IEEE 754 Revisions Committee ---------------------------- Wednesday March 15, 2006 from 1:00 PM to 5:00 PM Space Mountain Room, Building 14, Sun, Santa Clara phone: 866-839-8145 Int'l: 865-524-6352 access: 2107800 Host: Joe Darcy Proposed Agenda --------------- 1. Roll Call & Introductions 2. Show Patent Slides 3. Call Meeting to Order We have a quorum (30 voters, 3 non-voters) 4. I would like to have Microprocessor Standards Committee Chair Bob Davis act as meeting chair for today. 5. Approval of Agenda Markstein moved, Zuras seconded, 16+12 in favor. (http://nonabelian.com/754/agenda) 6. Approval of Minutes with clarification of merged minutes to be posted. Dave Hough moved to table approval of minutes until next month. Markstein seconded. passed w/o opposition. (http://nonabelian.com/754/minutes/754/060215) 7. Proposed Change in August Meeting Date (to August 9 & 10). no objection. Wednesday 10/18 full meeting, Fred Zemke at Oracle in Redwood City Thursday 10/19 draft review, Fred Zemke at Oracle in Redwood City 8. Request from Dr John Pryce Prof Kahan has invited the interval group to Berkeley. He suggests we table this issue until they are here. David Hough moved that we formally reject any all proposals for new features ... Zemke doesn't believe it is appropriate to have a motion that limits the behavior of the committee. Hough withdrew & moved that we resolve that it is the sense of the committee to discourage new features that we have not discussed before ... Cowlishaw moved that we table item 8, Godard seconded, 13+12 in favor. passed w no objections. (http://nonabelian.com/754/IEEErequest.pdf) (http://nonabelian.com/754/IntvlArithCsets.pdf) 9. Advisory Vote Results Announced (as of last night at 9:30) (http://754r.ucbtest.org/ballots/) 754p0 as draft: 3-17 failed. 3/14/06 10:09 Quad in draft: 15-0 passed. 3/14/06 10:09 FMA in draft: 15-0 passed. 3/14/06 10:10 Decimal in draft: 13-0 passed. 3/14/06 10:09 Conditionals in draft: 10-0 passed. 3/13/06 17:18 Expressions Control in: 5-3 passed. 3/13/06 17:20 Alternate Exceptions: 3-6 failed. 3/13/06 17:34 Non-Default Modes not : 7-2 passed. 3/13/06 17:24 Flag Status not export: 6-4 passed. 3/13/06 17:27 Kahan mentioned he had difficulty with his email & as a result could not present arguments on these items. Thomas pointed out that the first item is not advisory. Markstein mentioned a quorum affects all but the first issue. Davis pointed out that the count for a web vote is yes's versus no's. Non-votes & abstentions don't count. Cowlishaw mentioned that this is not how the vote was presented. Zuras said he was the one who said that & was wrong. Thomas wants the advisory votes qualified in the minutes to the extent that this was not the understanding of the committee at the time of the vote. 10. Starting Precedent: Text of proposed draft changes to accompany motions & amendments (not required today). Davis said proposed text should be posted well in advance of the meeting. 11. Suggestion for Collaboration 12. Motion: Shall we require a 2/3s majority of those present & voting to pass any changes or modifications to the current 754R working draft? (David Hough) (defeated at least twice in past, defeated 5-7 2/18/04) no second. David stated his motion as: motions to replace or modify the current 754r working draft require a 2/3s majority of these present and voting to pass. Dick seconded. Zuras made a point of order to the effect that Dave couldn't make this motion. Markstein moved to table this motion. Eric seconded. 14+10 in favor, 3 opposed, 1 abstention, motion passes 13. Motion: Shall we adopt the 2006-03-07 draft as the working draft? See http://754r.ucbtest.org/drafts/2006-03-07.sxw with full verbosity, http://754r.ucbtest.org/drafts/2006-03-07.pdf with suppressed verbosity. (David Hough) David stated this motion as: Replace 754r's current working draft 2005/10/20 with 2006/3/7. Godard seconded. Tang wanted clarification about the 10/20 draft as the working draft. Zuras suggested: make 754r's working draft the March 7, 2006 draft. Hough accepted the wording. 17. Amendment: If the 2006-03-07 motion has passed, Shall we, In the section on correctly rounded transcendental functions, limit the requirements for correct rounding to single precision and double precision binary functions, and only require well-rounding for the higher precisions and for the decimal radix? (Peter Markstein) (transcendental accepted in past) Markstein amended the draft in the section on correctly rounded transcendental functions to single & double precision functions & only require well-rounding for the higher precisions & for the decimal radix. Crawford seconded. Hack asked about detail to be addressed in the future. in favor: Markstein, Thomas opposed: Kahan, Godard, Postpischel, Hough, Kidder Markstein: Wants to restrict because only for single & double do we have algorithms that are known to work w/o arbitrary precision for all cases. Doesn't want a standard that requires us to do something we don't know how to do well. Kahan: is against this proposal because he wants to remove all accuracy specifications from this standard. There are better places than 754 for this. Thomas: in favor. Should be considered before accepting the draft. New procedures & new focus make us hopeful. But the issues are unchanged since November. The March draft is reasonable only if we pare it down to its essentials. Single & double is in the state of the art. Hough: says this is out of order. Adopt draft first. Godard: amendment ill-formed. Postpischel: doesn't know what algorithms exist. Kidder: against in the interest of getting the new draft first. Markstein: correctly rounded algorithms exist in some cases. Harrison: agrees with Kidder. Kahan: if we adopt a standard that includes an obligation to correctly round these functions will make life difficult to those who wish to vectorize these functions. Thomas: an alternative to this draft would be the 1985 draft & add items to it. Cowlishaw: Kahan's argument is stronger than Markstein's. Called the vote on amendment: 11 for, 8 against, passes. Original motion as amended. 16. Amendment: If the 2006-03-07 motion has passed, Shall we remove specification for alternate exception handling (Chapter 8) from the main body of the draft? (Jim Thomas) Thomas offers an amendment to remove chapter 8 (alternate exception handling). Markstein seconded. Thomas: in favor, chapter 8 describes exception handling in the case where default exception handling is not useful. The problem is it requires invention of new language constructs. Not a strong consensus in committee on this part of the draft. Therefore, concludes it isn't ready to go into the standard. This reduces it back to what was in 754 "in the 754 March draft style". An alternative is the examine other approaches that don't require new language constructs. Delp: wants a clarification from Thomas: does removing chapter 8 fall back to 754 (1985) or farther? Thomas: Going back to 754-1985 is my intention. Hough went to email archive for exact text of motion. Kahan: None of the things requested is unprecendented. The time to write portable exception handling far exceeds the time to write algorithms even when they are possible at all. We have to do something as what we have now is extremely unsatisfactory. Kidder: this section has only one 'shall' in it & the rest are 'should's. Also thinks amending is an inefficient way for this committee to proceed. Hough: agrees with Kidder & Kahan. Golliver: this (chapter 8) isn't going to solve the problem. It is reasonable that this is on the table to get closure. Kidder: the draft text is speaking to the language people (to provide these constructs) not hardware guys. Thomas: Simpathizes with Kahan but doesn't think this approach is going to be productive. Kahan: asked if examples of implementations were presented, would that allay your objection? Hough: answered no. Called for vote: 7+1 favor, 6+3 against, amendment fails. Golliver called the question. Godard seconded. 7+9 favor, none opposed, motion passes. Original motion as amended. 15+11 favor, 2 opposed, passes. pee break at 3:10, came back at 3:25 14. Motion: If the 2006-03-07 motion has failed (Peter Markstein) moot. 15. Motion: If the 2006-03-07 motion has passed, Shall we remove references to Counting Mode under Alternate Exception Handling and reinstate the scaled product functions as proposed in (David Hough) (specific list removed in past) http://754r.ucbtest.org/ballots-2005/withdrawn/scaledproduct.pdf? Hough: remove references to Counting Mode under Alternate Exception Handling and reinstate the scaled product functions as proposed in http://754r.ucbtest.org/ballots-2005/withdrawn/scaledproduct.pdf. Kahan seconded. Hough: counting mode came back when scaled product defeated. Thomas: how does this fit in with other operations? Kahan: this (counting mode) evolved from something I introduced in the IBM 709 in the 1960s. Okada: 'very rarely used' therefore belong in a library for those functions rather than in the standard. Zuras: These are the only known use of counting mode. shown Wigner formula. Kidder: discussed our thinking on this in the numeric subcommittee. These are the primary cases someone would want to handle. Cowlishaw: this motion is presented as an alternative. Golliver: counting modes is in our working draft. Kahan: magnitude of two complex variables. Zuras: Markov modelling the bio-informaticians use. Canon: This happens elsewhere as well. Golliver: implementation of counting mode has scope problems, scaled products are limited, much simpler in the tradeoff. Schwarz: Feels its difficult to implement. Difficult to round correctly when 'n' is unbounded. Doesn't feel its a valid operation for that reason. Kahan: possible with a 32-bit integer. Kidder: points out it is easier than Schwarz thinks to produce accurate results. Vote on the issue. 11 for, 5 against, passes. 16. Motion: If the 2006-03-07 motion has passed, Shall we remove specification for alternate exception handling (Chapter 8) from the main body of the draft? (Jim Thomas) Thomas: To remove the specification in section 8 in the manner described in the associated email from the main body of the draft. Markstein seconded. Kidder; moves to amend the motion to place the section 8 text in an annex. Markstein seconded. Markstein: moved to amend to change all 'shall's in section 8 to 'should's. Crawford seconded. Kidder accepts that as a friendly amendment. Now reads amend the motion to place the section 8 text to an annex & change all 'shall's to 'should's. Cowlishaw: It is a good idea we know where the text goes. Vote: 8+3 favor, 1+2 opposed, amendment passes. Back to main motion: to move text of section 8 to an annex, change 'shall's to 'should's, & make the edits listed in the email. Thomas: apologise for confusion. Kahan: believes this will delay progress for years. We should try for a consensus. Doesn't believe the language problem is as bad as Thomas believes. Golliver: fastest approach is to take existing solutions to language committees. Kahan: believes language people don't see customer demand. Kidder: Agrees with Golliver & the long view on this. Vote the issue: 11+4 for, 3+0 against, motion passes. 18. Motion: If the 2006-03-07 motion has passed, Shall we delete the section on Varying-Width Floating-Point Formats? (Roger Golliver) (tower defeated previously, varying adopted 10/14/04 w/o integers) Golliver: move to delete section 3.5 & the first bullet item on page 19. Crawford seconded. Zemke: was confused. Golliver: wants to find things to remove from the standard in an effort to agree on a draft quicker. There exist packages for this functionality. Thomas: I don't think we have ever identified the scope of implementations for this. Draft isn't consistent. Davis: speaking as originator of the document: to support Demming by providing a set of precisions that everyone could use portably. Originally the 'tower of power'. Kidder: we had MPFR (over GMP) in mind. Kahan: 854 can serve this function w/o language problems. Variable precision takes the precision as a more or less continuously variable parameter. Davis: pointed out that 854 is gone. Kahan: document still exist as guidance. Dynamic changing is also a problem. Vote: 11+7 favor, 3+1 opposed, motion passes. 19. Motion: If the 2006-03-07 motion has passed, Shall we delete the section on Tail Operations? (Roger Golliver) (adopted in past) Golliver: move to delete the tail operations. Markstein seconded. Golliver: same argument as before, there were functions without a lot of 'newness' in them. Kahan: we want the tail functions specified by name so that people need not write their own & the optimizer doesn't eliminate them on the grounds of computing zero. Kidder: agrees. These are known but not widely available so canonizing them allows the programmer to understand their use. Golliver: in the interest of time, he is willing to withdraw the motion. Thomas: system provided function is subject to optimization too. Golliver moves to withdraw, done w/o objection. Peter: move to take 2/3s motion off the table. Kidder: this modifies the P&P. Zemke: actions you take today could be overruled tomorrow. Therefore, motions to tie the bodies hands are out of order. If we change our procedures, that is of another level. Davis: This involves going to the MSC for this change. Hough: moves to withdraw the motion. Kidder seconded. None opposed, hearing none, it passes without opposition. Kahan: asks to consider the editor's frustration over the committee changing its mind often. Suggests we log the reasons for our decisions. Nelson: this sounds like a further burden on the editor. Crawford: observed we made progress today. Davis: wants to discuss the list of items to be considered versus those that are closed. What do we have to attack to finish the draft? Dave posted one list, Mike posted an update. Can be found by googling 754r issues. Mike showed his list & we may be able to go over it in the draft review tomorrow. Zuras would like to make the list explicit & start to assign tasks. Thomas believes this may look more rosy than it might seem. 20. Proposal: Self selecting groups to clean up drafts Hough: says this is moot. Kidder: wanted to talk about Dr Pryce's proposal. Zuras: only some of these cases make sense outside intervals. Kahan: one must use tagged intervals to include or exclude the endpoints to know the correct answer to these questions. In the multidimensional case, we need some generalization of intervals (to hyper- ellipsoids or something) to avoid an explosion of volume. Kidder: asked of Davis what we should do about people like Pryce. Davis: perhaps some can be put in an annex. Postpischel moved to ajorn, Dick seconded, done by acclimation. 21. Adjournment Next Meetings ------------- Wednesday 3/15 full meeting, Joe Darcy at Sun in Santa Clara Thursday 3/16 draft review, Joe Darcy at Sun in Santa Clara Wednesday 4/19 full meeting, Roger Golliver at Intel in Santa Thursday 4/20 draft review Wednesday 5/17 full meeting, Roger Golliver at Intel in Santa Thursday 5/18 draft review Wednesday 6/21 full meeting, Jim Thomas at HP in Cupertino Thursday 6/22 draft review Monday 7/10 6:00 MSC Wednesday 7/19 full meeting, David Hough at Sun in Menlo Park Thursday 7/20 draft review Wednesday 8/9 full meeting, Eric Postpischil at Apple in Ten Forward Thursday 8/10 draft review, Conference Room at 2 Infinite Loop, Cupertino Wednesday 9/20 full meeting, TBA Thursday 9/21 draft review Monday 10/9 6:00 MSC Wednesday 10/18 full meeting, Fred Zemke at Oracle in Redwood City Thursday 10/19 draft review, Fred Zemke at Oracle in Redwood City Wednesday 11/15 full meeting, Eric Postpischil at Apple in Ten Forward Thursday 11/16 draft review, Conference Room at 2 Infinite Loop, Cupertino Wednesday 12/20 full meeting, TBA Thursday 12/21 draft review 754's PAR expires 12/06 Monday 1/8 6:00 MSC Monday 4/9 6:00 MSC Monday 7/9 6:00 MSC Monday 10/8 6:00 MSC