This is the mail archive of the newlib@sourceware.org mailing list for the newlib project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: long double (was "strtold?")


On Thursday 02 April 2009 16:25:36 Howland Craig D (Craig) wrote:
> Ken Werner wrote Thursday, April 02, 2009 8:59 AM:
> >Why would anyone with a platform where there's no difference between
> > double and long double prefer the long double routine?
>
> The short, perhaps cryptic answer, is, when the size of double does
> not match the double functions but does match the long double functions.
>
> To elaborate, using some slightly different diction to hopefully make it
> clear, the Newlib math library currently has two set of math routines.
> The first works on IEEE binary64, and is normally named for the double
> functions.  The second set works on IEEE binary32 and is named for the
> float functions.  On platforms where double is binary32, the double
> functions are defined to be wrappers calling the hard-coded binary32
> ("float") routines, instead of calling the binary64 routines.
>
> If there ever were a platform on which double were neither binary64 nor
> binary32, Newlib would presently be out of business.  But, if at that
> time we had dual-use binary80/128 routines that were normally named for
> the long double functions and double were 80 or 128, then the double
> functions could be wrappers to the binary80/128 ("long double")
> routines.
>
> Craig

Thanks for your explanation. It should be sufficient to add these switches when 
the real long double code gets implemented I think.

Ken


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]