754R subcommittee meeting
The next 754R subcommittee meeting will be Thursday July 29,
at 1:00pm - 6:00 pm at Sun
Microsystems Menlo Park Building 14, in the Sandhill conference room 3591.
Go to the Building 14 lobby and call the conference room at 650-786-7242.
To buy lunch first in the company cafeteria, please arrive at the Building 14
lobby by 12:00 noon.
Please let me know if you are planning to attend in person or would like
to participate by telephone.
On Tuesday July 27, I. Godard, D. Hough, W. Kahan, and D. Zuras discussed
concurrency issues, as raised in
The most pressing change to the draft seems to be
to emphasize that modes and flags are to be thought of as variables like
other variables in a particular language, with defined scopes. The current
common implementation artifact of unique hardware
global modes and flags that can be
asynchronously modified with instantaneous global effect is not to be
considered mandatory or even desirable.
A related notion is that
part of the signature of a function under IEEE arithmetic, along with its
argument and result types, is an indication of whether
it inherits caller's modes on entry
it inherits caller's flags on entry
it sets caller's modes on exit
it sets caller's flags on exit
A programming environment should define the default cases and provide methods
of indicating non-default cases.
The next major issue lurking here is whether dynamic modes are mandatory
or not. Full run-time dynamic modes are a common implementation mechanism,
but are they required? In an environment of link-time code generation,
one can imagine that debugging foreign object code might only require
modes that can be changed at link time rather than run time.
We didn't get to any concrete standards language embodying these concepts.
We plan to continue this discussion on July 29.