Chipmunk speed questions « cocos2d for iPhone

cocos2d for iPhone » Programming » physics

Chipmunk speed questions

(9 posts) (3 voices)

Started 2 years ago by penguins
Latest reply from penguins

Tags:

penguins
Member
Hello,
In the process of making my first game, I noticed that chipmunk runs really slow on the device. My game is separated into levels, The levels vary, but max I have around 100 cpSegments. And around 14 cpBodies as objects, and then probably another 20 as bullets.
Is this too much for the iphone to handle??? (I have chipmunk stepping every-step.)

Another thing that'ss been driving me crazy about chipmunk is that my player, a circle made like so:
body = cpBodyNew(10.0, cpMomentForCircle(10.0, 29, 29, cpvzero));
Is moving through the boxes. But not as if there is nothing there. If it hits from above, it will work. But if it rams from a side, it starts to go through, then it like jumps. Its really odd. (Mind non of this happens on the simulator.)

Boxes are made like
body = cpBodyNew(10.0, cpMomentForPoly(10.0, num, verts, cpvzero));

Info:

Using cocos 8.2
Using chipmunk trunk pre 5.0 release.

I have been trying to figure this stuff out myself, and its just making me crazier and crazier.

My init Code :
```
space = cpSpaceNew();
	cpSpaceResizeActiveHash(space, 32, 120);
	cpSpaceResizeStaticHash(space, 48, 600);
	space->gravity = cpv(0,-200);
	space->iterations = 5;
```
Another weird problem I am having is adding a joint kills my fps. Like cuts it in half.
This is for dragging and dropping objects.
```
mouseMotor = cpDampedRotarySpringNew(mouseBody, draggedBody, draggedBody->a, 0, 100000);

			mouseJoint = cpPivotJointNew2(mouseBody, draggedBody, cpvzero, cpBodyWorld2Local(draggedBody, globalLocation));

			mouseJoint->maxForce = 50000.0f;
			mouseJoint->biasCoef = 0.15f;
			cpSpaceAddConstraint([g space], mouseMotor);
			cpSpaceAddConstraint([g space], mouseJoint);
```
Thanks for the help!!!!!!!!
(sorry this was so long)
Posted 2 years ago #
jordo
Member

hmmm....

My game uses chipmunk. I've got very large levels, and the total number of static cpSegments in most levels are anywhere from 2000-3000, with around 50 or so dynamic objects... (probably only 5-7 colliding at the same time however)...

Currently I've optimized the game to consistantly be taking around 30% CPU on a iphone 3G. After sharking it, physics simulation isn't even the main CPU hog. I'm pushing around 25 512x512 textures, and I'm guessing 100 or so sprites...

Posted 2 years ago #
zombie
Member

Calling cpspacestep each frame once is properbly one reason for your problems on the device. With lower framerates at 30 or even lower you call cpspacestep not often enough.

Most CPU time is chipmunk using for resolving collisions. So keep an eye on that.

Hope this is if any help,
zombie

Posted 2 years ago #
penguins
Member

Wow, Jordo, I am jeallous. I am using literally all of the unused system cpu.
I just tried building for release, and that helped some(a good amount). I am no longer crashing through stuff.

Zombie: What do you mean by not calling it every frame? Or calling it more than once? Wouldn't more than once hurt my performance? Just tried it, and it chopped my fps down to 9.

I have lots of collision callbacks to, for bullets and jump mechanics. Will those kill speeds?

My step loops consist of:
An iteration of all the objects in the scene calling there step commands. (Sadly, the objects are in a NSSet. Is this a huge hit in speed?);
for(id object in objects) {

Also, I tried using the performance tools, and I am not sure really what I am looking for/couldn't find anything using alot of cpu. There was libsystem.b.dylib that has using almost all cpu.

Thanks for the help!!

Posted 2 years ago #
jordo
Member

libSystem.B.dylib all the CPU? What about inside that? Change your shark sampling profile to 'Time Profile'..

When I shark my app, at a 1ms sampling interval I don't get any one symbol over 5-8%... Everything else is spread out from there...

libSystem.B.dylib is taking 3% CPU on mine, and inside there most of the time is spent in 'PushModelViewMatrix'...

Posted 2 years ago #
penguins
Member

Ok Here is what I got. I am extremely new to any sorta performance stuff, but I managed to figure out shark.

Here is what I got.
The symbol cpv had 12.5%, with in that, there was cpvmult, and cpvadd taking up the top 2 spots.

Then below that, cpArbiterApplyImpulse at 9.6%. This would make sense, because I am applying an impulse every accelerometer step. (This is probably bad right?) Inside that, there is cpSpaceStep, then all my methods that get stepped.

The next symbol, cpvadd is 5.8%
then cpvmult at 5.7%
cpvcross 2.5%
cpvperp 2.2%
cpBodyApplyImpulse 2.1%
cpBodyApplyBiasImpulse 2.1%
cpvdot 1.9%
cpvsub 1.9%
unKnown library and symbol at 1.8% (Should I care what that is???)
relative_velocity 1.6%
then finally a non physics, objc_msgSend at 1.4%
then more physics stuff.

All and all, my app is using 69.5% of my cpu.

@libSystem.B.dylib, your right. Its only taking up 1.3% (I was using, run with performance tools, cpu Sampler.)

So what does all this mean? That there is too much stuff that needs chipmunk?
So it probably isn't that nsset thing?

Thanks again for all your help.

Posted 2 years ago #
penguins
Member

Just did another quick test to verify my fears, its all cpSpaceStep(space, dt) fault. I turned it off, I can ran my most intensive level at a perfect 60 FPS.

I also found another cheap trick off of the chipmunk forums. I stuck all my bullets into the same shape group, eliminating the need for massive collision checks between bullets.

Here is the post if anyone cares.
http://www.slembcke.net/forums/viewtopic.php?f=1&t=649&p=2866&hilit=optimizing#p2866

I also found this post
http://www.slembcke.net/forums/viewtopic.php?f=1&t=679&p=3027&hilit=optimize#p3027
Which helps a lot. To summarize, adding -ffast-math and -O3 to cFlags.

That had a major boost in speed.

Thanks for the help all, and sorry for the inconvenience when a lot of answers were just over at another forum (but I did learn a lot from you too).

Posted 2 years ago #
zombie
Member

Hi,

I wanted to point out that it is often a bad idea to call cpSpaceStep exactly once per frame. Doing this will give you unrealistic physics the lower the framerates. For us calling cpSpaceStep less than every 1/60 second is giving us bad physics. I also recommend to call cpSpaceStep with a fixed timestep. Depending on your application, this can improve your performance.

Also try to eliminate all unnecessary collisions (as you have already done with putting things in groups). You can also use layers to fine-tune this even more.

- zombie

Posted 2 years ago #
penguins
Member

Thanks for the tip zombie.
I was actually thinking about using a fixed time step for cpSpaceStep, and you have inspired me to do so. I found the post discussing it here.

Not sure when I could use layers in my game, but i didn't even know they existed. I will keep in mind for my next game.
I have no time now, but I will get back to you when I try the fixed time step.

Thanks

Posted 2 years ago #

RSS feed for this topic

Reply

You must log in to post.