[Cfp-interest] WG14 IEEE 754-C binding meeting minutes 2017/07/11

[Rajan Bhakta]

  Attendees: Rajan, David H. Fred, Jim, David C., Ian

  New agenda items:
    None

  Last meeting action items:
    Jim: Reword 
http://wiki.edg.com/pub/CFP/WebHome/augop_spec-20170422.pdf to have the 
additions done in Annex F only. - Done
    Jim: Work http://wiki.edg.com/pub/CFP/WebHome/min-max_names.pdf into a 
specification for the main body of the standard. Also add a footnote 
saying for implementations without NaN's the functions have the same 
results. - Done
    All: Look into any other teleconferencing facility (Ex. Cisco Webex, 
Skype, etc.). - Not done

  New action items:
    Jim: Send the note (2017/06/28) drafted by Jim to WG14 for DR9 %a 
precision concerns.
    Jim: Re DR501: Make a new proposal for Part 3 for new macros 
(format-specific DECIMAL_DIG like macros)
    Jim: Check implications for tgmath with regards to the augmented 
precision functions.
    Jim: Augmented precision: Add in text to state that the functions 
force a particular rounding (ignoring static or dynamic rounding modes and 
implementation supported rounding modes).
    Jim: min-max: Add in a statement about the preferred exponent.
    Jim: min-max: Add a reference to the corresponding Annex F section for 
NaN treatment in the fmaximum/fminimum/fmaximum_mag/fminimum_mag 
functions.
    Jim: min-max: {fmaximum/fminimum/fmaximum_mag/fminimum_mag}_num 
functions: They determine the number -> they return the number
    All: min-max: Consider what to do for fmin/fmax functions in the C 
standard.
    Fred: Summarize what goes wrong for FLT_EVAL_METHOD for the things 
Fred has tested.
 
  Next Meetings:
    Tuesday August 29th, 2017, 12:00 EDT, 9:00 PST
    Same teleconference number.

  Discussion:
    IEEE 754 revision:
      Another month or two until ballot likely.
      Expecting 2018 publish

    C++ liaison:
      No update.
 
    C2x proposals:
      Part 5d (alternate exception handling):
        Leave it as is unless there is a proposal, some direction or 
implementation experience.
 
    DRs:
      DR9 (tgmath, arith conversions): Latest proposal from us is there. 
Looks good.
      DR11 (a-style formatting): Issue with the precision specification 
(binary vs decimal)
        Recommend it not be changed as per last meeting discussion.
        *Send the note (2017/06/28) drafted by Jim to WG14.
      DR501 (C, DECIMAL_DIG): 
        Let DECIMAL_DIG be obsolesced.
        *Make a new proposal for Part 3 for new macros (format-specific 
DECIMAL_DIG like macros)
 
    Binding for IEEE 754-2018:
      Augmented precision (
http://wiki.edg.com/pub/CFP/WebHome/augop_spec-20170619.pdf): 
        h, t names OK? Yes.
        Fred: Any mention for tgmath here? 
          Jim:No, not going to try to add these to tgmath.
          David H: Usually these functions are used with the widest 
hardware precision present. Type generic wouldn't add much here.
          Fred: We should add words to say why these are not in tgmath.
          *Check implications for tgmath with regards to the augmented 
precision functions
        Note that round to nearest, tie away from zero. This means it is 
independent of the current rounding mode (static and dynamic), and the 
implementation does not necessarily have to have support for that rounding 
mode, though the functions must have the results as if it is there.
        Can add these notes to F.10.13.
      min/max operations (
http://wiki.edg.com/pub/CFP/WebHome/min-max_spec-20170614.pdf): 
        In the rationale (paragraph 1), state it would be problematic 
rather than unusable.
        Even though these are optional, they are intended to be mandatory 
the next revision.
        Fred: Does the quantum matter for the zeros?
          Jim: Missing a statement about the preferred exponent here.
          Seems to be not specified in IEEE 754 so far. To be discussed in 
the IEEE 754 meeting.
        For the max/min functions it mentions differing from the 
max/min_num functions due to NaNs, but doesn't say what it does for NaNs.
        *Add a reference to the corresponding Annex F section for this in 
the fmaximum/fminimum/fmaximum_mag/fminimum_mag functions
        *{fmaximum/fminimum/fmaximum_mag/fminimum_mag}_num functions: They 
determine the number -> they return the number
        Should fmin/fmax be obsolesced? Been out for a while. Can propose 
that as well.
          The specification (7.12.12.2/3) could also be loosened for NaNs?

    Other:
      FLT_EVAL_METHOD (Willem Wakker):
        Likely most compilers do the constant evaluation wrong and should 
have FLT_EVAL_METHOD -1, though it may be correct for operations.
        The TS will allow you to request FLT_EVAL_METHOD 0.
        *Fred: Summarize what goes wrong for FLT_EVAL_METHOD for the 
things Fred has tested.
 
      floating vs floating-point:
        Inconsistent usage throughout the C standard. Can maybe say 
floating-point is for things that follow the floating point model.
 
      Constant rounding modes and tgmath (Joseph Myers) - Email from Jim 
on 2017/07/07:
        Deals with the example for this (cube root function).
        Change proposed in the email seems good.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20170711/cd202067/attachment.html