>   However, I should point out that the most common reason that
>   I have seen for wanting square root is to compute
>
>             sqrt (x*x + y*y)
>
>   (Eculidian Distance).
>   If anybody want's square root for that, they should be aware
>   that the full operation can be done with noticibly lower
>   overhead then by doing the multiplies, adds, and then
>   the square root.

Cool... care to elaborate?  Are there any really good approaches for cases
where, e.g., x and y are 8 bit quantities and where perfect rouding is not
required?  How much complexity do different degrees of refinement add?