I was educated. From that article I learned that if you bit copy a 32 bit IEEE754 float to a DINT and then right shift the DINT 1 bit, then subtract that from the magic hex number 5F3759DF , then take that result and bit copy it back to a float and square it, then multiply it by 1/2 the original float and subtract result from 1.5 and multiply that by itself again, you get a very good first approximation of the inverse square root of the float.
Pretty cool magic trick. I could totally use that somewhere.