This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC error ?
On Friday 07 June 2002 00:40, Hans-Joachim.Schmidt@bmf.bund.de wrote:
> Hello,
>
> there is a problem when I compute:
>
> double x,y,d,e;
>
> x=112.7;
> y=120.0;
>
> d=x/y-1;
> e=x/y; e--;
>
> d=0x1.70a3d70a3d711p-6 (output of %a format)
> e=0x1.70a3d70a3d700p-6
>
> and
> d-e=0x1.1000000000000p-54
>
>
> Does the output of 'd' and 'e' (in %a format) show that there is more than
> 1 bit difference ?
>
> We use SuSE Linux 8.0 (i386) - Kernel 2.4.18-64GB-SMP on a Pentium III
> (Coppermine).
>
>
It looks as if the division in the first expression has been rounded only to
64-bit precision, while the value of e has been rounded again to 53-bit
precision before being decremented. This would be an expected behavior, in
the absence of any instructions by you to the contrary. There was a thread
on this earlier this week, how the choice of precision mode mode is the
responsibility of the library, not gcc.
--
Tim Prince