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 1.4 by dl, Sat Sep 6 13:19:17 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 flecible
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	>	*   phaser.arrive(); // allow threads to start
106	>	*   int p = phaser.arriveAndDeregister(); // deregister self
107	>	*   otherActions(); // do other things while tasks execute
108	>	*   phaser.awaitAdvance(p); // wait for all tasks to arrive
109	>	* }
110	>	* </pre>
111	>	*
112	>	* <p>One way to cause a set of threads to repeatedly perform actions
113	>	* for a given number of iterations is to override <tt>onAdvance</tt>:
114	>	*
115	>	* <pre>
116	>	*  void startTasks(List&lt;Runnable&gt; list, final int iterations) {
117	>	*    final Phaser phaser = new Phaser() {
118	>	*       public boolean onAdvance(int phase, int registeredParties) {
119	>	*         return phase &gt;= iterations || registeredParties == 0;
120	>	*       }
121	>	*    };
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();
133		*   }
134	+	*   phaser.arriveAndDeregister(); // deregister self, don't wait
135		* }
136		* </pre>
137		*
138	+	* <p> To create a set of tasks using a tree of Phasers,
139	+	* you could use code of the following form, assuming a
140	+	* Task class with a constructor accepting a Phaser that
141	+	* it registers for upon construction:
142	+	* <pre>
143	+	*  void build(Task[] actions, int lo, int hi, Phaser b) {
144	+	*    int step = (hi - lo) / TASKS_PER_PHASER;
145	+	*    if (step &gt; 1) {
146	+	*       int i = lo;
147	+	*       while (i &lt; hi) {
148	+	*         int r = Math.min(i + step, hi);
149	+	*         build(actions, i, r, new Phaser(b));
150	+	*         i = r;
151	+	*       }
152	+	*    }
153	+	*    else {
154	+	*      for (int i = lo; i &lt; hi; ++i)
155	+	*        actions[i] = new Task(b);
156	+	*        // assumes new Task(b) performs b.register()
157	+	*    }
158	+	*  }
159	+	*  // .. initially called, for n tasks via
160	+	*  build(new Task[n], 0, n, new Phaser());
161	+	* </pre>
162	+	*
163	+	* The best value of <tt>TASKS_PER_PHASER</tt> depends mainly on
164	+	* expected barrier synchronization rates. A value as low as four may
165	+	* be appropriate for extremely small per-barrier task bodies (thus
166	+	* high rates), or up to hundreds for extremely large ones.
167	+	*
168	+	* </pre>
169	+	*
170		* <p><b>Implementation notes</b>: This implementation restricts the
171	<	* maximum number of parties to 65535. Attempts to register
172	<	* additional parties result in IllegalStateExceptions.
171	>	* maximum number of parties to 65535. Attempts to register additional
172	>	* parties result in IllegalStateExceptions. However, you can and
173	>	* should create tiered phasers to accommodate arbitrarily large sets
174	>	* of participants.
175		*/
176		public class Phaser {
177		/*
178		* This class implements an extension of X10 "clocks".  Thanks to
179	<	* Vijay Saraswat for the idea of applying it to ForkJoinTasks,
180	<	* and to Vivek Sarkar for enhancements to extend functionality.
179	>	* Vijay Saraswat for the idea, and to Vivek Sarkar for
180	>	* enhancements to extend functionality.
181		*/
182
183		/**
#	Line 133 | Line 190 | public class Phaser {
190		* * terminated -- set if barrier is terminated (1 bit)
191		*
192		* However, to efficiently maintain atomicity, these values are
193	<	* packed into a single AtomicLong. Termination uses the sign bit
194	<	* of 32 bit representation of phase, so phase is set to -1 on
195	<	* termination.
196	<	*/
197	<	private final AtomicLong state;
198	<
199	<	/**
143	<	* Head of Treiber stack for waiting nonFJ threads.
193	>	* packed into a single (atomic) long. Termination uses the sign
194	>	* bit of 32 bit representation of phase, so phase is set to -1 on
195	>	* termination. Good performace relies on keeping state decoding
196	>	* and encoding simple, and keeping race windows short.
197	>	*
198	>	* Note: there are some cheats in arrive() that rely on unarrived
199	>	* being lowest 16 bits.
200		*/
201	<	private final AtomicReference<QNode> head = new AtomicReference<QNode>();
201	>	private volatile long state;
202
203		private static final int ushortBits = 16;
204		private static final int ushortMask =  (1 << ushortBits) - 1;
#	Line 168 | Line 224 | public class Phaser {
224		return (((long)phase) << 32) | ((parties << 16) | unarrived);
225		}
226
227	+	private static long trippedStateFor(int phase, int parties) {
228	+	return (((long)phase) << 32) | ((parties << 16) | parties);
229	+	}
230	+
231		private static IllegalStateException badBounds(int parties, int unarrived) {
232	<	return new IllegalStateException("Attempt to set " + unarrived +
233	<	" unarrived of " + parties + " parties");
232	>	return new IllegalStateException
233	>	("Attempt to set " + unarrived +
234	>	" unarrived of " + parties + " parties");
235	>	}
236	>
237	>	/**
238	>	* The parent of this phaser, or null if none
239	>	*/
240	>	private final Phaser parent;
241	>
242	>	/**
243	>	* The root of Phaser tree. Equals this if not in a tree.  Used to
244	>	* support faster state push-down.
245	>	*/
246	>	private final Phaser root;
247	>
248	>	// Wait queues
249	>
250	>	/**
251	>	* Heads of Treiber stacks waiting for nonFJ threads. To eliminate
252	>	* contention while releasing some threads while adding others, we
253	>	* use two of them, alternating across even and odd phases.
254	>	*/
255	>	private final AtomicReference<QNode> evenQ = new AtomicReference<QNode>();
256	>	private final AtomicReference<QNode> oddQ  = new AtomicReference<QNode>();
257	>
258	>	private AtomicReference<QNode> queueFor(int phase) {
259	>	return (phase & 1) == 0? evenQ : oddQ;
260	>	}
261	>
262	>	/**
263	>	* Returns current state, first resolving lagged propagation from
264	>	* root if necessary.
265	>	*/
266	>	private long getReconciledState() {
267	>	return parent == null? state : reconcileState();
268	>	}
269	>
270	>	/**
271	>	* Recursively resolves state.
272	>	*/
273	>	private long reconcileState() {
274	>	Phaser p = parent;
275	>	long s = state;
276	>	if (p != null) {
277	>	while (unarrivedOf(s) == 0 && phaseOf(s) != phaseOf(root.state)) {
278	>	long parentState = p.getReconciledState();
279	>	int parentPhase = phaseOf(parentState);
280	>	int phase = phaseOf(s = state);
281	>	if (phase != parentPhase) {
282	>	long next = trippedStateFor(parentPhase, partiesOf(s));
283	>	if (casState(s, next)) {
284	>	releaseWaiters(phase);
285	>	s = next;
286	>	}
287	>	}
288	>	}
289	>	}
290	>	return s;
291		}
292
293		/**
294		* Creates a new Phaser without any initially registered parties,
295	<	* and initial phase number 0.
295	>	* initial phase number 0, and no parent.
296		*/
297		public Phaser() {
298	<	state = new AtomicLong(stateFor(0, 0, 0));
298	>	this(null);
299		}
300
301		/**
302		* Creates a new Phaser with the given numbers of registered
303	<	* unarrived parties and initial phase number 0.
303	>	* unarrived parties, initial phase number 0, and no parent.
304		* @param parties the number of parties required to trip barrier.
305		* @throws IllegalArgumentException if parties less than zero
306		* or greater than the maximum number of parties supported.
307		*/
308		public Phaser(int parties) {
309	+	this(null, parties);
310	+	}
311	+
312	+	/**
313	+	* Creates a new Phaser with the given parent, without any
314	+	* initially registered parties. If parent is non-null this phaser
315	+	* is registered with the parent and its initial phase number is
316	+	* the same as that of parent phaser.
317	+	* @param parent the parent phaser.
318	+	*/
319	+	public Phaser(Phaser parent) {
320	+	int phase = 0;
321	+	this.parent = parent;
322	+	if (parent != null) {
323	+	this.root = parent.root;
324	+	phase = parent.register();
325	+	}
326	+	else
327	+	this.root = this;
328	+	this.state = trippedStateFor(phase, 0);
329	+	}
330	+
331	+	/**
332	+	* Creates a new Phaser with the given parent and numbers of
333	+	* registered unarrived parties. If parent is non-null this phaser
334	+	* is registered with the parent and its initial phase number is
335	+	* the same as that of parent phaser.
336	+	* @param parent the parent phaser.
337	+	* @param parties the number of parties required to trip barrier.
338	+	* @throws IllegalArgumentException if parties less than zero
339	+	* or greater than the maximum number of parties supported.
340	+	*/
341	+	public Phaser(Phaser parent, int parties) {
342		if (parties < 0 || parties > ushortMask)
343		throw new IllegalArgumentException("Illegal number of parties");
344	<	state = new AtomicLong(stateFor(0, parties, parties));
344	>	int phase = 0;
345	>	this.parent = parent;
346	>	if (parent != null) {
347	>	this.root = parent.root;
348	>	phase = parent.register();
349	>	}
350	>	else
351	>	this.root = this;
352	>	this.state = trippedStateFor(phase, parties);
353		}
354
355		/**
#	Line 200 | Line 358 | public class Phaser {
358		* @throws IllegalStateException if attempting to register more
359		* than the maximum supported number of parties.
360		*/
361	<	public int register() { // increment both parties and unarrived
362	<	final AtomicLong state = this.state;
361	>	public int register() {
362	>	return doRegister(1);
363	>	}
364	>
365	>	/**
366	>	* Adds the given number of new unarrived parties to this phaser.
367	>	* @param parties the number of parties required to trip barrier.
368	>	* @return the current barrier phase number upon registration
369	>	* @throws IllegalStateException if attempting to register more
370	>	* than the maximum supported number of parties.
371	>	*/
372	>	public int bulkRegister(int parties) {
373	>	if (parties < 0)
374	>	throw new IllegalArgumentException();
375	>	if (parties == 0)
376	>	return getPhase();
377	>	return doRegister(parties);
378	>	}
379	>
380	>	/**
381	>	* Shared code for register, bulkRegister
382	>	*/
383	>	private int doRegister(int registrations) {
384	>	int phase;
385		for (;;) {
386	<	long s = state.get();
387	<	int phase = phaseOf(s);
388	<	int parties = partiesOf(s) + 1;
389	<	int unarrived = unarrivedOf(s) + 1;
386	>	long s = getReconciledState();
387	>	phase = phaseOf(s);
388	>	int unarrived = unarrivedOf(s) + registrations;
389	>	int parties = partiesOf(s) + registrations;
390	>	if (phase < 0)
391	>	break;
392		if (parties > ushortMask || unarrived > ushortMask)
393		throw badBounds(parties, unarrived);
394	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
395	<	return phase;
394	>	if (phase == phaseOf(root.state) &&
395	>	casState(s, stateFor(phase, parties, unarrived)))
396	>	break;
397		}
398	+	return phase;
399		}
400
401		/**
402		* Arrives at the barrier, but does not wait for others.  (You can
403		* in turn wait for others via {@link #awaitAdvance}).
404		*
405	<	* @return the current barrier phase number upon entry to
406	<	* this method, or a negative value if terminated;
407	<	* @throws IllegalStateException if the number of unarrived
408	<	* parties would become negative.
405	>	* @return the barrier phase number upon entry to this method, or a
406	>	* negative value if terminated;
407	>	* @throws IllegalStateException if not terminated and the number
408	>	* of unarrived parties would become negative.
409		*/
410	<	public int arrive() { // decrement unarrived. If zero, trip
411	<	final AtomicLong state = this.state;
410	>	public int arrive() {
411	>	int phase;
412		for (;;) {
413	<	long s = state.get();
414	<	int phase = phaseOf(s);
413	>	long s = state;
414	>	phase = phaseOf(s);
415		int parties = partiesOf(s);
416		int unarrived = unarrivedOf(s) - 1;
417	<	if (unarrived < 0)
418	<	throw badBounds(parties, unarrived);
419	<	if (unarrived == 0 && phase >= 0) {
420	<	trip(phase, parties);
421	<	return phase;
417	>	if (unarrived > 0) {        // Not the last arrival
418	>	if (casState(s, s - 1)) // s-1 adds one arrival
419	>	break;
420	>	}
421	>	else if (unarrived == 0) {  // the last arrival
422	>	Phaser par = parent;
423	>	if (par == null) {      // directly trip
424	>	if (casState
425	>	(s,
426	>	trippedStateFor(onAdvance(phase, parties)? -1 :
427	>	((phase + 1) & phaseMask), parties))) {
428	>	releaseWaiters(phase);
429	>	break;
430	>	}
431	>	}
432	>	else {                  // cascade to parent
433	>	if (casState(s, s - 1)) { // zeroes unarrived
434	>	par.arrive();
435	>	reconcileState();
436	>	break;
437	>	}
438	>	}
439		}
440	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
441	<	return phase;
440	>	else if (phase < 0) // Don't throw exception if terminated
441	>	break;
442	>	else if (phase != phaseOf(root.state)) // or if unreconciled
443	>	reconcileState();
444	>	else
445	>	throw badBounds(parties, unarrived);
446		}
447	+	return phase;
448		}
449
450		/**
451		* Arrives at the barrier, and deregisters from it, without
452	<	* waiting for others.
452	>	* waiting for others. Deregistration reduces number of parties
453	>	* required to trip the barrier in future phases.  If this phaser
454	>	* has a parent, and deregistration causes this phaser to have
455	>	* zero parties, this phaser is also deregistered from its parent.
456		*
457		* @return the current barrier phase number upon entry to
458		* this method, or a negative value if terminated;
459	<	* @throws IllegalStateException if the number of registered or
460	<	* unarrived parties would become negative.
459	>	* @throws IllegalStateException if not terminated and the number
460	>	* of registered or unarrived parties would become negative.
461		*/
462	<	public int arriveAndDeregister() { // Same as arrive, plus decrement parties
463	<	final AtomicLong state = this.state;
462	>	public int arriveAndDeregister() {
463	>	// similar code to arrive, but too different to merge
464	>	Phaser par = parent;
465	>	int phase;
466		for (;;) {
467	<	long s = state.get();
468	<	int phase = phaseOf(s);
467	>	long s = state;
468	>	phase = phaseOf(s);
469		int parties = partiesOf(s) - 1;
470		int unarrived = unarrivedOf(s) - 1;
471	<	if (parties < 0 || unarrived < 0)
472	<	throw badBounds(parties, unarrived);
473	<	if (unarrived == 0 && phase >= 0) {
474	<	trip(phase, parties);
475	<	return phase;
471	>	if (parties >= 0) {
472	>	if (unarrived > 0 || (unarrived == 0 && par != null)) {
473	>	if (casState
474	>	(s,
475	>	stateFor(phase, parties, unarrived))) {
476	>	if (unarrived == 0) {
477	>	par.arriveAndDeregister();
478	>	reconcileState();
479	>	}
480	>	break;
481	>	}
482	>	continue;
483	>	}
484	>	if (unarrived == 0) {
485	>	if (casState
486	>	(s,
487	>	trippedStateFor(onAdvance(phase, parties)? -1 :
488	>	((phase + 1) & phaseMask), parties))) {
489	>	releaseWaiters(phase);
490	>	break;
491	>	}
492	>	continue;
493	>	}
494	>	if (phase < 0)
495	>	break;
496	>	if (par != null && phase != phaseOf(root.state)) {
497	>	reconcileState();
498	>	continue;
499	>	}
500		}
501	<	if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
267	<	return phase;
501	>	throw badBounds(parties, unarrived);
502		}
503	+	return phase;
504		}
505
506		/**
507	<	* Arrives at the barrier and awaits others. Unlike other arrival
508	<	* methods, this method returns the arrival index of the
509	<	* caller. The caller tripping the barrier returns zero, the
510	<	* previous caller 1, and so on.
511	<	* @return the arrival index
512	<	* @throws IllegalStateException if the number of unarrived
513	<	* parties would become negative.
507	>	* Arrives at the barrier and awaits others. Equivalent in effect
508	>	* to <tt>awaitAdvance(arrive())</tt>.  If you instead need to
509	>	* await with interruption of timeout, and/or deregister upon
510	>	* arrival, you can arrange them using analogous constructions.
511	>	* @return the phase on entry to this method
512	>	* @throws IllegalStateException if not terminated and the number
513	>	* of unarrived parties would become negative.
514		*/
515		public int arriveAndAwaitAdvance() {
516	<	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	<	}
516	>	return awaitAdvance(arrive());
517		}
518
519		/**
520		* Awaits the phase of the barrier to advance from the given
521	<	* value, or returns immediately if this barrier is terminated.
521	>	* value, or returns immediately if argument is negative or this
522	>	* barrier is terminated.
523		* @param phase the phase on entry to this method
524		* @return the phase on exit from this method
525		*/
526		public int awaitAdvance(int phase) {
527		if (phase < 0)
528		return phase;
529	<	Thread current = Thread.currentThread();
530	<	if (current instanceof ForkJoinWorkerThread)
531	<	return helpingWait(phase);
532	<	if (untimedWait(current, phase, false))
533	<	current.interrupt();
534	<	return phaseOf(state.get());
529	>	long s = getReconciledState();
530	>	int p = phaseOf(s);
531	>	if (p != phase)
532	>	return p;
533	>	if (unarrivedOf(s) == 0)
534	>	parent.awaitAdvance(phase);
535	>	// Fall here even if parent waited, to reconcile and help release
536	>	return untimedWait(phase);
537		}
538
539		/**
540		* Awaits the phase of the barrier to advance from the given
541	<	* value, or returns immediately if this barrier is terminated, or
542	<	* throws InterruptedException if interrupted while waiting.
541	>	* value, or returns immediately if argumet is negative or this
542	>	* barrier is terminated, or throws InterruptedException if
543	>	* interrupted while waiting.
544		* @param phase the phase on entry to this method
545		* @return the phase on exit from this method
546		* @throws InterruptedException if thread interrupted while waiting
#	Line 325 | Line 548 | public class Phaser {
548		public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
549		if (phase < 0)
550		return phase;
551	<	Thread current = Thread.currentThread();
552	<	if (current instanceof ForkJoinWorkerThread)
553	<	return helpingWait(phase);
554	<	else if (Thread.interrupted() || untimedWait(current, phase, true))
555	<	throw new InterruptedException();
556	<	else
557	<	return phaseOf(state.get());
551	>	long s = getReconciledState();
552	>	int p = phaseOf(s);
553	>	if (p != phase)
554	>	return p;
555	>	if (unarrivedOf(s) != 0)
556	>	parent.awaitAdvanceInterruptibly(phase);
557	>	return interruptibleWait(phase);
558		}
559
560		/**
561		* Awaits the phase of the barrier to advance from the given value
562	<	* or the given timeout elapses, or returns immediately if this
563	<	* barrier is terminated.
562	>	* or the given timeout elapses, or returns immediately if
563	>	* argument is negative or this barrier is terminated.
564		* @param phase the phase on entry to this method
565		* @return the phase on exit from this method
566		* @throws InterruptedException if thread interrupted while waiting
#	Line 347 | Line 570 | public class Phaser {
570		throws InterruptedException, TimeoutException {
571		if (phase < 0)
572		return phase;
573	<	long nanos = unit.toNanos(timeout);
574	<	Thread current = Thread.currentThread();
575	<	if (current instanceof ForkJoinWorkerThread)
576	<	return timedHelpingWait(phase, nanos);
577	<	timedWait(current, phase, nanos);
578	<	return phaseOf(state.get());
573	>	long s = getReconciledState();
574	>	int p = phaseOf(s);
575	>	if (p != phase)
576	>	return p;
577	>	if (unarrivedOf(s) == 0)
578	>	parent.awaitAdvanceInterruptibly(phase, timeout, unit);
579	>	return timedWait(phase, unit.toNanos(timeout));
580		}
581
582		/**
583		* Forces this barrier to enter termination state. Counts of
584	<	* arrived and registered parties are unaffected. This method may
585	<	* be useful for coordinating recovery after one or more tasks
586	<	* encounter unexpected exceptions.
584	>	* arrived and registered parties are unaffected. If this phaser
585	>	* has a parent, it too is terminated. This method may be useful
586	>	* for coordinating recovery after one or more tasks encounter
587	>	* unexpected exceptions.
588		*/
589		public void forceTermination() {
365	–	final AtomicLong state = this.state;
590		for (;;) {
591	<	long s = state.get();
591	>	long s = getReconciledState();
592		int phase = phaseOf(s);
593		int parties = partiesOf(s);
594		int unarrived = unarrivedOf(s);
595		if (phase < 0 ||
596	<	state.compareAndSet(s, stateFor(-1, parties, unarrived))) {
597	<	if (head.get() != null)
598	<	releaseWaiters(-1);
596	>	casState(s, stateFor(-1, parties, unarrived))) {
597	>	releaseWaiters(0);
598	>	releaseWaiters(1);
599	>	if (parent != null)
600	>	parent.forceTermination();
601		return;
602		}
603		}
604		}
605
606		/**
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	–	/**
607		* Returns the current phase number. The maximum phase number is
608		* <tt>Integer.MAX_VALUE</tt>, after which it restarts at
609		* zero. Upon termination, the phase number is negative.
610		* @return the phase number, or a negative value if terminated
611		*/
612	<	public int getPhase() {
613	<	return phaseOf(state.get());
612	>	public final int getPhase() {
613	>	return phaseOf(getReconciledState());
614	>	}
615	>
616	>	/**
617	>	* Returns true if the current phase number equals the given phase.
618	>	* @param phase the phase
619	>	* @return true if the current phase number equals the given phase.
620	>	*/
621	>	public final boolean hasPhase(int phase) {
622	>	return phaseOf(getReconciledState()) == phase;
623		}
624
625		/**
#	Line 409 | Line 627 | public class Phaser {
627		* @return the number of parties
628		*/
629		public int getRegisteredParties() {
630	<	return partiesOf(state.get());
630	>	return partiesOf(state);
631		}
632
633		/**
#	Line 418 | Line 636 | public class Phaser {
636		* @return the number of arrived parties
637		*/
638		public int getArrivedParties() {
639	<	return arrivedOf(state.get());
639	>	return arrivedOf(state);
640		}
641
642		/**
#	Line 427 | Line 645 | public class Phaser {
645		* @return the number of unarrived parties
646		*/
647		public int getUnarrivedParties() {
648	<	return unarrivedOf(state.get());
648	>	return unarrivedOf(state);
649	>	}
650	>
651	>	/**
652	>	* Returns the parent of this phaser, or null if none.
653	>	* @return the parent of this phaser, or null if none.
654	>	*/
655	>	public Phaser getParent() {
656	>	return parent;
657	>	}
658	>
659	>	/**
660	>	* Returns the root ancestor of this phaser, which is the same as
661	>	* this phaser if it has no parent.
662	>	* @return the root ancestor of this phaser.
663	>	*/
664	>	public Phaser getRoot() {
665	>	return root;
666		}
667
668		/**
#	Line 435 | Line 670 | public class Phaser {
670		* @return true if this barrier has been terminated
671		*/
672		public boolean isTerminated() {
673	<	return phaseOf(state.get()) < 0;
673	>	return getPhase() < 0;
674		}
675
676		/**
#	Line 452 | Line 687 | public class Phaser {
687		* termination for other reasons should also preserve this
688		* property.
689		*
690	+	* <p> You may override this method to perform an action with side
691	+	* effects visible to participating tasks, but it is in general
692	+	* only sensible to do so in designs where all parties register
693	+	* before any arrive, and all <tt>awaitAdvance</tt> at each phase.
694	+	* Otherwise, you cannot ensure lack of interference. In
695	+	* particular, this method may be invoked more than once per
696	+	* transition if other parties successfully register while the
697	+	* invocation of this method is in progress, thus postponing the
698	+	* transition until those parties also arrive, re-triggering this
699	+	* method.
700	+	*
701		* @param phase the phase number on entering the barrier
702		* @param registeredParties the current number of registered
703		* parties.
#	Line 462 | Line 708 | public class Phaser {
708		}
709
710		/**
711	<	* Returns a string identifying this barrier, as well as its
711	>	* Returns a string identifying this phaser, as well as its
712		* state.  The state, in brackets, includes the String {@code
713		* "phase ="} followed by the phase number, {@code "parties ="}
714		* followed by the number of registered parties, and {@code
#	Line 471 | Line 717 | public class Phaser {
717		* @return a string identifying this barrier, as well as its state
718		*/
719		public String toString() {
720	<	long s = state.get();
720	>	long s = getReconciledState();
721		return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
722		}
723
724	<	// 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	<	}
724	>	// methods for waiting
725
726		/** The number of CPUs, for spin control */
727		static final int NCPUS = Runtime.getRuntime().availableProcessors();
#	Line 585 | Line 746 | public class Phaser {
746		static final long spinForTimeoutThreshold = 1000L;
747
748		/**
749	+	* Wait nodes for Treiber stack representing wait queue for non-FJ
750	+	* tasks.
751	+	*/
752	+	static final class QNode {
753	+	QNode next;
754	+	volatile Thread thread; // nulled to cancel wait
755	+	QNode() {
756	+	thread = Thread.currentThread();
757	+	}
758	+	void signal() {
759	+	Thread t = thread;
760	+	if (t != null) {
761	+	thread = null;
762	+	LockSupport.unpark(t);
763	+	}
764	+	}
765	+	}
766	+
767	+	/**
768	+	* Removes and signals waiting threads from wait queue
769	+	*/
770	+	private void releaseWaiters(int phase) {
771	+	AtomicReference<QNode> head = queueFor(phase);
772	+	QNode q;
773	+	while ((q = head.get()) != null) {
774	+	if (head.compareAndSet(q, q.next))
775	+	q.signal();
776	+	}
777	+	}
778	+
779	+	/**
780		* Enqueues node and waits unless aborted or signalled.
781		*/
782	<	private boolean untimedWait(Thread thread, int currentPhase,
783	<	boolean abortOnInterrupt) {
592	<	final AtomicReference<QNode> head = this.head;
593	<	final AtomicLong state = this.state;
594	<	boolean wasInterrupted = false;
782	>	private int untimedWait(int phase) {
783	>	int spins = maxUntimedSpins;
784		QNode node = null;
785	+	boolean interrupted = false;
786		boolean queued = false;
787	<	int spins = maxUntimedSpins;
788	<	while (phaseOf(state.get()) == currentPhase) {
789	<	QNode h;
790	<	if (node != null && queued) {
791	<	if (node.thread != null) {
792	<	LockSupport.park();
793	<	if (Thread.interrupted()) {
604	<	wasInterrupted = true;
605	<	if (abortOnInterrupt)
606	<	break;
607	<	}
787	>	int p;
788	>	while ((p = getPhase()) == phase) {
789	>	interrupted = Thread.interrupted();
790	>	if (node != null) {
791	>	if (!queued) {
792	>	AtomicReference<QNode> head = queueFor(phase);
793	>	queued = head.compareAndSet(node.next = head.get(), node);
794		}
795	+	else if (node.thread != null)
796	+	LockSupport.park(this);
797		}
798	<	else if ((h = head.get()) != null && h.phase != currentPhase) {
799	<	if (phaseOf(state.get()) == currentPhase) { // must recheck
800	<	if (head.compareAndSet(h, h.next)) {
801	<	Thread t = h.thread; // help clear out old waiters
802	<	if (t != null) {
803	<	h.thread = null;
804	<	LockSupport.unpark(t);
805	<	}
806	<	}
798	>	else if (spins <= 0)
799	>	node = new QNode();
800	>	else
801	>	--spins;
802	>	}
803	>	if (node != null)
804	>	node.thread = null;
805	>	if (interrupted)
806	>	Thread.currentThread().interrupt();
807	>	releaseWaiters(phase);
808	>	return p;
809	>	}
810	>
811	>	/**
812	>	* Messier interruptible version
813	>	*/
814	>	private int interruptibleWait(int phase) throws InterruptedException {
815	>	int spins = maxUntimedSpins;
816	>	QNode node = null;
817	>	boolean queued = false;
818	>	boolean interrupted = false;
819	>	int p;
820	>	while ((p = getPhase()) == phase) {
821	>	if (interrupted = Thread.interrupted())
822	>	break;
823	>	if (node != null) {
824	>	if (!queued) {
825	>	AtomicReference<QNode> head = queueFor(phase);
826	>	queued = head.compareAndSet(node.next = head.get(), node);
827		}
828	<	else
829	<	break;
828	>	else if (node.thread != null)
829	>	LockSupport.park(this);
830		}
623	–	else if (node != null)
624	–	queued = head.compareAndSet(node.next = h, node);
831		else if (spins <= 0)
832	<	node = new QNode(thread, currentPhase);
832	>	node = new QNode();
833		else
834		--spins;
835		}
836		if (node != null)
837		node.thread = null;
838	<	return wasInterrupted;
838	>	if (interrupted)
839	>	throw new InterruptedException();
840	>	releaseWaiters(phase);
841	>	return p;
842		}
843
844		/**
845	<	* Messier timeout version
845	>	* Even messier timeout version.
846		*/
847	<	private void timedWait(Thread thread, int currentPhase, long nanos)
847	>	private int timedWait(int phase, long nanos)
848		throws InterruptedException, TimeoutException {
849	<	final AtomicReference<QNode> head = this.head;
850	<	final AtomicLong state = this.state;
851	<	long lastTime = System.nanoTime();
852	<	QNode node = null;
853	<	boolean queued = false;
854	<	int spins = maxTimedSpins;
855	<	while (phaseOf(state.get()) == currentPhase) {
856	<	QNode h;
857	<	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
849	>	int p;
850	>	if ((p = getPhase()) == phase) {
851	>	long lastTime = System.nanoTime();
852	>	int spins = maxTimedSpins;
853	>	QNode node = null;
854	>	boolean queued = false;
855	>	boolean interrupted = false;
856	>	while ((p = getPhase()) == phase) {
857	>	if (interrupted = Thread.interrupted())
858		break;
859	<	}
860	<	else if (node != null && queued) {
861	<	if (node.thread != null &&
862	<	nanos > spinForTimeoutThreshold) {
863	<	//                LockSupport.parkNanos(this, nanos);
864	<	LockSupport.parkNanos(nanos);
865	<	if (Thread.interrupted()) {
866	<	node.thread = null;
666	<	throw new InterruptedException();
859	>	long now = System.nanoTime();
860	>	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	<	}
869	<	}
870	<	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	<	}
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	<	break;
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();
884	>	}
885	>	releaseWaiters(phase);
886	>	return p;
887	>	}
888	>
889	>	// Temporary Unsafe mechanics for preliminary release
890	>
891	>	static final Unsafe _unsafe;
892	>	static final long stateOffset;
893	>
894	>	static {
895	>	try {
896	>	if (Phaser.class.getClassLoader() != null) {
897	>	Field f = Unsafe.class.getDeclaredField("theUnsafe");
898	>	f.setAccessible(true);
899	>	_unsafe = (Unsafe)f.get(null);
900		}
683	–	else if (node != null)
684	–	queued = head.compareAndSet(node.next = h, node);
685	–	else if (spins <= 0)
686	–	node = new QNode(thread, currentPhase);
901		else
902	<	--spins;
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);
907		}
690	–	if (node != null)
691	–	node.thread = null;
908		}
909
910	+	final boolean casState(long cmp, long val) {
911	+	return _unsafe.compareAndSwapLong(this, stateOffset, cmp, val);
912	+	}
913		}

Diff Legend

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