george@cis.ohio-state.edu (Karl Kleinpaste) (06/01/89)
One of our grad students is trying to write some routines to do generic
parameter passing (i.e. give it the address of a variable, the size
and let bcopy store/fetch the value). In the process of doing this
we came across some interesting behavior with floating point variables
on the Multimax. The following piece of code
main()
{
float f = 4.0;
printf("float = %f\n", f);
printf("float = %x\n", f);
}
on a sun and a pyramid will print both the floating point value (4.0) and
then the raw hex value showing how the floating point number is stored.
On the Multimax the second value is reported as 0. Is this (a) a compiler
bug (b) a fuzzy area in the definition of C (c) other ?
Thanks,
---George Jones
OSU Computer & Inf. Science 2036 Neil Ave.,Columbus,Ohio 43210. 614-292-7325
george@cis.ohio-state.edu or ...!osu-cis!george
Wishing you joy as you celebrate Christmas !goldman@xenna.UUCP (Steve Goldman) (06/01/89)
Answer is (c) other. Multimax floating point is little endian. f is promoted
to double on each printf call. The second call dutifully prints the low order
32 bits of the mantissa which are zero. On a big endian machine you'd get the
high order bits which in this case are much more interesting.
Steve Goldman, Encore Computer Corp (919) 481-3730
901 Kildaire Farm rd., bldg D Cary, NC 27511 USA
arpa: goldman@Encore.COM (129.91.1.14)
uucp: {necntc,talcott,decvax,allegra}!encore!goldman
Steve Goldman, Encore Computer Corp (919) 481-3730
901 Kildaire Farm rd., bldg D Cary, NC 27511 USA
arpa: goldman@Encore.COM (129.91.1.14)
uucp: {necntc,talcott,decvax,allegra}!encore!goldmandrs@bnlux0.bnl.gov (David R. Stampf) (06/02/89)
In article <50537@tut.cis.ohio-state.edu> george@cis.ohio-state.edu (Karl Kleinpaste) writes: >on the Multimax. The following piece of code > >main() >{ > float f = 4.0; > > printf("float = %f\n", f); > printf("float = %x\n", f); >} > >on a sun and a pyramid will print both the floating point value (4.0) and >then the raw hex value showing how the floating point number is stored. >On the Multimax the second value is reported as 0. Is this (a) a compiler >bug (b) a fuzzy area in the definition of C (c) other ? > >Thanks, >---George Jones On the multimax, the sizeof(double) is 8 bytes, sizeof(long) is 4 bytes and I would venture a guess that the floating format has the words in opposite order from sun/pyramid. The printf converts f to a double, then pushes it on the stack which looks like this. 00000000 <- least sig. bits of 4.0 04010000 <- or something like this for msb The %x (or %lx) picks off the top one and obediently prints 0. If you had written instead: printf("%lx %lx\n",f); you would see the correct format. Had you said f = 2.0/3.0, you would have seen something that looked better, but it would have been wrong, since you would only see the lower bits. < dave
guy@auspex.auspex.com (Guy Harris) (06/03/89)
>If you had written instead: > > printf("%lx %lx\n",f); > > you would see the correct format. At least on some implementations, including, presumably, the Encore one. If you *really* want to dump the bits of a value in a way that doesn't depend on the details of parameter passing, write a "dump bits" routine that takes a "char *" (or "unsigned char *") and perhaps a byte count, and do dump_bits((char *)&f, sizeof f); or something like that.