今天面试遇到同学说做过内存优化,于是我一般都会问那 Bitmap 的像素内存存在哪?大多数同学都回答在 JAVA heap 里面,就比较尴尬,理论上你做内存优化,如果连图片这个内存大户内存存在哪都不清楚,实在不太能说得过去。
Bitmap可以说是Android/ target=_blank class=infotextkey>安卓里面最常见的内存消耗大户了,我们开发过程中遇到的oom问题很多都是由它引发的。谷歌官方也一直在迭代它的像素内存管理策略。从 Android 2.3.3以前的分配在native上,到2.3-7.1之间的分配在java堆上,又到8.0之后的回到native上。几度变迁,它的回收方法也在跟着变化。
2.3.3以前Bitmap的像素内存是分配在natvie上,而且不确定什么时候会被回收。根据官方文档的说法我们需要手动调用Bitmap.recycle()去回收:
https://developer.android.com/topic/performance/graphics/manage-memory
在 Android 2.3.3(API 级别 10)及更低版本上,位图的后备像素数据存储在本地内存中。它与存储在 Dalvik 堆中的位图本身是分开的。本地内存中的像素数据并不以可预测的方式释放,可能会导致应用短暂超出其内存限制并崩溃。
在 Android 2.3.3(API 级别 10)及更低版本上,建议使用 recycle()。如果您在应用中显示大量位图数据,则可能会遇到 OutOfMemoryError 错误。利用recycle() 方法,应用可以尽快回收内存。
注意:只有当您确定位图已不再使用时才应该使用 recycle()。如果您调用recycle() 并在稍后尝试绘制位图,则会收到错误:"Canvas: trying to use a recycled bitmap"。
虽然3.0~7.1的版本Bitmp的像素内存是分配在java堆上的,但是实际是在natvie层进行decode的,而且会在native层创建一个c++的对象和java层的Bitmap对象进行关联。
从BitmapFactory的源码我们可以看到它一路调用到nativeDecodeStream这个native方法:
// BitmapFactory.java
public static Bitmap decodeFile(String pathName, Options opts) {
...
stream = new FileInputStream(pathName);
bm = decodeStream(stream, null, opts);
...
return bm;
}
public static Bitmap decodeStream(InputStream is, Rect outPadding, Options opts) {
...
bm = decodeStreamInternal(is, outPadding, opts);
...
return bm;
}
private static Bitmap decodeStreamInternal(InputStream is, Rect outPadding, Options opts) {
...
return nativeDecodeStream(is, tempStorage, outPadding, opts);
}
nativeDecodeStream实际上会通过jni创建java堆的内存,然后读取io流解码图片将像素数据存到这个java堆内存里面:
// BitmapFactory.cpp
static jobject nativeDecodeStream(JNIEnv* env, jobject clazz, jobject is, jbyteArray storage,
jobject padding, jobject options) {
...
bitmap = doDecode(env, bufferedStream, padding, options);
...
return bitmap;
}
static jobject doDecode(JNIEnv* env, SkStreamRewindable* stream, jobject padding, jobject options) {
...
// outputAllocator是像素内存的分配器,会在java堆上创建内存给像素数据,可以通过BitmapFactory.Options.inBitmap复用前一个bitmap像素内存
SkBitmap::Allocator* outputAllocator = (javaBitmap != NULL) ?
(SkBitmap::Allocator*)&recyclingAllocator : (SkBitmap::Allocator*)&javaAllocator;
...
// 将内存分配器设置给解码器
decoder->setAllocator(outputAllocator);
...
//解码
if (decoder->decode(stream, &decodingBitmap, prefColorType, decodeMode)
!= SkImageDecoder::kSuccess) {
return nullObjectReturn("decoder->decode returned false");
}
...
return GraphicsJNI::createBitmap(env, javaAllocator.getStorageObjAndReset(),
bitmapCreateFlags, ninePatchChunk, ninePatchInsets, -1);
}
// Graphics.cpp
jobject GraphicsJNI::createBitmap(JNIEnv* env, android::Bitmap* bitmap,
int bitmapCreateFlags, jbyteArray ninePatchChunk, jobject ninePatchInsets,
int density) {
// java层的Bitmap对象实际上是natvie层new出来的
// native层也会创建一个android::Bitmap对象与java层的Bitmap对象绑定
// bitmap->javaByteArray()代码bitmap的像素数据其实是存在java层的byte数组中
jobject obj = env->NewObject(gBitmap_class, gBitmap_constructorMethodID,
reinterpret_cast<jlong>(bitmap), bitmap->javaByteArray(),
bitmap->width(), bitmap->height(), density, isMutable, isPremultiplied,
ninePatchChunk, ninePatchInsets);
...
return obj;
}
我们可以看最后会调用javaAllocator.getStorageObjAndReset()创建一个android::Bitmap类型的native层Bitmap对象,然后通过jni调用java层的Bitmap构造函数去创建java层的Bitmap对象,同时将native层的Bitmap对象保存到mNativePtr:
// Bitmap.java
// Convenience for JNI access
private final long mNativePtr;
/**
* Private constructor that must received an already allocated native bitmap
* int (pointer).
*/
// called from JNI
Bitmap(long nativeBitmap, byte[] buffer, int width, int height, int density,
boolean isMutable, boolean requestPremultiplied,
byte[] ninePatchChunk, NinePatch.InsetStruct ninePatchInsets) {
...
mNativePtr = nativeBitmap;
...
}
从上面的源码我们也能看出来,Bitmap的像素是存在java堆的,所以如果bitmap没有人使用了,垃圾回收器就能自动回收这块的内存,但是在native创建出来的nativeBitmap要怎么回收呢?从Bitmap的源码我们可以看到在Bitmap构造函数里面还会创建一个BitmapFinalizer去管理nativeBitmap:
/**
* Private constructor that must received an already allocated native bitmap
* int (pointer).
*/
// called from JNI
Bitmap(long nativeBitmap, byte[] buffer, int width, int height, int density,
boolean isMutable, boolean requestPremultiplied,
byte[] ninePatchChunk, NinePatch.InsetStruct ninePatchInsets) {
...
mNativePtr = nativeBitmap;
mFinalizer = new BitmapFinalizer(nativeBitmap);
...
}
BitmapFinalizer的原理十分简单。Bitmap对象被销毁的时候BitmapFinalizer也会同步被销毁,然后就可以在BitmapFinalizer.finalize()里面销毁native层的nativeBitmap:
private static class BitmapFinalizer {
private long mNativeBitmap;
...
BitmapFinalizer(long nativeBitmap) {
mNativeBitmap = nativeBitmap;
}
...
@Override
public void finalize() {
try {
super.finalize();
} catch (Throwable t) {
// Ignore
} finally {
setNativeAllocationByteCount(0);
nativeDestructor(mNativeBitmap);
mNativeBitmap = 0;
}
}
}
8.0以后像素内存又被放回了native上,所以依然需要在java层的Bitmap对象回收之后同步回收native的内存。
虽然BitmapFinalizer同样可以实现,但是Java的finalize方法实际上是不推荐使用的,所以谷歌也换了NativeAllocationRegistry去实现:
/**
* Private constructor that must received an already allocated native bitmap
* int (pointer).
*/
// called from JNI
Bitmap(long nativeBitmap, int width, int height, int density,
boolean isMutable, boolean requestPremultiplied,
...
mNativePtr = nativeBitmap;
long nativeSize = NATIVE_ALLOCATION_SIZE + getAllocationByteCount();
NativeAllocationRegistry registry = new NativeAllocationRegistry(
Bitmap.class.getClassLoader(), nativeGetNativeFinalizer(), nativeSize);
registry.registerNativeAllocation(this, nativeBitmap);
}
NativeAllocationRegistry底层实际上使用了sun.misc.Cleaner,可以为对象注册一个清理的Runnable。当对象内存被回收的时候jvm就会调用它。
import sun.misc.Cleaner;
public Runnable registerNativeAllocation(Object referent, Allocator allocator) {
...
CleanerThunk thunk = new CleanerThunk();
Cleaner cleaner = Cleaner.create(referent, thunk);
..
}
private class CleanerThunk implements Runnable {
...
public void run() {
if (nativePtr != 0) {
ApplyFreeFunction(freeFunction, nativePtr);
}
registerNativeFree(size);
}
...
}
这个Cleaner的原理也很暴力,首先它是一个虚引用,registerNativeAllocation实际上创建了一个Bitmap的虚引用:
// Cleaner.java
public class Cleaner extends PhantomReference {
...
public static Cleaner create(Object ob, Runnable thunk) {
...
return add(new Cleaner(ob, thunk));
}
...
private Cleaner(Object referent, Runnable thunk) {
super(referent, dummyQueue);
this.thunk = thunk;
}
...
public void clean() {
...
thunk.run();
...
}
...
}
虚引用的话我们都知道需要配合一个ReferenceQueue使用,当对象的引用被回收的时候,jvm就会将这个虚引用丢到ReferenceQueue里面。而ReferenceQueue在插入的时候居然通过instanceof判断了下是不是Cleaner:
// ReferenceQueue.java
private boolean enqueueLocked(Reference<? extends T> r) {
...
if (r instanceof Cleaner) {
Cleaner cl = (sun.misc.Cleaner) r;
cl.clean();
...
}
...
}
也就是说Bitmap对象被回收,就会触发Cleaner这个虚引用被丢入ReferenceQueue,而ReferenceQueue里面会判断丢进来的虚引用是不是Cleaner,如果是就调用Cleaner.clean()方法。而clean方法内部就会再去执行我们注册的清理的Runnable。