ms on the 4896×3264 image? That's a nice improvement.

Could you explain in the commit message a bit what the change is doing and why it improves the performance? Glancing over this, Isee it's moving the parent and size from the struct ufrec into the struct unionfind, turning them into a point for arrays and then initialising all elements in the list at once instead of looping over the array of struct ufrec?

[bouk]

Expanded the commit message a bit. The savings is indeed from using memset instead of a for loop

[christian-rauch]

Can you remove the concrete ~4% from the commit message again? I guess this value depends a lot on the image resolution and is not the same improvement in all cases.

[bouk]

Sure, removed

[christian-rauch]

Thanks!