IEEE 754 Revisions Committee, draft review, Mickey's Toon Town Room, Building 14, Sun, Santa Clara, Thursday March 16, 2006. Joe Darcy was our host. Roger Golliver, David Hough, Peter Markstein, Mike Cowlishaw, Craig Nelson, Jim Thomas, Dick Delp, Peter Tang, & Dan Zuras attended. Mark Erle, Jeff Kidder, & Eric Schwarz were on the phone. Dave wanted to change the date of the draft review relative to the main meeting. We began to construct an issues list with Mike's summary & Dave's draft. Tell people about last call for issues. Add email addresses to notes. (Mike) Mike Cowlishaw (Roger) "Golliver, Roger A" (Dave) David Hough 754R work <754r@ucbtest.org> (Peter M) Peter Markstein (Eric) Eric Schwarz (Peter T) "Tang, Peter" (Jim) Jim Thomas The Issues List --------------- 0.1 Abstract, to be added. Mike will look into what sections we need at the beginning of the standard as far as the IEEE is concerned. (Mike) 0.2 Keywords, to be added. (Mike) 0.4 Introduction, to be added. (Mike) 0.5 Forward, to be discussed. (Mike) 1.1, 1.2, 1.3 Scope, to be discussed. (Mike) 1.4 Language defined, to be discussed. (?) What does it mean for a language standard to conform. (Jim, Roger) 2.0 Definitions, there must be issues here. The definition of floating-point number. (Jim) Trailing significand needs to be defined. (Peter M) definition of status flag - is it required that the user be able to raise overflow or underflow without inexact? (Jim) 3.0 Background & Terminology Should +/-0 be at level 2. (Jim) 3.3 Decimal format encoding. Figure 1.1-10 drawing of decimal encoding. (Peter T, Mike) captions need fixing 4.0 Rounding & Modes - scope of chapter 4 (Peter M) 4.1 Rounding direction default for decimal should be language defined. (Dave) 4.4 Constant mode declaration. Needs discussion for implications. (Jim) Global modes is controversial - inheritance in the absense of any explicit statement. (Roger) 5.0 Summary of operations - issue with the number of operations especially the explicit width operations. (Jim, Roger) issue of cross radix operations (to be excluded). (Peter T) Intformat too broad? (Dave) int, unsigned, chars, et al., issue is conversions to those formats & the number of versions. Do we really have to have format-of tail operations? (Dave) the issue is are these of explicit width or generic and do they belong here or with scalb? Decimal tail operations (Peter T) Implicit width operations - are they feasible for c++? (Jim) talk to c++ guys & get back to us. Set-rounding-direction is operation but should be mode declaration (maybe)? (Dave) 5.2 Square root - old text about wider destinations (Roger) issue is inconsisent text, there may be other examples. 5.3 Floating-point format conversions - mixed radix conversions. (Peter T) rounding modes not detailed enough for decimal. 5.4.2 Conversion from floating-point to integer - didn't provide for round away. Do we need to? clarification (Peter M) Explicit versus implicit conversions. (Jim) 5.6.3 references to varying width format (global issue). (Dave) Issue with requiring extra digits to be treated as zero on external string conversions. (Jim) the issue is whether or not to completely specify the behavior when the number of digits exceeds some limit. (substantive) 5.10 there have been complaints about min/max functions. Signed zeros & totalorder as a basis is part of the issue. (Only an issue for default min/max) (Roger) Also the binding of min & max. (clarification) (Jim) 5.SP Scaled Product - not clear that you are allowed to do an inexact implementation here. (Eric) Error bound for these functions. (Eric) Issue with definition of 'n'. (Eric) License to reassociate. (Eric) Jim claims that text elsewhere is falsified by these operations, therefore need a new catagory operation. (Dave) (possibly new table) 7.0 Default Exception Handling - What is the scope of flags? (Dave) must know how it is to be used. How do you declare the scope where the flags are visible? (Jim) Need a constant mode declaration to insure flags are imported & exported as if they were global dynamic modes. (Dave) global flags: whether the statement represents a change in the standard. Jim raised some issue about consistency of global flag state that I didn't really understand. Also, as with modes, something about dynamic versus static & its unclear how it is to be used. R Required Functions & Predicates - Item 14 - do we need equivalent() or not? (Dave) logb - whether this is the exponent of the normalized input or not? (It is not now.) Should it be? (Jim) E Expression Evaluation - Ivan has issues with this entire concept. Jim says we need to say more about optimization. (Jim) Reproducible results is an issue. References to fixed width formats (all of our formats are fixed width now). (Dave) Widento modes - issue with declaration that languages prohibit or discourge mixed expressions with integer. (Jim) Also, requiring widento modes for every supported format. (Jim) T. Elementary Transcendental Functions - Eric is concerned about decimal quad for sin or cos for really big arguments & wants it to be invalid. (Eric) Domain restrictions on transcendental functions is an issue. (Eric) This section is binary only for the moment but there is an issue as to whether it should be generalized to decimal in the near future or not. (Peter M) N Non-Upward Performance Compatible Extensions - Should we have this at all & should there be shalls in it. (Eric) (Annex 8 Alternate Exception Handling) Annex T Traps - should it be just deleted. (Dave) Annex D Debugging - others think this should be deleted. Annex L Language Issues - just a bunch of position statements, either write something useful or delete it. (Dave) removing purple text (Dave, later) On to draft review. 1.3 exclusions - bullet for 'rules for optimizing expression evaluation' to be deleted. 1.4 language defined/implementation defined - split long sentence into two. Aspire to... goes to aspire towards. page 21, languages can not conform... Section 2 need to be alphabetized page 24, definition of basic formats - fixed goes to specified. page 25, definition of floating-point number page 26, alternate exception handling modes & widening modes. page 27, definition of signal - note that... page 27, definition of status flag - page 28, 3. Formats page 28, non-interchange formats page 28, storage formats - deleted destinations page 29, basic formats page 45, page 48, Peter wants to make minor changes here (negotiate with Mike). page 108 - page 111, page 110, inherited scope... page 111, defined by what is not defined? page 144, all supported binary... called it a day at 5:30.