ViewVC Help
View File | Revision Log | Show Annotations | Download File | Root Listing
root/jsr166/jsr166/src/jsr166y/Phaser.java
(Generate patch)

Comparing jsr166/src/jsr166y/Phaser.java (file contents):
Revision 1.3 by jsr166, Fri Jul 25 18:11:53 2008 UTC vs.
Revision 1.81 by jsr166, Fri Jul 15 18:49:12 2016 UTC

# Line 1 | Line 1
1   /*
2   * Written by Doug Lea with assistance from members of JCP JSR-166
3   * Expert Group and released to the public domain, as explained at
4 < * http://creativecommons.org/licenses/publicdomain
4 > * http://creativecommons.org/publicdomain/zero/1.0/
5   */
6  
7   package jsr166y;
8 < import jsr166y.forkjoin.*;
9 < import java.util.concurrent.*;
10 < import java.util.concurrent.atomic.*;
8 >
9 > import java.util.concurrent.TimeUnit;
10 > import java.util.concurrent.TimeoutException;
11 > import java.util.concurrent.atomic.AtomicReference;
12   import java.util.concurrent.locks.LockSupport;
13  
14   /**
15 < * A reusable synchronization barrier, similar in functionality to a
16 < * {@link java.util.concurrent.CyclicBarrier}, but supporting more
17 < * flexible usage.
15 > * A reusable synchronization barrier, similar in functionality to
16 > * {@link java.util.concurrent.CyclicBarrier CyclicBarrier} and
17 > * {@link java.util.concurrent.CountDownLatch CountDownLatch}
18 > * but supporting more flexible usage.
19   *
20 < * <ul>
20 > * <p><b>Registration.</b> Unlike the case for other barriers, the
21 > * number of parties <em>registered</em> to synchronize on a phaser
22 > * may vary over time.  Tasks may be registered at any time (using
23 > * methods {@link #register}, {@link #bulkRegister}, or forms of
24 > * constructors establishing initial numbers of parties), and
25 > * optionally deregistered upon any arrival (using {@link
26 > * #arriveAndDeregister}).  As is the case with most basic
27 > * synchronization constructs, registration and deregistration affect
28 > * only internal counts; they do not establish any further internal
29 > * bookkeeping, so tasks cannot query whether they are registered.
30 > * (However, you can introduce such bookkeeping by subclassing this
31 > * class.)
32   *
33 < * <li> The number of parties synchronizing on the barrier may vary
34 < * over time.  A task may register to be a party in a barrier at any
35 < * time, and may deregister upon arriving at the barrier.  As is the
36 < * case with most basic synchronization constructs, registration
37 < * and deregistration affect only internal counts; they do not
38 < * establish any further internal bookkeeping, so tasks cannot query
39 < * whether they are registered.
40 < *
41 < * <li> Each generation has an associated phase value, starting at
42 < * zero, and advancing when all parties reach the barrier (wrapping
43 < * around to zero after reaching <tt>Integer.MAX_VALUE</tt>).
31 < *
32 < * <li> Like a CyclicBarrier, a Phaser may be repeatedly awaited.
33 < * Method <tt>arriveAndAwaitAdvance</tt> has effect analogous to
34 < * <tt>CyclicBarrier.await</tt>.  However, Phasers separate two
35 < * aspects of coordination, that may be invoked independently:
33 > * <p><b>Synchronization.</b> Like a {@code CyclicBarrier}, a {@code
34 > * Phaser} may be repeatedly awaited.  Method {@link
35 > * #arriveAndAwaitAdvance} has effect analogous to {@link
36 > * java.util.concurrent.CyclicBarrier#await CyclicBarrier.await}. Each
37 > * generation of a phaser has an associated phase number. The phase
38 > * number starts at zero, and advances when all parties arrive at the
39 > * phaser, wrapping around to zero after reaching {@code
40 > * Integer.MAX_VALUE}. The use of phase numbers enables independent
41 > * control of actions upon arrival at a phaser and upon awaiting
42 > * others, via two kinds of methods that may be invoked by any
43 > * registered party:
44   *
45   * <ul>
46   *
47 < *   <li> Arriving at a barrier. Methods <tt>arrive</tt> and
48 < *       <tt>arriveAndDeregister</tt> do not block, but return
49 < *       the phase value on entry to the method.
50 < *
51 < *   <li> Awaiting others. Method <tt>awaitAdvance</tt> requires an
52 < *       argument indicating the entry phase, and returns when the
53 < *       barrier advances to a new phase.
47 > *   <li><b>Arrival.</b> Methods {@link #arrive} and
48 > *       {@link #arriveAndDeregister} record arrival.  These methods
49 > *       do not block, but return an associated <em>arrival phase
50 > *       number</em>; that is, the phase number of the phaser to which
51 > *       the arrival applied. When the final party for a given phase
52 > *       arrives, an optional action is performed and the phase
53 > *       advances.  These actions are performed by the party
54 > *       triggering a phase advance, and are arranged by overriding
55 > *       method {@link #onAdvance(int, int)}, which also controls
56 > *       termination. Overriding this method is similar to, but more
57 > *       flexible than, providing a barrier action to a {@code
58 > *       CyclicBarrier}.
59 > *
60 > *   <li><b>Waiting.</b> Method {@link #awaitAdvance} requires an
61 > *       argument indicating an arrival phase number, and returns when
62 > *       the phaser advances to (or is already at) a different phase.
63 > *       Unlike similar constructions using {@code CyclicBarrier},
64 > *       method {@code awaitAdvance} continues to wait even if the
65 > *       waiting thread is interrupted. Interruptible and timeout
66 > *       versions are also available, but exceptions encountered while
67 > *       tasks wait interruptibly or with timeout do not change the
68 > *       state of the phaser. If necessary, you can perform any
69 > *       associated recovery within handlers of those exceptions,
70 > *       often after invoking {@code forceTermination}.  Phasers may
71 > *       also be used by tasks executing in a {@link ForkJoinPool},
72 > *       which will ensure sufficient parallelism to execute tasks
73 > *       when others are blocked waiting for a phase to advance.
74 > *
75   * </ul>
76   *
77 + * <p><b>Termination.</b> A phaser may enter a <em>termination</em>
78 + * state, that may be checked using method {@link #isTerminated}. Upon
79 + * termination, all synchronization methods immediately return without
80 + * waiting for advance, as indicated by a negative return value.
81 + * Similarly, attempts to register upon termination have no effect.
82 + * Termination is triggered when an invocation of {@code onAdvance}
83 + * returns {@code true}. The default implementation returns {@code
84 + * true} if a deregistration has caused the number of registered
85 + * parties to become zero.  As illustrated below, when phasers control
86 + * actions with a fixed number of iterations, it is often convenient
87 + * to override this method to cause termination when the current phase
88 + * number reaches a threshold. Method {@link #forceTermination} is
89 + * also available to abruptly release waiting threads and allow them
90 + * to terminate.
91   *
92 < * <li> Barrier actions, performed by the task triggering a phase
93 < * advance while others may be waiting, are arranged by overriding
94 < * method <tt>onAdvance</tt>, that also controls termination.
95 < *
96 < * <li> Phasers may enter a <em>termination</em> state in which all
97 < * await actions immediately return, indicating (via a negative phase
98 < * value) that execution is complete.  Termination is triggered by
56 < * executing the overridable <tt>onAdvance</tt> method that is invoked
57 < * each time the barrier is tripped. When a Phaser is controlling an
58 < * action with a fixed number of iterations, it is often convenient to
59 < * override this method to cause termination when the current phase
60 < * number reaches a threshold.  Method <tt>forceTermination</tt> is
61 < * also available to assist recovery actions upon failure.
62 < *
63 < * <li> Unlike most synchronizers, a Phaser may also be used with
64 < * ForkJoinTasks (as well as plain threads).
65 < *
66 < * <li> By default, <tt>awaitAdvance</tt> continues to wait even if
67 < * the current thread is interrupted. And unlike the case in
68 < * CyclicBarriers, exceptions encountered while tasks wait
69 < * interruptibly or with timeout do not change the state of the
70 < * barrier. If necessary, you can perform any associated recovery
71 < * within handlers of those exceptions.
92 > * <p><b>Tiering.</b> Phasers may be <em>tiered</em> (i.e.,
93 > * constructed in tree structures) to reduce contention. Phasers with
94 > * large numbers of parties that would otherwise experience heavy
95 > * synchronization contention costs may instead be set up so that
96 > * groups of sub-phasers share a common parent.  This may greatly
97 > * increase throughput even though it incurs greater per-operation
98 > * overhead.
99   *
100 < * </ul>
100 > * <p>In a tree of tiered phasers, registration and deregistration of
101 > * child phasers with their parent are managed automatically.
102 > * Whenever the number of registered parties of a child phaser becomes
103 > * non-zero (as established in the {@link #Phaser(Phaser,int)}
104 > * constructor, {@link #register}, or {@link #bulkRegister}), the
105 > * child phaser is registered with its parent.  Whenever the number of
106 > * registered parties becomes zero as the result of an invocation of
107 > * {@link #arriveAndDeregister}, the child phaser is deregistered
108 > * from its parent.
109 > *
110 > * <p><b>Monitoring.</b> While synchronization methods may be invoked
111 > * only by registered parties, the current state of a phaser may be
112 > * monitored by any caller.  At any given moment there are {@link
113 > * #getRegisteredParties} parties in total, of which {@link
114 > * #getArrivedParties} have arrived at the current phase ({@link
115 > * #getPhase}).  When the remaining ({@link #getUnarrivedParties})
116 > * parties arrive, the phase advances.  The values returned by these
117 > * methods may reflect transient states and so are not in general
118 > * useful for synchronization control.  Method {@link #toString}
119 > * returns snapshots of these state queries in a form convenient for
120 > * informal monitoring.
121   *
122 < * <p><b>Sample usage:</b>
122 > * <p><b>Sample usages:</b>
123   *
124 < * <p>[todo: non-FJ example]
124 > * <p>A {@code Phaser} may be used instead of a {@code CountDownLatch}
125 > * to control a one-shot action serving a variable number of parties.
126 > * The typical idiom is for the method setting this up to first
127 > * register, then start the actions, then deregister, as in:
128   *
129 < * <p> A Phaser may be used to support a style of programming in
130 < * which a task waits for others to complete, without otherwise
131 < * needing to keep track of which tasks it is waiting for. This is
132 < * similar to the "sync" construct in Cilk and "clocks" in X10.
133 < * Special constructions based on such barriers are available using
134 < * the <tt>LinkedAsyncAction</tt> and <tt>CyclicAction</tt> classes,
135 < * but they can be useful in other contexts as well.  For a simple
136 < * (but not very useful) example, here is a variant of Fibonacci:
137 < *
138 < * <pre>
139 < * class BarrierFibonacci extends RecursiveAction {
140 < *   int argument, result;
91 < *   final Phaser parentBarrier;
92 < *   BarrierFibonacci(int n, Phaser parentBarrier) {
93 < *     this.argument = n;
94 < *     this.parentBarrier = parentBarrier;
95 < *     parentBarrier.register();
129 > *  <pre> {@code
130 > * void runTasks(List<Runnable> tasks) {
131 > *   final Phaser phaser = new Phaser(1); // "1" to register self
132 > *   // create and start threads
133 > *   for (final Runnable task : tasks) {
134 > *     phaser.register();
135 > *     new Thread() {
136 > *       public void run() {
137 > *         phaser.arriveAndAwaitAdvance(); // await all creation
138 > *         task.run();
139 > *       }
140 > *     }.start();
141   *   }
142 < *   protected void compute() {
143 < *     int n = argument;
144 < *     if (n &lt;= 1)
145 < *        result = n;
146 < *     else {
147 < *        Phaser childBarrier = new Phaser(1);
148 < *        BarrierFibonacci f1 = new BarrierFibonacci(n - 1, childBarrier);
149 < *        BarrierFibonacci f2 = new BarrierFibonacci(n - 2, childBarrier);
150 < *        f1.fork();
151 < *        f2.fork();
152 < *        childBarrier.arriveAndAwait();
153 < *        result = f1.result + f2.result;
142 > *
143 > *   // allow threads to start and deregister self
144 > *   phaser.arriveAndDeregister();
145 > * }}</pre>
146 > *
147 > * <p>One way to cause a set of threads to repeatedly perform actions
148 > * for a given number of iterations is to override {@code onAdvance}:
149 > *
150 > *  <pre> {@code
151 > * void startTasks(List<Runnable> tasks, final int iterations) {
152 > *   final Phaser phaser = new Phaser() {
153 > *     protected boolean onAdvance(int phase, int registeredParties) {
154 > *       return phase >= iterations || registeredParties == 0;
155   *     }
156 < *     parentBarrier.arriveAndDeregister();
156 > *   };
157 > *   phaser.register();
158 > *   for (final Runnable task : tasks) {
159 > *     phaser.register();
160 > *     new Thread() {
161 > *       public void run() {
162 > *         do {
163 > *           task.run();
164 > *           phaser.arriveAndAwaitAdvance();
165 > *         } while (!phaser.isTerminated());
166 > *       }
167 > *     }.start();
168   *   }
169 < * }
170 < * </pre>
169 > *   phaser.arriveAndDeregister(); // deregister self, don't wait
170 > * }}</pre>
171 > *
172 > * If the main task must later await termination, it
173 > * may re-register and then execute a similar loop:
174 > *  <pre> {@code
175 > *   // ...
176 > *   phaser.register();
177 > *   while (!phaser.isTerminated())
178 > *     phaser.arriveAndAwaitAdvance();}</pre>
179 > *
180 > * <p>Related constructions may be used to await particular phase numbers
181 > * in contexts where you are sure that the phase will never wrap around
182 > * {@code Integer.MAX_VALUE}. For example:
183 > *
184 > *  <pre> {@code
185 > * void awaitPhase(Phaser phaser, int phase) {
186 > *   int p = phaser.register(); // assumes caller not already registered
187 > *   while (p < phase) {
188 > *     if (phaser.isTerminated())
189 > *       // ... deal with unexpected termination
190 > *     else
191 > *       p = phaser.arriveAndAwaitAdvance();
192 > *   }
193 > *   phaser.arriveAndDeregister();
194 > * }}</pre>
195 > *
196 > * <p>To create a set of {@code n} tasks using a tree of phasers, you
197 > * could use code of the following form, assuming a Task class with a
198 > * constructor accepting a {@code Phaser} that it registers with upon
199 > * construction. After invocation of {@code build(new Task[n], 0, n,
200 > * new Phaser())}, these tasks could then be started, for example by
201 > * submitting to a pool:
202 > *
203 > *  <pre> {@code
204 > * void build(Task[] tasks, int lo, int hi, Phaser ph) {
205 > *   if (hi - lo > TASKS_PER_PHASER) {
206 > *     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
207 > *       int j = Math.min(i + TASKS_PER_PHASER, hi);
208 > *       build(tasks, i, j, new Phaser(ph));
209 > *     }
210 > *   } else {
211 > *     for (int i = lo; i < hi; ++i)
212 > *       tasks[i] = new Task(ph);
213 > *       // assumes new Task(ph) performs ph.register()
214 > *   }
215 > * }}</pre>
216 > *
217 > * The best value of {@code TASKS_PER_PHASER} depends mainly on
218 > * expected synchronization rates. A value as low as four may
219 > * be appropriate for extremely small per-phase task bodies (thus
220 > * high rates), or up to hundreds for extremely large ones.
221   *
222   * <p><b>Implementation notes</b>: This implementation restricts the
223 < * maximum number of parties to 65535. Attempts to register
224 < * additional parties result in IllegalStateExceptions.
223 > * maximum number of parties to 65535. Attempts to register additional
224 > * parties result in {@code IllegalStateException}. However, you can and
225 > * should create tiered phasers to accommodate arbitrarily large sets
226 > * of participants.
227 > *
228 > * @since 1.7
229 > * @author Doug Lea
230   */
231   public class Phaser {
232      /*
233       * This class implements an extension of X10 "clocks".  Thanks to
234 <     * Vijay Saraswat for the idea of applying it to ForkJoinTasks,
235 <     * and to Vivek Sarkar for enhancements to extend functionality.
234 >     * Vijay Saraswat for the idea, and to Vivek Sarkar for
235 >     * enhancements to extend functionality.
236       */
237  
238      /**
239 <     * Barrier state representation. Conceptually, a barrier contains
128 <     * four values:
239 >     * Primary state representation, holding four bit-fields:
240       *
241 <     * * parties -- the number of parties to wait (16 bits)
242 <     * * unarrived -- the number of parties yet to hit barrier (16 bits)
243 <     * * phase -- the generation of the barrier (31 bits)
244 <     * * terminated -- set if barrier is terminated (1 bit)
241 >     * unarrived  -- the number of parties yet to hit barrier (bits  0-15)
242 >     * parties    -- the number of parties to wait            (bits 16-31)
243 >     * phase      -- the generation of the barrier            (bits 32-62)
244 >     * terminated -- set if barrier is terminated             (bit  63 / sign)
245       *
246 <     * However, to efficiently maintain atomicity, these values are
247 <     * packed into a single AtomicLong. Termination uses the sign bit
248 <     * of 32 bit representation of phase, so phase is set to -1 on
249 <     * termination.
250 <     */
251 <    private final AtomicLong state;
252 <
253 <    /**
254 <     * Head of Treiber stack for waiting nonFJ threads.
255 <     */
256 <    private final AtomicReference<QNode> head = new AtomicReference<QNode>();
246 >     * Except that a phaser with no registered parties is
247 >     * distinguished by the otherwise illegal state of having zero
248 >     * parties and one unarrived parties (encoded as EMPTY below).
249 >     *
250 >     * To efficiently maintain atomicity, these values are packed into
251 >     * a single (atomic) long. Good performance relies on keeping
252 >     * state decoding and encoding simple, and keeping race windows
253 >     * short.
254 >     *
255 >     * All state updates are performed via CAS except initial
256 >     * registration of a sub-phaser (i.e., one with a non-null
257 >     * parent).  In this (relatively rare) case, we use built-in
258 >     * synchronization to lock while first registering with its
259 >     * parent.
260 >     *
261 >     * The phase of a subphaser is allowed to lag that of its
262 >     * ancestors until it is actually accessed -- see method
263 >     * reconcileState.
264 >     */
265 >    private volatile long state;
266 >
267 >    private static final int  MAX_PARTIES     = 0xffff;
268 >    private static final int  MAX_PHASE       = Integer.MAX_VALUE;
269 >    private static final int  PARTIES_SHIFT   = 16;
270 >    private static final int  PHASE_SHIFT     = 32;
271 >    private static final int  UNARRIVED_MASK  = 0xffff;      // to mask ints
272 >    private static final long PARTIES_MASK    = 0xffff0000L; // to mask longs
273 >    private static final long COUNTS_MASK     = 0xffffffffL;
274 >    private static final long TERMINATION_BIT = 1L << 63;
275 >
276 >    // some special values
277 >    private static final int  ONE_ARRIVAL     = 1;
278 >    private static final int  ONE_PARTY       = 1 << PARTIES_SHIFT;
279 >    private static final int  ONE_DEREGISTER  = ONE_ARRIVAL|ONE_PARTY;
280 >    private static final int  EMPTY           = 1;
281  
282 <    private static final int ushortBits = 16;
148 <    private static final int ushortMask =  (1 << ushortBits) - 1;
149 <    private static final int phaseMask = 0x7fffffff;
282 >    // The following unpacking methods are usually manually inlined
283  
284      private static int unarrivedOf(long s) {
285 <        return (int)(s & ushortMask);
285 >        int counts = (int)s;
286 >        return (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
287      }
288  
289      private static int partiesOf(long s) {
290 <        return (int)(s & (ushortMask << 16)) >>> 16;
290 >        return (int)s >>> PARTIES_SHIFT;
291      }
292  
293      private static int phaseOf(long s) {
294 <        return (int)(s >>> 32);
294 >        return (int)(s >>> PHASE_SHIFT);
295      }
296  
297      private static int arrivedOf(long s) {
298 <        return partiesOf(s) - unarrivedOf(s);
298 >        int counts = (int)s;
299 >        return (counts == EMPTY) ? 0 :
300 >            (counts >>> PARTIES_SHIFT) - (counts & UNARRIVED_MASK);
301      }
302  
303 <    private static long stateFor(int phase, int parties, int unarrived) {
304 <        return (((long)phase) << 32) | ((parties << 16) | unarrived);
305 <    }
303 >    /**
304 >     * The parent of this phaser, or null if none
305 >     */
306 >    private final Phaser parent;
307  
308 <    private static IllegalStateException badBounds(int parties, int unarrived) {
309 <        return new IllegalStateException("Attempt to set " + unarrived +
310 <                                         " unarrived of " + parties + " parties");
308 >    /**
309 >     * The root of phaser tree. Equals this if not in a tree.
310 >     */
311 >    private final Phaser root;
312 >
313 >    /**
314 >     * Heads of Treiber stacks for waiting threads. To eliminate
315 >     * contention when releasing some threads while adding others, we
316 >     * use two of them, alternating across even and odd phases.
317 >     * Subphasers share queues with root to speed up releases.
318 >     */
319 >    private final AtomicReference<QNode> evenQ;
320 >    private final AtomicReference<QNode> oddQ;
321 >
322 >    private AtomicReference<QNode> queueFor(int phase) {
323 >        return ((phase & 1) == 0) ? evenQ : oddQ;
324      }
325  
326      /**
327 <     * Creates a new Phaser without any initially registered parties,
178 <     * and initial phase number 0.
327 >     * Returns message string for bounds exceptions on arrival.
328       */
329 <    public Phaser() {
330 <        state = new AtomicLong(stateFor(0, 0, 0));
329 >    private String badArrive(long s) {
330 >        return "Attempted arrival of unregistered party for " +
331 >            stateToString(s);
332      }
333  
334      /**
335 <     * Creates a new Phaser with the given numbers of registered
186 <     * unarrived parties and initial phase number 0.
187 <     * @param parties the number of parties required to trip barrier.
188 <     * @throws IllegalArgumentException if parties less than zero
189 <     * or greater than the maximum number of parties supported.
335 >     * Returns message string for bounds exceptions on registration.
336       */
337 <    public Phaser(int parties) {
338 <        if (parties < 0 || parties > ushortMask)
339 <            throw new IllegalArgumentException("Illegal number of parties");
194 <        state = new AtomicLong(stateFor(0, parties, parties));
337 >    private String badRegister(long s) {
338 >        return "Attempt to register more than " +
339 >            MAX_PARTIES + " parties for " + stateToString(s);
340      }
341  
342      /**
343 <     * Adds a new unarrived party to this phaser.
344 <     * @return the current barrier phase number upon registration
345 <     * @throws IllegalStateException if attempting to register more
346 <     * than the maximum supported number of parties.
343 >     * Main implementation for methods arrive and arriveAndDeregister.
344 >     * Manually tuned to speed up and minimize race windows for the
345 >     * common case of just decrementing unarrived field.
346 >     *
347 >     * @param adjust value to subtract from state;
348 >     *               ONE_ARRIVAL for arrive,
349 >     *               ONE_DEREGISTER for arriveAndDeregister
350       */
351 <    public int register() { // increment both parties and unarrived
352 <        final AtomicLong state = this.state;
351 >    private int doArrive(int adjust) {
352 >        final Phaser root = this.root;
353          for (;;) {
354 <            long s = state.get();
355 <            int phase = phaseOf(s);
356 <            int parties = partiesOf(s) + 1;
357 <            int unarrived = unarrivedOf(s) + 1;
358 <            if (parties > ushortMask || unarrived > ushortMask)
359 <                throw badBounds(parties, unarrived);
360 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
354 >            long s = (root == this) ? state : reconcileState();
355 >            int phase = (int)(s >>> PHASE_SHIFT);
356 >            if (phase < 0)
357 >                return phase;
358 >            int counts = (int)s;
359 >            int unarrived = (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
360 >            if (unarrived <= 0)
361 >                throw new IllegalStateException(badArrive(s));
362 >            if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adjust)) {
363 >                if (unarrived == 1) {
364 >                    long n = s & PARTIES_MASK;  // base of next state
365 >                    int nextUnarrived = (int)n >>> PARTIES_SHIFT;
366 >                    if (root == this) {
367 >                        if (onAdvance(phase, nextUnarrived))
368 >                            n |= TERMINATION_BIT;
369 >                        else if (nextUnarrived == 0)
370 >                            n |= EMPTY;
371 >                        else
372 >                            n |= nextUnarrived;
373 >                        int nextPhase = (phase + 1) & MAX_PHASE;
374 >                        n |= (long)nextPhase << PHASE_SHIFT;
375 >                        UNSAFE.compareAndSwapLong(this, stateOffset, s, n);
376 >                        releaseWaiters(phase);
377 >                    }
378 >                    else if (nextUnarrived == 0) { // propagate deregistration
379 >                        phase = parent.doArrive(ONE_DEREGISTER);
380 >                        UNSAFE.compareAndSwapLong(this, stateOffset,
381 >                                                  s, s | EMPTY);
382 >                    }
383 >                    else
384 >                        phase = parent.doArrive(ONE_ARRIVAL);
385 >                }
386                  return phase;
387 +            }
388          }
389      }
390  
391      /**
392 <     * Arrives at the barrier, but does not wait for others.  (You can
219 <     * in turn wait for others via {@link #awaitAdvance}).
392 >     * Implementation of register, bulkRegister
393       *
394 <     * @return the current barrier phase number upon entry to
395 <     * this method, or a negative value if terminated;
223 <     * @throws IllegalStateException if the number of unarrived
224 <     * parties would become negative.
394 >     * @param registrations number to add to both parties and
395 >     * unarrived fields. Must be greater than zero.
396       */
397 <    public int arrive() { // decrement unarrived. If zero, trip
398 <        final AtomicLong state = this.state;
397 >    private int doRegister(int registrations) {
398 >        // adjustment to state
399 >        long adjust = ((long)registrations << PARTIES_SHIFT) | registrations;
400 >        final Phaser parent = this.parent;
401 >        int phase;
402          for (;;) {
403 <            long s = state.get();
404 <            int phase = phaseOf(s);
405 <            int parties = partiesOf(s);
406 <            int unarrived = unarrivedOf(s) - 1;
407 <            if (unarrived < 0)
408 <                throw badBounds(parties, unarrived);
409 <            if (unarrived == 0 && phase >= 0) {
410 <                trip(phase, parties);
411 <                return phase;
403 >            long s = (parent == null) ? state : reconcileState();
404 >            int counts = (int)s;
405 >            int parties = counts >>> PARTIES_SHIFT;
406 >            int unarrived = counts & UNARRIVED_MASK;
407 >            if (registrations > MAX_PARTIES - parties)
408 >                throw new IllegalStateException(badRegister(s));
409 >            phase = (int)(s >>> PHASE_SHIFT);
410 >            if (phase < 0)
411 >                break;
412 >            if (counts != EMPTY) {                  // not 1st registration
413 >                if (parent == null || reconcileState() == s) {
414 >                    if (unarrived == 0)             // wait out advance
415 >                        root.internalAwaitAdvance(phase, null);
416 >                    else if (UNSAFE.compareAndSwapLong(this, stateOffset,
417 >                                                       s, s + adjust))
418 >                        break;
419 >                }
420 >            }
421 >            else if (parent == null) {              // 1st root registration
422 >                long next = ((long)phase << PHASE_SHIFT) | adjust;
423 >                if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next))
424 >                    break;
425 >            }
426 >            else {
427 >                synchronized (this) {               // 1st sub registration
428 >                    if (state == s) {               // recheck under lock
429 >                        phase = parent.doRegister(1);
430 >                        if (phase < 0)
431 >                            break;
432 >                        // finish registration whenever parent registration
433 >                        // succeeded, even when racing with termination,
434 >                        // since these are part of the same "transaction".
435 >                        while (!UNSAFE.compareAndSwapLong
436 >                               (this, stateOffset, s,
437 >                                ((long)phase << PHASE_SHIFT) | adjust)) {
438 >                            s = state;
439 >                            phase = (int)(root.state >>> PHASE_SHIFT);
440 >                            // assert (int)s == EMPTY;
441 >                        }
442 >                        break;
443 >                    }
444 >                }
445              }
239            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
240                return phase;
446          }
447 +        return phase;
448      }
449  
450      /**
451 <     * Arrives at the barrier, and deregisters from it, without
452 <     * waiting for others.
451 >     * Resolves lagged phase propagation from root if necessary.
452 >     * Reconciliation normally occurs when root has advanced but
453 >     * subphasers have not yet done so, in which case they must finish
454 >     * their own advance by setting unarrived to parties (or if
455 >     * parties is zero, resetting to unregistered EMPTY state).
456       *
457 <     * @return the current barrier phase number upon entry to
249 <     * this method, or a negative value if terminated;
250 <     * @throws IllegalStateException if the number of registered or
251 <     * unarrived parties would become negative.
457 >     * @return reconciled state
458       */
459 <    public int arriveAndDeregister() { // Same as arrive, plus decrement parties
460 <        final AtomicLong state = this.state;
461 <        for (;;) {
462 <            long s = state.get();
463 <            int phase = phaseOf(s);
464 <            int parties = partiesOf(s) - 1;
465 <            int unarrived = unarrivedOf(s) - 1;
466 <            if (parties < 0 || unarrived < 0)
467 <                throw badBounds(parties, unarrived);
468 <            if (unarrived == 0 && phase >= 0) {
469 <                trip(phase, parties);
470 <                return phase;
471 <            }
472 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
473 <                return phase;
459 >    private long reconcileState() {
460 >        final Phaser root = this.root;
461 >        long s = state;
462 >        if (root != this) {
463 >            int phase, p;
464 >            // CAS to root phase with current parties, tripping unarrived
465 >            while ((phase = (int)(root.state >>> PHASE_SHIFT)) !=
466 >                   (int)(s >>> PHASE_SHIFT) &&
467 >                   !UNSAFE.compareAndSwapLong
468 >                   (this, stateOffset, s,
469 >                    s = (((long)phase << PHASE_SHIFT) |
470 >                         ((phase < 0) ? (s & COUNTS_MASK) :
471 >                          (((p = (int)s >>> PARTIES_SHIFT) == 0) ? EMPTY :
472 >                           ((s & PARTIES_MASK) | p))))))
473 >                s = state;
474 >        }
475 >        return s;
476 >    }
477 >
478 >    /**
479 >     * Creates a new phaser with no initially registered parties, no
480 >     * parent, and initial phase number 0. Any thread using this
481 >     * phaser will need to first register for it.
482 >     */
483 >    public Phaser() {
484 >        this(null, 0);
485 >    }
486 >
487 >    /**
488 >     * Creates a new phaser with the given number of registered
489 >     * unarrived parties, no parent, and initial phase number 0.
490 >     *
491 >     * @param parties the number of parties required to advance to the
492 >     * next phase
493 >     * @throws IllegalArgumentException if parties less than zero
494 >     * or greater than the maximum number of parties supported
495 >     */
496 >    public Phaser(int parties) {
497 >        this(null, parties);
498 >    }
499 >
500 >    /**
501 >     * Equivalent to {@link #Phaser(Phaser, int) Phaser(parent, 0)}.
502 >     *
503 >     * @param parent the parent phaser
504 >     */
505 >    public Phaser(Phaser parent) {
506 >        this(parent, 0);
507 >    }
508 >
509 >    /**
510 >     * Creates a new phaser with the given parent and number of
511 >     * registered unarrived parties.  When the given parent is non-null
512 >     * and the given number of parties is greater than zero, this
513 >     * child phaser is registered with its parent.
514 >     *
515 >     * @param parent the parent phaser
516 >     * @param parties the number of parties required to advance to the
517 >     * next phase
518 >     * @throws IllegalArgumentException if parties less than zero
519 >     * or greater than the maximum number of parties supported
520 >     */
521 >    public Phaser(Phaser parent, int parties) {
522 >        if (parties >>> PARTIES_SHIFT != 0)
523 >            throw new IllegalArgumentException("Illegal number of parties");
524 >        int phase = 0;
525 >        this.parent = parent;
526 >        if (parent != null) {
527 >            final Phaser root = parent.root;
528 >            this.root = root;
529 >            this.evenQ = root.evenQ;
530 >            this.oddQ = root.oddQ;
531 >            if (parties != 0)
532 >                phase = parent.doRegister(1);
533          }
534 +        else {
535 +            this.root = this;
536 +            this.evenQ = new AtomicReference<QNode>();
537 +            this.oddQ = new AtomicReference<QNode>();
538 +        }
539 +        this.state = (parties == 0) ? (long)EMPTY :
540 +            ((long)phase << PHASE_SHIFT) |
541 +            ((long)parties << PARTIES_SHIFT) |
542 +            ((long)parties);
543 +    }
544 +
545 +    /**
546 +     * Adds a new unarrived party to this phaser.  If an ongoing
547 +     * invocation of {@link #onAdvance} is in progress, this method
548 +     * may await its completion before returning.  If this phaser has
549 +     * a parent, and this phaser previously had no registered parties,
550 +     * this child phaser is also registered with its parent. If
551 +     * this phaser is terminated, the attempt to register has
552 +     * no effect, and a negative value is returned.
553 +     *
554 +     * @return the arrival phase number to which this registration
555 +     * applied.  If this value is negative, then this phaser has
556 +     * terminated, in which case registration has no effect.
557 +     * @throws IllegalStateException if attempting to register more
558 +     * than the maximum supported number of parties
559 +     */
560 +    public int register() {
561 +        return doRegister(1);
562 +    }
563 +
564 +    /**
565 +     * Adds the given number of new unarrived parties to this phaser.
566 +     * If an ongoing invocation of {@link #onAdvance} is in progress,
567 +     * this method may await its completion before returning.  If this
568 +     * phaser has a parent, and the given number of parties is greater
569 +     * than zero, and this phaser previously had no registered
570 +     * parties, this child phaser is also registered with its parent.
571 +     * If this phaser is terminated, the attempt to register has no
572 +     * effect, and a negative value is returned.
573 +     *
574 +     * @param parties the number of additional parties required to
575 +     * advance to the next phase
576 +     * @return the arrival phase number to which this registration
577 +     * applied.  If this value is negative, then this phaser has
578 +     * terminated, in which case registration has no effect.
579 +     * @throws IllegalStateException if attempting to register more
580 +     * than the maximum supported number of parties
581 +     * @throws IllegalArgumentException if {@code parties < 0}
582 +     */
583 +    public int bulkRegister(int parties) {
584 +        if (parties < 0)
585 +            throw new IllegalArgumentException();
586 +        if (parties == 0)
587 +            return getPhase();
588 +        return doRegister(parties);
589 +    }
590 +
591 +    /**
592 +     * Arrives at this phaser, without waiting for others to arrive.
593 +     *
594 +     * <p>It is a usage error for an unregistered party to invoke this
595 +     * method.  However, this error may result in an {@code
596 +     * IllegalStateException} only upon some subsequent operation on
597 +     * this phaser, if ever.
598 +     *
599 +     * @return the arrival phase number, or a negative value if terminated
600 +     * @throws IllegalStateException if not terminated and the number
601 +     * of unarrived parties would become negative
602 +     */
603 +    public int arrive() {
604 +        return doArrive(ONE_ARRIVAL);
605 +    }
606 +
607 +    /**
608 +     * Arrives at this phaser and deregisters from it without waiting
609 +     * for others to arrive. Deregistration reduces the number of
610 +     * parties required to advance in future phases.  If this phaser
611 +     * has a parent, and deregistration causes this phaser to have
612 +     * zero parties, this phaser is also deregistered from its parent.
613 +     *
614 +     * <p>It is a usage error for an unregistered party to invoke this
615 +     * method.  However, this error may result in an {@code
616 +     * IllegalStateException} only upon some subsequent operation on
617 +     * this phaser, if ever.
618 +     *
619 +     * @return the arrival phase number, or a negative value if terminated
620 +     * @throws IllegalStateException if not terminated and the number
621 +     * of registered or unarrived parties would become negative
622 +     */
623 +    public int arriveAndDeregister() {
624 +        return doArrive(ONE_DEREGISTER);
625      }
626  
627      /**
628 <     * Arrives at the barrier and awaits others. Unlike other arrival
629 <     * methods, this method returns the arrival index of the
630 <     * caller. The caller tripping the barrier returns zero, the
631 <     * previous caller 1, and so on.
632 <     * @return the arrival index
633 <     * @throws IllegalStateException if the number of unarrived
634 <     * parties would become negative.
628 >     * Arrives at this phaser and awaits others. Equivalent in effect
629 >     * to {@code awaitAdvance(arrive())}.  If you need to await with
630 >     * interruption or timeout, you can arrange this with an analogous
631 >     * construction using one of the other forms of the {@code
632 >     * awaitAdvance} method.  If instead you need to deregister upon
633 >     * arrival, use {@code awaitAdvance(arriveAndDeregister())}.
634 >     *
635 >     * <p>It is a usage error for an unregistered party to invoke this
636 >     * method.  However, this error may result in an {@code
637 >     * IllegalStateException} only upon some subsequent operation on
638 >     * this phaser, if ever.
639 >     *
640 >     * @return the arrival phase number, or the (negative)
641 >     * {@linkplain #getPhase() current phase} if terminated
642 >     * @throws IllegalStateException if not terminated and the number
643 >     * of unarrived parties would become negative
644       */
645      public int arriveAndAwaitAdvance() {
646 <        final AtomicLong state = this.state;
646 >        // Specialization of doArrive+awaitAdvance eliminating some reads/paths
647 >        final Phaser root = this.root;
648          for (;;) {
649 <            long s = state.get();
650 <            int phase = phaseOf(s);
651 <            int parties = partiesOf(s);
652 <            int unarrived = unarrivedOf(s) - 1;
653 <            if (unarrived < 0)
654 <                throw badBounds(parties, unarrived);
655 <            if (unarrived == 0 && phase >= 0) {
656 <                trip(phase, parties);
657 <                return 0;
658 <            }
659 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived))) {
660 <                awaitAdvance(phase);
661 <                return unarrived;
649 >            long s = (root == this) ? state : reconcileState();
650 >            int phase = (int)(s >>> PHASE_SHIFT);
651 >            if (phase < 0)
652 >                return phase;
653 >            int counts = (int)s;
654 >            int unarrived = (counts == EMPTY) ? 0 : (counts & UNARRIVED_MASK);
655 >            if (unarrived <= 0)
656 >                throw new IllegalStateException(badArrive(s));
657 >            if (UNSAFE.compareAndSwapLong(this, stateOffset, s,
658 >                                          s -= ONE_ARRIVAL)) {
659 >                if (unarrived > 1)
660 >                    return root.internalAwaitAdvance(phase, null);
661 >                if (root != this)
662 >                    return parent.arriveAndAwaitAdvance();
663 >                long n = s & PARTIES_MASK;  // base of next state
664 >                int nextUnarrived = (int)n >>> PARTIES_SHIFT;
665 >                if (onAdvance(phase, nextUnarrived))
666 >                    n |= TERMINATION_BIT;
667 >                else if (nextUnarrived == 0)
668 >                    n |= EMPTY;
669 >                else
670 >                    n |= nextUnarrived;
671 >                int nextPhase = (phase + 1) & MAX_PHASE;
672 >                n |= (long)nextPhase << PHASE_SHIFT;
673 >                if (!UNSAFE.compareAndSwapLong(this, stateOffset, s, n))
674 >                    return (int)(state >>> PHASE_SHIFT); // terminated
675 >                releaseWaiters(phase);
676 >                return nextPhase;
677              }
678          }
679      }
680  
681      /**
682 <     * Awaits the phase of the barrier to advance from the given
683 <     * value, or returns immediately if this barrier is terminated.
684 <     * @param phase the phase on entry to this method
685 <     * @return the phase on exit from this method
682 >     * Awaits the phase of this phaser to advance from the given phase
683 >     * value, returning immediately if the current phase is not equal
684 >     * to the given phase value or this phaser is terminated.
685 >     *
686 >     * @param phase an arrival phase number, or negative value if
687 >     * terminated; this argument is normally the value returned by a
688 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
689 >     * @return the next arrival phase number, or the argument if it is
690 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
691 >     * if terminated
692       */
693      public int awaitAdvance(int phase) {
694 +        final Phaser root = this.root;
695 +        long s = (root == this) ? state : reconcileState();
696 +        int p = (int)(s >>> PHASE_SHIFT);
697          if (phase < 0)
698              return phase;
699 <        Thread current = Thread.currentThread();
700 <        if (current instanceof ForkJoinWorkerThread)
701 <            return helpingWait(phase);
312 <        if (untimedWait(current, phase, false))
313 <            current.interrupt();
314 <        return phaseOf(state.get());
699 >        if (p == phase)
700 >            return root.internalAwaitAdvance(phase, null);
701 >        return p;
702      }
703  
704      /**
705 <     * Awaits the phase of the barrier to advance from the given
706 <     * value, or returns immediately if this barrier is terminated, or
707 <     * throws InterruptedException if interrupted while waiting.
708 <     * @param phase the phase on entry to this method
709 <     * @return the phase on exit from this method
705 >     * Awaits the phase of this phaser to advance from the given phase
706 >     * value, throwing {@code InterruptedException} if interrupted
707 >     * while waiting, or returning immediately if the current phase is
708 >     * not equal to the given phase value or this phaser is
709 >     * terminated.
710 >     *
711 >     * @param phase an arrival phase number, or negative value if
712 >     * terminated; this argument is normally the value returned by a
713 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
714 >     * @return the next arrival phase number, or the argument if it is
715 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
716 >     * if terminated
717       * @throws InterruptedException if thread interrupted while waiting
718       */
719 <    public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
719 >    public int awaitAdvanceInterruptibly(int phase)
720 >        throws InterruptedException {
721 >        final Phaser root = this.root;
722 >        long s = (root == this) ? state : reconcileState();
723 >        int p = (int)(s >>> PHASE_SHIFT);
724          if (phase < 0)
725              return phase;
726 <        Thread current = Thread.currentThread();
727 <        if (current instanceof ForkJoinWorkerThread)
728 <            return helpingWait(phase);
729 <        else if (Thread.interrupted() || untimedWait(current, phase, true))
730 <            throw new InterruptedException();
731 <        else
732 <            return phaseOf(state.get());
726 >        if (p == phase) {
727 >            QNode node = new QNode(this, phase, true, false, 0L);
728 >            p = root.internalAwaitAdvance(phase, node);
729 >            if (node.wasInterrupted)
730 >                throw new InterruptedException();
731 >        }
732 >        return p;
733      }
734  
735      /**
736 <     * Awaits the phase of the barrier to advance from the given value
737 <     * or the given timeout elapses, or returns immediately if this
738 <     * barrier is terminated.
739 <     * @param phase the phase on entry to this method
740 <     * @return the phase on exit from this method
736 >     * Awaits the phase of this phaser to advance from the given phase
737 >     * value or the given timeout to elapse, throwing {@code
738 >     * InterruptedException} if interrupted while waiting, or
739 >     * returning immediately if the current phase is not equal to the
740 >     * given phase value or this phaser is terminated.
741 >     *
742 >     * @param phase an arrival phase number, or negative value if
743 >     * terminated; this argument is normally the value returned by a
744 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
745 >     * @param timeout how long to wait before giving up, in units of
746 >     *        {@code unit}
747 >     * @param unit a {@code TimeUnit} determining how to interpret the
748 >     *        {@code timeout} parameter
749 >     * @return the next arrival phase number, or the argument if it is
750 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
751 >     * if terminated
752       * @throws InterruptedException if thread interrupted while waiting
753       * @throws TimeoutException if timed out while waiting
754       */
755 <    public int awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)
755 >    public int awaitAdvanceInterruptibly(int phase,
756 >                                         long timeout, TimeUnit unit)
757          throws InterruptedException, TimeoutException {
758 +        long nanos = unit.toNanos(timeout);
759 +        final Phaser root = this.root;
760 +        long s = (root == this) ? state : reconcileState();
761 +        int p = (int)(s >>> PHASE_SHIFT);
762          if (phase < 0)
763              return phase;
764 <        long nanos = unit.toNanos(timeout);
765 <        Thread current = Thread.currentThread();
766 <        if (current instanceof ForkJoinWorkerThread)
767 <            return timedHelpingWait(phase, nanos);
768 <        timedWait(current, phase, nanos);
769 <        return phaseOf(state.get());
764 >        if (p == phase) {
765 >            QNode node = new QNode(this, phase, true, true, nanos);
766 >            p = root.internalAwaitAdvance(phase, node);
767 >            if (node.wasInterrupted)
768 >                throw new InterruptedException();
769 >            else if (p == phase)
770 >                throw new TimeoutException();
771 >        }
772 >        return p;
773      }
774  
775      /**
776 <     * Forces this barrier to enter termination state. Counts of
777 <     * arrived and registered parties are unaffected. This method may
778 <     * be useful for coordinating recovery after one or more tasks
779 <     * encounter unexpected exceptions.
776 >     * Forces this phaser to enter termination state.  Counts of
777 >     * registered parties are unaffected.  If this phaser is a member
778 >     * of a tiered set of phasers, then all of the phasers in the set
779 >     * are terminated.  If this phaser is already terminated, this
780 >     * method has no effect.  This method may be useful for
781 >     * coordinating recovery after one or more tasks encounter
782 >     * unexpected exceptions.
783       */
784      public void forceTermination() {
785 <        final AtomicLong state = this.state;
786 <        for (;;) {
787 <            long s = state.get();
788 <            int phase = phaseOf(s);
789 <            int parties = partiesOf(s);
790 <            int unarrived = unarrivedOf(s);
791 <            if (phase < 0 ||
792 <                state.compareAndSet(s, stateFor(-1, parties, unarrived))) {
793 <                if (head.get() != null)
374 <                    releaseWaiters(-1);
785 >        // Only need to change root state
786 >        final Phaser root = this.root;
787 >        long s;
788 >        while ((s = root.state) >= 0) {
789 >            if (UNSAFE.compareAndSwapLong(root, stateOffset,
790 >                                          s, s | TERMINATION_BIT)) {
791 >                // signal all threads
792 >                releaseWaiters(0); // Waiters on evenQ
793 >                releaseWaiters(1); // Waiters on oddQ
794                  return;
795              }
796          }
797      }
798  
799      /**
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    /**
800       * Returns the current phase number. The maximum phase number is
801 <     * <tt>Integer.MAX_VALUE</tt>, after which it restarts at
802 <     * zero. Upon termination, the phase number is negative.
801 >     * {@code Integer.MAX_VALUE}, after which it restarts at
802 >     * zero. Upon termination, the phase number is negative,
803 >     * in which case the prevailing phase prior to termination
804 >     * may be obtained via {@code getPhase() + Integer.MIN_VALUE}.
805 >     *
806       * @return the phase number, or a negative value if terminated
807       */
808 <    public int getPhase() {
809 <        return phaseOf(state.get());
808 >    public final int getPhase() {
809 >        return (int)(root.state >>> PHASE_SHIFT);
810      }
811  
812      /**
813 <     * Returns the number of parties registered at this barrier.
813 >     * Returns the number of parties registered at this phaser.
814 >     *
815       * @return the number of parties
816       */
817      public int getRegisteredParties() {
818 <        return partiesOf(state.get());
818 >        return partiesOf(state);
819      }
820  
821      /**
822 <     * Returns the number of parties that have arrived at the current
823 <     * phase of this barrier.
822 >     * Returns the number of registered parties that have arrived at
823 >     * the current phase of this phaser. If this phaser has terminated,
824 >     * the returned value is meaningless and arbitrary.
825 >     *
826       * @return the number of arrived parties
827       */
828      public int getArrivedParties() {
829 <        return arrivedOf(state.get());
829 >        return arrivedOf(reconcileState());
830      }
831  
832      /**
833       * Returns the number of registered parties that have not yet
834 <     * arrived at the current phase of this barrier.
834 >     * arrived at the current phase of this phaser. If this phaser has
835 >     * terminated, the returned value is meaningless and arbitrary.
836 >     *
837       * @return the number of unarrived parties
838       */
839      public int getUnarrivedParties() {
840 <        return unarrivedOf(state.get());
840 >        return unarrivedOf(reconcileState());
841 >    }
842 >
843 >    /**
844 >     * Returns the parent of this phaser, or {@code null} if none.
845 >     *
846 >     * @return the parent of this phaser, or {@code null} if none
847 >     */
848 >    public Phaser getParent() {
849 >        return parent;
850      }
851  
852      /**
853 <     * Returns true if this barrier has been terminated.
854 <     * @return true if this barrier has been terminated
853 >     * Returns the root ancestor of this phaser, which is the same as
854 >     * this phaser if it has no parent.
855 >     *
856 >     * @return the root ancestor of this phaser
857 >     */
858 >    public Phaser getRoot() {
859 >        return root;
860 >    }
861 >
862 >    /**
863 >     * Returns {@code true} if this phaser has been terminated.
864 >     *
865 >     * @return {@code true} if this phaser has been terminated
866       */
867      public boolean isTerminated() {
868 <        return phaseOf(state.get()) < 0;
868 >        return root.state < 0L;
869      }
870  
871      /**
872 <     * Overridable method to perform an action upon phase advance, and
873 <     * to control termination. This method is invoked whenever the
874 <     * barrier is tripped (and thus all other waiting parties are
875 <     * dormant). If it returns true, then, rather than advance the
876 <     * phase number, this barrier will be set to a final termination
877 <     * state, and subsequent calls to <tt>isTerminated</tt> will
878 <     * return true.
879 <     *
880 <     * <p> The default version returns true when the number of
881 <     * registered parties is zero. Normally, overrides that arrange
882 <     * termination for other reasons should also preserve this
883 <     * property.
884 <     *
885 <     * @param phase the phase number on entering the barrier
886 <     * @param registeredParties the current number of registered
887 <     * parties.
888 <     * @return true if this barrier should terminate
872 >     * Overridable method to perform an action upon impending phase
873 >     * advance, and to control termination. This method is invoked
874 >     * upon arrival of the party advancing this phaser (when all other
875 >     * waiting parties are dormant).  If this method returns {@code
876 >     * true}, this phaser will be set to a final termination state
877 >     * upon advance, and subsequent calls to {@link #isTerminated}
878 >     * will return true. Any (unchecked) Exception or Error thrown by
879 >     * an invocation of this method is propagated to the party
880 >     * attempting to advance this phaser, in which case no advance
881 >     * occurs.
882 >     *
883 >     * <p>The arguments to this method provide the state of the phaser
884 >     * prevailing for the current transition.  The effects of invoking
885 >     * arrival, registration, and waiting methods on this phaser from
886 >     * within {@code onAdvance} are unspecified and should not be
887 >     * relied on.
888 >     *
889 >     * <p>If this phaser is a member of a tiered set of phasers, then
890 >     * {@code onAdvance} is invoked only for its root phaser on each
891 >     * advance.
892 >     *
893 >     * <p>To support the most common use cases, the default
894 >     * implementation of this method returns {@code true} when the
895 >     * number of registered parties has become zero as the result of a
896 >     * party invoking {@code arriveAndDeregister}.  You can disable
897 >     * this behavior, thus enabling continuation upon future
898 >     * registrations, by overriding this method to always return
899 >     * {@code false}:
900 >     *
901 >     * <pre> {@code
902 >     * Phaser phaser = new Phaser() {
903 >     *   protected boolean onAdvance(int phase, int parties) { return false; }
904 >     * }}</pre>
905 >     *
906 >     * @param phase the current phase number on entry to this method,
907 >     * before this phaser is advanced
908 >     * @param registeredParties the current number of registered parties
909 >     * @return {@code true} if this phaser should terminate
910       */
911      protected boolean onAdvance(int phase, int registeredParties) {
912 <        return registeredParties <= 0;
912 >        return registeredParties == 0;
913      }
914  
915      /**
916 <     * Returns a string identifying this barrier, as well as its
916 >     * Returns a string identifying this phaser, as well as its
917       * state.  The state, in brackets, includes the String {@code
918 <     * "phase ="} followed by the phase number, {@code "parties ="}
918 >     * "phase = "} followed by the phase number, {@code "parties = "}
919       * followed by the number of registered parties, and {@code
920 <     * "arrived ="} followed by the number of arrived parties
920 >     * "arrived = "} followed by the number of arrived parties.
921       *
922 <     * @return a string identifying this barrier, as well as its state
922 >     * @return a string identifying this phaser, as well as its state
923       */
924      public String toString() {
925 <        long s = state.get();
475 <        return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
925 >        return stateToString(reconcileState());
926      }
927  
478    // methods for tripping and waiting
479
928      /**
929 <     * Advance the current phase (or terminate)
929 >     * Implementation of toString and string-based error messages
930       */
931 <    private void trip(int phase, int parties) {
932 <        int next = onAdvance(phase, parties)? -1 : ((phase + 1) & phaseMask);
933 <        state.set(stateFor(next, parties, parties));
934 <        if (head.get() != null)
935 <            releaseWaiters(next);
931 >    private String stateToString(long s) {
932 >        return super.toString() +
933 >            "[phase = " + phaseOf(s) +
934 >            " parties = " + partiesOf(s) +
935 >            " arrived = " + arrivedOf(s) + "]";
936      }
937  
938 <    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 <    }
938 >    // Waiting mechanics
939  
940 <    private int timedHelpingWait(int phase, long nanos) throws TimeoutException {
941 <        final AtomicLong state = this.state;
942 <        long lastTime = System.nanoTime();
943 <        int p;
944 <        while ((p = phaseOf(state.get())) == phase) {
945 <            long now = System.nanoTime();
946 <            nanos -= now - lastTime;
947 <            lastTime = now;
948 <            if (nanos <= 0) {
949 <                if ((p = phaseOf(state.get())) == phase)
950 <                    throw new TimeoutException();
951 <                else
952 <                    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 <                }
940 >    /**
941 >     * Removes and signals threads from queue for phase.
942 >     */
943 >    private void releaseWaiters(int phase) {
944 >        QNode q;   // first element of queue
945 >        Thread t;  // its thread
946 >        AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
947 >        while ((q = head.get()) != null &&
948 >               q.phase != (int)(root.state >>> PHASE_SHIFT)) {
949 >            if (head.compareAndSet(q, q.next) &&
950 >                (t = q.thread) != null) {
951 >                q.thread = null;
952 >                LockSupport.unpark(t);
953              }
954          }
531        return p;
955      }
956  
957      /**
958 <     * Wait nodes for Treiber stack representing wait queue for non-FJ
959 <     * tasks. The waiting scheme is an adaptation of the one used in
960 <     * forkjoin.PoolBarrier.
958 >     * Variant of releaseWaiters that additionally tries to remove any
959 >     * nodes no longer waiting for advance due to timeout or
960 >     * interrupt. Currently, nodes are removed only if they are at
961 >     * head of queue, which suffices to reduce memory footprint in
962 >     * most usages.
963 >     *
964 >     * @return current phase on exit
965       */
966 <    static final class QNode {
967 <        QNode next;
968 <        volatile Thread thread; // nulled to cancel wait
969 <        final int phase;
970 <        QNode(Thread t, int c) {
971 <            thread = t;
972 <            phase = c;
973 <        }
974 <    }
975 <
976 <    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);
966 >    private int abortWait(int phase) {
967 >        AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
968 >        for (;;) {
969 >            Thread t;
970 >            QNode q = head.get();
971 >            int p = (int)(root.state >>> PHASE_SHIFT);
972 >            if (q == null || ((t = q.thread) != null && q.phase == p))
973 >                return p;
974 >            if (head.compareAndSet(q, q.next) && t != null) {
975 >                q.thread = null;
976 >                LockSupport.unpark(t);
977              }
978          }
979      }
980  
981      /** The number of CPUs, for spin control */
982 <    static final int NCPUS = Runtime.getRuntime().availableProcessors();
567 <
568 <    /**
569 <     * The number of times to spin before blocking in timed waits.
570 <     * The value is empirically derived.
571 <     */
572 <    static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
573 <
574 <    /**
575 <     * The number of times to spin before blocking in untimed waits.
576 <     * This is greater than timed value because untimed waits spin
577 <     * faster since they don't need to check times on each spin.
578 <     */
579 <    static final int maxUntimedSpins = maxTimedSpins * 32;
982 >    private static final int NCPU = Runtime.getRuntime().availableProcessors();
983  
984      /**
985 <     * The number of nanoseconds for which it is faster to spin
986 <     * rather than to use timed park. A rough estimate suffices.
985 >     * The number of times to spin before blocking while waiting for
986 >     * advance, per arrival while waiting. On multiprocessors, fully
987 >     * blocking and waking up a large number of threads all at once is
988 >     * usually a very slow process, so we use rechargeable spins to
989 >     * avoid it when threads regularly arrive: When a thread in
990 >     * internalAwaitAdvance notices another arrival before blocking,
991 >     * and there appear to be enough CPUs available, it spins
992 >     * SPINS_PER_ARRIVAL more times before blocking. The value trades
993 >     * off good-citizenship vs big unnecessary slowdowns.
994       */
995 <    static final long spinForTimeoutThreshold = 1000L;
995 >    static final int SPINS_PER_ARRIVAL = (NCPU < 2) ? 1 : 1 << 8;
996  
997      /**
998 <     * Enqueues node and waits unless aborted or signalled.
999 <     */
1000 <    private boolean untimedWait(Thread thread, int currentPhase,
1001 <                               boolean abortOnInterrupt) {
1002 <        final AtomicReference<QNode> head = this.head;
1003 <        final AtomicLong state = this.state;
1004 <        boolean wasInterrupted = false;
1005 <        QNode node = null;
1006 <        boolean queued = false;
1007 <        int spins = maxUntimedSpins;
1008 <        while (phaseOf(state.get()) == currentPhase) {
1009 <            QNode h;
1010 <            if (node != null && queued) {
1011 <                if (node.thread != null) {
1012 <                    LockSupport.park();
1013 <                    if (Thread.interrupted()) {
1014 <                        wasInterrupted = true;
1015 <                        if (abortOnInterrupt)
1016 <                            break;
1017 <                    }
998 >     * Possibly blocks and waits for phase to advance unless aborted.
999 >     * Call only on root phaser.
1000 >     *
1001 >     * @param phase current phase
1002 >     * @param node if non-null, the wait node to track interrupt and timeout;
1003 >     * if null, denotes noninterruptible wait
1004 >     * @return current phase
1005 >     */
1006 >    private int internalAwaitAdvance(int phase, QNode node) {
1007 >        // assert root == this;
1008 >        releaseWaiters(phase-1);          // ensure old queue clean
1009 >        boolean queued = false;           // true when node is enqueued
1010 >        int lastUnarrived = 0;            // to increase spins upon change
1011 >        int spins = SPINS_PER_ARRIVAL;
1012 >        long s;
1013 >        int p;
1014 >        while ((p = (int)((s = state) >>> PHASE_SHIFT)) == phase) {
1015 >            if (node == null) {           // spinning in noninterruptible mode
1016 >                int unarrived = (int)s & UNARRIVED_MASK;
1017 >                if (unarrived != lastUnarrived &&
1018 >                    (lastUnarrived = unarrived) < NCPU)
1019 >                    spins += SPINS_PER_ARRIVAL;
1020 >                boolean interrupted = Thread.interrupted();
1021 >                if (interrupted || --spins < 0) { // need node to record intr
1022 >                    node = new QNode(this, phase, false, false, 0L);
1023 >                    node.wasInterrupted = interrupted;
1024                  }
1025              }
1026 <            else if ((h = head.get()) != null && h.phase != currentPhase) {
1027 <                if (phaseOf(state.get()) == currentPhase) { // must recheck
1028 <                    if (head.compareAndSet(h, h.next)) {
1029 <                        Thread t = h.thread; // help clear out old waiters
1030 <                        if (t != null) {
1031 <                            h.thread = null;
1032 <                            LockSupport.unpark(t);
1033 <                        }
1034 <                    }
1026 >            else if (node.isReleasable()) // done or aborted
1027 >                break;
1028 >            else if (!queued) {           // push onto queue
1029 >                AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
1030 >                QNode q = node.next = head.get();
1031 >                if ((q == null || q.phase == phase) &&
1032 >                    (int)(state >>> PHASE_SHIFT) == phase) // avoid stale enq
1033 >                    queued = head.compareAndSet(q, node);
1034 >            }
1035 >            else {
1036 >                try {
1037 >                    ForkJoinPool.managedBlock(node);
1038 >                } catch (InterruptedException ie) {
1039 >                    node.wasInterrupted = true;
1040                  }
620                else
621                    break;
1041              }
623            else if (node != null)
624                queued = head.compareAndSet(node.next = h, node);
625            else if (spins <= 0)
626                node = new QNode(thread, currentPhase);
627            else
628                --spins;
1042          }
1043 <        if (node != null)
1044 <            node.thread = null;
1045 <        return wasInterrupted;
1043 >
1044 >        if (node != null) {
1045 >            if (node.thread != null)
1046 >                node.thread = null;       // avoid need for unpark()
1047 >            if (node.wasInterrupted && !node.interruptible)
1048 >                Thread.currentThread().interrupt();
1049 >            if (p == phase && (p = (int)(state >>> PHASE_SHIFT)) == phase)
1050 >                return abortWait(phase); // possibly clean up on abort
1051 >        }
1052 >        releaseWaiters(phase);
1053 >        return p;
1054      }
1055  
1056      /**
1057 <     * Messier timeout version
1057 >     * Wait nodes for Treiber stack representing wait queue
1058       */
1059 <    private void timedWait(Thread thread, int currentPhase, long nanos)
1060 <        throws InterruptedException, TimeoutException {
1061 <        final AtomicReference<QNode> head = this.head;
1062 <        final AtomicLong state = this.state;
1063 <        long lastTime = System.nanoTime();
1064 <        QNode node = null;
1065 <        boolean queued = false;
1066 <        int spins = maxTimedSpins;
1067 <        while (phaseOf(state.get()) == currentPhase) {
1068 <            QNode h;
1069 <            long now = System.nanoTime();
1070 <            nanos -= now - lastTime;
1071 <            lastTime = now;
1072 <            if (nanos <= 0) {
1073 <                if (node != null)
1074 <                    node.thread = null;
1075 <                if (phaseOf(state.get()) == currentPhase)
1076 <                    throw new TimeoutException();
1077 <                else
1078 <                    break;
1059 >    static final class QNode implements ForkJoinPool.ManagedBlocker {
1060 >        final Phaser phaser;
1061 >        final int phase;
1062 >        final boolean interruptible;
1063 >        final boolean timed;
1064 >        boolean wasInterrupted;
1065 >        long nanos;
1066 >        long lastTime;
1067 >        volatile Thread thread; // nulled to cancel wait
1068 >        QNode next;
1069 >
1070 >        QNode(Phaser phaser, int phase, boolean interruptible,
1071 >              boolean timed, long nanos) {
1072 >            this.phaser = phaser;
1073 >            this.phase = phase;
1074 >            this.interruptible = interruptible;
1075 >            this.nanos = nanos;
1076 >            this.timed = timed;
1077 >            this.lastTime = timed ? System.nanoTime() : 0L;
1078 >            thread = Thread.currentThread();
1079 >        }
1080 >
1081 >        public boolean isReleasable() {
1082 >            if (thread == null)
1083 >                return true;
1084 >            if (phaser.getPhase() != phase) {
1085 >                thread = null;
1086 >                return true;
1087              }
1088 <            else if (node != null && queued) {
1089 <                if (node.thread != null &&
1090 <                    nanos > spinForTimeoutThreshold) {
1091 <                    //                LockSupport.parkNanos(this, nanos);
1092 <                    LockSupport.parkNanos(nanos);
664 <                    if (Thread.interrupted()) {
665 <                        node.thread = null;
666 <                        throw new InterruptedException();
667 <                    }
668 <                }
1088 >            if (Thread.interrupted())
1089 >                wasInterrupted = true;
1090 >            if (wasInterrupted && interruptible) {
1091 >                thread = null;
1092 >                return true;
1093              }
1094 <            else if ((h = head.get()) != null && h.phase != currentPhase) {
1095 <                if (phaseOf(state.get()) == currentPhase) { // must recheck
1096 <                    if (head.compareAndSet(h, h.next)) {
1097 <                        Thread t = h.thread; // help clear out old waiters
1098 <                        if (t != null) {
1099 <                            h.thread = null;
1100 <                            LockSupport.unpark(t);
1101 <                        }
1102 <                    }
1094 >            if (timed) {
1095 >                if (nanos > 0L) {
1096 >                    long now = System.nanoTime();
1097 >                    nanos -= now - lastTime;
1098 >                    lastTime = now;
1099 >                }
1100 >                if (nanos <= 0L) {
1101 >                    thread = null;
1102 >                    return true;
1103                  }
680                else
681                    break;
1104              }
1105 <            else if (node != null)
1106 <                queued = head.compareAndSet(node.next = h, node);
1107 <            else if (spins <= 0)
1108 <                node = new QNode(thread, currentPhase);
1109 <            else
1110 <                --spins;
1105 >            return false;
1106 >        }
1107 >
1108 >        public boolean block() {
1109 >            if (isReleasable())
1110 >                return true;
1111 >            else if (!timed)
1112 >                LockSupport.park(this);
1113 >            else if (nanos > 0)
1114 >                LockSupport.parkNanos(this, nanos);
1115 >            return isReleasable();
1116 >        }
1117 >    }
1118 >
1119 >    // Unsafe mechanics
1120 >
1121 >    private static final sun.misc.Unsafe UNSAFE;
1122 >    private static final long stateOffset;
1123 >    static {
1124 >        try {
1125 >            UNSAFE = getUnsafe();
1126 >            Class<?> k = Phaser.class;
1127 >            stateOffset = UNSAFE.objectFieldOffset
1128 >                (k.getDeclaredField("state"));
1129 >        } catch (Exception e) {
1130 >            throw new Error(e);
1131          }
690        if (node != null)
691            node.thread = null;
1132      }
1133  
1134 +    /**
1135 +     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1136 +     * Replace with a simple call to Unsafe.getUnsafe when integrating
1137 +     * into a jdk.
1138 +     *
1139 +     * @return a sun.misc.Unsafe
1140 +     */
1141 +    private static sun.misc.Unsafe getUnsafe() {
1142 +        try {
1143 +            return sun.misc.Unsafe.getUnsafe();
1144 +        } catch (SecurityException tryReflectionInstead) {}
1145 +        try {
1146 +            return java.security.AccessController.doPrivileged
1147 +            (new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1148 +                public sun.misc.Unsafe run() throws Exception {
1149 +                    Class<sun.misc.Unsafe> k = sun.misc.Unsafe.class;
1150 +                    for (java.lang.reflect.Field f : k.getDeclaredFields()) {
1151 +                        f.setAccessible(true);
1152 +                        Object x = f.get(null);
1153 +                        if (k.isInstance(x))
1154 +                            return k.cast(x);
1155 +                    }
1156 +                    throw new NoSuchFieldError("the Unsafe");
1157 +                }});
1158 +        } catch (java.security.PrivilegedActionException e) {
1159 +            throw new RuntimeException("Could not initialize intrinsics",
1160 +                                       e.getCause());
1161 +        }
1162 +    }
1163   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines