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 1.2 by jsr166, Fri Jul 25 18:10:41 2008 UTC vs.
Revision 1.55 by dl, Mon Nov 15 12:51:54 2010 UTC

# Line 5 | Line 5
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 {@code Phaser} has an associated phase number. The
38 > * phase number starts at zero, and advances when all parties arrive
39 > * at the barrier, 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 barrier 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 at a
49 > *       barrier. These methods do not block, but return an associated
50 > *       <em>arrival phase number</em>; that is, the phase number of
51 > *       the barrier to which the arrival applied. When the final
52 > *       party for a given phase arrives, an optional barrier action
53 > *       is performed and the phase advances.  Barrier actions,
54 > *       performed by the party triggering a phase advance, are
55 > *       arranged by overriding method {@link #onAdvance(int, int)},
56 > *       which also controls termination. Overriding this method is
57 > *       similar to, but more flexible than, providing a barrier
58 > *       action to a {@code CyclicBarrier}.
59 > *
60 > *   <li> <b>Waiting.</b> Method {@link #awaitAdvance} requires an
61 > *       argument indicating an arrival phase number, and returns when
62 > *       the barrier 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 barrier. 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 {@code Phaser} may enter a
78 + * <em>termination</em> state in which all synchronization methods
79 + * immediately return without updating phaser state or waiting for
80 + * advance, and indicating (via a negative phase value) that execution
81 + * is complete.  Termination is triggered when an invocation of {@code
82 + * onAdvance} returns {@code true}.  As illustrated below, when
83 + * phasers control actions with a fixed number of iterations, it is
84 + * often convenient to override this method to cause termination when
85 + * the current phase number reaches a threshold. Method {@link
86 + * #forceTermination} is also available to abruptly release waiting
87 + * threads and allow them to terminate.
88   *
89 < * <li> Barrier actions, performed by the task triggering a phase
90 < * advance while others may be waiting, are arranged by overriding
91 < * method <tt>onAdvance</tt>, that also controls termination.
92 < *
93 < * <li> Phasers may enter a <em>termination</em> state in which all
94 < * await actions immediately return, indicating (via a negative phase
95 < * 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.
89 > * <p> <b>Tiering.</b> Phasers may be <em>tiered</em> (i.e., arranged
90 > * in tree structures) to reduce contention. Phasers with large
91 > * numbers of parties that would otherwise experience heavy
92 > * synchronization contention costs may instead be set up so that
93 > * groups of sub-phasers share a common parent.  This may greatly
94 > * increase throughput even though it incurs greater per-operation
95 > * overhead.
96   *
97 < * </ul>
97 > * <p><b>Monitoring.</b> While synchronization methods may be invoked
98 > * only by registered parties, the current state of a phaser may be
99 > * monitored by any caller.  At any given moment there are {@link
100 > * #getRegisteredParties} parties in total, of which {@link
101 > * #getArrivedParties} have arrived at the current phase ({@link
102 > * #getPhase}).  When the remaining ({@link #getUnarrivedParties})
103 > * parties arrive, the phase advances.  The values returned by these
104 > * methods may reflect transient states and so are not in general
105 > * useful for synchronization control.  Method {@link #toString}
106 > * returns snapshots of these state queries in a form convenient for
107 > * informal monitoring.
108   *
109 < * <p><b>Sample usage:</b>
109 > * <p><b>Sample usages:</b>
110   *
111 < * <p>[todo: non-FJ example]
111 > * <p>A {@code Phaser} may be used instead of a {@code CountDownLatch}
112 > * to control a one-shot action serving a variable number of parties.
113 > * The typical idiom is for the method setting this up to first
114 > * register, then start the actions, then deregister, as in:
115   *
116 < * <p> A Phaser may be used to support a style of programming in
117 < * which a task waits for others to complete, without otherwise
118 < * needing to keep track of which tasks it is waiting for. This is
119 < * similar to the "sync" construct in Cilk and "clocks" in X10.
120 < * Special constructions based on such barriers are available using
121 < * the <tt>LinkedAsyncAction</tt> and <tt>CyclicAction</tt> classes,
122 < * but they can be useful in other contexts as well.  For a simple
123 < * (but not very useful) example, here is a variant of Fibonacci:
124 < *
125 < * <pre>
126 < * class BarrierFibonacci extends RecursiveAction {
127 < *   int argument, result;
91 < *   final Phaser parentBarrier;
92 < *   BarrierFibonacci(int n, Phaser parentBarrier) {
93 < *     this.argument = n;
94 < *     this.parentBarrier = parentBarrier;
95 < *     parentBarrier.register();
116 > *  <pre> {@code
117 > * void runTasks(List<Runnable> tasks) {
118 > *   final Phaser phaser = new Phaser(1); // "1" to register self
119 > *   // create and start threads
120 > *   for (Runnable task : tasks) {
121 > *     phaser.register();
122 > *     new Thread() {
123 > *       public void run() {
124 > *         phaser.arriveAndAwaitAdvance(); // await all creation
125 > *         task.run();
126 > *       }
127 > *     }.start();
128   *   }
129 < *   protected void compute() {
130 < *     int n = argument;
131 < *     if (n &lt;= 1)
132 < *        result = n;
133 < *     else {
134 < *        Phaser childBarrier = new Phaser(1);
135 < *        BarrierFibonacci f1 = new BarrierFibonacci(n - 1, childBarrier);
136 < *        BarrierFibonacci f2 = new BarrierFibonacci(n - 2, childBarrier);
137 < *        f1.fork();
138 < *        f2.fork();
139 < *        childBarrier.arriveAndAwait();
140 < *        result = f1.result + f2.result;
129 > *
130 > *   // allow threads to start and deregister self
131 > *   phaser.arriveAndDeregister();
132 > * }}</pre>
133 > *
134 > * <p>One way to cause a set of threads to repeatedly perform actions
135 > * for a given number of iterations is to override {@code onAdvance}:
136 > *
137 > *  <pre> {@code
138 > * void startTasks(List<Runnable> tasks, final int iterations) {
139 > *   final Phaser phaser = new Phaser() {
140 > *     protected boolean onAdvance(int phase, int registeredParties) {
141 > *       return phase >= iterations || registeredParties == 0;
142   *     }
143 < *     parentBarrier.arriveAndDeregister();
143 > *   };
144 > *   phaser.register();
145 > *   for (final Runnable task : tasks) {
146 > *     phaser.register();
147 > *     new Thread() {
148 > *       public void run() {
149 > *         do {
150 > *           task.run();
151 > *           phaser.arriveAndAwaitAdvance();
152 > *         } while (!phaser.isTerminated());
153 > *       }
154 > *     }.start();
155 > *   }
156 > *   phaser.arriveAndDeregister(); // deregister self, don't wait
157 > * }}</pre>
158 > *
159 > * If the main task must later await termination, it
160 > * may re-register and then execute a similar loop:
161 > *  <pre> {@code
162 > *   // ...
163 > *   phaser.register();
164 > *   while (!phaser.isTerminated())
165 > *     phaser.arriveAndAwaitAdvance();}</pre>
166 > *
167 > * <p>Related constructions may be used to await particular phase numbers
168 > * in contexts where you are sure that the phase will never wrap around
169 > * {@code Integer.MAX_VALUE}. For example:
170 > *
171 > *  <pre> {@code
172 > * void awaitPhase(Phaser phaser, int phase) {
173 > *   int p = phaser.register(); // assumes caller not already registered
174 > *   while (p < phase) {
175 > *     if (phaser.isTerminated())
176 > *       // ... deal with unexpected termination
177 > *     else
178 > *       p = phaser.arriveAndAwaitAdvance();
179 > *   }
180 > *   phaser.arriveAndDeregister();
181 > * }}</pre>
182 > *
183 > *
184 > * <p>To create a set of tasks using a tree of phasers,
185 > * you could use code of the following form, assuming a
186 > * Task class with a constructor accepting a phaser that
187 > * it registers with upon construction:
188 > *
189 > *  <pre> {@code
190 > * void build(Task[] actions, int lo, int hi, Phaser ph) {
191 > *   if (hi - lo > TASKS_PER_PHASER) {
192 > *     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
193 > *       int j = Math.min(i + TASKS_PER_PHASER, hi);
194 > *       build(actions, i, j, new Phaser(ph));
195 > *     }
196 > *   } else {
197 > *     for (int i = lo; i < hi; ++i)
198 > *       actions[i] = new Task(ph);
199 > *       // assumes new Task(ph) performs ph.register()
200   *   }
201   * }
202 < * </pre>
202 > * // .. initially called, for n tasks via
203 > * build(new Task[n], 0, n, new Phaser());}</pre>
204 > *
205 > * The best value of {@code TASKS_PER_PHASER} depends mainly on
206 > * expected barrier synchronization rates. A value as low as four may
207 > * be appropriate for extremely small per-barrier task bodies (thus
208 > * high rates), or up to hundreds for extremely large ones.
209   *
210   * <p><b>Implementation notes</b>: This implementation restricts the
211 < * maximum number of parties to 65535. Attempts to register
212 < * additional parties result in IllegalStateExceptions.  
211 > * maximum number of parties to 65535. Attempts to register additional
212 > * parties result in {@code IllegalStateException}. However, you can and
213 > * should create tiered phasers to accommodate arbitrarily large sets
214 > * of participants.
215 > *
216 > * @since 1.7
217 > * @author Doug Lea
218   */
219   public class Phaser {
220      /*
221       * This class implements an extension of X10 "clocks".  Thanks to
222 <     * Vijay Saraswat for the idea of applying it to ForkJoinTasks,
223 <     * and to Vivek Sarkar for enhancements to extend functionality.
222 >     * Vijay Saraswat for the idea, and to Vivek Sarkar for
223 >     * enhancements to extend functionality.
224       */
225  
226      /**
227       * Barrier state representation. Conceptually, a barrier contains
228       * four values:
229 <     *
230 <     * * parties -- the number of parties to wait (16 bits)
231 <     * * unarrived -- the number of parties yet to hit barrier (16 bits)
232 <     * * phase -- the generation of the barrier (31 bits)
233 <     * * terminated -- set if barrier is terminated (1 bit)
229 >     *
230 >     * * unarrived -- the number of parties yet to hit barrier (bits  0-15)
231 >     * * parties -- the number of parties to wait              (bits 16-31)
232 >     * * phase -- the generation of the barrier                (bits 32-62)
233 >     * * terminated -- set if barrier is terminated            (bit  63 / sign)
234       *
235       * However, to efficiently maintain atomicity, these values are
236 <     * packed into a single AtomicLong. Termination uses the sign bit
237 <     * of 32 bit representation of phase, so phase is set to -1 on
238 <     * termination.
239 <     */
240 <    private final AtomicLong state;
241 <
242 <    /**
243 <     * Head of Treiber stack for waiting nonFJ threads.
244 <     */
245 <    private final AtomicReference<QNode> head = new AtomicReference<QNode>();
236 >     * packed into a single (atomic) long. Termination uses the sign
237 >     * bit of 32 bit representation of phase, so phase is set to -1 on
238 >     * termination. Good performance relies on keeping state decoding
239 >     * and encoding simple, and keeping race windows short.
240 >     */
241 >    private volatile long state;
242 >
243 >    private static final int  MAX_PARTIES    = 0xffff;
244 >    private static final int  MAX_PHASE      = 0x7fffffff;
245 >    private static final int  PARTIES_SHIFT  = 16;
246 >    private static final int  PHASE_SHIFT    = 32;
247 >    private static final int  UNARRIVED_MASK = 0xffff;
248 >    private static final int  PARTIES_MASK   = 0xffff0000;
249 >    private static final long LPARTIES_MASK  = 0xffff0000L; // long version
250 >    private static final long ONE_ARRIVAL    = 1L;
251 >    private static final long ONE_PARTY      = 1L << PARTIES_SHIFT;
252 >    private static final long TERMINATION_PHASE  = -1L << PHASE_SHIFT;
253  
254 <    private static final int ushortBits = 16;
148 <    private static final int ushortMask =  (1 << ushortBits) - 1;
149 <    private static final int phaseMask = 0x7fffffff;
254 >    // The following unpacking methods are usually manually inlined
255  
256      private static int unarrivedOf(long s) {
257 <        return (int)(s & ushortMask);
257 >        return ((int) s) & UNARRIVED_MASK;
258      }
259  
260      private static int partiesOf(long s) {
261 <        return (int)(s & (ushortMask << 16)) >>> 16;
261 >        return (((int) s) & PARTIES_MASK) >>> PARTIES_SHIFT;
262      }
263  
264      private static int phaseOf(long s) {
265 <        return (int)(s >>> 32);
265 >        return (int) (s >>> PHASE_SHIFT);
266      }
267  
268      private static int arrivedOf(long s) {
269          return partiesOf(s) - unarrivedOf(s);
270      }
271  
272 <    private static long stateFor(int phase, int parties, int unarrived) {
273 <        return (((long)phase) << 32) | ((parties << 16) | unarrived);
274 <    }
275 <
171 <    private static IllegalStateException badBounds(int parties, int unarrived) {
172 <        return new IllegalStateException("Attempt to set " + unarrived +
173 <                                         " unarrived of " + parties + " parties");
174 <    }
272 >    /**
273 >     * The parent of this phaser, or null if none
274 >     */
275 >    private final Phaser parent;
276  
277      /**
278 <     * Creates a new Phaser without any initially registered parties,
279 <     * and initial phase number 0.
278 >     * The root of phaser tree. Equals this if not in a tree.  Used to
279 >     * support faster state push-down.
280       */
281 <    public Phaser() {
181 <        state = new AtomicLong(stateFor(0, 0, 0));
182 <    }
281 >    private final Phaser root;
282  
283      /**
284 <     * Creates a new Phaser with the given numbers of registered
285 <     * unarrived parties and initial phase number 0.
286 <     * @param parties the number of parties required to trip barrier.
287 <     * @throws IllegalArgumentException if parties less than zero
189 <     * or greater than the maximum number of parties supported.
284 >     * Heads of Treiber stacks for waiting threads. To eliminate
285 >     * contention when releasing some threads while adding others, we
286 >     * use two of them, alternating across even and odd phases.
287 >     * Subphasers share queues with root to speed up releases.
288       */
289 <    public Phaser(int parties) {
290 <        if (parties < 0 || parties > ushortMask)
291 <            throw new IllegalArgumentException("Illegal number of parties");
292 <        state = new AtomicLong(stateFor(0, parties, parties));
289 >    private final AtomicReference<QNode> evenQ;
290 >    private final AtomicReference<QNode> oddQ;
291 >
292 >    private AtomicReference<QNode> queueFor(int phase) {
293 >        return ((phase & 1) == 0) ? evenQ : oddQ;
294      }
295  
296      /**
297 <     * Adds a new unarrived party to this phaser.
298 <     * @return the current barrier phase number upon registration
299 <     * @throws IllegalStateException if attempting to register more
300 <     * than the maximum supported number of parties.
297 >     * Main implementation for methods arrive and arriveAndDeregister.
298 >     * Manually tuned to speed up and minimize race windows for the
299 >     * common case of just decrementing unarrived field.
300 >     *
301 >     * @param adj - adjustment to apply to state -- either
302 >     * ONE_ARRIVAL (for arrive) or
303 >     * ONE_ARRIVAL|ONE_PARTY (for arriveAndDeregister)
304       */
305 <    public int register() { // increment both parties and unarrived
204 <        final AtomicLong state = this.state;
305 >    private int doArrive(long adj) {
306          for (;;) {
307 <            long s = state.get();
308 <            int phase = phaseOf(s);
309 <            int parties = partiesOf(s) + 1;
209 <            int unarrived = unarrivedOf(s) + 1;
210 <            if (parties > ushortMask || unarrived > ushortMask)
211 <                throw badBounds(parties, unarrived);
212 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
307 >            long s;
308 >            int phase, unarrived;
309 >            if ((phase = (int)((s = state) >>> PHASE_SHIFT)) < 0)
310                  return phase;
311 +            else if ((unarrived = ((int)s) & UNARRIVED_MASK) == 0)
312 +                checkBadArrive(s);
313 +            else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adj)) {
314 +                if (unarrived == 1) {
315 +                    Phaser par;
316 +                    long p = s & LPARTIES_MASK; // unshifted parties field
317 +                    long lu = p >>> PARTIES_SHIFT;
318 +                    int u = (int)lu;
319 +                    int nextPhase = (phase + 1) & MAX_PHASE;
320 +                    long next = ((long)nextPhase << PHASE_SHIFT) | p | lu;
321 +                    if ((par = parent) == null) {
322 +                        if (onAdvance(phase, u))
323 +                            next |= TERMINATION_PHASE; // obliterate phase
324 +                        UNSAFE.compareAndSwapLong(this, stateOffset, s, next);
325 +                        releaseWaiters(phase);
326 +                    }
327 +                    else {
328 +                        par.doArrive(u == 0?
329 +                                     ONE_ARRIVAL|ONE_PARTY : ONE_ARRIVAL);
330 +                        if ((int)(par.state >>> PHASE_SHIFT) != nextPhase ||
331 +                            ((int)(state >>> PHASE_SHIFT) != nextPhase &&
332 +                             !UNSAFE.compareAndSwapLong(this, stateOffset,
333 +                                                        s, next)))
334 +                            reconcileState();
335 +                    }
336 +                }
337 +                return phase;
338 +            }
339          }
340      }
341  
342      /**
343 <     * Arrives at the barrier, but does not wait for others.  (You can
344 <     * in turn wait for others via {@link #awaitAdvance}).
343 >     * Rechecks state and throws bounds exceptions on arrival -- called
344 >     * only if unarrived is apparently zero.
345 >     */
346 >    private void checkBadArrive(long s) {
347 >        if (reconcileState() == s)
348 >            throw new IllegalStateException
349 >                ("Attempted arrival of unregistered party for " +
350 >                 stateToString(s));
351 >    }
352 >
353 >    /**
354 >     * Implementation of register, bulkRegister
355       *
356 <     * @return the current barrier phase number upon entry to
222 <     * this method, or a negative value if terminated;
223 <     * @throws IllegalStateException if the number of unarrived
224 <     * parties would become negative.
356 >     * @param registrations number to add to both parties and unarrived fields
357       */
358 <    public int arrive() { // decrement unarrived. If zero, trip
359 <        final AtomicLong state = this.state;
358 >    private int doRegister(int registrations) {
359 >        long adj = (long)registrations; // adjustment to state
360 >        adj |= adj << PARTIES_SHIFT;
361 >        Phaser par = parent;
362          for (;;) {
363 <            long s = state.get();
364 <            int phase = phaseOf(s);
365 <            int parties = partiesOf(s);
232 <            int unarrived = unarrivedOf(s) - 1;
233 <            if (unarrived < 0)
234 <                throw badBounds(parties, unarrived);
235 <            if (unarrived == 0 && phase >= 0) {
236 <                trip(phase, parties);
363 >            int phase, parties;
364 >            long s = par == null? state : reconcileState();
365 >            if ((phase = (int)(s >>> PHASE_SHIFT)) < 0)
366                  return phase;
367 <            }
368 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
367 >            if ((parties = (((int)s) & PARTIES_MASK) >>> PARTIES_SHIFT) != 0 &&
368 >                (((int)s) & UNARRIVED_MASK) == 0)
369 >                internalAwaitAdvance(phase, null); // wait for onAdvance
370 >            else if (parties + registrations > MAX_PARTIES)
371 >                throw new IllegalStateException(badRegister(s));
372 >            else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s + adj))
373                  return phase;
374          }
375      }
376  
377      /**
378 <     * Arrives at the barrier, and deregisters from it, without
379 <     * waiting for others.
380 <     *
381 <     * @return the current barrier phase number upon entry to
382 <     * this method, or a negative value if terminated;
383 <     * @throws IllegalStateException if the number of registered or
384 <     * unarrived parties would become negative.
378 >     * Returns message string for out of bounds exceptions on registration.
379 >     */
380 >    private String badRegister(long s) {
381 >        return "Attempt to register more than " +
382 >            MAX_PARTIES + " parties for " + stateToString(s);
383 >    }
384 >
385 >    /**
386 >     * Recursively resolves lagged phase propagation from root if necessary.
387       */
388 <    public int arriveAndDeregister() { // Same as arrive, plus decrement parties
389 <        final AtomicLong state = this.state;
388 >    private long reconcileState() {
389 >        Phaser par = parent;
390 >        if (par == null)
391 >            return state;
392 >        Phaser rt = root;
393          for (;;) {
394 <            long s = state.get();
395 <            int phase = phaseOf(s);
396 <            int parties = partiesOf(s) - 1;
397 <            int unarrived = unarrivedOf(s) - 1;
398 <            if (parties < 0 || unarrived < 0)
399 <                throw badBounds(parties, unarrived);
400 <            if (unarrived == 0 && phase >= 0) {
401 <                trip(phase, parties);
402 <                return phase;
394 >            long s, u;
395 >            int phase, rPhase, pPhase;
396 >            if ((phase = (int)((s = state)>>> PHASE_SHIFT)) < 0 ||
397 >                (rPhase = (int)(rt.state >>> PHASE_SHIFT)) == phase)
398 >                return s;
399 >            long pState = par.parent == null? par.state : par.reconcileState();
400 >            if (state == s) {
401 >                if ((rPhase < 0 || (((int)s) & UNARRIVED_MASK) == 0) &&
402 >                    ((pPhase = (int)(pState >>> PHASE_SHIFT)) < 0 ||
403 >                     pPhase == ((phase + 1) & MAX_PHASE)))
404 >                    UNSAFE.compareAndSwapLong
405 >                        (this, stateOffset, s,
406 >                         (((long) pPhase) << PHASE_SHIFT) |
407 >                         (u = s & LPARTIES_MASK) |
408 >                         (u >>> PARTIES_SHIFT)); // reset unarrived to parties
409 >                else
410 >                    releaseWaiters(phase); // help release others
411              }
266            if (state.compareAndSet(s, stateFor(phase, parties, unarrived)))
267                return phase;
412          }
413      }
414  
415      /**
416 <     * Arrives at the barrier and awaits others. Unlike other arrival
417 <     * methods, this method returns the arrival index of the
418 <     * caller. The caller tripping the barrier returns zero, the
275 <     * previous caller 1, and so on.
276 <     * @return the arrival index
277 <     * @throws IllegalStateException if the number of unarrived
278 <     * parties would become negative.
416 >     * Creates a new phaser without any initially registered parties,
417 >     * initial phase number 0, and no parent. Any thread using this
418 >     * phaser will need to first register for it.
419       */
420 <    public int arriveAndAwaitAdvance() {
421 <        final AtomicLong state = this.state;
422 <        for (;;) {
423 <            long s = state.get();
424 <            int phase = phaseOf(s);
425 <            int parties = partiesOf(s);
426 <            int unarrived = unarrivedOf(s) - 1;
427 <            if (unarrived < 0)
428 <                throw badBounds(parties, unarrived);
429 <            if (unarrived == 0 && phase >= 0) {
430 <                trip(phase, parties);
431 <                return 0;
432 <            }
433 <            if (state.compareAndSet(s, stateFor(phase, parties, unarrived))) {
434 <                awaitAdvance(phase);
435 <                return unarrived;
436 <            }
420 >    public Phaser() {
421 >        this(null, 0);
422 >    }
423 >
424 >    /**
425 >     * Creates a new phaser with the given number of registered
426 >     * unarrived parties, initial phase number 0, and no parent.
427 >     *
428 >     * @param parties the number of parties required to trip barrier
429 >     * @throws IllegalArgumentException if parties less than zero
430 >     * or greater than the maximum number of parties supported
431 >     */
432 >    public Phaser(int parties) {
433 >        this(null, parties);
434 >    }
435 >
436 >    /**
437 >     * Creates a new phaser with the given parent, without any
438 >     * initially registered parties. If parent is non-null this phaser
439 >     * is registered with the parent and its initial phase number is
440 >     * the same as that of parent phaser.
441 >     *
442 >     * @param parent the parent phaser
443 >     */
444 >    public Phaser(Phaser parent) {
445 >        this(parent, 0);
446 >    }
447 >
448 >    /**
449 >     * Creates a new phaser with the given parent and number of
450 >     * registered unarrived parties. If parent is non-null, this phaser
451 >     * is registered with the parent and its initial phase number is
452 >     * the same as that of parent phaser.
453 >     *
454 >     * @param parent the parent phaser
455 >     * @param parties the number of parties required to trip barrier
456 >     * @throws IllegalArgumentException if parties less than zero
457 >     * or greater than the maximum number of parties supported
458 >     */
459 >    public Phaser(Phaser parent, int parties) {
460 >        if (parties >>> PARTIES_SHIFT != 0)
461 >            throw new IllegalArgumentException("Illegal number of parties");
462 >        int phase;
463 >        this.parent = parent;
464 >        if (parent != null) {
465 >            Phaser r = parent.root;
466 >            this.root = r;
467 >            this.evenQ = r.evenQ;
468 >            this.oddQ = r.oddQ;
469 >            phase = parent.register();
470 >        }
471 >        else {
472 >            this.root = this;
473 >            this.evenQ = new AtomicReference<QNode>();
474 >            this.oddQ = new AtomicReference<QNode>();
475 >            phase = 0;
476          }
477 +        long p = (long)parties;
478 +        this.state = (((long)phase) << PHASE_SHIFT) | p | (p << PARTIES_SHIFT);
479      }
480  
481      /**
482 <     * Awaits the phase of the barrier to advance from the given
483 <     * value, or returns immediately if this barrier is terminated.
484 <     * @param phase the phase on entry to this method
485 <     * @return the phase on exit from this method
482 >     * Adds a new unarrived party to this phaser.
483 >     * If an ongoing invocation of {@link #onAdvance} is in progress,
484 >     * this method may wait until its completion before registering.
485 >     *
486 >     * @return the arrival phase number to which this registration applied
487 >     * @throws IllegalStateException if attempting to register more
488 >     * than the maximum supported number of parties
489 >     */
490 >    public int register() {
491 >        return doRegister(1);
492 >    }
493 >
494 >    /**
495 >     * Adds the given number of new unarrived parties to this phaser.
496 >     * If an ongoing invocation of {@link #onAdvance} is in progress,
497 >     * this method may wait until its completion before registering.
498 >     *
499 >     * @param parties the number of additional parties required to trip barrier
500 >     * @return the arrival phase number to which this registration applied
501 >     * @throws IllegalStateException if attempting to register more
502 >     * than the maximum supported number of parties
503 >     * @throws IllegalArgumentException if {@code parties < 0}
504 >     */
505 >    public int bulkRegister(int parties) {
506 >        if (parties < 0)
507 >            throw new IllegalArgumentException();
508 >        if (parties > MAX_PARTIES)
509 >            throw new IllegalStateException(badRegister(state));
510 >        if (parties == 0)
511 >            return getPhase();
512 >        return doRegister(parties);
513 >    }
514 >
515 >    /**
516 >     * Arrives at the barrier, but does not wait for others.  (You can
517 >     * in turn wait for others via {@link #awaitAdvance}).  It is an
518 >     * unenforced usage error for an unregistered party to invoke this
519 >     * method.
520 >     *
521 >     * @return the arrival phase number, or a negative value if terminated
522 >     * @throws IllegalStateException if not terminated and the number
523 >     * of unarrived parties would become negative
524 >     */
525 >    public int arrive() {
526 >        return doArrive(ONE_ARRIVAL);
527 >    }
528 >
529 >    /**
530 >     * Arrives at the barrier and deregisters from it without waiting
531 >     * for others. Deregistration reduces the number of parties
532 >     * required to trip the barrier in future phases.  If this phaser
533 >     * has a parent, and deregistration causes this phaser to have
534 >     * zero parties, this phaser also arrives at and is deregistered
535 >     * from its parent.  It is an unenforced usage error for an
536 >     * unregistered party to invoke this method.
537 >     *
538 >     * @return the arrival phase number, or a negative value if terminated
539 >     * @throws IllegalStateException if not terminated and the number
540 >     * of registered or unarrived parties would become negative
541 >     */
542 >    public int arriveAndDeregister() {
543 >        return doArrive(ONE_ARRIVAL|ONE_PARTY);
544 >    }
545 >
546 >    /**
547 >     * Arrives at the barrier and awaits others. Equivalent in effect
548 >     * to {@code awaitAdvance(arrive())}.  If you need to await with
549 >     * interruption or timeout, you can arrange this with an analogous
550 >     * construction using one of the other forms of the {@code
551 >     * awaitAdvance} method.  If instead you need to deregister upon
552 >     * arrival, use {@link #arriveAndDeregister}. It is an unenforced
553 >     * usage error for an unregistered party to invoke this method.
554 >     *
555 >     * @return the arrival phase number, or a negative number if terminated
556 >     * @throws IllegalStateException if not terminated and the number
557 >     * of unarrived parties would become negative
558 >     */
559 >    public int arriveAndAwaitAdvance() {
560 >        return awaitAdvance(arrive());
561 >    }
562 >
563 >    /**
564 >     * Awaits the phase of the barrier to advance from the given phase
565 >     * value, returning immediately if the current phase of the
566 >     * barrier is not equal to the given phase value or this barrier
567 >     * is terminated.
568 >     *
569 >     * @param phase an arrival phase number, or negative value if
570 >     * terminated; this argument is normally the value returned by a
571 >     * previous call to {@code arrive} or its variants
572 >     * @return the next arrival phase number, or a negative value
573 >     * if terminated or argument is negative
574       */
575      public int awaitAdvance(int phase) {
576 +        int p;
577          if (phase < 0)
578              return phase;
579 <        Thread current = Thread.currentThread();
580 <        if (current instanceof ForkJoinWorkerThread)
581 <            return helpingWait(phase);
582 <        if (untimedWait(current, phase, false))
583 <            current.interrupt();
314 <        return phaseOf(state.get());
579 >        else if ((p = (int)((parent == null? state : reconcileState())
580 >                            >>> PHASE_SHIFT)) == phase)
581 >            return internalAwaitAdvance(phase, null);
582 >        else
583 >            return p;
584      }
585  
586      /**
587 <     * Awaits the phase of the barrier to advance from the given
588 <     * value, or returns immediately if this barrier is terminated, or
589 <     * throws InterruptedException if interrupted while waiting.
590 <     * @param phase the phase on entry to this method
591 <     * @return the phase on exit from this method
587 >     * Awaits the phase of the barrier to advance from the given phase
588 >     * value, throwing {@code InterruptedException} if interrupted
589 >     * while waiting, or returning immediately if the current phase of
590 >     * the barrier is not equal to the given phase value or this
591 >     * barrier is terminated.
592 >     *
593 >     * @param phase an arrival phase number, or negative value if
594 >     * terminated; this argument is normally the value returned by a
595 >     * previous call to {@code arrive} or its variants
596 >     * @return the next arrival phase number, or a negative value
597 >     * if terminated or argument is negative
598       * @throws InterruptedException if thread interrupted while waiting
599       */
600 <    public int awaitAdvanceInterruptibly(int phase) throws InterruptedException {
600 >    public int awaitAdvanceInterruptibly(int phase)
601 >        throws InterruptedException {
602 >        int p;
603          if (phase < 0)
604              return phase;
605 <        Thread current = Thread.currentThread();
606 <        if (current instanceof ForkJoinWorkerThread)
607 <            return helpingWait(phase);
608 <        else if (Thread.interrupted() || untimedWait(current, phase, true))
609 <            throw new InterruptedException();
610 <        else
611 <            return phaseOf(state.get());
605 >        if ((p = (int)((parent == null? state : reconcileState())
606 >                       >>> PHASE_SHIFT)) == phase) {
607 >            QNode node = new QNode(this, phase, true, false, 0L);
608 >            p = internalAwaitAdvance(phase, node);
609 >            if (node.wasInterrupted)
610 >                throw new InterruptedException();
611 >        }
612 >        return p;
613      }
614  
615      /**
616 <     * Awaits the phase of the barrier to advance from the given value
617 <     * or the given timeout elapses, or returns immediately if this
618 <     * barrier is terminated.
619 <     * @param phase the phase on entry to this method
620 <     * @return the phase on exit from this method
616 >     * Awaits the phase of the barrier to advance from the given phase
617 >     * value or the given timeout to elapse, throwing {@code
618 >     * InterruptedException} if interrupted while waiting, or
619 >     * returning immediately if the current phase of the barrier is
620 >     * not equal to the given phase value or this barrier is
621 >     * terminated.
622 >     *
623 >     * @param phase an arrival phase number, or negative value if
624 >     * terminated; this argument is normally the value returned by a
625 >     * previous call to {@code arrive} or its variants
626 >     * @param timeout how long to wait before giving up, in units of
627 >     *        {@code unit}
628 >     * @param unit a {@code TimeUnit} determining how to interpret the
629 >     *        {@code timeout} parameter
630 >     * @return the next arrival phase number, or a negative value
631 >     * if terminated or argument is negative
632       * @throws InterruptedException if thread interrupted while waiting
633       * @throws TimeoutException if timed out while waiting
634       */
635 <    public int awaitAdvanceInterruptibly(int phase, long timeout, TimeUnit unit)
635 >    public int awaitAdvanceInterruptibly(int phase,
636 >                                         long timeout, TimeUnit unit)
637          throws InterruptedException, TimeoutException {
638 +        long nanos = unit.toNanos(timeout);
639 +        int p;
640          if (phase < 0)
641              return phase;
642 <        long nanos = unit.toNanos(timeout);
643 <        Thread current = Thread.currentThread();
644 <        if (current instanceof ForkJoinWorkerThread)
645 <            return timedHelpingWait(phase, nanos);
646 <        timedWait(current, phase, nanos);
647 <        return phaseOf(state.get());
642 >        if ((p = (int)((parent == null? state : reconcileState())
643 >                       >>> PHASE_SHIFT)) == phase) {
644 >            QNode node = new QNode(this, phase, true, true, nanos);
645 >            p = internalAwaitAdvance(phase, node);
646 >            if (node.wasInterrupted)
647 >                throw new InterruptedException();
648 >            else if (p == phase)
649 >                throw new TimeoutException();
650 >        }
651 >        return p;
652      }
653  
654      /**
655 <     * Forces this barrier to enter termination state. Counts of
656 <     * arrived and registered parties are unaffected. This method may
657 <     * be useful for coordinating recovery after one or more tasks
655 >     * Forces this barrier to enter termination state.  Counts of
656 >     * arrived and registered parties are unaffected.  If this phaser
657 >     * is a member of a tiered set of phasers, then all of the phasers
658 >     * in the set are terminated.  If this phaser is already
659 >     * terminated, this method has no effect.  This method may be
660 >     * useful for coordinating recovery after one or more tasks
661       * encounter unexpected exceptions.
662       */
663      public void forceTermination() {
664 <        final AtomicLong state = this.state;
665 <        for (;;) {
666 <            long s = state.get();
667 <            int phase = phaseOf(s);
668 <            int parties = partiesOf(s);
669 <            int unarrived = unarrivedOf(s);
670 <            if (phase < 0 ||
671 <                state.compareAndSet(s, stateFor(-1, parties, unarrived))) {
373 <                if (head.get() != null)
374 <                    releaseWaiters(-1);
664 >        // Only need to change root state
665 >        final Phaser root = this.root;
666 >        long s;
667 >        while ((s = root.state) >= 0) {
668 >            if (UNSAFE.compareAndSwapLong(root, stateOffset,
669 >                                          s, s | TERMINATION_PHASE)) {
670 >                releaseWaiters(0); // signal all threads
671 >                releaseWaiters(1);
672                  return;
673              }
674          }
675      }
676  
677      /**
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    /**
678       * Returns the current phase number. The maximum phase number is
679 <     * <tt>Integer.MAX_VALUE</tt>, after which it restarts at
679 >     * {@code Integer.MAX_VALUE}, after which it restarts at
680       * zero. Upon termination, the phase number is negative.
681 +     *
682       * @return the phase number, or a negative value if terminated
683       */
684 <    public int getPhase() {
685 <        return phaseOf(state.get());
684 >    public final int getPhase() {
685 >        return (int)((parent==null? state : reconcileState()) >>> PHASE_SHIFT);
686      }
687  
688      /**
689       * Returns the number of parties registered at this barrier.
690 +     *
691       * @return the number of parties
692       */
693      public int getRegisteredParties() {
694 <        return partiesOf(state.get());
694 >        return partiesOf(parent==null? state : reconcileState());
695      }
696  
697      /**
698 <     * Returns the number of parties that have arrived at the current
699 <     * phase of this barrier.
698 >     * Returns the number of registered parties that have arrived at
699 >     * the current phase of this barrier.
700 >     *
701       * @return the number of arrived parties
702       */
703      public int getArrivedParties() {
704 <        return arrivedOf(state.get());
704 >        return arrivedOf(parent==null? state : reconcileState());
705      }
706  
707      /**
708       * Returns the number of registered parties that have not yet
709       * arrived at the current phase of this barrier.
710 +     *
711       * @return the number of unarrived parties
712       */
713      public int getUnarrivedParties() {
714 <        return unarrivedOf(state.get());
714 >        return unarrivedOf(parent==null? state : reconcileState());
715      }
716  
717      /**
718 <     * Returns true if this barrier has been terminated.
719 <     * @return true if this barrier has been terminated
718 >     * Returns the parent of this phaser, or {@code null} if none.
719 >     *
720 >     * @return the parent of this phaser, or {@code null} if none
721 >     */
722 >    public Phaser getParent() {
723 >        return parent;
724 >    }
725 >
726 >    /**
727 >     * Returns the root ancestor of this phaser, which is the same as
728 >     * this phaser if it has no parent.
729 >     *
730 >     * @return the root ancestor of this phaser
731 >     */
732 >    public Phaser getRoot() {
733 >        return root;
734 >    }
735 >
736 >    /**
737 >     * Returns {@code true} if this barrier has been terminated.
738 >     *
739 >     * @return {@code true} if this barrier has been terminated
740       */
741      public boolean isTerminated() {
742 <        return phaseOf(state.get()) < 0;
742 >        return (parent == null? state : reconcileState()) < 0;
743      }
744  
745      /**
746 <     * Overridable method to perform an action upon phase advance, and
747 <     * to control termination. This method is invoked whenever the
748 <     * barrier is tripped (and thus all other waiting parties are
749 <     * dormant). If it returns true, then, rather than advance the
750 <     * phase number, this barrier will be set to a final termination
751 <     * state, and subsequent calls to <tt>isTerminated</tt> will
752 <     * return true.
753 <     *
754 <     * <p> The default version returns true when the number of
746 >     * Overridable method to perform an action upon impending phase
747 >     * advance, and to control termination. This method is invoked
748 >     * upon arrival of the party tripping the barrier (when all other
749 >     * waiting parties are dormant).  If this method returns {@code
750 >     * true}, then, rather than advance the phase number, this barrier
751 >     * will be set to a final termination state, and subsequent calls
752 >     * to {@link #isTerminated} will return true. Any (unchecked)
753 >     * Exception or Error thrown by an invocation of this method is
754 >     * propagated to the party attempting to trip the barrier, in
755 >     * which case no advance occurs.
756 >     *
757 >     * <p>The arguments to this method provide the state of the phaser
758 >     * prevailing for the current transition.  The effects of invoking
759 >     * arrival, registration, and waiting methods on this Phaser from
760 >     * within {@code onAdvance} are unspecified and should not be
761 >     * relied on.
762 >     *
763 >     * <p>If this Phaser is a member of a tiered set of Phasers, then
764 >     * {@code onAdvance} is invoked only for its root Phaser on each
765 >     * advance.
766 >     *
767 >     * <p>The default version returns {@code true} when the number of
768       * registered parties is zero. Normally, overrides that arrange
769       * termination for other reasons should also preserve this
770       * property.
771       *
772       * @param phase the phase number on entering the barrier
773 <     * @param registeredParties the current number of registered
774 <     * parties.
458 <     * @return true if this barrier should terminate
773 >     * @param registeredParties the current number of registered parties
774 >     * @return {@code true} if this barrier should terminate
775       */
776      protected boolean onAdvance(int phase, int registeredParties) {
777          return registeredParties <= 0;
778      }
779  
780      /**
781 <     * Returns a string identifying this barrier, as well as its
781 >     * Returns a string identifying this phaser, as well as its
782       * state.  The state, in brackets, includes the String {@code
783 <     * "phase ="} followed by the phase number, {@code "parties ="}
783 >     * "phase = "} followed by the phase number, {@code "parties = "}
784       * followed by the number of registered parties, and {@code
785 <     * "arrived ="} followed by the number of arrived parties
785 >     * "arrived = "} followed by the number of arrived parties.
786       *
787       * @return a string identifying this barrier, as well as its state
788       */
789      public String toString() {
790 <        long s = state.get();
475 <        return super.toString() + "[phase = " + phaseOf(s) + " parties = " + partiesOf(s) + " arrived = " + arrivedOf(s) + "]";
790 >        return stateToString(reconcileState());
791      }
792  
478    // methods for tripping and waiting
479
793      /**
794 <     * Advance the current phase (or terminate)
794 >     * Implementation of toString and string-based error messages
795       */
796 <    private void trip(int phase, int parties) {
797 <        int next = onAdvance(phase, parties)? -1 : ((phase + 1) & phaseMask);
798 <        state.set(stateFor(next, parties, parties));
799 <        if (head.get() != null)
800 <            releaseWaiters(next);
796 >    private String stateToString(long s) {
797 >        return super.toString() +
798 >            "[phase = " + phaseOf(s) +
799 >            " parties = " + partiesOf(s) +
800 >            " arrived = " + arrivedOf(s) + "]";
801      }
802  
803 <    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 <    }
803 >    // Waiting mechanics
804  
805 <    private int timedHelpingWait(int phase, long nanos) throws TimeoutException {
806 <        final AtomicLong state = this.state;
807 <        long lastTime = System.nanoTime();
805 >    /**
806 >     * Removes and signals threads from queue for phase
807 >     */
808 >    private void releaseWaiters(int phase) {
809 >        AtomicReference<QNode> head = queueFor(phase);
810 >        QNode q;
811          int p;
812 <        while ((p = phaseOf(state.get())) == phase) {
813 <            long now = System.nanoTime();
814 <            nanos -= now - lastTime;
815 <            lastTime = now;
816 <            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 <            }
812 >        while ((q = head.get()) != null &&
813 >               ((p = q.phase) == phase ||
814 >                (int)(root.state >>> PHASE_SHIFT) != p)) {
815 >            if (head.compareAndSet(q, q.next))
816 >                q.signal();
817          }
531        return p;
818      }
819  
820      /**
821 <     * Wait nodes for Treiber stack representing wait queue for non-FJ
822 <     * tasks. The waiting scheme is an adaptation of the one used in
823 <     * forkjoin.PoolBarrier.
821 >     * Tries to enqueue given node in the appropriate wait queue.
822 >     *
823 >     * @return true if successful
824       */
825 <    static final class QNode {
826 <        QNode next;
827 <        volatile Thread thread; // nulled to cancel wait
828 <        final int phase;
829 <        QNode(Thread t, int c) {
830 <            thread = t;
831 <            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 <        }
825 >    private boolean tryEnqueue(int phase, QNode node) {
826 >        releaseWaiters(phase-1); // ensure old queue clean
827 >        AtomicReference<QNode> head = queueFor(phase);
828 >        QNode q = head.get();
829 >        return ((q == null || q.phase == phase) &&
830 >                (int)(root.state >>> PHASE_SHIFT) == phase &&
831 >                head.compareAndSet(node.next = q, node));
832      }
833  
834      /** The number of CPUs, for spin control */
835 <    static final int NCPUS = Runtime.getRuntime().availableProcessors();
835 >    private static final int NCPU = Runtime.getRuntime().availableProcessors();
836  
837      /**
838 <     * The number of times to spin before blocking in timed waits.
839 <     * The value is empirically derived.
838 >     * The number of times to spin before blocking while waiting for
839 >     * advance, per arrival while waiting. On multiprocessors, fully
840 >     * blocking and waking up a large number of threads all at once is
841 >     * usually a very slow process, so we use rechargeable spins to
842 >     * avoid it when threads regularly arrive: When a thread in
843 >     * internalAwaitAdvance notices another arrival before blocking,
844 >     * and there appear to be enough CPUs available, it spins
845 >     * SPINS_PER_ARRIVAL more times before blocking. Plus, even on
846 >     * uniprocessors, there is at least one intervening Thread.yield
847 >     * before blocking. The value trades off good-citizenship vs big
848 >     * unnecessary slowdowns.
849       */
850 <    static final int maxTimedSpins = (NCPUS < 2)? 0 : 32;
850 >    static final int SPINS_PER_ARRIVAL = (NCPU < 2) ? 1 : 1 << 8;
851  
852      /**
853 <     * The number of times to spin before blocking in untimed waits.
854 <     * This is greater than timed value because untimed waits spin
855 <     * faster since they don't need to check times on each spin.
856 <     */
857 <    static final int maxUntimedSpins = maxTimedSpins * 32;
853 >     * Possibly blocks and waits for phase to advance unless aborted.
854 >     *
855 >     * @param phase current phase
856 >     * @param node if non-null, the wait node to track interrupt and timeout;
857 >     * if null, denotes noninterruptible wait
858 >     * @return current phase
859 >     */
860 >    private int internalAwaitAdvance(int phase, QNode node) {
861 >        Phaser current = this;       // to eventually wait at root if tiered
862 >        boolean queued = false;      // true when node is enqueued
863 >        int lastUnarrived = -1;      // to increase spins upon change
864 >        int spins = SPINS_PER_ARRIVAL;
865 >        for (;;) {
866 >            int p, unarrived;
867 >            Phaser par;
868 >            long s = current.state;
869 >            if ((p = (int)(s >>> PHASE_SHIFT)) != phase) {
870 >                if (node != null)
871 >                    node.onRelease();
872 >                releaseWaiters(phase);
873 >                return p;
874 >            }
875 >            else if ((unarrived = ((int)s) & UNARRIVED_MASK) == 0 &&
876 >                     (par = current.parent) != null) {
877 >                current = par;       // if all arrived, use parent
878 >                par = par.parent;
879 >                lastUnarrived = -1;
880 >            }
881 >            else if (unarrived != lastUnarrived) {
882 >                if ((lastUnarrived = unarrived) < NCPU)
883 >                    spins += SPINS_PER_ARRIVAL;
884 >            }
885 >            else if (spins > 0) {
886 >                if (--spins == (SPINS_PER_ARRIVAL >>> 1))
887 >                    Thread.yield();  // yield midway through spin
888 >            }
889 >            else if (node == null)   // must be noninterruptible
890 >                node = new QNode(this, phase, false, false, 0L);
891 >            else if (node.isReleasable()) {
892 >                if ((int)(reconcileState() >>> PHASE_SHIFT) == phase)
893 >                    return phase;    // aborted
894 >            }
895 >            else if (!queued)
896 >                queued = tryEnqueue(phase, node);
897 >            else {
898 >                try {
899 >                    ForkJoinPool.managedBlock(node);
900 >                } catch (InterruptedException ie) {
901 >                    node.wasInterrupted = true;
902 >                }
903 >            }
904 >        }
905 >    }
906  
907      /**
908 <     * The number of nanoseconds for which it is faster to spin
583 <     * rather than to use timed park. A rough estimate suffices.
908 >     * Wait nodes for Treiber stack representing wait queue
909       */
910 <    static final long spinForTimeoutThreshold = 1000L;
910 >    static final class QNode implements ForkJoinPool.ManagedBlocker {
911 >        final Phaser phaser;
912 >        final int phase;
913 >        final boolean interruptible;
914 >        final boolean timed;
915 >        boolean wasInterrupted;
916 >        long nanos;
917 >        long lastTime;
918 >        volatile Thread thread; // nulled to cancel wait
919 >        QNode next;
920  
921 <    /**
922 <     * Enqueues node and waits unless aborted or signalled.
923 <     */
924 <    private boolean untimedWait(Thread thread, int currentPhase,
925 <                               boolean abortOnInterrupt) {
926 <        final AtomicReference<QNode> head = this.head;
927 <        final AtomicLong state = this.state;
928 <        boolean wasInterrupted = false;
929 <        QNode node = null;
930 <        boolean queued = false;
931 <        int spins = maxUntimedSpins;
932 <        while (phaseOf(state.get()) == currentPhase) {
933 <            QNode h;
934 <            if (node != null && queued) {
935 <                if (node.thread != null) {
936 <                    LockSupport.park();
937 <                    if (Thread.interrupted()) {
921 >        QNode(Phaser phaser, int phase, boolean interruptible,
922 >              boolean timed, long nanos) {
923 >            this.phaser = phaser;
924 >            this.phase = phase;
925 >            this.interruptible = interruptible;
926 >            this.nanos = nanos;
927 >            this.timed = timed;
928 >            this.lastTime = timed? System.nanoTime() : 0L;
929 >            thread = Thread.currentThread();
930 >        }
931 >
932 >        public boolean isReleasable() {
933 >            Thread t = thread;
934 >            if (t != null) {
935 >                if (phaser.getPhase() != phase)
936 >                    t = null;
937 >                else {
938 >                    if (Thread.interrupted())
939                          wasInterrupted = true;
940 <                        if (abortOnInterrupt)
941 <                            break;
942 <                    }
943 <                }
944 <            }
945 <            else if ((h = head.get()) != null && h.phase != currentPhase) {
946 <                if (phaseOf(state.get()) == currentPhase) { // must recheck
612 <                    if (head.compareAndSet(h, h.next)) {
613 <                        Thread t = h.thread; // help clear out old waiters
614 <                        if (t != null) {
615 <                            h.thread = null;
616 <                            LockSupport.unpark(t);
940 >                    if (interruptible && wasInterrupted)
941 >                        t = null;
942 >                    else if (timed) {
943 >                        if (nanos > 0) {
944 >                            long now = System.nanoTime();
945 >                            nanos -= now - lastTime;
946 >                            lastTime = now;
947                          }
948 +                        if (nanos <= 0)
949 +                            t = null;
950                      }
951                  }
952 <                else
953 <                    break;
952 >                if (t != null)
953 >                    return false;
954 >                thread = null;
955 >            }
956 >            return true;
957 >        }
958 >
959 >        public boolean block() {
960 >            if (isReleasable())
961 >                return true;
962 >            else if (!timed)
963 >                LockSupport.park(this);
964 >            else if (nanos > 0)
965 >                LockSupport.parkNanos(this, nanos);
966 >            return isReleasable();
967 >        }
968 >
969 >        void signal() {
970 >            Thread t = thread;
971 >            if (t != null) {
972 >                thread = null;
973 >                LockSupport.unpark(t);
974              }
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;
975          }
976 <        if (node != null)
977 <            node.thread = null;
978 <        return wasInterrupted;
976 >
977 >        void onRelease() { // actions upon return from internalAwaitAdvance
978 >            if (!interruptible && wasInterrupted)
979 >                Thread.currentThread().interrupt();
980 >            if (thread != null)
981 >                thread = null;
982 >        }
983 >
984 >    }
985 >
986 >    // Unsafe mechanics
987 >
988 >    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
989 >    private static final long stateOffset =
990 >        objectFieldOffset("state", Phaser.class);
991 >
992 >    private static long objectFieldOffset(String field, Class<?> klazz) {
993 >        try {
994 >            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
995 >        } catch (NoSuchFieldException e) {
996 >            // Convert Exception to corresponding Error
997 >            NoSuchFieldError error = new NoSuchFieldError(field);
998 >            error.initCause(e);
999 >            throw error;
1000 >        }
1001      }
1002  
1003      /**
1004 <     * Messier timeout version
1004 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1005 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
1006 >     * into a jdk.
1007 >     *
1008 >     * @return a sun.misc.Unsafe
1009       */
1010 <    private void timedWait(Thread thread, int currentPhase, long nanos)
1011 <        throws InterruptedException, TimeoutException {
1012 <        final AtomicReference<QNode> head = this.head;
1013 <        final AtomicLong state = this.state;
1014 <        long lastTime = System.nanoTime();
1015 <        QNode node = null;
1016 <        boolean queued = false;
1017 <        int spins = maxTimedSpins;
1018 <        while (phaseOf(state.get()) == currentPhase) {
1019 <            QNode h;
1020 <            long now = System.nanoTime();
1021 <            nanos -= now - lastTime;
1022 <            lastTime = now;
1023 <            if (nanos <= 0) {
1024 <                if (node != null)
1025 <                    node.thread = null;
1026 <                if (phaseOf(state.get()) == currentPhase)
655 <                    throw new TimeoutException();
656 <                else
657 <                    break;
658 <            }
659 <            else if (node != null && queued) {
660 <                if (node.thread != null &&
661 <                    nanos > spinForTimeoutThreshold) {
662 <                    //                LockSupport.parkNanos(this, nanos);
663 <                    LockSupport.parkNanos(nanos);
664 <                    if (Thread.interrupted()) {
665 <                        node.thread = null;
666 <                        throw new InterruptedException();
667 <                    }
668 <                }
1010 >    private static sun.misc.Unsafe getUnsafe() {
1011 >        try {
1012 >            return sun.misc.Unsafe.getUnsafe();
1013 >        } catch (SecurityException se) {
1014 >            try {
1015 >                return java.security.AccessController.doPrivileged
1016 >                    (new java.security
1017 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1018 >                        public sun.misc.Unsafe run() throws Exception {
1019 >                            java.lang.reflect.Field f = sun.misc
1020 >                                .Unsafe.class.getDeclaredField("theUnsafe");
1021 >                            f.setAccessible(true);
1022 >                            return (sun.misc.Unsafe) f.get(null);
1023 >                        }});
1024 >            } catch (java.security.PrivilegedActionException e) {
1025 >                throw new RuntimeException("Could not initialize intrinsics",
1026 >                                           e.getCause());
1027              }
670            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                        }
678                    }
679                }
680                else
681                    break;
682            }
683            else if (node != null)
684                queued = head.compareAndSet(node.next = h, node);
685            else if (spins <= 0)
686                node = new QNode(thread, currentPhase);
687            else
688                --spins;
1028          }
690        if (node != null)
691            node.thread = null;
1029      }
693
1030   }
695

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines