IEEE 754R notes from December 16, 2004

Notes taken by Dan Zuras.

December 16, 2004: Draft review

Notes for meeting at Intel in Building 12 in Santa Clara at 1:00 on Thursday 12/16/04. Jeff Kidder hosted us in room 209. Dick Delp, David Hough, Prof Kahan, Ivan Goddard Jon Okada, Jim Thomas, Mike Cowlishaw, Eric Schwartz, Peter Markstein, & Dan Zuras attended.

Dave started with Basic Formats.

Prof Kahan moved: Let us have a document so we can ask the graphics folks & audio people what they think of: 12/11 sig bits, arithmetic format, simulation. Motion passed.

Dave moved that Basic Formats be adopted. Peter amended that text be black. Motion passed.

Next, Underflow. Kahan pointed out that this is a mistake. After some considerable discussion, Jeff proposed we send the motion back to subcommittee.

Next to Expressions. The discussion started with the question (raised by Ivan) of whether or not we should generalize the notion of contractions from FMA & add/subtract (our list) to any operation just before an assignment.

At 3:30 we broke for a pee break & to change rooms.

Afterwards, I asked about guidence for the future of Annex Z. Jim would have FixedIntegers be a range converted to & from & would like to see justification of variable integers. Prof Kahan doesn't believe we should say anything about integers that we don't absolutely need to. Mike would say nothing about fixed integers at all & only mention arbitrary integers as a subset of the reals. Jon doesn't think we should think of the integers should be thought of as a subset of reals. And leave out arbitrary integers entirely. Fixed length as source or destination for conversions only. Jim doesn't want to say anything about infinite precision integers.

Jim moved to remove integers from Annex Z. Dave seconds. Passed with one objection.

Back to expressions. Ivan brought up a*b +/- c*d (with one rounding error) as another example of a possible contraction. We decided to go back to committee on this proposal. Jim pointed out that if the defining charicteristic of a contraction is that it is an expression evaluated with one rounding error.

Yesterday's bin list:

Today's bin list:

754 | revision | FAQ | references