[Cfp-interest 1283] Re: Links to WG14 papers about floating-point

[Jim Thomas]

> On Feb 2, 2019, at 5:58 PM, Fred J. Tydeman <tydeman at tybor.com> wrote:
> 
> On Thu, 24 Jan 2019 12:43:26 -0600 Rajan Bhakta wrote:
>> 
>>   Fred: Send links to papers of interest to CFP from N2319-N2323. 
> 
> My recent WG14 (C standard) papers related to floating-point:
> 
> N2319<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2319.htm> 2018/12/10 Tydeman, FE_TONEARESTFROMZERO w.r.t. FLT_ROUNDS

Removes footnote and says its to be part of paragraph 9, but I don’t see anything else about a paragraph 9.

> N2320<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2320.htm> 2018/12/10 Tydeman, Precision and NAN/NAN(...) and INF/INFINITY

The first scheme seems arbitrary in some respects and might break all existing implementations. The impact could be lessened by allowing either of the infinity or NaN forms when p >= 9. Why require the nan(n-char-sequence) form when the n-char-sequence is not required to have any meaning? 

> N2321<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2321.htm> 2018/12/10 Tydeman, Nextafter/nexttoward/nextup/nextdown

A patch for double-double. Why is this definition preferable? With this definition of nextup etc., how does one determine the actual next-up value? This is important for testing monotonicity, for example.

> N2323<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2323.htm> 2018/12/10 Tydeman, SD3 1: DR 440: Test macros for FP being 754 types

Agree with the need at least for a macro to say whether long double has an IEC 60559 format for annex F implementations.  

What does “… operations are those of an IEC 60559 basic, interchange or extended format” mean? Of what use is it without a binding? Interchange formats include basic formats. What if format is IEC 60559 but operations are not? Why mention operations?

IEC 60559 itself has formats, not types. 

Does saying the type's values are those of an IEC 60559 type/format mean that subnormals and signed zeros and infinities are values of the type?

How about just saying

0	type does not have an IEC 60559 format
1	type has an IEC 60559 format

The number for bits in an IEC 60559 extended format is not defined.

> N2325<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2325.htm> 2018/12/10 Tydeman, SD3 11: Maximum normalized FP number

More support for double-double.

> N2326<http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2326.htm> 2018/12/10 Tydeman, Merge DR 432+467

The footnote says “Some implementation may have types …”. It should be “Some implementations have types …” or “Implementations may have types …”

Instead, consider attaching the footnote to "floating types may be able to contain other kinds of floating-point numbers” and change the footnote to “Some implementations have types that include finite numbers with extra range and/or precision that are not covered by the model.”

- Jim Thomas



> 
> 
> ---
> Fred J. Tydeman        Tydeman Consulting
> tydeman at tybor.com      Testing, numerics, programming
> +1 (702) 608-6093      Vice-chair of PL22.11 (ANSI "C")
> Sample C99+FPCE tests: http://www.tybor.com
> Savers sleep well, investors eat well, spenders work forever.
> 
> _______________________________________________
> Cfp-interest mailing list
> Cfp-interest at oakapple.net
> http://mailman.oakapple.net/mailman/listinfo/cfp-interest