[Cfp-interest 3039] WG14 IEEE 754-C binding meeting minutes - 2024/03/13
Rajan Bhakta
rbhakta at us.ibm.com
Wed Mar 13 10:09:59 PDT 2024
Attendees: Rajan, Jim, Fred, Damian, Jerome, Joshua, David
New agenda items (https://wiki.edg.com/pub/CFP/WebHome/CFP%20meeting%20agenda-20240313-update.pdf):
None.
Previous meeting notes:
See CFP2976 (http://mailman.oakapple.net/pipermail/cfp-interest/2024-January/002990.html).
Next Meeting(s):
April 10, 2024, 4PM UTC
ISO Zoom teleconference
Please notify the group if this time slot does not work.
New action items:
Rajan: Send the WG14 editorial comments from CFP to CFP.
Rajan: For C2Y issue 5, reword H.3.6 and 5.2.5.3.2#28 to "If a signaling NaN macro (optionally preceded by the unary + or - operator) is used for initializing an object of the same type that has static or thread storage duration, the object is initialized with a signaling NaN value."
Jim: Fix the suggested changes section in CFP3020's paper to point to N3219 (instead of the incorrect N3619 as it currently is) and send it out to WG14.
Jim: Submit the paper resolving C2Y Issue 17 (CFP3022) to WG14.
Jim/Jerome/Damian: Follow up on C26 issue 1.
Fred: Add CFP3003 to the issues list.
Jim: Draft up changes to incorporate CFP3006.
Fred: Add CFP3007 to the C26 issues list.
Damian: Get a list of editorial issues in Annex G and send them out for future submission to WG14.
Action items to be carried over:
None.
C++ liaison:
None.
WG14 (added):
See CFP3008 and follow ons.
No editorial review group resolution meeting scheduled yet.
Next WG14 meeting (virtual) is June 10-14th, 2024.
^Rajan: Send the WG14 editorial comments from CFP to CFP.
C23 integration
C23 drafts:
C23 working draft n3219 - July 2, 2023 - For CFP review only. Do not distribute.
Carry over action items
None
Action items from previous meeting (Done unless stated otherwise)
Fred: C26: Issue 5: Are there any <math.h> macros with the same issue? Should words be added to an introduction section in <float.h>?
See [Cfp-interest 2999] <math.h> macros and exceptions (and follow on CFP 3031)
OK with using CFP 3031 as the direction for the resolution to the issue.
Jim: For SNAN macro recommended practice (in F.2.2#6 on), don't see the issue.
Fred: Will look at it.
Rajan: The H.3#6 should have the optional unary operator be after the initializing an object to not make that optional mandatory (in a specific way of reading it).
Jim: I'm OK with that.
Jerome: Doesn't adding the -/+ cause raising a signal?
Jim: No, this is not an expression. 754 has a set of operations that do not signal.
Damian: Yes, that's true. Things like copysign.
^Rajan: For C2Y issue 5, reword H.3.6 and 5.2.5.3.2#28 to "If a signaling NaN macro (optionally preceded by the unary + or - operator) is used for initializing an object of the same type that has static or thread storage duration, the object is initialized with a signaling NaN value."
Fred: C26: Issue 9: Look at original CFP messages to see if 3.10 (Correctly rounded definition) might cover it.
See [Cfp-interest 3000] CFP issue #9
Fred: Fine with it.
Rajan: It is in 3.12 in N3219.
Jim: C26: Issue 4: Draft a paper as per the resolution in the issues list.
See [Cfp-interest 3020 and follow ons] Re: printf and rounding recommendation
^Jim: Fix the suggested changes section in CFP3020's paper to point to N3219 (instead of the incorrect N3619 as it currently is) and send it out to WG14.
Jim: C26: Issue 17: Draft a paper as per the resolution in the issues list.
See [Cfp-interest 3022] C2Y Issue 17
^Jim: Submit the paper resolving C2Y Issue 17 (CFP3022) to WG14.
Jerome: C26: Issue 1: Get definitions of terms relating to the issue for 754 and C and regular math.
See [Cfp-interest 3016] Re: about C26 Issue 1
Jerome: Will need to remain an open issue until we get closure from the 754 people. We could make the changes I suggested in C, but better to wait for 754.
Jim: The 754 term we need to be consistent with is the divide-by-zero exception. It doesn't classify the different types of errors. C needs the categories due to errno.
Damian: I think singularity is exactly what you want.
Fred: So log(0) is a singularity error?
Jerome: Yes, on 754 systems.
Fred: If you make this change for log gamma, you need to do it for all other cases like log(0).
Jerome: Yes, I think you are right.
Jim: Wider issue, see CFP-2996.
Fred: Don't most mathematicians consider pow(0,0) to be 1?
Damian: Yes.
Jim: No. Because the limit does not exist.
David: It's an exception since there is no right answer. The default is arbitrary. You should get an exception for pow(0,0). I had forgotten that.
Joshua: All the pown, powr, and pow say they return 1 without exceptions in IEEE.
Jerome: I think some people wanted NaN since it is dangerous to say anything since 0, 1, or infinity are all valid.
Jim: Kahan said you had to have 1 as anything else would cause problems.
Jim: Now I'm thinking of leaving "mathematical function" as not fully defined.
Jerome: My sense of IEEE is that they don't like the calculus idea of infinity, but more the computation based numerics. If there is no language in the C standard to talk about the mathematical domain that lurks behind everything, then it would be hard to introduce it without a lot of work. And instead talk about the computer arithmetic domain.
Jim: The mathematics is definitely behind it. Like the discussion of poles, and log referring to the mathematical log functions.
^Jim/Jerome/Damian: Follow up on C26 issue 1.
TS-4 and TS-5 revisions
See [CFP 3015]
Jim: For TS-4, the use of "consider" is not accepted by ISO. The example for scaled_proddiff, we can say "The following computes a fragment of the Clebsch-Gordan calculation as a simplified example." instead.
Jim: For augmented arithmetic, an example has moved to 7.2. This makes it a more ISO conforming way of referring to the example.
C26 issues
Issues list
See https://wiki.edg.com/pub/CFP/WebHome/C26C.HTM
See [CFP 2992, 2994, 3003, 3005 and follow ups]
Jim: CFP3003 should be added to the issues list.
^Fred: Add CFP3003 to the issues list.
Jim: For CFP3006, Vincent has better words for inputs.
Rajan: This would change the "correctly rounded" with respect to inputs.
Jim: Yes, correct. We'd have to do that too.
^Jim: Draft up changes to incorporate CFP3006.
Jim: Similar for CFP3007. This issue there is he is saying infinity is a floating-point number, but it is not, at least how we have defined it.
^Fred: Add CFP3007 to the C26 issues list.
Issue 1: In progress (Jerome's item).
Issue 3: Jim: I have this as no need for change and closed.
Issue 4: Jim: Have a draft proposal for this. Action item.
Issue 9: Jim: We decided we could close this.
Issue 17: Jim: Have a draft proposal for this. Action item.
Jim: Propose we look at 11 and 14 next.
Imaginary types
See [N3206, CFP 2979, CFP 2997 and follow ups]
Annex G complex functions
See [CFP 3018, 3019, 3032, and follow ups]
Damian: Noticed some inconsistencies in Annex G.
Damian: CFP3037 has a summary of inconsistencies that are mostly editorial.
^Damian: Get a list of editorial issues in Annex G and send them out for future submission to WG14.
Others?
Other issues
IEEE 754 meetings
Damian, David, Jerome, Mike are attending.
Fred: IEEE explicitly asked for James Thomas for language input.
Accuracy of mathematical functions
See [CFP 3002]
Fred: Not very good at setting the flags.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20240313/92537089/attachment-0001.htm>
More information about the Cfp-interest
mailing list