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

Comparing jsr166/src/jsr166y/ForkJoinWorkerThread.java (file contents):
Revision 1.44 by jsr166, Wed Sep 1 20:15:43 2010 UTC vs.
Revision 1.69 by dl, Mon Feb 20 18:20:06 2012 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
11 import java.util.Random;
12 import java.util.Collection;
13 import java.util.concurrent.locks.LockSupport;
14
9   /**
10 < * A thread managed by a {@link ForkJoinPool}.  This class is
11 < * subclassable solely for the sake of adding functionality -- there
12 < * are no overridable methods dealing with scheduling or execution.
13 < * However, you can override initialization and termination methods
14 < * surrounding the main task processing loop.  If you do create such a
15 < * subclass, you will also need to supply a custom {@link
16 < * ForkJoinPool.ForkJoinWorkerThreadFactory} to use it in a {@code
17 < * ForkJoinPool}.
10 > * A thread managed by a {@link ForkJoinPool}, which executes
11 > * {@link ForkJoinTask}s.
12 > * This class is subclassable solely for the sake of adding
13 > * functionality -- there are no overridable methods dealing with
14 > * scheduling or execution.  However, you can override initialization
15 > * and termination methods surrounding the main task processing loop.
16 > * If you do create such a subclass, you will also need to supply a
17 > * custom {@link ForkJoinPool.ForkJoinWorkerThreadFactory} to use it
18 > * in a {@code ForkJoinPool}.
19   *
20   * @since 1.7
21   * @author Doug Lea
22   */
23   public class ForkJoinWorkerThread extends Thread {
24      /*
30     * Overview:
31     *
25       * ForkJoinWorkerThreads are managed by ForkJoinPools and perform
26 <     * ForkJoinTasks. This class includes bookkeeping in support of
27 <     * worker activation, suspension, and lifecycle control described
35 <     * in more detail in the internal documentation of class
36 <     * ForkJoinPool. And as described further below, this class also
37 <     * includes special-cased support for some ForkJoinTask
38 <     * methods. But the main mechanics involve work-stealing:
39 <     *
40 <     * Work-stealing queues are special forms of Deques that support
41 <     * only three of the four possible end-operations -- push, pop,
42 <     * and deq (aka steal), under the further constraints that push
43 <     * and pop are called only from the owning thread, while deq may
44 <     * be called from other threads.  (If you are unfamiliar with
45 <     * them, you probably want to read Herlihy and Shavit's book "The
46 <     * Art of Multiprocessor programming", chapter 16 describing these
47 <     * in more detail before proceeding.)  The main work-stealing
48 <     * queue design is roughly similar to those in the papers "Dynamic
49 <     * Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005
50 <     * (http://research.sun.com/scalable/pubs/index.html) and
51 <     * "Idempotent work stealing" by Michael, Saraswat, and Vechev,
52 <     * PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186).
53 <     * The main differences ultimately stem from gc requirements that
54 <     * we null out taken slots as soon as we can, to maintain as small
55 <     * a footprint as possible even in programs generating huge
56 <     * numbers of tasks. To accomplish this, we shift the CAS
57 <     * arbitrating pop vs deq (steal) from being on the indices
58 <     * ("base" and "sp") to the slots themselves (mainly via method
59 <     * "casSlotNull()"). So, both a successful pop and deq mainly
60 <     * entail a CAS of a slot from non-null to null.  Because we rely
61 <     * on CASes of references, we do not need tag bits on base or sp.
62 <     * They are simple ints as used in any circular array-based queue
63 <     * (see for example ArrayDeque).  Updates to the indices must
64 <     * still be ordered in a way that guarantees that sp == base means
65 <     * the queue is empty, but otherwise may err on the side of
66 <     * possibly making the queue appear nonempty when a push, pop, or
67 <     * deq have not fully committed. Note that this means that the deq
68 <     * operation, considered individually, is not wait-free. One thief
69 <     * cannot successfully continue until another in-progress one (or,
70 <     * if previously empty, a push) completes.  However, in the
71 <     * aggregate, we ensure at least probabilistic non-blockingness.
72 <     * If an attempted steal fails, a thief always chooses a different
73 <     * random victim target to try next. So, in order for one thief to
74 <     * progress, it suffices for any in-progress deq or new push on
75 <     * any empty queue to complete. One reason this works well here is
76 <     * that apparently-nonempty often means soon-to-be-stealable,
77 <     * which gives threads a chance to set activation status if
78 <     * necessary before stealing.
79 <     *
80 <     * This approach also enables support for "async mode" where local
81 <     * task processing is in FIFO, not LIFO order; simply by using a
82 <     * version of deq rather than pop when locallyFifo is true (as set
83 <     * by the ForkJoinPool).  This allows use in message-passing
84 <     * frameworks in which tasks are never joined.
85 <     *
86 <     * When a worker would otherwise be blocked waiting to join a
87 <     * task, it first tries a form of linear helping: Each worker
88 <     * records (in field currentSteal) the most recent task it stole
89 <     * from some other worker. Plus, it records (in field currentJoin)
90 <     * the task it is currently actively joining. Method joinTask uses
91 <     * these markers to try to find a worker to help (i.e., steal back
92 <     * a task from and execute it) that could hasten completion of the
93 <     * actively joined task. In essence, the joiner executes a task
94 <     * that would be on its own local deque had the to-be-joined task
95 <     * not been stolen. This may be seen as a conservative variant of
96 <     * the approach in Wagner & Calder "Leapfrogging: a portable
97 <     * technique for implementing efficient futures" SIGPLAN Notices,
98 <     * 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs
99 <     * in that: (1) We only maintain dependency links across workers
100 <     * upon steals, rather than use per-task bookkeeping.  This may
101 <     * require a linear scan of workers array to locate stealers, but
102 <     * usually doesn't because stealers leave hints (that may become
103 <     * stale/wrong) of where to locate them. This isolates cost to
104 <     * when it is needed, rather than adding to per-task overhead.
105 <     * (2) It is "shallow", ignoring nesting and potentially cyclic
106 <     * mutual steals.  (3) It is intentionally racy: field currentJoin
107 <     * is updated only while actively joining, which means that we
108 <     * miss links in the chain during long-lived tasks, GC stalls etc
109 <     * (which is OK since blocking in such cases is usually a good
110 <     * idea).  (4) We bound the number of attempts to find work (see
111 <     * MAX_HELP_DEPTH) and fall back to suspending the worker and if
112 <     * necessary replacing it with a spare (see
113 <     * ForkJoinPool.awaitJoin).
114 <     *
115 <     * Efficient implementation of these algorithms currently relies
116 <     * on an uncomfortable amount of "Unsafe" mechanics. To maintain
117 <     * correct orderings, reads and writes of variable base require
118 <     * volatile ordering.  Variable sp does not require volatile
119 <     * writes but still needs store-ordering, which we accomplish by
120 <     * pre-incrementing sp before filling the slot with an ordered
121 <     * store.  (Pre-incrementing also enables backouts used in
122 <     * joinTask.)  Because they are protected by volatile base reads,
123 <     * reads of the queue array and its slots by other threads do not
124 <     * need volatile load semantics, but writes (in push) require
125 <     * store order and CASes (in pop and deq) require (volatile) CAS
126 <     * semantics.  (Michael, Saraswat, and Vechev's algorithm has
127 <     * similar properties, but without support for nulling slots.)
128 <     * Since these combinations aren't supported using ordinary
129 <     * volatiles, the only way to accomplish these efficiently is to
130 <     * use direct Unsafe calls. (Using external AtomicIntegers and
131 <     * AtomicReferenceArrays for the indices and array is
132 <     * significantly slower because of memory locality and indirection
133 <     * effects.)
134 <     *
135 <     * Further, performance on most platforms is very sensitive to
136 <     * placement and sizing of the (resizable) queue array.  Even
137 <     * though these queues don't usually become all that big, the
138 <     * initial size must be large enough to counteract cache
139 <     * contention effects across multiple queues (especially in the
140 <     * presence of GC cardmarking). Also, to improve thread-locality,
141 <     * queues are initialized after starting.  All together, these
142 <     * low-level implementation choices produce as much as a factor of
143 <     * 4 performance improvement compared to naive implementations,
144 <     * and enable the processing of billions of tasks per second,
145 <     * sometimes at the expense of ugliness.
146 <     */
147 <
148 <    /**
149 <     * Generator for initial random seeds for random victim
150 <     * selection. This is used only to create initial seeds. Random
151 <     * steals use a cheaper xorshift generator per steal attempt. We
152 <     * expect only rare contention on seedGenerator, so just use a
153 <     * plain Random.
154 <     */
155 <    private static final Random seedGenerator = new Random();
156 <
157 <    /**
158 <     * The maximum stolen->joining link depth allowed in helpJoinTask.
159 <     * Depths for legitimate chains are unbounded, but we use a fixed
160 <     * constant to avoid (otherwise unchecked) cycles and bound
161 <     * staleness of traversal parameters at the expense of sometimes
162 <     * blocking when we could be helping.
163 <     */
164 <    private static final int MAX_HELP_DEPTH = 8;
165 <
166 <    /**
167 <     * Capacity of work-stealing queue array upon initialization.
168 <     * Must be a power of two. Initial size must be at least 4, but is
169 <     * padded to minimize cache effects.
170 <     */
171 <    private static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
172 <
173 <    /**
174 <     * Maximum work-stealing queue array size.  Must be less than or
175 <     * equal to 1 << 28 to ensure lack of index wraparound. (This
176 <     * is less than usual bounds, because we need leftshift by 3
177 <     * to be in int range).
178 <     */
179 <    private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
180 <
181 <    /**
182 <     * The pool this thread works in. Accessed directly by ForkJoinTask.
183 <     */
184 <    final ForkJoinPool pool;
185 <
186 <    /**
187 <     * The work-stealing queue array. Size must be a power of two.
188 <     * Initialized in onStart, to improve memory locality.
189 <     */
190 <    private ForkJoinTask<?>[] queue;
191 <
192 <    /**
193 <     * Index (mod queue.length) of least valid queue slot, which is
194 <     * always the next position to steal from if nonempty.
195 <     */
196 <    private volatile int base;
197 <
198 <    /**
199 <     * Index (mod queue.length) of next queue slot to push to or pop
200 <     * from. It is written only by owner thread, and accessed by other
201 <     * threads only after reading (volatile) base.  Both sp and base
202 <     * are allowed to wrap around on overflow, but (sp - base) still
203 <     * estimates size.
204 <     */
205 <    private int sp;
206 <
207 <    /**
208 <     * The index of most recent stealer, used as a hint to avoid
209 <     * traversal in method helpJoinTask. This is only a hint because a
210 <     * worker might have had multiple steals and this only holds one
211 <     * of them (usually the most current). Declared non-volatile,
212 <     * relying on other prevailing sync to keep reasonably current.
213 <     */
214 <    private int stealHint;
215 <
216 <    /**
217 <     * Run state of this worker. In addition to the usual run levels,
218 <     * tracks if this worker is suspended as a spare, and if it was
219 <     * killed (trimmed) while suspended. However, "active" status is
220 <     * maintained separately and modified only in conjunction with
221 <     * CASes of the pool's runState (which are currently sadly
222 <     * manually inlined for performance.)  Accessed directly by pool
223 <     * to simplify checks for normal (zero) status.
224 <     */
225 <    volatile int runState;
226 <
227 <    private static final int TERMINATING = 0x01;
228 <    private static final int TERMINATED  = 0x02;
229 <    private static final int SUSPENDED   = 0x04; // inactive spare
230 <    private static final int TRIMMED     = 0x08; // killed while suspended
231 <
232 <    /**
233 <     * Number of steals, transferred and reset in pool callbacks pool
234 <     * when idle Accessed directly by pool.
235 <     */
236 <    int stealCount;
237 <
238 <    /**
239 <     * Seed for random number generator for choosing steal victims.
240 <     * Uses Marsaglia xorshift. Must be initialized as nonzero.
241 <     */
242 <    private int seed;
243 <
244 <    /**
245 <     * Activity status. When true, this worker is considered active.
246 <     * Accessed directly by pool.  Must be false upon construction.
247 <     */
248 <    boolean active;
249 <
250 <    /**
251 <     * True if use local fifo, not default lifo, for local polling.
252 <     * Shadows value from ForkJoinPool.
253 <     */
254 <    private final boolean locallyFifo;
255 <
256 <    /**
257 <     * Index of this worker in pool array. Set once by pool before
258 <     * running, and accessed directly by pool to locate this worker in
259 <     * its workers array.
26 >     * ForkJoinTasks. For explanation, see the internal documentation
27 >     * of class ForkJoinPool.
28       */
261    int poolIndex;
29  
30 <    /**
31 <     * The last pool event waited for. Accessed only by pool in
265 <     * callback methods invoked within this thread.
266 <     */
267 <    int lastEventCount;
268 <
269 <    /**
270 <     * Encoded index and event count of next event waiter. Accessed
271 <     * only by ForkJoinPool for managing event waiters.
272 <     */
273 <    volatile long nextWaiter;
274 <
275 <    /**
276 <     * Number of times this thread suspended as spare. Accessed only
277 <     * by pool.
278 <     */
279 <    int spareCount;
280 <
281 <    /**
282 <     * Encoded index and count of next spare waiter. Accessed only
283 <     * by ForkJoinPool for managing spares.
284 <     */
285 <    volatile int nextSpare;
286 <
287 <    /**
288 <     * The task currently being joined, set only when actively trying
289 <     * to helpStealer. Written only by current thread, but read by
290 <     * others.
291 <     */
292 <    private volatile ForkJoinTask<?> currentJoin;
293 <
294 <    /**
295 <     * The task most recently stolen from another worker (or
296 <     * submission queue).  Written only by current thread, but read by
297 <     * others.
298 <     */
299 <    private volatile ForkJoinTask<?> currentSteal;
30 >    final ForkJoinPool.WorkQueue workQueue; // Work-stealing mechanics
31 >    final ForkJoinPool pool;                // the pool this thread works in
32  
33      /**
34       * Creates a ForkJoinWorkerThread operating in the given pool.
# Line 305 | Line 37 | public class ForkJoinWorkerThread extend
37       * @throws NullPointerException if pool is null
38       */
39      protected ForkJoinWorkerThread(ForkJoinPool pool) {
40 <        this.pool = pool;
309 <        this.locallyFifo = pool.locallyFifo;
40 >        super(pool.nextWorkerName());
41          setDaemon(true);
42 <        // To avoid exposing construction details to subclasses,
312 <        // remaining initialization is in start() and onStart()
313 <    }
314 <
315 <    /**
316 <     * Performs additional initialization and starts this thread
317 <     */
318 <    final void start(int poolIndex, UncaughtExceptionHandler ueh) {
319 <        this.poolIndex = poolIndex;
42 >        Thread.UncaughtExceptionHandler ueh = pool.ueh;
43          if (ueh != null)
44              setUncaughtExceptionHandler(ueh);
45 <        start();
45 >        this.pool = pool;
46 >        pool.registerWorker(this.workQueue = new ForkJoinPool.WorkQueue
47 >                            (pool, this, pool.localMode));
48      }
49  
325    // Public/protected methods
326
50      /**
51       * Returns the pool hosting this thread.
52       *
# Line 343 | Line 66 | public class ForkJoinWorkerThread extend
66       * @return the index number
67       */
68      public int getPoolIndex() {
69 <        return poolIndex;
69 >        return workQueue.poolIndex;
70      }
71  
72      /**
73       * Initializes internal state after construction but before
74       * processing any tasks. If you override this method, you must
75 <     * invoke super.onStart() at the beginning of the method.
75 >     * invoke {@code super.onStart()} at the beginning of the method.
76       * Initialization requires care: Most fields must have legal
77       * default values, to ensure that attempted accesses from other
78       * threads work correctly even before this thread starts
79       * processing tasks.
80       */
81      protected void onStart() {
359        int rs = seedGenerator.nextInt();
360        seed = rs == 0? 1 : rs; // seed must be nonzero
361
362        // Allocate name string and arrays in this thread
363        String pid = Integer.toString(pool.getPoolNumber());
364        String wid = Integer.toString(poolIndex);
365        setName("ForkJoinPool-" + pid + "-worker-" + wid);
366
367        queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
82      }
83  
84      /**
# Line 376 | Line 90 | public class ForkJoinWorkerThread extend
90       * to an unrecoverable error, or {@code null} if completed normally
91       */
92      protected void onTermination(Throwable exception) {
379        try {
380            ForkJoinPool p = pool;
381            if (active) {
382                int a; // inline p.tryDecrementActiveCount
383                active = false;
384                do {} while (!UNSAFE.compareAndSwapInt
385                             (p, poolRunStateOffset, a = p.runState, a - 1));
386            }
387            cancelTasks();
388            setTerminated();
389            p.workerTerminated(this);
390        } catch (Throwable ex) {        // Shouldn't ever happen
391            if (exception == null)      // but if so, at least rethrown
392                exception = ex;
393        } finally {
394            if (exception != null)
395                UNSAFE.throwException(exception);
396        }
93      }
94  
95      /**
96       * This method is required to be public, but should never be
97       * called explicitly. It performs the main run loop to execute
98 <     * ForkJoinTasks.
98 >     * {@link ForkJoinTask}s.
99       */
100      public void run() {
101          Throwable exception = null;
102          try {
103              onStart();
104 <            mainLoop();
104 >            pool.runWorker(workQueue);
105          } catch (Throwable ex) {
106              exception = ex;
107          } finally {
412            onTermination(exception);
413        }
414    }
415
416    // helpers for run()
417
418    /**
419     * Find and execute tasks and check status while running
420     */
421    private void mainLoop() {
422        boolean ran = false; // true if ran a task on last step
423        ForkJoinPool p = pool;
424        for (;;) {
425            p.preStep(this, ran);
426            if (runState != 0)
427                break;
428            ran = tryExecSteal() || tryExecSubmission();
429        }
430    }
431
432    /**
433     * Try to steal a task and execute it
434     *
435     * @return true if ran a task
436     */
437    private boolean tryExecSteal() {
438        ForkJoinTask<?> t;
439        if ((t = scan()) != null) {
440            t.quietlyExec();
441            UNSAFE.putOrderedObject(this, currentStealOffset, null);
442            if (sp != base)
443                execLocalTasks();
444            return true;
445        }
446        return false;
447    }
448
449    /**
450     * If a submission exists, try to activate and run it;
451     *
452     * @return true if ran a task
453     */
454    private boolean tryExecSubmission() {
455        ForkJoinPool p = pool;
456        while (p.hasQueuedSubmissions()) {
457            ForkJoinTask<?> t; int a;
458            if (active || // inline p.tryIncrementActiveCount
459                (active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
460                                                   a = p.runState, a + 1))) {
461                if ((t = p.pollSubmission()) != null) {
462                    UNSAFE.putOrderedObject(this, currentStealOffset, t);
463                    t.quietlyExec();
464                    UNSAFE.putOrderedObject(this, currentStealOffset, null);
465                    if (sp != base)
466                        execLocalTasks();
467                    return true;
468                }
469            }
470        }
471        return false;
472    }
473
474    /**
475     * Runs local tasks until queue is empty or shut down.  Call only
476     * while active.
477     */
478    private void execLocalTasks() {
479        while (runState == 0) {
480            ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask();
481            if (t != null)
482                t.quietlyExec();
483            else if (sp == base)
484                break;
485        }
486    }
487
488    /*
489     * Intrinsics-based atomic writes for queue slots. These are
490     * basically the same as methods in AtomicObjectArray, but
491     * specialized for (1) ForkJoinTask elements (2) requirement that
492     * nullness and bounds checks have already been performed by
493     * callers and (3) effective offsets are known not to overflow
494     * from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't
495     * need corresponding version for reads: plain array reads are OK
496     * because they protected by other volatile reads and are
497     * confirmed by CASes.
498     *
499     * Most uses don't actually call these methods, but instead contain
500     * inlined forms that enable more predictable optimization.  We
501     * don't define the version of write used in pushTask at all, but
502     * instead inline there a store-fenced array slot write.
503     */
504
505    /**
506     * CASes slot i of array q from t to null. Caller must ensure q is
507     * non-null and index is in range.
508     */
509    private static final boolean casSlotNull(ForkJoinTask<?>[] q, int i,
510                                             ForkJoinTask<?> t) {
511        return UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
512    }
513
514    /**
515     * Performs a volatile write of the given task at given slot of
516     * array q.  Caller must ensure q is non-null and index is in
517     * range. This method is used only during resets and backouts.
518     */
519    private static final void writeSlot(ForkJoinTask<?>[] q, int i,
520                                              ForkJoinTask<?> t) {
521        UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
522    }
523
524    // queue methods
525
526    /**
527     * Pushes a task. Call only from this thread.
528     *
529     * @param t the task. Caller must ensure non-null.
530     */
531    final void pushTask(ForkJoinTask<?> t) {
532        ForkJoinTask<?>[] q = queue;
533        int mask = q.length - 1; // implicit assert q != null
534        int s = sp++;            // ok to increment sp before slot write
535        UNSAFE.putOrderedObject(q, ((s & mask) << qShift) + qBase, t);
536        if ((s -= base) == 0)
537            pool.signalWork();   // was empty
538        else if (s == mask)
539            growQueue();         // is full
540    }
541
542    /**
543     * Tries to take a task from the base of the queue, failing if
544     * empty or contended. Note: Specializations of this code appear
545     * in locallyDeqTask and elsewhere.
546     *
547     * @return a task, or null if none or contended
548     */
549    final ForkJoinTask<?> deqTask() {
550        ForkJoinTask<?> t;
551        ForkJoinTask<?>[] q;
552        int b, i;
553        if (sp != (b = base) &&
554            (q = queue) != null && // must read q after b
555            (t = q[i = (q.length - 1) & b]) != null && base == b &&
556            UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) {
557            base = b + 1;
558            return t;
559        }
560        return null;
561    }
562
563    /**
564     * Tries to take a task from the base of own queue. Assumes active
565     * status.  Called only by current thread.
566     *
567     * @return a task, or null if none
568     */
569    final ForkJoinTask<?> locallyDeqTask() {
570        ForkJoinTask<?>[] q = queue;
571        if (q != null) {
572            ForkJoinTask<?> t;
573            int b, i;
574            while (sp != (b = base)) {
575                if ((t = q[i = (q.length - 1) & b]) != null && base == b &&
576                    UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase,
577                                                t, null)) {
578                    base = b + 1;
579                    return t;
580                }
581            }
582        }
583        return null;
584    }
585
586    /**
587     * Returns a popped task, or null if empty. Assumes active status.
588     * Called only by current thread.
589     */
590    private ForkJoinTask<?> popTask() {
591        ForkJoinTask<?>[] q = queue;
592        if (q != null) {
593            int s;
594            while ((s = sp) != base) {
595                int i = (q.length - 1) & --s;
596                long u = (i << qShift) + qBase; // raw offset
597                ForkJoinTask<?> t = q[i];
598                if (t == null)   // lost to stealer
599                    break;
600                if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
601                    sp = s; // putOrderedInt may encourage more timely write
602                    // UNSAFE.putOrderedInt(this, spOffset, s);
603                    return t;
604                }
605            }
606        }
607        return null;
608    }
609
610    /**
611     * Specialized version of popTask to pop only if topmost element
612     * is the given task. Called only by current thread while
613     * active.
614     *
615     * @param t the task. Caller must ensure non-null.
616     */
617    final boolean unpushTask(ForkJoinTask<?> t) {
618        int s;
619        ForkJoinTask<?>[] q = queue;
620        if ((s = sp) != base && q != null &&
621            UNSAFE.compareAndSwapObject
622            (q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) {
623            sp = s; // putOrderedInt may encourage more timely write
624            // UNSAFE.putOrderedInt(this, spOffset, s);
625            return true;
626        }
627        return false;
628    }
629
630    /**
631     * Returns next task or null if empty or contended
632     */
633    final ForkJoinTask<?> peekTask() {
634        ForkJoinTask<?>[] q = queue;
635        if (q == null)
636            return null;
637        int mask = q.length - 1;
638        int i = locallyFifo ? base : (sp - 1);
639        return q[i & mask];
640    }
641
642    /**
643     * Doubles queue array size. Transfers elements by emulating
644     * steals (deqs) from old array and placing, oldest first, into
645     * new array.
646     */
647    private void growQueue() {
648        ForkJoinTask<?>[] oldQ = queue;
649        int oldSize = oldQ.length;
650        int newSize = oldSize << 1;
651        if (newSize > MAXIMUM_QUEUE_CAPACITY)
652            throw new RejectedExecutionException("Queue capacity exceeded");
653        ForkJoinTask<?>[] newQ = queue = new ForkJoinTask<?>[newSize];
654
655        int b = base;
656        int bf = b + oldSize;
657        int oldMask = oldSize - 1;
658        int newMask = newSize - 1;
659        do {
660            int oldIndex = b & oldMask;
661            ForkJoinTask<?> t = oldQ[oldIndex];
662            if (t != null && !casSlotNull(oldQ, oldIndex, t))
663                t = null;
664            writeSlot(newQ, b & newMask, t);
665        } while (++b != bf);
666        pool.signalWork();
667    }
668
669    /**
670     * Computes next value for random victim probe in scan().  Scans
671     * don't require a very high quality generator, but also not a
672     * crummy one.  Marsaglia xor-shift is cheap and works well enough.
673     * Note: This is manually inlined in scan()
674     */
675    private static final int xorShift(int r) {
676        r ^= r << 13;
677        r ^= r >>> 17;
678        return r ^ (r << 5);
679    }
680
681    /**
682     * Tries to steal a task from another worker. Starts at a random
683     * index of workers array, and probes workers until finding one
684     * with non-empty queue or finding that all are empty.  It
685     * randomly selects the first n probes. If these are empty, it
686     * resorts to a circular sweep, which is necessary to accurately
687     * set active status. (The circular sweep uses steps of
688     * approximately half the array size plus 1, to avoid bias
689     * stemming from leftmost packing of the array in ForkJoinPool.)
690     *
691     * This method must be both fast and quiet -- usually avoiding
692     * memory accesses that could disrupt cache sharing etc other than
693     * those needed to check for and take tasks (or to activate if not
694     * already active). This accounts for, among other things,
695     * updating random seed in place without storing it until exit.
696     *
697     * @return a task, or null if none found
698     */
699    private ForkJoinTask<?> scan() {
700        ForkJoinPool p = pool;
701        ForkJoinWorkerThread[] ws;        // worker array
702        int n;                            // upper bound of #workers
703        if ((ws = p.workers) != null && (n = ws.length) > 1) {
704            boolean canSteal = active;    // shadow active status
705            int r = seed;                 // extract seed once
706            int mask = n - 1;
707            int j = -n;                   // loop counter
708            int k = r;                    // worker index, random if j < 0
709            for (;;) {
710                ForkJoinWorkerThread v = ws[k & mask];
711                r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift
712                ForkJoinTask<?>[] q; ForkJoinTask<?> t; int b, a;
713                if (v != null && (b = v.base) != v.sp &&
714                    (q = v.queue) != null) {
715                    int i = (q.length - 1) & b;
716                    long u = (i << qShift) + qBase; // raw offset
717                    int pid = poolIndex;
718                    if ((t = q[i]) != null) {
719                        if (!canSteal &&  // inline p.tryIncrementActiveCount
720                            UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
721                                                     a = p.runState, a + 1))
722                            canSteal = active = true;
723                        if (canSteal && v.base == b++ &&
724                            UNSAFE.compareAndSwapObject(q, u, t, null)) {
725                            v.base = b;
726                            v.stealHint = pid;
727                            UNSAFE.putOrderedObject(this,
728                                                    currentStealOffset, t);
729                            seed = r;
730                            ++stealCount;
731                            return t;
732                        }
733                    }
734                    j = -n;
735                    k = r;                // restart on contention
736                }
737                else if (++j <= 0)
738                    k = r;
739                else if (j <= n)
740                    k += (n >>> 1) | 1;
741                else
742                    break;
743            }
744        }
745        return null;
746    }
747
748    // Run State management
749
750    // status check methods used mainly by ForkJoinPool
751    final boolean isRunning()     { return runState == 0; }
752    final boolean isTerminating() { return (runState & TERMINATING) != 0; }
753    final boolean isTerminated()  { return (runState & TERMINATED) != 0; }
754    final boolean isSuspended()   { return (runState & SUSPENDED) != 0; }
755    final boolean isTrimmed()     { return (runState & TRIMMED) != 0; }
756
757    /**
758     * Sets state to TERMINATING. Does NOT unpark or interrupt
759     * to wake up if currently blocked. Callers must do so if desired.
760     */
761    final void shutdown() {
762        for (;;) {
763            int s = runState;
764            if ((s & (TERMINATING|TERMINATED)) != 0)
765                break;
766            if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended
767                if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
768                                             (s & ~SUSPENDED) |
769                                             (TRIMMED|TERMINATING)))
770                    break;
771            }
772            else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
773                                              s | TERMINATING))
774                break;
775        }
776    }
777
778    /**
779     * Sets state to TERMINATED. Called only by onTermination()
780     */
781    private void setTerminated() {
782        int s;
783        do {} while (!UNSAFE.compareAndSwapInt(this, runStateOffset,
784                                               s = runState,
785                                               s | (TERMINATING|TERMINATED)));
786    }
787
788    /**
789     * If suspended, tries to set status to unsuspended.
790     * Does NOT wake up if blocked.
791     *
792     * @return true if successful
793     */
794    final boolean tryUnsuspend() {
795        int s;
796        while (((s = runState) & SUSPENDED) != 0) {
797            if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
798                                         s & ~SUSPENDED))
799                return true;
800        }
801        return false;
802    }
803
804    /**
805     * Sets suspended status and blocks as spare until resumed
806     * or shutdown.
807     */
808    final void suspendAsSpare() {
809        for (;;) {                  // set suspended unless terminating
810            int s = runState;
811            if ((s & TERMINATING) != 0) { // must kill
812                if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
813                                             s | (TRIMMED | TERMINATING)))
814                    return;
815            }
816            else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s,
817                                              s | SUSPENDED))
818                break;
819        }
820        ForkJoinPool p = pool;
821        p.pushSpare(this);
822        while ((runState & SUSPENDED) != 0) {
823            if (p.tryAccumulateStealCount(this)) {
824                interrupted();          // clear/ignore interrupts
825                if ((runState & SUSPENDED) == 0)
826                    break;
827                LockSupport.park(this);
828            }
829        }
830    }
831
832    // Misc support methods for ForkJoinPool
833
834    /**
835     * Returns an estimate of the number of tasks in the queue.  Also
836     * used by ForkJoinTask.
837     */
838    final int getQueueSize() {
839        int n; // external calls must read base first
840        return (n = -base + sp) <= 0 ? 0 : n;
841    }
842
843    /**
844     * Removes and cancels all tasks in queue.  Can be called from any
845     * thread.
846     */
847    final void cancelTasks() {
848        ForkJoinTask<?> cj = currentJoin; // try to cancel ongoing tasks
849        if (cj != null) {
850            currentJoin = null;
851            cj.cancelIgnoringExceptions();
108              try {
109 <                this.interrupt(); // awaken wait
110 <            } catch (SecurityException ignore) {
111 <            }
112 <        }
113 <        ForkJoinTask<?> cs = currentSteal;
114 <        if (cs != null) {
859 <            currentSteal = null;
860 <            cs.cancelIgnoringExceptions();
861 <        }
862 <        while (base != sp) {
863 <            ForkJoinTask<?> t = deqTask();
864 <            if (t != null)
865 <                t.cancelIgnoringExceptions();
866 <        }
867 <    }
868 <
869 <    /**
870 <     * Drains tasks to given collection c.
871 <     *
872 <     * @return the number of tasks drained
873 <     */
874 <    final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
875 <        int n = 0;
876 <        while (base != sp) {
877 <            ForkJoinTask<?> t = deqTask();
878 <            if (t != null) {
879 <                c.add(t);
880 <                ++n;
881 <            }
882 <        }
883 <        return n;
884 <    }
885 <
886 <    // Support methods for ForkJoinTask
887 <
888 <    /**
889 <     * Gets and removes a local task.
890 <     *
891 <     * @return a task, if available
892 <     */
893 <    final ForkJoinTask<?> pollLocalTask() {
894 <        ForkJoinPool p = pool;
895 <        while (sp != base) {
896 <            int a; // inline p.tryIncrementActiveCount
897 <            if (active ||
898 <                (active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset,
899 <                                                   a = p.runState, a + 1)))
900 <                return locallyFifo ? locallyDeqTask() : popTask();
901 <        }
902 <        return null;
903 <    }
904 <
905 <    /**
906 <     * Gets and removes a local or stolen task.
907 <     *
908 <     * @return a task, if available
909 <     */
910 <    final ForkJoinTask<?> pollTask() {
911 <        ForkJoinTask<?> t = pollLocalTask();
912 <        if (t == null) {
913 <            t = scan();
914 <            // cannot retain/track/help steal
915 <            UNSAFE.putOrderedObject(this, currentStealOffset, null);
916 <        }
917 <        return t;
918 <    }
919 <
920 <    /**
921 <     * Possibly runs some tasks and/or blocks, until task is done.
922 <     *
923 <     * @param joinMe the task to join
924 <     */
925 <    final void joinTask(ForkJoinTask<?> joinMe) {
926 <        // currentJoin only written by this thread; only need ordered store
927 <        ForkJoinTask<?> prevJoin = currentJoin;
928 <        UNSAFE.putOrderedObject(this, currentJoinOffset, joinMe);
929 <        if (sp != base)
930 <            localHelpJoinTask(joinMe);
931 <        if (joinMe.status >= 0)
932 <            pool.awaitJoin(joinMe, this);
933 <        UNSAFE.putOrderedObject(this, currentJoinOffset, prevJoin);
934 <    }
935 <
936 <    /**
937 <     * Run tasks in local queue until given task is done.
938 <     *
939 <     * @param joinMe the task to join
940 <     */
941 <    private void localHelpJoinTask(ForkJoinTask<?> joinMe) {
942 <        int s;
943 <        ForkJoinTask<?>[] q;
944 <        while (joinMe.status >= 0 && (s = sp) != base && (q = queue) != null) {
945 <            int i = (q.length - 1) & --s;
946 <            long u = (i << qShift) + qBase; // raw offset
947 <            ForkJoinTask<?> t = q[i];
948 <            if (t == null)  // lost to a stealer
949 <                break;
950 <            if (UNSAFE.compareAndSwapObject(q, u, t, null)) {
951 <                /*
952 <                 * This recheck (and similarly in helpJoinTask)
953 <                 * handles cases where joinMe is independently
954 <                 * cancelled or forced even though there is other work
955 <                 * available. Back out of the pop by putting t back
956 <                 * into slot before we commit by writing sp.
957 <                 */
958 <                if (joinMe.status < 0) {
959 <                    UNSAFE.putObjectVolatile(q, u, t);
960 <                    break;
961 <                }
962 <                sp = s;
963 <                // UNSAFE.putOrderedInt(this, spOffset, s);
964 <                t.quietlyExec();
965 <            }
966 <        }
967 <    }
968 <
969 <    /**
970 <     * Unless terminating, tries to locate and help perform tasks for
971 <     * a stealer of the given task, or in turn one of its stealers.
972 <     * Traces currentSteal->currentJoin links looking for a thread
973 <     * working on a descendant of the given task and with a non-empty
974 <     * queue to steal back and execute tasks from.
975 <     *
976 <     * The implementation is very branchy to cope with potential
977 <     * inconsistencies or loops encountering chains that are stale,
978 <     * unknown, or of length greater than MAX_HELP_DEPTH links.  All
979 <     * of these cases are dealt with by just returning back to the
980 <     * caller, who is expected to retry if other join mechanisms also
981 <     * don't work out.
982 <     *
983 <     * @param joinMe the task to join
984 <     */
985 <    final void helpJoinTask(ForkJoinTask<?> joinMe) {
986 <        ForkJoinWorkerThread[] ws;
987 <        int n;
988 <        if (joinMe.status < 0)                // already done
989 <            return;
990 <        if ((runState & TERMINATING) != 0) {  // cancel if shutting down
991 <            joinMe.cancelIgnoringExceptions();
992 <            return;
993 <        }
994 <        if ((ws = pool.workers) == null || (n = ws.length) <= 1)
995 <            return;                           // need at least 2 workers
996 <
997 <        ForkJoinTask<?> task = joinMe;        // base of chain
998 <        ForkJoinWorkerThread thread = this;   // thread with stolen task
999 <        for (int d = 0; d < MAX_HELP_DEPTH; ++d) { // chain length
1000 <            // Try to find v, the stealer of task, by first using hint
1001 <            ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)];
1002 <            if (v == null || v.currentSteal != task) {
1003 <                for (int j = 0; ; ++j) {      // search array
1004 <                    if (j < n) {
1005 <                        ForkJoinTask<?> vs;
1006 <                        if ((v = ws[j]) != null &&
1007 <                            (vs = v.currentSteal) != null) {
1008 <                            if (joinMe.status < 0 || task.status < 0)
1009 <                                return;       // stale or done
1010 <                            if (vs == task) {
1011 <                                thread.stealHint = j;
1012 <                                break;        // save hint for next time
1013 <                            }
1014 <                        }
1015 <                    }
1016 <                    else
1017 <                        return;               // no stealer
1018 <                }
1019 <            }
1020 <            for (;;) { // Try to help v, using specialized form of deqTask
1021 <                if (joinMe.status < 0)
1022 <                    return;
1023 <                int b = v.base;
1024 <                ForkJoinTask<?>[] q = v.queue;
1025 <                if (b == v.sp || q == null)
1026 <                    break;
1027 <                int i = (q.length - 1) & b;
1028 <                long u = (i << qShift) + qBase;
1029 <                ForkJoinTask<?> t = q[i];
1030 <                int pid = poolIndex;
1031 <                ForkJoinTask<?> ps = currentSteal;
1032 <                if (task.status < 0)
1033 <                    return;                   // stale or done
1034 <                if (t != null && v.base == b++ &&
1035 <                    UNSAFE.compareAndSwapObject(q, u, t, null)) {
1036 <                    if (joinMe.status < 0) {
1037 <                        UNSAFE.putObjectVolatile(q, u, t);
1038 <                        return;               // back out on cancel
1039 <                    }
1040 <                    v.base = b;
1041 <                    v.stealHint = pid;
1042 <                    UNSAFE.putOrderedObject(this, currentStealOffset, t);
1043 <                    t.quietlyExec();
1044 <                    UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1045 <                }
1046 <            }
1047 <            // Try to descend to find v's stealer
1048 <            ForkJoinTask<?> next = v.currentJoin;
1049 <            if (task.status < 0 || next == null || next == task ||
1050 <                joinMe.status < 0)
1051 <                return;
1052 <            task = next;
1053 <            thread = v;
1054 <        }
1055 <    }
1056 <
1057 <    /**
1058 <     * Returns an estimate of the number of tasks, offset by a
1059 <     * function of number of idle workers.
1060 <     *
1061 <     * This method provides a cheap heuristic guide for task
1062 <     * partitioning when programmers, frameworks, tools, or languages
1063 <     * have little or no idea about task granularity.  In essence by
1064 <     * offering this method, we ask users only about tradeoffs in
1065 <     * overhead vs expected throughput and its variance, rather than
1066 <     * how finely to partition tasks.
1067 <     *
1068 <     * In a steady state strict (tree-structured) computation, each
1069 <     * thread makes available for stealing enough tasks for other
1070 <     * threads to remain active. Inductively, if all threads play by
1071 <     * the same rules, each thread should make available only a
1072 <     * constant number of tasks.
1073 <     *
1074 <     * The minimum useful constant is just 1. But using a value of 1
1075 <     * would require immediate replenishment upon each steal to
1076 <     * maintain enough tasks, which is infeasible.  Further,
1077 <     * partitionings/granularities of offered tasks should minimize
1078 <     * steal rates, which in general means that threads nearer the top
1079 <     * of computation tree should generate more than those nearer the
1080 <     * bottom. In perfect steady state, each thread is at
1081 <     * approximately the same level of computation tree. However,
1082 <     * producing extra tasks amortizes the uncertainty of progress and
1083 <     * diffusion assumptions.
1084 <     *
1085 <     * So, users will want to use values larger, but not much larger
1086 <     * than 1 to both smooth over transient shortages and hedge
1087 <     * against uneven progress; as traded off against the cost of
1088 <     * extra task overhead. We leave the user to pick a threshold
1089 <     * value to compare with the results of this call to guide
1090 <     * decisions, but recommend values such as 3.
1091 <     *
1092 <     * When all threads are active, it is on average OK to estimate
1093 <     * surplus strictly locally. In steady-state, if one thread is
1094 <     * maintaining say 2 surplus tasks, then so are others. So we can
1095 <     * just use estimated queue length (although note that (sp - base)
1096 <     * can be an overestimate because of stealers lagging increments
1097 <     * of base).  However, this strategy alone leads to serious
1098 <     * mis-estimates in some non-steady-state conditions (ramp-up,
1099 <     * ramp-down, other stalls). We can detect many of these by
1100 <     * further considering the number of "idle" threads, that are
1101 <     * known to have zero queued tasks, so compensate by a factor of
1102 <     * (#idle/#active) threads.
1103 <     */
1104 <    final int getEstimatedSurplusTaskCount() {
1105 <        return sp - base - pool.idlePerActive();
1106 <    }
1107 <
1108 <    /**
1109 <     * Runs tasks until {@code pool.isQuiescent()}.
1110 <     */
1111 <    final void helpQuiescePool() {
1112 <        ForkJoinTask<?> ps = currentSteal; // to restore below
1113 <        for (;;) {
1114 <            ForkJoinTask<?> t = pollLocalTask();
1115 <            if (t != null || (t = scan()) != null)
1116 <                t.quietlyExec();
1117 <            else {
1118 <                ForkJoinPool p = pool;
1119 <                int a; // to inline CASes
1120 <                if (active) {
1121 <                    if (!UNSAFE.compareAndSwapInt
1122 <                        (p, poolRunStateOffset, a = p.runState, a - 1))
1123 <                        continue;   // retry later
1124 <                    active = false; // inactivate
1125 <                    UNSAFE.putOrderedObject(this, currentStealOffset, ps);
1126 <                }
1127 <                if (p.isQuiescent()) {
1128 <                    active = true; // re-activate
1129 <                    do {} while (!UNSAFE.compareAndSwapInt
1130 <                                 (p, poolRunStateOffset, a = p.runState, a+1));
1131 <                    return;
1132 <                }
1133 <            }
1134 <        }
1135 <    }
1136 <
1137 <    // Unsafe mechanics
1138 <
1139 <    private static final sun.misc.Unsafe UNSAFE = getUnsafe();
1140 <    private static final long spOffset =
1141 <        objectFieldOffset("sp", ForkJoinWorkerThread.class);
1142 <    private static final long runStateOffset =
1143 <        objectFieldOffset("runState", ForkJoinWorkerThread.class);
1144 <    private static final long currentJoinOffset =
1145 <        objectFieldOffset("currentJoin", ForkJoinWorkerThread.class);
1146 <    private static final long currentStealOffset =
1147 <        objectFieldOffset("currentSteal", ForkJoinWorkerThread.class);
1148 <    private static final long qBase =
1149 <        UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
1150 <    private static final long poolRunStateOffset = // to inline CAS
1151 <        objectFieldOffset("runState", ForkJoinPool.class);
1152 <
1153 <    private static final int qShift;
1154 <
1155 <    static {
1156 <        int s = UNSAFE.arrayIndexScale(ForkJoinTask[].class);
1157 <        if ((s & (s-1)) != 0)
1158 <            throw new Error("data type scale not a power of two");
1159 <        qShift = 31 - Integer.numberOfLeadingZeros(s);
1160 <    }
1161 <
1162 <    private static long objectFieldOffset(String field, Class<?> klazz) {
1163 <        try {
1164 <            return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
1165 <        } catch (NoSuchFieldException e) {
1166 <            // Convert Exception to corresponding Error
1167 <            NoSuchFieldError error = new NoSuchFieldError(field);
1168 <            error.initCause(e);
1169 <            throw error;
1170 <        }
1171 <    }
1172 <
1173 <    /**
1174 <     * Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
1175 <     * Replace with a simple call to Unsafe.getUnsafe when integrating
1176 <     * into a jdk.
1177 <     *
1178 <     * @return a sun.misc.Unsafe
1179 <     */
1180 <    private static sun.misc.Unsafe getUnsafe() {
1181 <        try {
1182 <            return sun.misc.Unsafe.getUnsafe();
1183 <        } catch (SecurityException se) {
1184 <            try {
1185 <                return java.security.AccessController.doPrivileged
1186 <                    (new java.security
1187 <                     .PrivilegedExceptionAction<sun.misc.Unsafe>() {
1188 <                        public sun.misc.Unsafe run() throws Exception {
1189 <                            java.lang.reflect.Field f = sun.misc
1190 <                                .Unsafe.class.getDeclaredField("theUnsafe");
1191 <                            f.setAccessible(true);
1192 <                            return (sun.misc.Unsafe) f.get(null);
1193 <                        }});
1194 <            } catch (java.security.PrivilegedActionException e) {
1195 <                throw new RuntimeException("Could not initialize intrinsics",
1196 <                                           e.getCause());
109 >                onTermination(exception);
110 >            } catch (Throwable ex) {
111 >                if (exception == null)
112 >                    exception = ex;
113 >            } finally {
114 >                pool.deregisterWorker(this, exception);
115              }
116          }
117      }
118   }
119 +

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines