evil floating point bit level hacking

[dm319]

I was digging into the fast inverse square again last night, and made the personal discovery that Kahan was connected to this.

The algorithm reminded me so much of the CORDIC algorithm, in that it is equal parts a mixture of maths and computer science hacking.

And it ties in with the TVM stuff too, in that Kahan had a hand in improving the algorithm for the HP-27 and later, and also that Newton's method is used for the fast inverse square and often used to solve for i.