Why can't the floating point types compare by using memcmp() function?

Question

bool floatcmp(const float a, const float b)
{
    const void *p = (void*)&a;
    const void *q = (void*)&b;
    if (memcmp(p, q, sizeof(float)) == 0)
        return true;
    return false;
}

The example code is above, the man page says memcmp(x,y)==0 does not imply that x==y and floating point types often have a value called NaN (‘not a number’) with the property that NaN==NaN is false. But I change the type to void* and I think the compiler doesn't know a is a float number.

Answer 1

For IEEE 754 binary floating point there are two important cases where a bit by bit equality comparison will produce different results from a floating point equality comparison.

The first and probably most important is zero. IEEE 754 has representations for both positive zero and negative zero. These will compare equal in a floating point comparison but unequal in a bitwise comparison.

The second is NaNs, comparisions involving NaNs, even equality to another identical NaN are always false, but under bitwise comparison two identical NaNs will compare the same.