incorrect result from cos on non-x86 platform

[Paul Mackerras]

Ulrich Drepper writes:

> Philip Blundell wrote:
> > A Debian user reported that running this program on PowerPC or ARM,
> > using glibc 2.3.1:
> 
> I hope you verified that.
> 
> I passed the report on to the authors since I cannot see anything
> oviously wrong.

It happens on my powerbook (PowerPC laptop) running Debian.
Interestingly, it only happens for the one particular binary value:
0x3fe9a92cdb25a2e1, which is 0.80190127184058835.  Other binary values
in the vicinity don't provoke the bug.

Paul.