Possible ancient floating point patent workaround?
Reading around at another populair nerd site (yes, that one), a reader posted a link to the fast inverse squire root calculation used in the Quake3 source code.
What's so special about this calculation is that it performs a floating point calculation in an integer on a x86 by means of assembly language.
It appears to be so ancient that it was used even before 3dfx came around. (I smell prior art)
Now I was wondering if the same trick could somehow be used on modern GPU's...
PS: it can even be tweaked to deliver up to as much as 48 decimal precision without a table and still be twice as fast as floating point.