Tuesday, November 15, 2011

Compute the minimum or maximum of two integers without branching / operators

Source


On some rare machines where branching is expensive, the below obvious approach to find minimum can be slow as it uses branching.
/* The obvious approach to find minimum (involves branching) */
int min(int x, int y)
{
  return (x < y) ? x : y
}
Below are the methods to get minimum(or maximum) without using branching. Typically, the obvious approach is best, though.
Method 1(Use XOR and comparison operator)
Minimum of x and y will be
y ^ ((x ^ y) & -(x < y))
It works because if x < y, then -(x < y) will be all ones, so r = y ^ (x ^ y) & ~0 = y ^ x ^ y = x. Otherwise, if x >= y, then -(x < y) will be all zeros, so r = y ^ ((x ^ y) & 0) = y. On some machines, evaluating (x < y) as 0 or 1 requires a branch instruction, so there may be no advantage.
To find the maximum, use
x ^ ((x ^ y) & -(x < y));
#include<stdio.h>
 
/*Function to find minimum of x and y*/
int min(int x, int y)
{
  return y ^ ((x ^ y) & -(x < y));
}
 
/*Function to find maximum of x and y*/
int max(int x, int y)
{
  return x ^ ((x ^ y) & -(x < y));
}
 
/* Driver program to test above functions */
int main()
{
  int x = 15;
  int y = 6;
  printf("Minimum of %d and %d is ", x, y);
  printf("%d", min(x, y));
  printf("\nMaximum of %d and %d is ", x, y);
  printf("%d", max(x, y));
  getchar();
}
Method 2(Use subtraction and shift)
If we know that
INT_MIN <= (x - y) <= INT_MAX
, then we can use the following, which are faster because (x - y) only needs to be evaluated once.
Minimum of x and y will be
y + ((x - y) & ((x - y) >>(sizeof(int) * CHAR_BIT - 1)))
This method shifts the subtraction of x and y by 31 (if size of integer is 32). If (x-y) is smaller than 0, then (x -y)>>31 will be 1. If (x-y) is greater than or equal to 0, then (x -y)>>31 will be 0.
So if x >= y, we get minimum as y + (x-y)&0 which is y.
If x < y, we get minimum as y + (x-y)&1 which is x.
Similarly, to find the maximum use
x - ((x - y) & ((x - y) >> (sizeof(int) * CHAR_BIT - 1)))
#include<stdio.h>
#define CHAR_BIT 8
 
/*Function to find minimum of x and y*/
int min(int x, int y)
{
  return  y + ((x - y) & ((x - y) >>
            (sizeof(int) * CHAR_BIT - 1)));
}
 
/*Function to find maximum of x and y*/
int max(int x, int y)
{
  return x - ((x - y) & ((x - y) >>
            (sizeof(int) * CHAR_BIT - 1)));
}
 
/* Driver program to test above functions */
int main()
{
  int x = 15;
  int y = 6;
  printf("Minimum of %d and %d is ", x, y);
  printf("%d", min(x, y));
  printf("\nMaximum of %d and %d is ", x, y);
  printf("%d", max(x, y));
  getchar();
}
Note that the 1989 ANSI C specification doesn't specify the result of signed right-shift, so above method is not portable. If exceptions are thrown on overflows, then the values of x and y should be unsigned or cast to unsigned for the subtractions to avoid unnecessarily throwing an exception, however the right-shift needs a signed operand to produce all one bits when negative, so cast to signed there.