rint/nearbyint long-double

[Joseph S. Myers]

On Mon, 27 Feb 2012, David Miller wrote:

> Joseph, after your ring/nearbyint float fixes, the new tests you added
> to the math testsuite started failing for long double on sparc (both
> 32-bit and 64-bit).
> 
> I've tried to mimick your float fixes in the ldbl-128 routines.
> 
> Do you mind if I check this in to fix the testsuite failures?

I think this is fine according to the same logic I gave for my fix, 
although:

>  	if(j0<48) {
[...]
>  	} else if (j0>111) {

the code would look simpler (and more similar to the float case) if 
written with "j0 < 112" (i.e. the value might not already be an integer) 
for the first conditional and just "else" for the second (the case where 
if finite it's so large it's already an integer).

-- 
Joseph S. Myers
joseph@codesourcery.com