Funktion __ffsqrtf from "fpu.h" is not defined for 0

Rafeal,

That is interesting.  I just ran a simple test case.  __ffsqrt(0.0) returns the hex value 0x78F00000, which in floating point format maps to 1.#INF (according to CCS watch window).  The software team will need to take a look at this.  I suspect the problem has to do with the numerical method used to compute the sqrt.

For now, how about doing this:

 if(x == 0.0)
 {
  y = 0.0;
 }
 else
 {
  y = __ffsqrtf(x);
 }

I suppose the real issue here is how small can x be before the sqrt fails.  If you look at float.h in the compiler folder, it defines the smallest value as

#define FLT_MIN            1.175494351E-38F   /* SMALLEST POSITIVE VALUE   */

I tried __ffsqrtf() with this value, and it returns y=2.439455e-19, which is not quite correct (correct answer is 1.08420e-19).  This is no doubt due to numerical error.  At least the answer is a real value and not 1.#INF or something like that.  I also tried with x=1e-30, and it correctly returns y=1e-15.

I suppose you can pick a minimum value for x, and if x is smaller than that you return y=0.0.  Alternately, if x is smaller than the minimum, use the regular sqrt() function in the compiler RTS library instead, as controlled by your if() statement in your code.

Regards,

David