I ran across this article (and follow up article) today and had to read it though twice. It was simply great for a former graphics geek like me.


The article is basically about the history behind this little C hack:This code and its like are not seen in most modern programs. It may hide somewhere in the basement of a graphics library or physics engine but not in the application code.

The code is a simplified version of the Newton-Raphson's method applied to approximating 1/sqrt(x). It is an amazing little bit-twiddling hack that calculates a good "first guess" which optimizes the NR-Method so that it converges faster.


The article was great to read because it ran past the names of iconic names in graphics programming: John Carmack, Michael Abrash, Terje Mathisen (who I had never head of), and Gary Tarolli.


..you just don't see code like this anymore...

The *magic* behind the code is eventually ascribed to Clive Moler -- the inventor of MatLAB. Amazing little piece of code!

That is a pretty slick piece of code Nick. bit shifting is not utilized nearly enough in modern day coding. Also the reuse of the same variable allows for quick reading of variable values.
Knowing there's a better solution is one thing, finding it is another. I'll be adding this to my collection of math algorithms

Cool!

I wanted to post Duff's device, but couldn't remember it's name. Anyway, here it is, enjoy!

I hadn't see that before (even though I have both books mentioned... must be in the exercises ). I have actually used fall though in a similar manner (though I put my loop outside the switch -- this is much better) in a line drawing function. My program only broke it up to 4 conditions... there was some reason why but I don't remember what it was.

Anyway, I later learned faster ways to draw a line.

very cool little trick to keep in mind when you have a loop that the compiler can't really unroll for you.