18

I'm working on making a logical right shift function in C using only bitwise operators. Here's what I have:

int logical_right_shift(int x, int n)
{
    int size = sizeof(int); // size of int

    // arithmetic shifts to create logical shift, return 1 for true
    return (x >> n) & ~(((x >> (size << 3) - 1) << (size << 3) -1)) >> (n-1);
}

This actually works for all cases except if n = 0. I've been trying to figure out a way to fix it so it will work for n = 0 as well, but I'm stuck.

Ignacio Vazquez-Abrams
  • 776,304
  • 153
  • 1,341
  • 1,358
Dan
  • 181
  • 1
  • 1
  • 3

8 Answers8

37
int lsr(int x, int n)
{
  return (int)((unsigned int)x >> n);
}
Ignacio Vazquez-Abrams
  • 776,304
  • 153
  • 1,341
  • 1,358
  • 3
    This is the correct answer (modulo the unnecessary and ugly cast on the result). Right-shift of a negative value has implementation-defined behavior (or is it an implementation-defined value? I forget), so any implementation of your `lsr` function in terms of using the `>>` operator on a signed value is simply non-portable. – R.. GitHub STOP HELPING ICE Mar 09 '11 at 23:12
  • One issue with this solution: strictly speaking, it has implementation-defined behavior in the case of `n==0`, since casting from an `int` to `unsigned` and back results in implementation defined behavior if the original value is negative. The first conversion must happen modulo `UINT_MAX+1`, but the conversion back to signed `int` might simply be a reinterpretation of the representation, in which case the value would be changed. – R.. GitHub STOP HELPING ICE Mar 09 '11 at 23:14
  • @R.. Hi, can you give a example, or a snippet code, that can helps me to understand what you mean ? Thank you. – Jonguo Mar 03 '19 at 06:22
16

This is what you need:

int logical_right_shift(int x, int n)
{
    int size = sizeof(int) * 8; // usually sizeof(int) is 4 bytes (32 bits)
    return (x >> n) & ~(((0x1 << size) >> n) << 1);
}

Explain

x >> n shifts n bits right. However, if x is negative, the sign bit (left-most bit) will be copied to its right, for example:

Assume every int is 32 bits here, let
x     = -2147483648 (10000000 00000000 00000000 00000000), then
x >> 1 = -1073741824 (11000000 00000000 00000000 00000000)
x >> 2 = -536870912  (11100000 00000000 00000000 00000000)
and so on.

So we need to erase out those sign extra sign bits when n is negative.

Assume n = 5 here:

0x1 << size moves 1 to the left-most position:

(10000000 00000000 00000000 00000000)

((0x1 << size) >> n) << 1 copies 1 to its n-1 neighbors:

(11111000 00000000 00000000 00000000)

~((0x1 << size) >> n) << 1! reverses all bits:

(00000111 11111111 11111111 11111111)

so we finally obtain a mask to extract what really need from x >> n:

(x >> n) & ~(((0x1 << size) >> n) << 1)

the & operation does the trick.

And the total cost of this function is 6 operations.

mingyc
  • 526
  • 5
  • 13
  • 1
    I think this is correct. But can you do left shift by 32 times? That will be an undefined shift? – Timothy Leung Aug 17 '13 at 05:13
  • 3
    Yes, if `int` is, for example, 4 bytes, this fails. He needs to subtract 1 from `size` so that if `int` is 4 bytes, `size`is 31. – modulitos Jul 25 '14 at 02:29
6

Just store your int in an unsigned int, and perform >> upon it.

(The sign is not extended or preserved if you use unsigned int)

http://en.wikipedia.org/wiki/Logical_shift

Bernd Elkemann
  • 23,242
  • 4
  • 37
  • 66
3

I think problem is in your ">> (n-1)" part. If n is 0 then left part will be shift by -1. So,here is my solution

int logical_right_shift(int x, int n)
{
  int mask = ~(-1 << n) << (32 - n);
  return  ~mask & ( (x >> n) | mask); 
}
Levenson
  • 73
  • 5
  • It's an elegant solution as it involve no conditional statements, therefore no CPU branching. The final `| mask` operation is unnecessary, as every bit set to 1 by that OR operation will be set to 0 by the AND operation. The mask calculation could be simplified as follows: `int mask = int.MinValue >> (n - 1);` but my simplification will produce an incorrect result if n == 0, so it depends on your use case - this simplification would not be suitable for a library. – radfast Feb 05 '21 at 15:05
2

Derived from php's implementation of logical right shifting

function logical_right_shift( i , shift ) {

    if( i & 2147483648 ) {
        return ( i >> shift ) ^ ( 2147483648 >> ( shift - 1 ) );
    }

    return i >> shift;
}

For 32bit platforms only.

1

Milnex's answer is great and has an awesome explanation, but the implementation unfortunately fails due to the shift by total size. Here is a working version:

int logicalShift(int x, int n) {
  int totalBitsMinusOne = (sizeof(int) * 8) - 1; // usually sizeof(int) is 4 bytes (32 bits)
  return (x >> n) & ~(((0x1 << totalBitsMinusOne) >> n) << 1);
}

To have 1 as the most significant bit, and all zeroes elsewhere, we need to shift 0x1 by number of bits - 1. I am submitting my own answer because my edit to the accepted answer was somehow rejected.

modulitos
  • 14,737
  • 16
  • 67
  • 110
0

As with @Ignacio's comment, I don't know why you would want to do this (without just doing a cast to unsigned like in the other answers), but what about (assuming two's complement and binary, and that signed shifts are arithmetic):

(x >> n) + ((1 << (sizeof(int) * CHAR_BIT - n - 1)) << 1)

or:

(x >> n) ^ ((INT_MIN >> n) << 1)
Jeremiah Willcock
  • 30,161
  • 7
  • 76
  • 78
-1
int logicalShift(int x, int n) {
  int mask = x>>31<<31>>(n)<<1;
  return mask^(x>>n);
}

Only for 32 bits

Toch
  • 7
  • 1
  • 2
    Please add more information to explain your answer. Code-only answers are not well accepted here. –  Jan 22 '17 at 23:49
  • 1
    @MarkYisri Ironic since the top voted answer on this "thread" is code-only and perfectly, clearly, answers the question. – MD XF Sep 27 '17 at 01:46
  • This causes undefined behaviour if `x >> 31` results in a non-zero value – M.M Oct 12 '18 at 23:15