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 1.20 by jsr166, Sat Jul 25 00:34:00 2009 UTC vs.
Revision 1.73 by jsr166, Wed May 25 16:08:03 2011 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  
9 < import java.util.concurrent.*;
10 <
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
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>
21 < *
22 < * <li> The number of parties synchronizing on a phaser may vary over
23 < * time.  A task may register to be a party at any time, and may
24 < * deregister upon arriving at the barrier.  As is the case with most
25 < * basic synchronization constructs, registration and deregistration
26 < * affect only internal counts; they do not establish any further
27 < * internal bookkeeping, so tasks cannot query whether they are
28 < * registered. (However, you can introduce such bookkeeping by
29 < * subclassing this class.)
30 < *
31 < * <li> Each generation has an associated phase value, starting at
32 < * zero, and advancing when all parties reach the barrier (wrapping
33 < * around to zero after reaching {@code Integer.MAX_VALUE}).
34 < *
35 < * <li> Like a CyclicBarrier, a Phaser may be repeatedly awaited.
36 < * Method {@code arriveAndAwaitAdvance} has effect analogous to
37 < * {@code CyclicBarrier.await}.  However, Phasers separate two
38 < * aspects of coordination, that may also be invoked independently:
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 > * <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 {@code arrive} and
48 < *       {@code arriveAndDeregister} do not block, but return
49 < *       the phase value current upon entry to the method.
50 < *
51 < *   <li> Awaiting others. Method {@code awaitAdvance} requires an
52 < *       argument indicating the entry phase, and returns when the
53 < *       barrier advances to a new phase.
54 < * </ul>
55 < *
56 < *
57 < * <li> Barrier actions, performed by the task triggering a phase
58 < * advance while others may be waiting, are arranged by overriding
59 < * method {@code onAdvance}, that also controls termination.
60 < * Overriding this method may be used to similar but more flexible
61 < * effect as providing a barrier action to a CyclicBarrier.
62 < *
63 < * <li> Phasers may enter a <em>termination</em> state in which all
64 < * actions immediately return without updating phaser state or waiting
65 < * for advance, and indicating (via a negative phase value) that
66 < * execution is complete.  Termination is triggered by executing the
67 < * overridable {@code onAdvance} method that is invoked each time the
68 < * barrier is about to be tripped. When a Phaser is controlling an
69 < * action with a fixed number of iterations, it is often convenient to
70 < * override this method to cause termination when the current phase
71 < * number reaches a threshold. Method {@code forceTermination} is also
72 < * available to abruptly release waiting threads and allow them to
73 < * terminate.
69 < *
70 < * <li> Phasers may be tiered to reduce contention. Phasers with large
71 < * numbers of parties that would otherwise experience heavy
72 < * synchronization contention costs may instead be arranged in trees.
73 < * This will typically greatly increase throughput even though it
74 < * incurs somewhat greater per-operation overhead.
75 < *
76 < * <li> By default, {@code awaitAdvance} continues to wait even if
77 < * the waiting thread is interrupted. And unlike the case in
78 < * CyclicBarriers, exceptions encountered while tasks wait
79 < * interruptibly or with timeout do not change the state of the
80 < * barrier. If necessary, you can perform any associated recovery
81 < * within handlers of those exceptions, often after invoking
82 < * {@code forceTermination}.
83 < *
84 < * <li>Phasers ensure lack of starvation when used by ForkJoinTasks.
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 + * <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 + * <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 usages:</b>
123   *
124 < * <p>A Phaser may be used instead of a {@code CountDownLatch} to control
125 < * a one-shot action serving a variable number of parties. The typical
126 < * idiom is for the method setting this up to first register, then
127 < * start the actions, then deregister, as in:
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   *  <pre> {@code
130 < * void runTasks(List<Runnable> list) {
130 > * void runTasks(List<Runnable> tasks) {
131   *   final Phaser phaser = new Phaser(1); // "1" to register self
132 < *   for (Runnable r : list) {
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 < *         r.run();
104 < *         phaser.arriveAndDeregister();   // signal completion
138 > *         task.run();
139   *       }
140   *     }.start();
141   *   }
142   *
143 < *   doSomethingOnBehalfOfWorkers();
144 < *   phaser.arrive(); // allow threads to start
111 < *   int p = phaser.arriveAndDeregister(); // deregister self  ...
112 < *   p = phaser.awaitAdvance(p); // ... and await arrival
113 < *   otherActions(); // do other things while tasks execute
114 < *   phaser.awaitAdvance(p); // await final completion
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> list, final int iterations) {
151 > * void startTasks(List<Runnable> tasks, final int iterations) {
152   *   final Phaser phaser = new Phaser() {
153 < *     public boolean onAdvance(int phase, int registeredParties) {
153 > *     protected boolean onAdvance(int phase, int registeredParties) {
154   *       return phase >= iterations || registeredParties == 0;
155   *     }
156   *   };
157   *   phaser.register();
158 < *   for (Runnable r : list) {
158 > *   for (final Runnable task : tasks) {
159   *     phaser.register();
160   *     new Thread() {
161   *       public void run() {
162   *         do {
163 < *           r.run();
163 > *           task.run();
164   *           phaser.arriveAndAwaitAdvance();
165 < *         } while(!phaser.isTerminated();
165 > *         } while (!phaser.isTerminated());
166   *       }
167   *     }.start();
168   *   }
169   *   phaser.arriveAndDeregister(); // deregister self, don't wait
170   * }}</pre>
171   *
172 < * <p> To create a set of tasks using a tree of Phasers,
173 < * you could use code of the following form, assuming a
174 < * Task class with a constructor accepting a Phaser that
175 < * it registers for upon construction:
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 > *
197 > * <p>To create a set of {@code n} tasks using a tree of phasers, you
198 > * could use code of the following form, assuming a Task class with a
199 > * constructor accepting a {@code Phaser} that it registers with upon
200 > * construction. After invocation of {@code build(new Task[n], 0, n,
201 > * new Phaser())}, these tasks could then be started, for example by
202 > * submitting to a pool:
203 > *
204   *  <pre> {@code
205 < * void build(Task[] actions, int lo, int hi, Phaser b) {
206 < *   int step = (hi - lo) / TASKS_PER_PHASER;
207 < *   if (step > 1) {
208 < *     int i = lo;
209 < *     while (i < hi) {
152 < *       int r = Math.min(i + step, hi);
153 < *       build(actions, i, r, new Phaser(b));
154 < *       i = r;
205 > * void build(Task[] tasks, int lo, int hi, Phaser ph) {
206 > *   if (hi - lo > TASKS_PER_PHASER) {
207 > *     for (int i = lo; i < hi; i += TASKS_PER_PHASER) {
208 > *       int j = Math.min(i + TASKS_PER_PHASER, hi);
209 > *       build(tasks, i, j, new Phaser(ph));
210   *     }
211   *   } else {
212   *     for (int i = lo; i < hi; ++i)
213 < *       actions[i] = new Task(b);
214 < *       // assumes new Task(b) performs b.register()
213 > *       tasks[i] = new Task(ph);
214 > *       // assumes new Task(ph) performs ph.register()
215   *   }
216 < * }
162 < * // .. initially called, for n tasks via
163 < * build(new Task[n], 0, n, new Phaser());}</pre>
216 > * }}</pre>
217   *
218   * The best value of {@code TASKS_PER_PHASER} depends mainly on
219 < * expected barrier synchronization rates. A value as low as four may
220 < * be appropriate for extremely small per-barrier task bodies (thus
219 > * expected synchronization rates. A value as low as four may
220 > * be appropriate for extremely small per-phase task bodies (thus
221   * high rates), or up to hundreds for extremely large ones.
222   *
170 * </pre>
171 *
223   * <p><b>Implementation notes</b>: This implementation restricts the
224   * maximum number of parties to 65535. Attempts to register additional
225 < * parties result in IllegalStateExceptions. However, you can and
225 > * parties result in {@code IllegalStateException}. However, you can and
226   * should create tiered phasers to accommodate arbitrarily large sets
227   * of participants.
228   *
# Line 186 | Line 237 | public class Phaser {
237       */
238  
239      /**
240 <     * Barrier state representation. Conceptually, a barrier contains
190 <     * four values:
240 >     * Primary state representation, holding four bit-fields:
241       *
242 <     * * parties -- the number of parties to wait (16 bits)
243 <     * * unarrived -- the number of parties yet to hit barrier (16 bits)
244 <     * * phase -- the generation of the barrier (31 bits)
245 <     * * terminated -- set if barrier is terminated (1 bit)
246 <     *
247 <     * However, to efficiently maintain atomicity, these values are
248 <     * packed into a single (atomic) long. Termination uses the sign
249 <     * bit of 32 bit representation of phase, so phase is set to -1 on
250 <     * termination. Good performance relies on keeping state decoding
251 <     * and encoding simple, and keeping race windows short.
252 <     *
253 <     * Note: there are some cheats in arrive() that rely on unarrived
254 <     * count being lowest 16 bits.
242 >     * unarrived  -- the number of parties yet to hit barrier (bits  0-15)
243 >     * parties    -- the number of parties to wait            (bits 16-31)
244 >     * phase      -- the generation of the barrier            (bits 32-62)
245 >     * terminated -- set if barrier is terminated             (bit  63 / sign)
246 >     *
247 >     * Except that a phaser with no registered parties is
248 >     * distinguished by the otherwise illegal state of having zero
249 >     * parties and one unarrived parties (encoded as EMPTY below).
250 >     *
251 >     * To efficiently maintain atomicity, these values are packed into
252 >     * a single (atomic) long. Good performance relies on keeping
253 >     * state decoding and encoding simple, and keeping race windows
254 >     * short.
255 >     *
256 >     * All state updates are performed via CAS except initial
257 >     * registration of a sub-phaser (i.e., one with a non-null
258 >     * parent).  In this (relatively rare) case, we use built-in
259 >     * synchronization to lock while first registering with its
260 >     * parent.
261 >     *
262 >     * The phase of a subphaser is allowed to lag that of its
263 >     * ancestors until it is actually accessed -- see method
264 >     * reconcileState.
265       */
266      private volatile long state;
267  
268 <    private static final int ushortBits = 16;
269 <    private static final int ushortMask = 0xffff;
270 <    private static final int phaseMask  = 0x7fffffff;
268 >    private static final int  MAX_PARTIES     = 0xffff;
269 >    private static final int  MAX_PHASE       = Integer.MAX_VALUE;
270 >    private static final int  PARTIES_SHIFT   = 16;
271 >    private static final int  PHASE_SHIFT     = 32;
272 >    private static final int  UNARRIVED_MASK  = 0xffff;      // to mask ints
273 >    private static final long PARTIES_MASK    = 0xffff0000L; // to mask longs
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  EMPTY           = 1;
280 >
281 >    // The following unpacking methods are usually manually inlined
282  
283      private static int unarrivedOf(long s) {
284 <        return (int) (s & ushortMask);
284 >        int counts = (int)s;
285 >        return (counts == EMPTY) ? 0 : counts & UNARRIVED_MASK;
286      }
287  
288      private static int partiesOf(long s) {
289 <        return ((int) s) >>> 16;
289 >        return (int)s >>> PARTIES_SHIFT;
290      }
291  
292      private static int phaseOf(long s) {
293 <        return (int) (s >>> 32);
293 >        return (int)(s >>> PHASE_SHIFT);
294      }
295  
296      private static int arrivedOf(long s) {
297 <        return partiesOf(s) - unarrivedOf(s);
298 <    }
299 <
228 <    private static long stateFor(int phase, int parties, int unarrived) {
229 <        return ((((long) phase) << 32) | (((long) parties) << 16) |
230 <                (long) unarrived);
231 <    }
232 <
233 <    private static long trippedStateFor(int phase, int parties) {
234 <        long lp = (long) parties;
235 <        return (((long) phase) << 32) | (lp << 16) | lp;
236 <    }
237 <
238 <    /**
239 <     * Returns message string for bad bounds exceptions.
240 <     */
241 <    private static String badBounds(int parties, int unarrived) {
242 <        return ("Attempt to set " + unarrived +
243 <                " unarrived of " + parties + " parties");
297 >        int counts = (int)s;
298 >        return (counts == EMPTY) ? 0 :
299 >            (counts >>> PARTIES_SHIFT) - (counts & UNARRIVED_MASK);
300      }
301  
302      /**
# Line 249 | Line 305 | public class Phaser {
305      private final Phaser parent;
306  
307      /**
308 <     * The root of Phaser tree. Equals this if not in a tree.  Used to
253 <     * support faster state push-down.
308 >     * The root of phaser tree. Equals this if not in a tree.
309       */
310      private final Phaser root;
311  
257    // Wait queues
258
312      /**
313       * Heads of Treiber stacks for waiting threads. To eliminate
314 <     * contention while releasing some threads while adding others, we
314 >     * contention when releasing some threads while adding others, we
315       * use two of them, alternating across even and odd phases.
316 +     * Subphasers share queues with root to speed up releases.
317       */
318 <    private final AtomicReference<QNode> evenQ = new AtomicReference<QNode>();
319 <    private final AtomicReference<QNode> oddQ  = new AtomicReference<QNode>();
318 >    private final AtomicReference<QNode> evenQ;
319 >    private final AtomicReference<QNode> oddQ;
320  
321      private AtomicReference<QNode> queueFor(int phase) {
322          return ((phase & 1) == 0) ? evenQ : oddQ;
323      }
324  
325      /**
326 <     * Returns current state, first resolving lagged propagation from
273 <     * root if necessary.
326 >     * Returns message string for bounds exceptions on arrival.
327       */
328 <    private long getReconciledState() {
329 <        return (parent == null) ? state : reconcileState();
328 >    private String badArrive(long s) {
329 >        return "Attempted arrival of unregistered party for " +
330 >            stateToString(s);
331      }
332  
333      /**
334 <     * Recursively resolves state.
334 >     * Returns message string for bounds exceptions on registration.
335       */
336 <    private long reconcileState() {
337 <        Phaser p = parent;
338 <        long s = state;
339 <        if (p != null) {
340 <            while (unarrivedOf(s) == 0 && phaseOf(s) != phaseOf(root.state)) {
341 <                long parentState = p.getReconciledState();
342 <                int parentPhase = phaseOf(parentState);
343 <                int phase = phaseOf(s = state);
344 <                if (phase != parentPhase) {
345 <                    long next = trippedStateFor(parentPhase, partiesOf(s));
346 <                    if (casState(s, next)) {
347 <                        releaseWaiters(phase);
348 <                        s = next;
336 >    private String badRegister(long s) {
337 >        return "Attempt to register more than " +
338 >            MAX_PARTIES + " parties for " + stateToString(s);
339 >    }
340 >
341 >    /**
342 >     * Main implementation for methods arrive and arriveAndDeregister.
343 >     * Manually tuned to speed up and minimize race windows for the
344 >     * common case of just decrementing unarrived field.
345 >     *
346 >     * @param deregister false for arrive, true for arriveAndDeregister
347 >     */
348 >    private int doArrive(boolean deregister) {
349 >        int adj = deregister ? ONE_ARRIVAL|ONE_PARTY : ONE_ARRIVAL;
350 >        final Phaser root = this.root;
351 >        for (;;) {
352 >            long s = (root == this) ? state : reconcileState();
353 >            int phase = (int)(s >>> PHASE_SHIFT);
354 >            int counts = (int)s;
355 >            int unarrived = (counts & UNARRIVED_MASK) - 1;
356 >            if (phase < 0)
357 >                return phase;
358 >            else if (counts == EMPTY || unarrived < 0) {
359 >                if (root == this || reconcileState() == s)
360 >                    throw new IllegalStateException(badArrive(s));
361 >            }
362 >            else if (UNSAFE.compareAndSwapLong(this, stateOffset, s, s-=adj)) {
363 >                if (unarrived == 0) {
364 >                    long n = s & PARTIES_MASK;  // base of next state
365 >                    int nextUnarrived = (int)n >>> PARTIES_SHIFT;
366 >                    if (root != this)
367 >                        return parent.doArrive(nextUnarrived == 0);
368 >                    if (onAdvance(phase, nextUnarrived))
369 >                        n |= TERMINATION_BIT;
370 >                    else if (nextUnarrived == 0)
371 >                        n |= EMPTY;
372 >                    else
373 >                        n |= nextUnarrived;
374 >                    n |= (long)((phase + 1) & MAX_PHASE) << PHASE_SHIFT;
375 >                    UNSAFE.compareAndSwapLong(this, stateOffset, s, n);
376 >                    releaseWaiters(phase);
377 >                }
378 >                return phase;
379 >            }
380 >        }
381 >    }
382 >
383 >    /**
384 >     * Implementation of register, bulkRegister
385 >     *
386 >     * @param registrations number to add to both parties and
387 >     * unarrived fields. Must be greater than zero.
388 >     */
389 >    private int doRegister(int registrations) {
390 >        // adjustment to state
391 >        long adj = ((long)registrations << PARTIES_SHIFT) | registrations;
392 >        final Phaser parent = this.parent;
393 >        int phase;
394 >        for (;;) {
395 >            long s = state;
396 >            int counts = (int)s;
397 >            int parties = counts >>> PARTIES_SHIFT;
398 >            int unarrived = counts & UNARRIVED_MASK;
399 >            if (registrations > MAX_PARTIES - parties)
400 >                throw new IllegalStateException(badRegister(s));
401 >            else if ((phase = (int)(s >>> PHASE_SHIFT)) < 0)
402 >                break;
403 >            else if (counts != EMPTY) {             // not 1st registration
404 >                if (parent == null || reconcileState() == s) {
405 >                    if (unarrived == 0)             // wait out advance
406 >                        root.internalAwaitAdvance(phase, null);
407 >                    else if (UNSAFE.compareAndSwapLong(this, stateOffset,
408 >                                                       s, s + adj))
409 >                        break;
410 >                }
411 >            }
412 >            else if (parent == null) {              // 1st root registration
413 >                long next = ((long)phase << PHASE_SHIFT) | adj;
414 >                if (UNSAFE.compareAndSwapLong(this, stateOffset, s, next))
415 >                    break;
416 >            }
417 >            else {
418 >                synchronized (this) {               // 1st sub registration
419 >                    if (state == s) {               // recheck under lock
420 >                        parent.doRegister(1);
421 >                        do {                        // force current phase
422 >                            phase = (int)(root.state >>> PHASE_SHIFT);
423 >                            // assert phase < 0 || (int)state == EMPTY;
424 >                        } while (!UNSAFE.compareAndSwapLong
425 >                                 (this, stateOffset, state,
426 >                                  ((long)phase << PHASE_SHIFT) | adj));
427 >                        break;
428                      }
429                  }
430              }
431          }
432 +        return phase;
433 +    }
434 +
435 +    /**
436 +     * Resolves lagged phase propagation from root if necessary.
437 +     * Reconciliation normally occurs when root has advanced but
438 +     * subphasers have not yet done so, in which case they must finish
439 +     * their own advance by setting unarrived to parties (or if
440 +     * parties is zero, resetting to unregistered EMPTY state).
441 +     * However, this method may also be called when "floating"
442 +     * subphasers with possibly some unarrived parties are merely
443 +     * catching up to current phase, in which case counts are
444 +     * unaffected.
445 +     *
446 +     * @return reconciled state
447 +     */
448 +    private long reconcileState() {
449 +        final Phaser root = this.root;
450 +        long s = state;
451 +        if (root != this) {
452 +            int phase, u, p;
453 +            // CAS root phase with current parties; possibly trip unarrived
454 +            while ((phase = (int)(root.state >>> PHASE_SHIFT)) !=
455 +                   (int)(s >>> PHASE_SHIFT) &&
456 +                   !UNSAFE.compareAndSwapLong
457 +                   (this, stateOffset, s,
458 +                    s = (((long)phase << PHASE_SHIFT) |
459 +                         (s & PARTIES_MASK) |
460 +                         ((p = (int)s >>> PARTIES_SHIFT) == 0 ? EMPTY :
461 +                          (u = (int)s & UNARRIVED_MASK) == 0 ? p : u))))
462 +                s = state;
463 +        }
464          return s;
465      }
466  
467      /**
468 <     * Creates a new Phaser without any initially registered parties,
469 <     * initial phase number 0, and no parent. Any thread using this
470 <     * Phaser will need to first register for it.
468 >     * Creates a new phaser with no initially registered parties, no
469 >     * parent, and initial phase number 0. Any thread using this
470 >     * phaser will need to first register for it.
471       */
472      public Phaser() {
473 <        this(null);
473 >        this(null, 0);
474      }
475  
476      /**
477 <     * Creates a new Phaser with the given numbers of registered
478 <     * unarrived parties, initial phase number 0, and no parent.
477 >     * Creates a new phaser with the given number of registered
478 >     * unarrived parties, no parent, and initial phase number 0.
479       *
480 <     * @param parties the number of parties required to trip barrier
480 >     * @param parties the number of parties required to advance to the
481 >     * next phase
482       * @throws IllegalArgumentException if parties less than zero
483       * or greater than the maximum number of parties supported
484       */
# Line 321 | Line 487 | public class Phaser {
487      }
488  
489      /**
490 <     * Creates a new Phaser with the given parent, without any
325 <     * initially registered parties. If parent is non-null this phaser
326 <     * is registered with the parent and its initial phase number is
327 <     * the same as that of parent phaser.
490 >     * Equivalent to {@link #Phaser(Phaser, int) Phaser(parent, 0)}.
491       *
492       * @param parent the parent phaser
493       */
494      public Phaser(Phaser parent) {
495 <        int phase = 0;
333 <        this.parent = parent;
334 <        if (parent != null) {
335 <            this.root = parent.root;
336 <            phase = parent.register();
337 <        }
338 <        else
339 <            this.root = this;
340 <        this.state = trippedStateFor(phase, 0);
495 >        this(parent, 0);
496      }
497  
498      /**
499 <     * Creates a new Phaser with the given parent and numbers of
500 <     * registered unarrived parties. If parent is non-null, this phaser
501 <     * is registered with the parent and its initial phase number is
502 <     * the same as that of parent phaser.
499 >     * Creates a new phaser with the given parent and number of
500 >     * registered unarrived parties.  When the given parent is non-null
501 >     * and the given number of parties is greater than zero, this
502 >     * child phaser is registered with its parent.
503       *
504       * @param parent the parent phaser
505 <     * @param parties the number of parties required to trip barrier
505 >     * @param parties the number of parties required to advance to the
506 >     * next phase
507       * @throws IllegalArgumentException if parties less than zero
508       * or greater than the maximum number of parties supported
509       */
510      public Phaser(Phaser parent, int parties) {
511 <        if (parties < 0 || parties > ushortMask)
511 >        if (parties >>> PARTIES_SHIFT != 0)
512              throw new IllegalArgumentException("Illegal number of parties");
513          int phase = 0;
514          this.parent = parent;
515          if (parent != null) {
516 <            this.root = parent.root;
517 <            phase = parent.register();
516 >            final Phaser root = parent.root;
517 >            this.root = root;
518 >            this.evenQ = root.evenQ;
519 >            this.oddQ = root.oddQ;
520 >            if (parties != 0)
521 >                phase = parent.doRegister(1);
522          }
523 <        else
523 >        else {
524              this.root = this;
525 <        this.state = trippedStateFor(phase, parties);
525 >            this.evenQ = new AtomicReference<QNode>();
526 >            this.oddQ = new AtomicReference<QNode>();
527 >        }
528 >        this.state = (parties == 0) ? (long)EMPTY :
529 >            ((long)phase << PHASE_SHIFT) |
530 >            ((long)parties << PARTIES_SHIFT) |
531 >            ((long)parties);
532      }
533  
534      /**
535 <     * Adds a new unarrived party to this phaser.
536 <     *
537 <     * @return the current barrier phase number upon registration
535 >     * Adds a new unarrived party to this phaser.  If an ongoing
536 >     * invocation of {@link #onAdvance} is in progress, this method
537 >     * may await its completion before returning.  If this phaser has
538 >     * a parent, and this phaser previously had no registered parties,
539 >     * this child phaser is also registered with its parent. If
540 >     * this phaser is terminated, the attempt to register has
541 >     * no effect, and a negative value is returned.
542 >     *
543 >     * @return the arrival phase number to which this registration
544 >     * applied.  If this value is negative, then this phaser has
545 >     * terminated, in which case registration has no effect.
546       * @throws IllegalStateException if attempting to register more
547       * than the maximum supported number of parties
548       */
# Line 378 | Line 552 | public class Phaser {
552  
553      /**
554       * Adds the given number of new unarrived parties to this phaser.
555 <     *
556 <     * @param parties the number of parties required to trip barrier
557 <     * @return the current barrier phase number upon registration
555 >     * If an ongoing invocation of {@link #onAdvance} is in progress,
556 >     * this method may await its completion before returning.  If this
557 >     * phaser has a parent, and the given number of parties is greater
558 >     * than zero, and this phaser previously had no registered
559 >     * parties, this child phaser is also registered with its parent.
560 >     * If this phaser is terminated, the attempt to register has no
561 >     * effect, and a negative value is returned.
562 >     *
563 >     * @param parties the number of additional parties required to
564 >     * advance to the next phase
565 >     * @return the arrival phase number to which this registration
566 >     * applied.  If this value is negative, then this phaser has
567 >     * terminated, in which case registration has no effect.
568       * @throws IllegalStateException if attempting to register more
569       * than the maximum supported number of parties
570 +     * @throws IllegalArgumentException if {@code parties < 0}
571       */
572      public int bulkRegister(int parties) {
573          if (parties < 0)
# Line 393 | Line 578 | public class Phaser {
578      }
579  
580      /**
581 <     * Shared code for register, bulkRegister
582 <     */
583 <    private int doRegister(int registrations) {
584 <        int phase;
585 <        for (;;) {
586 <            long s = getReconciledState();
402 <            phase = phaseOf(s);
403 <            int unarrived = unarrivedOf(s) + registrations;
404 <            int parties = partiesOf(s) + registrations;
405 <            if (phase < 0)
406 <                break;
407 <            if (parties > ushortMask || unarrived > ushortMask)
408 <                throw new IllegalStateException(badBounds(parties, unarrived));
409 <            if (phase == phaseOf(root.state) &&
410 <                casState(s, stateFor(phase, parties, unarrived)))
411 <                break;
412 <        }
413 <        return phase;
414 <    }
415 <
416 <    /**
417 <     * Arrives at the barrier, but does not wait for others.  (You can
418 <     * in turn wait for others via {@link #awaitAdvance}).
581 >     * Arrives at this phaser, without waiting for others to arrive.
582 >     *
583 >     * <p>It is a usage error for an unregistered party to invoke this
584 >     * method.  However, this error may result in an {@code
585 >     * IllegalStateException} only upon some subsequent operation on
586 >     * this phaser, if ever.
587       *
588 <     * @return the barrier phase number upon entry to this method, or a
421 <     * negative value if terminated
588 >     * @return the arrival phase number, or a negative value if terminated
589       * @throws IllegalStateException if not terminated and the number
590       * of unarrived parties would become negative
591       */
592      public int arrive() {
593 <        int phase;
427 <        for (;;) {
428 <            long s = state;
429 <            phase = phaseOf(s);
430 <            if (phase < 0)
431 <                break;
432 <            int parties = partiesOf(s);
433 <            int unarrived = unarrivedOf(s) - 1;
434 <            if (unarrived > 0) {        // Not the last arrival
435 <                if (casState(s, s - 1)) // s-1 adds one arrival
436 <                    break;
437 <            }
438 <            else if (unarrived == 0) {  // the last arrival
439 <                Phaser par = parent;
440 <                if (par == null) {      // directly trip
441 <                    if (casState
442 <                        (s,
443 <                         trippedStateFor(onAdvance(phase, parties) ? -1 :
444 <                                         ((phase + 1) & phaseMask), parties))) {
445 <                        releaseWaiters(phase);
446 <                        break;
447 <                    }
448 <                }
449 <                else {                  // cascade to parent
450 <                    if (casState(s, s - 1)) { // zeroes unarrived
451 <                        par.arrive();
452 <                        reconcileState();
453 <                        break;
454 <                    }
455 <                }
456 <            }
457 <            else if (phase != phaseOf(root.state)) // or if unreconciled
458 <                reconcileState();
459 <            else
460 <                throw new IllegalStateException(badBounds(parties, unarrived));
461 <        }
462 <        return phase;
593 >        return doArrive(false);
594      }
595  
596      /**
597 <     * Arrives at the barrier, and deregisters from it, without
598 <     * waiting for others. Deregistration reduces number of parties
599 <     * required to trip the barrier in future phases.  If this phaser
597 >     * Arrives at this phaser and deregisters from it without waiting
598 >     * for others to arrive. Deregistration reduces the number of
599 >     * parties required to advance in future phases.  If this phaser
600       * has a parent, and deregistration causes this phaser to have
601       * zero parties, this phaser is also deregistered from its parent.
602       *
603 <     * @return the current barrier phase number upon entry to
604 <     * this method, or a negative value if terminated
603 >     * <p>It is a usage error for an unregistered party to invoke this
604 >     * method.  However, this error may result in an {@code
605 >     * IllegalStateException} only upon some subsequent operation on
606 >     * this phaser, if ever.
607 >     *
608 >     * @return the arrival phase number, or a negative value if terminated
609       * @throws IllegalStateException if not terminated and the number
610       * of registered or unarrived parties would become negative
611       */
612      public int arriveAndDeregister() {
613 <        // similar code to arrive, but too different to merge
479 <        Phaser par = parent;
480 <        int phase;
481 <        for (;;) {
482 <            long s = state;
483 <            phase = phaseOf(s);
484 <            if (phase < 0)
485 <                break;
486 <            int parties = partiesOf(s) - 1;
487 <            int unarrived = unarrivedOf(s) - 1;
488 <            if (parties >= 0) {
489 <                if (unarrived > 0 || (unarrived == 0 && par != null)) {
490 <                    if (casState
491 <                        (s,
492 <                         stateFor(phase, parties, unarrived))) {
493 <                        if (unarrived == 0) {
494 <                            par.arriveAndDeregister();
495 <                            reconcileState();
496 <                        }
497 <                        break;
498 <                    }
499 <                    continue;
500 <                }
501 <                if (unarrived == 0) {
502 <                    if (casState
503 <                        (s,
504 <                         trippedStateFor(onAdvance(phase, parties) ? -1 :
505 <                                         ((phase + 1) & phaseMask), parties))) {
506 <                        releaseWaiters(phase);
507 <                        break;
508 <                    }
509 <                    continue;
510 <                }
511 <                if (par != null && phase != phaseOf(root.state)) {
512 <                    reconcileState();
513 <                    continue;
514 <                }
515 <            }
516 <            throw new IllegalStateException(badBounds(parties, unarrived));
517 <        }
518 <        return phase;
613 >        return doArrive(true);
614      }
615  
616      /**
617 <     * Arrives at the barrier and awaits others. Equivalent in effect
618 <     * to {@code awaitAdvance(arrive())}.  If you instead need to
619 <     * await with interruption of timeout, and/or deregister upon
620 <     * arrival, you can arrange them using analogous constructions.
617 >     * Arrives at this phaser and awaits others. Equivalent in effect
618 >     * to {@code awaitAdvance(arrive())}.  If you need to await with
619 >     * interruption or timeout, you can arrange this with an analogous
620 >     * construction using one of the other forms of the {@code
621 >     * awaitAdvance} method.  If instead you need to deregister upon
622 >     * arrival, use {@code awaitAdvance(arriveAndDeregister())}.
623       *
624 <     * @return the phase on entry to this method
624 >     * <p>It is a usage error for an unregistered party to invoke this
625 >     * method.  However, this error may result in an {@code
626 >     * IllegalStateException} only upon some subsequent operation on
627 >     * this phaser, if ever.
628 >     *
629 >     * @return the arrival phase number, or the (negative)
630 >     * {@linkplain #getPhase() current phase} if terminated
631       * @throws IllegalStateException if not terminated and the number
632       * of unarrived parties would become negative
633       */
634      public int arriveAndAwaitAdvance() {
635 <        return awaitAdvance(arrive());
635 >        // Specialization of doArrive+awaitAdvance eliminating some reads/paths
636 >        final Phaser root = this.root;
637 >        for (;;) {
638 >            long s = (root == this) ? state : reconcileState();
639 >            int phase = (int)(s >>> PHASE_SHIFT);
640 >            int counts = (int)s;
641 >            int unarrived = (counts & UNARRIVED_MASK) - 1;
642 >            if (phase < 0)
643 >                return phase;
644 >            else if (counts == EMPTY || unarrived < 0) {
645 >                if (reconcileState() == s)
646 >                    throw new IllegalStateException(badArrive(s));
647 >            }
648 >            else if (UNSAFE.compareAndSwapLong(this, stateOffset, s,
649 >                                               s -= ONE_ARRIVAL)) {
650 >                if (unarrived != 0)
651 >                    return root.internalAwaitAdvance(phase, null);
652 >                if (root != this)
653 >                    return parent.arriveAndAwaitAdvance();
654 >                long n = s & PARTIES_MASK;  // base of next state
655 >                int nextUnarrived = (int)n >>> PARTIES_SHIFT;
656 >                if (onAdvance(phase, nextUnarrived))
657 >                    n |= TERMINATION_BIT;
658 >                else if (nextUnarrived == 0)
659 >                    n |= EMPTY;
660 >                else
661 >                    n |= nextUnarrived;
662 >                int nextPhase = (phase + 1) & MAX_PHASE;
663 >                n |= (long)nextPhase << PHASE_SHIFT;
664 >                if (!UNSAFE.compareAndSwapLong(this, stateOffset, s, n))
665 >                    return (int)(state >>> PHASE_SHIFT); // terminated
666 >                releaseWaiters(phase);
667 >                return nextPhase;
668 >            }
669 >        }
670      }
671  
672      /**
673 <     * Awaits the phase of the barrier to advance from the given
674 <     * value, or returns immediately if argument is negative or this
675 <     * barrier is terminated.
676 <     *
677 <     * @param phase the phase on entry to this method
678 <     * @return the phase on exit from this method
673 >     * Awaits the phase of this phaser to advance from the given phase
674 >     * value, returning immediately if the current phase is not equal
675 >     * to the given phase value or this phaser is terminated.
676 >     *
677 >     * @param phase an arrival phase number, or negative value if
678 >     * terminated; this argument is normally the value returned by a
679 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
680 >     * @return the next arrival phase number, or the argument if it is
681 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
682 >     * if terminated
683       */
684      public int awaitAdvance(int phase) {
685 +        final Phaser root = this.root;
686 +        long s = (root == this) ? state : reconcileState();
687 +        int p = (int)(s >>> PHASE_SHIFT);
688          if (phase < 0)
689              return phase;
690 <        long s = getReconciledState();
691 <        int p = phaseOf(s);
692 <        if (p != phase)
549 <            return p;
550 <        if (unarrivedOf(s) == 0 && parent != null)
551 <            parent.awaitAdvance(phase);
552 <        // Fall here even if parent waited, to reconcile and help release
553 <        return untimedWait(phase);
690 >        if (p == phase)
691 >            return root.internalAwaitAdvance(phase, null);
692 >        return p;
693      }
694  
695      /**
696 <     * Awaits the phase of the barrier to advance from the given
697 <     * value, or returns immediately if argument is negative or this
698 <     * barrier is terminated, or throws InterruptedException if
699 <     * interrupted while waiting.
700 <     *
701 <     * @param phase the phase on entry to this method
702 <     * @return the phase on exit from this method
696 >     * Awaits the phase of this phaser to advance from the given phase
697 >     * value, throwing {@code InterruptedException} if interrupted
698 >     * while waiting, or returning immediately if the current phase is
699 >     * not equal to the given phase value or this phaser is
700 >     * terminated.
701 >     *
702 >     * @param phase an arrival phase number, or negative value if
703 >     * terminated; this argument is normally the value returned by a
704 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
705 >     * @return the next arrival phase number, or the argument if it is
706 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
707 >     * if terminated
708       * @throws InterruptedException if thread interrupted while waiting
709       */
710      public int awaitAdvanceInterruptibly(int phase)
711          throws InterruptedException {
712 +        final Phaser root = this.root;
713 +        long s = (root == this) ? state : reconcileState();
714 +        int p = (int)(s >>> PHASE_SHIFT);
715          if (phase < 0)
716              return phase;
717 <        long s = getReconciledState();
718 <        int p = phaseOf(s);
719 <        if (p != phase)
720 <            return p;
721 <        if (unarrivedOf(s) == 0 && parent != null)
722 <            parent.awaitAdvanceInterruptibly(phase);
723 <        return interruptibleWait(phase);
717 >        if (p == phase) {
718 >            QNode node = new QNode(this, phase, true, false, 0L);
719 >            p = root.internalAwaitAdvance(phase, node);
720 >            if (node.wasInterrupted)
721 >                throw new InterruptedException();
722 >        }
723 >        return p;
724      }
725  
726      /**
727 <     * Awaits the phase of the barrier to advance from the given value
728 <     * or the given timeout elapses, or returns immediately if
729 <     * argument is negative or this barrier is terminated.
730 <     *
731 <     * @param phase the phase on entry to this method
732 <     * @return the phase on exit from this method
727 >     * Awaits the phase of this phaser to advance from the given phase
728 >     * value or the given timeout to elapse, throwing {@code
729 >     * InterruptedException} if interrupted while waiting, or
730 >     * returning immediately if the current phase is not equal to the
731 >     * given phase value or this phaser is terminated.
732 >     *
733 >     * @param phase an arrival phase number, or negative value if
734 >     * terminated; this argument is normally the value returned by a
735 >     * previous call to {@code arrive} or {@code arriveAndDeregister}.
736 >     * @param timeout how long to wait before giving up, in units of
737 >     *        {@code unit}
738 >     * @param unit a {@code TimeUnit} determining how to interpret the
739 >     *        {@code timeout} parameter
740 >     * @return the next arrival phase number, or the argument if it is
741 >     * negative, or the (negative) {@linkplain #getPhase() current phase}
742 >     * if terminated
743       * @throws InterruptedException if thread interrupted while waiting
744       * @throws TimeoutException if timed out while waiting
745       */
746      public int awaitAdvanceInterruptibly(int phase,
747                                           long timeout, TimeUnit unit)
748          throws InterruptedException, TimeoutException {
749 +        long nanos = unit.toNanos(timeout);
750 +        final Phaser root = this.root;
751 +        long s = (root == this) ? state : reconcileState();
752 +        int p = (int)(s >>> PHASE_SHIFT);
753          if (phase < 0)
754              return phase;
755 <        long s = getReconciledState();
756 <        int p = phaseOf(s);
757 <        if (p != phase)
758 <            return p;
759 <        if (unarrivedOf(s) == 0 && parent != null)
760 <            parent.awaitAdvanceInterruptibly(phase, timeout, unit);
761 <        return timedWait(phase, unit.toNanos(timeout));
755 >        if (p == phase) {
756 >            QNode node = new QNode(this, phase, true, true, nanos);
757 >            p = root.internalAwaitAdvance(phase, node);
758 >            if (node.wasInterrupted)
759 >                throw new InterruptedException();
760 >            else if (p == phase)
761 >                throw new TimeoutException();
762 >        }
763 >        return p;
764      }
765  
766      /**
767 <     * Forces this barrier to enter termination state. Counts of
768 <     * arrived and registered parties are unaffected. If this phaser
769 <     * has a parent, it too is terminated. This method may be useful
770 <     * for coordinating recovery after one or more tasks encounter
767 >     * Forces this phaser to enter termination state.  Counts of
768 >     * registered parties are unaffected.  If this phaser is a member
769 >     * of a tiered set of phasers, then all of the phasers in the set
770 >     * are terminated.  If this phaser is already terminated, this
771 >     * method has no effect.  This method may be useful for
772 >     * coordinating recovery after one or more tasks encounter
773       * unexpected exceptions.
774       */
775      public void forceTermination() {
776 <        for (;;) {
777 <            long s = getReconciledState();
778 <            int phase = phaseOf(s);
779 <            int parties = partiesOf(s);
780 <            int unarrived = unarrivedOf(s);
781 <            if (phase < 0 ||
782 <                casState(s, stateFor(-1, parties, unarrived))) {
776 >        // Only need to change root state
777 >        final Phaser root = this.root;
778 >        long s;
779 >        while ((s = root.state) >= 0) {
780 >            if (UNSAFE.compareAndSwapLong(root, stateOffset,
781 >                                          s, s | TERMINATION_BIT)) {
782 >                // signal all threads
783                  releaseWaiters(0);
784                  releaseWaiters(1);
620                if (parent != null)
621                    parent.forceTermination();
785                  return;
786              }
787          }
# Line 627 | Line 790 | public class Phaser {
790      /**
791       * Returns the current phase number. The maximum phase number is
792       * {@code Integer.MAX_VALUE}, after which it restarts at
793 <     * zero. Upon termination, the phase number is negative.
793 >     * zero. Upon termination, the phase number is negative,
794 >     * in which case the prevailing phase prior to termination
795 >     * may be obtained via {@code getPhase() + Integer.MIN_VALUE}.
796       *
797       * @return the phase number, or a negative value if terminated
798       */
799      public final int getPhase() {
800 <        return phaseOf(getReconciledState());
800 >        return (int)(root.state >>> PHASE_SHIFT);
801      }
802  
803      /**
804 <     * Returns {@code true} if the current phase number equals the given phase.
640 <     *
641 <     * @param phase the phase
642 <     * @return {@code true} if the current phase number equals the given phase
643 <     */
644 <    public final boolean hasPhase(int phase) {
645 <        return phaseOf(getReconciledState()) == phase;
646 <    }
647 <
648 <    /**
649 <     * Returns the number of parties registered at this barrier.
804 >     * Returns the number of parties registered at this phaser.
805       *
806       * @return the number of parties
807       */
# Line 655 | Line 810 | public class Phaser {
810      }
811  
812      /**
813 <     * Returns the number of parties that have arrived at the current
814 <     * phase of this barrier.
813 >     * Returns the number of registered parties that have arrived at
814 >     * the current phase of this phaser. If this phaser has terminated,
815 >     * the returned value is meaningless and arbitrary.
816       *
817       * @return the number of arrived parties
818       */
819      public int getArrivedParties() {
820 <        return arrivedOf(state);
820 >        return arrivedOf(reconcileState());
821      }
822  
823      /**
824       * Returns the number of registered parties that have not yet
825 <     * arrived at the current phase of this barrier.
825 >     * arrived at the current phase of this phaser. If this phaser has
826 >     * terminated, the returned value is meaningless and arbitrary.
827       *
828       * @return the number of unarrived parties
829       */
830      public int getUnarrivedParties() {
831 <        return unarrivedOf(state);
831 >        return unarrivedOf(reconcileState());
832      }
833  
834      /**
835 <     * Returns the parent of this phaser, or null if none.
835 >     * Returns the parent of this phaser, or {@code null} if none.
836       *
837 <     * @return the parent of this phaser, or null if none
837 >     * @return the parent of this phaser, or {@code null} if none
838       */
839      public Phaser getParent() {
840          return parent;
# Line 694 | Line 851 | public class Phaser {
851      }
852  
853      /**
854 <     * Returns {@code true} if this barrier has been terminated.
854 >     * Returns {@code true} if this phaser has been terminated.
855       *
856 <     * @return {@code true} if this barrier has been terminated
856 >     * @return {@code true} if this phaser has been terminated
857       */
858      public boolean isTerminated() {
859 <        return getPhase() < 0;
859 >        return root.state < 0L;
860      }
861  
862      /**
863 <     * Overridable method to perform an action upon phase advance, and
864 <     * to control termination. This method is invoked whenever the
865 <     * barrier is tripped (and thus all other waiting parties are
866 <     * dormant). If it returns true, then, rather than advance the
867 <     * phase number, this barrier will be set to a final termination
868 <     * state, and subsequent calls to {@code isTerminated} will
869 <     * return true.
870 <     *
871 <     * <p> The default version returns true when the number of
872 <     * registered parties is zero. Normally, overrides that arrange
873 <     * termination for other reasons should also preserve this
874 <     * property.
875 <     *
876 <     * <p> You may override this method to perform an action with side
877 <     * effects visible to participating tasks, but it is in general
878 <     * only sensible to do so in designs where all parties register
879 <     * before any arrive, and all {@code awaitAdvance} at each phase.
880 <     * Otherwise, you cannot ensure lack of interference. In
881 <     * particular, this method may be invoked more than once per
882 <     * transition if other parties successfully register while the
883 <     * invocation of this method is in progress, thus postponing the
884 <     * transition until those parties also arrive, re-triggering this
885 <     * method.
863 >     * Overridable method to perform an action upon impending phase
864 >     * advance, and to control termination. This method is invoked
865 >     * upon arrival of the party advancing this phaser (when all other
866 >     * waiting parties are dormant).  If this method returns {@code
867 >     * true}, this phaser will be set to a final termination state
868 >     * upon advance, and subsequent calls to {@link #isTerminated}
869 >     * will return true. Any (unchecked) Exception or Error thrown by
870 >     * an invocation of this method is propagated to the party
871 >     * attempting to advance this phaser, in which case no advance
872 >     * occurs.
873 >     *
874 >     * <p>The arguments to this method provide the state of the phaser
875 >     * prevailing for the current transition.  The effects of invoking
876 >     * arrival, registration, and waiting methods on this phaser from
877 >     * within {@code onAdvance} are unspecified and should not be
878 >     * relied on.
879 >     *
880 >     * <p>If this phaser is a member of a tiered set of phasers, then
881 >     * {@code onAdvance} is invoked only for its root phaser on each
882 >     * advance.
883 >     *
884 >     * <p>To support the most common use cases, the default
885 >     * implementation of this method returns {@code true} when the
886 >     * number of registered parties has become zero as the result of a
887 >     * party invoking {@code arriveAndDeregister}.  You can disable
888 >     * this behavior, thus enabling continuation upon future
889 >     * registrations, by overriding this method to always return
890 >     * {@code false}:
891 >     *
892 >     * <pre> {@code
893 >     * Phaser phaser = new Phaser() {
894 >     *   protected boolean onAdvance(int phase, int parties) { return false; }
895 >     * }}</pre>
896       *
897 <     * @param phase the phase number on entering the barrier
897 >     * @param phase the current phase number on entry to this method,
898 >     * before this phaser is advanced
899       * @param registeredParties the current number of registered parties
900 <     * @return {@code true} if this barrier should terminate
900 >     * @return {@code true} if this phaser should terminate
901       */
902      protected boolean onAdvance(int phase, int registeredParties) {
903 <        return registeredParties <= 0;
903 >        return registeredParties == 0;
904      }
905  
906      /**
# Line 742 | Line 910 | public class Phaser {
910       * followed by the number of registered parties, and {@code
911       * "arrived = "} followed by the number of arrived parties.
912       *
913 <     * @return a string identifying this barrier, as well as its state
913 >     * @return a string identifying this phaser, as well as its state
914       */
915      public String toString() {
916 <        long s = getReconciledState();
916 >        return stateToString(reconcileState());
917 >    }
918 >
919 >    /**
920 >     * Implementation of toString and string-based error messages
921 >     */
922 >    private String stateToString(long s) {
923          return super.toString() +
924              "[phase = " + phaseOf(s) +
925              " parties = " + partiesOf(s) +
926              " arrived = " + arrivedOf(s) + "]";
927      }
928  
929 <    // methods for waiting
929 >    // Waiting mechanics
930 >
931 >    /**
932 >     * Removes and signals threads from queue for phase.
933 >     */
934 >    private void releaseWaiters(int phase) {
935 >        QNode q;   // first element of queue
936 >        Thread t;  // its thread
937 >        AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
938 >        while ((q = head.get()) != null &&
939 >               q.phase != (int)(root.state >>> PHASE_SHIFT)) {
940 >            if (head.compareAndSet(q, q.next) &&
941 >                (t = q.thread) != null) {
942 >                q.thread = null;
943 >                LockSupport.unpark(t);
944 >            }
945 >        }
946 >    }
947 >
948 >    /**
949 >     * Variant of releaseWaiters that additionally tries to remove any
950 >     * nodes no longer waiting for advance due to timeout or
951 >     * interrupt. Currently, nodes are removed only if they are at
952 >     * head of queue, which suffices to reduce memory footprint in
953 >     * most usages.
954 >     *
955 >     * @return current phase on exit
956 >     */
957 >    private int abortWait(int phase) {
958 >        AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
959 >        for (;;) {
960 >            Thread t;
961 >            QNode q = head.get();
962 >            int p = (int)(root.state >>> PHASE_SHIFT);
963 >            if (q == null || ((t = q.thread) != null && q.phase == p))
964 >                return p;
965 >            if (head.compareAndSet(q, q.next) && t != null) {
966 >                q.thread = null;
967 >                LockSupport.unpark(t);
968 >            }
969 >        }
970 >    }
971 >
972 >    /** The number of CPUs, for spin control */
973 >    private static final int NCPU = Runtime.getRuntime().availableProcessors();
974 >
975 >    /**
976 >     * The number of times to spin before blocking while waiting for
977 >     * advance, per arrival while waiting. On multiprocessors, fully
978 >     * blocking and waking up a large number of threads all at once is
979 >     * usually a very slow process, so we use rechargeable spins to
980 >     * avoid it when threads regularly arrive: When a thread in
981 >     * internalAwaitAdvance notices another arrival before blocking,
982 >     * and there appear to be enough CPUs available, it spins
983 >     * SPINS_PER_ARRIVAL more times before blocking. The value trades
984 >     * off good-citizenship vs big unnecessary slowdowns.
985 >     */
986 >    static final int SPINS_PER_ARRIVAL = (NCPU < 2) ? 1 : 1 << 8;
987 >
988 >    /**
989 >     * Possibly blocks and waits for phase to advance unless aborted.
990 >     * Call only from root node.
991 >     *
992 >     * @param phase current phase
993 >     * @param node if non-null, the wait node to track interrupt and timeout;
994 >     * if null, denotes noninterruptible wait
995 >     * @return current phase
996 >     */
997 >    private int internalAwaitAdvance(int phase, QNode node) {
998 >        releaseWaiters(phase-1);          // ensure old queue clean
999 >        boolean queued = false;           // true when node is enqueued
1000 >        int lastUnarrived = 0;            // to increase spins upon change
1001 >        int spins = SPINS_PER_ARRIVAL;
1002 >        long s;
1003 >        int p;
1004 >        while ((p = (int)((s = state) >>> PHASE_SHIFT)) == phase) {
1005 >            if (node == null) {           // spinning in noninterruptible mode
1006 >                int unarrived = (int)s & UNARRIVED_MASK;
1007 >                if (unarrived != lastUnarrived &&
1008 >                    (lastUnarrived = unarrived) < NCPU)
1009 >                    spins += SPINS_PER_ARRIVAL;
1010 >                boolean interrupted = Thread.interrupted();
1011 >                if (interrupted || --spins < 0) { // need node to record intr
1012 >                    node = new QNode(this, phase, false, false, 0L);
1013 >                    node.wasInterrupted = interrupted;
1014 >                }
1015 >            }
1016 >            else if (node.isReleasable()) // done or aborted
1017 >                break;
1018 >            else if (!queued) {           // push onto queue
1019 >                AtomicReference<QNode> head = (phase & 1) == 0 ? evenQ : oddQ;
1020 >                QNode q = node.next = head.get();
1021 >                if ((q == null || q.phase == phase) &&
1022 >                    (int)(state >>> PHASE_SHIFT) == phase) // avoid stale enq
1023 >                    queued = head.compareAndSet(q, node);
1024 >            }
1025 >            else {
1026 >                try {
1027 >                    ForkJoinPool.managedBlock(node);
1028 >                } catch (InterruptedException ie) {
1029 >                    node.wasInterrupted = true;
1030 >                }
1031 >            }
1032 >        }
1033 >
1034 >        if (node != null) {
1035 >            if (node.thread != null)
1036 >                node.thread = null;       // avoid need for unpark()
1037 >            if (node.wasInterrupted && !node.interruptible)
1038 >                Thread.currentThread().interrupt();
1039 >            if (p == phase && (p = (int)(state >>> PHASE_SHIFT)) == phase)
1040 >                return abortWait(phase); // possibly clean up on abort
1041 >        }
1042 >        releaseWaiters(phase);
1043 >        return p;
1044 >    }
1045  
1046      /**
1047       * Wait nodes for Treiber stack representing wait queue
# Line 760 | Line 1049 | public class Phaser {
1049      static final class QNode implements ForkJoinPool.ManagedBlocker {
1050          final Phaser phaser;
1051          final int phase;
763        final long startTime;
764        final long nanos;
765        final boolean timed;
1052          final boolean interruptible;
1053 <        volatile boolean wasInterrupted = false;
1053 >        final boolean timed;
1054 >        boolean wasInterrupted;
1055 >        long nanos;
1056 >        long lastTime;
1057          volatile Thread thread; // nulled to cancel wait
1058          QNode next;
1059 +
1060          QNode(Phaser phaser, int phase, boolean interruptible,
1061 <              boolean timed, long startTime, long nanos) {
1061 >              boolean timed, long nanos) {
1062              this.phaser = phaser;
1063              this.phase = phase;
774            this.timed = timed;
1064              this.interruptible = interruptible;
776            this.startTime = startTime;
1065              this.nanos = nanos;
1066 +            this.timed = timed;
1067 +            this.lastTime = timed ? System.nanoTime() : 0L;
1068              thread = Thread.currentThread();
1069          }
1070 +
1071          public boolean isReleasable() {
1072 <            return (thread == null ||
1073 <                    phaser.getPhase() != phase ||
1074 <                    (interruptible && wasInterrupted) ||
1075 <                    (timed && (nanos - (System.nanoTime() - startTime)) <= 0));
1076 <        }
786 <        public boolean block() {
787 <            if (Thread.interrupted()) {
788 <                wasInterrupted = true;
789 <                if (interruptible)
790 <                    return true;
791 <            }
792 <            if (!timed)
793 <                LockSupport.park(this);
794 <            else {
795 <                long waitTime = nanos - (System.nanoTime() - startTime);
796 <                if (waitTime <= 0)
797 <                    return true;
798 <                LockSupport.parkNanos(this, waitTime);
1072 >            if (thread == null)
1073 >                return true;
1074 >            if (phaser.getPhase() != phase) {
1075 >                thread = null;
1076 >                return true;
1077              }
1078 <            return isReleasable();
1079 <        }
1080 <        void signal() {
803 <            Thread t = thread;
804 <            if (t != null) {
1078 >            if (Thread.interrupted())
1079 >                wasInterrupted = true;
1080 >            if (wasInterrupted && interruptible) {
1081                  thread = null;
1082 <                LockSupport.unpark(t);
1082 >                return true;
1083              }
1084 <        }
1085 <        boolean doWait() {
1086 <            if (thread != null) {
1087 <                try {
1088 <                    ForkJoinPool.managedBlock(this, false);
1089 <                } catch (InterruptedException ie) {
1084 >            if (timed) {
1085 >                if (nanos > 0L) {
1086 >                    long now = System.nanoTime();
1087 >                    nanos -= now - lastTime;
1088 >                    lastTime = now;
1089 >                }
1090 >                if (nanos <= 0L) {
1091 >                    thread = null;
1092 >                    return true;
1093                  }
1094              }
1095 <            return wasInterrupted;
1095 >            return false;
1096          }
1097  
1098 <    }
1099 <
1100 <    /**
1101 <     * Removes and signals waiting threads from wait queue.
1102 <     */
1103 <    private void releaseWaiters(int phase) {
1104 <        AtomicReference<QNode> head = queueFor(phase);
1105 <        QNode q;
827 <        while ((q = head.get()) != null) {
828 <            if (head.compareAndSet(q, q.next))
829 <                q.signal();
1098 >        public boolean block() {
1099 >            if (isReleasable())
1100 >                return true;
1101 >            else if (!timed)
1102 >                LockSupport.park(this);
1103 >            else if (nanos > 0)
1104 >                LockSupport.parkNanos(this, nanos);
1105 >            return isReleasable();
1106          }
1107      }
1108  
1109 <    /**
834 <     * Tries to enqueue given node in the appropriate wait queue.
835 <     *
836 <     * @return true if successful
837 <     */
838 <    private boolean tryEnqueue(QNode node) {
839 <        AtomicReference<QNode> head = queueFor(node.phase);
840 <        return head.compareAndSet(node.next = head.get(), node);
841 <    }
1109 >    // Unsafe mechanics
1110  
1111 <    /**
1112 <     * Enqueues node and waits unless aborted or signalled.
1113 <     *
1114 <     * @return current phase
1115 <     */
1116 <    private int untimedWait(int phase) {
1117 <        QNode node = null;
1118 <        boolean queued = false;
1119 <        boolean interrupted = false;
1120 <        int p;
853 <        while ((p = getPhase()) == phase) {
854 <            if (Thread.interrupted())
855 <                interrupted = true;
856 <            else if (node == null)
857 <                node = new QNode(this, phase, false, false, 0, 0);
858 <            else if (!queued)
859 <                queued = tryEnqueue(node);
860 <            else
861 <                interrupted = node.doWait();
1111 >    private static final sun.misc.Unsafe UNSAFE;
1112 >    private static final long stateOffset;
1113 >    static {
1114 >        try {
1115 >            UNSAFE = getUnsafe();
1116 >            Class k = Phaser.class;
1117 >            stateOffset = UNSAFE.objectFieldOffset
1118 >                (k.getDeclaredField("state"));
1119 >        } catch (Exception e) {
1120 >            throw new Error(e);
1121          }
863        if (node != null)
864            node.thread = null;
865        releaseWaiters(phase);
866        if (interrupted)
867            Thread.currentThread().interrupt();
868        return p;
1122      }
1123  
1124      /**
1125 <     * Interruptible version
1126 <     * @return current phase
1127 <     */
1128 <    private int interruptibleWait(int phase) throws InterruptedException {
1129 <        QNode node = null;
877 <        boolean queued = false;
878 <        boolean interrupted = false;
879 <        int p;
880 <        while ((p = getPhase()) == phase && !interrupted) {
881 <            if (Thread.interrupted())
882 <                interrupted = true;
883 <            else if (node == null)
884 <                node = new QNode(this, phase, true, false, 0, 0);
885 <            else if (!queued)
886 <                queued = tryEnqueue(node);
887 <            else
888 <                interrupted = node.doWait();
889 <        }
890 <        if (node != null)
891 <            node.thread = null;
892 <        if (p != phase || (p = getPhase()) != phase)
893 <            releaseWaiters(phase);
894 <        if (interrupted)
895 <            throw new InterruptedException();
896 <        return p;
897 <    }
898 <
899 <    /**
900 <     * Timeout version.
901 <     * @return current phase
1125 >     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1126 >     * Replace with a simple call to Unsafe.getUnsafe when integrating
1127 >     * into a jdk.
1128 >     *
1129 >     * @return a sun.misc.Unsafe
1130       */
903    private int timedWait(int phase, long nanos)
904        throws InterruptedException, TimeoutException {
905        long startTime = System.nanoTime();
906        QNode node = null;
907        boolean queued = false;
908        boolean interrupted = false;
909        int p;
910        while ((p = getPhase()) == phase && !interrupted) {
911            if (Thread.interrupted())
912                interrupted = true;
913            else if (nanos - (System.nanoTime() - startTime) <= 0)
914                break;
915            else if (node == null)
916                node = new QNode(this, phase, true, true, startTime, nanos);
917            else if (!queued)
918                queued = tryEnqueue(node);
919            else
920                interrupted = node.doWait();
921        }
922        if (node != null)
923            node.thread = null;
924        if (p != phase || (p = getPhase()) != phase)
925            releaseWaiters(phase);
926        if (interrupted)
927            throw new InterruptedException();
928        if (p == phase)
929            throw new TimeoutException();
930        return p;
931    }
932
933    // Unsafe mechanics for jsr166y 3rd party package.
1131      private static sun.misc.Unsafe getUnsafe() {
1132          try {
1133              return sun.misc.Unsafe.getUnsafe();
1134          } catch (SecurityException se) {
1135              try {
1136                  return java.security.AccessController.doPrivileged
1137 <                    (new java.security.PrivilegedExceptionAction<sun.misc.Unsafe>() {
1137 >                    (new java.security
1138 >                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1139                          public sun.misc.Unsafe run() throws Exception {
1140 <                            return getUnsafeByReflection();
1140 >                            java.lang.reflect.Field f = sun.misc
1141 >                                .Unsafe.class.getDeclaredField("theUnsafe");
1142 >                            f.setAccessible(true);
1143 >                            return (sun.misc.Unsafe) f.get(null);
1144                          }});
1145              } catch (java.security.PrivilegedActionException e) {
1146                  throw new RuntimeException("Could not initialize intrinsics",
# Line 947 | Line 1148 | public class Phaser {
1148              }
1149          }
1150      }
950
951    private static sun.misc.Unsafe getUnsafeByReflection()
952            throws NoSuchFieldException, IllegalAccessException {
953        java.lang.reflect.Field f =
954            sun.misc.Unsafe.class.getDeclaredField("theUnsafe");
955        f.setAccessible(true);
956        return (sun.misc.Unsafe) f.get(null);
957    }
958
959    private static long fieldOffset(String fieldName, Class<?> klazz) {
960        try {
961            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(fieldName));
962        } catch (NoSuchFieldException e) {
963            // Convert Exception to Error
964            NoSuchFieldError error = new NoSuchFieldError(fieldName);
965            error.initCause(e);
966            throw error;
967        }
968    }
969
970    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
971    static final long stateOffset =
972        fieldOffset("state", Phaser.class);
973
974    final boolean casState(long cmp, long val) {
975        return UNSAFE.compareAndSwapLong(this, stateOffset, cmp, val);
976    }
1151   }

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines