[ace123]

If you are only using the sqrt function, you might like to see the implementation of InvSqrt():
http://www.beyond3d.com/articles/fastinvsqrt/

here is an implementation that I compiled on my first controller (the scary constant is a standard IEEE754 floating point number, so it will work on every microprocessor the same way).

I would definitely extensively test this method before putting it to use (i.e. don't hook up motors until you are sure it gives the expected answers), but I believe that it is just a first-order Newton's method approximation, with the constant being a first guess I believe.

I'm pretty sure it will take up less space than the included math library.

float invSqrt(float magnitudeSquared) {
    float xhalf;
    float temp;
    long i;
    
    // magic follows:
    temp = magnitudeSquared;
    xhalf = 0.5f*temp;
    i = *(long*)&temp;
    i = 0x5f3759df - (i>>1);
    temp = *(float*)&i;
    temp = temp*(1.5f - xhalf*temp*temp);
    
    // temp now contains the 1/sqrt(magnitudeSquared)
    return temp;
}

Definitely works on the computer:

5629.000:     InvSqrt 0.013318; 1/sqrt 0.013329
281.0000:     InvSqrt 0.059639; 1/sqrt 0.059655
28.00000:     InvSqrt 0.188722; 1/sqrt 0.188982
6.000000:     InvSqrt 0.407681; 1/sqrt 0.408248
4.000000:     InvSqrt 0.499154; 1/sqrt 0.500000
2.017857:     InvSqrt 0.703756; 1/sqrt 0.703971
0.250000:     InvSqrt 1.996614; 1/sqrt 2.000000
0.166667:     InvSqrt 2.445322; 1/sqrt 2.449490
0.000000:     InvSqrt 19817753709685768192; 1/sqrt inf