[Cfp-interest] AI about specification for inexact
Jim Thomas
jaswthomas at sbcglobal.net
Tue Jun 26 14:32:22 PDT 2018
Action item:
> Jim: Specification for inexact: Make the editorial change as given in the modified agenda with links.
>
The agreed to add the following in F.3:
[4b] The reduction functions (7.12.13b) raise the “inexact” floating-point exception in conjunction with overflow and underflow; otherwise, whether they raise the “inexact” floating-point exception is unspecified. The other functions listed in the table in F.3 of IEC 60559-recommended operations should raise the” inexact” floating-point exception if the result is inexact, and should not raise the” inexact” floating-point exception if the result is exact, unless explicitly stated otherwise.
We (at least I) hadn’t considered that F.3 is just about the binding to IEC 60559 and this isn't binding information. Here’s an alternative suggestion that is a more complete treatment of inexact, to be placed along with the specification for the other exceptions in F.10.
In TS 18661-4, clause 7, before “After F.10.1.7, insert the following”, insert:
After F.10#7, insert the paragraph:
[7a] The functions listed in the table in F.3 of IEC 60559-required operations raise the “inexact” floating-point exception whenever a returned numerical result does not equal the mathematical result. The IEC 60559-recommended reduction functions (7.12.13b) raise the “inexact” floating-point exception in conjunction with overflow and underflow; otherwise, whether they raise the “inexact” floating-point exception is unspecified. The other functions listed in the table in F.3 of IEC 60559-recommended operations should raise the” inexact” floating-point exception if the result is inexact, and should not raise the” inexact” floating-point exception if the result is exact, unless explicitly specified otherwise.
What do you think?
- Jim Thomas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20180626/f9cd7e09/attachment-0001.html
More information about the Cfp-interest
mailing list