Stylish
Member
Hello,
Since I have updated my iPhone 4 to iOS 5 I have noticed my game is running slowly. Sometimes it run well but the execution becomes slow. It seems that the game is using too much resources but it runs well with iOS4!!
Somebody has noticed this?
Best regards
This sounds an awful lot like the issue we were having. SPPoolObject has a memory leak in iOS 5 which degrades performance and will eventually lead to an "Out of Memory" crash.
See this post for more information and a temporary work-around:
http://forum.sparrow-framework.org/topic/sppoolobject-leaking
dpizzle
Member
I had this same issue. Tweens seem to stutter in iOS5. I followed bsowers advice and put DISABLE_MEMORY_POOLING into Preprocessor Macros and changed typedef NSObject SPPoolObject; to #define SPPoolObject NSObject.
Things seem to be working now. Does this leak memory by doing this or have any other side effects? I can't run instruments with memory pooling disabled.
Would love an official word from Daniel on this and if he's working on a fix.
I tested on my 3GS after upgrading to iOS 5 and I got the same result, the game starts stuttering....and memory leaks are abundant from SPPoolObject (quickly reach into the MB range)
UPDATE: The above fix worked for me as well, but I am waiting to test on an iOS 4.2 device (the minimum that Pandemic supports) to see if it has any adverse effects.
I've done pretty extensive tests in Leaks, and from what I can tell, disabling memory pooling does not cause any leaks or have other adverse side effects (aside from the performance hit you'll take from not having object pooling, which is a necessary evil right now).
foxyfox
Member
I'm really a newbie in Obj-C.
Can someone please let me know where exactly I should put DISABLE_MEMORY_POOLING in SPPoolObject.h?
Below is the code at the top of SPPoolObject.h
=====================
#import <Foundation/Foundation.h>
@class SPPoolObject;
typedef struct
{
Class poolClass;
SPPoolObject *lastElement;
} SPPoolInfo;
#ifndef DISABLE_MEMORY_POOLING
====================
should I take out the #ifndef DISABLE_MEMORY_POOLING?
dpizzle
Member
Thanks bsowers, appreciate the response.
foxyfox, I put it in Preprocessor Macros and it worked for me.
Try this, click your project at the top, click build settings on the right and search for Preprocessor Macros. Under debug and release enter DISABLE_MEMORY_POOLING
The advantages of memory pooling are avoiding the cost of allocation. Objects are added to a pool instead of being deallocated, then when you need a new one they are simply taken from the pool of already allocated objects. Obviously, your memory footprint will be larger, but Sparrow is designed to clear this pool if memory warnings come.
PS All you need to do is write #define DISABLE_MEMORY_POOLING in SPPoolObject.h (under the #import line).
http://forum.sparrow-framework.org/topic/arc-and-ios-5
Sparrow 1.x will stay with iOS 4 and OpenGL 1.1 and Sparrow 2.0 will start off with iOS 5 and OpenGL ES 2.
It is purely a memory issue, and I am looking at it right now but it is very strange. 100% of the leak comes from the function NSAllocateObject, but...It doesn't seem like this is a real leak. It looks like Instruments is getting fooled by the lack of a call to [super dealloc] and thus an object residing in memory with zero references. The history of an SPPoolObject looks like this:
Malloc 1 +[SPPoolObject allocWithZone:]
Autorelease +[SPRectangle rectangleWithX:y:width:height:]
Release 0 -[NSAutoreleasePool release]
Autorelease +[SPRectangle rectangleWithX:y:width:height:]
Release 0 -[NSAutoreleasePool release]
etc, etc, etc which looks exactly like what it is supposed to do..
If that's the case though, then why the performance drop...I don't know...
I've had some downtime so I spent about 2 hours analyzing this problem. I stripped out all of the internal calls so the buck stops at allocWithZone....as in the following (instead of NSAllocateObject):
size_t size = class_getInstanceSize([self class]); if(size < 16) size = 16; id result = (id)malloc(size); bzero(result,size); if(result) result->isa = [self class]; return result;
In the SPRectangle methods, etc, I changed them to directly call allocWithZone:nil and the result was the same...same leaks. This is quite puzzling....I checked the assembly too and it is line per line the same instructions in both SDKs.
UPDATE: I am pretty sure I found the function responsible for the performance hits, but I don't know how it relates to the memory leaks (I'm convinced they are false positives). If I run the Time Profiler on Pandemic in simulator 4.3, all is well. If I run it in simulator 5 (or device 5) I notice a new function that wasn't present in 4.3. It is a C++ function from the libobjc.A.dylib library called (it is really long):
objc::DenseMap<objc_object*, unsigned long, true, objc::DenseMapInfo<objc_object*>, objc::DenseMapInfo<unsigned long> >::LookupBucketFor(objc_object*, const&, std::pair<objc_object*, unsigned long>*&) const
If I disable memory pooling, this function stays in about 3rd place in the low single digit percent of CPU time. If I enable memory pooling, it starts to skyrocket and gets up to 20% or more (over double the time spent on objc_msgsend).
FURTHER UPDATE: The source code for the function in question (Open Source component released by apple)
http://www.opensource.apple.com/source/objc4/objc4-493.9/runtime/llvm-DenseMap.h
It is called from the method objc_rootReleaseWasZero in this (new to iOS 5) file:
http://www.opensource.apple.com/source/objc4/objc4-493.9/runtime/objc-arr.mm
I think I was able to correct the performance issue (the leaks still appear but I no longer believe them, and the game no longer stutters and the above mentioned function is under control). I have absolutely no clue why this works, as in theory it should do absolutely nothing...but somehow it does (for me at least, I'd love to have you all try it).
Click on your Sparrow project on the left, and choose the "Sparrow" target from the middle pane. Go to "Build Phases" and expand "Compile Sources". Double click on SPPoolObject.m and type in "-fno-objc-arc" (no quotes) to add a compiler flag explicitly disabling ARC for that file (which should be disabled by default already....wtf). Then try and see how it works.
Oh I don't know if the compiler will accept that flag pre iOS 5 or not....
You must log in to post.
