On Fri, Oct 24, 2008 at 09:37:41AM +0200, Abdelrazak Younes wrote:
> Nope, I would just test the difference to be less than 'epsilon'
> instead. And 'epsilon = 1e-307' for a double IIRC.
You don't recall correctly. That is the smallest number representable
with a double. The precision epsilon is the smallest number such that
1 + epsilon != 1. The attached C program will tell you what epsilon is.
> And FYI a double is a
> double is a 64bit on any platform ;-)
Yep.
--
Enrico
#include <stdio.h>
int main(void)
{
double u;
double eps;
double unit = 1.0;
do {
eps = unit;
unit = 0.5 * unit;
u = 1.0 - unit;
} while (u < 1.0);
printf("eps = %9.2e\n", 2.0 * eps);
return 0;
}
