Re: math function accuracy -- was Re: text2interval again /
On 2013-03-12 12:05:07 -0500, Ralph Baker Kearfott wrote:
> On 03/12/2013 11:08 AM, Vincent Lefevre wrote:
> >But for 754-conforming P1788 implementations, you can add the
> >associated "tightest" requirements for each supported function
> >(this includes the required functions +, -, *, /, etc.).
I think that it should be exactly the required functions, since
it may not be obvious to know whether some recommended function
is supported or note (in particular if this can change from a
version of the math library to another one, or if there is a
platform dependent behavior).
It is important that implementation-defined functions do not have
to follow this requirement: there may exist a correctly-rounded
version while a tight interval version would not be computable.
> Personally, I like the idea of accuracy as specified in Table 8 tied to
> the 754 requirements the optional elementary functions,
> for 754 data type; that is, if a function is provided for
> a 754 data type, it shall be tight. For a general data type,
> a simple containment requirement could suffice. If there are no objections
> or controversy, I suspect John will write this into the draft
> and we will vote directly on the wording. Otherwise, someone
> should formulate and post motions.
Please take my above point into account.
> I personally think having several levels of accuracy for general
> data types could easily add complication and controversy and
> thus derail the standard at this point, unless we can do it
> clearly and simply.
I think that the standard shouldn't forbid different levels of
accuracy. But there doesn't need to be some special support in
P1788. The language would just provide different bindings...
--
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)