[PATCH v4 06/21] math: Remove the SVID error handling from remainderf

[Adhemerval Zanella Netto]

On 29/10/25 13:18, Wilco Dijkstra wrote:
> Hi Adhemerval,
> 
> Same comments as for remainder:
> 
>        if (__glibc_unlikely (hx >= EXPONENT_MASK))
> -	return (x * p) / (x * p);
> +	return __math_invalidf (x);
>  
> If is redundant.

As for remainder, I had to keep it to avoid a 'isnan' check after
the fmod operation to avoid FE_INVALID if 'x' is NaN.

> 
>        x = fabs (__ieee754_fmodf (x, p + p)); /* now x < 2p */
> 
> __fmodf

Ack.

> 
>        if (x + x > p)
> @@ -53,8 +55,12 @@ __ieee754_remainderf(float x, float p)
>    else
>      {
>        /* |x| not finite or |y| is NaN or 0 */
> -      if ((hx >= EXPONENT_MASK || (hp - 1) >= EXPONENT_MASK))
> -	return (x * p) / (x * p);
> +      if (__glibc_unlikely (hx >= EXPONENT_MASK || (hp - 1) >= EXPONENT_MASK))
> +	{
> +	  if (__glibc_unlikely (is_inf (hp)))
> +	    return __math_invalidf (x);
> +	  return (x * p) / (x * p);
> +	}
> 
> Should be:
> 
> if (__glibc_unlikely (hx >= EXPONENT_MASK || hp > EXPONENT_MASK))
>   return __math_invalidf (x * p);

Ack.

> 
> Cheers,
> Wilco