ViewVC Help

View File | Revision Log | Show Annotations | Download File | Root Listing

root/jsr166/jsr166/src/jsr166y/ForkJoinWorkerThread.java

Comparing jsr166/src/jsr166y/ForkJoinWorkerThread.java (file contents):
Revision 1.23 by dl, Fri Jul 31 16:27:08 2009 UTC vs.
Revision 1.72 by dl, Mon Nov 19 18:12:42 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.Collection;
9	>	import java.util.concurrent.atomic.AtomicInteger;
10
11		/**
12	<	* A thread managed by a {@link ForkJoinPool}.  This class is
13	<	* subclassable solely for the sake of adding functionality -- there
14	<	* are no overridable methods dealing with scheduling or
15	<	* execution. However, you can override initialization and termination
16	<	* methods surrounding the main task processing loop.  If you do
17	<	* create such a subclass, you will also need to supply a custom
18	<	* ForkJoinWorkerThreadFactory to use it in a ForkJoinPool.
12	>	* A thread managed by a {@link ForkJoinPool}, which executes
13	>	* {@link ForkJoinTask}s.
14	>	* This class is subclassable solely for the sake of adding
15	>	* functionality -- there are no overridable methods dealing with
16	>	* scheduling or execution.  However, you can override initialization
17	>	* and termination methods surrounding the main task processing loop.
18	>	* If you do create such a subclass, you will also need to supply a
19	>	* custom {@link ForkJoinPool.ForkJoinWorkerThreadFactory} to use it
20	>	* in a {@code ForkJoinPool}.
21		*
22		* @since 1.7
23		* @author Doug Lea
24		*/
25		public class ForkJoinWorkerThread extends Thread {
26		/*
27	<	* Algorithm overview:
28	<	*
29	<	* 1. Work-Stealing: Work-stealing queues are special forms of
30	<	* Deques that support only three of the four possible
31	<	* end-operations -- push, pop, and deq (aka steal), and only do
32	<	* so under the constraints that push and pop are called only from
33	<	* the owning thread, while deq may be called from other threads.
34	<	* (If you are unfamiliar with them, you probably want to read
35	<	* Herlihy and Shavit's book "The Art of Multiprocessor
36	<	* programming", chapter 16 describing these in more detail before
37	<	* proceeding.)  The main work-stealing queue design is roughly
38	<	* similar to "Dynamic Circular Work-Stealing Deque" by David
39	<	* Chase and Yossi Lev, SPAA 2005
40	<	* (http://research.sun.com/scalable/pubs/index.html).  The main
41	<	* difference ultimately stems from gc requirements that we null
42	<	* out taken slots as soon as we can, to maintain as small a
43	<	* footprint as possible even in programs generating huge numbers
44	<	* of tasks. To accomplish this, we shift the CAS arbitrating pop
45	<	* vs deq (steal) from being on the indices ("base" and "sp") to
46	<	* the slots themselves (mainly via method "casSlotNull()"). So,
47	<	* both a successful pop and deq mainly entail CAS'ing a non-null
48	<	* slot to null.  Because we rely on CASes of references, we do
49	<	* not need tag bits on base or sp.  They are simple ints as used
50	<	* in any circular array-based queue (see for example ArrayDeque).
51	<	* Updates to the indices must still be ordered in a way that
52	<	* guarantees that (sp - base) > 0 means the queue is empty, but
53	<	* otherwise may err on the side of possibly making the queue
54	<	* appear nonempty when a push, pop, or deq have not fully
55	<	* committed. Note that this means that the deq operation,
56	<	* considered individually, is not wait-free. One thief cannot
57	<	* successfully continue until another in-progress one (or, if
58	<	* previously empty, a push) completes.  However, in the
59	<	* aggregate, we ensure at least probabilistic non-blockingness. If
60	<	* an attempted steal fails, a thief always chooses a different
61	<	* random victim target to try next. So, in order for one thief to
62	<	* progress, it suffices for any in-progress deq or new push on
63	<	* any empty queue to complete. One reason this works well here is
64	<	* that apparently-nonempty often means soon-to-be-stealable,
65	<	* which gives threads a chance to activate if necessary before
66	<	* stealing (see below).
67	<	*
68	<	* This approach also enables support for "async mode" where local
69	<	* task processing is in FIFO, not LIFO order; simply by using a
70	<	* version of deq rather than pop when locallyFifo is true (as set
71	<	* by the ForkJoinPool).  This allows use in message-passing
72	<	* frameworks in which tasks are never joined.
73	<	*
74	<	* Efficient implementation of this approach currently relies on
75	<	* an uncomfortable amount of "Unsafe" mechanics. To maintain
76	<	* correct orderings, reads and writes of variable base require
77	<	* volatile ordering.  Variable sp does not require volatile write
78	<	* but needs cheaper store-ordering on writes.  Because they are
79	<	* protected by volatile base reads, reads of the queue array and
80	<	* its slots do not need volatile load semantics, but writes (in
81	<	* push) require store order and CASes (in pop and deq) require
82	<	* (volatile) CAS semantics. Since these combinations aren't
83	<	* supported using ordinary volatiles, the only way to accomplish
84	<	* these efficiently is to use direct Unsafe calls. (Using external
85	<	* AtomicIntegers and AtomicReferenceArrays for the indices and
86	<	* array is significantly slower because of memory locality and
87	<	* indirection effects.) Further, performance on most platforms is
88	<	* very sensitive to placement and sizing of the (resizable) queue
89	<	* array.  Even though these queues don't usually become all that
90	<	* big, the initial size must be large enough to counteract cache
91	<	* contention effects across multiple queues (especially in the
92	<	* presence of GC cardmarking). Also, to improve thread-locality,
93	<	* queues are currently initialized immediately after the thread
94	<	* gets the initial signal to start processing tasks.  However,
95	<	* all queue-related methods except pushTask are written in a way
96	<	* that allows them to instead be lazily allocated and/or disposed
97	<	* of when empty. All together, these low-level implementation
98	<	* choices produce as much as a factor of 4 performance
99	<	* improvement compared to naive implementations, and enable the
100	<	* processing of billions of tasks per second, sometimes at the
101	<	* expense of ugliness.
27	>	* ForkJoinWorkerThreads are managed by ForkJoinPools and perform
28	>	* ForkJoinTasks. For explanation, see the internal documentation
29	>	* of class ForkJoinPool.
30		*
31	<	* 2. Run control: The primary run control is based on a global
32	<	* counter (activeCount) held by the pool. It uses an algorithm
33	<	* similar to that in Herlihy and Shavit section 17.6 to cause
34	<	* threads to eventually block when all threads declare they are
35	<	* inactive. (See variable "scans".)  For this to work, threads
36	<	* must be declared active when executing tasks, and before
37	<	* stealing a task. They must be inactive before blocking on the
110	<	* Pool Barrier (awaiting a new submission or other Pool
111	<	* event). In between, there is some free play which we take
112	<	* advantage of to avoid contention and rapid flickering of the
113	<	* global activeCount: If inactive, we activate only if a victim
114	<	* queue appears to be nonempty (see above).  Similarly, a thread
115	<	* tries to inactivate only after a full scan of other threads.
116	<	* The net effect is that contention on activeCount is rarely a
117	<	* measurable performance issue. (There are also a few other cases
118	<	* where we scan for work rather than retry/block upon
119	<	* contention.)
120	<	*
121	<	* 3. Selection control. We maintain policy of always choosing to
122	<	* run local tasks rather than stealing, and always trying to
123	<	* steal tasks before trying to run a new submission. All steals
124	<	* are currently performed in randomly-chosen deq-order. It may be
125	<	* worthwhile to bias these with locality / anti-locality
126	<	* information, but doing this well probably requires more
127	<	* lower-level information from JVMs than currently provided.
31	>	* This class just maintains links to its pool and WorkQueue.  The
32	>	* pool field is set upon construction, but the workQueue field is
33	>	* not set until the thread has started (unless forced early by a
34	>	* subclass constructor call to poolIndex()).  This provides
35	>	* better memory placement (because this thread allocates queue
36	>	* and bookkeeping fields) but because the field is non-final, we
37	>	* require that it never be accessed except by the owning thread.
38		*/
39
40	<	/**
41	<	* Capacity of work-stealing queue array upon initialization.
132	<	* Must be a power of two. Initial size must be at least 2, but is
133	<	* padded to minimize cache effects.
134	<	*/
135	<	private static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
40	>	final ForkJoinPool pool;                // the pool this thread works in
41	>	ForkJoinPool.WorkQueue workQueue;       // Work-stealing mechanics
42
43		/**
44	<	* Maximum work-stealing queue array size.  Must be less than or
139	<	* equal to 1 << 28 to ensure lack of index wraparound. (This
140	<	* is less than usual bounds, because we need leftshift by 3
141	<	* to be in int range).
44	>	* Sequence number for creating worker Names
45		*/
46	<	private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
144	<
145	<	/**
146	<	* The pool this thread works in. Accessed directly by ForkJoinTask.
147	<	*/
148	<	final ForkJoinPool pool;
149	<
150	<	/**
151	<	* The work-stealing queue array. Size must be a power of two.
152	<	* Initialized when thread starts, to improve memory locality.
153	<	*/
154	<	private ForkJoinTask<?>[] queue;
155	<
156	<	/**
157	<	* Index (mod queue.length) of next queue slot to push to or pop
158	<	* from. It is written only by owner thread, via ordered store.
159	<	* Both sp and base are allowed to wrap around on overflow, but
160	<	* (sp - base) still estimates size.
161	<	*/
162	<	private volatile int sp;
163	<
164	<	/**
165	<	* Index (mod queue.length) of least valid queue slot, which is
166	<	* always the next position to steal from if nonempty.
167	<	*/
168	<	private volatile int base;
169	<
170	<	/**
171	<	* Activity status. When true, this worker is considered active.
172	<	* Must be false upon construction. It must be true when executing
173	<	* tasks, and BEFORE stealing a task. It must be false before
174	<	* calling pool.sync.
175	<	*/
176	<	private boolean active;
177	<
178	<	/**
179	<	* Run state of this worker. Supports simple versions of the usual
180	<	* shutdown/shutdownNow control.
181	<	*/
182	<	private volatile int runState;
183	<
184	<	/**
185	<	* Seed for random number generator for choosing steal victims.
186	<	* Uses Marsaglia xorshift. Must be nonzero upon initialization.
187	<	*/
188	<	private int seed;
189	<
190	<	/**
191	<	* Number of steals, transferred to pool when idle
192	<	*/
193	<	private int stealCount;
194	<
195	<	/**
196	<	* Index of this worker in pool array. Set once by pool before
197	<	* running, and accessed directly by pool during cleanup etc.
198	<	*/
199	<	int poolIndex;
200	<
201	<	/**
202	<	* The last barrier event waited for. Accessed in pool callback
203	<	* methods, but only by current thread.
204	<	*/
205	<	long lastEventCount;
206	<
207	<	/**
208	<	* True if use local fifo, not default lifo, for local polling
209	<	*/
210	<	private boolean locallyFifo;
46	>	private static final AtomicInteger threadNumber = new AtomicInteger();
47
48		/**
49		* Creates a ForkJoinWorkerThread operating in the given pool.
#	Line 216 | Line 52 | public class ForkJoinWorkerThread extend
52		* @throws NullPointerException if pool is null
53		*/
54		protected ForkJoinWorkerThread(ForkJoinPool pool) {
55	<	if (pool == null) throw new NullPointerException();
55	>	super(pool.workerNamePrefix.concat(Integer.toString(threadNumber.incrementAndGet())));
56	>	setDaemon(true);
57		this.pool = pool;
58	<	// Note: poolIndex is set by pool during construction
59	<	// Remaining initialization is deferred to onStart
58	>	Thread.UncaughtExceptionHandler ueh = pool.ueh;
59	>	if (ueh != null)
60	>	setUncaughtExceptionHandler(ueh);
61		}
62
225	–	// Public access methods
226	–
63		/**
64		* Returns the pool hosting this thread.
65		*
#	Line 243 | Line 79 | public class ForkJoinWorkerThread extend
79		* @return the index number
80		*/
81		public int getPoolIndex() {
82	<	return poolIndex;
83	<	}
84	<
85	<	/**
86	<	* Establishes local first-in-first-out scheduling mode for forked
251	<	* tasks that are never joined.
252	<	*
253	<	* @param async if true, use locally FIFO scheduling
254	<	*/
255	<	void setAsyncMode(boolean async) {
256	<	locallyFifo = async;
257	<	}
258	<
259	<	// Runstate management
260	<
261	<	// Runstate values. Order matters
262	<	private static final int RUNNING     = 0;
263	<	private static final int SHUTDOWN    = 1;
264	<	private static final int TERMINATING = 2;
265	<	private static final int TERMINATED  = 3;
266	<
267	<	final boolean isShutdown()    { return runState >= SHUTDOWN;  }
268	<	final boolean isTerminating() { return runState >= TERMINATING;  }
269	<	final boolean isTerminated()  { return runState == TERMINATED; }
270	<	final boolean shutdown()      { return transitionRunStateTo(SHUTDOWN); }
271	<	final boolean shutdownNow()   { return transitionRunStateTo(TERMINATING); }
272	<
273	<	/**
274	<	* Transitions to at least the given state.
275	<	*
276	<	* @return {@code true} if not already at least at given state
277	<	*/
278	<	private boolean transitionRunStateTo(int state) {
279	<	for (;;) {
280	<	int s = runState;
281	<	if (s >= state)
282	<	return false;
283	<	if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, state))
284	<	return true;
285	<	}
286	<	}
287	<
288	<	/**
289	<	* Tries to set status to active; fails on contention.
290	<	*/
291	<	private boolean tryActivate() {
292	<	if (!active) {
293	<	if (!pool.tryIncrementActiveCount())
294	<	return false;
295	<	active = true;
296	<	}
297	<	return true;
298	<	}
299	<
300	<	/**
301	<	* Tries to set status to inactive; fails on contention.
302	<	*/
303	<	private boolean tryInactivate() {
304	<	if (active) {
305	<	if (!pool.tryDecrementActiveCount())
306	<	return false;
307	<	active = false;
308	<	}
309	<	return true;
310	<	}
311	<
312	<	/**
313	<	* Computes next value for random victim probe.  Scans don't
314	<	* require a very high quality generator, but also not a crummy
315	<	* one.  Marsaglia xor-shift is cheap and works well.
316	<	*/
317	<	private static int xorShift(int r) {
318	<	r ^= (r << 13);
319	<	r ^= (r >>> 17);
320	<	return r ^ (r << 5);
321	<	}
322	<
323	<	// Lifecycle methods
324	<
325	<	/**
326	<	* This method is required to be public, but should never be
327	<	* called explicitly. It performs the main run loop to execute
328	<	* ForkJoinTasks.
329	<	*/
330	<	public void run() {
331	<	Throwable exception = null;
332	<	try {
333	<	onStart();
334	<	pool.sync(this); // await first pool event
335	<	mainLoop();
336	<	} catch (Throwable ex) {
337	<	exception = ex;
338	<	} finally {
339	<	onTermination(exception);
340	<	}
341	<	}
342	<
343	<	/**
344	<	* Executes tasks until shut down.
345	<	*/
346	<	private void mainLoop() {
347	<	while (!isShutdown()) {
348	<	ForkJoinTask<?> t = pollTask();
349	<	if (t != null || (t = pollSubmission()) != null)
350	<	t.quietlyExec();
351	<	else if (tryInactivate())
352	<	pool.sync(this);
82	>	// force early registration if called before started
83	>	ForkJoinPool.WorkQueue q;
84	>	if ((q = workQueue) == null) {
85	>	pool.registerWorker(this);
86	>	q = workQueue;
87		}
88	+	return q.poolIndex;
89		}
90
91		/**
92		* Initializes internal state after construction but before
93		* processing any tasks. If you override this method, you must
94	<	* invoke super.onStart() at the beginning of the method.
94	>	* invoke {@code super.onStart()} at the beginning of the method.
95		* Initialization requires care: Most fields must have legal
96		* default values, to ensure that attempted accesses from other
97		* threads work correctly even before this thread starts
98		* processing tasks.
99		*/
100		protected void onStart() {
366	–	// Allocate while starting to improve chances of thread-local
367	–	// isolation
368	–	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
369	–	// Initial value of seed need not be especially random but
370	–	// should differ across workers and must be nonzero
371	–	int p = poolIndex + 1;
372	–	seed = p + (p << 8) + (p << 16) + (p << 24); // spread bits
101		}
102
103		/**
#	Line 381 | Line 109 | public class ForkJoinWorkerThread extend
109		* to an unrecoverable error, or {@code null} if completed normally
110		*/
111		protected void onTermination(Throwable exception) {
384	–	// Execute remaining local tasks unless aborting or terminating
385	–	while (exception == null &&  !pool.isTerminating() && base != sp) {
386	–	try {
387	–	ForkJoinTask<?> t = popTask();
388	–	if (t != null)
389	–	t.quietlyExec();
390	–	} catch (Throwable ex) {
391	–	exception = ex;
392	–	}
393	–	}
394	–	// Cancel other tasks, transition status, notify pool, and
395	–	// propagate exception to uncaught exception handler
396	–	try {
397	–	do {} while (!tryInactivate()); // ensure inactive
398	–	cancelTasks();
399	–	runState = TERMINATED;
400	–	pool.workerTerminated(this);
401	–	} catch (Throwable ex) {        // Shouldn't ever happen
402	–	if (exception == null)      // but if so, at least rethrown
403	–	exception = ex;
404	–	} finally {
405	–	if (exception != null)
406	–	ForkJoinTask.rethrowException(exception);
407	–	}
408	–	}
409	–
410	–	// Intrinsics-based support for queue operations.
411	–
412	–	/**
413	–	* Adds in store-order the given task at given slot of q to null.
414	–	* Caller must ensure q is non-null and index is in range.
415	–	*/
416	–	private static void setSlot(ForkJoinTask<?>[] q, int i,
417	–	ForkJoinTask<?> t) {
418	–	UNSAFE.putOrderedObject(q, (i << qShift) + qBase, t);
419	–	}
420	–
421	–	/**
422	–	* CAS given slot of q to null. Caller must ensure q is non-null
423	–	* and index is in range.
424	–	*/
425	–	private static boolean casSlotNull(ForkJoinTask<?>[] q, int i,
426	–	ForkJoinTask<?> t) {
427	–	return UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
428	–	}
429	–
430	–	/**
431	–	* Sets sp in store-order.
432	–	*/
433	–	private void storeSp(int s) {
434	–	UNSAFE.putOrderedInt(this, spOffset, s);
112		}
113
437	–	// Main queue methods
438	–
114		/**
115	<	* Pushes a task. Called only by current thread.
116	<	*
117	<	* @param t the task. Caller must ensure non-null.
443	<	*/
444	<	final void pushTask(ForkJoinTask<?> t) {
445	<	ForkJoinTask<?>[] q = queue;
446	<	int mask = q.length - 1;
447	<	int s = sp;
448	<	setSlot(q, s & mask, t);
449	<	storeSp(++s);
450	<	if ((s -= base) == 1)
451	<	pool.signalWork();
452	<	else if (s >= mask)
453	<	growQueue();
454	<	}
455	<
456	<	/**
457	<	* Tries to take a task from the base of the queue, failing if
458	<	* either empty or contended.
459	<	*
460	<	* @return a task, or null if none or contended
461	<	*/
462	<	final ForkJoinTask<?> deqTask() {
463	<	ForkJoinTask<?> t;
464	<	ForkJoinTask<?>[] q;
465	<	int i;
466	<	int b;
467	<	if (sp != (b = base) &&
468	<	(q = queue) != null && // must read q after b
469	<	(t = q[i = (q.length - 1) & b]) != null &&
470	<	casSlotNull(q, i, t)) {
471	<	base = b + 1;
472	<	return t;
473	<	}
474	<	return null;
475	<	}
476	<
477	<	/**
478	<	* Tries to take a task from the base of own queue, activating if
479	<	* necessary, failing only if empty. Called only by current thread.
480	<	*
481	<	* @return a task, or null if none
482	<	*/
483	<	final ForkJoinTask<?> locallyDeqTask() {
484	<	int b;
485	<	while (sp != (b = base)) {
486	<	if (tryActivate()) {
487	<	ForkJoinTask<?>[] q = queue;
488	<	int i = (q.length - 1) & b;
489	<	ForkJoinTask<?> t = q[i];
490	<	if (t != null && casSlotNull(q, i, t)) {
491	<	base = b + 1;
492	<	return t;
493	<	}
494	<	}
495	<	}
496	<	return null;
497	<	}
498	<
499	<	/**
500	<	* Returns a popped task, or null if empty. Ensures active status
501	<	* if non-null. Called only by current thread.
502	<	*/
503	<	final ForkJoinTask<?> popTask() {
504	<	int s = sp;
505	<	while (s != base) {
506	<	if (tryActivate()) {
507	<	ForkJoinTask<?>[] q = queue;
508	<	int mask = q.length - 1;
509	<	int i = (s - 1) & mask;
510	<	ForkJoinTask<?> t = q[i];
511	<	if (t == null || !casSlotNull(q, i, t))
512	<	break;
513	<	storeSp(s - 1);
514	<	return t;
515	<	}
516	<	}
517	<	return null;
518	<	}
519	<
520	<	/**
521	<	* Specialized version of popTask to pop only if
522	<	* topmost element is the given task. Called only
523	<	* by current thread while active.
524	<	*
525	<	* @param t the task. Caller must ensure non-null.
526	<	*/
527	<	final boolean unpushTask(ForkJoinTask<?> t) {
528	<	ForkJoinTask<?>[] q = queue;
529	<	int mask = q.length - 1;
530	<	int s = sp - 1;
531	<	if (casSlotNull(q, s & mask, t)) {
532	<	storeSp(s);
533	<	return true;
534	<	}
535	<	return false;
536	<	}
537	<
538	<	/**
539	<	* Returns next task or null if empty or contended
540	<	*/
541	<	final ForkJoinTask<?> peekTask() {
542	<	ForkJoinTask<?>[] q = queue;
543	<	if (q == null)
544	<	return null;
545	<	int mask = q.length - 1;
546	<	int i = locallyFifo ? base : (sp - 1);
547	<	return q[i & mask];
548	<	}
549	<
550	<	/**
551	<	* Doubles queue array size. Transfers elements by emulating
552	<	* steals (deqs) from old array and placing, oldest first, into
553	<	* new array.
554	<	*/
555	<	private void growQueue() {
556	<	ForkJoinTask<?>[] oldQ = queue;
557	<	int oldSize = oldQ.length;
558	<	int newSize = oldSize << 1;
559	<	if (newSize > MAXIMUM_QUEUE_CAPACITY)
560	<	throw new RejectedExecutionException("Queue capacity exceeded");
561	<	ForkJoinTask<?>[] newQ = queue = new ForkJoinTask<?>[newSize];
562	<
563	<	int b = base;
564	<	int bf = b + oldSize;
565	<	int oldMask = oldSize - 1;
566	<	int newMask = newSize - 1;
567	<	do {
568	<	int oldIndex = b & oldMask;
569	<	ForkJoinTask<?> t = oldQ[oldIndex];
570	<	if (t != null && !casSlotNull(oldQ, oldIndex, t))
571	<	t = null;
572	<	setSlot(newQ, b & newMask, t);
573	<	} while (++b != bf);
574	<	pool.signalWork();
575	<	}
576	<
577	<	/**
578	<	* Tries to steal a task from another worker. Starts at a random
579	<	* index of workers array, and probes workers until finding one
580	<	* with non-empty queue or finding that all are empty.  It
581	<	* randomly selects the first n probes. If these are empty, it
582	<	* resorts to a full circular traversal, which is necessary to
583	<	* accurately set active status by caller. Also restarts if pool
584	<	* events occurred since last scan, which forces refresh of
585	<	* workers array, in case barrier was associated with resize.
586	<	*
587	<	* This method must be both fast and quiet -- usually avoiding
588	<	* memory accesses that could disrupt cache sharing etc other than
589	<	* those needed to check for and take tasks. This accounts for,
590	<	* among other things, updating random seed in place without
591	<	* storing it until exit.
592	<	*
593	<	* @return a task, or null if none found
594	<	*/
595	<	private ForkJoinTask<?> scan() {
596	<	ForkJoinTask<?> t = null;
597	<	int r = seed;                    // extract once to keep scan quiet
598	<	ForkJoinWorkerThread[] ws;       // refreshed on outer loop
599	<	int mask;                        // must be power 2 minus 1 and > 0
600	<	outer:do {
601	<	if ((ws = pool.workers) != null && (mask = ws.length - 1) > 0) {
602	<	int idx = r;
603	<	int probes = ~mask;      // use random index while negative
604	<	for (;;) {
605	<	r = xorShift(r);     // update random seed
606	<	ForkJoinWorkerThread v = ws[mask & idx];
607	<	if (v == null || v.sp == v.base) {
608	<	if (probes <= mask)
609	<	idx = (probes++ < 0) ? r : (idx + 1);
610	<	else
611	<	break;
612	<	}
613	<	else if (!tryActivate() || (t = v.deqTask()) == null)
614	<	continue outer;  // restart on contention
615	<	else
616	<	break outer;
617	<	}
618	<	}
619	<	} while (pool.hasNewSyncEvent(this)); // retry on pool events
620	<	seed = r;
621	<	return t;
622	<	}
623	<
624	<	/**
625	<	* Gets and removes a local or stolen task.
626	<	*
627	<	* @return a task, if available
628	<	*/
629	<	final ForkJoinTask<?> pollTask() {
630	<	ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask();
631	<	if (t == null && (t = scan()) != null)
632	<	++stealCount;
633	<	return t;
634	<	}
635	<
636	<	/**
637	<	* Gets a local task.
638	<	*
639	<	* @return a task, if available
640	<	*/
641	<	final ForkJoinTask<?> pollLocalTask() {
642	<	return locallyFifo ? locallyDeqTask() : popTask();
643	<	}
644	<
645	<	/**
646	<	* Returns a pool submission, if one exists, activating first.
647	<	*
648	<	* @return a submission, if available
649	<	*/
650	<	private ForkJoinTask<?> pollSubmission() {
651	<	ForkJoinPool p = pool;
652	<	while (p.hasQueuedSubmissions()) {
653	<	ForkJoinTask<?> t;
654	<	if (tryActivate() && (t = p.pollSubmission()) != null)
655	<	return t;
656	<	}
657	<	return null;
658	<	}
659	<
660	<	// Methods accessed only by Pool
661	<
662	<	/**
663	<	* Removes and cancels all tasks in queue.  Can be called from any
664	<	* thread.
665	<	*/
666	<	final void cancelTasks() {
667	<	ForkJoinTask<?> t;
668	<	while (base != sp && (t = deqTask()) != null)
669	<	t.cancelIgnoringExceptions();
670	<	}
671	<
672	<	/**
673	<	* Drains tasks to given collection c.
674	<	*
675	<	* @return the number of tasks drained
676	<	*/
677	<	final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
678	<	int n = 0;
679	<	ForkJoinTask<?> t;
680	<	while (base != sp && (t = deqTask()) != null) {
681	<	c.add(t);
682	<	++n;
683	<	}
684	<	return n;
685	<	}
686	<
687	<	/**
688	<	* Gets and clears steal count for accumulation by pool.  Called
689	<	* only when known to be idle (in pool.sync and termination).
690	<	*/
691	<	final int getAndClearStealCount() {
692	<	int sc = stealCount;
693	<	stealCount = 0;
694	<	return sc;
695	<	}
696	<
697	<	/**
698	<	* Returns {@code true} if at least one worker in the given array
699	<	* appears to have at least one queued task.
700	<	*
701	<	* @param ws array of workers
702	<	*/
703	<	static boolean hasQueuedTasks(ForkJoinWorkerThread[] ws) {
704	<	if (ws != null) {
705	<	int len = ws.length;
706	<	for (int j = 0; j < 2; ++j) { // need two passes for clean sweep
707	<	for (int i = 0; i < len; ++i) {
708	<	ForkJoinWorkerThread w = ws[i];
709	<	if (w != null && w.sp != w.base)
710	<	return true;
711	<	}
712	<	}
713	<	}
714	<	return false;
715	<	}
716	<
717	<	// Support methods for ForkJoinTask
718	<
719	<	/**
720	<	* Returns an estimate of the number of tasks in the queue.
721	<	*/
722	<	final int getQueueSize() {
723	<	// suppress momentarily negative values
724	<	return Math.max(0, sp - base);
725	<	}
726	<
727	<	/**
728	<	* Returns an estimate of the number of tasks, offset by a
729	<	* function of number of idle workers.
730	<	*/
731	<	final int getEstimatedSurplusTaskCount() {
732	<	// The halving approximates weighting idle vs non-idle workers
733	<	return (sp - base) - (pool.getIdleThreadCount() >>> 1);
734	<	}
735	<
736	<	/**
737	<	* Scans, returning early if joinMe done.
738	<	*/
739	<	final ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
740	<	ForkJoinTask<?> t = pollTask();
741	<	if (t != null && joinMe.status < 0 && sp == base) {
742	<	pushTask(t); // unsteal if done and this task would be stealable
743	<	t = null;
744	<	}
745	<	return t;
746	<	}
747	<
748	<	/**
749	<	* Runs tasks until {@code pool.isQuiescent()}.
750	<	*/
751	<	final void helpQuiescePool() {
752	<	for (;;) {
753	<	ForkJoinTask<?> t = pollTask();
754	<	if (t != null)
755	<	t.quietlyExec();
756	<	else if (tryInactivate() && pool.isQuiescent())
757	<	break;
758	<	}
759	<	do {} while (!tryActivate()); // re-activate on exit
760	<	}
761	<
762	<	// Unsafe mechanics
763	<
764	<	private static final sun.misc.Unsafe UNSAFE = getUnsafe();
765	<	private static final long spOffset =
766	<	objectFieldOffset("sp", ForkJoinWorkerThread.class);
767	<	private static final long runStateOffset =
768	<	objectFieldOffset("runState", ForkJoinWorkerThread.class);
769	<	private static final long qBase;
770	<	private static final int qShift;
771	<
772	<	static {
773	<	qBase = UNSAFE.arrayBaseOffset(ForkJoinTask[].class);
774	<	int s = UNSAFE.arrayIndexScale(ForkJoinTask[].class);
775	<	if ((s & (s-1)) != 0)
776	<	throw new Error("data type scale not a power of two");
777	<	qShift = 31 - Integer.numberOfLeadingZeros(s);
778	<	}
779	<
780	<	private static long objectFieldOffset(String field, Class<?> klazz) {
781	<	try {
782	<	return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field));
783	<	} catch (NoSuchFieldException e) {
784	<	// Convert Exception to corresponding Error
785	<	NoSuchFieldError error = new NoSuchFieldError(field);
786	<	error.initCause(e);
787	<	throw error;
788	<	}
789	<	}
790	<
791	<	/**
792	<	* Returns a sun.misc.Unsafe.  Suitable for use in a 3rd party package.
793	<	* Replace with a simple call to Unsafe.getUnsafe when integrating
794	<	* into a jdk.
795	<	*
796	<	* @return a sun.misc.Unsafe
115	>	* This method is required to be public, but should never be
116	>	* called explicitly. It performs the main run loop to execute
117	>	* {@link ForkJoinTask}s.
118		*/
119	<	private static sun.misc.Unsafe getUnsafe() {
119	>	public void run() {
120	>	Throwable exception = null;
121		try {
122	<	return sun.misc.Unsafe.getUnsafe();
123	<	} catch (SecurityException se) {
122	>	pool.registerWorker(this);
123	>	onStart();
124	>	pool.runWorker(workQueue);
125	>	} catch (Throwable ex) {
126	>	exception = ex;
127	>	} finally {
128		try {
129	<	return java.security.AccessController.doPrivileged
130	<	(new java.security
131	<	.PrivilegedExceptionAction<sun.misc.Unsafe>() {
132	<	public sun.misc.Unsafe run() throws Exception {
133	<	java.lang.reflect.Field f = sun.misc
134	<	.Unsafe.class.getDeclaredField("theUnsafe");
809	<	f.setAccessible(true);
810	<	return (sun.misc.Unsafe) f.get(null);
811	<	}});
812	<	} catch (java.security.PrivilegedActionException e) {
813	<	throw new RuntimeException("Could not initialize intrinsics",
814	<	e.getCause());
129	>	onTermination(exception);
130	>	} catch (Throwable ex) {
131	>	if (exception == null)
132	>	exception = ex;
133	>	} finally {
134	>	pool.deregisterWorker(this, exception);
135		}
136		}
137		}

Diff Legend

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