svn commit: r230193 - head/lib/libc/sparc64/fpu

[David Schultz]

On Tue, Jan 17, 2012, Bruce Evans wrote:
> On Mon, 16 Jan 2012, David Schultz wrote:
> 
> >Log:
> > Computations on NaNs are supposed to return one of the input NaNs 
> > unchanged.
> > Fix a few places in the sparc64 floating-point emulator where this wasn't
> > being handled properly.
> >
> > Submitted by:	bde
> 
> Thanks.  The only remaining large bug that I noticed near this is that
> without -mhard-quad-float, signaling NaNs are not quieted and (IIRC)
> FE_INVALID is not raised.
> 
> BTW, NetBSD in 2005 uses Hauser soft-float for long doubles on sparc64,
> and at least the MI parts of it are almost identical with the Hauser
> soft-float in FreeBSD.  But FreeBSD uses a completely different version
> of soft-float for sparc64.  It was apparently what NetBSD was using
> in 2002 when it was imported into FreeBSD.  Perhaps the Hauser version
> is better (more correct or faster).  However, the other version is
> much simpler and looks much nicer -- it was originally from Berkeley
> and has almost perfect KNF formatting (over 95% of lines are perfectly
> formatted accoring to knfom; that is much better than 4.4BSD-Lite2
> sys/kern/*.c (88%) and FreeBSD-current sys/kern/*.c (89%, not counting
> kern_intr.c which is about 0% after a single C++ comment in it confuses
> indent(1)) and contrib/nvi/*/*.c (94%).  Hauser soft-float has a Gnuish
> style with 4-char indents amd is 26% KNF (probably mainly for the the
> empty lines and some comments).

softfloat is probably better.  The style of contrib sources is
what it is, and we have worse in the tree.

That said, I don't have the cycles right now to fix what ain't
broken.  Moving all of the libc/quad/ floating-point routines to
softfloat is more important, because that stuff *is* broken.