How to set / change the absolute value of variable?

Question

Everyone knows the abs() function to get the absolute value of variables. Is there a similar function or efficient way of setting it ignoring the sign?

Practical example: Force minimum value for motor speed control

if (abs(speed) < 50 && speed != 0)
{
    if (speed < 0)
        speed = -50;
    else
        speed = 50;
}

I am looking for a standardized function or a clever bit manipulation one-liner. Wrapping the above if into a function or macro is obvious.

`speed = (1 - 2*(speed < 0))*50` will do the inner bit. The expression `speed < 0` evaluates to `1` if `speed < 0` and `0` otherwise. — Peter, Apr 06 '18 at 11:00

phuclv · Accepted Answer · 2018-04-10T16:20:23.287

3

It's basically the signum function

speed = 50*sgn(speed);

There's a standard function for this

speed = copysign(50, speed);

however this isn't quite efficient for integer types.

The sgn() function can be implemented using various other ways that you can find here

speed = ((speed > 0) - (speed < 0))*50;

edited Apr 10 '18 at 16:20

answered Apr 10 '18 at 16:15

phuclv

37,963
15
156
475

More importantly, going via a floating-point type will lose precision from large enough integers. – Toby Speight Apr 10 '18 at 18:17
@TobySpeight a double-precision type can store exactly all integers up to 2^53, so it'll work well for an `int32_t` – phuclv Apr 11 '18 at 01:39

kabanus · Answer 2 · 2018-04-06T11:02:33.687

2

Combining with answers you can find around SO, for example Fast sign of integer in C:

speed=((speed > 0) - (speed < 0)) * theNewValue;

You basically just want to keep the sign, and multiply it by a new norm.

edited Apr 06 '18 at 11:02

answered Apr 06 '18 at 10:56

kabanus

24,623
6
41
74

Toby Speight · Answer 3 · 2018-04-11T17:37:14.240

You're looking for an integer equivalent of the standard copysign() function.

The simplest way to do that is simply to multiply the absolute value of the magnitude argument by the sign of the sign argument, using one of the standard methods to obtain that:

int setabs(int magnitude, int sign)
{
     return magnitude
        * ((magnitude > 0) - (magnitude < 0))
        * ((sign > 0) - (sign < 0));
}

Tests to demonstrate:

int main()
{
    return
        + (setabs(50, 1) != 50)
        + (setabs(50, 0) != 0)
        + (setabs(50, -1) != -50)
        + (setabs(-50, 1) != 50)
        + (setabs(-50, 0) != 0)
        + (setabs(-50, -1) != -50);
}

If multiplications are expensive for you, then we can use subtraction of two booleans to make a slightly cheaper version (passing the same set of tests):

int setabs(int magnitude, int sign)
{
    return magnitude
        * ((magnitude > 0 == sign >= 0) - (magnitude > 0 == sign <= 0));
}

Any bit-manipulation techniques will depend on which representation is used by your platform for negative integers (sign-magnitude, one's-complement, two's-complement, ...), and will require casting the arguments to unsigned types (or rely on undefined behaviour).

If you want 0 to be considered "positive" for the sign, modify the tests and swap `>` with `>=`. — Toby Speight, Apr 11 '18 at 20:23

score 0 · Answer 4 · answered Apr 10 '18 at 14:48

0

It depends on your platform. Since you haven't said what system you're targeting, I'm going to assume your integers use sign+magnitude representation. ;-)

int setabs(int sign, int magnitude)
{
    static const unsigned mask =- 0^0;
    assert(mask); // in case it's used on weird (two's complement) systems
    return (int)((unsigned)sign&mask | (unsigned)magnitude&~mask);
}

You can then write

speed = setabs(speed, 50);

answered Apr 10 '18 at 14:48

Toby Speight

27,591
48
66
103

You must have some coding error here. `assert(mask);` is like `assert(0);` Perhaps `- 0` was to be `~ 0`? Maybe use `mask = UINT_MAX - INT_MAX;` – chux - Reinstate Monica Apr 10 '18 at 17:59
I am not trying to say `-0` and `~0` are the same. Yet `unsigned mask =- 0^0;` is zero with 2's complement - the far most common. This leads to `assert(mask);` failing to assert and causing a compile time failure. – chux - Reinstate Monica Apr 10 '18 at 18:17
1

Yes, that's why we assert - to make sure this isn't run on a two's complement system, as written in the comment. – Toby Speight Apr 10 '18 at 18:18
Try running the code. - on a 2's complement machine. – chux - Reinstate Monica Apr 10 '18 at 18:18
Sleep well my friend, Review it another day. – chux - Reinstate Monica Apr 10 '18 at 18:42
Just had chance to try it on a 2's-complement machine, and it successfully aborts for me. It's only a run-time failure, but if we used C++, we could `static_assert` the same thing to get compile-time failure (which is obviously better). Or we could fake it with the array dimension trick that I vaguely remember seeing once. – Toby Speight Apr 10 '18 at 22:04
Note that C has `static_assert` since C11. – chux - Reinstate Monica Apr 11 '18 at 17:01

How to set / change the absolute value of variable?

4 Answers4