[Cfp-interest 2800] WG14 IEEE 754-C binding meeting minutes - 2023/07/05
Rajan Bhakta
rbhakta at us.ibm.com
Wed Jul 5 09:57:34 PDT 2023
Attendees: Rajan, Jim, Fred, Damian, Henry Kleynhans (Bloomberg), Mike, David, Ian
New agenda items (https://wiki.edg.com/pub/CFP/WebHome/CFP_meeting_agenda-20230705-update.pdf):
None.
Next Meeting(s):
August 2, 2023, 4PM UTC
ISO Zoom teleconference
Please notify the group if this time slot does not work.
New action items:
Fred: Send any issues found with incorrect CD2 comment applications (Ex. DECIMAL_DIG has two entries in the suffix) to JeanHeyd and cc WG14 before the July 10th.
Rajan: Submit CFP2741 as part of the US national body as an editorial comment for the DIS ballot.
Jim: Let the editor know of the page 675 "floating point" -> "floating-point" change
Jim: Update TS part 5 as per CFP2790's final suggestion.
Jim: Follow up with Hubert to see when he wants input on CFP2799.
C++ liaison:
None.
C23:
C2X working draft N3149 (for CFP only)
https://wiki.edg.com/pub/CFP/WebHome/N3149.pdf
Review of post-CD2 drafts
[Cfp-interest 2791] new C2X working draft, review
Jim: I had reviewed the changes and the corrections made by the editor after the initial set of comments. Converged very well.
Fred: I found a misapplied CD2 change.
Jim: Hold onto it. We don't want to miss it.
Jim: Can do DIS comments, C2Y list, or no action for any future issues.
Fred: They are editorial. Ex. DECIMAL_DIG has two entries in the suffix.
Jim: Send a message to JeanHeyd and cc WG14 with anything found before the 10th.
^Fred: Send any issues found with incorrect CD2 comment applications (Ex. DECIMAL_DIG has two entries in the suffix) to JeanHeyd and cc WG14 before the July 10th.
Carry-over action items results:
Damian: Rework the carg description to say phase instead of phase angle using the spreadsheet form.
See [Cfp-interest 2734] and follow ups.
Damian: I don't know of any Australian NB for WG14. Hence can't submit the comment.
Damian: Calling it phase vs phase angle matches closer to NIST.
Rajan: Expressed in radians is also there.
David: May not be editorial due to radians.
^Rajan: Submit CFP2741 as part of the US national body as an editorial comment for the DIS ballot.
David H: Get an example for the scaled reduction functions (perhaps by asking Jason or Jim Demmel or looking into the IEEE references) by July. - Not done.
See https://754r.ucbtest.org/background/traps-and-wraps.txt
David H: Get an example for the augmented arithmetic functions (perhaps by asking Jason or Jim or looking into the IEEE references) by July. - Not done.
All: TS part 5: Consider what to do for issue 3 (which headers contain which pragmas) as per the spreadsheet (current vs alternative 1).
[Cfp-interest 2755] pragma spreadsheet Jim Thomas
[Cfp-interest 2790] action item about TS-5, pragma prefixes Jim Thomas
Action items results (from previous meeting):
Jim: Add in an agenda item for Mike's Arith 2023 report after their September meeting.
Mike: Cannot go anymore.
Rajan: Drop this item.
Jim: Update CFP recommendations for CD2 ballot comments with:
See [Cfp-interest 2780] CFP response to NB comments for CD2
GB-005: Mike: Perhaps say "make" instead of "return" (alternative is "provide").
WG14 made it "produce".
CA-022: Make a note of the difference - Need a hyphen.
US-087: Clarify that there should not be any replacement (original text is fine), just adding the new recommended practice words.
US-117: CFP recommendation: “If a floating result overflows and default rounding is in effect and the integer expression math_errhandling & MATH_ERRNO is nonzero, then the integer expression errno acquires the value ERANGE."
GB-188: Change "ininfinite" to "isinfinte" in the proposed change.
US-201: Annex G does *NOT* specify the behavior of SNANs (typo, need to fix)
All accepted.
Fred: Validate the list of functions for GB-181's proposed response.
[Cfp-interest 2776] GB 181: Missing from B.30 Fred J. Tydeman
Fred: wcstol->wcstold.
Accepted by WG14 with the typo correction.
Rajan: For US-117, at the WG14 meeting say an alternative is having comma's after the "and"s.
Fred: Add CFP2745 to the list of issues for C2Y.
- Still open.
Jim: We will talk about this later. May result in dropping or a new action item.
Other issues:
No common real type
[Cfp-interest 2745] Re: double-double OP float128_t
Rajan: One implementation defines extended rules for a fixed point OP floating point and has this as an extension to the standard..
Jim: If long double is float-128, while double is double-double in some cases, in this situation, an implementation can do similar to what Rajan said and give rules to handle it however they want.
Fred: I can capture this.
Which IEC 60559 formats
[Cfp-interest 2746] {FLT, DBL, LDBL}_IS_IEC_60559 macros: clarification needed
Rajan: I think Fred asked for these.
Fred: This is supposed to exclude abrupt underflow.
Jim: The definition got changed to be just about the format, not operations.
Consider the issue for C2Y.
Minimum positive subnormals
[Cfp-interest 2747] about FLT_TRUE_MIN and DBL_TRUE_MIN in 5.2.4.2.2p29 example
Jim: Assuming default rounding and no wide evaluation, converting the decimal values to binary would work. Presumably they could be 1 and 5.
Fred: But implementations are allowed to use wide precision.
Jim: But then it wouldn't be right anyways.
Jim: Anyone want to consider shortening these for C2Y?
No consensus. Keep as is.
FP_CONTRACT pragma for freestanding
[Cfp-interest 2793] FP_CONTRACT pragma and <math.h>
Jim: I said for Annex F conforming implementations, FP_CONTRACT is available.
Fred: We should just disassociate pragma's from headers.
Jim: We could consider this for C2Y. This means everyone has to support the pragmas.
Floating vs floating-point
[Cfp-interest 2794] "floating" vs "floating-point", and hyphens
Mike: We brought this up 3 years ago but decided not to press it then.
Mike: I checked most of them and it looks good.
Jim: Good candidate for C2Y.
Mike: The other comment about floating-point in CD2. This could be an extension of that.
Jim: The last one can be done since it is another instance of what was proposed. The rest go beyond it.
Mike: For results is one set. The other set is leaving out the hyphen.
Jim: Yes, that last set was the what the comment was about. The rest is wider. Floating types are a C defined term. The comment is about how floating is used in other cases. It is not comprehensive and we need to clarify what they should be applied to.
Mike: I agree. We can go through the list and do the "floating point" -> "floating-point" ones only and leave the rest.
Jim: Mike, can you go through those and get that list?
Mike: Hard to tell with page numbers. Can Vincent do this? Get the list of things that don't change the meaning of "floating".
^Jim: Let the editor know of the page 675 "floating point" -> "floating-point" change
Jim: Look through the list of items in CFP-2794 items and let the editor know of any "floating point" -> "floating-point" changes needed.
Seems like the page 675 one is the only one.
Non-IEC 60559 long double with Annex F
[Cfp-interest 2795] operations on long double when not an IEC 60559 format
Fred: Perhaps add one sentence to say it is implementation defined.
Jim: It takes more than that since need to know which values are covered.
Damian: Saying it is implementation defined is too broad a brush.
C2Y issue.
TS part 4 revision:
See CFP2710.
Jim: Added slots for the examples.
TS part 5 revision:
See CFP2753.
Jim: Big issue was the evaluation macros.
No issues with the changes.
Issue 1: We wanted to bring this up for CD2 but dropped it somehow. Should this be DIS or C2Y?
David: I don't see how this justifies another whole draft.
C2Y issue.
Issue 3: Pragmas and headers and prefixes
See CFP2755, CFP2790
Fred: float.h seems better for stand-alone pragmas controlling floating-point expressions.
Jim: Issue with requirements for freestanding. We rejected it last time.
David: I don't see any issue with the changes proposed.
^Jim: Update TS part 5 as per CFP2790's final suggestion.
Atomic fetch add
See CFP2799 Hubert Tong's email.
Damian: Requires two operands since min/max.
David: If it is atomic, you can't get an and b atomically and has to be handled at a higher level.
Jim: I'm assuming they're integer equivalents of this?
Rajan: No, integers only have add/sub/or/xor/and. Not a min/max thing.
^Jim: Follow up with Hubert to see when he wants input on CFP2799.
David: If atomic floating points were an issue for applications, this would have come up a long time ago. For integers it makes sense. Seems not to be a problem that needs to be solved.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20230705/ef4e0f11/attachment-0001.htm>
More information about the Cfp-interest
mailing list