[Cfp-interest 1855] Re: Negative zero

[Jim Thomas]

> On Nov 21, 2020, at 4:30 PM, Damian McGuckin <damianm at esi.com.au> wrote:
> 
> On Sat, 21 Nov 2020, Jim Thomas wrote:
> 
>> In a recent post to STDS-754, Vincent Lef?vre wrote:
>> 
>>      I think that the terminology should be:
>>       * negative: equivalent to "less than zero", i.e. "< 0";
>>       * negative sign: can be used for any FP number; -0 has a
>>         negative sign, but NaN does not have a sign;
>>       * negative sign bit: can be used for any FP datum.
>> Do we agree with this interpretation of ?negative?? 
>> (I do, except for the last item. Do qNaN and sNaN data in an extended format
>> necessarily have a sign bit?
> 
> None of NaN, qNaN and sNan have a sign. However, they do have a sign bit. Why should that be any different for extended format (even though IEC60559
> does not specify what it is)?

754 formats have to include at least one qNaN representation and at least one sNaN representation. For interchange formats, encoding are given and these include sign bits for NaNs. For extended formats, encodings are not given; an implementation could have sNaN and qNaN with signless encodings. 

- Jim Thomas

> 
>> A sign bit pertains to encoding and 754 doesn?t specify an encoding for extended formats. I think this is a 754 issue.
> 
> Yes. So, C23 should not try and resolve it. But does that mean we need to explicitly say that it is undefined. Out of curiosity, how many chipsets in use today still use an extended IEC60559 format?
> 
>> Does C require opposite-signed zeros to compare equal? Does Annex F say this explicitly? Annex F says it implicitly by reference to IEC 60559, so it should be editorial to say it explicitly in Annex F.
> 
> Sounds wise although I never know when to editorialize and when to not.
> 
>> We might want to add a (foot)note to point out that negative zero and NaNs
>> with a negative sign bit are not negative values.
> 
> Yes. Why not within the text so it is glaringly obvious?
> 
>> Have we been careful enough about distinguishing the sign of a value and the bit sign of the representation? For example, 7.12.3.7 says
>> The signbit macro returns a nonzero value if and only if the sign of its
>> argument value is negative.
> 
> As MikeC correctly notes, the signbit() macro is testing the signbit of the encoding, not its negativity.
> 
> But JimT's original question was whether we had been careful enough to distinguish between the sign of a value and the sign bit within Annex F. Asking ourselves is probably counterproductive as we are all too close to it. We should each probably ask a few of our colleagues to get a better idea that this had been achieved.
> 
> Regards - Damian
> 
> Pacific Engineering Systems International, 277-279 Broadway, Glebe NSW 2037
> Ph:+61-2-8571-0847 .. Fx:+61-2-9692-9623 | unsolicited email not wanted here
> Views & opinions here are mine and not those of any past or present employer