Are two doubles of equal value guaranteed to have the same bit pattern?

Question

在C中，是否保证两个值相等的双精度数（ double1 == double2 ）将具有相同的位模式？

Answer 1

There is no such guarantee.

For example, in IEEE floating point format , there exists the concept of negative 0. It compares equal to positive 0 but has a different representation.

Here's an example of that:

#include <stdio.h>

int main()
{
    unsigned long long *px, *py;
    double x = 0.0, y = -0.0;
    px = (unsigned long long *)&x;
    py = (unsigned long long *)&y;

    printf("sizeof(double)=%zu\n",sizeof(double));
    printf("sizeof(unsigned long long)=%zu\n",sizeof(unsigned long long));

    printf("x=%f,y=%f,equal=%d\n",x,y,(x==y));
    printf("x=%016llx,y=%016llx\n",*px,*py);

    return 0;
}

Output:

sizeof(double)=8
sizeof(unsigned long long)=8
x=0.000000,y=-0.000000,equal=1
x=0000000000000000,y=8000000000000000

EDIT:

Here's a revised example that doesn't rely on type punning:

#include <stdio.h>

void print_bytes(char *name, void *p, size_t size)
{
    size_t i;
    unsigned char *pdata = p;
    printf("%s =", name);
    for (i=0; i<size; i++) {
        printf(" %02x", pdata[i]);
    }
    printf("\n");
}

int main()
{
    double x = 0.0, y = -0.0;

    printf("x=%f,y=%f,equal=%d\n",x,y,(x==y));
    print_bytes("x", &x, sizeof(x));
    print_bytes("y", &y, sizeof(y));

    return 0;
}

Output:

x=0.000000,y=-0.000000,equal=1
x = 00 00 00 00 00 00 00 00
y = 00 00 00 00 00 00 00 80

You can see here the difference in representation between the two. One has the sign bit set while the other doesn't.

Answer 2

Yes, with the exception of zero. There is only one way to represent a particular IEEE float or double value. If there were two ways to represent the same value then that would be wasteful and the IEEE formats are very efficient.

Zero is a special case because there are positive and negative values of zero and they will compare equal, and they are equal for most purposes. There are a few rare cases (1/0) where they will give different results - positive versus negative infinity.

The previous answer focused unnecessarily on the zero case and failed, as far as I can tell, to answer the case for the other 4 billion floats and 16 billion billion doubles.