[Bug math/28472] pow(10, i) accuracy

["vincent-srcware at vinc17 dot net" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28472

Vincent Lefèvre <vincent-srcware at vinc17 dot net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vincent-srcware at vinc17 dot net

--- Comment #14 from Vincent Lefèvre <vincent-srcware at vinc17 dot net> ---
(In reply to b. from comment #13)
> (In reply to joseph@codesourcery.com from comment #12)
> > I don't consider special-casing 10 in pow to be reasonable.  
> As M Welinder pointed out in C#2 it's an IEEE recommendation to have these
> values correct, and - again - it's important in rounding other values.

What IEEE 754 recommends is that some math functions, like pow and exp10, be
correctly rounded. If a function is not correctly rounded, there are no
additional recommendations at all. In particular, IEEE 754 does not recommend
anything for some particular values, like powers of 10. Having special-casing
may be counter-productive, because this can make the function less smooth at
these points and/or break monotonicity (though monotonicity is not guaranteed).
What could be done is to make such functions more accurate, say ensure faithful
rounding for round-to-nearest (so with an error strictly less than 1 ulp of the
exact result and of the computed result), even though this would make them a
bit slower. Users who do not care about the accuracy could still write faster
(and less accurate) functions than those already provided, anyway.

-- 
You are receiving this mail because:
You are on the CC list for the bug.