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 1.5 by dl, Sun Sep 7 11:24:26 2008 UTC vs.
Revision 1.24 by jsr166, Mon Jul 27 21:41:53 2009 UTC

#	Line 5 | Line 5
5		*/
6
7		package jsr166y;
8	+
9		import java.util.concurrent.*;
10	<	import java.util.concurrent.atomic.*;
10	>
11	>	import java.util.concurrent.atomic.AtomicReference;
12		import java.util.concurrent.locks.LockSupport;
11	–	import sun.misc.Unsafe;
12	–	import java.lang.reflect.*;
13
14		/**
15		* A reusable synchronization barrier, similar in functionality to a
16	<	* {@link java.util.concurrent.CyclicBarrier} and {@link
17	<	* java.util.concurrent.CountDownLatch} but supporting more flexible
18	<	* usage.
16	>	* {@link java.util.concurrent.CyclicBarrier CyclicBarrier} and
17	>	* {@link java.util.concurrent.CountDownLatch CountDownLatch}
18	>	* but supporting more flexible usage.
19		*
20		* <ul>
21		*
#	Line 25 | Line 25 | import java.lang.reflect.*;
25		* basic synchronization constructs, registration and deregistration
26		* affect only internal counts; they do not establish any further
27		* internal bookkeeping, so tasks cannot query whether they are
28	<	* registered. (However, you can introduce such bookkeeping in by
28	>	* registered. (However, you can introduce such bookkeeping by
29		* subclassing this class.)
30		*
31		* <li> Each generation has an associated phase value, starting at
32		* zero, and advancing when all parties reach the barrier (wrapping
33	<	* around to zero after reaching <tt>Integer.MAX_VALUE</tt>).
33	>	* around to zero after reaching {@code Integer.MAX_VALUE}).
34		*
35	<	* <li> Like a CyclicBarrier, a Phaser may be repeatedly awaited.
36	<	* Method <tt>arriveAndAwaitAdvance</tt> has effect analogous to
37	<	* <tt>CyclicBarrier.await</tt>.  However, Phasers separate two
38	<	* aspects of coordination, that may also be invoked independently:
35	>	* <li> Like a {@code CyclicBarrier}, a Phaser may be repeatedly
36	>	* awaited.  Method {@link #arriveAndAwaitAdvance} has effect
37	>	* analogous to {@link java.util.concurrent.CyclicBarrier#await
38	>	* CyclicBarrier.await}.  However, phasers separate two aspects of
39	>	* coordination, that may also be invoked independently:
40		*
41		* <ul>
42		*
43	<	*   <li> Arriving at a barrier. Methods <tt>arrive</tt> and
44	<	*       <tt>arriveAndDeregister</tt> do not block, but return
43	>	*   <li> Arriving at a barrier. Methods {@link #arrive} and
44	>	*       {@link #arriveAndDeregister} do not block, but return
45		*       the phase value current upon entry to the method.
46		*
47	<	*   <li> Awaiting others. Method <tt>awaitAdvance</tt> requires an
47	>	*   <li> Awaiting others. Method {@link #awaitAdvance} requires an
48		*       argument indicating the entry phase, and returns when the
49		*       barrier advances to a new phase.
50		* </ul>
#	Line 51 | Line 52 | import java.lang.reflect.*;
52		*
53		* <li> Barrier actions, performed by the task triggering a phase
54		* advance while others may be waiting, are arranged by overriding
55	<	* method <tt>onAdvance</tt>, that also controls termination.
55	>	* method {@link #onAdvance}, that also controls termination.
56		* Overriding this method may be used to similar but more flexible
57	<	* effect as providing a barrier action to a CyclicBarrier.
57	>	* effect as providing a barrier action to a {@code CyclicBarrier}.
58		*
59		* <li> Phasers may enter a <em>termination</em> state in which all
60	<	* await actions immediately return, indicating (via a negative phase
61	<	* value) that execution is complete.  Termination is triggered by
62	<	* executing the overridable <tt>onAdvance</tt> method that is invoked
63	<	* each time the barrier is about to be tripped. When a Phaser is
64	<	* controlling an action with a fixed number of iterations, it is
65	<	* often convenient to override this method to cause termination when
66	<	* the current phase number reaches a threshold. Method
67	<	* <tt>forceTermination</tt> is also available to abruptly release
68	<	* waiting threads and allow them to terminate.
60	>	* actions immediately return without updating phaser state or waiting
61	>	* for advance, and indicating (via a negative phase value) that
62	>	* execution is complete.  Termination is triggered by executing the
63	>	* overridable {@code onAdvance} method that is invoked each time the
64	>	* barrier is about to be tripped. When a phaser is controlling an
65	>	* action with a fixed number of iterations, it is often convenient to
66	>	* override this method to cause termination when the current phase
67	>	* number reaches a threshold. Method {@link #forceTermination} is also
68	>	* available to abruptly release waiting threads and allow them to
69	>	* terminate.
70		*
71		* <li> Phasers may be tiered to reduce contention. Phasers with large
72		* numbers of parties that would otherwise experience heavy
#	Line 72 | Line 74 | import java.lang.reflect.*;
74		* This will typically greatly increase throughput even though it
75		* incurs somewhat greater per-operation overhead.
76		*
77	<	* <li> By default, <tt>awaitAdvance</tt> continues to wait even if
77	>	* <li> By default, {@code awaitAdvance} continues to wait even if
78		* the waiting thread is interrupted. And unlike the case in
79	<	* CyclicBarriers, exceptions encountered while tasks wait
79	>	* {@code CyclicBarrier}, exceptions encountered while tasks wait
80		* interruptibly or with timeout do not change the state of the
81		* barrier. If necessary, you can perform any associated recovery
82		* within handlers of those exceptions, often after invoking
83	<	* <tt>forceTermination</tt>.
83	>	* {@code forceTermination}.
84	>	*
85	>	* <li>Phasers ensure lack of starvation when used by ForkJoinTasks.
86		*
87		* </ul>
88		*
89		* <p><b>Sample usages:</b>
90		*
91	<	* <p>A Phaser may be used instead of a <tt>CountdownLatch</tt> to control
92	<	* a one-shot action serving a variable number of parties. The typical
93	<	* idiom is for the method setting this up to first register, then
94	<	* start the actions, then deregister, as in:
95	<	*
96	<	* <pre>
97	<	*  void runTasks(List&lt;Runnable&gt; list) {
98	<	*    final Phaser phaser = new Phaser(1); // "1" to register self
99	<	*    for (Runnable r : list) {
100	<	*      phaser.register();
101	<	*      new Thread() {
102	<	*        public void run() {
103	<	*          phaser.arriveAndAwaitAdvance(); // await all creation
104	<	*          r.run();
105	<	*          phaser.arriveAndDeregister();   // signal completion
106	<	*        }
107	<	*      }.start();
91	>	* <p>A {@code Phaser} may be used instead of a {@code CountDownLatch}
92	>	* to control a one-shot action serving a variable number of
93	>	* parties. The typical idiom is for the method setting this up to
94	>	* first register, then start the actions, then deregister, as in:
95	>	*
96	>	*  <pre> {@code
97	>	* void runTasks(List<Runnable> list) {
98	>	*   final Phaser phaser = new Phaser(1); // "1" to register self
99	>	*   for (Runnable r : list) {
100	>	*     phaser.register();
101	>	*     new Thread() {
102	>	*       public void run() {
103	>	*         phaser.arriveAndAwaitAdvance(); // await all creation
104	>	*         r.run();
105	>	*         phaser.arriveAndDeregister();   // signal completion
106	>	*       }
107	>	*     }.start();
108		*   }
109	+	*
110	+	*   doSomethingOnBehalfOfWorkers();
111		*   phaser.arrive(); // allow threads to start
112	<	*   int p = phaser.arriveAndDeregister(); // deregister self
112	>	*   int p = phaser.arriveAndDeregister(); // deregister self  ...
113	>	*   p = phaser.awaitAdvance(p); // ... and await arrival
114		*   otherActions(); // do other things while tasks execute
115	<	*   phaser.awaitAdvance(p); // wait for all tasks to arrive
116	<	* }
110	<	* </pre>
115	>	*   phaser.awaitAdvance(p); // await final completion
116	>	* }}</pre>
117		*
118		* <p>One way to cause a set of threads to repeatedly perform actions
119	<	* for a given number of iterations is to override <tt>onAdvance</tt>:
119	>	* for a given number of iterations is to override {@code onAdvance}:
120		*
121	<	* <pre>
122	<	*  void startTasks(List&lt;Runnable&gt; list, final int iterations) {
123	<	*    final Phaser phaser = new Phaser() {
124	<	*       public boolean onAdvance(int phase, int registeredParties) {
125	<	*         return phase &gt;= iterations || registeredParties == 0;
121	>	*  <pre> {@code
122	>	* void startTasks(List<Runnable> list, final int iterations) {
123	>	*   final Phaser phaser = new Phaser() {
124	>	*     public boolean onAdvance(int phase, int registeredParties) {
125	>	*       return phase >= iterations || registeredParties == 0;
126	>	*     }
127	>	*   };
128	>	*   phaser.register();
129	>	*   for (Runnable r : list) {
130	>	*     phaser.register();
131	>	*     new Thread() {
132	>	*       public void run() {
133	>	*         do {
134	>	*           r.run();
135	>	*           phaser.arriveAndAwaitAdvance();
136	>	*         } while(!phaser.isTerminated();
137		*       }
138	<	*    };
122	<	*    phaser.register();
123	<	*    for (Runnable r : list) {
124	<	*      phaser.register();
125	<	*      new Thread() {
126	<	*        public void run() {
127	<	*           do {
128	<	*             r.run();
129	<	*             phaser.arriveAndAwaitAdvance();
130	<	*           } while(!phaser.isTerminated();
131	<	*        }
132	<	*      }.start();
138	>	*     }.start();
139		*   }
140		*   phaser.arriveAndDeregister(); // deregister self, don't wait
141	<	* }
136	<	* </pre>
141	>	* }}</pre>
142		*
143	<	* <p> To create a set of tasks using a tree of Phasers,
143	>	* <p> To create a set of tasks using a tree of phasers,
144		* you could use code of the following form, assuming a
145	<	* Task class with a constructor accepting a Phaser that
145	>	* Task class with a constructor accepting a phaser that
146		* it registers for upon construction:
147	<	* <pre>
148	<	*  void build(Task[] actions, int lo, int hi, Phaser b) {
149	<	*    int step = (hi - lo) / TASKS_PER_PHASER;
150	<	*    if (step &gt; 1) {
151	<	*       int i = lo;
152	<	*       while (i &lt; hi) {
153	<	*         int r = Math.min(i + step, hi);
154	<	*         build(actions, i, r, new Phaser(b));
155	<	*         i = r;
156	<	*       }
157	<	*    }
158	<	*    else {
159	<	*      for (int i = lo; i &lt; hi; ++i)
160	<	*        actions[i] = new Task(b);
161	<	*        // assumes new Task(b) performs b.register()
162	<	*    }
163	<	*  }
164	<	*  // .. initially called, for n tasks via
160	<	*  build(new Task[n], 0, n, new Phaser());
161	<	* </pre>
147	>	*  <pre> {@code
148	>	* void build(Task[] actions, int lo, int hi, Phaser b) {
149	>	*   int step = (hi - lo) / TASKS_PER_PHASER;
150	>	*   if (step > 1) {
151	>	*     int i = lo;
152	>	*     while (i < hi) {
153	>	*       int r = Math.min(i + step, hi);
154	>	*       build(actions, i, r, new Phaser(b));
155	>	*       i = r;
156	>	*     }
157	>	*   } else {
158	>	*     for (int i = lo; i < hi; ++i)
159	>	*       actions[i] = new Task(b);
160	>	*       // assumes new Task(b) performs b.register()
161	>	*   }
162	>	* }
163	>	* // .. initially called, for n tasks via
164	>	* build(new Task[n], 0, n, new Phaser());}</pre>
165		*
166	<	* The best value of <tt>TASKS_PER_PHASER</tt> depends mainly on
166	>	* The best value of {@code TASKS_PER_PHASER} depends mainly on
167		* expected barrier synchronization rates. A value as low as four may
168		* be appropriate for extremely small per-barrier task bodies (thus
169		* high rates), or up to hundreds for extremely large ones.
#	Line 172 | Line 175 | import java.lang.reflect.*;
175		* parties result in IllegalStateExceptions. However, you can and
176		* should create tiered phasers to accommodate arbitrarily large sets
177		* of participants.
178	+	*
179	+	* @since 1.7
180	+	* @author Doug Lea
181		*/
182		public class Phaser {
183		/*
#	Line 192 | Line 198 | public class Phaser {
198		* However, to efficiently maintain atomicity, these values are
199		* packed into a single (atomic) long. Termination uses the sign
200		* bit of 32 bit representation of phase, so phase is set to -1 on
201	<	* termination. Good performace relies on keeping state decoding
201	>	* termination. Good performance relies on keeping state decoding
202		* and encoding simple, and keeping race windows short.
203		*
204		* Note: there are some cheats in arrive() that rely on unarrived
205	<	* being lowest 16 bits.
205	>	* count being lowest 16 bits.
206		*/
207		private volatile long state;
208
209		private static final int ushortBits = 16;
210	<	private static final int ushortMask =  (1 << ushortBits) - 1;
211	<	private static final int phaseMask = 0x7fffffff;
210	>	private static final int ushortMask = 0xffff;
211	>	private static final int phaseMask  = 0x7fffffff;
212
213		private static int unarrivedOf(long s) {
214	<	return (int)(s & ushortMask);
214	>	return (int) (s & ushortMask);
215		}
216
217		private static int partiesOf(long s) {
218	<	return (int)(s & (ushortMask << 16)) >>> 16;
218	>	return ((int) s) >>> 16;
219		}
220
221		private static int phaseOf(long s) {
222	<	return (int)(s >>> 32);
222	>	return (int) (s >>> 32);
223		}
224
225		private static int arrivedOf(long s) {
#	Line 221 | Line 227 | public class Phaser {
227		}
228
229		private static long stateFor(int phase, int parties, int unarrived) {
230	<	return (((long)phase) << 32) | ((parties << 16) | unarrived);
230	>	return ((((long) phase) << 32) | (((long) parties) << 16) |
231	>	(long) unarrived);
232		}
233
234		private static long trippedStateFor(int phase, int parties) {
235	<	return (((long)phase) << 32) | ((parties << 16) | parties);
235	>	long lp = (long) parties;
236	>	return (((long) phase) << 32) | (lp << 16) | lp;
237		}
238
239	<	private static IllegalStateException badBounds(int parties, int unarrived) {
240	<	return new IllegalStateException
241	<	("Attempt to set " + unarrived +
242	<	" unarrived of " + parties + " parties");
239	>	/**
240	>	* Returns message string for bad bounds exceptions.
241	>	*/
242	>	private static String badBounds(int parties, int unarrived) {
243	>	return ("Attempt to set " + unarrived +
244	>	" unarrived of " + parties + " parties");
245		}
246
247		/**
#	Line 240 | Line 250 | public class Phaser {
250		private final Phaser parent;
251
252		/**
253	<	* The root of Phaser tree. Equals this if not in a tree.  Used to
253	>	* The root of phaser tree. Equals this if not in a tree.  Used to
254		* support faster state push-down.
255		*/
256		private final Phaser root;
#	Line 248 | Line 258 | public class Phaser {
258		// Wait queues
259
260		/**
261	<	* Heads of Treiber stacks waiting for nonFJ threads. To eliminate
261	>	* Heads of Treiber stacks for waiting threads. To eliminate
262		* contention while releasing some threads while adding others, we
263		* use two of them, alternating across even and odd phases.
264		*/
#	Line 256 | Line 266 | public class Phaser {
266		private final AtomicReference<QNode> oddQ  = new AtomicReference<QNode>();
267
268		private AtomicReference<QNode> queueFor(int phase) {
269	<	return (phase & 1) == 0? evenQ : oddQ;
269	>	return ((phase & 1) == 0) ? evenQ : oddQ;
270		}
271
272		/**
#	Line 264 | Line 274 | public class Phaser {
274		* root if necessary.
275		*/
276		private long getReconciledState() {
277	<	return parent == null? state : reconcileState();
277	>	return (parent == null) ? state : reconcileState();
278		}
279
280		/**
#	Line 291 | Line 301 | public class Phaser {
301		}
302
303		/**
304	<	* Creates a new Phaser without any initially registered parties,
305	<	* initial phase number 0, and no parent.
304	>	* Creates a new phaser without any initially registered parties,
305	>	* initial phase number 0, and no parent. Any thread using this
306	>	* phaser will need to first register for it.
307		*/
308		public Phaser() {
309		this(null);
310		}
311
312		/**
313	<	* Creates a new Phaser with the given numbers of registered
313	>	* Creates a new phaser with the given numbers of registered
314		* unarrived parties, initial phase number 0, and no parent.
315	<	* @param parties the number of parties required to trip barrier.
315	>	*
316	>	* @param parties the number of parties required to trip barrier
317		* @throws IllegalArgumentException if parties less than zero
318	<	* or greater than the maximum number of parties supported.
318	>	* or greater than the maximum number of parties supported
319		*/
320		public Phaser(int parties) {
321		this(null, parties);
322		}
323
324		/**
325	<	* Creates a new Phaser with the given parent, without any
325	>	* Creates a new phaser with the given parent, without any
326		* initially registered parties. If parent is non-null this phaser
327		* is registered with the parent and its initial phase number is
328		* the same as that of parent phaser.
329	<	* @param parent the parent phaser.
329	>	*
330	>	* @param parent the parent phaser
331		*/
332		public Phaser(Phaser parent) {
333		int phase = 0;
#	Line 329 | Line 342 | public class Phaser {
342		}
343
344		/**
345	<	* Creates a new Phaser with the given parent and numbers of
346	<	* registered unarrived parties. If parent is non-null this phaser
345	>	* Creates a new phaser with the given parent and numbers of
346	>	* registered unarrived parties. If parent is non-null, this phaser
347		* is registered with the parent and its initial phase number is
348		* the same as that of parent phaser.
349	<	* @param parent the parent phaser.
350	<	* @param parties the number of parties required to trip barrier.
349	>	*
350	>	* @param parent the parent phaser
351	>	* @param parties the number of parties required to trip barrier
352		* @throws IllegalArgumentException if parties less than zero
353	<	* or greater than the maximum number of parties supported.
353	>	* or greater than the maximum number of parties supported
354		*/
355		public Phaser(Phaser parent, int parties) {
356		if (parties < 0 || parties > ushortMask)
#	Line 354 | Line 368 | public class Phaser {
368
369		/**
370		* Adds a new unarrived party to this phaser.
371	+	*
372		* @return the current barrier phase number upon registration
373		* @throws IllegalStateException if attempting to register more
374	<	* than the maximum supported number of parties.
374	>	* than the maximum supported number of parties
375		*/
376		public int register() {
377		return doRegister(1);
#	Line 364 | Line 379 | public class Phaser {
379
380		/**
381		* Adds the given number of new unarrived parties to this phaser.
382	<	* @param parties the number of parties required to trip barrier.
382	>	*
383	>	* @param parties the number of parties required to trip barrier
384		* @return the current barrier phase number upon registration
385		* @throws IllegalStateException if attempting to register more
386	<	* than the maximum supported number of parties.
386	>	* than the maximum supported number of parties
387		*/
388		public int bulkRegister(int parties) {
389		if (parties < 0)
#	Line 390 | Line 406 | public class Phaser {
406		if (phase < 0)
407		break;
408		if (parties > ushortMask || unarrived > ushortMask)
409	<	throw badBounds(parties, unarrived);
409	>	throw new IllegalStateException(badBounds(parties, unarrived));
410		if (phase == phaseOf(root.state) &&
411		casState(s, stateFor(phase, parties, unarrived)))
412		break;
#	Line 403 | Line 419 | public class Phaser {
419		* in turn wait for others via {@link #awaitAdvance}).
420		*
421		* @return the barrier phase number upon entry to this method, or a
422	<	* negative value if terminated;
422	>	* negative value if terminated
423		* @throws IllegalStateException if not terminated and the number
424	<	* of unarrived parties would become negative.
424	>	* of unarrived parties would become negative
425		*/
426		public int arrive() {
427		int phase;
428		for (;;) {
429		long s = state;
430		phase = phaseOf(s);
431	+	if (phase < 0)
432	+	break;
433		int parties = partiesOf(s);
434		int unarrived = unarrivedOf(s) - 1;
435		if (unarrived > 0) {        // Not the last arrival
#	Line 423 | Line 441 | public class Phaser {
441		if (par == null) {      // directly trip
442		if (casState
443		(s,
444	<	trippedStateFor(onAdvance(phase, parties)? -1 :
444	>	trippedStateFor(onAdvance(phase, parties) ? -1 :
445		((phase + 1) & phaseMask), parties))) {
446		releaseWaiters(phase);
447		break;
#	Line 437 | Line 455 | public class Phaser {
455		}
456		}
457		}
440	–	else if (phase < 0) // Don't throw exception if terminated
441	–	break;
458		else if (phase != phaseOf(root.state)) // or if unreconciled
459		reconcileState();
460		else
461	<	throw badBounds(parties, unarrived);
461	>	throw new IllegalStateException(badBounds(parties, unarrived));
462		}
463		return phase;
464		}
#	Line 455 | Line 471 | public class Phaser {
471		* zero parties, this phaser is also deregistered from its parent.
472		*
473		* @return the current barrier phase number upon entry to
474	<	* this method, or a negative value if terminated;
474	>	* this method, or a negative value if terminated
475		* @throws IllegalStateException if not terminated and the number
476	<	* of registered or unarrived parties would become negative.
476	>	* of registered or unarrived parties would become negative
477		*/
478		public int arriveAndDeregister() {
479		// similar code to arrive, but too different to merge
#	Line 466 | Line 482 | public class Phaser {
482		for (;;) {
483		long s = state;
484		phase = phaseOf(s);
485	+	if (phase < 0)
486	+	break;
487		int parties = partiesOf(s) - 1;
488		int unarrived = unarrivedOf(s) - 1;
489		if (parties >= 0) {
#	Line 484 | Line 502 | public class Phaser {
502		if (unarrived == 0) {
503		if (casState
504		(s,
505	<	trippedStateFor(onAdvance(phase, parties)? -1 :
505	>	trippedStateFor(onAdvance(phase, parties) ? -1 :
506		((phase + 1) & phaseMask), parties))) {
507		releaseWaiters(phase);
508		break;
509		}
510		continue;
511		}
494	–	if (phase < 0)
495	–	break;
512		if (par != null && phase != phaseOf(root.state)) {
513		reconcileState();
514		continue;
515		}
516		}
517	<	throw badBounds(parties, unarrived);
517	>	throw new IllegalStateException(badBounds(parties, unarrived));
518		}
519		return phase;
520		}
521
522		/**
523		* Arrives at the barrier and awaits others. Equivalent in effect
524	<	* to <tt>awaitAdvance(arrive())</tt>.  If you instead need to
524	>	* to {@code awaitAdvance(arrive())}.  If you instead need to
525		* await with interruption of timeout, and/or deregister upon
526		* arrival, you can arrange them using analogous constructions.
527	+	*
528		* @return the phase on entry to this method
529		* @throws IllegalStateException if not terminated and the number
530	<	* of unarrived parties would become negative.
530	>	* of unarrived parties would become negative
531		*/
532		public int arriveAndAwaitAdvance() {
533		return awaitAdvance(arrive());
#	Line 520 | Line 537 | public class Phaser {
537		* Awaits the phase of the barrier to advance from the given
538		* value, or returns immediately if argument is negative or this
539		* barrier is terminated.
540	+	*
541		* @param phase the phase on entry to this method
542		* @return the phase on exit from this method
543		*/
#	Line 530 | Line 548 | public class Phaser {
548		int p = phaseOf(s);
549		if (p != phase)
550		return p;
551	<	if (unarrivedOf(s) == 0)
551	>	if (unarrivedOf(s) == 0 && parent != null)
552		parent.awaitAdvance(phase);
553		// Fall here even if parent waited, to reconcile and help release
554		return untimedWait(phase);
#	Line 538 | Line 556 | public class Phaser {
556
557		/**
558		* Awaits the phase of the barrier to advance from the given
559	<	* value, or returns immediately if argumet is negative or this
559	>	* value, or returns immediately if argument is negative or this
560		* barrier is terminated, or throws InterruptedException if
561		* interrupted while waiting.
562	+	*
563		* @param phase the phase on entry to this method
564		* @return the phase on exit from this method
565		* @throws InterruptedException if thread interrupted while waiting
566		*/
567	<	public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
567	>	public int awaitAdvanceInterruptibly(int phase)
568	>	throws InterruptedException {
569		if (phase < 0)
570		return phase;
571		long s = getReconciledState();
572		int p = phaseOf(s);
573		if (p != phase)
574		return p;
575	<	if (unarrivedOf(s) != 0)
575	>	if (unarrivedOf(s) == 0 && parent != null)
576		parent.awaitAdvanceInterruptibly(phase);
577		return interruptibleWait(phase);
578		}
#	Line 561 | Line 581 | public class Phaser {
581		* Awaits the phase of the barrier to advance from the given value
582		* or the given timeout elapses, or returns immediately if
583		* argument is negative or this barrier is terminated.
584	+	*
585		* @param phase the phase on entry to this method
586		* @return the phase on exit from this method
587		* @throws InterruptedException if thread interrupted while waiting
588		* @throws TimeoutException if timed out while waiting
589		*/
590	<	public int awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)
590	>	public int awaitAdvanceInterruptibly(int phase,
591	>	long timeout, TimeUnit unit)
592		throws InterruptedException, TimeoutException {
593		if (phase < 0)
594		return phase;
#	Line 574 | Line 596 | public class Phaser {
596		int p = phaseOf(s);
597		if (p != phase)
598		return p;
599	<	if (unarrivedOf(s) == 0)
599	>	if (unarrivedOf(s) == 0 && parent != null)
600		parent.awaitAdvanceInterruptibly(phase, timeout, unit);
601		return timedWait(phase, unit.toNanos(timeout));
602		}
#	Line 605 | Line 627 | public class Phaser {
627
628		/**
629		* Returns the current phase number. The maximum phase number is
630	<	* <tt>Integer.MAX_VALUE</tt>, after which it restarts at
630	>	* {@code Integer.MAX_VALUE}, after which it restarts at
631		* zero. Upon termination, the phase number is negative.
632	+	*
633		* @return the phase number, or a negative value if terminated
634		*/
635		public final int getPhase() {
#	Line 614 | Line 637 | public class Phaser {
637		}
638
639		/**
640	<	* Returns true if the current phase number equals the given phase.
640	>	* Returns {@code true} if the current phase number equals the given phase.
641	>	*
642		* @param phase the phase
643	<	* @return true if the current phase number equals the given phase.
643	>	* @return {@code true} if the current phase number equals the given phase
644		*/
645		public final boolean hasPhase(int phase) {
646		return phaseOf(getReconciledState()) == phase;
#	Line 624 | Line 648 | public class Phaser {
648
649		/**
650		* Returns the number of parties registered at this barrier.
651	+	*
652		* @return the number of parties
653		*/
654		public int getRegisteredParties() {
#	Line 633 | Line 658 | public class Phaser {
658		/**
659		* Returns the number of parties that have arrived at the current
660		* phase of this barrier.
661	+	*
662		* @return the number of arrived parties
663		*/
664		public int getArrivedParties() {
#	Line 642 | Line 668 | public class Phaser {
668		/**
669		* Returns the number of registered parties that have not yet
670		* arrived at the current phase of this barrier.
671	+	*
672		* @return the number of unarrived parties
673		*/
674		public int getUnarrivedParties() {
#	Line 649 | Line 676 | public class Phaser {
676		}
677
678		/**
679	<	* Returns the parent of this phaser, or null if none.
680	<	* @return the parent of this phaser, or null if none.
679	>	* Returns the parent of this phaser, or {@code null} if none.
680	>	*
681	>	* @return the parent of this phaser, or {@code null} if none
682		*/
683		public Phaser getParent() {
684		return parent;
#	Line 659 | Line 687 | public class Phaser {
687		/**
688		* Returns the root ancestor of this phaser, which is the same as
689		* this phaser if it has no parent.
690	<	* @return the root ancestor of this phaser.
690	>	*
691	>	* @return the root ancestor of this phaser
692		*/
693		public Phaser getRoot() {
694		return root;
695		}
696
697		/**
698	<	* Returns true if this barrier has been terminated.
699	<	* @return true if this barrier has been terminated
698	>	* Returns {@code true} if this barrier has been terminated.
699	>	*
700	>	* @return {@code true} if this barrier has been terminated
701		*/
702		public boolean isTerminated() {
703		return getPhase() < 0;
#	Line 677 | Line 707 | public class Phaser {
707		* Overridable method to perform an action upon phase advance, and
708		* to control termination. This method is invoked whenever the
709		* barrier is tripped (and thus all other waiting parties are
710	<	* dormant). If it returns true, then, rather than advance the
711	<	* phase number, this barrier will be set to a final termination
712	<	* state, and subsequent calls to <tt>isTerminated</tt> will
713	<	* return true.
710	>	* dormant). If it returns {@code true}, then, rather than advance
711	>	* the phase number, this barrier will be set to a final
712	>	* termination state, and subsequent calls to {@link #isTerminated}
713	>	* will return true.
714		*
715	<	* <p> The default version returns true when the number of
715	>	* <p> The default version returns {@code true} when the number of
716		* registered parties is zero. Normally, overrides that arrange
717		* termination for other reasons should also preserve this
718		* property.
#	Line 690 | Line 720 | public class Phaser {
720		* <p> You may override this method to perform an action with side
721		* effects visible to participating tasks, but it is in general
722		* only sensible to do so in designs where all parties register
723	<	* before any arrive, and all <tt>awaitAdvance</tt> at each phase.
723	>	* before any arrive, and all {@link #awaitAdvance} at each phase.
724		* Otherwise, you cannot ensure lack of interference. In
725		* particular, this method may be invoked more than once per
726		* transition if other parties successfully register while the
#	Line 699 | Line 729 | public class Phaser {
729		* method.
730		*
731		* @param phase the phase number on entering the barrier
732	<	* @param registeredParties the current number of registered
733	<	* parties.
704	<	* @return true if this barrier should terminate
732	>	* @param registeredParties the current number of registered parties
733	>	* @return {@code true} if this barrier should terminate
734		*/
735		protected boolean onAdvance(int phase, int registeredParties) {
736		return registeredParties <= 0;
#	Line 710 | Line 739 | public class Phaser {
739		/**
740		* Returns a string identifying this phaser, as well as its
741		* state.  The state, in brackets, includes the String {@code
742	<	* "phase ="} followed by the phase number, {@code "parties ="}
742	>	* "phase = "} followed by the phase number, {@code "parties = "}
743		* followed by the number of registered parties, and {@code
744	<	* "arrived ="} followed by the number of arrived parties
744	>	* "arrived = "} followed by the number of arrived parties.
745		*
746		* @return a string identifying this barrier, as well as its state
747		*/
748		public String toString() {
749		long s = getReconciledState();
750	<	return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
750	>	return super.toString() +
751	>	"[phase = " + phaseOf(s) +
752	>	" parties = " + partiesOf(s) +
753	>	" arrived = " + arrivedOf(s) + "]";
754		}
755
756		// methods for waiting
757
726	–	/** The number of CPUs, for spin control */
727	–	static final int NCPUS = Runtime.getRuntime().availableProcessors();
728	–
729	–	/**
730	–	* The number of times to spin before blocking in timed waits.
731	–	* The value is empirically derived.
732	–	*/
733	–	static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
734	–
735	–	/**
736	–	* The number of times to spin before blocking in untimed waits.
737	–	* This is greater than timed value because untimed waits spin
738	–	* faster since they don't need to check times on each spin.
739	–	*/
740	–	static final int maxUntimedSpins = maxTimedSpins * 32;
741	–
742	–	/**
743	–	* The number of nanoseconds for which it is faster to spin
744	–	* rather than to use timed park. A rough estimate suffices.
745	–	*/
746	–	static final long spinForTimeoutThreshold = 1000L;
747	–
758		/**
759	<	* Wait nodes for Treiber stack representing wait queue for non-FJ
750	<	* tasks.
759	>	* Wait nodes for Treiber stack representing wait queue
760		*/
761	<	static final class QNode {
762	<	QNode next;
761	>	static final class QNode implements ForkJoinPool.ManagedBlocker {
762	>	final Phaser phaser;
763	>	final int phase;
764	>	final long startTime;
765	>	final long nanos;
766	>	final boolean timed;
767	>	final boolean interruptible;
768	>	volatile boolean wasInterrupted = false;
769		volatile Thread thread; // nulled to cancel wait
770	<	QNode() {
770	>	QNode next;
771	>	QNode(Phaser phaser, int phase, boolean interruptible,
772	>	boolean timed, long startTime, long nanos) {
773	>	this.phaser = phaser;
774	>	this.phase = phase;
775	>	this.timed = timed;
776	>	this.interruptible = interruptible;
777	>	this.startTime = startTime;
778	>	this.nanos = nanos;
779		thread = Thread.currentThread();
780		}
781	+	public boolean isReleasable() {
782	+	return (thread == null ||
783	+	phaser.getPhase() != phase ||
784	+	(interruptible && wasInterrupted) ||
785	+	(timed && (nanos - (System.nanoTime() - startTime)) <= 0));
786	+	}
787	+	public boolean block() {
788	+	if (Thread.interrupted()) {
789	+	wasInterrupted = true;
790	+	if (interruptible)
791	+	return true;
792	+	}
793	+	if (!timed)
794	+	LockSupport.park(this);
795	+	else {
796	+	long waitTime = nanos - (System.nanoTime() - startTime);
797	+	if (waitTime <= 0)
798	+	return true;
799	+	LockSupport.parkNanos(this, waitTime);
800	+	}
801	+	return isReleasable();
802	+	}
803		void signal() {
804		Thread t = thread;
805		if (t != null) {
#	Line 762 | Line 807 | public class Phaser {
807		LockSupport.unpark(t);
808		}
809		}
810	+	boolean doWait() {
811	+	if (thread != null) {
812	+	try {
813	+	ForkJoinPool.managedBlock(this, false);
814	+	} catch (InterruptedException ie) {
815	+	}
816	+	}
817	+	return wasInterrupted;
818	+	}
819	+
820		}
821
822		/**
823	<	* Removes and signals waiting threads from wait queue
823	>	* Removes and signals waiting threads from wait queue.
824		*/
825		private void releaseWaiters(int phase) {
826		AtomicReference<QNode> head = queueFor(phase);
#	Line 777 | Line 832 | public class Phaser {
832		}
833
834		/**
835	+	* Tries to enqueue given node in the appropriate wait queue.
836	+	*
837	+	* @return true if successful
838	+	*/
839	+	private boolean tryEnqueue(QNode node) {
840	+	AtomicReference<QNode> head = queueFor(node.phase);
841	+	return head.compareAndSet(node.next = head.get(), node);
842	+	}
843	+
844	+	/**
845		* Enqueues node and waits unless aborted or signalled.
846	+	*
847	+	* @return current phase
848		*/
849		private int untimedWait(int phase) {
783	–	int spins = maxUntimedSpins;
850		QNode node = null;
785	–	boolean interrupted = false;
851		boolean queued = false;
852	+	boolean interrupted = false;
853		int p;
854		while ((p = getPhase()) == phase) {
855	<	interrupted = Thread.interrupted();
856	<	if (node != null) {
857	<	if (!queued) {
858	<	AtomicReference<QNode> head = queueFor(phase);
859	<	queued = head.compareAndSet(node.next = head.get(), node);
860	<	}
795	<	else if (node.thread != null)
796	<	LockSupport.park(this);
797	<	}
798	<	else if (spins <= 0)
799	<	node = new QNode();
855	>	if (Thread.interrupted())
856	>	interrupted = true;
857	>	else if (node == null)
858	>	node = new QNode(this, phase, false, false, 0, 0);
859	>	else if (!queued)
860	>	queued = tryEnqueue(node);
861		else
862	<	--spins;
862	>	interrupted = node.doWait();
863		}
864		if (node != null)
865		node.thread = null;
866	+	releaseWaiters(phase);
867		if (interrupted)
868		Thread.currentThread().interrupt();
807	–	releaseWaiters(phase);
869		return p;
870		}
871
872		/**
873	<	* Messier interruptible version
873	>	* Interruptible version
874	>	* @return current phase
875		*/
876		private int interruptibleWait(int phase) throws InterruptedException {
815	–	int spins = maxUntimedSpins;
877		QNode node = null;
878		boolean queued = false;
879		boolean interrupted = false;
880		int p;
881	<	while ((p = getPhase()) == phase) {
882	<	if (interrupted = Thread.interrupted())
883	<	break;
884	<	if (node != null) {
885	<	if (!queued) {
886	<	AtomicReference<QNode> head = queueFor(phase);
887	<	queued = head.compareAndSet(node.next = head.get(), node);
827	<	}
828	<	else if (node.thread != null)
829	<	LockSupport.park(this);
830	<	}
831	<	else if (spins <= 0)
832	<	node = new QNode();
881	>	while ((p = getPhase()) == phase && !interrupted) {
882	>	if (Thread.interrupted())
883	>	interrupted = true;
884	>	else if (node == null)
885	>	node = new QNode(this, phase, true, false, 0, 0);
886	>	else if (!queued)
887	>	queued = tryEnqueue(node);
888		else
889	<	--spins;
889	>	interrupted = node.doWait();
890		}
891		if (node != null)
892		node.thread = null;
893	+	if (p != phase || (p = getPhase()) != phase)
894	+	releaseWaiters(phase);
895		if (interrupted)
896		throw new InterruptedException();
840	–	releaseWaiters(phase);
897		return p;
898		}
899
900		/**
901	<	* Even messier timeout version.
901	>	* Timeout version.
902	>	* @return current phase
903		*/
904		private int timedWait(int phase, long nanos)
905		throws InterruptedException, TimeoutException {
906	+	long startTime = System.nanoTime();
907	+	QNode node = null;
908	+	boolean queued = false;
909	+	boolean interrupted = false;
910		int p;
911	<	if ((p = getPhase()) == phase) {
912	<	long lastTime = System.nanoTime();
913	<	int spins = maxTimedSpins;
914	<	QNode node = null;
915	<	boolean queued = false;
916	<	boolean interrupted = false;
917	<	while ((p = getPhase()) == phase) {
918	<	if (interrupted = Thread.interrupted())
919	<	break;
920	<	long now = System.nanoTime();
921	<	if ((nanos -= now - lastTime) <= 0)
861	<	break;
862	<	lastTime = now;
863	<	if (node != null) {
864	<	if (!queued) {
865	<	AtomicReference<QNode> head = queueFor(phase);
866	<	queued = head.compareAndSet(node.next = head.get(), node);
867	<	}
868	<	else if (node.thread != null &&
869	<	nanos > spinForTimeoutThreshold) {
870	<	LockSupport.parkNanos(this, nanos);
871	<	}
872	<	}
873	<	else if (spins <= 0)
874	<	node = new QNode();
875	<	else
876	<	--spins;
877	<	}
878	<	if (node != null)
879	<	node.thread = null;
880	<	if (interrupted)
881	<	throw new InterruptedException();
882	<	if (p == phase && (p = getPhase()) == phase)
883	<	throw new TimeoutException();
911	>	while ((p = getPhase()) == phase && !interrupted) {
912	>	if (Thread.interrupted())
913	>	interrupted = true;
914	>	else if (nanos - (System.nanoTime() - startTime) <= 0)
915	>	break;
916	>	else if (node == null)
917	>	node = new QNode(this, phase, true, true, startTime, nanos);
918	>	else if (!queued)
919	>	queued = tryEnqueue(node);
920	>	else
921	>	interrupted = node.doWait();
922		}
923	<	releaseWaiters(phase);
923	>	if (node != null)
924	>	node.thread = null;
925	>	if (p != phase || (p = getPhase()) != phase)
926	>	releaseWaiters(phase);
927	>	if (interrupted)
928	>	throw new InterruptedException();
929	>	if (p == phase)
930	>	throw new TimeoutException();
931		return p;
932		}
933
934	<	// Temporary Unsafe mechanics for preliminary release
934	>	// Unsafe mechanics
935	>
936	>	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
937	>	private static final long stateOffset =
938	>	objectFieldOffset("state", Phaser.class);
939
940	<	static final Unsafe _unsafe;
941	<	static final long stateOffset;
940	>	private final boolean casState(long cmp, long val) {
941	>	return UNSAFE.compareAndSwapLong(this, stateOffset, cmp, val);
942	>	}
943
944	<	static {
944	>	private static long objectFieldOffset(String field, Class<?> klazz) {
945		try {
946	<	if (Phaser.class.getClassLoader() != null) {
947	<	Field f = Unsafe.class.getDeclaredField("theUnsafe");
948	<	f.setAccessible(true);
949	<	_unsafe = (Unsafe)f.get(null);
950	<	}
951	<	else
902	<	_unsafe = Unsafe.getUnsafe();
903	<	stateOffset = _unsafe.objectFieldOffset
904	<	(Phaser.class.getDeclaredField("state"));
905	<	} catch (Exception e) {
906	<	throw new RuntimeException("Could not initialize intrinsics", e);
946	>	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
947	>	} catch (NoSuchFieldException e) {
948	>	// Convert Exception to corresponding Error
949	>	NoSuchFieldError error = new NoSuchFieldError(field);
950	>	error.initCause(e);
951	>	throw error;
952		}
953		}
954
955	<	final boolean casState(long cmp, long val) {
956	<	return _unsafe.compareAndSwapLong(this, stateOffset, cmp, val);
955	>	/**
956	>	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
957	>	* Replace with a simple call to Unsafe.getUnsafe when integrating
958	>	* into a jdk.
959	>	*
960	>	* @return a sun.misc.Unsafe
961	>	*/
962	>	private static sun.misc.Unsafe getUnsafe() {
963	>	try {
964	>	return sun.misc.Unsafe.getUnsafe();
965	>	} catch (SecurityException se) {
966	>	try {
967	>	return java.security.AccessController.doPrivileged
968	>	(new java.security
969	>	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
970	>	public sun.misc.Unsafe run() throws Exception {
971	>	java.lang.reflect.Field f = sun.misc
972	>	.Unsafe.class.getDeclaredField("theUnsafe");
973	>	f.setAccessible(true);
974	>	return (sun.misc.Unsafe) f.get(null);
975	>	}});
976	>	} catch (java.security.PrivilegedActionException e) {
977	>	throw new RuntimeException("Could not initialize intrinsics",
978	>	e.getCause());
979	>	}
980	>	}
981		}
982		}

Diff Legend

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