RC2: Performance issues

[UrbanMonk]

It's beta 21 and I'm still experiencing performance issues on the iPhone 3GS

On the iPhone 3GS there are random skips in the game where the objects appear to move about 30 cycles ahead, almost like the app stops rendering the screen for a second and all the objects skip.

Every now and then all the objects on the screen will be destroyed just leaving the counter and a few other GUI objects.

The frame rate fluctuates between a full 60 down to about 2 every 2-3 minutes.

All of these symptoms started when I resized the app to 960x640 so I could support retina and the iPad resolutions.


I've already sent a link to my mfa to Andos I believe, but I have yet to receive a response, so I'm hoping someone can either fix this problem, or perhaps give me an idea as to where the issue is coming from.


Here's some info about the app.

-No scrolling
-only a few alpha channels (5-6)
-No more than 75 objects at a time on screen
-No String objects
-2 counters
-No extensions used
-Music plays in the background (when it's removed there is no change in performance)
-Lot's of sounds (when they're removed there is no change in performance)
-One loop to load the levels from a global string, but then no loops during gameplay. (Removing this does not change performance either)
-App is set to run at 60 fps
-No background objects at all
-Largest object is 240x90

So basically this app should be running full speed with no issues, so this is really sad.

The only thing I can think of is that I'm compiling it wrong or something, so here are my settings at compile time.

-Device
-Release
-Optimized (armv7)
-Color reduction
-Multitask (enabled)
-Landscape (left)
-iAd's on, bottom of screen

If any of the developers want the latest mfa let me know.

Hope this can be resolved, because if the exporter can't handle this I won't be able to use it.

Thanks
Jacob

[Anders]

I have tested your app and profiled it quite a bit and it is very difficult to pinpoint why there are slowdowns. I cannot reproduce the skipping though.

I found out that only certain levels seem to cause the slowdown from 60 to 40 FPS on my device and it goes back to 60 FPS once the next level is loaded. The weird part is that there is absolutely no frame change involved - only object creation/destruction. It is very difficult to see which object start causing the slowdown and what happens that resolves it. It doesn't seem to be a CPU or GPU wise performance problem either.

[UrbanMonk]

The performance issue isn't on a per-level basis for me. It happens even on the same level.

I just bought an iPhone 4 yesterday and I'm happy to report that there are no slowdowns, but the object skipping bug is still there.

I think it's a problem with the bouncy ball movement actually because every now and then it'll travel through other objects without performing a collision, and jump to the other side of the screen.

What could possibly be causing this?


Also I should add that the game works perfectly on JavaME phones once I drop the fps down to 30, and lower the resolution.

[BHGames]

I see a birthday cake beside your nickname, so Happy birthday!

Maybe the tunneling effect is caused on the bouncy ball movement. There was a bug a few betas ago for path movement where the movement wasn't taking into account the elapsed time, so perhaps something similar could be happening here (?)

[Anders]

Happy birthday UrbanMonk

No BHGames, the actual framerate drops.
I profiled the runtime a lot today and implemented some tweaks that should prevent the graphics card doing unnecessary work during the draw-cycle. Hopefully it will help. I also have another tweak in the workings that should make retina games work even better on 3GS devices.

[BHGames]

> No BHGames, the actual framerate drops.

Then, do you guys have any "catch-up" code?

[UrbanMonk]

Thanks! I'm having a good birthday so far. No work!


Actually I think BHGames might be right about that timing bit since that might explain the jumping ball issue.

On my iPhone 4 there is no framerate issues, but now the ball is randomly jumping like the runtime skips a few cycles every now and then.

I was thinking about just going back to the lower resolution since it didn't have any problems then. Should I? Or will this be fixed?

[Bruto]

yay happy birthday urbanmonk!

[netninja]

I actually have a very similar setup to the game that UrbanMonk described, except that I don't have iAds and my game runs in portrait. Besides that, it sounds very similar in terms of the inner workings. My game natively runs at 320x480.

I also experience random framerate drops after the game has been running mostly consistently (60fps on an iPhone 4).

Maybe there is something else inside our code which is similar? Do you have any activating or deactivating of groups? I use the list object in my game too if that makes a difference.

I have bouncy ball movement for some objects in my game which also appear to skip, however I changed to that movement from "per pixel" movement (i.e. Always set x position to x position + 2), which in my tests was 30% slower than bouncy ball movement, in terms of how many objects of that type I could have onscreen before slow down.

(Btw, Happy Birthday UrbanMonk!)

[AndyH]

I've been testing on a 2nd and 4th gen iTouch and notice frame drops from time to time. The 2nd gen, with lots going on tends to drop to around 30fps at the worst case but normally hovers around 40 to 50 on average while the 4th gen sometimes dips below 50 but is mostly stable at 60. I don't see any specific reason for the drops in frame rate in that the conditions when it occurs do not seem consistent with the number of graphics (active or background) on screen. It's hard to tell how events are affecting it either (I'm not using fastloops).

Interesting to hear about your experience with using actions to change the x and y position of objects as opposed to built in movements. I would comment that by default the built in movements would use timer based movements at 50 fps unless you have changed those settings but I can understand that using actions on many objects may not be as efficient as letting MMF handle movements in a more native way. BTW, I counter the frame rate changes by using a couple of global values to keep track of the milliseconds (timer) difference between frames and use this with the frame rate MMF reports to calculate a multiplying factor that I then apply to all my action based movements. It is working out really well for me and the apparant speed of the 2nd gen keeps up with the 4th gen at all times. I must say I'm impressed in the iTouch's perform in this regard as the perception of smoothness in the animation/movement is not really noticable.