Re: Lack of fabsf on Solaris (patch included)

[Cedric Berger <cedric at wireless-networks dot com>]

Martin Kahlert wrote:
> 
> On Thu, Jan 25, 2001 at 05:47:23AM -0800, Cedric Berger wrote:
> > Why not simply this:
> >
> >    jfloat java::lang::Math::abs(jfloat x)
> >    {
> >       return x < 0.0F ? -x : x;
> >    }
> >
> > And removing all this autoconf mess?
> 
> Would be nice. But i don't know, if it's correct:
> 
> Are the float types symmetric?

Yes, they are (bit 31 is the sign bit). 
This could suggest a more efficient code, if the compiler is smart
enough to keep x in a register:
	
   jfloat java::lang::Math::abs(jfloat x)
   {
        return *((uint32_t *)&x) ^ 0x80000000U;
   }

 
> it gets inlined in gcc on some OSes (x86 for example).

Good point, I didn't knew that.

Cedric