Mail Archives: djgpp/1995/02/08/02:17:49
On Tue, 7 Feb 1995, Stephen Turnbull wrote:
> I think that this has been discussed on the list. Try searching the
> archives (see the URL below if you don't want to download the 8MB
> monster from Clarkson).
I would love to, but I have a dial-up terminal account with my service
provider, and they run the Lynx WWW browser (is that the correct term)
and it seems to be one-way. That is, I tried to search your archives but
Lynx seems not to want to send input from me to your server. Also, I am
limited to 2MB disk space on their Sun, so downloading 8MB (and then
transferring it with zmodem!) is out of the question. Maybe I just have
to learn more about using Lynx? Or maybe I need a proper SLIP account,
bu that seems to mean windoze...
> Sure, it's dangerous, but what choice does he
> have if he wants efficient code (ie, having the FPU trap it)?
Surely the point here is to _not_ have the FPU trap an exception, so that
Nan or +-Inf can stay on the FPU stack and be returned (but then how do
you handle a NaN value, surely printf() won't understand it?) without
raising an exception.
> I believe that affine means that there are plus and minus infinity;
> projective (like complex) only has one infinity. (That's what it
> means in math, IEEE definitions may be something else again.) So this
> is not going to return NaN.
My math never got that far, although I did do some set theory (or it did
me...). I always assume affine meant something like "not finite" and
projective meant something like "close as I can come". But as I said, I
never knew what those terms meant. Anyway, the fellow who asked wanted
to get Inf or -Inf as answers, not just NaN.
Thanks for the info. I keep trying to learn more about the '87, it
possesses some very powerful instructions and is far more configurable
than most people realize, but it must be one of the least documented
processors around!
Bill Davidson
bdavidson AT ra DOT isisnet DOT com
- Raw text -