Timothy Lochner

What was your method for sqrt approximation?

I will have to go back and look. Been long enough that I don’t recall. I’ll follow up when I can get to my computer.

It looks like I settled on a recursive newtonian approximation. Code is similar to this: https://gist.github.com/alexshtf/eb5128b3e3e143187794 … FWIW, I didn't bother trying to make it fast. I wanted constexpr and never got around to making a faster approximation.

For a real-time approximation, I would probably unroll it to be non-recursive, and iterate a fixed number of times, probably templating on # of iterations so that I could have a few version. sqrt_fast and sqrt_precise, for instance.

Do you have a fast sqrt() function (or approximation) for double precision floats? All the ones I'm finding online are for 32 bit, or for the inverse sq. root which is what I don't want to calculate.

I don’t. You’re right about there being a lot of 32 bit ones, and they usually rely on UB, reinterpret-casting and bit shifting for their speed. Sorry.

For instance, if you iterate sqrt8 (babylonian method) a fixed number of times instead of comparing to see if it's close enough, you may get a reasonable speed/precision tradeoff.