Re: [PATCH] libm: switch sf_log1p from double error routines to float

Corinna Vinschen <[email protected]> Mon, 2 Dec 2019 10:05:25 +0100

Newsgroups	gmane.comp.lib.newlib
Message-ID	<[email protected]>

On Nov 29 11:23, Keith Packard wrote:
> sf_log1p was using __math_divzero and __math_invalid, which
> drag in a pile of double-precision code. Switch to using the
> single-precision variants. This also required making those
> available in __OBSOLETE_MATH mode.
> 
> Signed-off-by: Keith Packard <[email protected]>
> ---
>  newlib/libm/common/math_errf.c | 2 --
>  newlib/libm/common/sf_log1p.c  | 4 ++--
>  2 files changed, 2 insertions(+), 4 deletions(-)
> 
> diff --git a/newlib/libm/common/math_errf.c b/newlib/libm/common/math_errf.c
> index 762fc2799..53c68b1cf 100644
> --- a/newlib/libm/common/math_errf.c
> +++ b/newlib/libm/common/math_errf.c
> @@ -75,7 +75,6 @@ __math_oflowf (uint32_t sign)
>    return xflowf (sign, 0x1p97f);
>  }
>  
> -#if !__OBSOLETE_MATH
>  HIDDEN float
>  __math_divzerof (uint32_t sign)
>  {
> @@ -89,4 +88,3 @@ __math_invalidf (float x)
>    float y = (x - x) / (x - x);
>    return isnan (x) ? y : with_errnof (y, EDOM);
>  }
> -#endif /* !__OBSOLETE_MATH */
> diff --git a/newlib/libm/common/sf_log1p.c b/newlib/libm/common/sf_log1p.c
> index d86768082..d326e00ba 100644
> --- a/newlib/libm/common/sf_log1p.c
> +++ b/newlib/libm/common/sf_log1p.c
> @@ -56,9 +56,9 @@ static float zero = 0.0;
>  	if (hx < 0x3ed413d7) {			/* x < 0.41422  */
>  	    if(ax>=0x3f800000) {		/* x <= -1.0 */
>  		if(x==(float)-1.0)
> -		    return __math_divzero (1); /* log1p(-1)=-inf */
> +		    return __math_divzerof (1); /* log1p(-1)=-inf */
>  		else
> -		    return __math_invalid (x);	/* log1p(x<-1)=NaN */
> +		    return __math_invalidf (x);	/* log1p(x<-1)=NaN */
>  	    }
>  	    if(ax<0x31000000) {			/* |x| < 2**-29 */
>  		if(two25+x>zero			/* raise inexact */
> -- 
> 2.24.0

Pushed.


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

signature.asc (application/pgp-signature, 833 B)