precision of bfloat functions



Sheldon Newhouse wrote:
> Hello,
>  Is it the case that the so-called intrinsic transcendental functions 
> (line exp, log, sin, etc) are accurate to the ulp in the various precisions?
>
>   
I think the routines try to be accurate to 1 ulp (or less), but I'm not
sure.  I think it's fairly difficult for things like sin of large
arguments where you need many digits of pi to do argument reduction
correctly.

Ray