Does rearranging a conditional evaluation speed up a loop?

Asked 2009-04-09T13:13:58.837

Bit of a weird one: I was told a while ago by a friend that rearranging this example for loop from :

for(int i = 0; i < constant; ++i) {
    // code...
}

to:

for(int i = 0; constant > i; ++i) {
    // code...
}

would slightly increase performance in C++. I don't see how comparing a constant value to a variable is faster than vice-versa, and some rudimentary tests I ran didn't show any difference in speed between the two implementations. The same was also true of testing this Python while loop:

while i < constant:
    # code...
    i += 1

vs:

while constant > i:
    # code...
    i += 1

Am I wrong? Are my simple tests not enough to determine the speed variation? Is this true of other languages? Or is this just a new best practice?

Edit Report

4 Answers

My suspicion is your friend is 100% wrong. But I wouldn't trust my opinion anymore than I would trust your friend. In fact, if there is a performance problem there is only one person you should trust.

The Profiler

This is only way you can ever claim with any authority that one way is or is not faster than another.

answered 2009-04-09T13:16:52.293

Assuming short-circuit evaluation, the only time this should make much of a difference is if you have a call to a slow function in your loop. For example, if you had a function that queried a value from your database, and returned it, then this:

while(bContinue && QueryStatusFromDatabase==1){
}  //while

Would be much faster than:

while(QueryStatusFromDatabase==1 && bContinue){
}  //while

Even though they are logically identical.

That's because the first one can stop as soon as a simple boolean is FALSE - the query only has to run when the boolean is TRUE, but the second one will always run the query.

Unless you have a need to squeeze every possible CPU cycle out of your loop, then those extreme cases are probably the only ones worth spending your time on. Think of it this way: To make up the time you spent asking this question would probably take several billion iterations of your loop.

Worst of all is when you have a function as a condition, and that function has side-effects that are secretly expected by some other place in the code. So when you make your little optimization, the side effects only happen some of the time, and your code breaks in weird ways. But that's a bit of a tangent. The short answer to your question is "Sometimes, but it usually doesn't matter."

answered 2009-04-09T13:28:15.667

Comparing against 0 is very fast, so this would actually be slightly faster:

for (int i = constant; i > 0; --i)
{ 
  //yo
}

I think it is better to use != in any case, as it makes off by one errors easier to detect and is the only way to use iterators with non-contiguous data structures, like linked lists.

answered 2009-08-03T06:15:04.397

-1

This is absolutely a case of micro-optimization and really doesn't need to be done.

It is true that (especially) in C++ there is a small performance difference between a post-increment operation and a pre-increment operation but that difference in today's compilers is generally negligible. The reason for changing the order of the conditional is due to the change from post- to pre-increment.

answered 2009-04-09T13:18:47.407

Does rearranging a conditional evaluation speed up a loop?

4 Answers

The Profiler

Your Answer