Date: Mon, 6 Nov 1995 20:47:43 +1100 From: Bruce Evans <bde@zeta.org.au> To: bde@zeta.org.au, markd@grizzly.com Cc: freebsd-hackers@freebsd.org, jkh@time.cdrom.com, julian@ref.tfs.com Subject: Re: NPX still broken in 2.1.0-951104-SNAP... Message-ID: <199511060947.UAA10166@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>>I think the current behaviour (trapping) is more useful for
>>most programs. Programs that can handle IEEE arithmetic
>>should do something special to check and get it at confiuration
>>time since it is not guaranteed.
>Its incompatible with every Unix system I have access to. See enclosed table.
>No system core dumps! I don't have any standards handy, but SCO, which
>is X/Open, ANSI, and POSIX compatible sets errno and returns NaN.
The ANSI standard says that math functions shall not cause any externally
visible exceptions; for domain errors, errno shall be set to EDOM and an
implementation-defined value shall be returned; for range errors, errno
shall be set to ERANGE and a correctly signed +-HUGE_VAL shall be returned.
What do these systems do for common application errors? E.g.,
/* 1. Overrun i387 stack. */
main() { int i; for (i = 0; i < 1000; ++i) foo(); } double foo() { return 0; }
/* 2. Overflow. */
#include <math.h>
main() { double volatile d; d = pow(1e6, 1e6); d += d; d += d; }
/* 3. Invalid operation. */
#include <math.h>
main() { double volatile d; d = pow(1e6, 1e6); d -= d; d -= d; }
(I assume that HUGE_VAL is +Infinity...)
Systems that use __IBCS_NPXCW__ for the default control word have to do
extra work in the library functions so that the library functions don't
trap. I think some of the i386 Unixes do this.
>>The change similar to mapping page 0 and putting 0's there so
>>that strcmp("foo", NULL) works right.
>>Surely such a simple change that might help some people
>>is worth puting in now? ;-)
>I don't see this as being analogous. strcmp to NULL is an undefined, invalid
>operation. IMHO calling acos with an out of range operand is an error case
>akin to calling open with an invalid file name. The big difference between
>this and strcmp is that you can read a man page and find out the expected
>error respones.
The problem is that the control word is global so changing it affects the
behaviour of d += d and d-= d above as well as fixing the library functions.
>It seems pretty straight forward to modify k_standard.c to return NaN
>in POSIX_MODE. Making the modified POSIX mode the default, and changing the
The correct value isn't passed to __kernel_standard() so one has to be
invented. This is fairly easy for acos(|x|>1) because there is only
one reasonable value (the default NaN). Other cases are more difficult.
E.g., for pow(0, negative), the correct value is +HUGE_VAL for +0 and
-HUGE_VAL for -0 but __kernel_standard always returns -HUGE_VAL. There
are 19 different cases for the `IEEE' version of pow(), but only 6 cases
for pow() in __kernel_standard().
>----------------------------------------------
>SCO 3.2v5:
>acos (2.0) = nan, errno = 33
Correct.
>----------------------------------------------
>BSDI 1.1:
>acos (2.0) = NaN, errno = 0
Wrong errno (from BSD libm?).
>----------------------------------------------
>HP-UX A.09.01:
>acos: DOMAIN error
>acos (2.0) = 0, errno = 33
Bogus side effect (extra output).
>----------------------------------------------
>IRIX 5.3
>acos (2.0) = nan0x7fffffff, errno = 33
>----------------------------------------------
>OSF/1 V3.0
>acos (2.0) = 0, errno = 33
>----------------------------------------------
>ULTRIX 4.3
>acos (2.0) = NaN, errno = 33
Correct.
>----------------------------------------------
>SunOS olden 4.1.3_U1 1 sun4c
>acos: DOMAIN error
>acos (2.0) = NaN, errno = 33
>----------------------------------------------
Bogus side effect.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199511060947.UAA10166>