Re: [PATCH] fix llrint and lrint for 52 <= exponent <= 62

["Matthias J. Kannwischer" <matthias at kannwischer dot eu>]

Hi Corinna, 

sure, it's attached - sorry for that. 

Best, 

Matthias 

On Tue, May 29, 2018 at 03:25:21PM +0200, Corinna Vinschen wrote:
> Hi Matthias,
> 
> On May 14 14:38, Matthias J. Kannwischer wrote:
> > lrint and llrint produce wrong results if the exponent is between 52 and 62. 
> > e.g. llrint(2.308393893084546048E18) returns 146666059061806080 instead of 2308393893084546048. 
> > Can you confirm this?
> > I've tracked this down to two mistakes in s_llrint.c/s_lrint.c (one missing 0 and one uint32 overflow), see patch below. 
> > 
> > Cheers, 
> > Matthias
> > 
> > ---
> >  newlib/libm/common/s_llrint.c | 4 ++--
> >  newlib/libm/common/s_lrint.c  | 4 ++--
> >  2 files changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/newlib/libm/common/s_llrint.c b/newlib/libm/common/s_llrint.c
> > index 8b8a846ae..72452dbe9 100644
> > --- a/newlib/libm/common/s_llrint.c
> > +++ b/newlib/libm/common/s_llrint.c
> > @@ -93,9 +93,9 @@ long long int
> >        if (j0 >= 52)
> >         /* 64bit return: j0 in [52,62] */
> >         /* 64bit return: left shift amt in [32,42] */
> > -        result = ((long long int) ((i0 & 0x000fffff) | 0x0010000) << (j0 - 20)) | 
> > +        result = ((long long int) ((i0 & 0x000fffff) | 0x00100000) << (j0 - 20)) |
> >                 /* 64bit return: right shift amt in [0,10] */
> > -                   (i1 << (j0 - 52));
> > +                   ((long long int) i1 << (j0 - 52));
> >        else
> >          {
> >           /* 64bit return: j0 in [20,51] */
> > diff --git a/newlib/libm/common/s_lrint.c b/newlib/libm/common/s_lrint.c
> > index 9d2cb7306..b37f50fd4 100644
> > --- a/newlib/libm/common/s_lrint.c
> > +++ b/newlib/libm/common/s_lrint.c
> > @@ -131,9 +131,9 @@ TWO52[2]={
> >        if (j0 >= 52)
> >         /* 64bit return: j0 in [52,62] */
> >         /* 64bit return: left shift amt in [32,42] */
> > -        result = ((long int) ((i0 & 0x000fffff) | 0x0010000) << (j0 - 20)) | 
> > +        result = ((long int) ((i0 & 0x000fffff) | 0x00100000) << (j0 - 20)) |
> >                 /* 64bit return: right shift amt in [0,10] */
> > -                   (i1 << (j0 - 52));
> > +                   ((long int) i1 << (j0 - 52));
> >        else
> >          {
> >           /* 32bit return: j0 in [20,30] */
> > -- 
> > 2.14.3
> > 
> 
> Looks right to me, but can you resend the patch as attachment, please?
> Your MUA apparently replaced tabs with spaces.
> 
> 
> Thanks,
> Corinna
> 
> -- 
> Corinna Vinschen
> Cygwin Maintainer
> Red Hat

Attachment: 0001-fix-llrint-and-lrint-for-52-exponent-62.patch
Description: Text document

Attachment: signature.asc
Description: PGP signature