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 1.3 by jsr166, Fri Jul 25 18:11:53 2008 UTC vs.
Revision 1.6 by dl, Tue Oct 28 23:03:24 2008 UTC

#	Line 5 | Line 5
5		*/
6
7		package jsr166y;
8	–	import jsr166y.forkjoin.*;
8		import java.util.concurrent.*;
9		import java.util.concurrent.atomic.*;
10		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}, but supporting more
17	<	* flexible usage.
16	>	* {@link java.util.concurrent.CyclicBarrier} and {@link
17	>	* java.util.concurrent.CountDownLatch} but supporting more flexible
18	>	* usage.
19		*
20		* <ul>
21		*
22	<	* <li> The number of parties synchronizing on the barrier may vary
23	<	* over time.  A task may register to be a party in a barrier at any
24	<	* time, and may deregister upon arriving at the barrier.  As is the
25	<	* case with most basic synchronization constructs, registration
26	<	* and deregistration affect only internal counts; they do not
27	<	* establish any further internal bookkeeping, so tasks cannot query
28	<	* whether they are registered.
22	>	* <li> The number of parties synchronizing on a phaser may vary over
23	>	* time.  A task may register to be a party at any time, and may
24	>	* deregister upon arriving at the barrier.  As is the case with most
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
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
#	Line 32 | Line 35 | import java.util.concurrent.locks.LockSu
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 be invoked independently:
38	>	* aspects of coordination, that may also be invoked independently:
39		*
40		* <ul>
41		*
42		*   <li> Arriving at a barrier. Methods <tt>arrive</tt> and
43		*       <tt>arriveAndDeregister</tt> do not block, but return
44	<	*       the phase value on entry to the method.
44	>	*       the phase value current upon entry to the method.
45		*
46		*   <li> Awaiting others. Method <tt>awaitAdvance</tt> requires an
47		*       argument indicating the entry phase, and returns when the
#	Line 49 | Line 52 | import java.util.concurrent.locks.LockSu
52		* <li> Barrier actions, performed by the task triggering a phase
53		* advance while others may be waiting, are arranged by overriding
54		* method <tt>onAdvance</tt>, that also controls termination.
55	+	* Overriding this method may be used to similar but more flexible
56	+	* effect as providing a barrier action to a CyclicBarrier.
57		*
58		* <li> Phasers may enter a <em>termination</em> state in which all
59		* await actions immediately return, indicating (via a negative phase
60		* value) that execution is complete.  Termination is triggered by
61		* executing the overridable <tt>onAdvance</tt> method that is invoked
62	<	* each time the barrier is tripped. When a Phaser is controlling an
63	<	* action with a fixed number of iterations, it is often convenient to
64	<	* override this method to cause termination when the current phase
65	<	* number reaches a threshold.  Method <tt>forceTermination</tt> is
66	<	* also available to assist recovery actions upon failure.
67	<	*
68	<	* <li> Unlike most synchronizers, a Phaser may also be used with
69	<	* ForkJoinTasks (as well as plain threads).
62	>	* each time the barrier is about to be tripped. When a Phaser is
63	>	* controlling an action with a fixed number of iterations, it is
64	>	* often convenient to override this method to cause termination when
65	>	* the current phase number reaches a threshold. Method
66	>	* <tt>forceTermination</tt> is also available to abruptly release
67	>	* waiting threads and allow them to terminate.
68	>	*
69	>	* <li> Phasers may be tiered to reduce contention. Phasers with large
70	>	* numbers of parties that would otherwise experience heavy
71	>	* synchronization contention costs may instead be arranged in trees.
72	>	* This will typically greatly increase throughput even though it
73	>	* incurs somewhat greater per-operation overhead.
74		*
75		* <li> By default, <tt>awaitAdvance</tt> continues to wait even if
76	<	* the current thread is interrupted. And unlike the case in
76	>	* the waiting thread is interrupted. And unlike the case in
77		* CyclicBarriers, exceptions encountered while tasks wait
78		* interruptibly or with timeout do not change the state of the
79		* barrier. If necessary, you can perform any associated recovery
80	<	* within handlers of those exceptions.
80	>	* within handlers of those exceptions, often after invoking
81	>	* <tt>forceTermination</tt>.
82		*
83		* </ul>
84		*
85	<	* <p><b>Sample usage:</b>
76	<	*
77	<	* <p>[todo: non-FJ example]
85	>	* <p><b>Sample usages:</b>
86		*
87	<	* <p> A Phaser may be used to support a style of programming in
88	<	* which a task waits for others to complete, without otherwise
89	<	* needing to keep track of which tasks it is waiting for. This is
90	<	* similar to the "sync" construct in Cilk and "clocks" in X10.
83	<	* Special constructions based on such barriers are available using
84	<	* the <tt>LinkedAsyncAction</tt> and <tt>CyclicAction</tt> classes,
85	<	* but they can be useful in other contexts as well.  For a simple
86	<	* (but not very useful) example, here is a variant of Fibonacci:
87	>	* <p>A Phaser may be used instead of a <tt>CountdownLatch</tt> to control
88	>	* a one-shot action serving a variable number of parties. The typical
89	>	* idiom is for the method setting this up to first register, then
90	>	* start the actions, then deregister, as in:
91		*
92		* <pre>
93	<	* class BarrierFibonacci extends RecursiveAction {
94	<	*   int argument, result;
95	<	*   final Phaser parentBarrier;
96	<	*   BarrierFibonacci(int n, Phaser parentBarrier) {
97	<	*     this.argument = n;
98	<	*     this.parentBarrier = parentBarrier;
99	<	*     parentBarrier.register();
93	>	*  void runTasks(List&lt;Runnable&gt; list) {
94	>	*    final Phaser phaser = new Phaser(1); // "1" to register self
95	>	*    for (Runnable r : list) {
96	>	*      phaser.register();
97	>	*      new Thread() {
98	>	*        public void run() {
99	>	*          phaser.arriveAndAwaitAdvance(); // await all creation
100	>	*          r.run();
101	>	*          phaser.arriveAndDeregister();   // signal completion
102	>	*        }
103	>	*      }.start();
104		*   }
105	<	*   protected void compute() {
106	<	*     int n = argument;
107	<	*     if (n &lt;= 1)
108	<	*        result = n;
109	<	*     else {
110	<	*        Phaser childBarrier = new Phaser(1);
111	<	*        BarrierFibonacci f1 = new BarrierFibonacci(n - 1, childBarrier);
112	<	*        BarrierFibonacci f2 = new BarrierFibonacci(n - 2, childBarrier);
113	<	*        f1.fork();
114	<	*        f2.fork();
115	<	*        childBarrier.arriveAndAwait();
116	<	*        result = f1.result + f2.result;
117	<	*     }
118	<	*     parentBarrier.arriveAndDeregister();
105	>	*
106	>	*   doSomethingOnBehalfOfWorkers();
107	>	*   phaser.arrive(); // allow threads to start
108	>	*   int p = phaser.arriveAndDeregister(); // deregister self  ...
109	>	*   p = phaser.awaitAdvance(p); // ... and await arrival
110	>	*   otherActions(); // do other things while tasks execute
111	>	*   phaser.awaitAdvance(p); // awit final completion
112	>	* }
113	>	* </pre>
114	>	*
115	>	* <p>One way to cause a set of threads to repeatedly perform actions
116	>	* for a given number of iterations is to override <tt>onAdvance</tt>:
117	>	*
118	>	* <pre>
119	>	*  void startTasks(List&lt;Runnable&gt; list, final int iterations) {
120	>	*    final Phaser phaser = new Phaser() {
121	>	*       public boolean onAdvance(int phase, int registeredParties) {
122	>	*         return phase &gt;= iterations || registeredParties == 0;
123	>	*       }
124	>	*    };
125	>	*    phaser.register();
126	>	*    for (Runnable r : list) {
127	>	*      phaser.register();
128	>	*      new Thread() {
129	>	*        public void run() {
130	>	*           do {
131	>	*             r.run();
132	>	*             phaser.arriveAndAwaitAdvance();
133	>	*           } while(!phaser.isTerminated();
134	>	*        }
135	>	*      }.start();
136		*   }
137	+	*   phaser.arriveAndDeregister(); // deregister self, don't wait
138		* }
139		* </pre>
140		*
141	+	* <p> To create a set of tasks using a tree of Phasers,
142	+	* you could use code of the following form, assuming a
143	+	* Task class with a constructor accepting a Phaser that
144	+	* it registers for upon construction:
145	+	* <pre>
146	+	*  void build(Task[] actions, int lo, int hi, Phaser b) {
147	+	*    int step = (hi - lo) / TASKS_PER_PHASER;
148	+	*    if (step &gt; 1) {
149	+	*       int i = lo;
150	+	*       while (i &lt; hi) {
151	+	*         int r = Math.min(i + step, hi);
152	+	*         build(actions, i, r, new Phaser(b));
153	+	*         i = r;
154	+	*       }
155	+	*    }
156	+	*    else {
157	+	*      for (int i = lo; i &lt; hi; ++i)
158	+	*        actions[i] = new Task(b);
159	+	*        // assumes new Task(b) performs b.register()
160	+	*    }
161	+	*  }
162	+	*  // .. initially called, for n tasks via
163	+	*  build(new Task[n], 0, n, new Phaser());
164	+	* </pre>
165	+	*
166	+	* The best value of <tt>TASKS_PER_PHASER</tt> 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.
170	+	*
171	+	* </pre>
172	+	*
173		* <p><b>Implementation notes</b>: This implementation restricts the
174	<	* maximum number of parties to 65535. Attempts to register
175	<	* additional parties result in IllegalStateExceptions.
174	>	* maximum number of parties to 65535. Attempts to register additional
175	>	* parties result in IllegalStateExceptions. However, you can and
176	>	* should create tiered phasers to accommodate arbitrarily large sets
177	>	* of participants.
178		*/
179		public class Phaser {
180		/*
181		* This class implements an extension of X10 "clocks".  Thanks to
182	<	* Vijay Saraswat for the idea of applying it to ForkJoinTasks,
183	<	* and to Vivek Sarkar for enhancements to extend functionality.
182	>	* Vijay Saraswat for the idea, and to Vivek Sarkar for
183	>	* enhancements to extend functionality.
184		*/
185
186		/**
#	Line 133 | Line 193 | public class Phaser {
193		* * terminated -- set if barrier is terminated (1 bit)
194		*
195		* However, to efficiently maintain atomicity, these values are
196	<	* packed into a single AtomicLong. Termination uses the sign bit
197	<	* of 32 bit representation of phase, so phase is set to -1 on
198	<	* termination.
199	<	*/
200	<	private final AtomicLong state;
201	<
202	<	/**
143	<	* Head of Treiber stack for waiting nonFJ threads.
196	>	* packed into a single (atomic) long. Termination uses the sign
197	>	* bit of 32 bit representation of phase, so phase is set to -1 on
198	>	* termination. Good performace relies on keeping state decoding
199	>	* and encoding simple, and keeping race windows short.
200	>	*
201	>	* Note: there are some cheats in arrive() that rely on unarrived
202	>	* being lowest 16 bits.
203		*/
204	<	private final AtomicReference<QNode> head = new AtomicReference<QNode>();
204	>	private volatile long state;
205
206		private static final int ushortBits = 16;
207		private static final int ushortMask =  (1 << ushortBits) - 1;
#	Line 168 | Line 227 | public class Phaser {
227		return (((long)phase) << 32) | ((parties << 16) | unarrived);
228		}
229
230	+	private static long trippedStateFor(int phase, int parties) {
231	+	return (((long)phase) << 32) | ((parties << 16) | parties);
232	+	}
233	+
234		private static IllegalStateException badBounds(int parties, int unarrived) {
235	<	return new IllegalStateException("Attempt to set " + unarrived +
236	<	" unarrived of " + parties + " parties");
235	>	return new IllegalStateException
236	>	("Attempt to set " + unarrived +
237	>	" unarrived of " + parties + " parties");
238	>	}
239	>
240	>	/**
241	>	* The parent of this phaser, or null if none
242	>	*/
243	>	private final Phaser parent;
244	>
245	>	/**
246	>	* The root of Phaser tree. Equals this if not in a tree.  Used to
247	>	* support faster state push-down.
248	>	*/
249	>	private final Phaser root;
250	>
251	>	// Wait queues
252	>
253	>	/**
254	>	* Heads of Treiber stacks waiting for nonFJ threads. To eliminate
255	>	* contention while releasing some threads while adding others, we
256	>	* use two of them, alternating across even and odd phases.
257	>	*/
258	>	private final AtomicReference<QNode> evenQ = new AtomicReference<QNode>();
259	>	private final AtomicReference<QNode> oddQ  = new AtomicReference<QNode>();
260	>
261	>	private AtomicReference<QNode> queueFor(int phase) {
262	>	return (phase & 1) == 0? evenQ : oddQ;
263	>	}
264	>
265	>	/**
266	>	* Returns current state, first resolving lagged propagation from
267	>	* root if necessary.
268	>	*/
269	>	private long getReconciledState() {
270	>	return parent == null? state : reconcileState();
271	>	}
272	>
273	>	/**
274	>	* Recursively resolves state.
275	>	*/
276	>	private long reconcileState() {
277	>	Phaser p = parent;
278	>	long s = state;
279	>	if (p != null) {
280	>	while (unarrivedOf(s) == 0 && phaseOf(s) != phaseOf(root.state)) {
281	>	long parentState = p.getReconciledState();
282	>	int parentPhase = phaseOf(parentState);
283	>	int phase = phaseOf(s = state);
284	>	if (phase != parentPhase) {
285	>	long next = trippedStateFor(parentPhase, partiesOf(s));
286	>	if (casState(s, next)) {
287	>	releaseWaiters(phase);
288	>	s = next;
289	>	}
290	>	}
291	>	}
292	>	}
293	>	return s;
294		}
295
296		/**
297		* Creates a new Phaser without any initially registered parties,
298	<	* and initial phase number 0.
298	>	* initial phase number 0, and no parent.
299		*/
300		public Phaser() {
301	<	state = new AtomicLong(stateFor(0, 0, 0));
301	>	this(null);
302		}
303
304		/**
305		* Creates a new Phaser with the given numbers of registered
306	<	* unarrived parties and initial phase number 0.
306	>	* unarrived parties, initial phase number 0, and no parent.
307		* @param parties the number of parties required to trip barrier.
308		* @throws IllegalArgumentException if parties less than zero
309		* or greater than the maximum number of parties supported.
310		*/
311		public Phaser(int parties) {
312	+	this(null, parties);
313	+	}
314	+
315	+	/**
316	+	* Creates a new Phaser with the given parent, without any
317	+	* initially registered parties. If parent is non-null this phaser
318	+	* is registered with the parent and its initial phase number is
319	+	* the same as that of parent phaser.
320	+	* @param parent the parent phaser.
321	+	*/
322	+	public Phaser(Phaser parent) {
323	+	int phase = 0;
324	+	this.parent = parent;
325	+	if (parent != null) {
326	+	this.root = parent.root;
327	+	phase = parent.register();
328	+	}
329	+	else
330	+	this.root = this;
331	+	this.state = trippedStateFor(phase, 0);
332	+	}
333	+
334	+	/**
335	+	* Creates a new Phaser with the given parent and numbers of
336	+	* registered unarrived parties. If parent is non-null this phaser
337	+	* is registered with the parent and its initial phase number is
338	+	* the same as that of parent phaser.
339	+	* @param parent the parent phaser.
340	+	* @param parties the number of parties required to trip barrier.
341	+	* @throws IllegalArgumentException if parties less than zero
342	+	* or greater than the maximum number of parties supported.
343	+	*/
344	+	public Phaser(Phaser parent, int parties) {
345		if (parties < 0 || parties > ushortMask)
346		throw new IllegalArgumentException("Illegal number of parties");
347	<	state = new AtomicLong(stateFor(0, parties, parties));
347	>	int phase = 0;
348	>	this.parent = parent;
349	>	if (parent != null) {
350	>	this.root = parent.root;
351	>	phase = parent.register();
352	>	}
353	>	else
354	>	this.root = this;
355	>	this.state = trippedStateFor(phase, parties);
356		}
357
358		/**
#	Line 200 | Line 361 | public class Phaser {
361		* @throws IllegalStateException if attempting to register more
362		* than the maximum supported number of parties.
363		*/
364	<	public int register() { // increment both parties and unarrived
365	<	final AtomicLong state = this.state;
364	>	public int register() {
365	>	return doRegister(1);
366	>	}
367	>
368	>	/**
369	>	* Adds the given number of new unarrived parties to this phaser.
370	>	* @param parties the number of parties required to trip barrier.
371	>	* @return the current barrier phase number upon registration
372	>	* @throws IllegalStateException if attempting to register more
373	>	* than the maximum supported number of parties.
374	>	*/
375	>	public int bulkRegister(int parties) {
376	>	if (parties < 0)
377	>	throw new IllegalArgumentException();
378	>	if (parties == 0)
379	>	return getPhase();
380	>	return doRegister(parties);
381	>	}
382	>
383	>	/**
384	>	* Shared code for register, bulkRegister
385	>	*/
386	>	private int doRegister(int registrations) {
387	>	int phase;
388		for (;;) {
389	<	long s = state.get();
390	<	int phase = phaseOf(s);
391	<	int parties = partiesOf(s) + 1;
392	<	int unarrived = unarrivedOf(s) + 1;
389	>	long s = getReconciledState();
390	>	phase = phaseOf(s);
391	>	int unarrived = unarrivedOf(s) + registrations;
392	>	int parties = partiesOf(s) + registrations;
393	>	if (phase < 0)
394	>	break;
395		if (parties > ushortMask || unarrived > ushortMask)
396		throw badBounds(parties, unarrived);
397	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
398	<	return phase;
397	>	if (phase == phaseOf(root.state) &&
398	>	casState(s, stateFor(phase, parties, unarrived)))
399	>	break;
400		}
401	+	return phase;
402		}
403
404		/**
405		* Arrives at the barrier, but does not wait for others.  (You can
406		* in turn wait for others via {@link #awaitAdvance}).
407		*
408	<	* @return the current barrier phase number upon entry to
409	<	* this method, or a negative value if terminated;
410	<	* @throws IllegalStateException if the number of unarrived
411	<	* parties would become negative.
408	>	* @return the barrier phase number upon entry to this method, or a
409	>	* negative value if terminated;
410	>	* @throws IllegalStateException if not terminated and the number
411	>	* of unarrived parties would become negative.
412		*/
413	<	public int arrive() { // decrement unarrived. If zero, trip
414	<	final AtomicLong state = this.state;
413	>	public int arrive() {
414	>	int phase;
415		for (;;) {
416	<	long s = state.get();
417	<	int phase = phaseOf(s);
416	>	long s = state;
417	>	phase = phaseOf(s);
418		int parties = partiesOf(s);
419		int unarrived = unarrivedOf(s) - 1;
420	<	if (unarrived < 0)
421	<	throw badBounds(parties, unarrived);
422	<	if (unarrived == 0 && phase >= 0) {
423	<	trip(phase, parties);
424	<	return phase;
420	>	if (unarrived > 0) {        // Not the last arrival
421	>	if (casState(s, s - 1)) // s-1 adds one arrival
422	>	break;
423	>	}
424	>	else if (unarrived == 0) {  // the last arrival
425	>	Phaser par = parent;
426	>	if (par == null) {      // directly trip
427	>	if (casState
428	>	(s,
429	>	trippedStateFor(onAdvance(phase, parties)? -1 :
430	>	((phase + 1) & phaseMask), parties))) {
431	>	releaseWaiters(phase);
432	>	break;
433	>	}
434	>	}
435	>	else {                  // cascade to parent
436	>	if (casState(s, s - 1)) { // zeroes unarrived
437	>	par.arrive();
438	>	reconcileState();
439	>	break;
440	>	}
441	>	}
442		}
443	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
444	<	return phase;
443	>	else if (phase < 0) // Don't throw exception if terminated
444	>	break;
445	>	else if (phase != phaseOf(root.state)) // or if unreconciled
446	>	reconcileState();
447	>	else
448	>	throw badBounds(parties, unarrived);
449		}
450	+	return phase;
451		}
452
453		/**
454		* Arrives at the barrier, and deregisters from it, without
455	<	* waiting for others.
455	>	* waiting for others. Deregistration reduces number of parties
456	>	* required to trip the barrier in future phases.  If this phaser
457	>	* has a parent, and deregistration causes this phaser to have
458	>	* zero parties, this phaser is also deregistered from its parent.
459		*
460		* @return the current barrier phase number upon entry to
461		* this method, or a negative value if terminated;
462	<	* @throws IllegalStateException if the number of registered or
463	<	* unarrived parties would become negative.
462	>	* @throws IllegalStateException if not terminated and the number
463	>	* of registered or unarrived parties would become negative.
464		*/
465	<	public int arriveAndDeregister() { // Same as arrive, plus decrement parties
466	<	final AtomicLong state = this.state;
465	>	public int arriveAndDeregister() {
466	>	// similar code to arrive, but too different to merge
467	>	Phaser par = parent;
468	>	int phase;
469		for (;;) {
470	<	long s = state.get();
471	<	int phase = phaseOf(s);
470	>	long s = state;
471	>	phase = phaseOf(s);
472		int parties = partiesOf(s) - 1;
473		int unarrived = unarrivedOf(s) - 1;
474	<	if (parties < 0 || unarrived < 0)
475	<	throw badBounds(parties, unarrived);
476	<	if (unarrived == 0 && phase >= 0) {
477	<	trip(phase, parties);
478	<	return phase;
474	>	if (parties >= 0) {
475	>	if (unarrived > 0 || (unarrived == 0 && par != null)) {
476	>	if (casState
477	>	(s,
478	>	stateFor(phase, parties, unarrived))) {
479	>	if (unarrived == 0) {
480	>	par.arriveAndDeregister();
481	>	reconcileState();
482	>	}
483	>	break;
484	>	}
485	>	continue;
486	>	}
487	>	if (unarrived == 0) {
488	>	if (casState
489	>	(s,
490	>	trippedStateFor(onAdvance(phase, parties)? -1 :
491	>	((phase + 1) & phaseMask), parties))) {
492	>	releaseWaiters(phase);
493	>	break;
494	>	}
495	>	continue;
496	>	}
497	>	if (phase < 0)
498	>	break;
499	>	if (par != null && phase != phaseOf(root.state)) {
500	>	reconcileState();
501	>	continue;
502	>	}
503		}
504	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
267	<	return phase;
504	>	throw badBounds(parties, unarrived);
505		}
506	+	return phase;
507		}
508
509		/**
510	<	* Arrives at the barrier and awaits others. Unlike other arrival
511	<	* methods, this method returns the arrival index of the
512	<	* caller. The caller tripping the barrier returns zero, the
513	<	* previous caller 1, and so on.
514	<	* @return the arrival index
515	<	* @throws IllegalStateException if the number of unarrived
516	<	* parties would become negative.
510	>	* Arrives at the barrier and awaits others. Equivalent in effect
511	>	* to <tt>awaitAdvance(arrive())</tt>.  If you instead need to
512	>	* await with interruption of timeout, and/or deregister upon
513	>	* arrival, you can arrange them using analogous constructions.
514	>	* @return the phase on entry to this method
515	>	* @throws IllegalStateException if not terminated and the number
516	>	* of unarrived parties would become negative.
517		*/
518		public int arriveAndAwaitAdvance() {
519	<	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	<	}
519	>	return awaitAdvance(arrive());
520		}
521
522		/**
523		* Awaits the phase of the barrier to advance from the given
524	<	* value, or returns immediately if this barrier is terminated.
524	>	* value, or returns immediately if argument is negative or this
525	>	* barrier is terminated.
526		* @param phase the phase on entry to this method
527		* @return the phase on exit from this method
528		*/
529		public int awaitAdvance(int phase) {
530		if (phase < 0)
531		return phase;
532	<	Thread current = Thread.currentThread();
533	<	if (current instanceof ForkJoinWorkerThread)
534	<	return helpingWait(phase);
535	<	if (untimedWait(current, phase, false))
536	<	current.interrupt();
537	<	return phaseOf(state.get());
532	>	long s = getReconciledState();
533	>	int p = phaseOf(s);
534	>	if (p != phase)
535	>	return p;
536	>	if (unarrivedOf(s) == 0)
537	>	parent.awaitAdvance(phase);
538	>	// Fall here even if parent waited, to reconcile and help release
539	>	return untimedWait(phase);
540		}
541
542		/**
543		* Awaits the phase of the barrier to advance from the given
544	<	* value, or returns immediately if this barrier is terminated, or
545	<	* throws InterruptedException if interrupted while waiting.
544	>	* value, or returns immediately if argumet is negative or this
545	>	* barrier is terminated, or throws InterruptedException if
546	>	* interrupted while waiting.
547		* @param phase the phase on entry to this method
548		* @return the phase on exit from this method
549		* @throws InterruptedException if thread interrupted while waiting
#	Line 325 | Line 551 | public class Phaser {
551		public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
552		if (phase < 0)
553		return phase;
554	<	Thread current = Thread.currentThread();
555	<	if (current instanceof ForkJoinWorkerThread)
556	<	return helpingWait(phase);
557	<	else if (Thread.interrupted() || untimedWait(current, phase, true))
558	<	throw new InterruptedException();
559	<	else
560	<	return phaseOf(state.get());
554	>	long s = getReconciledState();
555	>	int p = phaseOf(s);
556	>	if (p != phase)
557	>	return p;
558	>	if (unarrivedOf(s) != 0)
559	>	parent.awaitAdvanceInterruptibly(phase);
560	>	return interruptibleWait(phase);
561		}
562
563		/**
564		* Awaits the phase of the barrier to advance from the given value
565	<	* or the given timeout elapses, or returns immediately if this
566	<	* barrier is terminated.
565	>	* or the given timeout elapses, or returns immediately if
566	>	* argument is negative or this barrier is terminated.
567		* @param phase the phase on entry to this method
568		* @return the phase on exit from this method
569		* @throws InterruptedException if thread interrupted while waiting
#	Line 347 | Line 573 | public class Phaser {
573		throws InterruptedException, TimeoutException {
574		if (phase < 0)
575		return phase;
576	<	long nanos = unit.toNanos(timeout);
577	<	Thread current = Thread.currentThread();
578	<	if (current instanceof ForkJoinWorkerThread)
579	<	return timedHelpingWait(phase, nanos);
580	<	timedWait(current, phase, nanos);
581	<	return phaseOf(state.get());
576	>	long s = getReconciledState();
577	>	int p = phaseOf(s);
578	>	if (p != phase)
579	>	return p;
580	>	if (unarrivedOf(s) == 0)
581	>	parent.awaitAdvanceInterruptibly(phase, timeout, unit);
582	>	return timedWait(phase, unit.toNanos(timeout));
583		}
584
585		/**
586		* Forces this barrier to enter termination state. Counts of
587	<	* arrived and registered parties are unaffected. This method may
588	<	* be useful for coordinating recovery after one or more tasks
589	<	* encounter unexpected exceptions.
587	>	* arrived and registered parties are unaffected. If this phaser
588	>	* has a parent, it too is terminated. This method may be useful
589	>	* for coordinating recovery after one or more tasks encounter
590	>	* unexpected exceptions.
591		*/
592		public void forceTermination() {
365	–	final AtomicLong state = this.state;
593		for (;;) {
594	<	long s = state.get();
594	>	long s = getReconciledState();
595		int phase = phaseOf(s);
596		int parties = partiesOf(s);
597		int unarrived = unarrivedOf(s);
598		if (phase < 0 ||
599	<	state.compareAndSet(s, stateFor(-1, parties, unarrived))) {
600	<	if (head.get() != null)
601	<	releaseWaiters(-1);
599	>	casState(s, stateFor(-1, parties, unarrived))) {
600	>	releaseWaiters(0);
601	>	releaseWaiters(1);
602	>	if (parent != null)
603	>	parent.forceTermination();
604		return;
605		}
606		}
607		}
608
609		/**
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	–	/**
610		* Returns the current phase number. The maximum phase number is
611		* <tt>Integer.MAX_VALUE</tt>, after which it restarts at
612		* zero. Upon termination, the phase number is negative.
613		* @return the phase number, or a negative value if terminated
614		*/
615	<	public int getPhase() {
616	<	return phaseOf(state.get());
615	>	public final int getPhase() {
616	>	return phaseOf(getReconciledState());
617	>	}
618	>
619	>	/**
620	>	* Returns true if the current phase number equals the given phase.
621	>	* @param phase the phase
622	>	* @return true if the current phase number equals the given phase.
623	>	*/
624	>	public final boolean hasPhase(int phase) {
625	>	return phaseOf(getReconciledState()) == phase;
626		}
627
628		/**
#	Line 409 | Line 630 | public class Phaser {
630		* @return the number of parties
631		*/
632		public int getRegisteredParties() {
633	<	return partiesOf(state.get());
633	>	return partiesOf(state);
634		}
635
636		/**
#	Line 418 | Line 639 | public class Phaser {
639		* @return the number of arrived parties
640		*/
641		public int getArrivedParties() {
642	<	return arrivedOf(state.get());
642	>	return arrivedOf(state);
643		}
644
645		/**
#	Line 427 | Line 648 | public class Phaser {
648		* @return the number of unarrived parties
649		*/
650		public int getUnarrivedParties() {
651	<	return unarrivedOf(state.get());
651	>	return unarrivedOf(state);
652	>	}
653	>
654	>	/**
655	>	* Returns the parent of this phaser, or null if none.
656	>	* @return the parent of this phaser, or null if none.
657	>	*/
658	>	public Phaser getParent() {
659	>	return parent;
660	>	}
661	>
662	>	/**
663	>	* Returns the root ancestor of this phaser, which is the same as
664	>	* this phaser if it has no parent.
665	>	* @return the root ancestor of this phaser.
666	>	*/
667	>	public Phaser getRoot() {
668	>	return root;
669		}
670
671		/**
#	Line 435 | Line 673 | public class Phaser {
673		* @return true if this barrier has been terminated
674		*/
675		public boolean isTerminated() {
676	<	return phaseOf(state.get()) < 0;
676	>	return getPhase() < 0;
677		}
678
679		/**
#	Line 452 | Line 690 | public class Phaser {
690		* termination for other reasons should also preserve this
691		* property.
692		*
693	+	* <p> You may override this method to perform an action with side
694	+	* effects visible to participating tasks, but it is in general
695	+	* only sensible to do so in designs where all parties register
696	+	* before any arrive, and all <tt>awaitAdvance</tt> at each phase.
697	+	* Otherwise, you cannot ensure lack of interference. In
698	+	* particular, this method may be invoked more than once per
699	+	* transition if other parties successfully register while the
700	+	* invocation of this method is in progress, thus postponing the
701	+	* transition until those parties also arrive, re-triggering this
702	+	* method.
703	+	*
704		* @param phase the phase number on entering the barrier
705		* @param registeredParties the current number of registered
706		* parties.
#	Line 462 | Line 711 | public class Phaser {
711		}
712
713		/**
714	<	* Returns a string identifying this barrier, as well as its
714	>	* Returns a string identifying this phaser, as well as its
715		* state.  The state, in brackets, includes the String {@code
716		* "phase ="} followed by the phase number, {@code "parties ="}
717		* followed by the number of registered parties, and {@code
#	Line 471 | Line 720 | public class Phaser {
720		* @return a string identifying this barrier, as well as its state
721		*/
722		public String toString() {
723	<	long s = state.get();
723	>	long s = getReconciledState();
724		return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
725		}
726
727	<	// methods for tripping and waiting
479	<
480	<	/**
481	<	* Advance the current phase (or terminate)
482	<	*/
483	<	private void trip(int phase, int parties) {
484	<	int next = onAdvance(phase, parties)? -1 : ((phase + 1) & phaseMask);
485	<	state.set(stateFor(next, parties, parties));
486	<	if (head.get() != null)
487	<	releaseWaiters(next);
488	<	}
489	<
490	<	private int helpingWait(int phase) {
491	<	final AtomicLong state = this.state;
492	<	int p;
493	<	while ((p = phaseOf(state.get())) == phase) {
494	<	ForkJoinTask<?> t = ForkJoinWorkerThread.pollTask();
495	<	if (t != null) {
496	<	if ((p = phaseOf(state.get())) == phase)
497	<	t.exec();
498	<	else {   // push task and exit if barrier advanced
499	<	t.fork();
500	<	break;
501	<	}
502	<	}
503	<	}
504	<	return p;
505	<	}
506	<
507	<	private int timedHelpingWait(int phase, long nanos) throws TimeoutException {
508	<	final AtomicLong state = this.state;
509	<	long lastTime = System.nanoTime();
510	<	int p;
511	<	while ((p = phaseOf(state.get())) == phase) {
512	<	long now = System.nanoTime();
513	<	nanos -= now - lastTime;
514	<	lastTime = now;
515	<	if (nanos <= 0) {
516	<	if ((p = phaseOf(state.get())) == phase)
517	<	throw new TimeoutException();
518	<	else
519	<	break;
520	<	}
521	<	ForkJoinTask<?> t = ForkJoinWorkerThread.pollTask();
522	<	if (t != null) {
523	<	if ((p = phaseOf(state.get())) == phase)
524	<	t.exec();
525	<	else {   // push task and exit if barrier advanced
526	<	t.fork();
527	<	break;
528	<	}
529	<	}
530	<	}
531	<	return p;
532	<	}
533	<
534	<	/**
535	<	* 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.
538	<	*/
539	<	static final class QNode {
540	<	QNode next;
541	<	volatile Thread thread; // nulled to cancel wait
542	<	final int phase;
543	<	QNode(Thread t, int c) {
544	<	thread = t;
545	<	phase = c;
546	<	}
547	<	}
548	<
549	<	private void releaseWaiters(int currentPhase) {
550	<	final AtomicReference<QNode> head = this.head;
551	<	QNode p;
552	<	while ((p = head.get()) != null && p.phase != currentPhase) {
553	<	if (head.compareAndSet(p, null)) {
554	<	do {
555	<	Thread t = p.thread;
556	<	if (t != null) {
557	<	p.thread = null;
558	<	LockSupport.unpark(t);
559	<	}
560	<	} while ((p = p.next) != null);
561	<	}
562	<	}
563	<	}
727	>	// methods for waiting
728
729		/** The number of CPUs, for spin control */
730		static final int NCPUS = Runtime.getRuntime().availableProcessors();
#	Line 585 | Line 749 | public class Phaser {
749		static final long spinForTimeoutThreshold = 1000L;
750
751		/**
752	+	* Wait nodes for Treiber stack representing wait queue for non-FJ
753	+	* tasks.
754	+	*/
755	+	static final class QNode {
756	+	QNode next;
757	+	volatile Thread thread; // nulled to cancel wait
758	+	QNode() {
759	+	thread = Thread.currentThread();
760	+	}
761	+	void signal() {
762	+	Thread t = thread;
763	+	if (t != null) {
764	+	thread = null;
765	+	LockSupport.unpark(t);
766	+	}
767	+	}
768	+	}
769	+
770	+	/**
771	+	* Removes and signals waiting threads from wait queue
772	+	*/
773	+	private void releaseWaiters(int phase) {
774	+	AtomicReference<QNode> head = queueFor(phase);
775	+	QNode q;
776	+	while ((q = head.get()) != null) {
777	+	if (head.compareAndSet(q, q.next))
778	+	q.signal();
779	+	}
780	+	}
781	+
782	+	/**
783		* Enqueues node and waits unless aborted or signalled.
784		*/
785	<	private boolean untimedWait(Thread thread, int currentPhase,
786	<	boolean abortOnInterrupt) {
592	<	final AtomicReference<QNode> head = this.head;
593	<	final AtomicLong state = this.state;
594	<	boolean wasInterrupted = false;
785	>	private int untimedWait(int phase) {
786	>	int spins = maxUntimedSpins;
787		QNode node = null;
788	+	boolean interrupted = false;
789		boolean queued = false;
790	<	int spins = maxUntimedSpins;
791	<	while (phaseOf(state.get()) == currentPhase) {
792	<	QNode h;
793	<	if (node != null && queued) {
794	<	if (node.thread != null) {
795	<	LockSupport.park();
796	<	if (Thread.interrupted()) {
604	<	wasInterrupted = true;
605	<	if (abortOnInterrupt)
606	<	break;
607	<	}
790	>	int p;
791	>	while ((p = getPhase()) == phase) {
792	>	interrupted = Thread.interrupted();
793	>	if (node != null) {
794	>	if (!queued) {
795	>	AtomicReference<QNode> head = queueFor(phase);
796	>	queued = head.compareAndSet(node.next = head.get(), node);
797		}
798	+	else if (node.thread != null)
799	+	LockSupport.park(this);
800		}
801	<	else if ((h = head.get()) != null && h.phase != currentPhase) {
802	<	if (phaseOf(state.get()) == currentPhase) { // must recheck
803	<	if (head.compareAndSet(h, h.next)) {
804	<	Thread t = h.thread; // help clear out old waiters
805	<	if (t != null) {
806	<	h.thread = null;
807	<	LockSupport.unpark(t);
808	<	}
809	<	}
801	>	else if (spins <= 0)
802	>	node = new QNode();
803	>	else
804	>	--spins;
805	>	}
806	>	if (node != null)
807	>	node.thread = null;
808	>	if (interrupted)
809	>	Thread.currentThread().interrupt();
810	>	releaseWaiters(phase);
811	>	return p;
812	>	}
813	>
814	>	/**
815	>	* Messier interruptible version
816	>	*/
817	>	private int interruptibleWait(int phase) throws InterruptedException {
818	>	int spins = maxUntimedSpins;
819	>	QNode node = null;
820	>	boolean queued = false;
821	>	boolean interrupted = false;
822	>	int p;
823	>	while ((p = getPhase()) == phase) {
824	>	if (interrupted = Thread.interrupted())
825	>	break;
826	>	if (node != null) {
827	>	if (!queued) {
828	>	AtomicReference<QNode> head = queueFor(phase);
829	>	queued = head.compareAndSet(node.next = head.get(), node);
830		}
831	<	else
832	<	break;
831	>	else if (node.thread != null)
832	>	LockSupport.park(this);
833		}
623	–	else if (node != null)
624	–	queued = head.compareAndSet(node.next = h, node);
834		else if (spins <= 0)
835	<	node = new QNode(thread, currentPhase);
835	>	node = new QNode();
836		else
837		--spins;
838		}
839		if (node != null)
840		node.thread = null;
841	<	return wasInterrupted;
841	>	if (interrupted)
842	>	throw new InterruptedException();
843	>	releaseWaiters(phase);
844	>	return p;
845		}
846
847		/**
848	<	* Messier timeout version
848	>	* Even messier timeout version.
849		*/
850	<	private void timedWait(Thread thread, int currentPhase, long nanos)
850	>	private int timedWait(int phase, long nanos)
851		throws InterruptedException, TimeoutException {
852	<	final AtomicReference<QNode> head = this.head;
853	<	final AtomicLong state = this.state;
854	<	long lastTime = System.nanoTime();
855	<	QNode node = null;
856	<	boolean queued = false;
857	<	int spins = maxTimedSpins;
858	<	while (phaseOf(state.get()) == currentPhase) {
859	<	QNode h;
860	<	long now = System.nanoTime();
649	<	nanos -= now - lastTime;
650	<	lastTime = now;
651	<	if (nanos <= 0) {
652	<	if (node != null)
653	<	node.thread = null;
654	<	if (phaseOf(state.get()) == currentPhase)
655	<	throw new TimeoutException();
656	<	else
852	>	int p;
853	>	if ((p = getPhase()) == phase) {
854	>	long lastTime = System.nanoTime();
855	>	int spins = maxTimedSpins;
856	>	QNode node = null;
857	>	boolean queued = false;
858	>	boolean interrupted = false;
859	>	while ((p = getPhase()) == phase) {
860	>	if (interrupted = Thread.interrupted())
861		break;
862	<	}
863	<	else if (node != null && queued) {
864	<	if (node.thread != null &&
865	<	nanos > spinForTimeoutThreshold) {
866	<	//                LockSupport.parkNanos(this, nanos);
867	<	LockSupport.parkNanos(nanos);
868	<	if (Thread.interrupted()) {
869	<	node.thread = null;
666	<	throw new InterruptedException();
862	>	long now = System.nanoTime();
863	>	if ((nanos -= now - lastTime) <= 0)
864	>	break;
865	>	lastTime = now;
866	>	if (node != null) {
867	>	if (!queued) {
868	>	AtomicReference<QNode> head = queueFor(phase);
869	>	queued = head.compareAndSet(node.next = head.get(), node);
870		}
871	<	}
872	<	}
873	<	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	<	}
871	>	else if (node.thread != null &&
872	>	nanos > spinForTimeoutThreshold) {
873	>	LockSupport.parkNanos(this, nanos);
874		}
875		}
876	+	else if (spins <= 0)
877	+	node = new QNode();
878		else
879	<	break;
879	>	--spins;
880	>	}
881	>	if (node != null)
882	>	node.thread = null;
883	>	if (interrupted)
884	>	throw new InterruptedException();
885	>	if (p == phase && (p = getPhase()) == phase)
886	>	throw new TimeoutException();
887	>	}
888	>	releaseWaiters(phase);
889	>	return p;
890	>	}
891	>
892	>	// Temporary Unsafe mechanics for preliminary release
893	>
894	>	static final Unsafe _unsafe;
895	>	static final long stateOffset;
896	>
897	>	static {
898	>	try {
899	>	if (Phaser.class.getClassLoader() != null) {
900	>	Field f = Unsafe.class.getDeclaredField("theUnsafe");
901	>	f.setAccessible(true);
902	>	_unsafe = (Unsafe)f.get(null);
903		}
683	–	else if (node != null)
684	–	queued = head.compareAndSet(node.next = h, node);
685	–	else if (spins <= 0)
686	–	node = new QNode(thread, currentPhase);
904		else
905	<	--spins;
905	>	_unsafe = Unsafe.getUnsafe();
906	>	stateOffset = _unsafe.objectFieldOffset
907	>	(Phaser.class.getDeclaredField("state"));
908	>	} catch (Exception e) {
909	>	throw new RuntimeException("Could not initialize intrinsics", e);
910		}
690	–	if (node != null)
691	–	node.thread = null;
911		}
912
913	+	final boolean casState(long cmp, long val) {
914	+	return _unsafe.compareAndSwapLong(this, stateOffset, cmp, val);
915	+	}
916		}

Diff Legend

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