[PATCH] Fix modf/f for NaN input

Corinna Vinschen vinschen@redhat.com
Thu Mar 19 15:43:07 GMT 2020


On Mar 18 14:18, Fabian Schriever wrote:
> For NaN input the modf/f procedures should return NaN instead of zero
> with the sign of the input.
> ---
>  newlib/libm/common/s_modf.c  | 1 +
>  newlib/libm/common/sf_modf.c | 1 +
>  2 files changed, 2 insertions(+)
> 
> diff --git a/newlib/libm/common/s_modf.c b/newlib/libm/common/s_modf.c
> index 8551a99e4..c948b8525 100644
> --- a/newlib/libm/common/s_modf.c
> +++ b/newlib/libm/common/s_modf.c
> @@ -100,6 +100,7 @@ static double one = 1.0;
>  	} else if (j0>51) {		/* no fraction part */
>  	    __uint32_t high;
>  	    *iptr = x*one;
> +	    if (__fpclassifyd(x) == FP_NAN) return x+x; /* x is NaN, return NaN */
>  	    GET_HIGH_WORD(high,x);
>  	    INSERT_WORDS(x,high&0x80000000,0);	/* return +-0 */
>  	    return x;
> diff --git a/newlib/libm/common/sf_modf.c b/newlib/libm/common/sf_modf.c
> index 6c64e3fa0..ae970762b 100644
> --- a/newlib/libm/common/sf_modf.c
> +++ b/newlib/libm/common/sf_modf.c
> @@ -52,6 +52,7 @@ static float one = 1.0;
>  	} else {			/* no fraction part */
>  	    __uint32_t ix;
>  	    *iptr = x*one;
> +	    if (__fpclassifyf(x) == FP_NAN) return x+x; /* x is NaN, return NaN */
>  	    GET_FLOAT_WORD(ix,x);
>  	    SET_FLOAT_WORD(x,ix&0x80000000);	/* return +-0 */
>  	    return x;
> -- 
> 2.24.1.windows.2
> 

Pushed.


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://sourceware.org/pipermail/newlib/attachments/20200319/927c7215/attachment.sig>


More information about the Newlib mailing list