[Cfp-interest 2727] WG14 IEEE 754-C binding meeting minutes 2023/02/28
Jim Thomas
jaswthomas at sbcglobal.net
Sat Mar 4 08:39:38 PST 2023
Corrections and additions welcome.
=============
CFP meeting 2023/02/28
Damian McGuckin,
David Hough,
Fred Tydeman,
Hans Boehm,
Ian McIntosh,
Jim Thomas,
Mike Cowlishaw,
Paul Zimmermann,
Rajan Bhakta,
Vivian Van Loan
March 29, 8 AM PDT / 11 PM EDT / 3 PM UTC - next meeting
Floating point rounding modes - lots of email discussion.
Hans presented slides (will be emailed to reflector after meeting).
fenv rounding issues.
interacts with constexpr.
FENV_ACCESS not supported in C++
Uses:
interval arithmetic
run same code with different rounding modes to assess the effect of round-off errors
Problems
no constant folding
effect on library functions
rounding of signed constants
global FP state
Solutions
deprecate fesetround()
add static rounding to C++
correctly rounded math functions with extra argument
(to indicate rounding direction)
correctly rounded FP types
David Hough: Dynamic rounding and compilers are a bad mix.
Static rounding modes are much better.
Likes dynamic exception modes (for debugging)
WG14 meeting results: all CFP suggestions accepted.
Carry-over Action Items:
David Hough not done yet.
Action Items from last meeting
All completed
N3105 accept by WG14
Jim to submit [Cfp-interest 2684] as a CD2 comment.
Jim to submit [Cfp-interest 2686] as a CD2 comment.
"floating types" and float_t, double_t.
Jim to submit CD2 comment to italicize “real floating types” in 6.2.5#14.
Re definition of float_t, double_t
Fred: Microsoft eval method is not any of float, double, long double,
so, what should float_t be?
No resolution yet for CD2 comment
To be continued in email discussion
"width" issue
No resolution yet for CD2 comment
To be continued in email discussion
TS Part 4 -- minor formatting updates.
TS Part 5 -- redone as four separate features.
It still has a few issues to be resolved.
Fred Tydeman
Jim Thomas
Rajan Bhakta
More information about the Cfp-interest
mailing list