Re: Motion P1788/M0045.02:ExactDotProductRevision -- revised text
Marco, P1788
On 4 Jul 2013, at 15:40, Marco Nehmeier wrote:
> I'm not happy with the sentence
>
> "If such correctly rounded operations are provided by the underlying 754 system, these shall be used; otherwise they shall be provided by the implementation."
>
> in the first paragraph.
>
> This implies that a P1788 implementation with its own implementation of these four operations
> is not 754-conforming on a system which provides these four operations by itself.
> And this would avoid improvements and optimizations...
> E.g. on a system with a poorly (in terms of performance/speed) implementation of these functions
> we are not allowed to use optimized algorithms and call it 754-conforming.
>
> In my opinion this part of the standard should be written that it ensures that these four operations
> are provided correctly rounded... and nothing more.
> If they are provided by the underlying system, by a 3rd party library, or by a piece of code of our
> own implementation is negligible
["irrelevant"? JDP]
> ...
Actually, I agree. The original wording
> An implementation that provides 754-conforming interval types shall provide the four reduction operations sum, dot, sumSquare and sumAbs of IEEE 754-2008 §9.4, correctly rounded. These shall be provided for the parent formats of each such type.
is better and I will go back to it.
If there is a danger of clashes such as Vincent suggests, just name the 1788 ones differently! The underlying 754 system may call it dot(), the new one can be dot1788(), etc.
Ned Nedialkov's comments, based on much experience, suggest this sort of thing will be way down our list of problems when trying to achieve portability.
John Pryce