IEEE 754R notes from November 17-18, 2004

Notes taken by Dan Zuras.

November 17, 2004: Draft review

Notes for meeting at Sun in 16 Network Circle in Menlo Park at 1:00 on Tuesday 11/17/04. David Hough hosted us in the Martin's Beach Room in building 16. Mike Cowlishaw, Jim Thomas, Neil Toda, & Dan Zuras attended. Eric Schwartz, Jeff Kidder, Leonard Tsai were on the phone.

We began on the draft. Page 9. On Page 12, we discussed the details of the various flavors of FMA (+/-)a*b +/- c. Jim asked if we should require all of the flavors. The first sentance on page 39 caused a lengthy discussion of the contents of the table on page 40 & 41. Mike brought up the issue of the possibility of something like (float)min(x,y) where x & y are doubles. As we have always thought of min & max as selection operations we have never considered the possibility of CHANGING the value of an operand as it passes through a min/max operation. I'm not sure we should but Mike says he already does something like this in decimal. Much discussion on all the colorful text about conversions on page 46. Mike wanted to put in a paragraph about decimal representations WRT string formats. Mike changed the paragraph he was supposed to on page 70.

November 18, 2004: Discussion

Notes for meeting at Sun in 17 Network Circle in Menlo Park at 1:00 on Tuesday 11/18/04. David Hough hosted us in the Beach Room in building 17. Alex Liu, Mike Cowlishaw, Jim Thomas, Joe Darcy, Ivan Goddard & Dan Zuras attended. Eric Schwartz, Leonard Tsai, & Jeff Kidder were on the phone.

Annex D went into the standard with little discussion.

Jim raised the issue of splitting the standard into the 1985 text + quad, min/max, & FMA & an Annex D for decimal. Jim wants to raise this as an agenda item in the December meeting. Find out with IEEE if we can have two votes.

The letter is to be shipped.

We finished the agenda & went on to BigDecimal, the topic that got us stuck yesterday. (Joe is in the room now.)

Ivan had a new topic after the 3:00 break. It was general background information about the Open Multi-Processing (OMP) standard for supporting MP annotations for Fortran. He thinks we should distinguish the highest precision available from the rest & relax the rules along these line: it should not support: modes, traps, gradual underflow, operations need not correctly round & reassociation should be permitted. Most of it exists today & is outside the standard. We are trying to bring some of it into the standard. Jeff characterized the problem as licensing expressions to be evaluated in another way so long as it is more precise than the operands. We spent a good deal of time discussing various expressions & problems with their evaluation.

Jeff wanted to look Annex L. Ivan is here so this is a good thing. Dave mentioned his "no action at a distance" notion.

We ended about 5:00.

754 | revision | FAQ | references