Standard Text Version 8.0 - 12.12.9 Numeric functions of intervals
My comments on 12.12.9 Numeric functions of intervals:
* In the first paragraph,
It shall return a result in a number format F that should be
compatible with T
is ambiguous. It could mean:
1. For any number format F such that the function returns a result
in F, F should be compatible with T.
2. There should exist a number format F compatible with T such that
the function returns a result in F.
Obviously, we want (2), but due to the way it is said in the current
document, this is not clear.
* It should also be clear that the number format F (or the set of
numbers formats F) is common to every numeric function.
* It seems that for rounding, the sign of an inexact zero is not
specified. IEEE 754-2008 says "Inexact numeric floating-point results
always have the same sign as the unrounded result." So, it could be
added: If F has signed zeros, an inexact zero has the same sign as
the unrounded result. Or something equivalent.
* For every numeric function, I wouldn't require correct rounding, except
if T is a 754-conforming type. Correct rounding (in the given direction)
is recommended. Satisfying the direction (except for mid) is required.
* For rad(x), if r = 0 and F has signed zeros, r shall be +0.
--
Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)