Date: Tue, 16 May 2000 17:22:40 -0400 (EDT)
Message-Id: <200005162122.RAA20137@indy.delorie.com>
From: Eli Zaretskii <eliz AT delorie DOT com>
To: "Dieter Buerssner" <buers AT gmx DOT de>
CC: djgpp-workers AT delorie DOT com
In-reply-to: <200005161320.JAA11285@delorie.com> (buers@gmx.de)
Subject: Re: Math functions
References: <200005151237 DOT PAA11341 AT is DOT elta DOT co DOT il> <200005161320 DOT JAA11285 AT delorie DOT com>
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

> From: "Dieter Buerssner" <buers AT gmx DOT de>
> Date: Tue, 16 May 2000 15:26:20 +0200
>
> > > When you "silently" return a NaN, say by log(-1.0), the bit will not 
> > > be set in the status word.
> > 
> > Are you sure?  
> 
> I tested it on my AMD K6-2. I will append some source.
> 
> > The Intel manual seems to say otherwise: it says that
> > any invalid operation sets the appropriate bit in the status word.
> 
> Loading a NaN into a FPU register seems not to be an invalid 
> operation (which makes sense to me).

Intel's text is limited to a QNaN, the one which is produced as
invalid operations' masked response.  If you tried that with an SNaN,
perhaps the behavior is different.  But the FPU itself never produces
an SNaN.

> I find it rediculous, that a portable program must be 
> aware of pow(NaN, 0) == 1.0.

Me too.  I think this is one of the cases where we don't comply with
these requirements.

> > I believe that you interpret the word ``exception'' too literally
> > here.  The C99 standard does not specify what does ``exception''
> > mean. 
> 
> This perhaps is a quality of implementation issue. In Annex H 
> (informative) I read:
> 
> C does require that SIGFPE be the signal corresponding to arithmetic 
> exceptions, if there is any signal raised for them.

Sure, but note two things: (1) Annex H is _informative_, not
normative, and (2) it says "if there is any signal raised".

> Again, I don't think, that exceptions should be the default. Just 
> give people, who wants them, the choice.

If it can be done without compromising the default, I don't object.

> With the following program I get the output:
> 
> after silent inf   x = Inf, sw = 0x00
> after trapping inf x = Inf, sw = 0x04
> after silent NaN   x = NaN, sw = 0x00
> after trapping NaN x = NaN, sw = 0x01

I get the same, but we were talking about a NaN that resulted from an
invalid operation, like sqrt(-1).  Loading a NaN forcibly is hardly an
interesting case, right?