1

Problem in a nutshell

I have been building an IOS application in recent weeks and have run into some trouble.The application is plays an animation by manipulating and then drawing an image raster multiple times per second. The image is drawn by assigning it to a UIViews CALayer like so self.layer.contents = (id)pimage.CGImage; The calculation and rendering are seperated in two CADisplayLinks.

This animation technique achieves a satisfactory performance on the IPhone 6.1 simulator but when it is build on the physical device (Iphone 4s running IOS 6.1.3) it experiences a significant slow down. The slow down is so bad that it actually makes the application unusable.

Suspected Issues

I have read, in this question Difference of memory organization between iOS device and iPhone simulator , that the simulator is allowed to use far more memory than the actual device. However, while observing my apps memory usage in in "instruments", I noticed that the total memory usage never exceeds 3Mbs. So Im unsure if that is actually the problem but it's probably worth pointing out.

According to this question, Does the iOS-Simulator use multiple cores? , the IOS simulator runs of an intel chip while actual my device uses an apple A5 chip. I suspect that this may also be the cause of the slowdown.

I am considering rewriting the animation in Open GL, however Id first like to try and improve the existing code before I take any drastic steps.

Any help in identifying what the problem is would be greatly appreciated.

Update

Thanks to all those who offered suggestions. I discovered while profiling that the main bottleneck was actually clearing the image raster for the next animation. I decided to rewrite the rendering of the animations in opengl. It didn't take as long as anticipated. The app now achieves a pretty good level of performance and is a little bit simpler.

Community
  • 1
  • 1
jonahhooper
  • 11
  • 3
  • Have you run it in instruments on beth device and simulator? I have found that calculations run much much faster on the simulator than device, but openGL style things run faster on the device. Using the timing instrument see what function is taking all the processing time and focus on that first (or post it here so we can see why it's slow). – Putz1103 Jun 05 '13 at 18:40
  • 1
    The simulator also uses the full potential of your Mac's CPU while on a real iOS device you have a much less powerful CPU. You have officially discovered why you must test apps on a real device. The simulator is fine for basic logic testing but not for performance testing. – rmaddy Jun 05 '13 at 18:42

2 Answers2

2

This is a classic problem. The simulator is using the resource of your high-powered workstation/laptop.

Unfortunately the only solutions is to go back and optimize your code, especially the display stuff.

Typically, you want to try to minimize the drawing time from the computation time, which it sounds like you are doing, but make sure you don't compute on the main thread.

dispatch_queue_t queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_HIGH, 0ul);
dispatch_async(queue, ^{
    // Do the computation
});

You can use instruments while running on the device, so the CoreGraphics instruments is available to see what is using all the time and point to the offending code. Unfortunately, you probably already know what it is and it's just going to come down to optimizations.

mprivat
  • 21,582
  • 4
  • 54
  • 64
1

The slowdown is most likely related to blitting the images. I assume you are using a series of still images that get changed in the display look callback. I believe that if you can use CALayers that get added to your primary view/layer (while removing the old one), and which contain already CGImageRefs, you can then use CGContextDrawImage() to blit the image in the layer's drawInContext method. Set the context to use copy not blend, so it just replaces the old bits.

You can use a dispatch queue to create CALayer subclasses containing an image on a secondary thread, then of course the drawing happens on the main queue. You can use some throttling to maintain a queue of CALayers of 10 or so, and replenishing them as they are consumed.

if this doesn't do it then OpenGL may help, but again none of this helps moving bits between the processor and the GPU (since you are using stacks of images, not just animating one).

David H
  • 40,852
  • 12
  • 92
  • 138