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 1.63 by dl, Mon Nov 29 15:47:19 2010 UTC vs.
Revision 1.81 by jsr166, Fri Jul 15 18:49:12 2016 UTC

#	Line 1 | Line 1
1		/*
2		* Written by Doug Lea with assistance from members of JCP JSR-166
3		* Expert Group and released to the public domain, as explained at
4	<	* http://creativecommons.org/licenses/publicdomain
4	>	* http://creativecommons.org/publicdomain/zero/1.0/
5		*/
6
7		package jsr166y;
#	Line 17 | Line 17 | import java.util.concurrent.locks.LockSu
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
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
#	Line 30 | Line 30 | import java.util.concurrent.locks.LockSu
30		* (However, you can introduce such bookkeeping by subclassing this
31		* class.)
32		*
33	<	* <p> <b>Synchronization.</b> Like a {@code CyclicBarrier}, a {@code
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
#	Line 44 | Line 44 | import java.util.concurrent.locks.LockSu
44		*
45		* <ul>
46		*
47	<	*   <li> <b>Arrival.</b> Methods {@link #arrive} and
47	>	*   <li><b>Arrival.</b> Methods {@link #arrive} and
48		*       {@link #arriveAndDeregister} record arrival.  These methods
49		*       do not block, but return an associated <em>arrival phase
50		*       number</em>; that is, the phase number of the phaser to which
#	Line 57 | Line 57 | import java.util.concurrent.locks.LockSu
57		*       flexible than, providing a barrier action to a {@code
58		*       CyclicBarrier}.
59		*
60	<	*   <li> <b>Waiting.</b> Method {@link #awaitAdvance} requires an
60	>	*   <li><b>Waiting.</b> Method {@link #awaitAdvance} requires an
61		*       argument indicating an arrival phase number, and returns when
62		*       the phaser advances to (or is already at) a different phase.
63		*       Unlike similar constructions using {@code CyclicBarrier},
#	Line 74 | Line 74 | import java.util.concurrent.locks.LockSu
74		*
75		* </ul>
76		*
77	<	* <p> <b>Termination.</b> A phaser may enter a <em>termination</em>
78	<	* state in which all synchronization methods immediately return
79	<	* without updating phaser state or waiting for advance, and
80	<	* indicating (via a negative phase value) that execution is complete.
77	>	* <p><b>Termination.</b> A phaser may enter a <em>termination</em>
78	>	* state, that may be checked using method {@link #isTerminated}. Upon
79	>	* termination, all synchronization methods immediately return without
80	>	* waiting for advance, as indicated by a negative return value.
81	>	* Similarly, attempts to register upon termination have no effect.
82		* Termination is triggered when an invocation of {@code onAdvance}
83		* returns {@code true}. The default implementation returns {@code
84		* true} if a deregistration has caused the number of registered
#	Line 88 | Line 89 | import java.util.concurrent.locks.LockSu
89		* also available to abruptly release waiting threads and allow them
90		* to terminate.
91		*
92	<	* <p> <b>Tiering.</b> Phasers may be <em>tiered</em> (i.e.,
92	>	* <p><b>Tiering.</b> Phasers may be <em>tiered</em> (i.e.,
93		* constructed in tree structures) to reduce contention. Phasers with
94		* large numbers of parties that would otherwise experience heavy
95		* synchronization contention costs may instead be set up so that
#	Line 96 | Line 97 | import java.util.concurrent.locks.LockSu
97		* increase throughput even though it incurs greater per-operation
98		* overhead.
99		*
100	+	* <p>In a tree of tiered phasers, registration and deregistration of
101	+	* child phasers with their parent are managed automatically.
102	+	* Whenever the number of registered parties of a child phaser becomes
103	+	* non-zero (as established in the {@link #Phaser(Phaser,int)}
104	+	* constructor, {@link #register}, or {@link #bulkRegister}), the
105	+	* child phaser is registered with its parent.  Whenever the number of
106	+	* registered parties becomes zero as the result of an invocation of
107	+	* {@link #arriveAndDeregister}, the child phaser is deregistered
108	+	* from its parent.
109	+	*
110		* <p><b>Monitoring.</b> While synchronization methods may be invoked
111		* only by registered parties, the current state of a phaser may be
112		* monitored by any caller.  At any given moment there are {@link
#	Line 119 | Line 130 | import java.util.concurrent.locks.LockSu
130		* void runTasks(List<Runnable> tasks) {
131		*   final Phaser phaser = new Phaser(1); // "1" to register self
132		*   // create and start threads
133	<	*   for (Runnable task : tasks) {
133	>	*   for (final Runnable task : tasks) {
134		*     phaser.register();
135		*     new Thread() {
136		*       public void run() {
#	Line 182 | Line 193 | import java.util.concurrent.locks.LockSu
193		*   phaser.arriveAndDeregister();
194		* }}</pre>
195		*
196	<	*
197	<	* <p>To create a set of tasks using a tree of phasers,
198	<	* you could use code of the following form, assuming a
199	<	* Task class with a constructor accepting a {@code Phaser} that
200	<	* it registers with upon construction:
196	>	* <p>To create a set of {@code n} tasks using a tree of phasers, you
197	>	* could use code of the following form, assuming a Task class with a
198	>	* constructor accepting a {@code Phaser} that it registers with upon
199	>	* construction. After invocation of {@code build(new Task[n], 0, n,
200	>	* new Phaser())}, these tasks could then be started, for example by
201	>	* submitting to a pool:
202		*
203		*  <pre> {@code
204	<	* void build(Task[] actions, int lo, int hi, Phaser ph) {
204	>	* void build(Task[] tasks, int lo, int hi, Phaser ph) {
205		*   if (hi - lo > TASKS_PER_PHASER) {
206		*     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
207		*       int j = Math.min(i + TASKS_PER_PHASER, hi);
208	<	*       build(actions, i, j, new Phaser(ph));
208	>	*       build(tasks, i, j, new Phaser(ph));
209		*     }
210		*   } else {
211		*     for (int i = lo; i < hi; ++i)
212	<	*       actions[i] = new Task(ph);
212	>	*       tasks[i] = new Task(ph);
213		*       // assumes new Task(ph) performs ph.register()
214		*   }
215	<	* }
204	<	* // .. initially called, for n tasks via
205	<	* build(new Task[n], 0, n, new Phaser());}</pre>
215	>	* }}</pre>
216		*
217		* The best value of {@code TASKS_PER_PHASER} depends mainly on
218		* expected synchronization rates. A value as low as four may
#	Line 226 | Line 236 | public class Phaser {
236		*/
237
238		/**
239	<	* Primary state representation, holding four fields:
239	>	* Primary state representation, holding four bit-fields:
240		*
241	<	* * unarrived -- the number of parties yet to hit barrier (bits  0-15)
242	<	* * parties -- the number of parties to wait              (bits 16-31)
243	<	* * phase -- the generation of the barrier                (bits 32-62)
244	<	* * terminated -- set if barrier is terminated            (bit  63 / sign)
245	<	*
246	<	* However, to efficiently maintain atomicity, these values are
247	<	* packed into a single (atomic) long. Termination uses the sign
248	<	* bit of 32 bit representation of phase, so phase is set to -1 on
249	<	* termination. Good performance relies on keeping state decoding
250	<	* and encoding simple, and keeping race windows short.
241	>	* unarrived  -- the number of parties yet to hit barrier (bits  0-15)
242	>	* parties    -- the number of parties to wait            (bits 16-31)
243	>	* phase      -- the generation of the barrier            (bits 32-62)
244	>	* terminated -- set if barrier is terminated             (bit  63 / sign)
245	>	*
246	>	* Except that a phaser with no registered parties is
247	>	* distinguished by the otherwise illegal state of having zero
248	>	* parties and one unarrived parties (encoded as EMPTY below).
249	>	*
250	>	* To efficiently maintain atomicity, these values are packed into
251	>	* a single (atomic) long. Good performance relies on keeping
252	>	* state decoding and encoding simple, and keeping race windows
253	>	* short.
254	>	*
255	>	* All state updates are performed via CAS except initial
256	>	* registration of a sub-phaser (i.e., one with a non-null
257	>	* parent).  In this (relatively rare) case, we use built-in
258	>	* synchronization to lock while first registering with its
259	>	* parent.
260	>	*
261	>	* The phase of a subphaser is allowed to lag that of its
262	>	* ancestors until it is actually accessed -- see method
263	>	* reconcileState.
264		*/
265		private volatile long state;
266
267		private static final int  MAX_PARTIES     = 0xffff;
268	<	private static final int  MAX_PHASE       = 0x7fffffff;
268	>	private static final int  MAX_PHASE       = Integer.MAX_VALUE;
269		private static final int  PARTIES_SHIFT   = 16;
270		private static final int  PHASE_SHIFT     = 32;
271		private static final int  UNARRIVED_MASK  = 0xffff;      // to mask ints
272		private static final long PARTIES_MASK    = 0xffff0000L; // to mask longs
273	<	private static final long ONE_ARRIVAL     = 1L;
251	<	private static final long ONE_PARTY       = 1L << PARTIES_SHIFT;
273	>	private static final long COUNTS_MASK     = 0xffffffffL;
274		private static final long TERMINATION_BIT = 1L << 63;
275
276	+	// some special values
277	+	private static final int  ONE_ARRIVAL     = 1;
278	+	private static final int  ONE_PARTY       = 1 << PARTIES_SHIFT;
279	+	private static final int  ONE_DEREGISTER  = ONE_ARRIVAL|ONE_PARTY;
280	+	private static final int  EMPTY           = 1;
281	+
282		// The following unpacking methods are usually manually inlined
283
284		private static int unarrivedOf(long s) {
285	<	return (int)s & UNARRIVED_MASK;
285	>	int counts = (int)s;
286	>	return (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
287		}
288
289		private static int partiesOf(long s) {
#	Line 262 | Line 291 | public class Phaser {
291		}
292
293		private static int phaseOf(long s) {
294	<	return (int) (s >>> PHASE_SHIFT);
294	>	return (int)(s >>> PHASE_SHIFT);
295		}
296
297		private static int arrivedOf(long s) {
298	<	return partiesOf(s) - unarrivedOf(s);
298	>	int counts = (int)s;
299	>	return (counts == EMPTY) ? 0 :
300	>	(counts >>> PARTIES_SHIFT) - (counts & UNARRIVED_MASK);
301		}
302
303		/**
#	Line 275 | Line 306 | public class Phaser {
306		private final Phaser parent;
307
308		/**
309	<	* The root of phaser tree. Equals this if not in a tree.  Used to
279	<	* support faster state push-down.
309	>	* The root of phaser tree. Equals this if not in a tree.
310		*/
311		private final Phaser root;
312
#	Line 314 | Line 344 | public class Phaser {
344		* Manually tuned to speed up and minimize race windows for the
345		* common case of just decrementing unarrived field.
346		*
347	<	* @param adj - adjustment to apply to state -- either
348	<	* ONE_ARRIVAL (for arrive) or
349	<	* ONE_ARRIVAL|ONE_PARTY (for arriveAndDeregister)
347	>	* @param adjust value to subtract from state;
348	>	*               ONE_ARRIVAL for arrive,
349	>	*               ONE_DEREGISTER for arriveAndDeregister
350		*/
351	<	private int doArrive(long adj) {
351	>	private int doArrive(int adjust) {
352	>	final Phaser root = this.root;
353		for (;;) {
354	<	long s = state;
324	<	int unarrived = (int)s & UNARRIVED_MASK;
354	>	long s = (root == this) ? state : reconcileState();
355		int phase = (int)(s >>> PHASE_SHIFT);
356		if (phase < 0)
357		return phase;
358	<	else if (unarrived == 0) {
359	<	if (reconcileState() == s)     // recheck
360	<	throw new IllegalStateException(badArrive(s));
361	<	}
362	<	else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adj)) {
358	>	int counts = (int)s;
359	>	int unarrived = (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
360	>	if (unarrived <= 0)
361	>	throw new IllegalStateException(badArrive(s));
362	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adjust)) {
363		if (unarrived == 1) {
364	<	long p = s & PARTIES_MASK; // unshifted parties field
365	<	long lu = p >>> PARTIES_SHIFT;
366	<	int u = (int)lu;
367	<	int nextPhase = (phase + 1) & MAX_PHASE;
368	<	long next = ((long)nextPhase << PHASE_SHIFT) | p | lu;
369	<	final Phaser parent = this.parent;
370	<	if (parent == null) {
371	<	if (onAdvance(phase, u))
372	<	next |= TERMINATION_BIT;
373	<	UNSAFE.compareAndSwapLong(this, stateOffset, s, next);
364	>	long n = s & PARTIES_MASK;  // base of next state
365	>	int nextUnarrived = (int)n >>> PARTIES_SHIFT;
366	>	if (root == this) {
367	>	if (onAdvance(phase, nextUnarrived))
368	>	n |= TERMINATION_BIT;
369	>	else if (nextUnarrived == 0)
370	>	n |= EMPTY;
371	>	else
372	>	n |= nextUnarrived;
373	>	int nextPhase = (phase + 1) & MAX_PHASE;
374	>	n |= (long)nextPhase << PHASE_SHIFT;
375	>	UNSAFE.compareAndSwapLong(this, stateOffset, s, n);
376		releaseWaiters(phase);
377		}
378	<	else {
379	<	parent.doArrive((u == 0) ?
380	<	ONE_ARRIVAL|ONE_PARTY : ONE_ARRIVAL);
381	<	if ((int)(parent.state >>> PHASE_SHIFT) != nextPhase)
350	<	reconcileState();
351	<	else if (state == s)
352	<	UNSAFE.compareAndSwapLong(this, stateOffset, s,
353	<	next);
378	>	else if (nextUnarrived == 0) { // propagate deregistration
379	>	phase = parent.doArrive(ONE_DEREGISTER);
380	>	UNSAFE.compareAndSwapLong(this, stateOffset,
381	>	s, s | EMPTY);
382		}
383	+	else
384	+	phase = parent.doArrive(ONE_ARRIVAL);
385		}
386		return phase;
387		}
#	Line 366 | Line 396 | public class Phaser {
396		*/
397		private int doRegister(int registrations) {
398		// adjustment to state
399	<	long adj = ((long)registrations << PARTIES_SHIFT) | registrations;
399	>	long adjust = ((long)registrations << PARTIES_SHIFT) | registrations;
400		final Phaser parent = this.parent;
401	+	int phase;
402		for (;;) {
403		long s = (parent == null) ? state : reconcileState();
404	<	int parties = (int)s >>> PARTIES_SHIFT;
405	<	int phase = (int)(s >>> PHASE_SHIFT);
406	<	if (phase < 0)
407	<	return phase;
377	<	else if (registrations > MAX_PARTIES - parties)
404	>	int counts = (int)s;
405	>	int parties = counts >>> PARTIES_SHIFT;
406	>	int unarrived = counts & UNARRIVED_MASK;
407	>	if (registrations > MAX_PARTIES - parties)
408		throw new IllegalStateException(badRegister(s));
409	<	else if ((parties == 0 && parent == null) || // first reg of root
410	<	((int)s & UNARRIVED_MASK) != 0) {   // not advancing
411	<	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s + adj))
412	<	return phase;
409	>	phase = (int)(s >>> PHASE_SHIFT);
410	>	if (phase < 0)
411	>	break;
412	>	if (counts != EMPTY) {                  // not 1st registration
413	>	if (parent == null || reconcileState() == s) {
414	>	if (unarrived == 0)             // wait out advance
415	>	root.internalAwaitAdvance(phase, null);
416	>	else if (UNSAFE.compareAndSwapLong(this, stateOffset,
417	>	s, s + adjust))
418	>	break;
419	>	}
420	>	}
421	>	else if (parent == null) {              // 1st root registration
422	>	long next = ((long)phase << PHASE_SHIFT) | adjust;
423	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next))
424	>	break;
425		}
426	<	else if (parties != 0)               // wait for onAdvance
427	<	root.internalAwaitAdvance(phase, null);
428	<	else {                               // 1st registration of child
429	<	synchronized (this) {            // register parent first
430	<	if (reconcileState() == s) { // recheck under lock
431	<	parent.doRegister(1);    // OK if throws IllegalState
432	<	for (;;) {               // simpler form of outer loop
433	<	s = reconcileState();
434	<	phase = (int)(s >>> PHASE_SHIFT);
435	<	if (phase < 0 ||
436	<	UNSAFE.compareAndSwapLong(this, stateOffset,
437	<	s, s + adj))
438	<	return phase;
426	>	else {
427	>	synchronized (this) {               // 1st sub registration
428	>	if (state == s) {               // recheck under lock
429	>	phase = parent.doRegister(1);
430	>	if (phase < 0)
431	>	break;
432	>	// finish registration whenever parent registration
433	>	// succeeded, even when racing with termination,
434	>	// since these are part of the same "transaction".
435	>	while (!UNSAFE.compareAndSwapLong
436	>	(this, stateOffset, s,
437	>	((long)phase << PHASE_SHIFT) | adjust)) {
438	>	s = state;
439	>	phase = (int)(root.state >>> PHASE_SHIFT);
440	>	// assert (int)s == EMPTY;
441		}
442	+	break;
443		}
444		}
445		}
446		}
447	+	return phase;
448		}
449
450		/**
451	<	* Recursively resolves lagged phase propagation from root if necessary.
451	>	* Resolves lagged phase propagation from root if necessary.
452	>	* Reconciliation normally occurs when root has advanced but
453	>	* subphasers have not yet done so, in which case they must finish
454	>	* their own advance by setting unarrived to parties (or if
455	>	* parties is zero, resetting to unregistered EMPTY state).
456	>	*
457	>	* @return reconciled state
458		*/
459		private long reconcileState() {
460	<	Phaser par = parent;
460	>	final Phaser root = this.root;
461		long s = state;
462	<	if (par != null) {
463	<	Phaser rt = root;
464	<	int phase, rPhase;
465	<	while ((phase = (int)(s >>> PHASE_SHIFT)) >= 0 &&
466	<	(rPhase = (int)(rt.state >>> PHASE_SHIFT)) != phase) {
467	<	if (par != rt && (int)(par.state >>> PHASE_SHIFT) != rPhase)
468	<	par.reconcileState();
469	<	else if (rPhase < 0 || ((int)s & UNARRIVED_MASK) == 0) {
470	<	long u = s & PARTIES_MASK; // reset unarrived to parties
471	<	long next = ((((long) rPhase) << PHASE_SHIFT) | u |
472	<	(u >>> PARTIES_SHIFT));
421	<	UNSAFE.compareAndSwapLong(this, stateOffset, s, next);
422	<	}
462	>	if (root != this) {
463	>	int phase, p;
464	>	// CAS to root phase with current parties, tripping unarrived
465	>	while ((phase = (int)(root.state >>> PHASE_SHIFT)) !=
466	>	(int)(s >>> PHASE_SHIFT) &&
467	>	!UNSAFE.compareAndSwapLong
468	>	(this, stateOffset, s,
469	>	s = (((long)phase << PHASE_SHIFT) |
470	>	((phase < 0) ? (s & COUNTS_MASK) :
471	>	(((p = (int)s >>> PARTIES_SHIFT) == 0) ? EMPTY :
472	>	((s & PARTIES_MASK) | p))))))
473		s = state;
424	–	}
474		}
475		return s;
476		}
#	Line 459 | Line 508 | public class Phaser {
508
509		/**
510		* Creates a new phaser with the given parent and number of
511	<	* registered unarrived parties. Registration and deregistration
512	<	* of this child phaser with its parent are managed automatically.
513	<	* If the given parent is non-null, whenever this child phaser has
465	<	* any registered parties (as established in this constructor,
466	<	* {@link #register}, or {@link #bulkRegister}), this child phaser
467	<	* is registered with its parent. Whenever the number of
468	<	* registered parties becomes zero as the result of an invocation
469	<	* of {@link #arriveAndDeregister}, this child phaser is
470	<	* deregistered from its parent.
511	>	* registered unarrived parties.  When the given parent is non-null
512	>	* and the given number of parties is greater than zero, this
513	>	* child phaser is registered with its parent.
514		*
515		* @param parent the parent phaser
516		* @param parties the number of parties required to advance to the
#	Line 478 | Line 521 | public class Phaser {
521		public Phaser(Phaser parent, int parties) {
522		if (parties >>> PARTIES_SHIFT != 0)
523		throw new IllegalArgumentException("Illegal number of parties");
524	<	long s = ((long) parties) | (((long) parties) << PARTIES_SHIFT);
524	>	int phase = 0;
525		this.parent = parent;
526		if (parent != null) {
527	<	Phaser r = parent.root;
528	<	this.root = r;
529	<	this.evenQ = r.evenQ;
530	<	this.oddQ = r.oddQ;
527	>	final Phaser root = parent.root;
528	>	this.root = root;
529	>	this.evenQ = root.evenQ;
530	>	this.oddQ = root.oddQ;
531		if (parties != 0)
532	<	s |= ((long)(parent.doRegister(1))) << PHASE_SHIFT;
532	>	phase = parent.doRegister(1);
533		}
534		else {
535		this.root = this;
536		this.evenQ = new AtomicReference<QNode>();
537		this.oddQ = new AtomicReference<QNode>();
538		}
539	<	this.state = s;
539	>	this.state = (parties == 0) ? (long)EMPTY :
540	>	((long)phase << PHASE_SHIFT) |
541	>	((long)parties << PARTIES_SHIFT) |
542	>	((long)parties);
543		}
544
545		/**
#	Line 501 | Line 547 | public class Phaser {
547		* invocation of {@link #onAdvance} is in progress, this method
548		* may await its completion before returning.  If this phaser has
549		* a parent, and this phaser previously had no registered parties,
550	<	* this phaser is also registered with its parent.
551	<	*
552	<	* @return the arrival phase number to which this registration applied
550	>	* this child phaser is also registered with its parent. If
551	>	* this phaser is terminated, the attempt to register has
552	>	* no effect, and a negative value is returned.
553	>	*
554	>	* @return the arrival phase number to which this registration
555	>	* applied.  If this value is negative, then this phaser has
556	>	* terminated, in which case registration has no effect.
557		* @throws IllegalStateException if attempting to register more
558		* than the maximum supported number of parties
559		*/
#	Line 515 | Line 565 | public class Phaser {
565		* Adds the given number of new unarrived parties to this phaser.
566		* If an ongoing invocation of {@link #onAdvance} is in progress,
567		* this method may await its completion before returning.  If this
568	<	* phaser has a parent, and the given number of parities is
569	<	* greater than zero, and this phaser previously had no registered
570	<	* parties, this phaser is also registered with its parent.
568	>	* phaser has a parent, and the given number of parties is greater
569	>	* than zero, and this phaser previously had no registered
570	>	* parties, this child phaser is also registered with its parent.
571	>	* If this phaser is terminated, the attempt to register has no
572	>	* effect, and a negative value is returned.
573		*
574		* @param parties the number of additional parties required to
575		* advance to the next phase
576	<	* @return the arrival phase number to which this registration applied
576	>	* @return the arrival phase number to which this registration
577	>	* applied.  If this value is negative, then this phaser has
578	>	* terminated, in which case registration has no effect.
579		* @throws IllegalStateException if attempting to register more
580		* than the maximum supported number of parties
581		* @throws IllegalArgumentException if {@code parties < 0}
#	Line 567 | Line 621 | public class Phaser {
621		* of registered or unarrived parties would become negative
622		*/
623		public int arriveAndDeregister() {
624	<	return doArrive(ONE_ARRIVAL|ONE_PARTY);
624	>	return doArrive(ONE_DEREGISTER);
625		}
626
627		/**
#	Line 583 | Line 637 | public class Phaser {
637		* IllegalStateException} only upon some subsequent operation on
638		* this phaser, if ever.
639		*
640	<	* @return the arrival phase number, or a negative number if terminated
640	>	* @return the arrival phase number, or the (negative)
641	>	* {@linkplain #getPhase() current phase} if terminated
642		* @throws IllegalStateException if not terminated and the number
643		* of unarrived parties would become negative
644		*/
645		public int arriveAndAwaitAdvance() {
646	<	return awaitAdvance(doArrive(ONE_ARRIVAL));
646	>	// Specialization of doArrive+awaitAdvance eliminating some reads/paths
647	>	final Phaser root = this.root;
648	>	for (;;) {
649	>	long s = (root == this) ? state : reconcileState();
650	>	int phase = (int)(s >>> PHASE_SHIFT);
651	>	if (phase < 0)
652	>	return phase;
653	>	int counts = (int)s;
654	>	int unarrived = (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
655	>	if (unarrived <= 0)
656	>	throw new IllegalStateException(badArrive(s));
657	>	if (UNSAFE.compareAndSwapLong(this, stateOffset, s,
658	>	s -= ONE_ARRIVAL)) {
659	>	if (unarrived > 1)
660	>	return root.internalAwaitAdvance(phase, null);
661	>	if (root != this)
662	>	return parent.arriveAndAwaitAdvance();
663	>	long n = s & PARTIES_MASK;  // base of next state
664	>	int nextUnarrived = (int)n >>> PARTIES_SHIFT;
665	>	if (onAdvance(phase, nextUnarrived))
666	>	n |= TERMINATION_BIT;
667	>	else if (nextUnarrived == 0)
668	>	n |= EMPTY;
669	>	else
670	>	n |= nextUnarrived;
671	>	int nextPhase = (phase + 1) & MAX_PHASE;
672	>	n |= (long)nextPhase << PHASE_SHIFT;
673	>	if (!UNSAFE.compareAndSwapLong(this, stateOffset, s, n))
674	>	return (int)(state >>> PHASE_SHIFT); // terminated
675	>	releaseWaiters(phase);
676	>	return nextPhase;
677	>	}
678	>	}
679		}
680
681		/**
#	Line 599 | Line 686 | public class Phaser {
686		* @param phase an arrival phase number, or negative value if
687		* terminated; this argument is normally the value returned by a
688		* previous call to {@code arrive} or {@code arriveAndDeregister}.
689	<	* @return the next arrival phase number, or a negative value
690	<	* if terminated or argument is negative
689	>	* @return the next arrival phase number, or the argument if it is
690	>	* negative, or the (negative) {@linkplain #getPhase() current phase}
691	>	* if terminated
692		*/
693		public int awaitAdvance(int phase) {
694	<	Phaser rt;
695	<	int p = (int)(state >>> PHASE_SHIFT);
694	>	final Phaser root = this.root;
695	>	long s = (root == this) ? state : reconcileState();
696	>	int p = (int)(s >>> PHASE_SHIFT);
697		if (phase < 0)
698		return phase;
699	<	if (p == phase &&
700	<	(p = (int)((rt = root).state >>> PHASE_SHIFT)) == phase)
612	<	return rt.internalAwaitAdvance(phase, null);
699	>	if (p == phase)
700	>	return root.internalAwaitAdvance(phase, null);
701		return p;
702		}
703
#	Line 623 | Line 711 | public class Phaser {
711		* @param phase an arrival phase number, or negative value if
712		* terminated; this argument is normally the value returned by a
713		* previous call to {@code arrive} or {@code arriveAndDeregister}.
714	<	* @return the next arrival phase number, or a negative value
715	<	* if terminated or argument is negative
714	>	* @return the next arrival phase number, or the argument if it is
715	>	* negative, or the (negative) {@linkplain #getPhase() current phase}
716	>	* if terminated
717		* @throws InterruptedException if thread interrupted while waiting
718		*/
719		public int awaitAdvanceInterruptibly(int phase)
720		throws InterruptedException {
721	<	Phaser rt;
722	<	int p = (int)(state >>> PHASE_SHIFT);
721	>	final Phaser root = this.root;
722	>	long s = (root == this) ? state : reconcileState();
723	>	int p = (int)(s >>> PHASE_SHIFT);
724		if (phase < 0)
725		return phase;
726	<	if (p == phase &&
637	<	(p = (int)((rt = root).state >>> PHASE_SHIFT)) == phase) {
726	>	if (p == phase) {
727		QNode node = new QNode(this, phase, true, false, 0L);
728	<	p = rt.internalAwaitAdvance(phase, node);
728	>	p = root.internalAwaitAdvance(phase, node);
729		if (node.wasInterrupted)
730		throw new InterruptedException();
731		}
#	Line 657 | Line 746 | public class Phaser {
746		*        {@code unit}
747		* @param unit a {@code TimeUnit} determining how to interpret the
748		*        {@code timeout} parameter
749	<	* @return the next arrival phase number, or a negative value
750	<	* if terminated or argument is negative
749	>	* @return the next arrival phase number, or the argument if it is
750	>	* negative, or the (negative) {@linkplain #getPhase() current phase}
751	>	* if terminated
752		* @throws InterruptedException if thread interrupted while waiting
753		* @throws TimeoutException if timed out while waiting
754		*/
#	Line 666 | Line 756 | public class Phaser {
756		long timeout, TimeUnit unit)
757		throws InterruptedException, TimeoutException {
758		long nanos = unit.toNanos(timeout);
759	<	Phaser rt;
760	<	int p = (int)(state >>> PHASE_SHIFT);
759	>	final Phaser root = this.root;
760	>	long s = (root == this) ? state : reconcileState();
761	>	int p = (int)(s >>> PHASE_SHIFT);
762		if (phase < 0)
763		return phase;
764	<	if (p == phase &&
674	<	(p = (int)((rt = root).state >>> PHASE_SHIFT)) == phase) {
764	>	if (p == phase) {
765		QNode node = new QNode(this, phase, true, true, nanos);
766	<	p = rt.internalAwaitAdvance(phase, node);
766	>	p = root.internalAwaitAdvance(phase, node);
767		if (node.wasInterrupted)
768		throw new InterruptedException();
769		else if (p == phase)
#	Line 684 | Line 774 | public class Phaser {
774
775		/**
776		* Forces this phaser to enter termination state.  Counts of
777	<	* arrived and registered parties are unaffected.  If this phaser
778	<	* is a member of a tiered set of phasers, then all of the phasers
779	<	* in the set are terminated.  If this phaser is already
780	<	* terminated, this method has no effect.  This method may be
781	<	* useful for coordinating recovery after one or more tasks
782	<	* encounter unexpected exceptions.
777	>	* registered parties are unaffected.  If this phaser is a member
778	>	* of a tiered set of phasers, then all of the phasers in the set
779	>	* are terminated.  If this phaser is already terminated, this
780	>	* method has no effect.  This method may be useful for
781	>	* coordinating recovery after one or more tasks encounter
782	>	* unexpected exceptions.
783		*/
784		public void forceTermination() {
785		// Only need to change root state
#	Line 698 | Line 788 | public class Phaser {
788		while ((s = root.state) >= 0) {
789		if (UNSAFE.compareAndSwapLong(root, stateOffset,
790		s, s | TERMINATION_BIT)) {
791	<	releaseWaiters(0); // signal all threads
792	<	releaseWaiters(1);
791	>	// signal all threads
792	>	releaseWaiters(0); // Waiters on evenQ
793	>	releaseWaiters(1); // Waiters on oddQ
794		return;
795		}
796		}
#	Line 729 | Line 820 | public class Phaser {
820
821		/**
822		* Returns the number of registered parties that have arrived at
823	<	* the current phase of this phaser.
823	>	* the current phase of this phaser. If this phaser has terminated,
824	>	* the returned value is meaningless and arbitrary.
825		*
826		* @return the number of arrived parties
827		*/
828		public int getArrivedParties() {
829	<	long s = state;
738	<	int u = unarrivedOf(s); // only reconcile if possibly needed
739	<	return (u != 0 || parent == null) ?
740	<	partiesOf(s) - u :
741	<	arrivedOf(reconcileState());
829	>	return arrivedOf(reconcileState());
830		}
831
832		/**
833		* Returns the number of registered parties that have not yet
834	<	* arrived at the current phase of this phaser.
834	>	* arrived at the current phase of this phaser. If this phaser has
835	>	* terminated, the returned value is meaningless and arbitrary.
836		*
837		* @return the number of unarrived parties
838		*/
839		public int getUnarrivedParties() {
840	<	int u = unarrivedOf(state);
752	<	return (u != 0 || parent == null) ? u : unarrivedOf(reconcileState());
840	>	return unarrivedOf(reconcileState());
841		}
842
843		/**
#	Line 785 | Line 873 | public class Phaser {
873		* advance, and to control termination. This method is invoked
874		* upon arrival of the party advancing this phaser (when all other
875		* waiting parties are dormant).  If this method returns {@code
876	<	* true}, then, rather than advance the phase number, this phaser
877	<	* will be set to a final termination state, and subsequent calls
878	<	* to {@link #isTerminated} will return true. Any (unchecked)
879	<	* Exception or Error thrown by an invocation of this method is
880	<	* propagated to the party attempting to advance this phaser, in
881	<	* which case no advance occurs.
876	>	* true}, this phaser will be set to a final termination state
877	>	* upon advance, and subsequent calls to {@link #isTerminated}
878	>	* will return true. Any (unchecked) Exception or Error thrown by
879	>	* an invocation of this method is propagated to the party
880	>	* attempting to advance this phaser, in which case no advance
881	>	* occurs.
882		*
883		* <p>The arguments to this method provide the state of the phaser
884		* prevailing for the current transition.  The effects of invoking
#	Line 854 | Line 942 | public class Phaser {
942		*/
943		private void releaseWaiters(int phase) {
944		QNode q;   // first element of queue
857	–	int p;     // its phase
945		Thread t;  // its thread
946		AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
947		while ((q = head.get()) != null &&
948	<	((p = q.phase) == phase ||
862	<	(int)(root.state >>> PHASE_SHIFT) != p)) {
948	>	q.phase != (int)(root.state >>> PHASE_SHIFT)) {
949		if (head.compareAndSet(q, q.next) &&
950		(t = q.thread) != null) {
951		q.thread = null;
#	Line 868 | Line 954 | public class Phaser {
954		}
955		}
956
957	+	/**
958	+	* Variant of releaseWaiters that additionally tries to remove any
959	+	* nodes no longer waiting for advance due to timeout or
960	+	* interrupt. Currently, nodes are removed only if they are at
961	+	* head of queue, which suffices to reduce memory footprint in
962	+	* most usages.
963	+	*
964	+	* @return current phase on exit
965	+	*/
966	+	private int abortWait(int phase) {
967	+	AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
968	+	for (;;) {
969	+	Thread t;
970	+	QNode q = head.get();
971	+	int p = (int)(root.state >>> PHASE_SHIFT);
972	+	if (q == null || ((t = q.thread) != null && q.phase == p))
973	+	return p;
974	+	if (head.compareAndSet(q, q.next) && t != null) {
975	+	q.thread = null;
976	+	LockSupport.unpark(t);
977	+	}
978	+	}
979	+	}
980	+
981		/** The number of CPUs, for spin control */
982		private static final int NCPU = Runtime.getRuntime().availableProcessors();
983
#	Line 886 | Line 996 | public class Phaser {
996
997		/**
998		* Possibly blocks and waits for phase to advance unless aborted.
999	<	* Call only from root node.
999	>	* Call only on root phaser.
1000		*
1001		* @param phase current phase
1002		* @param node if non-null, the wait node to track interrupt and timeout;
#	Line 894 | Line 1004 | public class Phaser {
1004		* @return current phase
1005		*/
1006		private int internalAwaitAdvance(int phase, QNode node) {
1007	+	// assert root == this;
1008		releaseWaiters(phase-1);          // ensure old queue clean
1009		boolean queued = false;           // true when node is enqueued
1010		int lastUnarrived = 0;            // to increase spins upon change
#	Line 935 | Line 1046 | public class Phaser {
1046		node.thread = null;       // avoid need for unpark()
1047		if (node.wasInterrupted && !node.interruptible)
1048		Thread.currentThread().interrupt();
1049	<	if ((p = (int)(state >>> PHASE_SHIFT)) == phase)
1050	<	return p;                 // recheck abort
1049	>	if (p == phase && (p = (int)(state >>> PHASE_SHIFT)) == phase)
1050	>	return abortWait(phase); // possibly clean up on abort
1051		}
1052		releaseWaiters(phase);
1053		return p;
#	Line 963 | Line 1074 | public class Phaser {
1074		this.interruptible = interruptible;
1075		this.nanos = nanos;
1076		this.timed = timed;
1077	<	this.lastTime = timed? System.nanoTime() : 0L;
1077	>	this.lastTime = timed ? System.nanoTime() : 0L;
1078		thread = Thread.currentThread();
1079		}
1080
#	Line 1007 | Line 1118 | public class Phaser {
1118
1119		// Unsafe mechanics
1120
1121	<	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1122	<	private static final long stateOffset =
1123	<	objectFieldOffset("state", Phaser.class);
1013	<
1014	<	private static long objectFieldOffset(String field, Class<?> klazz) {
1121	>	private static final sun.misc.Unsafe UNSAFE;
1122	>	private static final long stateOffset;
1123	>	static {
1124		try {
1125	<	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1126	<	} catch (NoSuchFieldException e) {
1127	<	// Convert Exception to corresponding Error
1128	<	NoSuchFieldError error = new NoSuchFieldError(field);
1129	<	error.initCause(e);
1130	<	throw error;
1125	>	UNSAFE = getUnsafe();
1126	>	Class<?> k = Phaser.class;
1127	>	stateOffset = UNSAFE.objectFieldOffset
1128	>	(k.getDeclaredField("state"));
1129	>	} catch (Exception e) {
1130	>	throw new Error(e);
1131		}
1132		}
1133
#	Line 1032 | Line 1141 | public class Phaser {
1141		private static sun.misc.Unsafe getUnsafe() {
1142		try {
1143		return sun.misc.Unsafe.getUnsafe();
1144	<	} catch (SecurityException se) {
1145	<	try {
1146	<	return java.security.AccessController.doPrivileged
1147	<	(new java.security
1148	<	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1149	<	public sun.misc.Unsafe run() throws Exception {
1150	<	java.lang.reflect.Field f = sun.misc
1151	<	.Unsafe.class.getDeclaredField("theUnsafe");
1152	<	f.setAccessible(true);
1153	<	return (sun.misc.Unsafe) f.get(null);
1154	<	}});
1155	<	} catch (java.security.PrivilegedActionException e) {
1156	<	throw new RuntimeException("Could not initialize intrinsics",
1157	<	e.getCause());
1158	<	}
1144	>	} catch (SecurityException tryReflectionInstead) {}
1145	>	try {
1146	>	return java.security.AccessController.doPrivileged
1147	>	(new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1148	>	public sun.misc.Unsafe run() throws Exception {
1149	>	Class<sun.misc.Unsafe> k = sun.misc.Unsafe.class;
1150	>	for (java.lang.reflect.Field f : k.getDeclaredFields()) {
1151	>	f.setAccessible(true);
1152	>	Object x = f.get(null);
1153	>	if (k.isInstance(x))
1154	>	return k.cast(x);
1155	>	}
1156	>	throw new NoSuchFieldError("the Unsafe");
1157	>	}});
1158	>	} catch (java.security.PrivilegedActionException e) {
1159	>	throw new RuntimeException("Could not initialize intrinsics",
1160	>	e.getCause());
1161		}
1162		}
1163		}

Diff Legend

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