private function i_hitTestCircle2(clip1:Sprite, clip2:Sprite):Boolean
{
	var dx:Number = clip2.x - clip1.x;
	var dy:Number = clip2.y  - clip1.y ;

	var radii:Number = ((clip1.width * .5) + (clip2.width * .5));
	return ((dx * dx) + (dy * dy) < radii * radii);
}

hitTest (50000 iterations)
Player version: MAC 10,1,51,28 (regular)
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
hitTest                                                      54     0.00
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
hitTestCircle (50000 iterations)
Player version: MAC 10,1,51,28 (regular)
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
hitTestCircle                                               133     0.00
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
hitTestCircle2 (50000 iterations)
Player version: MAC 10,1,51,28 (regular)
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
hitTestCircle2                                               56     0.00
–––––––––––––––––––––––––––––––––––––––––––––––

Sure, Box2D actually does the right thing w.r.t collision (and PushButton is integrated with it out of box). Even if you don’t care about the physical response part, the algorithms that Box2D uses to _detect_ collisions are smart and would get you the performance improvement I described.

Before we go 1.0 we are going to do a major code style update to match Flex style (http://code.google.com/p/pushbuttonengine/issues/detail?id=56) as well as reorganize the codebase to make it easier to work with… Our bad. :)

That is a good point. Here is the result with clips that are about 3 times larger:

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
checkBitmapDataHitInternal (10000 iterations)
Uses BitmapData.hitTest to check for collision. BitmapData not cached.
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
[function]                                                 8087     0.81
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
checkBitmapDataHitDynamic (10000 iterations)
Uses BitmapData.hitTest to check for collision. BitmapData not cached. M
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
[function]                                                 1848     0.18
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Note, that is with cacheAsBitmap set to true.

mike chambers

mesh@adobe.com

You are exactly correct. Setting cacheAsBitmap = true on the author time clips brings the performance to the same level as the dynamic clips.

Here is the relevant section:

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
checkBitmapDataHitDynamic (10000 iterations)
Uses BitmapData.hitTest to check for collision. BitmapData not cached. M
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
[function]                                                 1844     0.18
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
checkBitmapDataHitInternal - cacheAsBitmap (10000 iterations)
Uses BitmapData.hitTest to check for collision. BitmapData not cached.
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––
method...................................................ttl ms...avg ms
[function]                                                 1805     0.18
––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Ill update the post to reflect this. Good catch.

mike chambers

mesh@adobe.com

Thats not completely true because speed of BitmapData.hitTest is largely depending on the size of objects. For objects with size of only couple of pixels, its lightning-fast. But try objects of hundreds or thousands of pixels and it gets so slow that it cant even finish this test.

Other collision techniques, like math-based or checking bounds, are not affected by the size.

So you cant just compare these results directly.

Feel free to rewrite the circle collision and Ill add it to the test.

Thanks…

mike chambers

mesh@adobe.com

Yeah, that should be faster, and is easy. Ill add another test using that.

@SP

The goal of this wasnt to super optimize everything, and find the absolute fastest way to do that (since that will largely depend on the application / content / game).

The goal was to give a baseline of relative performance between different high level APIs and techniques

mike chambers

mesh@adobe.com

There’s a LOT of room for optimizations in that collision code though.
Tonypa’s circle-circle collision code could be around 10 times faster (maybe more?).
It could be improved even more by replacing the
var r:Number = clip1.width / 2 + clip2.width / 2;
by
var r:Number = (clip1.width + clip2.width) * 0.5;

If instead of using DiplayObject.x, y, width or height you create a custom circle class with those properties it would improve a LOT more. DiplayObject.x and DiplayObject.y are slow, but DiplayObject.width and DiplayObject.height are EXTREMELY slow (seriously, ridiculously slow).
And if you really need speed, you can inline the code eliminating those 10000 function calls.

The same applies for many of the other checks!

Make sure you are not using the debug player in the tests! It gives awful results.

SP.