On Mon, Sep 20, 2010 at 05:52:06PM -0400, bearophile wrote:
> Recently Bradley Mitchell in D.learn newsgroup has tried to implement the
> Quake fast inverse square root algorithm in D, and has found D lack the C++
> reinterpret cast:
> http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.D.learn&article_id=21901
Did anyone suggest porting the C++ line for line?
int xi = *reinterpret_cast<int *>( &x );
and
int xi = cast(int)cast(void*)x;
aren't the same thing!
Maybe he tried this, but webnews is crap and won't load up the rest of that
thread.
It seems like
int xi = * cast(int*) (&x);
should do the trick. And same for the cast back to float.
So the final program is:
============
import std.stdio;
float fastInvSqrt( float x )
{
const int INV_SQRT_N = 1597292357;
const float MULT = 1.000363245811462f;
float mx = 0.5f * MULT * x;
int xi = * cast(int*) (&x); /// note this line!
xi = INV_SQRT_N - (xi >> 1);
x = xi;
x = * cast(float*) (&xi); /// this one too!
return x * (1.5f * MULT - mx * x * x);
}
void main(string[] args)
{
float a = fastInvSqrt( 9.0f );
writefln("%f", a);
}
===========
Running it gives the same result as the C++ version on my box.
This looks like human error in the porting to me.
