[Mesa-dev] [PATCH 1/2] mesa: add signbit() macro

[Jose Fonseca]

----- Original Message -----
> On 09/24/2012 02:53 PM, Jose Fonseca wrote:
> >> From signbit manpage:
> >
> >    This is not the same as x<  0.0, because IEEE 754 floating point
> >    allows zero to be signed.   The  comparison  -0.0<   0.0  is
> >       false, but signbit(-0.0) will return a nonzero value.
> >
> > I think that for consistency (and especially because MSVC ends not
> > being as widely tested as gcc), we should match the spec. This can
> > easily be done using an inline and an union.
> 
> The DIFFERENT_SIGNS and IS_NEGATIVE macro are only used in the tnl
> clipping code and I don't think the signbit() of 0 vs -0 would make
> any real difference in the end.
> 
> But I see your point.  Who knows where signbit might be used in the
> future and consistency is good.  The question then is how closely do
> we want to follow Linux's example?  The code in math.h checks
> sizeof(x) to handle float and double (and long double) specially:
> 
> /* Return nonzero value if sign of X is negative.  */
> # ifdef __NO_LONG_DOUBLE_MATH
> #  define signbit(x) \
>       (sizeof (x) == sizeof (float) ? __signbitf (x) : __signbit (x))
> # else
> #  define signbit(x) \
>       (sizeof (x) == sizeof (float)					      \
>        ? __signbitf (x)							      \
>        : sizeof (x) == sizeof (double)					      \
>        ? __signbit (x) : __signbitl (x))
> # endif

I see, it's polymorphic macro, that works with all float types...

One MSVC implement one can often see in google is 

  #  define signbit(x) (_copysign(1.0, (x)))

floats are converted to double so it is correct for at least float and doubles (and I don't think we'll ever need to care for long doubles). I'm not sure if it is slower for single precision due to the double converstion though...

Jose