ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/jsr166/jsr166/src/jsr166y/Phaser.java

Comparing jsr166/src/jsr166y/Phaser.java (file contents):
Revision 1.4 by dl, Sat Sep 6 13:19:17 2008 UTC vs.
Revision 1.50 by dl, Sat Nov 6 16:12:10 2010 UTC

#	Line 5 | Line 5
5		*/
6
7		package jsr166y;
8	+
9		import java.util.concurrent.*;
10	<	import java.util.concurrent.atomic.*;
10	>	import java.util.concurrent.atomic.AtomicReference;
11		import java.util.concurrent.locks.LockSupport;
11	–	import sun.misc.Unsafe;
12	–	import java.lang.reflect.*;
12
13		/**
14	<	* A reusable synchronization barrier, similar in functionality to a
15	<	* {@link java.util.concurrent.CyclicBarrier} and {@link
16	<	* java.util.concurrent.CountDownLatch} but supporting more flexible
17	<	* usage.
14	>	* A reusable synchronization barrier, similar in functionality to
15	>	* {@link java.util.concurrent.CyclicBarrier CyclicBarrier} and
16	>	* {@link java.util.concurrent.CountDownLatch CountDownLatch}
17	>	* but supporting more flexible usage.
18	>	*
19	>	* <p> <b>Registration.</b> Unlike the case for other barriers, the
20	>	* number of parties <em>registered</em> to synchronize on a phaser
21	>	* may vary over time.  Tasks may be registered at any time (using
22	>	* methods {@link #register}, {@link #bulkRegister}, or forms of
23	>	* constructors establishing initial numbers of parties), and
24	>	* optionally deregistered upon any arrival (using {@link
25	>	* #arriveAndDeregister}).  As is the case with most basic
26	>	* synchronization constructs, registration and deregistration affect
27	>	* only internal counts; they do not establish any further internal
28	>	* bookkeeping, so tasks cannot query whether they are registered.
29	>	* (However, you can introduce such bookkeeping by subclassing this
30	>	* class.)
31	>	*
32	>	* <p> <b>Synchronization.</b> Like a {@code CyclicBarrier}, a {@code
33	>	* Phaser} may be repeatedly awaited.  Method {@link
34	>	* #arriveAndAwaitAdvance} has effect analogous to {@link
35	>	* java.util.concurrent.CyclicBarrier#await CyclicBarrier.await}. Each
36	>	* generation of a {@code Phaser} has an associated phase number. The
37	>	* phase number starts at zero, and advances when all parties arrive
38	>	* at the barrier, wrapping around to zero after reaching {@code
39	>	* Integer.MAX_VALUE}. The use of phase numbers enables independent
40	>	* control of actions upon arrival at a barrier and upon awaiting
41	>	* others, via two kinds of methods that may be invoked by any
42	>	* registered party:
43		*
44		* <ul>
45		*
46	<	* <li> The number of parties synchronizing on a phaser may vary over
47	<	* time.  A task may register to be a party at any time, and may
48	<	* deregister upon arriving at the barrier.  As is the case with most
49	<	* basic synchronization constructs, registration and deregistration
50	<	* affect only internal counts; they do not establish any further
51	<	* internal bookkeeping, so tasks cannot query whether they are
52	<	* registered. (However, you can introduce such bookkeeping in by
53	<	* subclassing this class.)
54	<	*
55	<	* <li> Each generation has an associated phase value, starting at
56	<	* zero, and advancing when all parties reach the barrier (wrapping
57	<	* around to zero after reaching <tt>Integer.MAX_VALUE</tt>).
58	<	*
59	<	* <li> Like a CyclicBarrier, a Phaser may be repeatedly awaited.
60	<	* Method <tt>arriveAndAwaitAdvance</tt> has effect analogous to
61	<	* <tt>CyclicBarrier.await</tt>.  However, Phasers separate two
62	<	* aspects of coordination, that may also be invoked independently:
63	<	*
64	<	* <ul>
46	>	*   <li> <b>Arrival.</b> Methods {@link #arrive} and
47	>	*       {@link #arriveAndDeregister} record arrival at a
48	>	*       barrier. These methods do not block, but return an associated
49	>	*       <em>arrival phase number</em>; that is, the phase number of
50	>	*       the barrier to which the arrival applied. When the final
51	>	*       party for a given phase arrives, an optional barrier action
52	>	*       is performed and the phase advances.  Barrier actions,
53	>	*       performed by the party triggering a phase advance, are
54	>	*       arranged by overriding method {@link #onAdvance(int, int)},
55	>	*       which also controls termination. Overriding this method is
56	>	*       similar to, but more flexible than, providing a barrier
57	>	*       action to a {@code CyclicBarrier}.
58	>	*
59	>	*   <li> <b>Waiting.</b> Method {@link #awaitAdvance} requires an
60	>	*       argument indicating an arrival phase number, and returns when
61	>	*       the barrier advances to (or is already at) a different phase.
62	>	*       Unlike similar constructions using {@code CyclicBarrier},
63	>	*       method {@code awaitAdvance} continues to wait even if the
64	>	*       waiting thread is interrupted. Interruptible and timeout
65	>	*       versions are also available, but exceptions encountered while
66	>	*       tasks wait interruptibly or with timeout do not change the
67	>	*       state of the barrier. If necessary, you can perform any
68	>	*       associated recovery within handlers of those exceptions,
69	>	*       often after invoking {@code forceTermination}.  Phasers may
70	>	*       also be used by tasks executing in a {@link ForkJoinPool},
71	>	*       which will ensure sufficient parallelism to execute tasks
72	>	*       when others are blocked waiting for a phase to advance.
73		*
42	–	*   <li> Arriving at a barrier. Methods <tt>arrive</tt> and
43	–	*       <tt>arriveAndDeregister</tt> do not block, but return
44	–	*       the phase value current upon entry to the method.
45	–	*
46	–	*   <li> Awaiting others. Method <tt>awaitAdvance</tt> requires an
47	–	*       argument indicating the entry phase, and returns when the
48	–	*       barrier advances to a new phase.
74		* </ul>
75		*
76	<	*
77	<	* <li> Barrier actions, performed by the task triggering a phase
78	<	* advance while others may be waiting, are arranged by overriding
79	<	* method <tt>onAdvance</tt>, that also controls termination.
80	<	* Overriding this method may be used to similar but more flecible
81	<	* effect as providing a barrier action to a CyclicBarrier.
82	<	*
58	<	* <li> Phasers may enter a <em>termination</em> state in which all
59	<	* await actions immediately return, indicating (via a negative phase
60	<	* value) that execution is complete.  Termination is triggered by
61	<	* executing the overridable <tt>onAdvance</tt> method that is invoked
62	<	* each time the barrier is about to be tripped. When a Phaser is
63	<	* controlling an action with a fixed number of iterations, it is
76	>	* <p> <b>Termination.</b> A {@code Phaser} may enter a
77	>	* <em>termination</em> state in which all synchronization methods
78	>	* immediately return without updating phaser state or waiting for
79	>	* advance, and indicating (via a negative phase value) that execution
80	>	* is complete.  Termination is triggered when an invocation of {@code
81	>	* onAdvance} returns {@code true}.  As illustrated below, when
82	>	* phasers control actions with a fixed number of iterations, it is
83		* often convenient to override this method to cause termination when
84	<	* the current phase number reaches a threshold. Method
85	<	* <tt>forceTermination</tt> is also available to abruptly release
86	<	* waiting threads and allow them to terminate.
84	>	* the current phase number reaches a threshold. Method {@link
85	>	* #forceTermination} is also available to abruptly release waiting
86	>	* threads and allow them to terminate.
87		*
88	<	* <li> Phasers may be tiered to reduce contention. Phasers with large
88	>	* <p> <b>Tiering.</b> Phasers may be <em>tiered</em> (i.e., arranged
89	>	* in tree structures) to reduce contention. Phasers with large
90		* numbers of parties that would otherwise experience heavy
91	<	* synchronization contention costs may instead be arranged in trees.
92	<	* This will typically greatly increase throughput even though it
93	<	* incurs somewhat greater per-operation overhead.
94	<	*
95	<	* <li> By default, <tt>awaitAdvance</tt> continues to wait even if
96	<	* the waiting thread is interrupted. And unlike the case in
97	<	* CyclicBarriers, exceptions encountered while tasks wait
98	<	* interruptibly or with timeout do not change the state of the
99	<	* barrier. If necessary, you can perform any associated recovery
100	<	* within handlers of those exceptions, often after invoking
101	<	* <tt>forceTermination</tt>.
102	<	*
103	<	* </ul>
91	>	* synchronization contention costs may instead be set up so that
92	>	* groups of sub-phasers share a common parent.  This may greatly
93	>	* increase throughput even though it incurs greater per-operation
94	>	* overhead.
95	>	*
96	>	* <p><b>Monitoring.</b> While synchronization methods may be invoked
97	>	* only by registered parties, the current state of a phaser may be
98	>	* monitored by any caller.  At any given moment there are {@link
99	>	* #getRegisteredParties} parties in total, of which {@link
100	>	* #getArrivedParties} have arrived at the current phase ({@link
101	>	* #getPhase}).  When the remaining ({@link #getUnarrivedParties})
102	>	* parties arrive, the phase advances.  The values returned by these
103	>	* methods may reflect transient states and so are not in general
104	>	* useful for synchronization control.  Method {@link #toString}
105	>	* returns snapshots of these state queries in a form convenient for
106	>	* informal monitoring.
107		*
108		* <p><b>Sample usages:</b>
109		*
110	<	* <p>A Phaser may be used instead of a <tt>CountdownLatch</tt> to control
111	<	* a one-shot action serving a variable number of parties. The typical
112	<	* idiom is for the method setting this up to first register, then
113	<	* start the actions, then deregister, as in:
114	<	*
115	<	* <pre>
116	<	*  void runTasks(List&lt;Runnable&gt; list) {
117	<	*    final Phaser phaser = new Phaser(1); // "1" to register self
118	<	*    for (Runnable r : list) {
119	<	*      phaser.register();
120	<	*      new Thread() {
121	<	*        public void run() {
122	<	*          phaser.arriveAndAwaitAdvance(); // await all creation
123	<	*          r.run();
124	<	*          phaser.arriveAndDeregister();   // signal completion
125	<	*        }
126	<	*      }.start();
110	>	* <p>A {@code Phaser} may be used instead of a {@code CountDownLatch}
111	>	* to control a one-shot action serving a variable number of parties.
112	>	* The typical idiom is for the method setting this up to first
113	>	* register, then start the actions, then deregister, as in:
114	>	*
115	>	*  <pre> {@code
116	>	* void runTasks(List<Runnable> tasks) {
117	>	*   final Phaser phaser = new Phaser(1); // "1" to register self
118	>	*   // create and start threads
119	>	*   for (Runnable task : tasks) {
120	>	*     phaser.register();
121	>	*     new Thread() {
122	>	*       public void run() {
123	>	*         phaser.arriveAndAwaitAdvance(); // await all creation
124	>	*         task.run();
125	>	*       }
126	>	*     }.start();
127		*   }
128	<	*   phaser.arrive(); // allow threads to start
129	<	*   int p = phaser.arriveAndDeregister(); // deregister self
130	<	*   otherActions(); // do other things while tasks execute
131	<	*   phaser.awaitAdvance(p); // wait for all tasks to arrive
109	<	* }
110	<	* </pre>
128	>	*
129	>	*   // allow threads to start and deregister self
130	>	*   phaser.arriveAndDeregister();
131	>	* }}</pre>
132		*
133		* <p>One way to cause a set of threads to repeatedly perform actions
134	<	* for a given number of iterations is to override <tt>onAdvance</tt>:
134	>	* for a given number of iterations is to override {@code onAdvance}:
135		*
136	<	* <pre>
137	<	*  void startTasks(List&lt;Runnable&gt; list, final int iterations) {
138	<	*    final Phaser phaser = new Phaser() {
139	<	*       public boolean onAdvance(int phase, int registeredParties) {
140	<	*         return phase &gt;= iterations || registeredParties == 0;
136	>	*  <pre> {@code
137	>	* void startTasks(List<Runnable> tasks, final int iterations) {
138	>	*   final Phaser phaser = new Phaser() {
139	>	*     protected boolean onAdvance(int phase, int registeredParties) {
140	>	*       return phase >= iterations || registeredParties == 0;
141	>	*     }
142	>	*   };
143	>	*   phaser.register();
144	>	*   for (final Runnable task : tasks) {
145	>	*     phaser.register();
146	>	*     new Thread() {
147	>	*       public void run() {
148	>	*         do {
149	>	*           task.run();
150	>	*           phaser.arriveAndAwaitAdvance();
151	>	*         } while (!phaser.isTerminated());
152		*       }
153	<	*    };
122	<	*    phaser.register();
123	<	*    for (Runnable r : list) {
124	<	*      phaser.register();
125	<	*      new Thread() {
126	<	*        public void run() {
127	<	*           do {
128	<	*             r.run();
129	<	*             phaser.arriveAndAwaitAdvance();
130	<	*           } while(!phaser.isTerminated();
131	<	*        }
132	<	*      }.start();
153	>	*     }.start();
154		*   }
155		*   phaser.arriveAndDeregister(); // deregister self, don't wait
156	<	* }
157	<	* </pre>
156	>	* }}</pre>
157	>	*
158	>	* If the main task must later await termination, it
159	>	* may re-register and then execute a similar loop:
160	>	*  <pre> {@code
161	>	*   // ...
162	>	*   phaser.register();
163	>	*   while (!phaser.isTerminated())
164	>	*     phaser.arriveAndAwaitAdvance();}</pre>
165	>	*
166	>	* <p>Related constructions may be used to await particular phase numbers
167	>	* in contexts where you are sure that the phase will never wrap around
168	>	* {@code Integer.MAX_VALUE}. For example:
169	>	*
170	>	*  <pre> {@code
171	>	* void awaitPhase(Phaser phaser, int phase) {
172	>	*   int p = phaser.register(); // assumes caller not already registered
173	>	*   while (p < phase) {
174	>	*     if (phaser.isTerminated())
175	>	*       // ... deal with unexpected termination
176	>	*     else
177	>	*       p = phaser.arriveAndAwaitAdvance();
178	>	*   }
179	>	*   phaser.arriveAndDeregister();
180	>	* }}</pre>
181		*
182	<	* <p> To create a set of tasks using a tree of Phasers,
182	>	*
183	>	* <p>To create a set of tasks using a tree of phasers,
184		* you could use code of the following form, assuming a
185	<	* Task class with a constructor accepting a Phaser that
186	<	* it registers for upon construction:
142	<	* <pre>
143	<	*  void build(Task[] actions, int lo, int hi, Phaser b) {
144	<	*    int step = (hi - lo) / TASKS_PER_PHASER;
145	<	*    if (step &gt; 1) {
146	<	*       int i = lo;
147	<	*       while (i &lt; hi) {
148	<	*         int r = Math.min(i + step, hi);
149	<	*         build(actions, i, r, new Phaser(b));
150	<	*         i = r;
151	<	*       }
152	<	*    }
153	<	*    else {
154	<	*      for (int i = lo; i &lt; hi; ++i)
155	<	*        actions[i] = new Task(b);
156	<	*        // assumes new Task(b) performs b.register()
157	<	*    }
158	<	*  }
159	<	*  // .. initially called, for n tasks via
160	<	*  build(new Task[n], 0, n, new Phaser());
161	<	* </pre>
185	>	* Task class with a constructor accepting a phaser that
186	>	* it registers with upon construction:
187		*
188	<	* The best value of <tt>TASKS_PER_PHASER</tt> depends mainly on
188	>	*  <pre> {@code
189	>	* void build(Task[] actions, int lo, int hi, Phaser ph) {
190	>	*   if (hi - lo > TASKS_PER_PHASER) {
191	>	*     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
192	>	*       int j = Math.min(i + TASKS_PER_PHASER, hi);
193	>	*       build(actions, i, j, new Phaser(ph));
194	>	*     }
195	>	*   } else {
196	>	*     for (int i = lo; i < hi; ++i)
197	>	*       actions[i] = new Task(ph);
198	>	*       // assumes new Task(ph) performs ph.register()
199	>	*   }
200	>	* }
201	>	* // .. initially called, for n tasks via
202	>	* build(new Task[n], 0, n, new Phaser());}</pre>
203	>	*
204	>	* The best value of {@code TASKS_PER_PHASER} depends mainly on
205		* expected barrier synchronization rates. A value as low as four may
206		* be appropriate for extremely small per-barrier task bodies (thus
207		* high rates), or up to hundreds for extremely large ones.
208		*
168	–	* </pre>
169	–	*
209		* <p><b>Implementation notes</b>: This implementation restricts the
210		* maximum number of parties to 65535. Attempts to register additional
211	<	* parties result in IllegalStateExceptions. However, you can and
211	>	* parties result in {@code IllegalStateException}. However, you can and
212		* should create tiered phasers to accommodate arbitrarily large sets
213		* of participants.
214	+	*
215	+	* @since 1.7
216	+	* @author Doug Lea
217		*/
218		public class Phaser {
219		/*
#	Line 192 | Line 234 | public class Phaser {
234		* However, to efficiently maintain atomicity, these values are
235		* packed into a single (atomic) long. Termination uses the sign
236		* bit of 32 bit representation of phase, so phase is set to -1 on
237	<	* termination. Good performace relies on keeping state decoding
237	>	* termination. Good performance relies on keeping state decoding
238		* and encoding simple, and keeping race windows short.
239		*
240		* Note: there are some cheats in arrive() that rely on unarrived
241	<	* being lowest 16 bits.
241	>	* count being lowest 16 bits.
242		*/
243		private volatile long state;
244
245	<	private static final int ushortBits = 16;
246	<	private static final int ushortMask =  (1 << ushortBits) - 1;
205	<	private static final int phaseMask = 0x7fffffff;
245	>	private static final int ushortMask = 0xffff;
246	>	private static final int phaseMask  = 0x7fffffff;
247
248		private static int unarrivedOf(long s) {
249	<	return (int)(s & ushortMask);
249	>	return (int) (s & ushortMask);
250		}
251
252		private static int partiesOf(long s) {
253	<	return (int)(s & (ushortMask << 16)) >>> 16;
253	>	return ((int) s) >>> 16;
254		}
255
256		private static int phaseOf(long s) {
257	<	return (int)(s >>> 32);
257	>	return (int) (s >>> 32);
258		}
259
260		private static int arrivedOf(long s) {
#	Line 221 | Line 262 | public class Phaser {
262		}
263
264		private static long stateFor(int phase, int parties, int unarrived) {
265	<	return (((long)phase) << 32) | ((parties << 16) | unarrived);
265	>	return ((((long) phase) << 32) | (((long) parties) << 16) |
266	>	(long) unarrived);
267		}
268
269		private static long trippedStateFor(int phase, int parties) {
270	<	return (((long)phase) << 32) | ((parties << 16) | parties);
270	>	long lp = (long) parties;
271	>	return (((long) phase) << 32) | (lp << 16) | lp;
272		}
273
274	<	private static IllegalStateException badBounds(int parties, int unarrived) {
275	<	return new IllegalStateException
276	<	("Attempt to set " + unarrived +
277	<	" unarrived of " + parties + " parties");
274	>	/**
275	>	* Returns message string for bad bounds exceptions.
276	>	*/
277	>	private static String badBounds(int parties, int unarrived) {
278	>	return ("Attempt to set " + unarrived +
279	>	" unarrived of " + parties + " parties");
280		}
281
282		/**
#	Line 240 | Line 285 | public class Phaser {
285		private final Phaser parent;
286
287		/**
288	<	* The root of Phaser tree. Equals this if not in a tree.  Used to
288	>	* The root of phaser tree. Equals this if not in a tree.  Used to
289		* support faster state push-down.
290		*/
291		private final Phaser root;
#	Line 248 | Line 293 | public class Phaser {
293		// Wait queues
294
295		/**
296	<	* Heads of Treiber stacks waiting for nonFJ threads. To eliminate
297	<	* contention while releasing some threads while adding others, we
296	>	* Heads of Treiber stacks for waiting threads. To eliminate
297	>	* contention when releasing some threads while adding others, we
298		* use two of them, alternating across even and odd phases.
299	+	* Subphasers share queues with root to speed up releases.
300		*/
301	<	private final AtomicReference<QNode> evenQ = new AtomicReference<QNode>();
302	<	private final AtomicReference<QNode> oddQ  = new AtomicReference<QNode>();
301	>	private final AtomicReference<QNode> evenQ;
302	>	private final AtomicReference<QNode> oddQ;
303
304		private AtomicReference<QNode> queueFor(int phase) {
305	<	return (phase & 1) == 0? evenQ : oddQ;
305	>	return ((phase & 1) == 0) ? evenQ : oddQ;
306		}
307
308		/**
#	Line 264 | Line 310 | public class Phaser {
310		* root if necessary.
311		*/
312		private long getReconciledState() {
313	<	return parent == null? state : reconcileState();
313	>	return (parent == null) ? state : reconcileState();
314		}
315
316		/**
317		* Recursively resolves state.
318		*/
319		private long reconcileState() {
320	<	Phaser p = parent;
320	>	Phaser par = parent;
321		long s = state;
322	<	if (p != null) {
323	<	while (unarrivedOf(s) == 0 && phaseOf(s) != phaseOf(root.state)) {
324	<	long parentState = p.getReconciledState();
325	<	int parentPhase = phaseOf(parentState);
326	<	int phase = phaseOf(s = state);
327	<	if (phase != parentPhase) {
322	>	if (par != null) {
323	>	int phase, rootPhase;
324	>	while ((phase = phaseOf(s)) >= 0 &&
325	>	(rootPhase = phaseOf(root.state)) != phase &&
326	>	(rootPhase < 0 || unarrivedOf(s) == 0)) {
327	>	int parentPhase = phaseOf(par.getReconciledState());
328	>	if (parentPhase != phase) {
329		long next = trippedStateFor(parentPhase, partiesOf(s));
330	<	if (casState(s, next)) {
331	<	releaseWaiters(phase);
285	<	s = next;
286	<	}
330	>	if (state == s)
331	>	UNSAFE.compareAndSwapLong(this, stateOffset, s, next);
332		}
333	+	s = state;
334		}
335		}
336		return s;
337		}
338
339		/**
340	<	* Creates a new Phaser without any initially registered parties,
341	<	* initial phase number 0, and no parent.
340	>	* Creates a new phaser without any initially registered parties,
341	>	* initial phase number 0, and no parent. Any thread using this
342	>	* phaser will need to first register for it.
343		*/
344		public Phaser() {
345	<	this(null);
345	>	this(null, 0);
346		}
347
348		/**
349	<	* Creates a new Phaser with the given numbers of registered
349	>	* Creates a new phaser with the given number of registered
350		* unarrived parties, initial phase number 0, and no parent.
351	<	* @param parties the number of parties required to trip barrier.
351	>	*
352	>	* @param parties the number of parties required to trip barrier
353		* @throws IllegalArgumentException if parties less than zero
354	<	* or greater than the maximum number of parties supported.
354	>	* or greater than the maximum number of parties supported
355		*/
356		public Phaser(int parties) {
357		this(null, parties);
358		}
359
360		/**
361	<	* Creates a new Phaser with the given parent, without any
361	>	* Creates a new phaser with the given parent, without any
362		* initially registered parties. If parent is non-null this phaser
363		* is registered with the parent and its initial phase number is
364		* the same as that of parent phaser.
365	<	* @param parent the parent phaser.
365	>	*
366	>	* @param parent the parent phaser
367		*/
368		public Phaser(Phaser parent) {
369	<	int phase = 0;
321	<	this.parent = parent;
322	<	if (parent != null) {
323	<	this.root = parent.root;
324	<	phase = parent.register();
325	<	}
326	<	else
327	<	this.root = this;
328	<	this.state = trippedStateFor(phase, 0);
369	>	this(parent, 0);
370		}
371
372		/**
373	<	* Creates a new Phaser with the given parent and numbers of
374	<	* registered unarrived parties. If parent is non-null this phaser
373	>	* Creates a new phaser with the given parent and number of
374	>	* registered unarrived parties. If parent is non-null, this phaser
375		* is registered with the parent and its initial phase number is
376		* the same as that of parent phaser.
377	<	* @param parent the parent phaser.
378	<	* @param parties the number of parties required to trip barrier.
377	>	*
378	>	* @param parent the parent phaser
379	>	* @param parties the number of parties required to trip barrier
380		* @throws IllegalArgumentException if parties less than zero
381	<	* or greater than the maximum number of parties supported.
381	>	* or greater than the maximum number of parties supported
382		*/
383		public Phaser(Phaser parent, int parties) {
384		if (parties < 0 || parties > ushortMask)
385		throw new IllegalArgumentException("Illegal number of parties");
386	<	int phase = 0;
386	>	int phase;
387		this.parent = parent;
388		if (parent != null) {
389	<	this.root = parent.root;
389	>	Phaser r = parent.root;
390	>	this.root = r;
391	>	this.evenQ = r.evenQ;
392	>	this.oddQ = r.oddQ;
393		phase = parent.register();
394		}
395	<	else
395	>	else {
396		this.root = this;
397	+	this.evenQ = new AtomicReference<QNode>();
398	+	this.oddQ = new AtomicReference<QNode>();
399	+	phase = 0;
400	+	}
401		this.state = trippedStateFor(phase, parties);
402		}
403
404		/**
405		* Adds a new unarrived party to this phaser.
406	<	* @return the current barrier phase number upon registration
406	>	* If an ongoing invocation of {@link #onAdvance} is in progress,
407	>	* this method may wait until its completion before registering.
408	>	*
409	>	* @return the arrival phase number to which this registration applied
410		* @throws IllegalStateException if attempting to register more
411	<	* than the maximum supported number of parties.
411	>	* than the maximum supported number of parties
412		*/
413		public int register() {
414		return doRegister(1);
#	Line 364 | Line 416 | public class Phaser {
416
417		/**
418		* Adds the given number of new unarrived parties to this phaser.
419	<	* @param parties the number of parties required to trip barrier.
420	<	* @return the current barrier phase number upon registration
419	>	* If an ongoing invocation of {@link #onAdvance} is in progress,
420	>	* this method may wait until its completion before registering.
421	>	*
422	>	* @param parties the number of additional parties required to trip barrier
423	>	* @return the arrival phase number to which this registration applied
424		* @throws IllegalStateException if attempting to register more
425	<	* than the maximum supported number of parties.
425	>	* than the maximum supported number of parties
426	>	* @throws IllegalArgumentException if {@code parties < 0}
427		*/
428		public int bulkRegister(int parties) {
429		if (parties < 0)
#	Line 381 | Line 437 | public class Phaser {
437		* Shared code for register, bulkRegister
438		*/
439		private int doRegister(int registrations) {
440	+	Phaser par = parent;
441	+	long s;
442		int phase;
443	<	for (;;) {
444	<	long s = getReconciledState();
445	<	phase = phaseOf(s);
446	<	int unarrived = unarrivedOf(s) + registrations;
447	<	int parties = partiesOf(s) + registrations;
448	<	if (phase < 0)
449	<	break;
450	<	if (parties > ushortMask || unarrived > ushortMask)
451	<	throw badBounds(parties, unarrived);
452	<	if (phase == phaseOf(root.state) &&
453	<	casState(s, stateFor(phase, parties, unarrived)))
454	<	break;
443	>	while ((phase = phaseOf(s = par==null? state:reconcileState())) >= 0) {
444	>	int p = partiesOf(s);
445	>	int u = unarrivedOf(s);
446	>	int unarrived = u + registrations;
447	>	int parties = p + registrations;
448	>	if (u == 0 && p != 0)  // if tripped, wait for advance
449	>	untimedWait(phase);
450	>	else if (parties > ushortMask)
451	>	throw new IllegalStateException(badBounds(parties, unarrived));
452	>	else if (par == null || phaseOf(root.state) == phase) {
453	>	long next = stateFor(phase, parties, unarrived);
454	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next))
455	>	break;
456	>	}
457		}
458		return phase;
459		}
460
461		/**
462		* Arrives at the barrier, but does not wait for others.  (You can
463	<	* in turn wait for others via {@link #awaitAdvance}).
463	>	* in turn wait for others via {@link #awaitAdvance}).  It is an
464	>	* unenforced usage error for an unregistered party to invoke this
465	>	* method.
466		*
467	<	* @return the barrier phase number upon entry to this method, or a
406	<	* negative value if terminated;
467	>	* @return the arrival phase number, or a negative value if terminated
468		* @throws IllegalStateException if not terminated and the number
469	<	* of unarrived parties would become negative.
469	>	* of unarrived parties would become negative
470		*/
471		public int arrive() {
472	+	Phaser par = parent;
473	+	long s;
474		int phase;
475	<	for (;;) {
413	<	long s = state;
414	<	phase = phaseOf(s);
475	>	while ((phase = phaseOf(s = par==null? state:reconcileState())) >= 0) {
476		int parties = partiesOf(s);
477		int unarrived = unarrivedOf(s) - 1;
478	<	if (unarrived > 0) {        // Not the last arrival
479	<	if (casState(s, s - 1)) // s-1 adds one arrival
480	<	break;
478	>	if (unarrived > 0) {                // Not the last arrival
479	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s - 1))
480	>	break;                      // s-1 adds one arrival
481		}
482	<	else if (unarrived == 0) {  // the last arrival
483	<	Phaser par = parent;
484	<	if (par == null) {      // directly trip
485	<	if (casState
486	<	(s,
487	<	trippedStateFor(onAdvance(phase, parties)? -1 :
488	<	((phase + 1) & phaseMask), parties))) {
489	<	releaseWaiters(phase);
490	<	break;
430	<	}
431	<	}
432	<	else {                  // cascade to parent
433	<	if (casState(s, s - 1)) { // zeroes unarrived
434	<	par.arrive();
435	<	reconcileState();
436	<	break;
437	<	}
482	>	else if (unarrived < 0)
483	>	throw new IllegalStateException(badBounds(parties, unarrived));
484	>	else if (par == null) {             // directly trip
485	>	long next = trippedStateFor(onAdvance(phase, parties) ? -1 :
486	>	((phase + 1) & phaseMask),
487	>	parties);
488	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next)) {
489	>	releaseWaiters(phase);
490	>	break;
491		}
492		}
493	<	else if (phase < 0) // Don't throw exception if terminated
494	<	break;
495	<	else if (phase != phaseOf(root.state)) // or if unreconciled
493	>	else if (phaseOf(root.state) == phase &&
494	>	UNSAFE.compareAndSwapLong(this, stateOffset, s, s - 1)) {
495	>	par.arrive();                   // cascade to parent
496		reconcileState();
497	<	else
498	<	throw badBounds(parties, unarrived);
497	>	break;
498	>	}
499		}
500		return phase;
501		}
502
503		/**
504	<	* Arrives at the barrier, and deregisters from it, without
505	<	* waiting for others. Deregistration reduces number of parties
504	>	* Arrives at the barrier and deregisters from it without waiting
505	>	* for others. Deregistration reduces the number of parties
506		* required to trip the barrier in future phases.  If this phaser
507		* has a parent, and deregistration causes this phaser to have
508	<	* zero parties, this phaser is also deregistered from its parent.
508	>	* zero parties, this phaser also arrives at and is deregistered
509	>	* from its parent.  It is an unenforced usage error for an
510	>	* unregistered party to invoke this method.
511		*
512	<	* @return the current barrier phase number upon entry to
458	<	* this method, or a negative value if terminated;
512	>	* @return the arrival phase number, or a negative value if terminated
513		* @throws IllegalStateException if not terminated and the number
514	<	* of registered or unarrived parties would become negative.
514	>	* of registered or unarrived parties would become negative
515		*/
516		public int arriveAndDeregister() {
517	<	// similar code to arrive, but too different to merge
517	>	// similar to arrive, but too different to merge
518		Phaser par = parent;
519	+	long s;
520		int phase;
521	<	for (;;) {
467	<	long s = state;
468	<	phase = phaseOf(s);
521	>	while ((phase = phaseOf(s = par==null? state:reconcileState())) >= 0) {
522		int parties = partiesOf(s) - 1;
523		int unarrived = unarrivedOf(s) - 1;
524	<	if (parties >= 0) {
525	<	if (unarrived > 0 || (unarrived == 0 && par != null)) {
526	<	if (casState
474	<	(s,
475	<	stateFor(phase, parties, unarrived))) {
476	<	if (unarrived == 0) {
477	<	par.arriveAndDeregister();
478	<	reconcileState();
479	<	}
480	<	break;
481	<	}
482	<	continue;
483	<	}
484	<	if (unarrived == 0) {
485	<	if (casState
486	<	(s,
487	<	trippedStateFor(onAdvance(phase, parties)? -1 :
488	<	((phase + 1) & phaseMask), parties))) {
489	<	releaseWaiters(phase);
490	<	break;
491	<	}
492	<	continue;
493	<	}
494	<	if (phase < 0)
524	>	if (unarrived > 0) {
525	>	long next = stateFor(phase, parties, unarrived);
526	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next))
527		break;
528	<	if (par != null && phase != phaseOf(root.state)) {
528	>	}
529	>	else if (unarrived < 0)
530	>	throw new IllegalStateException(badBounds(parties, unarrived));
531	>	else if (par == null) {
532	>	long next = trippedStateFor(onAdvance(phase, parties)? -1:
533	>	(phase + 1) & phaseMask,
534	>	parties);
535	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next)) {
536	>	releaseWaiters(phase);
537	>	break;
538	>	}
539	>	}
540	>	else if (phaseOf(root.state) == phase) {
541	>	long next = stateFor(phase, parties, 0);
542	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next)) {
543	>	if (parties == 0)
544	>	par.arriveAndDeregister();
545	>	else
546	>	par.arrive();
547		reconcileState();
548	<	continue;
548	>	break;
549		}
550		}
501	–	throw badBounds(parties, unarrived);
551		}
552		return phase;
553		}
554
555		/**
556		* Arrives at the barrier and awaits others. Equivalent in effect
557	<	* to <tt>awaitAdvance(arrive())</tt>.  If you instead need to
558	<	* await with interruption of timeout, and/or deregister upon
559	<	* arrival, you can arrange them using analogous constructions.
560	<	* @return the phase on entry to this method
557	>	* to {@code awaitAdvance(arrive())}.  If you need to await with
558	>	* interruption or timeout, you can arrange this with an analogous
559	>	* construction using one of the other forms of the {@code
560	>	* awaitAdvance} method.  If instead you need to deregister upon
561	>	* arrival, use {@link #arriveAndDeregister}. It is an unenforced
562	>	* usage error for an unregistered party to invoke this method.
563	>	*
564	>	* @return the arrival phase number, or a negative number if terminated
565		* @throws IllegalStateException if not terminated and the number
566	<	* of unarrived parties would become negative.
566	>	* of unarrived parties would become negative
567		*/
568		public int arriveAndAwaitAdvance() {
569		return awaitAdvance(arrive());
570		}
571
572		/**
573	<	* Awaits the phase of the barrier to advance from the given
574	<	* value, or returns immediately if argument is negative or this
575	<	* barrier is terminated.
576	<	* @param phase the phase on entry to this method
577	<	* @return the phase on exit from this method
573	>	* Awaits the phase of the barrier to advance from the given phase
574	>	* value, returning immediately if the current phase of the
575	>	* barrier is not equal to the given phase value or this barrier
576	>	* is terminated.
577	>	*
578	>	* @param phase an arrival phase number, or negative value if
579	>	* terminated; this argument is normally the value returned by a
580	>	* previous call to {@code arrive} or its variants
581	>	* @return the next arrival phase number, or a negative value
582	>	* if terminated or argument is negative
583		*/
584		public int awaitAdvance(int phase) {
585		if (phase < 0)
586		return phase;
587	<	long s = getReconciledState();
530	<	int p = phaseOf(s);
587	>	int p = getPhase();
588		if (p != phase)
589		return p;
533	–	if (unarrivedOf(s) == 0)
534	–	parent.awaitAdvance(phase);
535	–	// Fall here even if parent waited, to reconcile and help release
590		return untimedWait(phase);
591		}
592
593		/**
594	<	* Awaits the phase of the barrier to advance from the given
595	<	* value, or returns immediately if argumet is negative or this
596	<	* barrier is terminated, or throws InterruptedException if
597	<	* interrupted while waiting.
598	<	* @param phase the phase on entry to this method
599	<	* @return the phase on exit from this method
594	>	* Awaits the phase of the barrier to advance from the given phase
595	>	* value, throwing {@code InterruptedException} if interrupted
596	>	* while waiting, or returning immediately if the current phase of
597	>	* the barrier is not equal to the given phase value or this
598	>	* barrier is terminated.
599	>	*
600	>	* @param phase an arrival phase number, or negative value if
601	>	* terminated; this argument is normally the value returned by a
602	>	* previous call to {@code arrive} or its variants
603	>	* @return the next arrival phase number, or a negative value
604	>	* if terminated or argument is negative
605		* @throws InterruptedException if thread interrupted while waiting
606		*/
607	<	public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
607	>	public int awaitAdvanceInterruptibly(int phase)
608	>	throws InterruptedException {
609		if (phase < 0)
610		return phase;
611	<	long s = getReconciledState();
552	<	int p = phaseOf(s);
611	>	int p = getPhase();
612		if (p != phase)
613		return p;
555	–	if (unarrivedOf(s) != 0)
556	–	parent.awaitAdvanceInterruptibly(phase);
614		return interruptibleWait(phase);
615		}
616
617		/**
618	<	* Awaits the phase of the barrier to advance from the given value
619	<	* or the given timeout elapses, or returns immediately if
620	<	* argument is negative or this barrier is terminated.
621	<	* @param phase the phase on entry to this method
622	<	* @return the phase on exit from this method
618	>	* Awaits the phase of the barrier to advance from the given phase
619	>	* value or the given timeout to elapse, throwing {@code
620	>	* InterruptedException} if interrupted while waiting, or
621	>	* returning immediately if the current phase of the barrier is
622	>	* not equal to the given phase value or this barrier is
623	>	* terminated.
624	>	*
625	>	* @param phase an arrival phase number, or negative value if
626	>	* terminated; this argument is normally the value returned by a
627	>	* previous call to {@code arrive} or its variants
628	>	* @param timeout how long to wait before giving up, in units of
629	>	*        {@code unit}
630	>	* @param unit a {@code TimeUnit} determining how to interpret the
631	>	*        {@code timeout} parameter
632	>	* @return the next arrival phase number, or a negative value
633	>	* if terminated or argument is negative
634		* @throws InterruptedException if thread interrupted while waiting
635		* @throws TimeoutException if timed out while waiting
636		*/
637	<	public int awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)
637	>	public int awaitAdvanceInterruptibly(int phase,
638	>	long timeout, TimeUnit unit)
639		throws InterruptedException, TimeoutException {
640	+	long nanos = unit.toNanos(timeout);
641		if (phase < 0)
642		return phase;
643	<	long s = getReconciledState();
574	<	int p = phaseOf(s);
643	>	int p = getPhase();
644		if (p != phase)
645		return p;
646	<	if (unarrivedOf(s) == 0)
578	<	parent.awaitAdvanceInterruptibly(phase, timeout, unit);
579	<	return timedWait(phase, unit.toNanos(timeout));
646	>	return timedWait(phase, nanos);
647		}
648
649		/**
#	Line 587 | Line 654 | public class Phaser {
654		* unexpected exceptions.
655		*/
656		public void forceTermination() {
657	<	for (;;) {
658	<	long s = getReconciledState();
659	<	int phase = phaseOf(s);
660	<	int parties = partiesOf(s);
661	<	int unarrived = unarrivedOf(s);
662	<	if (phase < 0 ||
663	<	casState(s, stateFor(-1, parties, unarrived))) {
664	<	releaseWaiters(0);
665	<	releaseWaiters(1);
599	<	if (parent != null)
600	<	parent.forceTermination();
601	<	return;
602	<	}
603	<	}
657	>	Phaser r = root;    // force at root then reconcile
658	>	long s;
659	>	while (phaseOf(s = r.state) >= 0)
660	>	UNSAFE.compareAndSwapLong(r, stateOffset, s,
661	>	stateFor(-1, partiesOf(s),
662	>	unarrivedOf(s)));
663	>	reconcileState();
664	>	releaseWaiters(0);  // ensure wakeups on both queues
665	>	releaseWaiters(1);
666		}
667
668		/**
669		* Returns the current phase number. The maximum phase number is
670	<	* <tt>Integer.MAX_VALUE</tt>, after which it restarts at
670	>	* {@code Integer.MAX_VALUE}, after which it restarts at
671		* zero. Upon termination, the phase number is negative.
672	+	*
673		* @return the phase number, or a negative value if terminated
674		*/
675		public final int getPhase() {
#	Line 614 | Line 677 | public class Phaser {
677		}
678
679		/**
617	–	* Returns true if the current phase number equals the given phase.
618	–	* @param phase the phase
619	–	* @return true if the current phase number equals the given phase.
620	–	*/
621	–	public final boolean hasPhase(int phase) {
622	–	return phaseOf(getReconciledState()) == phase;
623	–	}
624	–
625	–	/**
680		* Returns the number of parties registered at this barrier.
681	+	*
682		* @return the number of parties
683		*/
684		public int getRegisteredParties() {
685	<	return partiesOf(state);
685	>	return partiesOf(getReconciledState());
686		}
687
688		/**
689	<	* Returns the number of parties that have arrived at the current
690	<	* phase of this barrier.
689	>	* Returns the number of registered parties that have arrived at
690	>	* the current phase of this barrier.
691	>	*
692		* @return the number of arrived parties
693		*/
694		public int getArrivedParties() {
695	<	return arrivedOf(state);
695	>	return arrivedOf(getReconciledState());
696		}
697
698		/**
699		* Returns the number of registered parties that have not yet
700		* arrived at the current phase of this barrier.
701	+	*
702		* @return the number of unarrived parties
703		*/
704		public int getUnarrivedParties() {
705	<	return unarrivedOf(state);
705	>	return unarrivedOf(getReconciledState());
706		}
707
708		/**
709	<	* Returns the parent of this phaser, or null if none.
710	<	* @return the parent of this phaser, or null if none.
709	>	* Returns the parent of this phaser, or {@code null} if none.
710	>	*
711	>	* @return the parent of this phaser, or {@code null} if none
712		*/
713		public Phaser getParent() {
714		return parent;
#	Line 659 | Line 717 | public class Phaser {
717		/**
718		* Returns the root ancestor of this phaser, which is the same as
719		* this phaser if it has no parent.
720	<	* @return the root ancestor of this phaser.
720	>	*
721	>	* @return the root ancestor of this phaser
722		*/
723		public Phaser getRoot() {
724		return root;
725		}
726
727		/**
728	<	* Returns true if this barrier has been terminated.
729	<	* @return true if this barrier has been terminated
728	>	* Returns {@code true} if this barrier has been terminated.
729	>	*
730	>	* @return {@code true} if this barrier has been terminated
731		*/
732		public boolean isTerminated() {
733		return getPhase() < 0;
734		}
735
736		/**
737	<	* Overridable method to perform an action upon phase advance, and
738	<	* to control termination. This method is invoked whenever the
739	<	* barrier is tripped (and thus all other waiting parties are
740	<	* dormant). If it returns true, then, rather than advance the
741	<	* phase number, this barrier will be set to a final termination
742	<	* state, and subsequent calls to <tt>isTerminated</tt> will
743	<	* return true.
737	>	* Overridable method to perform an action upon impending phase
738	>	* advance, and to control termination. This method is invoked
739	>	* upon arrival of the party tripping the barrier (when all other
740	>	* waiting parties are dormant).  If this method returns {@code
741	>	* true}, then, rather than advance the phase number, this barrier
742	>	* will be set to a final termination state, and subsequent calls
743	>	* to {@link #isTerminated} will return true. Any (unchecked)
744	>	* Exception or Error thrown by an invocation of this method is
745	>	* propagated to the party attempting to trip the barrier, in
746	>	* which case no advance occurs.
747	>	*
748	>	* <p>The arguments to this method provide the state of the phaser
749	>	* prevailing for the current transition.  The results and effects
750	>	* of invoking phase-related methods (including {@code getPhase}
751	>	* as well as arrival, registration, and waiting methods) from
752	>	* within {@code onAdvance} are unspecified and should not be
753	>	* relied on. Similarly, while it is possible to override this
754	>	* method to produce side-effects visible to participating tasks,
755	>	* it is in general safe to do so only in designs in which all
756	>	* parties register before any arrive, and all {@link
757	>	* #awaitAdvance} at each phase.
758		*
759	<	* <p> The default version returns true when the number of
759	>	* <p>The default version returns {@code true} when the number of
760		* registered parties is zero. Normally, overrides that arrange
761		* termination for other reasons should also preserve this
762		* property.
763		*
690	–	* <p> You may override this method to perform an action with side
691	–	* effects visible to participating tasks, but it is in general
692	–	* only sensible to do so in designs where all parties register
693	–	* before any arrive, and all <tt>awaitAdvance</tt> at each phase.
694	–	* Otherwise, you cannot ensure lack of interference. In
695	–	* particular, this method may be invoked more than once per
696	–	* transition if other parties successfully register while the
697	–	* invocation of this method is in progress, thus postponing the
698	–	* transition until those parties also arrive, re-triggering this
699	–	* method.
700	–	*
764		* @param phase the phase number on entering the barrier
765	<	* @param registeredParties the current number of registered
766	<	* parties.
704	<	* @return true if this barrier should terminate
765	>	* @param registeredParties the current number of registered parties
766	>	* @return {@code true} if this barrier should terminate
767		*/
768		protected boolean onAdvance(int phase, int registeredParties) {
769		return registeredParties <= 0;
#	Line 710 | Line 772 | public class Phaser {
772		/**
773		* Returns a string identifying this phaser, as well as its
774		* state.  The state, in brackets, includes the String {@code
775	<	* "phase ="} followed by the phase number, {@code "parties ="}
775	>	* "phase = "} followed by the phase number, {@code "parties = "}
776		* followed by the number of registered parties, and {@code
777	<	* "arrived ="} followed by the number of arrived parties
777	>	* "arrived = "} followed by the number of arrived parties.
778		*
779		* @return a string identifying this barrier, as well as its state
780		*/
781		public String toString() {
782		long s = getReconciledState();
783	<	return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
783	>	return super.toString() +
784	>	"[phase = " + phaseOf(s) +
785	>	" parties = " + partiesOf(s) +
786	>	" arrived = " + arrivedOf(s) + "]";
787		}
788
789		// methods for waiting
790
726	–	/** The number of CPUs, for spin control */
727	–	static final int NCPUS = Runtime.getRuntime().availableProcessors();
728	–
791		/**
792	<	* The number of times to spin before blocking in timed waits.
731	<	* The value is empirically derived.
792	>	* Wait nodes for Treiber stack representing wait queue
793		*/
794	<	static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
794	>	static final class QNode implements ForkJoinPool.ManagedBlocker {
795	>	final Phaser phaser;
796	>	final int phase;
797	>	final long startTime;
798	>	final long nanos;
799	>	final boolean timed;
800	>	final boolean interruptible;
801	>	volatile boolean wasInterrupted = false;
802	>	volatile Thread thread; // nulled to cancel wait
803	>	QNode next;
804
805	<	/**
806	<	* The number of times to spin before blocking in untimed waits.
807	<	* This is greater than timed value because untimed waits spin
808	<	* faster since they don't need to check times on each spin.
809	<	*/
810	<	static final int maxUntimedSpins = maxTimedSpins * 32;
805	>	QNode(Phaser phaser, int phase, boolean interruptible,
806	>	boolean timed, long startTime, long nanos) {
807	>	this.phaser = phaser;
808	>	this.phase = phase;
809	>	this.timed = timed;
810	>	this.interruptible = interruptible;
811	>	this.startTime = startTime;
812	>	this.nanos = nanos;
813	>	thread = Thread.currentThread();
814	>	}
815
816	<	/**
817	<	* The number of nanoseconds for which it is faster to spin
818	<	* rather than to use timed park. A rough estimate suffices.
819	<	*/
820	<	static final long spinForTimeoutThreshold = 1000L;
816	>	public boolean isReleasable() {
817	>	return (thread == null ||
818	>	phaser.getPhase() != phase ||
819	>	(interruptible && wasInterrupted) ||
820	>	(timed && (nanos - (System.nanoTime() - startTime)) <= 0));
821	>	}
822
823	<	/**
824	<	* Wait nodes for Treiber stack representing wait queue for non-FJ
825	<	* tasks.
826	<	*/
827	<	static final class QNode {
828	<	QNode next;
829	<	volatile Thread thread; // nulled to cancel wait
830	<	QNode() {
831	<	thread = Thread.currentThread();
823	>	public boolean block() {
824	>	if (Thread.interrupted()) {
825	>	wasInterrupted = true;
826	>	if (interruptible)
827	>	return true;
828	>	}
829	>	if (!timed)
830	>	LockSupport.park(this);
831	>	else {
832	>	long waitTime = nanos - (System.nanoTime() - startTime);
833	>	if (waitTime <= 0)
834	>	return true;
835	>	LockSupport.parkNanos(this, waitTime);
836	>	}
837	>	return isReleasable();
838		}
839	+
840		void signal() {
841		Thread t = thread;
842		if (t != null) {
#	Line 762 | Line 844 | public class Phaser {
844		LockSupport.unpark(t);
845		}
846		}
847	+
848	+	boolean doWait() {
849	+	if (thread != null) {
850	+	try {
851	+	ForkJoinPool.managedBlock(this);
852	+	} catch (InterruptedException ie) {
853	+	wasInterrupted = true; // can't currently happen
854	+	}
855	+	}
856	+	return wasInterrupted;
857	+	}
858		}
859
860		/**
861	<	* Removes and signals waiting threads from wait queue
861	>	* Removes and signals waiting threads from wait queue.
862		*/
863		private void releaseWaiters(int phase) {
864		AtomicReference<QNode> head = queueFor(phase);
#	Line 777 | Line 870 | public class Phaser {
870		}
871
872		/**
873	+	* Tries to enqueue given node in the appropriate wait queue.
874	+	*
875	+	* @return true if successful
876	+	*/
877	+	private boolean tryEnqueue(QNode node) {
878	+	AtomicReference<QNode> head = queueFor(node.phase);
879	+	return head.compareAndSet(node.next = head.get(), node);
880	+	}
881	+
882	+	/**
883	+	* The number of times to spin before blocking waiting for advance.
884	+	*/
885	+	static final int MAX_SPINS =
886	+	Runtime.getRuntime().availableProcessors() == 1 ? 0 : 1 << 8;
887	+
888	+	/**
889		* Enqueues node and waits unless aborted or signalled.
890	+	*
891	+	* @return current phase
892		*/
893		private int untimedWait(int phase) {
783	–	int spins = maxUntimedSpins;
894		QNode node = null;
785	–	boolean interrupted = false;
895		boolean queued = false;
896	+	boolean interrupted = false;
897	+	int spins = MAX_SPINS;
898		int p;
899		while ((p = getPhase()) == phase) {
900	<	interrupted = Thread.interrupted();
901	<	if (node != null) {
902	<	if (!queued) {
903	<	AtomicReference<QNode> head = queueFor(phase);
904	<	queued = head.compareAndSet(node.next = head.get(), node);
794	<	}
795	<	else if (node.thread != null)
796	<	LockSupport.park(this);
900	>	if (Thread.interrupted())
901	>	interrupted = true;
902	>	else if (spins > 0) {
903	>	if (--spins == 0)
904	>	Thread.yield();
905		}
906	<	else if (spins <= 0)
907	<	node = new QNode();
908	<	else
909	<	--spins;
906	>	else if (node == null)
907	>	node = new QNode(this, phase, false, false, 0, 0);
908	>	else if (!queued)
909	>	queued = tryEnqueue(node);
910	>	else if (node.doWait())
911	>	interrupted = true;
912		}
913		if (node != null)
914		node.thread = null;
915	+	releaseWaiters(phase);
916		if (interrupted)
917		Thread.currentThread().interrupt();
807	–	releaseWaiters(phase);
918		return p;
919		}
920
921		/**
922	<	* Messier interruptible version
922	>	* Interruptible version
923	>	* @return current phase
924		*/
925		private int interruptibleWait(int phase) throws InterruptedException {
815	–	int spins = maxUntimedSpins;
926		QNode node = null;
927		boolean queued = false;
928		boolean interrupted = false;
929	+	int spins = MAX_SPINS;
930		int p;
931	<	while ((p = getPhase()) == phase) {
932	<	if (interrupted = Thread.interrupted())
933	<	break;
934	<	if (node != null) {
935	<	if (!queued) {
936	<	AtomicReference<QNode> head = queueFor(phase);
826	<	queued = head.compareAndSet(node.next = head.get(), node);
827	<	}
828	<	else if (node.thread != null)
829	<	LockSupport.park(this);
931	>	while ((p = getPhase()) == phase && !interrupted) {
932	>	if (Thread.interrupted())
933	>	interrupted = true;
934	>	else if (spins > 0) {
935	>	if (--spins == 0)
936	>	Thread.yield();
937		}
938	<	else if (spins <= 0)
939	<	node = new QNode();
940	<	else
941	<	--spins;
938	>	else if (node == null)
939	>	node = new QNode(this, phase, true, false, 0, 0);
940	>	else if (!queued)
941	>	queued = tryEnqueue(node);
942	>	else if (node.doWait())
943	>	interrupted = true;
944		}
945		if (node != null)
946		node.thread = null;
947	+	if (p != phase || (p = getPhase()) != phase)
948	+	releaseWaiters(phase);
949		if (interrupted)
950		throw new InterruptedException();
840	–	releaseWaiters(phase);
951		return p;
952		}
953
954		/**
955	<	* Even messier timeout version.
955	>	* Timeout version.
956	>	* @return current phase
957		*/
958		private int timedWait(int phase, long nanos)
959		throws InterruptedException, TimeoutException {
960	+	long startTime = System.nanoTime();
961	+	QNode node = null;
962	+	boolean queued = false;
963	+	boolean interrupted = false;
964	+	int spins = MAX_SPINS;
965		int p;
966	<	if ((p = getPhase()) == phase) {
967	<	long lastTime = System.nanoTime();
968	<	int spins = maxTimedSpins;
969	<	QNode node = null;
970	<	boolean queued = false;
971	<	boolean interrupted = false;
972	<	while ((p = getPhase()) == phase) {
973	<	if (interrupted = Thread.interrupted())
974	<	break;
975	<	long now = System.nanoTime();
976	<	if ((nanos -= now - lastTime) <= 0)
977	<	break;
978	<	lastTime = now;
979	<	if (node != null) {
980	<	if (!queued) {
865	<	AtomicReference<QNode> head = queueFor(phase);
866	<	queued = head.compareAndSet(node.next = head.get(), node);
867	<	}
868	<	else if (node.thread != null &&
869	<	nanos > spinForTimeoutThreshold) {
870	<	LockSupport.parkNanos(this, nanos);
871	<	}
872	<	}
873	<	else if (spins <= 0)
874	<	node = new QNode();
875	<	else
876	<	--spins;
877	<	}
878	<	if (node != null)
879	<	node.thread = null;
880	<	if (interrupted)
881	<	throw new InterruptedException();
882	<	if (p == phase && (p = getPhase()) == phase)
883	<	throw new TimeoutException();
966	>	while ((p = getPhase()) == phase && !interrupted) {
967	>	if (Thread.interrupted())
968	>	interrupted = true;
969	>	else if (nanos - (System.nanoTime() - startTime) <= 0)
970	>	break;
971	>	else if (spins > 0) {
972	>	if (--spins == 0)
973	>	Thread.yield();
974	>	}
975	>	else if (node == null)
976	>	node = new QNode(this, phase, true, true, startTime, nanos);
977	>	else if (!queued)
978	>	queued = tryEnqueue(node);
979	>	else if (node.doWait())
980	>	interrupted = true;
981		}
982	<	releaseWaiters(phase);
982	>	if (node != null)
983	>	node.thread = null;
984	>	if (p != phase || (p = getPhase()) != phase)
985	>	releaseWaiters(phase);
986	>	if (interrupted)
987	>	throw new InterruptedException();
988	>	if (p == phase)
989	>	throw new TimeoutException();
990		return p;
991		}
992
993	<	// Temporary Unsafe mechanics for preliminary release
993	>	// Unsafe mechanics
994
995	<	static final Unsafe _unsafe;
996	<	static final long stateOffset;
995	>	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
996	>	private static final long stateOffset =
997	>	objectFieldOffset("state", Phaser.class);
998
999	<	static {
999	>	private static long objectFieldOffset(String field, Class<?> klazz) {
1000		try {
1001	<	if (Phaser.class.getClassLoader() != null) {
1002	<	Field f = Unsafe.class.getDeclaredField("theUnsafe");
1003	<	f.setAccessible(true);
1004	<	_unsafe = (Unsafe)f.get(null);
1005	<	}
1006	<	else
902	<	_unsafe = Unsafe.getUnsafe();
903	<	stateOffset = _unsafe.objectFieldOffset
904	<	(Phaser.class.getDeclaredField("state"));
905	<	} catch (Exception e) {
906	<	throw new RuntimeException("Could not initialize intrinsics", e);
1001	>	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1002	>	} catch (NoSuchFieldException e) {
1003	>	// Convert Exception to corresponding Error
1004	>	NoSuchFieldError error = new NoSuchFieldError(field);
1005	>	error.initCause(e);
1006	>	throw error;
1007		}
1008		}
1009
1010	<	final boolean casState(long cmp, long val) {
1011	<	return _unsafe.compareAndSwapLong(this, stateOffset, cmp, val);
1010	>	/**
1011	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1012	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
1013	>	* into a jdk.
1014	>	*
1015	>	* @return a sun.misc.Unsafe
1016	>	*/
1017	>	private static sun.misc.Unsafe getUnsafe() {
1018	>	try {
1019	>	return sun.misc.Unsafe.getUnsafe();
1020	>	} catch (SecurityException se) {
1021	>	try {
1022	>	return java.security.AccessController.doPrivileged
1023	>	(new java.security
1024	>	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1025	>	public sun.misc.Unsafe run() throws Exception {
1026	>	java.lang.reflect.Field f = sun.misc
1027	>	.Unsafe.class.getDeclaredField("theUnsafe");
1028	>	f.setAccessible(true);
1029	>	return (sun.misc.Unsafe) f.get(null);
1030	>	}});
1031	>	} catch (java.security.PrivilegedActionException e) {
1032	>	throw new RuntimeException("Could not initialize intrinsics",
1033	>	e.getCause());
1034	>	}
1035	>	}
1036		}
1037		}

Diff Legend

–	Removed lines
+	Added lines
<	Changed lines
>	Changed lines