[Cfp-interest 2152] Re: defect in ISO C about underflow with exact subnormal

[Jim Thomas]

The relevant text noted in [Cfp-interest 2106] is:

> F.9.2 "Expression transformations":
> 
>  1 × x and x/1 → x     The expressions 1 × x, x/1, and x may be
>                        regarded as equivalent (on IEC 60559 machines,
>                        among others).399)
> 
>  399) Strict support for signaling NaNs — not required by this
>  specification — would invalidate these and other transformations
>  that remove arithmetic operators.
> 


F.8 #2 says “… The specification in this annex assumes IEC 60559 default exception handling ….” Given that, I think the statement is correct.

Since we do mention alternate exception handling in other contexts, we might consider expanding the footnote:

399) Strict support for signaling NaNs and alternate exception handling — neither required by this specification — would invalidate these and other transformations that remove arithmetic operators.

- Jim Thomas

> On Sep 23, 2021, at 6:35 AM, Vincent Lefevre <vincent at vinc17.net> wrote:
> 
> On 2021-09-12 18:39:08 -0700, Jim Thomas wrote:
>> C Annex F requires default exception handling and doesn’t provide a
>> way to change to alternate exception handling. In this context the
>> statement is correct.
> 
> The C standard does not provide a way to change to alternate
> exception handling, but it allows an implementation to do so.
> This is the problem here.
> 
> There are notes in the standard about that, e.g. 220 in C17 "[...]
> enabled traps for floating-point exceptions [...]" and 222 "IEC 60559
> systems have a default non-stop mode, and typically at least one other
> mode for trap handling or aborting".
> 
> There's also F.8 "Floating-point environment", which says "It includes
> also IEC 60559 dynamic rounding precision and trap enablement modes,
> if the implementation supports them."
> 
> F.8.3 makes this clear:
> 
>  At program startup the floating-point environment is initialized as
>  prescribed by IEC 60559:
> [...]
>  - Trapping or stopping (if supported) is disabled on all
>    floating-point exceptions.
> 
> If the intent were to disallow trapping, the standard would not say
> "At program startup", but for the whole program execution.
> 
> -- 
> 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)
> _______________________________________________
> Cfp-interest mailing list
> Cfp-interest at oakapple.net
> http://mailman.oakapple.net/mailman/listinfo/cfp-interest

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20210926/3fdbe756/attachment.htm>