++i is faster (or) i++ is faster in a for loop???

[saiavinashiitr]

In a for loop like for(i=0;i<=10;i++) and for(i=0;i<=10;++i).... i have heard saying that writing ++i in the for loop is faster rather than writing i++.... is that true??... is there any explaination for that???

[Banfa]

The question is not "++i is faster (or) i++ is faster in a for loop" but "++i is or i++ faster", i.e. being in a for loop is irrelevant to the speed of the operation.

Taking the 2 operators then what they do is

1. ++i - increment the value of i and return the new value
2. i++ - increment the value of i and return the original value

In 1 having incremented the value of i you can then just use the value of i, however in i++ having incremented the value of i you have to return the original value so you have to deal with 2 values at the same time, the original and new value of i represented by the pseudo code

++i
{
    INCREMENT i
    RETURN i
}
 
i++
{
    result = i
    INCREMENT i
    RETURN result
}
 

You can see an extra variable is created and copied to so conceptually ++i is faster.

However we have yet to talk about type because this only really comes into play for classes that have defined operator++() and operator++(int). In the 2 for loops you give if i is an int (or any other POD type) then the compiler/optimiser is clever enough to see that you don't use the return value at all and optimise away the code that does the copy operation for the i++ implementation. However a classes implementation of the operators is always going to have to do a copy to correctly implement i++.

So the real answer is that ++i is faster than i++ for a class (that supports those operations) but whether it makes a difference to you run time speed is going to be rather dependent on the complexity of the class, e.g. for a simple class holding a single int you will see little difference between ++i and i++ for a complex class a lot of data (or component classes) you might see some difference, especially in a for loop that is repeatedly doing the unrequired copies.

As a matter of course I personally always use ++i in for loops for iterators.

[Luuk]

#include <stdio.h>
#include <sys/time.h>
 
int main() {
        struct timeval start, stop;
        int i;
        long long int u1, u2;
 
        gettimeofday(&start, NULL);
        for(i=0;i<=1000000;i++) { }
        gettimeofday(&stop, NULL);
        u1 = stop.tv_usec - start.tv_usec;
        //printf("%lld\n", stop.tv_usec - start.tv_usec);
 
        gettimeofday(&start, NULL);
        for(i=0;i<=1000000;++i) { }
        gettimeofday(&stop, NULL);
        u2 = stop.tv_usec - start.tv_usec;
        //printf("%lld\n", stop.tv_usec - start.tv_usec);
        //printf("\n");
        printf("%lld %lld %lld\n", u1, u2, u1 - u2);
}
 

luuk@opensuse:~/tmp> for f in {1..10}; do ./test1; done
4389 4168 221
3769 3775 -6
4357 4390 -33
4408 4386 22
4410 3794 616
3762 3849 -87
4572 4384 188
4412 4386 26
4139 3759 380
3770 4178 -408
luuk@opensuse:~/tmp>
 

so, b......t ;-)