This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug middle-end/21718] real.c rounding not perfect
- From: "neil at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 18 Oct 2007 15:24:18 -0000
- Subject: [Bug middle-end/21718] real.c rounding not perfect
- References: <bug-21718-6@http.gcc.gnu.org/bugzilla/>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Comment #6 from neil at gcc dot gnu dot org 2007-10-18 15:24 -------
(In reply to comment #5)
> I believe more than 160 bits are required to get even single-precision numbers
> right with DECIMAL_DIG digits precision and an exponent. I'm going to try and
> prove this by finding an example. It could be hard as I believe 160 is only
> just below the required number.
Here's an example to prove this assertion. When compiled with GCC 4.1.2 or
4.1.3 with -std=c99, assuming a correctly-rounding libc (e.g. NetBSD's; glibc
also seems to get this correct) you get the following output:
0x1.8p-147 0x1.4p-147 8.40779078594890243e-45
So not only is it rounded incorrectly, but the number it is rounded to, when
converted back to decimal, does not even match the input number in the first
digit.
#include <stdio.h>
#include <stdlib.h>
int main (void)
{
float f1 = 7.7071415537864938e-45;
float f2 = strtof ("7.7071415537864938e-45", NULL);
printf( "%a %a %0.18g\n", f1, f2, f1);
return 0;
}
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21718