Re: Decorated intervals ARE intervals...
> Subject: Re: Decorated intervals ARE intervals...
> From: John Pryce <j.d.pryce@xxxxxxxxxxxx>
> Date: Sat, 2 Jul 2011 14:49:58 +0100
> To: stds-1788 <stds-1788@xxxxxxxxxxxxxxxxx>
>
> Dan & all
>
> On 2 Jul 2011, at 01:04, Dan Zuras Intervals wrote:
> > I have stated this before & the current discussion only
> > strengthens my opinion that decorated intervals should
> > BE intervals. Full stop. Certainly at level 1 but as
> > far down as we can push it as well.
>
> It would be nice if the standard could take that view, but what
> about the "17-byte problem"? I.e., decorating the most popular
> kind of interval (a pair of doubles) looks as if it incurs severe
> storage and communication penalties.
>
> That's why the decoration system includes the Hayes/Neumaier
> "bare object arithmetic" scheme, which otherwise would be pointless.
>
> As someone who has long experience of how computer systems evolve,
> will you hazard to predict whether the 17-byte problem will be
> solved by hardware advances or software advances (so we needn't
> worry about it much); or will continue to be a problem?
>
> John
John,
You are giving me a level 4 objection to a level 1 problem.
But let me answer it all the same.
There are many solutions starting with aligning 17-byte
objects to 32-byte windows.
But the one we should be considering is the use of optimizing
compilers that recognise when the decoration may be neglected
(perhaps nearly always) & trims the decorated interval down
to 16-bytes as well as eliminating the decoration computation
that is known by the compiler to be unneeded.
It gives the implementers something to brag about when trying
to differentiate themselves from the competition & it is done
in a way that is both invisible to the user & irrelevant to
the goal of assured computing.
And, isn't assured computing our number 1 goal here?
For if we give the programmer ways to routinely defeat that
goal then none of our work is worth the effort.
Is that not so?
Dan