Comments on: XOR Swap

By: jackson

jackson — Sat, 23 Oct 2010 00:21:09 +0000

I always use the latest release player. At the time, that was Flash Player 10.0. Since I never test on a debug player, I didn’t know that XOR is faster there. Thanks for the explanation.

By: skyboy

skyboy — Fri, 22 Oct 2010 23:58:01 +0000

Were you using an old version of the debug player? Those results look far too far apart, but in both tests XOR swap is faster in the debug player, which is probably where the people who use it got idea that three XOR and three assignments is faster than only three assignments.

By: Bob

Bob — Tue, 22 Sep 2009 22:00:04 +0000

There are some situations where the XOR swap is handy albeit not necessarily in actionscript. One particularly useful situation is in embedded systems where you do not have any room on the stack or heap to allocate a swap variable, so you are forced to do things the hard way.

On some architecture, xor swap is a lot faster than a temp var, but this is usually truer in the lower level systems (assembler etc).

By: jackson

jackson — Tue, 22 Sep 2009 16:36:49 +0000

Wow, those are some really surprising results! To test, I tried it out in the standalone players on the same Windows box as in the article and got this:

Player	Assign Swap Time	Assign Swap Time (instancing)	XOR Swap Time
Release	237	269	280
Debug	11308	11350	11141

Those debug times look kind of crazy to me, but they are repeatable. The release times are pretty close to what I get with the browser plugin though. I wonder what accounts for our difference, especially since we have such similar machines. Thanks for your comment and your result times.

By: grapefrukt

grapefrukt — Tue, 22 Sep 2009 08:28:19 +0000

Strangely I get these times in the standalone (release) player:
Assign swap (including instancing) time: 414
Assign swap time: 320
XOR swap time: 357

What really surprised me, even though I knew that the debug player is a bit slower, is that if i run it there (still the code compiled in release) I get these results:
Assign swap (including instancing) time: 890
Assign swap time: 790
XOR swap time: 625

That’s almost twice the execution time. I really need to set up my dev enviroment to launch both the debug and standalone players.

All this was run on my 2.2 Ghz Intel Core 2 Duo with 3.5GB RAM running Windows XP SP3

By: jackson

jackson — Mon, 21 Sep 2009 16:28:58 +0000

Good point. My reason for leaving out that case was to make sure I was testing just the swap algorithm and not the variable instantiation. But since you expressed interest and I never did the test, I figured I might as well:

beforeTime = getTimer();
for (i = 0; i < SWAP_REPS; ++i)
{
    var temp2:int = a;
    a = b;
    b = temp;
}
logger.appendText("Assign swap time (new temp): " + (getTimer()-beforeTime) + "\n");

On the same Windows machine, I’m getting nearly identical times to the other version with a temporary variable. I’ve run it a couple dozen times and the difference is +/- 2 ms.

Thanks for the suggestion!

By: grapefrukt

grapefrukt — Mon, 21 Sep 2009 15:16:09 +0000

I think you should do a version with the temporary var instantiated inside the loop, i’m not able to test whether it would make any difference, but it should atleast make the comparison a bit more fair since that’s the normal use case.

But then again, since you’re doing a hundred million iterations normal might be a misnomer ;)