Submitter:Fred Tydeman (tydeman@tybor.com)
Author:CFP group
Submission Date: 2022-05-11
Document: WG14 N2981
Title: N2981: Make *_HAS_SUBNORM be obsolescent
Reference Documents: C N2797 (http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2797.htm); C++ (https://wg21.link/p2569r0)

Summary: Make the *_HAS_SUBNORM macros in <float.h> be an obsolescent feature.

Discussion

When the red part of paper N2797 was discussed at the November 2021 meeting (section 5.43 of minutes N2921), the committee did not take a vote. Instead, they asked the author to liaise with C++ (this paper is the result of that liaise). In addition, the sense of the committee was removing without first making them obsolescent/deprecated was a bad thing to do. There was some voice to add a runtime query function. There was some voice to further specify these macros.

When the green part of paper N2797 was discussed at the February 2022 meeting (section 5.8.1 in minutes N2941), the committee voted to accept.

The feedback from the SG6 (the C++ numerics sub-group) via the liaison group is:

• The C++ standard defers the *_HAS_SUBNORM macros entirely to C (there are no direct references to them in C++); the only (indirect) mention is in [cfloat.syn]:

  "The header <cfloat> defines all macros the same as the C standard library header <float.h>."

• C++ does not care if C were to remove the *_HAS_SUBNORM macros.
• C++ has their own means (traits) to define support for subnormals.

There are several ways subnormals are "supported" in the field:

• Partial support - hardware has encodings, but operations "fail".
  • Operands are flushed to zero; results are kept.
  • Operands are kept; results are flushed to zero.
  • Some operations flush, others do not flush.
  • Results suffer double rounding.
  • Support can be changed at runtime (not by means in Stardard C).
• Not at all. There are no hardware encodings of subnormals.
• Full support as per IEEE-754.

Proposal

Add to the end of 5.2.4.2.2, paragraph 14 as indicated:

The presence or absence of subnormal numbers is characterized by the implementation-defined values of FLT_HAS_SUBNORM, DBL_HAS_SUBNORM, and LDBL_HAS_SUBNORM:
-1 indeterminable 27)
0 absent (type does not support subnormal numbers)28)
1 present (type does support subnormal numbers)
The use of FLT_HAS_SUBNORM, DBL_HAS_SUBNORM, and LDBL_HAS_SUBNORM macros is an obsolescent feature.

Alter two footnotes attached to that paragraph as indicated:

27) Characterization as indeterminable is intended if floating-point operations do not consistently interpret subnormal representations (operands and/or results) as zero, nor as nonzero.
28) Characterization as absent is intended if noall floating-point operations produce subnormal results from non-subnormal inputs, even if the type format includes representations of subnormal numberstreat subnormal operands and results as zero.

Add to 7.32.5 Characteristics of floating types <float.h> a new paragraph: The use of FLT_HAS_SUBNORM, DBL_HAS_SUBNORM, and LDBL_HAS_SUBNORM macros is an obsolescent feature.

Straw poll

• Should the proposal be added to C23?