[Cfp-interest 2134] Re: AI about normal and subnormal classification

[Jim Thomas]

There’s still the issue discussed in the CFP thread “Subnormals in decimal” about defining “normal" and “subnormal" in terms on “normalized”. Here is a alternate version that does not depend on “normalized”:

[0] Floating-point values can be classified as NaN, infinite, normal, subnormal, or zero, or into other implementation-defined categories.
Numbers whose magnitude is at least b^(emin-1) and at most (1 - b^(-p))*b^emax, where b, p, emin, and emax are as defined in 5.2.4.2.2, are classified as normal. Nonzero numbers whose magnitude is less than b^(emin-1) are classified as subnormal. 

But I doubt this would be as readily meaningful to C standard readers. The significance of the bounds for normal numbers is to delineate the range of normalizable numbers. The IEC 60559 definition of “normal” includes: "Normal numbers can use the full precision available in a format."

Note that the normal number bounds here are the defining expressions for macros, e.g. FLT_MIN and FLT_NORM_MAX, that give the minimum positive normalized floating-point number and the maximum normalized floating-point number. 

- Jim Thomas

> On Sep 4, 2021, at 11:28 AM, Jim Thomas <jaswthomas at sbcglobal.net> wrote:
> 
> Action item:
>>    All: Look over CFP2096 and give feedback within 2 weeks.
> 
> 
> After more consideration, I think it would be confusing and risky to introduce into 5.2.4.2.2 definitions for normal and subnormal numbers that are different from the 5.2.4.2.2 definitions of normalized and subnormal floating-point numbers. The problem pointed out by Vincent is that the “normal" classification category in 7.12 is not defined. We can address this issue in 7.12 without having to change 5.2.4.2.2 (which is referenced explicitly and implicitly through the standard). This is the approach taken in the draft below.
> 
> - Jim Thomas
> 
> https://wiki.edg.com/pub/CFP/WebHome/C23_proposal_-_Normal_and_subnormal_classification-20210904.pdf <https://wiki.edg.com/pub/CFP/WebHome/C23_proposal_-_Normal_and_subnormal_classification-20210904.pdf>
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20210908/517c2e10/attachment-0001.htm>