Re: [PATCH] Fix ceill() for ldbl-96

[Andreas Jaeger <aj at suse dot de>]

Is this patch now correct or not?  The testcase I gave showed that it
fails ceill(0.25)...

Andreas

Gwenole Beauchesne <gbeauchesne@mandrakesoft.com> writes:

> Hi,
>
> ceil(x) is defined so that ceil(x) = n / n - 1 < x <= n, where n is
> integral, x a real number. The following patch fixes this for ldbl-96 
> targets (e.g. AMD64), in particular ceill(0.25).
>
> 2003-11-12  Gwenole Beauchesne  <gbeauchesne@mandrakesoft.com>
>
> 	* sysdeps/ieee754/ldbl-96/s_ceill.c (__ceill): Fix 0 < x < 1 case.
> 	
> --- glibc-2.3.2/sysdeps/ieee754/ldbl-96/s_ceill.c.ceill-fix	2000-10-01 00:01:23.000000000 +0200
> +++ glibc-2.3.2/sysdeps/ieee754/ldbl-96/s_ceill.c	2003-11-12 23:45:59.000000000 +0100
> @@ -50,9 +50,10 @@ static long double huge = 1.0e4930;
>  	j0 = (se&0x7fff)-0x3fff;
>  	if(j0<31) {
>  	    if(j0<0) { 	/* raise inexact if x != 0 */
> -		if(huge+x>0.0) {/* return 0*sign(x) if |x|<1 */
> -		    if(sx) {se=0x8000;i0=0;i1=0;}
> -		    else if((i0|i1)!=0) { se=0x3fff;i0=0;i1=0;}
> +		if(huge+x>0.0) {
> +		  /* return -0.0 if -1 < x < 0, 1.0 if 0 < x < 1.0 */
> +		  if(sx) {se=0x8000;i0=0;i1=0;}
> +		  else if((i0|i1)!=0) { se=0x3fff;i0=0x80000000;i1=0;}
>  		}
>  	    } else {
>  		i = (0x7fffffff)>>j0;
>
>

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

Attachment: pgp00000.pgp
Description: PGP signature