Re: text2interval again / a digression on fixed-point iterations, Mathematica Intervals, point-value vs interval constants
Richard
On 8 Mar 2013, at 19:25, Richard Fateman wrote:
> I found a sqrt routine that makes an effort to find out the rounding mode for one step in
> its calculation, but, as far as I can tell with a cursory inspection, it is not doing so to
> affect the final result, but to converge to the expected round-to-nearest regardless of
> the existing rounding mode.
I don't know what is actually out there for sin/cos/exp..., but it is quite explicit in 754-2008 that a conforming floating point shall support all the standard rounding modes for sqrt, see 2nd para of §4.3, and §5.4.1. Further, §9.1 seems to specify that *if* a conforming implementation provides any function in Table 9.1 including sin/cos/exp, it shall support all the standard rounding modes for it.
John