That one is not that complicated if you don’t think about the math. It’s basically just if we interpret the float as int and add a magic number we have a good estimation.
From what I remember at least, it’s been a little while since I implemented it.