[Cfp-interest 3151] Re: midpoint function

[Vincent Lefevre]

On 2024-07-19 13:09:37 +0100, Mike Cowlishaw wrote:
[quoting David Hough]
> > In my view, the usual criterion is that the function is 
> > simple enough to be put into hardware that produces a 
> > correctly rounded result at a cost smaller 
> > than its value to applications.   If it's not critical to 
> > performance, or if it can't be correctly rounded in hardware at 
> > reasonable cost, it's not a good candidate.
> 
> I'm a little uncomfortable with this as a criterion.  Would it not exclude
> most trigonometric functions -- which are available in many programming
> languages, and in IEEE-754 too?

Change David's criterion a bit: An operation that is simple enough
to be implemented in hardware but whose software implementation
(using the available operations) would be significantly slower is
a good candidate.

-- 
Vincent Lefèvre <vincent at vinc17.net> - Web: <https://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <https://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)