GCD
全称Grand Central Dispatch 。应用程序以Block
块对象的形式提交任务到队列,GCD
提供并管理的这些队列。 提交到GCD
队列的工作在完全在系统管理的线程池上执行。
GCD
的一些常用方法:
Dispatch Queue
1. dispatch_get_main_queue
返回主线程,由系统创建,是一个串行队列。
2. dispatch_get_global_queue
dispatch_get_global_queue(long identifier, unsigned long flags);
返回具有指定优先级的系统定义的全局并发队列。其中flags
参数为保留位,供将来使用, 传递除零以外的任何值可能会导致 NULL
返回值。
优先级分为四种:高、默认、低以及后台
#define DISPATCH_QUEUE_PRIORITY_HIGH 2
#define DISPATCH_QUEUE_PRIORITY_DEFAULT 0
#define DISPATCH_QUEUE_PRIORITY_LOW (-2)
#define DISPATCH_QUEUE_PRIORITY_BACKGROUND INT16_MIN
对应 Qos的优先级为:
dispatch_queue_priority_t | QoS |
---|---|
HIGH | NSQualityOfServiceUserInitiated |
DEFAULT | NSQualityOfServiceDefault |
LOW | NSQualityOfServiceUtility |
BACKGROUND | NSQualityOfServiceBackground |
-
QOS_CLASS_USER_INTERACTIVE
最高优先级,即使在争用情况下也可以运行几乎所有可用的系统CPU和I / O带宽。所以使用时应限于与用户的关键交互,例如处理主事件循环上的事件,视图绘制,动画等。 -
QOS_CLASS_USER_INITIATED
低于用户的关键交互,但相对高于系统上的其他工作,使用于持续时间短的操作 -
QOS_CLASS_UTILITY
用于一些可能需要花点时间的任务,这些任务不需要马上返回结果,比如下载任务等 -
QOS_CLASS_BACKGROUND
用于完全不紧急的任务,磁盘 I/O后台备份等用这个,使用此QOS类表明工作应以最节能和最有效的方式运行 -
QOS_CLASS_DEFAULT
优先级介于user-initiated 和 utility,由pthread_create()
创建的线程没有指定QOS的属性将默认为QOS_CLASS_DEFAULT
, 此值不应用作工作分类,只应在传播或恢复系统提供的QOS类值时进行设置
3. dispatch_queue_create
dispatch_queue_create(const char *_Nullable label, dispatch_queue_attr_t _Nullable attr);
其中label
:队列名称,用于debug
时唯一标示此队列,建议使用反向DNS命名样式(com.example.myqueue),因为应用程序、库和框架等都可以创建自己的队列
attr
:DISPATCH_QUEUE_SERIAL
or NULL
为串行队列,DISPATCH_QUEUE_CONCURRENT
为并行队列
iOS 8之后可以使用dispatch_queue_attr_make_with_qos_class
方法来生成dispatch_queue_attr_t
dispatch_queue_attr_make_with_qos_class(dispatch_queue_attr_t _Nullable attr,
dispatch_qos_class_t qos_class, int relative_priority)
relative_priority
为相对优先级(相对于指定优先级的负偏移量,因为优先级qos_class
实际就是一个 int
的枚举),传递大于零或小于QOS_MIN_RELATIVE_PRIORITY
(-15)的值会导致返回NULL
;一般直接设置为0
还可以通过dispatch_set_target_queue
设置优先级
dispatch_queue_t serialQueue = dispatch_queue_create("com.example.MyQueue",NULL);
dispatch_queue_t globalQueue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_DEFAULT,0);
//serialQueue现在的优先级跟globalQueue的优先级一样
dispatch_set_target_queue(serialQueue, globalQueue);
这个方法的前提是线程创建时没有指定优先级,但是系统还是建议使用dispatch_queue_attr_make_with_qos_class
来设置优先级
另外,dispatch_set_target_queue
有一个属性,就是如果一个串行队列,他的目标队列是另一个串行队列,那么提交到这个串行队列的 block
块不会与提交到目标队列或者具有相同目标队列的任何其他队列的block
块同时调用,比较绕口,如下代码:
dispatch_queue_t targetQueue = dispatch_queue_create("com.yxw.target_queue", DISPATCH_QUEUE_SERIAL);
dispatch_queue_t queue1 = dispatch_queue_create("com.yxw.queue1", DISPATCH_QUEUE_SERIAL);
dispatch_queue_t queue2 = dispatch_queue_create("com.yxw.queue2", DISPATCH_QUEUE_CONCURRENT);
dispatch_set_target_queue(queue1, targetQueue);
dispatch_set_target_queue(queue2, targetQueue);
dispatch_async(queue1, ^{
NSLog(@"target do job1");
[NSThread sleepForTimeInterval:5.f];
});
dispatch_async(queue2, ^{
NSLog(@"target do job2");
[NSThread sleepForTimeInterval:2.f];
});
dispatch_async(queue2, ^{
NSLog(@"target do job3");
[NSThread sleepForTimeInterval:1.f];
});
dispatch_async(targetQueue, ^{
NSLog(@"target do job4");
[NSThread sleepForTimeInterval:2.f];
});
执行结果:
YMultiThreadDemo[4068:382944] target do job1
YMultiThreadDemo[4068:382944] target do job2
YMultiThreadDemo[4068:382944] target do job3
YMultiThreadDemo[4068:382944] target do job4
4. dispatch_barrier_async
dispatch_barrier_async
用于等待队列前面的任务执行完毕后自己才执行,而它后面的任务也需等待它完成之后才执行。如数据的读写:
dispatch_queue_t dbQueue = dispatch_queue_create("com.yxw.database", DISPATCH_QUEUE_CONCURRENT);
dispatch_async(dbQueue, ^{
NSLog(@"DB reading data1");
[NSThread sleepForTimeInterval:1.];
NSLog(@"DB read data1 completed");
});
dispatch_async(dbQueue, ^{
NSLog(@"DB reading data2");
[NSThread sleepForTimeInterval:2.];
NSLog(@"DB read data2 completed");
});
dispatch_barrier_async(dbQueue, ^{
NSLog(@"DB writing data1");
[NSThread sleepForTimeInterval:1.];
NSLog(@"DB write data1 completed");
});
dispatch_async(dbQueue, ^{
NSLog(@"DB reading data3");
[NSThread sleepForTimeInterval:1.];
});
执行结果:
YMultiThreadDemo[4239:399611] DB reading data1
YMultiThreadDemo[4239:399637] DB reading data2
YMultiThreadDemo[4239:399611] DB read data1 completed
YMultiThreadDemo[4239:399637] DB read data2 completed
YMultiThreadDemo[4239:399637] DB writing data1
YMultiThreadDemo[4239:399637] DB write data1 completed
YMultiThreadDemo[4239:399637] DB reading data3
5. dispatch_queue_set_specific 、dispatch_get_specific
dispatch_queue_set_specific(dispatch_queue_t queue, const void *key,
void *_Nullable context, dispatch_function_t _Nullable destructor)
dispatch_queue_get_specific(dispatch_queue_t queue, const void *key)
dispatch_queue_set_specific
用系统唯一的key
将特定的上下文与queue
关联,dispatch_queue_get_specific
则是通过这个key
返回关联的上下文,如:
dispatch_queue_set_specific(queue, kDispatchQueueSpecificKey, (__bridge void *)self, NULL);
YGCDViewController *wself =
(__bridge id)dispatch_queue_get_specific(queue, kDispatchQueueSpecificKey);
NSLog(@"get specific wself title %@",wself.navigationItem.title);
6. dispatch_apply
dispatch_apply(size_t iterations,
dispatch_queue_t DISPATCH_APPLY_QUEUE_ARG_NULLABILITY queue,
DISPATCH_NOESCAPE void (^block)(size_t))
dispatch_apply
类似一个循环,在指定的queue
中执行iterations
后返回,如果指定的队列是并发的,则会同时调用block
块,这样就要求block
是可重入的
dispatch_apply(5, serialQueue, ^(size_t i) {
NSLog(@"apply run %zi times",i);
});
执行结果:
YMultiThreadDemo[4961:481865] apply run 0 times
YMultiThreadDemo[4961:481865] apply run 1 times
YMultiThreadDemo[4961:481865] apply run 2 times
YMultiThreadDemo[4961:481865] apply run 3 times
YMultiThreadDemo[4961:481865] apply run 4 times
Dispatch Block
dispatch_block_create(dispatch_block_flags_t flags, dispatch_block_t block);
返回的块已经由Block_copy
拷贝到堆中,返回块的第一次执行完成才会响应dispatch_block_wait()
或者dispatch_block_notify()
方法
flags
参数:
-
DISPATCH_BLOCK_BARRIER
,提交到并行队列时,这个标志相当于dispatch_barrier_async()
-
DISPATCH_BLOCK_INHERIT_QOS_CLASS
,异步提交到队列时默认使用 -
DISPATCH_BLOCK_ENFORCE_QOS_CLASS
,同步提交到队列时默认使用
1. dispatch_block_wait 与 dispatch_block_notify
dispatch_block_wait(dispatch_block_t block, dispatch_time_t timeout)
dispatch_block_notify(dispatch_block_t block, dispatch_queue_t queue,
dispatch_block_t notification_block)
dispatch_block_wait
:等待指定的block
块执行完,或者指定的timeout
已经超出,这个方法会阻塞当前线程,所以不要放在主线程中,另外,用dispatch_block_cancel
方法取消的 block
也算执行完成
dispatch_block_notify
:功能跟上面方法类似,等待指定的block
块执行完,然后在queue
中执行notification_block
2. dispatch_block_cancel
iOS 8 之后,可以通过dispatch_block_cancel
方法取消待执行的block
,已经在执行的block
没有影响,与取消block
相关联的任何资源将被延迟释放,直到下一次尝试执行该block
上面三个方法的代码如下:
dispatch_queue_t queue = dispatch_queue_create("com.yxw.queue", DISPATCH_QUEUE_SERIAL);
dispatch_block_t block = dispatch_block_create(0, ^{
NSLog(@"before block sleep");
[NSThread sleepForTimeInterval:2.];
NSLog(@"after block sleep");
});
dispatch_block_t block2 = dispatch_block_create(0, ^{
NSLog(@"before block2 sleep");
[NSThread sleepForTimeInterval:10.];
NSLog(@"after block2 sleep");
});
dispatch_async(queue, block);
dispatch_async(queue, block2);
dispatch_async(globalQueue, ^{
//等待block执行完毕
dispatch_block_wait(block, DISPATCH_TIME_FOREVER);
NSLog(@"block wait coutinue");
});
dispatch_block_notify(block, globalQueue, ^{
NSLog(@"block notify coutinue");
});
dispatch_async(globalQueue, ^{
[NSThread sleepForTimeInterval:0.5];
dispatch_block_cancel(block2);
dispatch_block_cancel(block);
});
dispatch_block_notify(block2, globalQueue, ^{
NSLog(@"block2 notify coutinue");
});
执行结果如下:
YMultiThreadDemo[2690:227780] before block sleep
YMultiThreadDemo[2690:227780] after block sleep
YMultiThreadDemo[2690:227780] block notify coutinue
YMultiThreadDemo[2690:227959] block wait coutinue
YMultiThreadDemo[2690:227962] block2 notify coutinue
可以看到,block2
取消成功,block2
的notify
执行了,block
取消无效
下面的代码测试只有block
的第一次执行完成才会notify
:
dispatch_block_t block3 = dispatch_block_create(0, ^{
NSLog(@"before block3 sleep");
[NSThread sleepForTimeInterval:1.];
NSLog(@"after block3 sleep");
});
dispatch_async(globalQueue, block3);
dispatch_block_notify(block3, globalQueue, ^{
NSLog(@"block3 notify coutinue");
});
dispatch_async(globalQueue, ^{
[NSThread sleepForTimeInterval:3.];
block3();
});
执行结果:
YMultiThreadDemo[4569:356960] before block3 sleep
YMultiThreadDemo[4569:356960] after block3 sleep
YMultiThreadDemo[4569:356960] block3 notify coutinue
YMultiThreadDemo[4569:356961] before block3 sleep
YMultiThreadDemo[4569:356961] after block3 sleep
Dispatch Group
当我们想在gcd queue中所有的任务执行完毕之后做些特定事情的时候,也就是队列的同步问题,如果队列是串行的话,那将该操作最后添加到队列中即可,但如果队列是并行队列的话,这时候就可以利用dispatch_group来实现了,dispatch_group能很方便的解决同步的问题。dispatch_group_create可以创建一个group对象,然后可以添加block到该组里面,下面看下它的一些用法:
1. dispatch_group_wait 与 dispatch_group_notify
跟上面的dispatch_block_wait
、 dispatch_block_notify
类似,但这里是等待前面提交到group
的block
执行完
代码如下:
dispatch_group_t group = dispatch_group_create();
dispatch_group_async(group,globalQueue,^{ NSLog(@"group block1"); });
dispatch_group_async(group,globalQueue,^{ NSLog(@"group block2"); });
dispatch_async(globalQueue, ^{
dispatch_group_wait(group, DISPATCH_TIME_FOREVER);
NSLog(@"group wait done");
});
dispatch_group_notify(group, globalQueue, ^{
NSLog(@"group notify done");
});
执行结果:
YMultiThreadDemo[4877:387857] group block1
YMultiThreadDemo[4877:387857] group block2
YMultiThreadDemo[4877:387857] group wait done
YMultiThreadDemo[4877:387858] group notify done
2. dispatch_group_enter 与 dispatch_group_leave
假如我们不想使用dispatch_group_async异步的将任务丢到group中去执行,这时候就需要用到dispatch_group_enter跟dispatch_group_leave方法,这两个方法要配对出现,以下这两种方法是等价的:
dispatch_group_async(group, queue, ^{
});
dispatch_group_enter(group);
dispatch_async(queue, ^{
dispatch_group_leave(group);
});
dispatch_once
dispatch_once(dispatch_once_t *predicate,
DISPATCH_NOESCAPE dispatch_block_t block);
在 APP 生命周期内,保证只执行一次。一般用于单例的初始化。如果从多个线程同时调用,则此函数将同步等待,所以能保证唯一单例。
predicate
与dispatch_once
一起使用的谓词。 它必须初始化为零,而且必须是静态或全局变量。因为dispatch_once_t
实际就是long
类型,所以传进去的是指向dispatch_once_t
的指针,用于表示block
块是否已完成(执行完predicate
为-1)。
如下:
static dispatch_once_t onceToken;
dispatch_once(&onceToken, ^{
});
NSLog(@"onceToken %zi",onceToken);
执行完后onceToken
为-1,表示执行完,下次调用就不会再执行block
块了。
参考文献
https://developer.apple.com/documentation/dispatch?language=objc
网友评论