raw partial April 15 Minutes
[There were three note-takers. Here are my notes.]
IEEE 754r Meeting minutes March 17, 2004
Sun, Santa Clara
Dan Zuras Self
David Hough Sun
Dick Delp Self
Eric Schwartz IBM
Jeff Kidder Intel (minute taker)
Jim Thomas HP
Joe Darcy Sun
John Okada HP
John Hauser self
Matthew Applegate Arithmatica (host)
Michael Parks Sun
Mike Cowlishaw IBM
Peter Markstein HP
W Kahan Univ of Calif
Note taker: Jeff Kidder
Called to order 10:05
* Draft review
3.0 At level 3 represent qNaN vs sNaN first at level 3.
(in the draft it was just NaN vs sNaN). This change
rippled through in a few other places.
"digit string" or "digit sequence"?
5.4 changed "as specified in 7" changed to "as specified in section 7"
5.5 accepted (with change of "number" to "representation" in title and
in the body with some wording change)
5.11 Accepted (including "canonical")
changed to "result representable entities depend only on the operand
"whether or not" accepted
Move conversion from a binary format to a decimal format to first table
Still need to resolve the max(NaN,NaN) and min(NaN,NaN) cases.
7.1 accepted three changes
Appendix 1 accepted
*MSC report out [Dan]
Asked about extension on par and deadline. Approved extension to June
Asked about naming and structure.
They suggested put everything in the normative text and
what needed to be done to comply. In particular, this
included the language text "Appendix L".
"Upward performance compatible" could be confined
to the "lower conformance level". The tighter constraints
would be included at a "higher conformance level".
E.g., sections 1-8 are the lower conformance level.
sections 1-10 are the higher conformance level.
Short term, call them section N and section L for now.
Could have an informative appendix about C99.
Should appendix 1 be in a section?
Should we treat the appendices as "specification under development".
Have both an appendix N and a section N. Things move into the section
when we agree they should be in the higher conformance level.
* Should and shall
One can have an appendix is deprecated and should use annex.
Could include normative text but generally isn't.
* Name of the split
* NaN Proposal
2. Conversions (binary: leading bits, decimal: trailing bits)
3. sNaN encoding
3.1 preserve common QNaN bit
3.2 All 1 sNaNs?
3.3 Distinguish sNaN/quieted sNaN/ NaN from invalid/ usrNaN?)
4. Integer versions of payload functions
Next meeting May 12-13 1-5 p.m. at Intel, Jeff Kidder hosting.