On 4/24/11, Ted Unangst <[email protected]> wrote: > On Sun, Apr 24, 2011 at 12:27 PM, Martynas Venckus <[email protected]> > wrote: >>> I think we should. Arch specific notes in man pages are not without >>> precedent. If something can happen, the man pages on my system should >>> tell me about it. >>> >>> ifconfig tells me about all sorts of cool things my computer can't do. :) >> >> The manual page documents standard IEEE 754 features, supported by >> our implementation. >> >> Alpha's FP_X_IOV should not be used; our architecture doesn't even >> maintain it properly since we call softfloat routines for conversions. >> >> We don't document it to discourage ever use it; we have plenty of >> precedents for those. > > OK, that's reasonable. This is a little outside my territory, but the > i386 denormalization exception does sound like something that would be > of interest to somebody who cares about floating point.
The amd64/i386 denormalization exception can be documented if clearly marked as MD extension. Maybe like this; OK? Index: fpgetmask.3 =================================================================== RCS file: /cvs/src/lib/libc/gen/fpgetmask.3,v retrieving revision 1.6 diff -u -r1.6 fpgetmask.3 --- fpgetmask.3 23 Apr 2011 21:37:14 -0000 1.6 +++ fpgetmask.3 24 Apr 2011 18:39:33 -0000 @@ -78,6 +78,7 @@ .It Dv FP_X_OFL Ta Overflow .It Dv FP_X_UFL Ta Underflow .It Dv FP_X_IMP Ta Imprecision (inexact) +.It Dv FP_X_DNML Ta Denormalization Pq amd64 and i386 only .El .Pp The
