serge@pacific.UUCP (serge) (01/09/86)
Is there some sort of standard in unix or IEEE for how to deal with the
following floating point issues. If there is no standard, what is the
"preferred" action? I am especially interested in how 68000-based
systems behave, but would also be interested in knowing how other Unix
systems deal with it.
double a, b, c;
[assignment to b and c]
c = a / b;
1) If `a' == 1.0 and `b' == 0.0 , should the divide-by-zero cause
a SIGFPE? (on some Unix systems it does not). If a signal
should not be generated, then what value should be
assigned to `c'?
2) If SIGFPE is being ignored or caught (perhaps there are two
answers), what value should be assigned to `c'? I suppose the
possibilities are infinity, a NaN, or some random value since
one could claim that dividing by zero produces an undefined result.
3) If `a' and `b' above both contain infinity, then what should
happen (infinity divided by infinity is a Nan)? Should SIGFPE
be generated in this case? What if it is ignored or caught?
4) If you manage to get the value of infinity or a nan into a
a float or double variable, what should printf(3) do with it
if you pass this value to it? I have seen various Unix systems
deal with this in one of three ways:
a) generate a SIGFPE
b) print a large number (for infinity)
c) go into an infinite loop (usually in ecvt(3))
Please respond by E-mail and I will post a summary if there is enough
interest. I apologize if this has all been discussed before (if it
has, would someone please let me know what, if anything, was decided).
Thanks!
Serge Issakow
--
sdcsvax \
>!pacific!serge
s3sun /
lee@haddock.UUCP (01/10/86)
Is there some sort of standard in unix or IEEE for how to deal with the following floating point issues. If there is no standard, what is the "preferred" action? Yes there is an IEEE standard. No there is not a Unix standard, tho there are Unix IEEE systems and other Unix systems to compare to. double a, b, c; [assignment to b and c] c = a / b; 1) If `a' == 1.0 and `b' == 0.0 , should the divide-by-zero cause a SIGFPE? (on some Unix systems it does not). If a signal should not be generated, then what value should be assigned to `c'? If the system is IEEE, the operation should not cause a signal and +INF should be assigned to c. All non-IEEE systems I know about generate signals. 2) If SIGFPE is being ignored or caught (perhaps there are two answers), what value should be assigned to `c'? I suppose the possibilities are infinity, a NaN, or some random value since one could claim that dividing by zero produces an undefined result. The IEEE spec defines this; I believe it is still +INF but I would have to look it up. In a non-IEEE system, usually the hardware defines the result. 3) If `a' and `b' above both contain infinity, then what should happen (infinity divided by infinity is a Nan)? Should SIGFPE be generated in this case? What if it is ignored or caught? And so on. The IEEE spec (get 10.0 or later, if there is one) is quite explicit about the values and signals. There is assumed to be a capability to change whether or not actions such as divide-by-zero will cause signals, but such a facility is highly non-portable and most systems have not been extended in this way. 4) If you manage to get the value of infinity or a nan into a a float or double variable, what should printf(3) do with it if you pass this value to it? I have seen various Unix systems deal with this in one of three ways: a) generate a SIGFPE b) print a large number (for infinity) c) go into an infinite loop (usually in ecvt(3)) The AT&T SystemV code has a special line in the entry to the printf formatter that looks at the input value and if it is a NaN or INF, it kills itself with a SIGFPE. This code is "standard" in the sense that AT&T supports IEEE machines and that is what they do.
spaf@gatech.CSNET (Gene Spafford) (01/14/86)
You can obtain a copy of ANSI/IEEE Standard 754-1985, "IEEE Standard for Binary Floating Point Arithmetic" by sending $6.50 plus $2 handling to: IEEE Service Center 445 Hoes Lane Piscataway, NJ 08854 NJ residents must add 6% sales tax to the $6.50 (= $6.89). IEEE members get a 10% discount (on the $6.50, *not* on the $2 handling) if you include your member number with the order. Canadian orders can be mailed to: IEEE Canadian Regional Office 7061 Yonge St. Thornhill, Ont. L3T 2A6 and the conversion rate is US$ * 1.33 = CDN$ There are quantity discounts for 10 or more copies. Details can be directed to +1 201 981 0060. I have all this information handy because I just ordered a copy.... -- Gene "the end is in sight" Spafford The Clouds Project, School of ICS, Georgia Tech, Atlanta GA 30332-0280 CSNet: Spaf @ GATech ARPA: Spaf%GATech.CSNet @ Relay.CS.NET uucp: ...!{akgua,decvax,hplabs,ihnp4,linus,seismo,ulysses}!gatech!spaf