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 1.2 by jsr166, Fri Jul 25 18:10:41 2008 UTC vs.
Revision 1.57 by dl, Fri Nov 19 16:03:24 2010 UTC

#	Line 5 | Line 5
5		*/
6
7		package jsr166y;
8	<	import jsr166y.forkjoin.*;
9	<	import java.util.concurrent.*;
10	<	import java.util.concurrent.atomic.*;
8	>
9	>	import java.util.concurrent.TimeUnit;
10	>	import java.util.concurrent.TimeoutException;
11	>	import java.util.concurrent.atomic.AtomicReference;
12		import java.util.concurrent.locks.LockSupport;
13
14		/**
15	<	* A reusable synchronization barrier, similar in functionality to a
16	<	* {@link java.util.concurrent.CyclicBarrier}, but supporting more
17	<	* flexible usage.
15	>	* A reusable synchronization barrier, similar in functionality to
16	>	* {@link java.util.concurrent.CyclicBarrier CyclicBarrier} and
17	>	* {@link java.util.concurrent.CountDownLatch CountDownLatch}
18	>	* but supporting more flexible usage.
19	>	*
20	>	* <p> <b>Registration.</b> Unlike the case for other barriers, the
21	>	* number of parties <em>registered</em> to synchronize on a phaser
22	>	* may vary over time.  Tasks may be registered at any time (using
23	>	* methods {@link #register}, {@link #bulkRegister}, or forms of
24	>	* constructors establishing initial numbers of parties), and
25	>	* optionally deregistered upon any arrival (using {@link
26	>	* #arriveAndDeregister}).  As is the case with most basic
27	>	* synchronization constructs, registration and deregistration affect
28	>	* only internal counts; they do not establish any further internal
29	>	* bookkeeping, so tasks cannot query whether they are registered.
30	>	* (However, you can introduce such bookkeeping by subclassing this
31	>	* class.)
32	>	*
33	>	* <p> <b>Synchronization.</b> Like a {@code CyclicBarrier}, a {@code
34	>	* Phaser} may be repeatedly awaited.  Method {@link
35	>	* #arriveAndAwaitAdvance} has effect analogous to {@link
36	>	* java.util.concurrent.CyclicBarrier#await CyclicBarrier.await}. Each
37	>	* generation of a {@code Phaser} has an associated phase number. The
38	>	* phase number starts at zero, and advances when all parties arrive
39	>	* at the barrier, wrapping around to zero after reaching {@code
40	>	* Integer.MAX_VALUE}. The use of phase numbers enables independent
41	>	* control of actions upon arrival at a barrier and upon awaiting
42	>	* others, via two kinds of methods that may be invoked by any
43	>	* registered party:
44		*
45		* <ul>
46		*
47	<	* <li> The number of parties synchronizing on the barrier may vary
48	<	* over time.  A task may register to be a party in a barrier at any
49	<	* time, and may deregister upon arriving at the barrier.  As is the
50	<	* case with most basic synchronization constructs, registration
51	<	* and deregistration affect only internal counts; they do not
52	<	* establish any further internal bookkeeping, so tasks cannot query
53	<	* whether they are registered.
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 be invoked independently:
63	<	*
64	<	* <ul>
47	>	*   <li> <b>Arrival.</b> Methods {@link #arrive} and
48	>	*       {@link #arriveAndDeregister} record arrival at a
49	>	*       barrier. These methods do not block, but return an associated
50	>	*       <em>arrival phase number</em>; that is, the phase number of
51	>	*       the barrier to which the arrival applied. When the final
52	>	*       party for a given phase arrives, an optional barrier action
53	>	*       is performed and the phase advances.  Barrier actions,
54	>	*       performed by the party triggering a phase advance, are
55	>	*       arranged by overriding method {@link #onAdvance(int, int)},
56	>	*       which also controls termination. Overriding this method is
57	>	*       similar to, but more flexible than, providing a barrier
58	>	*       action to a {@code CyclicBarrier}.
59	>	*
60	>	*   <li> <b>Waiting.</b> Method {@link #awaitAdvance} requires an
61	>	*       argument indicating an arrival phase number, and returns when
62	>	*       the barrier advances to (or is already at) a different phase.
63	>	*       Unlike similar constructions using {@code CyclicBarrier},
64	>	*       method {@code awaitAdvance} continues to wait even if the
65	>	*       waiting thread is interrupted. Interruptible and timeout
66	>	*       versions are also available, but exceptions encountered while
67	>	*       tasks wait interruptibly or with timeout do not change the
68	>	*       state of the barrier. If necessary, you can perform any
69	>	*       associated recovery within handlers of those exceptions,
70	>	*       often after invoking {@code forceTermination}.  Phasers may
71	>	*       also be used by tasks executing in a {@link ForkJoinPool},
72	>	*       which will ensure sufficient parallelism to execute tasks
73	>	*       when others are blocked waiting for a phase to advance.
74		*
39	–	*   <li> Arriving at a barrier. Methods <tt>arrive</tt> and
40	–	*       <tt>arriveAndDeregister</tt> do not block, but return
41	–	*       the phase value on entry to the method.
42	–	*
43	–	*   <li> Awaiting others. Method <tt>awaitAdvance</tt> requires an
44	–	*       argument indicating the entry phase, and returns when the
45	–	*       barrier advances to a new phase.
75		* </ul>
76		*
77	+	* <p> <b>Termination.</b> A {@code Phaser} may enter a
78	+	* <em>termination</em> state in which all synchronization methods
79	+	* immediately return without updating phaser state or waiting for
80	+	* advance, and indicating (via a negative phase value) that execution
81	+	* is complete.  Termination is triggered when an invocation of {@code
82	+	* onAdvance} returns {@code true}.  As illustrated below, when
83	+	* phasers control actions with a fixed number of iterations, it is
84	+	* often convenient to override this method to cause termination when
85	+	* the current phase number reaches a threshold. Method {@link
86	+	* #forceTermination} is also available to abruptly release waiting
87	+	* threads and allow them to terminate.
88	+	*
89	+	* <p> <b>Tiering.</b> Phasers may be <em>tiered</em> (i.e., arranged
90	+	* in tree structures) to reduce contention. Phasers with large
91	+	* numbers of parties that would otherwise experience heavy
92	+	* synchronization contention costs may instead be set up so that
93	+	* groups of sub-phasers share a common parent.  This may greatly
94	+	* increase throughput even though it incurs greater per-operation
95	+	* overhead.
96	+	*
97	+	* <p><b>Monitoring.</b> While synchronization methods may be invoked
98	+	* only by registered parties, the current state of a phaser may be
99	+	* monitored by any caller.  At any given moment there are {@link
100	+	* #getRegisteredParties} parties in total, of which {@link
101	+	* #getArrivedParties} have arrived at the current phase ({@link
102	+	* #getPhase}).  When the remaining ({@link #getUnarrivedParties})
103	+	* parties arrive, the phase advances.  The values returned by these
104	+	* methods may reflect transient states and so are not in general
105	+	* useful for synchronization control.  Method {@link #toString}
106	+	* returns snapshots of these state queries in a form convenient for
107	+	* informal monitoring.
108	+	*
109	+	* <p><b>Sample usages:</b>
110	+	*
111	+	* <p>A {@code Phaser} may be used instead of a {@code CountDownLatch}
112	+	* to control a one-shot action serving a variable number of parties.
113	+	* The typical idiom is for the method setting this up to first
114	+	* register, then start the actions, then deregister, as in:
115	+	*
116	+	*  <pre> {@code
117	+	* void runTasks(List<Runnable> tasks) {
118	+	*   final Phaser phaser = new Phaser(1); // "1" to register self
119	+	*   // create and start threads
120	+	*   for (Runnable task : tasks) {
121	+	*     phaser.register();
122	+	*     new Thread() {
123	+	*       public void run() {
124	+	*         phaser.arriveAndAwaitAdvance(); // await all creation
125	+	*         task.run();
126	+	*       }
127	+	*     }.start();
128	+	*   }
129		*
130	<	* <li> Barrier actions, performed by the task triggering a phase
131	<	* advance while others may be waiting, are arranged by overriding
132	<	* method <tt>onAdvance</tt>, that also controls termination.
133	<	*
134	<	* <li> Phasers may enter a <em>termination</em> state in which all
135	<	* await actions immediately return, indicating (via a negative phase
136	<	* value) that execution is complete.  Termination is triggered by
137	<	* executing the overridable <tt>onAdvance</tt> method that is invoked
138	<	* each time the barrier is tripped. When a Phaser is controlling an
139	<	* action with a fixed number of iterations, it is often convenient to
140	<	* override this method to cause termination when the current phase
141	<	* number reaches a threshold.  Method <tt>forceTermination</tt> is
142	<	* also available to assist recovery actions upon failure.
143	<	*
144	<	* <li> Unlike most synchronizers, a Phaser may also be used with
145	<	* ForkJoinTasks (as well as plain threads).
146	<	*
147	<	* <li> By default, <tt>awaitAdvance</tt> continues to wait even if
148	<	* the current thread is interrupted. And unlike the case in
149	<	* CyclicBarriers, exceptions encountered while tasks wait
150	<	* interruptibly or with timeout do not change the state of the
151	<	* barrier. If necessary, you can perform any associated recovery
152	<	* within handlers of those exceptions.
153	<	*
154	<	* </ul>
130	>	*   // allow threads to start and deregister self
131	>	*   phaser.arriveAndDeregister();
132	>	* }}</pre>
133	>	*
134	>	* <p>One way to cause a set of threads to repeatedly perform actions
135	>	* for a given number of iterations is to override {@code onAdvance}:
136	>	*
137	>	*  <pre> {@code
138	>	* void startTasks(List<Runnable> tasks, final int iterations) {
139	>	*   final Phaser phaser = new Phaser() {
140	>	*     protected boolean onAdvance(int phase, int registeredParties) {
141	>	*       return phase >= iterations || registeredParties == 0;
142	>	*     }
143	>	*   };
144	>	*   phaser.register();
145	>	*   for (final Runnable task : tasks) {
146	>	*     phaser.register();
147	>	*     new Thread() {
148	>	*       public void run() {
149	>	*         do {
150	>	*           task.run();
151	>	*           phaser.arriveAndAwaitAdvance();
152	>	*         } while (!phaser.isTerminated());
153	>	*       }
154	>	*     }.start();
155	>	*   }
156	>	*   phaser.arriveAndDeregister(); // deregister self, don't wait
157	>	* }}</pre>
158		*
159	<	* <p><b>Sample usage:</b>
159	>	* If the main task must later await termination, it
160	>	* may re-register and then execute a similar loop:
161	>	*  <pre> {@code
162	>	*   // ...
163	>	*   phaser.register();
164	>	*   while (!phaser.isTerminated())
165	>	*     phaser.arriveAndAwaitAdvance();}</pre>
166	>	*
167	>	* <p>Related constructions may be used to await particular phase numbers
168	>	* in contexts where you are sure that the phase will never wrap around
169	>	* {@code Integer.MAX_VALUE}. For example:
170	>	*
171	>	*  <pre> {@code
172	>	* void awaitPhase(Phaser phaser, int phase) {
173	>	*   int p = phaser.register(); // assumes caller not already registered
174	>	*   while (p < phase) {
175	>	*     if (phaser.isTerminated())
176	>	*       // ... deal with unexpected termination
177	>	*     else
178	>	*       p = phaser.arriveAndAwaitAdvance();
179	>	*   }
180	>	*   phaser.arriveAndDeregister();
181	>	* }}</pre>
182		*
77	–	* <p>[todo: non-FJ example]
183		*
184	<	* <p> A Phaser may be used to support a style of programming in
185	<	* which a task waits for others to complete, without otherwise
186	<	* needing to keep track of which tasks it is waiting for. This is
187	<	* similar to the "sync" construct in Cilk and "clocks" in X10.
188	<	* Special constructions based on such barriers are available using
189	<	* the <tt>LinkedAsyncAction</tt> and <tt>CyclicAction</tt> classes,
190	<	* but they can be useful in other contexts as well.  For a simple
191	<	* (but not very useful) example, here is a variant of Fibonacci:
192	<	*
193	<	* <pre>
194	<	* class BarrierFibonacci extends RecursiveAction {
90	<	*   int argument, result;
91	<	*   final Phaser parentBarrier;
92	<	*   BarrierFibonacci(int n, Phaser parentBarrier) {
93	<	*     this.argument = n;
94	<	*     this.parentBarrier = parentBarrier;
95	<	*     parentBarrier.register();
96	<	*   }
97	<	*   protected void compute() {
98	<	*     int n = argument;
99	<	*     if (n &lt;= 1)
100	<	*        result = n;
101	<	*     else {
102	<	*        Phaser childBarrier = new Phaser(1);
103	<	*        BarrierFibonacci f1 = new BarrierFibonacci(n - 1, childBarrier);
104	<	*        BarrierFibonacci f2 = new BarrierFibonacci(n - 2, childBarrier);
105	<	*        f1.fork();
106	<	*        f2.fork();
107	<	*        childBarrier.arriveAndAwait();
108	<	*        result = f1.result + f2.result;
184	>	* <p>To create a set of tasks using a tree of phasers,
185	>	* you could use code of the following form, assuming a
186	>	* Task class with a constructor accepting a phaser that
187	>	* it registers with upon construction:
188	>	*
189	>	*  <pre> {@code
190	>	* void build(Task[] actions, int lo, int hi, Phaser ph) {
191	>	*   if (hi - lo > TASKS_PER_PHASER) {
192	>	*     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
193	>	*       int j = Math.min(i + TASKS_PER_PHASER, hi);
194	>	*       build(actions, i, j, new Phaser(ph));
195		*     }
196	<	*     parentBarrier.arriveAndDeregister();
196	>	*   } else {
197	>	*     for (int i = lo; i < hi; ++i)
198	>	*       actions[i] = new Task(ph);
199	>	*       // assumes new Task(ph) performs ph.register()
200		*   }
201		* }
202	<	* </pre>
202	>	* // .. initially called, for n tasks via
203	>	* build(new Task[n], 0, n, new Phaser());}</pre>
204	>	*
205	>	* The best value of {@code TASKS_PER_PHASER} depends mainly on
206	>	* expected barrier synchronization rates. A value as low as four may
207	>	* be appropriate for extremely small per-barrier task bodies (thus
208	>	* high rates), or up to hundreds for extremely large ones.
209		*
210		* <p><b>Implementation notes</b>: This implementation restricts the
211	<	* maximum number of parties to 65535. Attempts to register
212	<	* additional parties result in IllegalStateExceptions.
211	>	* maximum number of parties to 65535. Attempts to register additional
212	>	* parties result in {@code IllegalStateException}. However, you can and
213	>	* should create tiered phasers to accommodate arbitrarily large sets
214	>	* of participants.
215	>	*
216	>	* @since 1.7
217	>	* @author Doug Lea
218		*/
219		public class Phaser {
220		/*
221		* This class implements an extension of X10 "clocks".  Thanks to
222	<	* Vijay Saraswat for the idea of applying it to ForkJoinTasks,
223	<	* and to Vivek Sarkar for enhancements to extend functionality.
222	>	* Vijay Saraswat for the idea, and to Vivek Sarkar for
223	>	* enhancements to extend functionality.
224		*/
225
226		/**
227		* Barrier state representation. Conceptually, a barrier contains
228		* four values:
229	<	*
230	<	* * parties -- the number of parties to wait (16 bits)
231	<	* * unarrived -- the number of parties yet to hit barrier (16 bits)
232	<	* * phase -- the generation of the barrier (31 bits)
233	<	* * terminated -- set if barrier is terminated (1 bit)
229	>	*
230	>	* * unarrived -- the number of parties yet to hit barrier (bits  0-15)
231	>	* * parties -- the number of parties to wait              (bits 16-31)
232	>	* * phase -- the generation of the barrier                (bits 32-62)
233	>	* * terminated -- set if barrier is terminated            (bit  63 / sign)
234		*
235		* However, to efficiently maintain atomicity, these values are
236	<	* packed into a single AtomicLong. Termination uses the sign bit
237	<	* of 32 bit representation of phase, so phase is set to -1 on
238	<	* termination.
239	<	*/
240	<	private final AtomicLong state;
236	>	* packed into a single (atomic) long. Termination uses the sign
237	>	* bit of 32 bit representation of phase, so phase is set to -1 on
238	>	* termination. Good performance relies on keeping state decoding
239	>	* and encoding simple, and keeping race windows short.
240	>	*/
241	>	private volatile long state;
242	>
243	>	private static final int  MAX_PARTIES    = 0xffff;
244	>	private static final int  MAX_PHASE      = 0x7fffffff;
245	>	private static final int  PARTIES_SHIFT  = 16;
246	>	private static final int  PHASE_SHIFT    = 32;
247	>	private static final int  UNARRIVED_MASK = 0xffff;
248	>	private static final long PARTIES_MASK   = 0xffff0000L; // for masking long
249	>	private static final long ONE_ARRIVAL    = 1L;
250	>	private static final long ONE_PARTY      = 1L << PARTIES_SHIFT;
251	>	private static final long TERMINATION_PHASE  = -1L << PHASE_SHIFT;
252
253	<	/**
143	<	* Head of Treiber stack for waiting nonFJ threads.
144	<	*/
145	<	private final AtomicReference<QNode> head = new AtomicReference<QNode>();
146	<
147	<	private static final int ushortBits = 16;
148	<	private static final int ushortMask =  (1 << ushortBits) - 1;
149	<	private static final int phaseMask = 0x7fffffff;
253	>	// The following unpacking methods are usually manually inlined
254
255		private static int unarrivedOf(long s) {
256	<	return (int)(s & ushortMask);
256	>	return (int)s & UNARRIVED_MASK;
257		}
258
259		private static int partiesOf(long s) {
260	<	return (int)(s & (ushortMask << 16)) >>> 16;
260	>	return (int)s >>> PARTIES_SHIFT;
261		}
262
263		private static int phaseOf(long s) {
264	<	return (int)(s >>> 32);
264	>	return (int) (s >>> PHASE_SHIFT);
265		}
266
267		private static int arrivedOf(long s) {
268		return partiesOf(s) - unarrivedOf(s);
269		}
270
271	<	private static long stateFor(int phase, int parties, int unarrived) {
272	<	return (((long)phase) << 32) | ((parties << 16) | unarrived);
271	>	/**
272	>	* The parent of this phaser, or null if none
273	>	*/
274	>	private final Phaser parent;
275	>
276	>	/**
277	>	* The root of phaser tree. Equals this if not in a tree.  Used to
278	>	* support faster state push-down.
279	>	*/
280	>	private final Phaser root;
281	>
282	>	/**
283	>	* Heads of Treiber stacks for waiting threads. To eliminate
284	>	* contention when releasing some threads while adding others, we
285	>	* use two of them, alternating across even and odd phases.
286	>	* Subphasers share queues with root to speed up releases.
287	>	*/
288	>	private final AtomicReference<QNode> evenQ;
289	>	private final AtomicReference<QNode> oddQ;
290	>
291	>	private AtomicReference<QNode> queueFor(int phase) {
292	>	return ((phase & 1) == 0) ? evenQ : oddQ;
293	>	}
294	>
295	>	/**
296	>	* Main implementation for methods arrive and arriveAndDeregister.
297	>	* Manually tuned to speed up and minimize race windows for the
298	>	* common case of just decrementing unarrived field.
299	>	*
300	>	* @param adj - adjustment to apply to state -- either
301	>	* ONE_ARRIVAL (for arrive) or
302	>	* ONE_ARRIVAL|ONE_PARTY (for arriveAndDeregister)
303	>	*/
304	>	private int doArrive(long adj) {
305	>	for (;;) {
306	>	long s = state;
307	>	int phase = (int)(s >>> PHASE_SHIFT);
308	>	if (phase < 0)
309	>	return phase;
310	>	int unarrived = (int)s & UNARRIVED_MASK;
311	>	if (unarrived == 0)
312	>	checkBadArrive(s);
313	>	else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adj)) {
314	>	if (unarrived == 1) {
315	>	long p = s & PARTIES_MASK; // unshifted parties field
316	>	long lu = p >>> PARTIES_SHIFT;
317	>	int u = (int)lu;
318	>	int nextPhase = (phase + 1) & MAX_PHASE;
319	>	long next = ((long)nextPhase << PHASE_SHIFT) | p | lu;
320	>	final Phaser parent = this.parent;
321	>	if (parent == null) {
322	>	if (onAdvance(phase, u))
323	>	next |= TERMINATION_PHASE; // obliterate phase
324	>	UNSAFE.compareAndSwapLong(this, stateOffset, s, next);
325	>	releaseWaiters(phase);
326	>	}
327	>	else {
328	>	parent.doArrive((u == 0) ?
329	>	ONE_ARRIVAL|ONE_PARTY : ONE_ARRIVAL);
330	>	if ((int)(parent.state >>> PHASE_SHIFT) != nextPhase ||
331	>	((int)(state >>> PHASE_SHIFT) != nextPhase &&
332	>	!UNSAFE.compareAndSwapLong(this, stateOffset,
333	>	s, next)))
334	>	reconcileState();
335	>	}
336	>	}
337	>	return phase;
338	>	}
339	>	}
340	>	}
341	>
342	>	/**
343	>	* Rechecks state and throws bounds exceptions on arrival -- called
344	>	* only if unarrived is apparently zero.
345	>	*/
346	>	private void checkBadArrive(long s) {
347	>	if (reconcileState() == s)
348	>	throw new IllegalStateException
349	>	("Attempted arrival of unregistered party for " +
350	>	stateToString(s));
351	>	}
352	>
353	>	/**
354	>	* Implementation of register, bulkRegister
355	>	*
356	>	* @param registrations number to add to both parties and unarrived fields
357	>	*/
358	>	private int doRegister(int registrations) {
359	>	// assert registrations > 0;
360	>	// adjustment to state
361	>	long adj = ((long)registrations << PARTIES_SHIFT) | registrations;
362	>	final Phaser parent = this.parent;
363	>	for (;;) {
364	>	long s = (parent == null) ? state : reconcileState();
365	>	int phase = (int)(s >>> PHASE_SHIFT);
366	>	if (phase < 0)
367	>	return phase;
368	>	int parties = (int)s >>> PARTIES_SHIFT;
369	>	if (parties != 0 && ((int)s & UNARRIVED_MASK) == 0)
370	>	internalAwaitAdvance(phase, null); // wait for onAdvance
371	>	else if (registrations > MAX_PARTIES - parties)
372	>	throw new IllegalStateException(badRegister(s));
373	>	else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s + adj))
374	>	return phase;
375	>	}
376	>	}
377	>
378	>	/**
379	>	* Returns message string for out of bounds exceptions on registration.
380	>	*/
381	>	private String badRegister(long s) {
382	>	return "Attempt to register more than " +
383	>	MAX_PARTIES + " parties for " + stateToString(s);
384		}
385
386	<	private static IllegalStateException badBounds(int parties, int unarrived) {
387	<	return new IllegalStateException("Attempt to set " + unarrived +
388	<	" unarrived of " + parties + " parties");
386	>	/**
387	>	* Recursively resolves lagged phase propagation from root if necessary.
388	>	*/
389	>	private long reconcileState() {
390	>	Phaser par = parent;
391	>	long s = state;
392	>	if (par != null) {
393	>	Phaser rt = root;
394	>	int phase, rPhase;
395	>	while ((phase = (int)(s >>> PHASE_SHIFT)) >= 0 &&
396	>	(rPhase = (int)(rt.state >>> PHASE_SHIFT)) != phase) {
397	>	if ((int)(par.state >>> PHASE_SHIFT) != rPhase)
398	>	par.reconcileState();
399	>	else if (rPhase < 0 || ((int)s & UNARRIVED_MASK) == 0) {
400	>	long u = s & PARTIES_MASK; // reset unarrived to parties
401	>	long next = ((((long) rPhase) << PHASE_SHIFT) | u |
402	>	(u >>> PARTIES_SHIFT));
403	>	if (state == s &&
404	>	UNSAFE.compareAndSwapLong(this, stateOffset,
405	>	s, s = next))
406	>	break;
407	>	}
408	>	s = state;
409	>	}
410	>	}
411	>	return s;
412		}
413
414		/**
415	<	* Creates a new Phaser without any initially registered parties,
416	<	* and initial phase number 0.
415	>	* Creates a new phaser without any initially registered parties,
416	>	* initial phase number 0, and no parent. Any thread using this
417	>	* phaser will need to first register for it.
418		*/
419		public Phaser() {
420	<	state = new AtomicLong(stateFor(0, 0, 0));
420	>	this(null, 0);
421		}
422
423		/**
424	<	* Creates a new Phaser with the given numbers of registered
425	<	* unarrived parties and initial phase number 0.
426	<	* @param parties the number of parties required to trip barrier.
424	>	* Creates a new phaser with the given number of registered
425	>	* unarrived parties, initial phase number 0, and no parent.
426	>	*
427	>	* @param parties the number of parties required to trip barrier
428		* @throws IllegalArgumentException if parties less than zero
429	<	* or greater than the maximum number of parties supported.
429	>	* or greater than the maximum number of parties supported
430		*/
431		public Phaser(int parties) {
432	<	if (parties < 0 || parties > ushortMask)
432	>	this(null, parties);
433	>	}
434	>
435	>	/**
436	>	* Creates a new phaser with the given parent, without any
437	>	* initially registered parties. If parent is non-null this phaser
438	>	* is registered with the parent and its initial phase number is
439	>	* the same as that of parent phaser.
440	>	*
441	>	* @param parent the parent phaser
442	>	*/
443	>	public Phaser(Phaser parent) {
444	>	this(parent, 0);
445	>	}
446	>
447	>	/**
448	>	* Creates a new phaser with the given parent and number of
449	>	* registered unarrived parties. If parent is non-null, this phaser
450	>	* is registered with the parent and its initial phase number is
451	>	* the same as that of parent phaser.
452	>	*
453	>	* @param parent the parent phaser
454	>	* @param parties the number of parties required to trip barrier
455	>	* @throws IllegalArgumentException if parties less than zero
456	>	* or greater than the maximum number of parties supported
457	>	*/
458	>	public Phaser(Phaser parent, int parties) {
459	>	if (parties >>> PARTIES_SHIFT != 0)
460		throw new IllegalArgumentException("Illegal number of parties");
461	<	state = new AtomicLong(stateFor(0, parties, parties));
461	>	int phase;
462	>	this.parent = parent;
463	>	if (parent != null) {
464	>	Phaser r = parent.root;
465	>	this.root = r;
466	>	this.evenQ = r.evenQ;
467	>	this.oddQ = r.oddQ;
468	>	phase = parent.register();
469	>	}
470	>	else {
471	>	this.root = this;
472	>	this.evenQ = new AtomicReference<QNode>();
473	>	this.oddQ = new AtomicReference<QNode>();
474	>	phase = 0;
475	>	}
476	>	long p = (long)parties;
477	>	this.state = (((long)phase) << PHASE_SHIFT) | p | (p << PARTIES_SHIFT);
478		}
479
480		/**
481		* Adds a new unarrived party to this phaser.
482	<	* @return the current barrier phase number upon registration
482	>	* If an ongoing invocation of {@link #onAdvance} is in progress,
483	>	* this method may wait until its completion before registering.
484	>	*
485	>	* @return the arrival phase number to which this registration applied
486		* @throws IllegalStateException if attempting to register more
487	<	* than the maximum supported number of parties.
487	>	* than the maximum supported number of parties
488		*/
489	<	public int register() { // increment both parties and unarrived
490	<	final AtomicLong state = this.state;
205	<	for (;;) {
206	<	long s = state.get();
207	<	int phase = phaseOf(s);
208	<	int parties = partiesOf(s) + 1;
209	<	int unarrived = unarrivedOf(s) + 1;
210	<	if (parties > ushortMask || unarrived > ushortMask)
211	<	throw badBounds(parties, unarrived);
212	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
213	<	return phase;
214	<	}
489	>	public int register() {
490	>	return doRegister(1);
491		}
492
493		/**
494	<	* Arrives at the barrier, but does not wait for others.  (You can
495	<	* in turn wait for others via {@link #awaitAdvance}).
494	>	* Adds the given number of new unarrived parties to this phaser.
495	>	* If an ongoing invocation of {@link #onAdvance} is in progress,
496	>	* this method may wait until its completion before registering.
497		*
498	<	* @return the current barrier phase number upon entry to
499	<	* this method, or a negative value if terminated;
500	<	* @throws IllegalStateException if the number of unarrived
501	<	* parties would become negative.
498	>	* @param parties the number of additional parties required to trip barrier
499	>	* @return the arrival phase number to which this registration applied
500	>	* @throws IllegalStateException if attempting to register more
501	>	* than the maximum supported number of parties
502	>	* @throws IllegalArgumentException if {@code parties < 0}
503		*/
504	<	public int arrive() { // decrement unarrived. If zero, trip
505	<	final AtomicLong state = this.state;
506	<	for (;;) {
507	<	long s = state.get();
508	<	int phase = phaseOf(s);
509	<	int parties = partiesOf(s);
232	<	int unarrived = unarrivedOf(s) - 1;
233	<	if (unarrived < 0)
234	<	throw badBounds(parties, unarrived);
235	<	if (unarrived == 0 && phase >= 0) {
236	<	trip(phase, parties);
237	<	return phase;
238	<	}
239	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
240	<	return phase;
241	<	}
504	>	public int bulkRegister(int parties) {
505	>	if (parties < 0)
506	>	throw new IllegalArgumentException();
507	>	if (parties == 0)
508	>	return getPhase();
509	>	return doRegister(parties);
510		}
511
512		/**
513	<	* Arrives at the barrier, and deregisters from it, without
514	<	* waiting for others.
513	>	* Arrives at the barrier, but does not wait for others.  (You can
514	>	* in turn wait for others via {@link #awaitAdvance}).  It is an
515	>	* unenforced usage error for an unregistered party to invoke this
516	>	* method.
517		*
518	<	* @return the current barrier phase number upon entry to
519	<	* this method, or a negative value if terminated;
520	<	* @throws IllegalStateException if the number of registered or
521	<	* unarrived parties would become negative.
522	<	*/
523	<	public int arriveAndDeregister() { // Same as arrive, plus decrement parties
524	<	final AtomicLong state = this.state;
525	<	for (;;) {
526	<	long s = state.get();
527	<	int phase = phaseOf(s);
528	<	int parties = partiesOf(s) - 1;
529	<	int unarrived = unarrivedOf(s) - 1;
530	<	if (parties < 0 || unarrived < 0)
531	<	throw badBounds(parties, unarrived);
532	<	if (unarrived == 0 && phase >= 0) {
533	<	trip(phase, parties);
534	<	return phase;
535	<	}
536	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
537	<	return phase;
538	<	}
518	>	* @return the arrival phase number, or a negative value if terminated
519	>	* @throws IllegalStateException if not terminated and the number
520	>	* of unarrived parties would become negative
521	>	*/
522	>	public int arrive() {
523	>	return doArrive(ONE_ARRIVAL);
524	>	}
525	>
526	>	/**
527	>	* Arrives at the barrier and deregisters from it without waiting
528	>	* for others. Deregistration reduces the number of parties
529	>	* required to trip the barrier in future phases.  If this phaser
530	>	* has a parent, and deregistration causes this phaser to have
531	>	* zero parties, this phaser also arrives at and is deregistered
532	>	* from its parent.  It is an unenforced usage error for an
533	>	* unregistered party to invoke this method.
534	>	*
535	>	* @return the arrival phase number, or a negative value if terminated
536	>	* @throws IllegalStateException if not terminated and the number
537	>	* of registered or unarrived parties would become negative
538	>	*/
539	>	public int arriveAndDeregister() {
540	>	return doArrive(ONE_ARRIVAL|ONE_PARTY);
541		}
542
543		/**
544	<	* Arrives at the barrier and awaits others. Unlike other arrival
545	<	* methods, this method returns the arrival index of the
546	<	* caller. The caller tripping the barrier returns zero, the
547	<	* previous caller 1, and so on.
548	<	* @return the arrival index
549	<	* @throws IllegalStateException if the number of unarrived
550	<	* parties would become negative.
544	>	* Arrives at the barrier and awaits others. Equivalent in effect
545	>	* to {@code awaitAdvance(arrive())}.  If you need to await with
546	>	* interruption or timeout, you can arrange this with an analogous
547	>	* construction using one of the other forms of the {@code
548	>	* awaitAdvance} method.  If instead you need to deregister upon
549	>	* arrival, use {@link #arriveAndDeregister}. It is an unenforced
550	>	* usage error for an unregistered party to invoke this method.
551	>	*
552	>	* @return the arrival phase number, or a negative number if terminated
553	>	* @throws IllegalStateException if not terminated and the number
554	>	* of unarrived parties would become negative
555		*/
556		public int arriveAndAwaitAdvance() {
557	<	final AtomicLong state = this.state;
282	<	for (;;) {
283	<	long s = state.get();
284	<	int phase = phaseOf(s);
285	<	int parties = partiesOf(s);
286	<	int unarrived = unarrivedOf(s) - 1;
287	<	if (unarrived < 0)
288	<	throw badBounds(parties, unarrived);
289	<	if (unarrived == 0 && phase >= 0) {
290	<	trip(phase, parties);
291	<	return 0;
292	<	}
293	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived))) {
294	<	awaitAdvance(phase);
295	<	return unarrived;
296	<	}
297	<	}
557	>	return awaitAdvance(arrive());
558		}
559
560		/**
561	<	* Awaits the phase of the barrier to advance from the given
562	<	* value, or returns immediately if this barrier is terminated.
563	<	* @param phase the phase on entry to this method
564	<	* @return the phase on exit from this method
561	>	* Awaits the phase of the barrier to advance from the given phase
562	>	* value, returning immediately if the current phase of the
563	>	* barrier is not equal to the given phase value or this barrier
564	>	* is terminated.
565	>	*
566	>	* @param phase an arrival phase number, or negative value if
567	>	* terminated; this argument is normally the value returned by a
568	>	* previous call to {@code arrive} or its variants
569	>	* @return the next arrival phase number, or a negative value
570	>	* if terminated or argument is negative
571		*/
572		public int awaitAdvance(int phase) {
573		if (phase < 0)
574		return phase;
575	<	Thread current = Thread.currentThread();
576	<	if (current instanceof ForkJoinWorkerThread)
577	<	return helpingWait(phase);
312	<	if (untimedWait(current, phase, false))
313	<	current.interrupt();
314	<	return phaseOf(state.get());
575	>	long s = (parent == null) ? state : reconcileState();
576	>	int p = (int)(s >>> PHASE_SHIFT);
577	>	return (p != phase) ? p : internalAwaitAdvance(phase, null);
578		}
579
580		/**
581	<	* Awaits the phase of the barrier to advance from the given
582	<	* value, or returns immediately if this barrier is terminated, or
583	<	* throws InterruptedException if interrupted while waiting.
584	<	* @param phase the phase on entry to this method
585	<	* @return the phase on exit from this method
581	>	* Awaits the phase of the barrier to advance from the given phase
582	>	* value, throwing {@code InterruptedException} if interrupted
583	>	* while waiting, or returning immediately if the current phase of
584	>	* the barrier is not equal to the given phase value or this
585	>	* barrier is terminated.
586	>	*
587	>	* @param phase an arrival phase number, or negative value if
588	>	* terminated; this argument is normally the value returned by a
589	>	* previous call to {@code arrive} or its variants
590	>	* @return the next arrival phase number, or a negative value
591	>	* if terminated or argument is negative
592		* @throws InterruptedException if thread interrupted while waiting
593		*/
594	<	public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
594	>	public int awaitAdvanceInterruptibly(int phase)
595	>	throws InterruptedException {
596		if (phase < 0)
597		return phase;
598	<	Thread current = Thread.currentThread();
599	<	if (current instanceof ForkJoinWorkerThread)
600	<	return helpingWait(phase);
601	<	else if (Thread.interrupted() || untimedWait(current, phase, true))
602	<	throw new InterruptedException();
603	<	else
604	<	return phaseOf(state.get());
598	>	long s = (parent == null) ? state : reconcileState();
599	>	int p = (int)(s >>> PHASE_SHIFT);
600	>	if (p == phase) {
601	>	QNode node = new QNode(this, phase, true, false, 0L);
602	>	p = internalAwaitAdvance(phase, node);
603	>	if (node.wasInterrupted)
604	>	throw new InterruptedException();
605	>	}
606	>	return p;
607		}
608
609		/**
610	<	* Awaits the phase of the barrier to advance from the given value
611	<	* or the given timeout elapses, or returns immediately if this
612	<	* barrier is terminated.
613	<	* @param phase the phase on entry to this method
614	<	* @return the phase on exit from this method
610	>	* Awaits the phase of the barrier to advance from the given phase
611	>	* value or the given timeout to elapse, throwing {@code
612	>	* InterruptedException} if interrupted while waiting, or
613	>	* returning immediately if the current phase of the barrier is
614	>	* not equal to the given phase value or this barrier is
615	>	* terminated.
616	>	*
617	>	* @param phase an arrival phase number, or negative value if
618	>	* terminated; this argument is normally the value returned by a
619	>	* previous call to {@code arrive} or its variants
620	>	* @param timeout how long to wait before giving up, in units of
621	>	*        {@code unit}
622	>	* @param unit a {@code TimeUnit} determining how to interpret the
623	>	*        {@code timeout} parameter
624	>	* @return the next arrival phase number, or a negative value
625	>	* if terminated or argument is negative
626		* @throws InterruptedException if thread interrupted while waiting
627		* @throws TimeoutException if timed out while waiting
628		*/
629	<	public int awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)
629	>	public int awaitAdvanceInterruptibly(int phase,
630	>	long timeout, TimeUnit unit)
631		throws InterruptedException, TimeoutException {
632		if (phase < 0)
633		return phase;
634	<	long nanos = unit.toNanos(timeout);
635	<	Thread current = Thread.currentThread();
636	<	if (current instanceof ForkJoinWorkerThread)
637	<	return timedHelpingWait(phase, nanos);
638	<	timedWait(current, phase, nanos);
639	<	return phaseOf(state.get());
634	>	long s = (parent == null) ? state : reconcileState();
635	>	int p = (int)(s >>> PHASE_SHIFT);
636	>	if (p == phase) {
637	>	long nanos = unit.toNanos(timeout);
638	>	QNode node = new QNode(this, phase, true, true, nanos);
639	>	p = internalAwaitAdvance(phase, node);
640	>	if (node.wasInterrupted)
641	>	throw new InterruptedException();
642	>	else if (p == phase)
643	>	throw new TimeoutException();
644	>	}
645	>	return p;
646		}
647
648		/**
649	<	* Forces this barrier to enter termination state. Counts of
650	<	* arrived and registered parties are unaffected. This method may
651	<	* be useful for coordinating recovery after one or more tasks
649	>	* Forces this barrier to enter termination state.  Counts of
650	>	* arrived and registered parties are unaffected.  If this phaser
651	>	* is a member of a tiered set of phasers, then all of the phasers
652	>	* in the set are terminated.  If this phaser is already
653	>	* terminated, this method has no effect.  This method may be
654	>	* useful for coordinating recovery after one or more tasks
655		* encounter unexpected exceptions.
656		*/
657		public void forceTermination() {
658	<	final AtomicLong state = this.state;
659	<	for (;;) {
660	<	long s = state.get();
661	<	int phase = phaseOf(s);
662	<	int parties = partiesOf(s);
663	<	int unarrived = unarrivedOf(s);
664	<	if (phase < 0 ||
665	<	state.compareAndSet(s, stateFor(-1, parties, unarrived))) {
373	<	if (head.get() != null)
374	<	releaseWaiters(-1);
658	>	// Only need to change root state
659	>	final Phaser root = this.root;
660	>	long s;
661	>	while ((s = root.state) >= 0) {
662	>	if (UNSAFE.compareAndSwapLong(root, stateOffset,
663	>	s, s | TERMINATION_PHASE)) {
664	>	releaseWaiters(0); // signal all threads
665	>	releaseWaiters(1);
666		return;
667		}
668		}
669		}
670
671		/**
381	–	* Resets the barrier with the given numbers of registered unarrived
382	–	* parties and phase number 0. This method allows repeated reuse
383	–	* of this barrier, but only if it is somehow known not to be in
384	–	* use for other purposes.
385	–	* @param parties the number of parties required to trip barrier.
386	–	* @throws IllegalArgumentException if parties less than zero
387	–	* or greater than the maximum number of parties supported.
388	–	*/
389	–	public void reset(int parties) {
390	–	if (parties < 0 || parties > ushortMask)
391	–	throw new IllegalArgumentException("Illegal number of parties");
392	–	state.set(stateFor(0, parties, parties));
393	–	if (head.get() != null)
394	–	releaseWaiters(0);
395	–	}
396	–
397	–	/**
672		* Returns the current phase number. The maximum phase number is
673	<	* <tt>Integer.MAX_VALUE</tt>, after which it restarts at
673	>	* {@code Integer.MAX_VALUE}, after which it restarts at
674		* zero. Upon termination, the phase number is negative.
675	+	*
676		* @return the phase number, or a negative value if terminated
677		*/
678	<	public int getPhase() {
679	<	return phaseOf(state.get());
678	>	public final int getPhase() {
679	>	return (int)(root.state >>> PHASE_SHIFT);
680		}
681
682		/**
683		* Returns the number of parties registered at this barrier.
684	+	*
685		* @return the number of parties
686		*/
687		public int getRegisteredParties() {
688	<	return partiesOf(state.get());
688	>	return partiesOf(state);
689		}
690
691		/**
692	<	* Returns the number of parties that have arrived at the current
693	<	* phase of this barrier.
692	>	* Returns the number of registered parties that have arrived at
693	>	* the current phase of this barrier.
694	>	*
695		* @return the number of arrived parties
696		*/
697		public int getArrivedParties() {
698	<	return arrivedOf(state.get());
698	>	return arrivedOf(parent==null? state : reconcileState());
699		}
700
701		/**
702		* Returns the number of registered parties that have not yet
703		* arrived at the current phase of this barrier.
704	+	*
705		* @return the number of unarrived parties
706		*/
707		public int getUnarrivedParties() {
708	<	return unarrivedOf(state.get());
708	>	return unarrivedOf(parent==null? state : reconcileState());
709		}
710
711		/**
712	<	* Returns true if this barrier has been terminated.
713	<	* @return true if this barrier has been terminated
712	>	* Returns the parent of this phaser, or {@code null} if none.
713	>	*
714	>	* @return the parent of this phaser, or {@code null} if none
715	>	*/
716	>	public Phaser getParent() {
717	>	return parent;
718	>	}
719	>
720	>	/**
721	>	* Returns the root ancestor of this phaser, which is the same as
722	>	* this phaser if it has no parent.
723	>	*
724	>	* @return the root ancestor of this phaser
725	>	*/
726	>	public Phaser getRoot() {
727	>	return root;
728	>	}
729	>
730	>	/**
731	>	* Returns {@code true} if this barrier has been terminated.
732	>	*
733	>	* @return {@code true} if this barrier has been terminated
734		*/
735		public boolean isTerminated() {
736	<	return phaseOf(state.get()) < 0;
736	>	return root.state < 0L;
737		}
738
739		/**
740	<	* Overridable method to perform an action upon phase advance, and
741	<	* to control termination. This method is invoked whenever the
742	<	* barrier is tripped (and thus all other waiting parties are
743	<	* dormant). If it returns true, then, rather than advance the
744	<	* phase number, this barrier will be set to a final termination
745	<	* state, and subsequent calls to <tt>isTerminated</tt> will
746	<	* return true.
747	<	*
748	<	* <p> The default version returns true when the number of
740	>	* Overridable method to perform an action upon impending phase
741	>	* advance, and to control termination. This method is invoked
742	>	* upon arrival of the party tripping the barrier (when all other
743	>	* waiting parties are dormant).  If this method returns {@code
744	>	* true}, then, rather than advance the phase number, this barrier
745	>	* will be set to a final termination state, and subsequent calls
746	>	* to {@link #isTerminated} will return true. Any (unchecked)
747	>	* Exception or Error thrown by an invocation of this method is
748	>	* propagated to the party attempting to trip the barrier, in
749	>	* which case no advance occurs.
750	>	*
751	>	* <p>The arguments to this method provide the state of the phaser
752	>	* prevailing for the current transition.  The effects of invoking
753	>	* arrival, registration, and waiting methods on this Phaser from
754	>	* within {@code onAdvance} are unspecified and should not be
755	>	* relied on.
756	>	*
757	>	* <p>If this Phaser is a member of a tiered set of Phasers, then
758	>	* {@code onAdvance} is invoked only for its root Phaser on each
759	>	* advance.
760	>	*
761	>	* <p>The default version returns {@code true} when the number of
762		* registered parties is zero. Normally, overrides that arrange
763		* termination for other reasons should also preserve this
764		* property.
765		*
766		* @param phase the phase number on entering the barrier
767	<	* @param registeredParties the current number of registered
768	<	* parties.
458	<	* @return true if this barrier should terminate
767	>	* @param registeredParties the current number of registered parties
768	>	* @return {@code true} if this barrier should terminate
769		*/
770		protected boolean onAdvance(int phase, int registeredParties) {
771		return registeredParties <= 0;
772		}
773
774		/**
775	<	* Returns a string identifying this barrier, as well as its
775	>	* Returns a string identifying this phaser, as well as its
776		* state.  The state, in brackets, includes the String {@code
777	<	* "phase ="} followed by the phase number, {@code "parties ="}
777	>	* "phase = "} followed by the phase number, {@code "parties = "}
778		* followed by the number of registered parties, and {@code
779	<	* "arrived ="} followed by the number of arrived parties
779	>	* "arrived = "} followed by the number of arrived parties.
780		*
781		* @return a string identifying this barrier, as well as its state
782		*/
783		public String toString() {
784	<	long s = state.get();
475	<	return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
784	>	return stateToString(reconcileState());
785		}
786
478	–	// methods for tripping and waiting
479	–
787		/**
788	<	* Advance the current phase (or terminate)
788	>	* Implementation of toString and string-based error messages
789		*/
790	<	private void trip(int phase, int parties) {
791	<	int next = onAdvance(phase, parties)? -1 : ((phase + 1) & phaseMask);
792	<	state.set(stateFor(next, parties, parties));
793	<	if (head.get() != null)
794	<	releaseWaiters(next);
790	>	private String stateToString(long s) {
791	>	return super.toString() +
792	>	"[phase = " + phaseOf(s) +
793	>	" parties = " + partiesOf(s) +
794	>	" arrived = " + arrivedOf(s) + "]";
795		}
796
797	<	private int helpingWait(int phase) {
798	<	final AtomicLong state = this.state;
797	>	// Waiting mechanics
798	>
799	>	/**
800	>	* Removes and signals threads from queue for phase.
801	>	*/
802	>	private void releaseWaiters(int phase) {
803	>	AtomicReference<QNode> head = queueFor(phase);
804	>	QNode q;
805		int p;
806	<	while ((p = phaseOf(state.get())) == phase) {
807	<	ForkJoinTask<?> t = ForkJoinWorkerThread.pollTask();
808	<	if (t != null) {
809	<	if ((p = phaseOf(state.get())) == phase)
810	<	t.exec();
498	<	else {   // push task and exit if barrier advanced
499	<	t.fork();
500	<	break;
501	<	}
502	<	}
806	>	while ((q = head.get()) != null &&
807	>	((p = q.phase) == phase ||
808	>	(int)(root.state >>> PHASE_SHIFT) != p)) {
809	>	if (head.compareAndSet(q, q.next))
810	>	q.signal();
811		}
504	–	return p;
812		}
813
814	<	private int timedHelpingWait(int phase, long nanos) throws TimeoutException {
815	<	final AtomicLong state = this.state;
816	<	long lastTime = System.nanoTime();
814	>	/** The number of CPUs, for spin control */
815	>	private static final int NCPU = Runtime.getRuntime().availableProcessors();
816	>
817	>	/**
818	>	* The number of times to spin before blocking while waiting for
819	>	* advance, per arrival while waiting. On multiprocessors, fully
820	>	* blocking and waking up a large number of threads all at once is
821	>	* usually a very slow process, so we use rechargeable spins to
822	>	* avoid it when threads regularly arrive: When a thread in
823	>	* internalAwaitAdvance notices another arrival before blocking,
824	>	* and there appear to be enough CPUs available, it spins
825	>	* SPINS_PER_ARRIVAL more times before blocking. Plus, even on
826	>	* uniprocessors, there is at least one intervening Thread.yield
827	>	* before blocking. The value trades off good-citizenship vs big
828	>	* unnecessary slowdowns.
829	>	*/
830	>	static final int SPINS_PER_ARRIVAL = (NCPU < 2) ? 1 : 1 << 8;
831	>
832	>	/**
833	>	* Possibly blocks and waits for phase to advance unless aborted.
834	>	*
835	>	* @param phase current phase
836	>	* @param node if non-null, the wait node to track interrupt and timeout;
837	>	* if null, denotes noninterruptible wait
838	>	* @return current phase
839	>	*/
840	>	private int internalAwaitAdvance(int phase, QNode node) {
841	>	Phaser current = this;       // to eventually wait at root if tiered
842	>	boolean queued = false;      // true when node is enqueued
843	>	int lastUnarrived = -1;      // to increase spins upon change
844	>	int spins = SPINS_PER_ARRIVAL;
845	>	long s;
846		int p;
847	<	while ((p = phaseOf(state.get())) == phase) {
848	<	long now = System.nanoTime();
849	<	nanos -= now - lastTime;
850	<	lastTime = now;
851	<	if (nanos <= 0) {
852	<	if ((p = phaseOf(state.get())) == phase)
853	<	throw new TimeoutException();
854	<	else
847	>	while ((p = (int)((s = current.state) >>> PHASE_SHIFT)) == phase) {
848	>	Phaser par;
849	>	int unarrived = (int)s & UNARRIVED_MASK;
850	>	if (unarrived != lastUnarrived) {
851	>	if (lastUnarrived == -1) // ensure old queue clean
852	>	releaseWaiters(phase-1);
853	>	if ((lastUnarrived = unarrived) < NCPU)
854	>	spins += SPINS_PER_ARRIVAL;
855	>	}
856	>	else if (unarrived == 0 && (par = current.parent) != null) {
857	>	current = par;       // if all arrived, use parent
858	>	par = par.parent;
859	>	lastUnarrived = -1;
860	>	}
861	>	else if (spins > 0) {
862	>	if (--spins == (SPINS_PER_ARRIVAL >>> 1))
863	>	Thread.yield();  // yield midway through spin
864	>	}
865	>	else if (node == null)   // must be noninterruptible
866	>	node = new QNode(this, phase, false, false, 0L);
867	>	else if (node.isReleasable()) {
868	>	if ((p = (int)(root.state >>> PHASE_SHIFT)) != phase)
869		break;
870	+	else
871	+	return phase;    // aborted
872		}
873	<	ForkJoinTask<?> t = ForkJoinWorkerThread.pollTask();
874	<	if (t != null) {
875	<	if ((p = phaseOf(state.get())) == phase)
876	<	t.exec();
877	<	else {   // push task and exit if barrier advanced
878	<	t.fork();
879	<	break;
873	>	else if (!queued) {      // push onto queue
874	>	AtomicReference<QNode> head = queueFor(phase);
875	>	QNode q = head.get();
876	>	if (q == null || q.phase == phase) {
877	>	node.next = q;
878	>	if ((p = (int)(root.state >>> PHASE_SHIFT)) != phase)
879	>	break;       // recheck to avoid stale enqueue
880	>	else
881	>	queued = head.compareAndSet(q, node);
882	>	}
883	>	}
884	>	else {
885	>	try {
886	>	ForkJoinPool.managedBlock(node);
887	>	} catch (InterruptedException ie) {
888	>	node.wasInterrupted = true;
889		}
890		}
891		}
892	+	releaseWaiters(phase);
893	+	if (node != null)
894	+	node.onRelease();
895		return p;
896		}
897
898		/**
899	<	* Wait nodes for Treiber stack representing wait queue for non-FJ
536	<	* tasks. The waiting scheme is an adaptation of the one used in
537	<	* forkjoin.PoolBarrier.
899	>	* Wait nodes for Treiber stack representing wait queue
900		*/
901	<	static final class QNode {
902	<	QNode next;
541	<	volatile Thread thread; // nulled to cancel wait
901	>	static final class QNode implements ForkJoinPool.ManagedBlocker {
902	>	final Phaser phaser;
903		final int phase;
904	<	QNode(Thread t, int c) {
905	<	thread = t;
906	<	phase = c;
904	>	final boolean interruptible;
905	>	final boolean timed;
906	>	boolean wasInterrupted;
907	>	long nanos;
908	>	long lastTime;
909	>	volatile Thread thread; // nulled to cancel wait
910	>	QNode next;
911	>
912	>	QNode(Phaser phaser, int phase, boolean interruptible,
913	>	boolean timed, long nanos) {
914	>	this.phaser = phaser;
915	>	this.phase = phase;
916	>	this.interruptible = interruptible;
917	>	this.nanos = nanos;
918	>	this.timed = timed;
919	>	this.lastTime = timed? System.nanoTime() : 0L;
920	>	thread = Thread.currentThread();
921		}
547	–	}
922
923	<	private void releaseWaiters(int currentPhase) {
924	<	final AtomicReference<QNode> head = this.head;
925	<	QNode p;
926	<	while ((p = head.get()) != null && p.phase != currentPhase) {
927	<	if (head.compareAndSet(p, null)) {
928	<	do {
929	<	Thread t = p.thread;
930	<	if (t != null) {
931	<	p.thread = null;
932	<	LockSupport.unpark(t);
923	>	public boolean isReleasable() {
924	>	Thread t = thread;
925	>	if (t != null) {
926	>	if (phaser.getPhase() != phase)
927	>	t = null;
928	>	else {
929	>	if (Thread.interrupted())
930	>	wasInterrupted = true;
931	>	if (interruptible && wasInterrupted)
932	>	t = null;
933	>	else if (timed) {
934	>	if (nanos > 0) {
935	>	long now = System.nanoTime();
936	>	nanos -= now - lastTime;
937	>	lastTime = now;
938	>	}
939	>	if (nanos <= 0)
940	>	t = null;
941		}
942	<	} while ((p = p.next) != null);
942	>	}
943	>	if (t != null)
944	>	return false;
945	>	thread = null;
946		}
947	+	return true;
948		}
563	–	}
949
950	<	/** The number of CPUs, for spin control */
951	<	static final int NCPUS = Runtime.getRuntime().availableProcessors();
950	>	public boolean block() {
951	>	if (isReleasable())
952	>	return true;
953	>	else if (!timed)
954	>	LockSupport.park(this);
955	>	else if (nanos > 0)
956	>	LockSupport.parkNanos(this, nanos);
957	>	return isReleasable();
958	>	}
959
960	<	/**
961	<	* The number of times to spin before blocking in timed waits.
962	<	* The value is empirically derived.
963	<	*/
964	<	static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
960	>	void signal() {
961	>	Thread t = thread;
962	>	if (t != null) {
963	>	thread = null;
964	>	LockSupport.unpark(t);
965	>	}
966	>	}
967
968	<	/**
969	<	* The number of times to spin before blocking in untimed waits.
970	<	* This is greater than timed value because untimed waits spin
971	<	* faster since they don't need to check times on each spin.
972	<	*/
973	<	static final int maxUntimedSpins = maxTimedSpins * 32;
968	>	void onRelease() { // actions upon return from internalAwaitAdvance
969	>	if (!interruptible && wasInterrupted)
970	>	Thread.currentThread().interrupt();
971	>	if (thread != null)
972	>	thread = null;
973	>	}
974
975	<	/**
582	<	* The number of nanoseconds for which it is faster to spin
583	<	* rather than to use timed park. A rough estimate suffices.
584	<	*/
585	<	static final long spinForTimeoutThreshold = 1000L;
975	>	}
976
977	<	/**
978	<	* Enqueues node and waits unless aborted or signalled.
979	<	*/
980	<	private boolean untimedWait(Thread thread, int currentPhase,
981	<	boolean abortOnInterrupt) {
982	<	final AtomicReference<QNode> head = this.head;
983	<	final AtomicLong state = this.state;
984	<	boolean wasInterrupted = false;
985	<	QNode node = null;
986	<	boolean queued = false;
987	<	int spins = maxUntimedSpins;
988	<	while (phaseOf(state.get()) == currentPhase) {
989	<	QNode h;
990	<	if (node != null && queued) {
601	<	if (node.thread != null) {
602	<	LockSupport.park();
603	<	if (Thread.interrupted()) {
604	<	wasInterrupted = true;
605	<	if (abortOnInterrupt)
606	<	break;
607	<	}
608	<	}
609	<	}
610	<	else if ((h = head.get()) != null && h.phase != currentPhase) {
611	<	if (phaseOf(state.get()) == currentPhase) { // must recheck
612	<	if (head.compareAndSet(h, h.next)) {
613	<	Thread t = h.thread; // help clear out old waiters
614	<	if (t != null) {
615	<	h.thread = null;
616	<	LockSupport.unpark(t);
617	<	}
618	<	}
619	<	}
620	<	else
621	<	break;
622	<	}
623	<	else if (node != null)
624	<	queued = head.compareAndSet(node.next = h, node);
625	<	else if (spins <= 0)
626	<	node = new QNode(thread, currentPhase);
627	<	else
628	<	--spins;
977	>	// Unsafe mechanics
978	>
979	>	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
980	>	private static final long stateOffset =
981	>	objectFieldOffset("state", Phaser.class);
982	>
983	>	private static long objectFieldOffset(String field, Class<?> klazz) {
984	>	try {
985	>	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
986	>	} catch (NoSuchFieldException e) {
987	>	// Convert Exception to corresponding Error
988	>	NoSuchFieldError error = new NoSuchFieldError(field);
989	>	error.initCause(e);
990	>	throw error;
991		}
630	–	if (node != null)
631	–	node.thread = null;
632	–	return wasInterrupted;
992		}
993
994		/**
995	<	* Messier timeout version
995	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
996	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
997	>	* into a jdk.
998	>	*
999	>	* @return a sun.misc.Unsafe
1000		*/
1001	<	private void timedWait(Thread thread, int currentPhase, long nanos)
1002	<	throws InterruptedException, TimeoutException {
1003	<	final AtomicReference<QNode> head = this.head;
1004	<	final AtomicLong state = this.state;
1005	<	long lastTime = System.nanoTime();
1006	<	QNode node = null;
1007	<	boolean queued = false;
1008	<	int spins = maxTimedSpins;
1009	<	while (phaseOf(state.get()) == currentPhase) {
1010	<	QNode h;
1011	<	long now = System.nanoTime();
1012	<	nanos -= now - lastTime;
1013	<	lastTime = now;
1014	<	if (nanos <= 0) {
1015	<	if (node != null)
1016	<	node.thread = null;
1017	<	if (phaseOf(state.get()) == currentPhase)
655	<	throw new TimeoutException();
656	<	else
657	<	break;
658	<	}
659	<	else if (node != null && queued) {
660	<	if (node.thread != null &&
661	<	nanos > spinForTimeoutThreshold) {
662	<	//                LockSupport.parkNanos(this, nanos);
663	<	LockSupport.parkNanos(nanos);
664	<	if (Thread.interrupted()) {
665	<	node.thread = null;
666	<	throw new InterruptedException();
667	<	}
668	<	}
1001	>	private static sun.misc.Unsafe getUnsafe() {
1002	>	try {
1003	>	return sun.misc.Unsafe.getUnsafe();
1004	>	} catch (SecurityException se) {
1005	>	try {
1006	>	return java.security.AccessController.doPrivileged
1007	>	(new java.security
1008	>	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1009	>	public sun.misc.Unsafe run() throws Exception {
1010	>	java.lang.reflect.Field f = sun.misc
1011	>	.Unsafe.class.getDeclaredField("theUnsafe");
1012	>	f.setAccessible(true);
1013	>	return (sun.misc.Unsafe) f.get(null);
1014	>	}});
1015	>	} catch (java.security.PrivilegedActionException e) {
1016	>	throw new RuntimeException("Could not initialize intrinsics",
1017	>	e.getCause());
1018		}
670	–	else if ((h = head.get()) != null && h.phase != currentPhase) {
671	–	if (phaseOf(state.get()) == currentPhase) { // must recheck
672	–	if (head.compareAndSet(h, h.next)) {
673	–	Thread t = h.thread; // help clear out old waiters
674	–	if (t != null) {
675	–	h.thread = null;
676	–	LockSupport.unpark(t);
677	–	}
678	–	}
679	–	}
680	–	else
681	–	break;
682	–	}
683	–	else if (node != null)
684	–	queued = head.compareAndSet(node.next = h, node);
685	–	else if (spins <= 0)
686	–	node = new QNode(thread, currentPhase);
687	–	else
688	–	--spins;
1019		}
690	–	if (node != null)
691	–	node.thread = null;
1020		}
693	–
1021		}
695	–

Diff Legend

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