error: invalid operands to binary expression ('float' and 'float') return (x & (1 << 31)) == 0

Question

Hi，I have a code like this:

STATIC bool is_pos_float(float x) {
  return (x & (1 << 31)) == 0;
}

But after compile, it shows:

error: invalid operands to binary expression ('float' and 'float')
  return (x & (1 << 31)) == 0;

What's the problem?

You need to get a bitwise copy of the floating point `x` as an integer. Floating point values makes no sense in bitwise operations. Also note that your code relies on implementation-specific details of the floating point format. Lastly, if you want to know if a floating point value is not negative, why not simply make that comparison? As in `x >= 0.0f`? — Some programmer dude, Feb 08 '18 at 07:50
You can't use floats in a bitwise operation would be my first guess. Error message seems pretty clear. — Mad Physicist, Feb 08 '18 at 07:52
Possible duplicate of [How to perform a bitwise operation on floating point numbers](https://stackoverflow.com/questions/1723575/how-to-perform-a-bitwise-operation-on-floating-point-numbers) — xskxzr, Feb 08 '18 at 08:04

Jive Dadson · Accepted Answer · 2018-02-08T08:53:47.123

2

The left operand of built-in operator& must be an integral type, not floating_point. Do this instead.

inline bool is_pos_float(float x) {
  return x > 0.0f;
}

Edit. Assuming that what the OP really wants is to muck around in the floating point format, I think this will work if the machine is Little Endian.

bool high_bit_zero(float x) {
    constexpr unsigned sz = sizeof(float);
    using raw = unsigned char[sz];
    raw *c = reinterpret_cast<raw*>(&x);
    return !((*c)[sz-1] & (1 << 7));
}

edited Feb 08 '18 at 08:53

answered Feb 08 '18 at 08:06

Jive Dadson

16,680
9
52
65

Fewest assumptions about representation here – Mad Physicist Feb 08 '18 at 08:15
Despite the function name, the question's code looks like it attempts to return true for `0.0f`. Your answer won't do that. May be worth mentioning in your answer. – Feb 08 '18 at 08:24
@hvd - The `high_bit_zero` function I added returns `false` for -0.0f and `true` for +0.0f. – Jive Dadson Feb 08 '18 at 08:52

Arash · Answer 2 · 2018-02-08T08:24:37.210

-1

What do you plan to do? Playing with the bits of a float variable???

If you are planning to make sure x is positive or zero, the solution is using !(x<0.0f).

Converting float to int causes neglecting small numbers between -1 and +1 which does not work too.

If you insist on doing something hacky, have a look at IEEE-754 standard:

#include <iostream>
using namespace std;

static bool is_posz_float(float x)
{
  static_assert(sizeof(float) == 4, "Unexpected type size!");
  union IEEE754{
    float number;
    unsigned char bytes[sizeof(float)];
  };
  IEEE754 a;
  a.number=x;
  return (a.bytes[sizeof(float)-1] & (1 << 7)) == 0;
}

void test(float x)
{
    if(is_posz_float(x))
        cout<<x<<" is a positive number or zero."<<endl;
    else
        cout<<x<<" is a negative number."<<endl;
}


int main() {
    test(0.1f);
    test(0.0f);
    test(3.14f);
    test(-0.11);
    return 0;
}

The results:

0.1 is a positive number or zero.
0 is a positive number or zero.
3.14 is a positive number or zero.
-0.11 is a negative number.

edited Feb 08 '18 at 08:24

answered Feb 08 '18 at 07:53

Arash

2,114
11
16

1

`int(x)` does something very different from the conversion you are implying. – Mad Physicist Feb 08 '18 at 07:56
@MadPhysicist, Thanks, have a look at my fix. – Arash Feb 08 '18 at 08:07
Assigning to one element of a union, and reading another is undefined behaviour. You either need memcpy into a char array, or `const auto c = `reinterpret_cast(&x);` and then use `c` as a pointer to the char array. – Martin Bonner supports Monica Feb 08 '18 at 08:17
1

Why did you suggest `!(x<0.0f)` instead of just using `>=0`? – Mad Physicist Feb 08 '18 at 08:17
Your test cases need to consider #+INF, #-INF, and #NAN. They are the tricky ones. – Martin Bonner supports Monica Feb 08 '18 at 08:18
@MadPhysicist, because the original code does something similar. Anyway, I am relaying on the optimizer. – Arash Feb 08 '18 at 08:18
@MartinBonner. Is the undefined behavior just a C++ thing? I thought it was the whole point of unions in C. – Mad Physicist Feb 08 '18 at 08:19
what's the point of the assert and the hard coded size when you can use `sizeof`? – Mad Physicist Feb 08 '18 at 08:20
@MartinBonner, Where is it stated that union has an undefined behavior? – Arash Feb 08 '18 at 08:21
@MadPhysicist, It reminds you that this computer is not standard. IEEE754 might not apply. – Arash Feb 08 '18 at 08:21
I use [n4296](http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4296.pdf) which is the pre-C++14 standard. All other standards have similar wording. Section 9.5 [class.union] says "In a union, at most one of the non-static data members can be active at any time, that is, the value of at most one of the non-static data members can be stored in a union at any time." – Martin Bonner supports Monica Feb 08 '18 at 09:37

error: invalid operands to binary expression ('float' and 'float') return (x & (1 << 31)) == 0

2 Answers2