释放双眼，带上耳机，听听看~！
2
3
4
2 implements ConcurrentMap<K,V>, Serializable {
3
4
一些成员变量
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2private static final int MAXIMUM_CAPACITY = 1 << 30;
3//默认容量
4private static final int DEFAULT_CAPACITY = 16;
5//扩容因子，当达到容量达到n*LOAD_FACTOR时，就会扩容
6private static final float LOAD_FACTOR = 0.75f;
7//超过这个值会扩容或建红黑树
8static final int TREEIFY_THRESHOLD = 8;
9//从链表升级为红黑树的阈值
10static final int MIN_TREEIFY_CAPACITY = 64;
11//多线程扩容，这个表示每个线程最少负责迁移数据的数量
12private static final int MIN_TRANSFER_STRIDE = 16;
13//下面都是hash值
14static final int MOVED = -1; //表示在扩容
15static final int TREEBIN = -2; //表示是树节点
16static final int RESERVED = -3;
17//hash值取除最高一位以外的31位
18static final int HASH_BITS = 0x7fffffff;
19//CPU数量
20static final int NCPU = Runtime.getRuntime().availableProcessors();
21//装载数据的数组
22 transient volatile Node<K,V>[] table;
23//扩容用的新数组
24 private transient volatile Node<K,V>[] nextTable;
25//没有并发时的计数
26 private transient volatile long baseCount;
27//达到该值就会扩容
28 private transient volatile int sizeCtl;
29
30
构造方法
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
2 if (initialCapacity < 0)
3 throw new IllegalArgumentException();
4 //如果initialCapacity大于MAXIMUM_CAPACITY的一半就直接赋值为MAXIMUM_CAPACITY
5 //否则赋值为1.5倍initialCapacity+1，然后向上去最近2的次方的数
6 int cap = ((initialCapacity >= (MAXIMUM_CAPACITY >>> 1)) ?
7 MAXIMUM_CAPACITY :
8 tableSizeFor(initialCapacity + (initialCapacity >>> 1) + 1));
9 this.sizeCtl = cap;
10 }
11
12public ConcurrentHashMap(int initialCapacity,
13 float loadFactor, int concurrencyLevel) {
14 if (!(loadFactor > 0.0f) || initialCapacity < 0 || concurrencyLevel <= 0)
15 throw new IllegalArgumentException();
16 if (initialCapacity < concurrencyLevel) // Use at least as many bins
17 initialCapacity = concurrencyLevel; // as estimated threads
18 long size = (long)(1.0 + (long)initialCapacity / loadFactor);
19 int cap = (size >= (long)MAXIMUM_CAPACITY) ?
20 MAXIMUM_CAPACITY : tableSizeFor((int)size);
21 this.sizeCtl = cap;
22 }
23
24
数据都封装到Node类中
2
3
4
5
6
7
8
9
10
11
12
13
14
15
2 final int hash;//哈希值
3 final K key;//键
4 volatile V val;//数据
5 volatile Node<K,V> next;//下一个节点
6 Node(int hash, K key, V val, Node<K,V> next) {
7 this.hash = hash;
8 this.key = key;
9 this.val = val;
10 this.next = next;
11 }
12 ...
13}
14
15
下面从增删查分析源码
分析之前，先了解下ConcurrentHashMap类中CAS操作，主要有以下三个方法
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
2 //返回第i个Node元素，i左移ASHIFT位+第一个元素的地址ABASE
3 return (Node<K,V>)U.getObjectVolatile(tab, ((long)i << ASHIFT) + ABASE);
4 }
5
6 static final <K,V> boolean casTabAt(Node<K,V>[] tab, int i,
7 Node<K,V> c, Node<K,V> v) {
8 //CAS方式更新第i个位置的元素
9 return U.compareAndSwapObject(tab, ((long)i << ASHIFT) + ABASE, c, v);
10 }
11 static final <K,V> void setTabAt(Node<K,V>[] tab, int i, Node<K,V> v) {
12 //cas赋值数组第i个元素
13 U.putObjectVolatile(tab, ((long)i << ASHIFT) + ABASE, v);
14 }
15
16
ASHIFT、ABASE在静态代码块中进行赋值
2
3
4
5
6
7
8
9
10
11
12
2 //arrayBaseOffset: 返回当前数组第一个元素地址相对于数组起始地址的偏移值
3 ABASE = U.arrayBaseOffset(ak);
4 //arrayIndexScale: 返回当前数组一个元素占用的字节数,在本例中返回4。
5 int scale = U.arrayIndexScale(ak);
6 if ((scale & (scale - 1)) != 0)//必须是2的指数
7 throw new Error("data type scale not a power of two");
8 //Integer.numberOfLeadingZeros：返回int数字的高位直到第一个非0位的位数，如4，返回29（100，前面29个0）
9 //ASHIFT作用：可方便地利用左移ASHIFT位拿到数组第i个元素的地址偏移量
10 ASHIFT = 31 - Integer.numberOfLeadingZeros(scale);
11
12
这里CAS操作数组，可以这么理解：拿到数组第一个元素的偏移地址p，然后拿到每个数据所占的字节数c，那么第i个元素偏移地址就是p+i*c。所以可以利用左移ASHIFT位来拿到第i个元素的偏移量，再加上ABASE，就是第i个元素相对于数组起始地址的偏移量。
1. 增加数据
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
2 //插入数据，存在就覆盖
3 return putVal(key, value, false);
4 }
5
6final V putVal(K key, V value, boolean onlyIfAbsent) {
7 if (key == null || value == null) throw new NullPointerException();
8//【标记1】对key的哈希值进行处理，使得更离散
9 int hash = spread(key.hashCode());
10 int binCount = 0;//统计链表长度
11 for (Node<K,V>[] tab = table;;) {
12 Node<K,V> f; int n, i, fh;
13 if (tab == null || (n = tab.length) == 0)
14 tab = initTable();//【标记2】一开始没初始化数组
15 else if ((f = tabAt(tab, i = (n - 1) & hash)) == null) {//用(n - 1) & hash取到数组下标，发现其元素为null
16 if (casTabAt(tab, i, null,
17 new Node<K,V>(hash, key, value, null)))
18 break;//直接CAS成功添加了一个新节点
19 }
20 else if ((fh = f.hash) == MOVED)//MOVED是数据迁移标志
21 tab = helpTransfer(tab, f);//【标记3】正在数据迁移，去帮助数据迁移
22 else {
23 V oldVal = null;
24 synchronized (f) {//锁住当前节点
25 if (tabAt(tab, i) == f) {
26 if (fh >= 0) {//说明是链表结构
27 binCount = 1;//计算链表节点数量
28 for (Node<K,V> e = f;; ++binCount) {
29 K ek;
30 if (e.hash == hash &&
31 ((ek = e.key) == key ||
32 (ek != null && key.equals(ek)))) {
33 oldVal = e.val;//找到原来的值
34 if (!onlyIfAbsent)
35 e.val = value;//只用onlyIfAbsent为false才更新值
36 break;
37 }
38 Node<K,V> pred = e;
39 if ((e = e.next) == null) {//到了链尾
40 //这里新建一个节点插入
41 pred.next = new Node<K,V>(hash, key,
42 value, null);
43 break;
44 }
45 }
46 }
47 else if (f instanceof TreeBin) {
48 Node<K,V> p;
49 binCount = 2;
50 if ((p = ((TreeBin<K,V>)f).putTreeVal(hash, key,
51 value)) != null) {
52 oldVal = p.val;
53 if (!onlyIfAbsent)
54 p.val = value;
55 }
56 }
57 }
58 }
59 if (binCount != 0) {
60 if (binCount >= TREEIFY_THRESHOLD)
61 treeifyBin(tab, i);//【标记4】超过阀值开始考虑建红黑树
62 if (oldVal != null)
63 return oldVal;
64 break;
65 }
66 }
67 }
68 //当对应的key原先不存在才到这里
69//【标记5】增加计数
70 addCount(1L, binCount);
71 return null;
72 }
73
74
ConcurrentHashMap插入需要保证线程安全，所以操作数据用了CAS方式，另外当插入的时候如果正好在扩容，就会去帮忙扩容，最后插入成功后会进行判断是否需要扩容，是否要增加计数。
【标记1】hash值处理，使映射更加离散
2
3
4
5
6
7
2 //异或，然后再取低31位
3 //HASH_BITS = 0x7fffffff;
4 return (h ^ (h >>> 16)) & HASH_BITS;
5}
6
7
【标记2】初始化数组
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
2 Node<K,V>[] tab; int sc;
3 while ((tab = table) == null || tab.length == 0) {
4 if ((sc = sizeCtl) < 0)
5 Thread.yield(); // 被其他线程抢先了，睡眠一下，醒来再自旋判断
6 else if (U.compareAndSwapInt(this, SIZECTL, sc, -1)) {//CAS方式设置为-1，防止多线程同时初始化
7 try {
8 if ((tab = table) == null || tab.length == 0) {
9 //如果sizeCtl小于等于0 ，会赋值为DEFAULT_CAPACITY
10 int n = (sc > 0) ? sc : DEFAULT_CAPACITY;
11 //申请数组空间
12 @SuppressWarnings("unchecked")
13 Node<K,V>[] nt = (Node<K,V>[])new Node<?,?>[n];
14 table = tab = nt;
15 //n-0.25n = 0.75n
16 sc = n - (n >>> 2);
17 }
18 } finally {
19 sizeCtl = sc;
20 }
21 break;
22 }
23 }
24 return tab;
25 }
26
27
数组一开始初始化容量是根据sizeCtl的值来决定的，然后sizeCtl会被更新为容量的0.75倍。
【标记3】帮助数据迁移
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
2 Node<K,V>[] nextTab; int sc;
3 //必须要原先tab不为null，而且f是ForwardingNode，并且新数组已经被赋值
4 if (tab != null && (f instanceof ForwardingNode) &&
5 (nextTab = ((ForwardingNode<K,V>)f).nextTable) != null) {
6 int rs = resizeStamp(tab.length);
7 while (nextTab == nextTable && table == tab &&
8 (sc = sizeCtl) < 0) {
9 //下面的条件不怎么看懂，大概是在控制帮助数据迁移的线程数量
10 // 最后的transferIndex <= 0说明数据迁移已经全部由其他线程完成或正在进行
11 if ((sc >>> RESIZE_STAMP_SHIFT) != rs || sc == rs + 1 ||
12 sc == rs + MAX_RESIZERS || transferIndex <= 0)
13 break;
14 if (U.compareAndSwapInt(this, SIZECTL, sc, sc + 1)) {
15 transfer(tab, nextTab);//进行数据迁移
16 break;
17 }
18 }
19 return nextTab;
20 }
21//上面条件不满足就返回旧的数组
22 return table;
23 }
24
25
【标记4】超过阀值开始考虑建红黑树
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
2 Node<K,V> b; int n, sc;
3 if (tab != null) {
4 if ((n = tab.length) < MIN_TREEIFY_CAPACITY)
5 tryPresize(n << 1);//小于MIN_TREEIFY_CAPACITY就只是数组扩容
6 else if ((b = tabAt(tab, index)) != null && b.hash >= 0) {
7 ...链表转红黑树
8 }
9 }
10 }
11
12private final void tryPresize(int size) {
13 //如果size大于一般MAXIMUM_CAPACITY就直接取MAXIMUM_CAPACITY，
14 //否则1.5倍size+1，然后向上取最近的2的次方的数
15 int c = (size >= (MAXIMUM_CAPACITY >>> 1)) ? MAXIMUM_CAPACITY :
16 tableSizeFor(size + (size >>> 1) + 1);//
17 int sc;
18 while ((sc = sizeCtl) >= 0) {//大于0说明没有正在扩容
19 Node<K,V>[] tab = table; int n;
20 //table为null说明是从ConcurrentHashMap(Map<? extends K, ? extends V> m)构造方法进来的
21 if (tab == null || (n = tab.length) == 0) {
22 n = (sc > c) ? sc : c;
23 if (U.compareAndSwapInt(this, SIZECTL, sc, -1)) {
24 //赋值为-1，待会可以退出出循环，且阻止其他线程进入这
25 try {
26 if (table == tab) {
27 @SuppressWarnings("unchecked")
28 Node<K,V>[] nt = (Node<K,V>[])new Node<?,?>[n];
29 table = nt;
30 sc = n - (n >>> 2);//实际是0.75n
31 }
32 } finally {
33 sizeCtl = sc;
34 }
35 }
36 }
37 else if (c <= sc || n >= MAXIMUM_CAPACITY)//小于阈值或大于最大容量就不在扩容了
38 break;
39 else if (tab == table) {
40 int rs = resizeStamp(n);
41 if (sc < 0) {//小于0说明已经开始数据迁移了，新的数组肯定不为空
42 Node<K,V>[] nt;
43 if ((sc >>> RESIZE_STAMP_SHIFT) != rs || sc == rs + 1 ||
44 sc == rs + MAX_RESIZERS || (nt = nextTable) == null ||
45 transferIndex <= 0)
46 break;
47 if (U.compareAndSwapInt(this, SIZECTL, sc, sc + 1))
48 transfer(tab, nt);//开始数据迁移
49 }
50 //sc还大于0，说明现在才进行数据迁移，而新的数组还没申请
51 else if (U.compareAndSwapInt(this, SIZECTL, sc,
52 (rs << RESIZE_STAMP_SHIFT) + 2))
53 transfer(tab, null);
54 }
55 }
56 }
57
58
真正开始数据迁移是在transfer()方法
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
2 int n = tab.length, stride;
3 //cpu数量为1就赋值为n,否则赋值为(n >>> 3) / NCPU ，但最小为MIN_TRANSFER_STRIDE
4 if ((stride = (NCPU > 1) ? (n >>> 3) / NCPU : n) < MIN_TRANSFER_STRIDE)
5 stride = MIN_TRANSFER_STRIDE; // subdivide range
6 if (nextTab == null) { //第一个线程开始扩容时，nextTab为null
7 try {
8 @SuppressWarnings("unchecked")
9 //容量为原来的两倍
10 Node<K,V>[] nt = (Node<K,V>[])new Node<?,?>[n << 1];
11 nextTab = nt;
12 } catch (Throwable ex) { // try to cope with OOME
13 sizeCtl = Integer.MAX_VALUE;
14 return;
15 }
16 nextTable = nextTab;
17 transferIndex = n;//从数组末尾开始数据迁移
18 }
19 int nextn = nextTab.length;
20 //用于标记旧数组中某位置已完成数据迁移
21 ForwardingNode<K,V> fwd = new ForwardingNode<K,V>(nextTab);
22 boolean advance = true;
23 boolean finishing = false; // to ensure sweep before committing nextTab
24 for (int i = 0, bound = 0;;) {
25 Node<K,V> f; int fh;
26 while (advance) {
27 int nextIndex, nextBound;
28 if (--i >= bound || finishing)//一开始-1会小于0
29 advance = false;//小于界限或已经完成就赋值为false跳出循环
30 //transferIndex<=0说明所有数据迁移都有其他线程完成或进行中
31 else if ((nextIndex = transferIndex) <= 0) {
32 i = -1;
33 advance = false;
34 }
35 //CAS更新transferIndex值
36 else if (U.compareAndSwapInt
37 (this, TRANSFERINDEX, nextIndex,
38 nextBound = (nextIndex > stride ?
39 nextIndex - stride : 0))) {
40 bound = nextBound;//赋值迁移数据的界限
41 i = nextIndex - 1;//赋值迁移数据的开始下标
42 advance = false;//跳出循环开始迁移
43 }
44 }
45 if (i < 0 || i >= n || i + n >= nextn) {
46 int sc;
47 if (finishing) {//数据迁移完成
48 nextTable = null;//
49 table = nextTab;
50 sizeCtl = (n << 1) - (n >>> 1);//0.75n
51 return;
52 }
53 if (U.compareAndSwapInt(this, SIZECTL, sc = sizeCtl, sc - 1)) {
54 if ((sc - 2) != resizeStamp(n) << RESIZE_STAMP_SHIFT)
55 return;//这里大概会筛选出一个线程进行一遍数据迁移检查
56 finishing = advance = true;
57 i = n; // 赋值为n，重新检查一遍是否所有数据已经迁移完成
58 }
59 }
60 else if ((f = tabAt(tab, i)) == null)
61 //原先为null，说明没有数据可迁移，直接cas赋值为fwd
62 advance = casTabAt(tab, i, null, fwd);
63 else if ((fh = f.hash) == MOVED)
64 //已经是ForwardNode，进入下次循环
65 advance = true; // already processed
66 else {
67 synchronized (f) {//锁住f
68 if (tabAt(tab, i) == f) {//如果不等，说明第i个节点已变成ForwardingNode了
69 //分成两条链表
70 //ln是表示原先节点的哈希值&n为0
71 //hn表示原先节点的哈希值&n为1
72 Node<K,V> ln, hn;
73 if (fh >= 0) {//说明是链表
74 int runBit = fh & n;
75 Node<K,V> lastRun = f;
76 for (Node<K,V> p = f.next; p != null; p = p.next) {
77 int b = p.hash & n;
78 if (b != runBit) {
79 runBit = b;
80 lastRun = p;
81 }
82 }
83 //lastRun节点及其后面的节点的hash&n都是0或都是1
84 if (runBit == 0) {
85 ln = lastRun;
86 hn = null;
87 }
88 else {
89 hn = lastRun;
90 ln = null;
91 }
92 for (Node<K,V> p = f; p != lastRun; p = p.next) {
93 int ph = p.hash; K pk = p.key; V pv = p.val;
94 if ((ph & n) == 0)
95 ln = new Node<K,V>(ph, pk, pv, ln);
96 else
97 hn = new Node<K,V>(ph, pk, pv, hn);
98 }
99 //设置新数组的第i位元素为ln链表
100 setTabAt(nextTab, i, ln);
101 //设置新数组的第i+n位元素为hn链表
102 setTabAt(nextTab, i + n, hn);
103 //设置旧数组中第i个元素为ForwardingNode
104 setTabAt(tab, i, fwd);
105 advance = true;
106 }
107 else if (f instanceof TreeBin) {//红黑树操作
108 ...
109 }
110 }
111 }
112 }
113 }
114 }
115
116
2. 删除操作
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
2 return replaceNode(key, null, null);
3 }
4
5final V replaceNode(Object key, V value, Object cv) {
6 int hash = spread(key.hashCode());
7 for (Node<K,V>[] tab = table;;) {
8 Node<K,V> f; int n, i, fh;
9 if (tab == null || (n = tab.length) == 0 ||
10 (f = tabAt(tab, i = (n - 1) & hash)) == null)
11 break;//数组为null或数组第(n - 1) & hash下标下的元素为null直接退出
12 else if ((fh = f.hash) == MOVED)
13 tab = helpTransfer(tab, f);//帮忙迁移数据
14 else {
15 V oldVal = null;
16 boolean validated = false;
17 synchronized (f) {
18 //多线程原因要判断是否是同个对象，因为可能被其他线程删除、更换了原来的元素
19 if (tabAt(tab, i) == f) {
20 if (fh >= 0) {//链表结构
21 validated = true;//表示真正操作了
22 for (Node<K,V> e = f, pred = null;;) {
23 K ek;
24 if (e.hash == hash &&
25 ((ek = e.key) == key ||
26 (ek != null && key.equals(ek)))) {
27 V ev = e.val;
28 //cv为null或当和原先的值和cv相等才替换
29 if (cv == null || cv == ev ||
30 (ev != null && cv.equals(ev))) {
31 oldVal = ev;
32 if (value != null)
33 e.val = value;
34 else if (pred != null)//断开链表
35 pred.next = e.next;
36 else//是链表头，直接把表头赋值为下一个节点
37 setTabAt(tab, i, e.next);
38 }
39 break;
40 }
41 pred = e;
42 if ((e = e.next) == null)
43 break;//这里就是没有找到要删除的元素
44 }
45 }
46 else if (f instanceof TreeBin) {//红黑树操作
47 validated = true;
48 TreeBin<K,V> t = (TreeBin<K,V>)f;
49 TreeNode<K,V> r, p;
50 if ((r = t.root) != null &&
51 (p = r.findTreeNode(hash, key, null)) != null) {
52 V pv = p.val;
53 if (cv == null || cv == pv ||
54 (pv != null && cv.equals(pv))) {
55 oldVal = pv;
56 if (value != null)
57 p.val = value;
58 else if (t.removeTreeNode(p))
59 setTabAt(tab, i, untreeify(t.first));
60 }
61 }
62 }
63 }
64 }
65 if (validated) {//如果真正操作了
66 if (oldVal != null) {//原先不为null
67 if (value == null)//而且新赋值为null说明是删除元素
68 addCount(-1L, -1);//【标记5】减少计数
69 return oldVal;//返回旧的值
70 }
71 break;
72 }
73 }
74 }
75 return null;
76 }
77
78
3. 查询数据
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
2 Node<K,V>[] tab; Node<K,V> e, p; int n, eh; K ek;
3 int h = spread(key.hashCode());
4 if ((tab = table) != null && (n = tab.length) > 0 &&
5 (e = tabAt(tab, (n - 1) & h)) != null) {
6 //第一个元素就是要找的元素
7 if ((eh = e.hash) == h) {
8 if ((ek = e.key) == key || (ek != null && key.equals(ek)))
9 return e.val;
10 }
11 else if (eh < 0)//说明不是链表结构，调用不同节点类型的find方法
12 return (p = e.find(h, key)) != null ? p.val : null;
13 //到这里说明是链表结构
14 while ((e = e.next) != null) {
15 if (e.hash == h &&
16 ((ek = e.key) == key || (ek != null && key.equals(ek))))
17 return e.val;
18 }
19 }
20 return null;//没有找到元素
21 }
22
23
4. 计数操作
ConcurrentHashMap的计数方式是，先尝试在baseCount（一个int类型）上计数、如果遇到并发就在尝试在CounterCell上计数。而ConcurrentHashMap要拿到当前所有元素个数的时候就是baseCount+每个CounterCell的计数
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
2 volatile long value;
3 CounterCell(long x) { value = x; }
4 }
5
6final long sumCount() {
7 CounterCell[] as = counterCells; CounterCell a;
8 long sum = baseCount;
9 if (as != null) {
10 for (int i = 0; i < as.length; ++i) {
11 if ((a = as[i]) != null)
12 sum += a.value;
13 }
14 }
15 //sum=baseCount + counterCells[0..n].value
16 return sum;
17 }
18
19
文中的【标记5】处是进行计数操作，代码如下，其中参数x表示要增加的计数（正负都有可能），check大于0表示要检查是否达到阈值扩容
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
2 CounterCell[] as; long b, s;
3 if ((as = counterCells) != null ||
4 !U.compareAndSwapLong(this, BASECOUNT, b = baseCount, s = b + x)) {
5 //当counterCells为null或CAS设置baseCount的值失败时，进入下面
6 CounterCell a; long v; int m;
7 boolean uncontended = true;
8 //uncontended为false条件：当counterCells不为null，而且指定下标的计数器不为null，然后cas更新计数器值失败
9 if (as == null || (m = as.length - 1) < 0 ||
10 (a = as[ThreadLocalRandom.getProbe() & m]) == null ||
11 !(uncontended =
12 U.compareAndSwapLong(a, CELLVALUE, v = a.value, v + x))) {
13 //ThreadLocalRandom.getProbe() & m：定位要用哪个计数器
14 //如果需要用到的计数器为null或没有累加成功就进入这里
15 //【标记6】全力以赴计数操作
16 fullAddCount(x, uncontended);
17 return;
18 }
19 if (check <= 1)
20 return;
21 s = sumCount();//计算一下容量，为下面如果check>=0时，用其来判断是否要扩容
22 }
23 if (check >= 0) {//check大于等于0才检查是否要扩容
24 Node<K,V>[] tab, nt; int n, sc;
25 while (s >= (long)(sc = sizeCtl) && (tab = table) != null &&
26 (n = tab.length) < MAXIMUM_CAPACITY) {
27 int rs = resizeStamp(n);
28 if (sc < 0) {
29 if ((sc >>> RESIZE_STAMP_SHIFT) != rs || sc == rs + 1 ||
30 sc == rs + MAX_RESIZERS || (nt = nextTable) == null ||
31 transferIndex <= 0)
32 break;
33 if (U.compareAndSwapInt(this, SIZECTL, sc, sc + 1))
34 transfer(tab, nt);
35 }
36 else if (U.compareAndSwapInt(this, SIZECTL, sc,
37 (rs << RESIZE_STAMP_SHIFT) + 2))
38 transfer(tab, null);
39 s = sumCount();//再次计算容量,看是否继续扩容
40 }
41 }
42 }
43
44
【标记6】处，当未能直接在baseCount上计数，且不能成功在CounterCell[]数组的指定下标计数器上计数，就会进入下面方法
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
2 int h;
3 if ((h = ThreadLocalRandom.getProbe()) == 0) {
4 ThreadLocalRandom.localInit(); // 初始化线程的Probe值
5 h = ThreadLocalRandom.getProbe();
6 wasUncontended = true;
7 }
8 boolean collide = false; // True if last slot nonempty
9 for (;;) {
10 CounterCell[] as; CounterCell a; int n; long v;
11 if ((as = counterCells) != null && (n = as.length) > 0) {
12 if ((a = as[(n - 1) & h]) == null) {//当指定下标的计数器还是null
13 if (cellsBusy == 0) { // 为0则表示还没其他线程抢到锁
14 CounterCell r = new CounterCell(x); // Optimistic create
15 if (cellsBusy == 0 &&
16 U.compareAndSwapInt(this, CELLSBUSY, 0, 1)) {//设置cellsBusy标志位为1
17 boolean created = false;
18 try { // Recheck under lock
19 CounterCell[] rs; int m, j;
20 if ((rs = counterCells) != null &&
21 (m = rs.length) > 0 &&
22 rs[j = (m - 1) & h] == null) {
23 rs[j] = r;//给指定下标元素赋值一个新的CounterCell实例
24 created = true;
25 }
26 } finally {
27 cellsBusy = 0;
28 }
29 if (created)
30 break;
31 continue; // Slot is now non-empty
32 }
33 }
34 collide = false;
35 }
36 //进入下面说明指定下标的CounterCell不为空了
37 else if (!wasUncontended) // 调该方法前CAS失败
38 wasUncontended = true; // Continue after rehash
39 else if (U.compareAndSwapLong(a, CELLVALUE, v = a.value, v + x))
40 break;//赋值成功
41 else if (counterCells != as || n >= NCPU)
42 collide = false; // At max size or stale
43 else if (!collide)
44 collide = true;
45 else if (cellsBusy == 0 &&
46 U.compareAndSwapInt(this, CELLSBUSY, 0, 1)) {//到这里竞争太厉害，要扩容？
47 try {
48 if (counterCells == as) {// Expand table unless stale
49 //扩容为原来的两倍
50 CounterCell[] rs = new CounterCell[n << 1];
51 for (int i = 0; i < n; ++i)
52 rs[i] = as[i];//数据迁移
53 counterCells = rs;
54 }
55 } finally {
56 cellsBusy = 0;
57 }
58 collide = false;
59 continue; // Retry with expanded table
60 }
61 //重新随机一个probe
62 h = ThreadLocalRandom.advanceProbe(h);
63 }
64 //这里是counterCells数组还没初始化
65 else if (cellsBusy == 0 && counterCells == as &&
66 U.compareAndSwapInt(this, CELLSBUSY, 0, 1)) {
67 boolean init = false;
68 try { // Initialize table
69 if (counterCells == as) {//判断是否还是空
70 CounterCell[] rs = new CounterCell[2];//一开始只有2个
71 rs[h & 1] = new CounterCell(x);//根据最低位选一个计数器
72 counterCells = rs;
73 init = true;
74 }
75 } finally {
76 cellsBusy = 0;
77 }
78 if (init)
79 break;//直接退出了
80 }
81 else if (U.compareAndSwapLong(this, BASECOUNT, v = baseCount, v + x))
82 break; //在baseCount上计数成功
83 }
84 }
85
86
