Re: Quick inverse square root

["Jeffrey R. Carter" <spam.jrcarter.not@spam.not.acm.org>]

On 1/2/21 11:26 PM, Matt Borchers wrote:
> 
>      generic
>          type F is digits <>;
>          type U is mod <>;
>          magic : U;
>      function G_Q_INV_SQRT( a : F ) return F;
> 
> I write the body like this:
> 
>      function G_Q_INV_SQRT( a : F ) return F is
>          y : F := a;
>          i : U;
>          for i'Address use y'Address;
>      begin
>          i := magic - shift_right( i, 1 );
>          return y * (1.5 - (0.5 * a * y * y));
>      end G_Q_INV_SQRT;
> 
>      function QUICK_INVERSE_SQRT is
>          new G_Q_INV_SQRT( FLOAT, UNSIGNED_32, 16#5F3759DF# );
> 
> This won't compile because the type U is not valid for the call to "shift_right".  How do I overcome this obstacle?

Make it an explicit generic formal function parameter:

    with function Shift_Right (...) return ...;

> Once that is overcome, is there a way I can eliminate having to pass in the unsigned type along with the floating point type?  That seems like the programmer would require internal knowledge to make use of the generic.  Any thoughts on how to get the compiler to compute the magic number in the generic at compile time?

You would want to make use of the attributes of floating point types in ARM A.5.3

http://www.ada-auth.org/standards/rm12_w_tc1/html/RM-A-5-3.html

Whether these provide the information you need is another question. I don't see 
how you could declare the modular type in the generic.

-- 
Jeff Carter
"One day he told me he was a gynecologist. He
couldn't speak no foreign languages."
Take the Money and Run
147