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 1.9 by jsr166, Mon Jul 20 22:26:03 2009 UTC vs.
Revision 1.73 by dl, Wed Nov 21 19:54:39 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	–	import java.util.*;
9	–	import java.util.concurrent.*;
10	–	import java.util.concurrent.atomic.*;
11	–	import java.util.concurrent.locks.*;
12	–	import sun.misc.Unsafe;
13	–	import java.lang.reflect.*;
8
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
13	<	* execution. However, you can override initialization and termination
14	<	* methods surrounding the main task processing loop.  If you do
15	<	* create such a subclass, you will also need to supply a custom
16	<	* ForkJoinWorkerThreadFactory to use it in a 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		/*
25	<	* Algorithm overview:
26	<	*
27	<	* 1. Work-Stealing: Work-stealing queues are special forms of
28	<	* Deques that support only three of the four possible
29	<	* end-operations -- push, pop, and deq (aka steal), and only do
30	<	* so under the constraints that push and pop are called only from
31	<	* the owning thread, while deq may be called from other threads.
32	<	* (If you are unfamiliar with them, you probably want to read
33	<	* Herlihy and Shavit's book "The Art of Multiprocessor
34	<	* 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 nonnull
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	<	* Efficient implementation of this approach currently relies on
69	<	* an uncomfortable amount of "Unsafe" mechanics. To maintain
70	<	* correct orderings, reads and writes of variable base require
71	<	* volatile ordering.  Variable sp does not require volatile write
72	<	* but needs cheaper store-ordering on writes.  Because they are
73	<	* protected by volatile base reads, reads of the queue array and
74	<	* its slots do not need volatile load semantics, but writes (in
75	<	* push) require store order and CASes (in pop and deq) require
76	<	* (volatile) CAS semantics. Since these combinations aren't
77	<	* supported using ordinary volatiles, the only way to accomplish
78	<	* these efficiently is to use direct Unsafe calls. (Using external
79	<	* AtomicIntegers and AtomicReferenceArrays for the indices and
80	<	* array is significantly slower because of memory locality and
81	<	* indirection effects.) Further, performance on most platforms is
82	<	* very sensitive to placement and sizing of the (resizable) queue
83	<	* array.  Even though these queues don't usually become all that
84	<	* big, the initial size must be large enough to counteract cache
85	<	* contention effects across multiple queues (especially in the
86	<	* presence of GC cardmarking). Also, to improve thread-locality,
87	<	* queues are currently initialized immediately after the thread
88	<	* gets the initial signal to start processing tasks.  However,
89	<	* all queue-related methods except pushTask are written in a way
90	<	* that allows them to instead be lazily allocated and/or disposed
91	<	* of when empty. All together, these low-level implementation
92	<	* choices produce as much as a factor of 4 performance
93	<	* improvement compared to naive implementations, and enable the
94	<	* processing of billions of tasks per second, sometimes at the
95	<	* expense of ugliness.
96	<	*
97	<	* 2. Run control: The primary run control is based on a global
98	<	* counter (activeCount) held by the pool. It uses an algorithm
99	<	* similar to that in Herlihy and Shavit section 17.6 to cause
100	<	* threads to eventually block when all threads declare they are
101	<	* inactive. (See variable "scans".)  For this to work, threads
102	<	* must be declared active when executing tasks, and before
103	<	* stealing a task. They must be inactive before blocking on the
104	<	* Pool Barrier (awaiting a new submission or other Pool
105	<	* event). In between, there is some free play which we take
106	<	* advantage of to avoid contention and rapid flickering of the
107	<	* global activeCount: If inactive, we activate only if a victim
108	<	* queue appears to be nonempty (see above).  Similarly, a thread
109	<	* tries to inactivate only after a full scan of other threads.
110	<	* The net effect is that contention on activeCount is rarely a
111	<	* measurable performance issue. (There are also a few other cases
112	<	* where we scan for work rather than retry/block upon
113	<	* contention.)
114	<	*
115	<	* 3. Selection control. We maintain policy of always choosing to
116	<	* run local tasks rather than stealing, and always trying to
117	<	* steal tasks before trying to run a new submission. All steals
118	<	* are currently performed in randomly-chosen deq-order. It may be
119	<	* worthwhile to bias these with locality / anti-locality
120	<	* information, but doing this well probably requires more
121	<	* lower-level information from JVMs than currently provided.
122	<	*/
123	<
124	<	/**
125	<	* Capacity of work-stealing queue array upon initialization.
126	<	* Must be a power of two. Initial size must be at least 2, but is
127	<	* padded to minimize cache effects.
128	<	*/
129	<	private static final int INITIAL_QUEUE_CAPACITY = 1 << 13;
130	<
131	<	/**
132	<	* Maximum work-stealing queue array size.  Must be less than or
133	<	* equal to 1 << 28 to ensure lack of index wraparound. (This
134	<	* is less than usual bounds, because we need leftshift by 3
135	<	* to be in int range).
136	<	*/
137	<	private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
138	<
139	<	/**
140	<	* The pool this thread works in. Accessed directly by ForkJoinTask
141	<	*/
142	<	final ForkJoinPool pool;
143	<
144	<	/**
145	<	* The work-stealing queue array. Size must be a power of two.
146	<	* Initialized when thread starts, to improve memory locality.
147	<	*/
148	<	private ForkJoinTask<?>[] queue;
149	<
150	<	/**
151	<	* Index (mod queue.length) of next queue slot to push to or pop
152	<	* from. It is written only by owner thread, via ordered store.
153	<	* Both sp and base are allowed to wrap around on overflow, but
154	<	* (sp - base) still estimates size.
155	<	*/
156	<	private volatile int sp;
157	<
158	<	/**
159	<	* Index (mod queue.length) of least valid queue slot, which is
160	<	* always the next position to steal from if nonempty.
161	<	*/
162	<	private volatile int base;
163	<
164	<	/**
165	<	* Activity status. When true, this worker is considered active.
166	<	* Must be false upon construction. It must be true when executing
167	<	* tasks, and BEFORE stealing a task. It must be false before
168	<	* calling pool.sync
169	<	*/
170	<	private boolean active;
171	<
172	<	/**
173	<	* Run state of this worker. Supports simple versions of the usual
174	<	* shutdown/shutdownNow control.
175	<	*/
176	<	private volatile int runState;
177	<
178	<	/**
179	<	* Seed for random number generator for choosing steal victims.
180	<	* Uses Marsaglia xorshift. Must be nonzero upon initialization.
181	<	*/
182	<	private int seed;
183	<
184	<	/**
185	<	* Number of steals, transferred to pool when idle
25	>	* ForkJoinWorkerThreads are managed by ForkJoinPools and perform
26	>	* ForkJoinTasks. For explanation, see the internal documentation
27	>	* of class ForkJoinPool.
28	>	*
29	>	* This class just maintains links to its pool and WorkQueue.  The
30	>	* pool field is set immediately upon construction, but the
31	>	* workQueue field is not set until a call to registerWorker
32	>	* completes. This leads to a visibility race, that is tolerated
33	>	* by requiring that the workQueue field is only accessed by the
34	>	* owning thread.
35		*/
187	–	private int stealCount;
36
37	<	/**
38	<	* Index of this worker in pool array. Set once by pool before
191	<	* running, and accessed directly by pool during cleanup etc
192	<	*/
193	<	int poolIndex;
194	<
195	<	/**
196	<	* The last barrier event waited for. Accessed in pool callback
197	<	* methods, but only by current thread.
198	<	*/
199	<	long lastEventCount;
200	<
201	<	/**
202	<	* True if use local fifo, not default lifo, for local polling
203	<	*/
204	<	private boolean locallyFifo;
37	>	final ForkJoinPool pool;                // the pool this thread works in
38	>	final ForkJoinPool.WorkQueue workQueue; // work-stealing mechanics
39
40		/**
41		* Creates a ForkJoinWorkerThread operating in the given pool.
42	+	*
43		* @param pool the pool this thread works in
44		* @throws NullPointerException if pool is null
45		*/
46		protected ForkJoinWorkerThread(ForkJoinPool pool) {
47	<	if (pool == null) throw new NullPointerException();
47	>	// Use a placeholder until a useful name can be set in registerWorker
48	>	super("aForkJoinWorkerThread");
49		this.pool = pool;
50	<	// Note: poolIndex is set by pool during construction
215	<	// Remaining initialization is deferred to onStart
50	>	this.workQueue = pool.registerWorker(this);
51		}
52
218	–	// Public access methods
219	–
53		/**
54	<	* Returns the pool hosting this thread
54	>	* Returns the pool hosting this thread.
55	>	*
56		* @return the pool
57		*/
58		public ForkJoinPool getPool() {
#	Line 231 | Line 65 | public class ForkJoinWorkerThread extend
65		* threads (minus one) that have ever been created in the pool.
66		* This method may be useful for applications that track status or
67		* collect results per-worker rather than per-task.
68	<	* @return the index number.
68	>	*
69	>	* @return the index number
70		*/
71		public int getPoolIndex() {
72	<	return poolIndex;
238	<	}
239	<
240	<	/**
241	<	* Establishes local first-in-first-out scheduling mode for forked
242	<	* tasks that are never joined.
243	<	* @param async if true, use locally FIFO scheduling
244	<	*/
245	<	void setAsyncMode(boolean async) {
246	<	locallyFifo = async;
247	<	}
248	<
249	<	// Runstate management
250	<
251	<	// Runstate values. Order matters
252	<	private static final int RUNNING     = 0;
253	<	private static final int SHUTDOWN    = 1;
254	<	private static final int TERMINATING = 2;
255	<	private static final int TERMINATED  = 3;
256	<
257	<	final boolean isShutdown()    { return runState >= SHUTDOWN;  }
258	<	final boolean isTerminating() { return runState >= TERMINATING;  }
259	<	final boolean isTerminated()  { return runState == TERMINATED; }
260	<	final boolean shutdown()      { return transitionRunStateTo(SHUTDOWN); }
261	<	final boolean shutdownNow()   { return transitionRunStateTo(TERMINATING); }
262	<
263	<	/**
264	<	* Transition to at least the given state. Return true if not
265	<	* already at least given state.
266	<	*/
267	<	private boolean transitionRunStateTo(int state) {
268	<	for (;;) {
269	<	int s = runState;
270	<	if (s >= state)
271	<	return false;
272	<	if (_unsafe.compareAndSwapInt(this, runStateOffset, s, state))
273	<	return true;
274	<	}
275	<	}
276	<
277	<	/**
278	<	* Try to set status to active; fail on contention
279	<	*/
280	<	private boolean tryActivate() {
281	<	if (!active) {
282	<	if (!pool.tryIncrementActiveCount())
283	<	return false;
284	<	active = true;
285	<	}
286	<	return true;
287	<	}
288	<
289	<	/**
290	<	* Try to set status to active; fail on contention
291	<	*/
292	<	private boolean tryInactivate() {
293	<	if (active) {
294	<	if (!pool.tryDecrementActiveCount())
295	<	return false;
296	<	active = false;
297	<	}
298	<	return true;
299	<	}
300	<
301	<	/**
302	<	* Computes next value for random victim probe. Scans don't
303	<	* require a very high quality generator, but also not a crummy
304	<	* one. Marsaglia xor-shift is cheap and works well.
305	<	*/
306	<	private static int xorShift(int r) {
307	<	r ^= r << 1;
308	<	r ^= r >>> 3;
309	<	r ^= r << 10;
310	<	return r;
311	<	}
312	<
313	<	// Lifecycle methods
314	<
315	<	/**
316	<	* This method is required to be public, but should never be
317	<	* called explicitly. It performs the main run loop to execute
318	<	* ForkJoinTasks.
319	<	*/
320	<	public void run() {
321	<	Throwable exception = null;
322	<	try {
323	<	onStart();
324	<	pool.sync(this); // await first pool event
325	<	mainLoop();
326	<	} catch (Throwable ex) {
327	<	exception = ex;
328	<	} finally {
329	<	onTermination(exception);
330	<	}
331	<	}
332	<
333	<	/**
334	<	* Execute tasks until shut down.
335	<	*/
336	<	private void mainLoop() {
337	<	while (!isShutdown()) {
338	<	ForkJoinTask<?> t = pollTask();
339	<	if (t != null || (t = pollSubmission()) != null)
340	<	t.quietlyExec();
341	<	else if (tryInactivate())
342	<	pool.sync(this);
343	<	}
72	>	return workQueue.poolIndex;
73		}
74
75		/**
76		* Initializes internal state after construction but before
77		* processing any tasks. If you override this method, you must
78	<	* invoke super.onStart() at the beginning of the method.
78	>	* invoke {@code super.onStart()} at the beginning of the method.
79		* Initialization requires care: Most fields must have legal
80		* default values, to ensure that attempted accesses from other
81		* threads work correctly even before this thread starts
82		* processing tasks.
83		*/
84		protected void onStart() {
356	–	// Allocate while starting to improve chances of thread-local
357	–	// isolation
358	–	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
359	–	// Initial value of seed need not be especially random but
360	–	// should differ across workers and must be nonzero
361	–	int p = poolIndex + 1;
362	–	seed = p + (p << 8) + (p << 16) + (p << 24); // spread bits
85		}
86
87		/**
88	<	* Perform cleanup associated with termination of this worker
88	>	* Performs cleanup associated with termination of this worker
89		* thread.  If you override this method, you must invoke
90	<	* super.onTermination at the end of the overridden method.
90	>	* {@code super.onTermination} at the end of the overridden method.
91		*
92		* @param exception the exception causing this thread to abort due
93	<	* to an unrecoverable error, or null if completed normally.
93	>	* to an unrecoverable error, or {@code null} if completed normally
94		*/
95		protected void onTermination(Throwable exception) {
374	–	// Execute remaining local tasks unless aborting or terminating
375	–	while (exception == null &&  !pool.isTerminating() && base != sp) {
376	–	try {
377	–	ForkJoinTask<?> t = popTask();
378	–	if (t != null)
379	–	t.quietlyExec();
380	–	} catch(Throwable ex) {
381	–	exception = ex;
382	–	}
383	–	}
384	–	// Cancel other tasks, transition status, notify pool, and
385	–	// propagate exception to uncaught exception handler
386	–	try {
387	–	do;while (!tryInactivate()); // ensure inactive
388	–	cancelTasks();
389	–	runState = TERMINATED;
390	–	pool.workerTerminated(this);
391	–	} catch (Throwable ex) {        // Shouldn't ever happen
392	–	if (exception == null)      // but if so, at least rethrown
393	–	exception = ex;
394	–	} finally {
395	–	if (exception != null)
396	–	ForkJoinTask.rethrowException(exception);
397	–	}
398	–	}
399	–
400	–	// Intrinsics-based support for queue operations.
401	–
402	–	/**
403	–	* Add in store-order the given task at given slot of q to
404	–	* null. Caller must ensure q is nonnull and index is in range.
405	–	*/
406	–	private static void setSlot(ForkJoinTask<?>[] q, int i,
407	–	ForkJoinTask<?> t){
408	–	_unsafe.putOrderedObject(q, (i << qShift) + qBase, t);
409	–	}
410	–
411	–	/**
412	–	* CAS given slot of q to null. Caller must ensure q is nonnull
413	–	* and index is in range.
414	–	*/
415	–	private static boolean casSlotNull(ForkJoinTask<?>[] q, int i,
416	–	ForkJoinTask<?> t) {
417	–	return _unsafe.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
96		}
97
98		/**
99	<	* Sets sp in store-order.
100	<	*/
101	<	private void storeSp(int s) {
424	<	_unsafe.putOrderedInt(this, spOffset, s);
425	<	}
426	<
427	<	// Main queue methods
428	<
429	<	/**
430	<	* Pushes a task. Called only by current thread.
431	<	* @param t the task. Caller must ensure nonnull
432	<	*/
433	<	final void pushTask(ForkJoinTask<?> t) {
434	<	ForkJoinTask<?>[] q = queue;
435	<	int mask = q.length - 1;
436	<	int s = sp;
437	<	setSlot(q, s & mask, t);
438	<	storeSp(++s);
439	<	if ((s -= base) == 1)
440	<	pool.signalWork();
441	<	else if (s >= mask)
442	<	growQueue();
443	<	}
444	<
445	<	/**
446	<	* Tries to take a task from the base of the queue, failing if
447	<	* either empty or contended.
448	<	* @return a task, or null if none or contended.
449	<	*/
450	<	final ForkJoinTask<?> deqTask() {
451	<	ForkJoinTask<?> t;
452	<	ForkJoinTask<?>[] q;
453	<	int i;
454	<	int b;
455	<	if (sp != (b = base) &&
456	<	(q = queue) != null && // must read q after b
457	<	(t = q[i = (q.length - 1) & b]) != null &&
458	<	casSlotNull(q, i, t)) {
459	<	base = b + 1;
460	<	return t;
461	<	}
462	<	return null;
463	<	}
464	<
465	<	/**
466	<	* Returns a popped task, or null if empty. Ensures active status
467	<	* if nonnull. Called only by current thread.
468	<	*/
469	<	final ForkJoinTask<?> popTask() {
470	<	int s = sp;
471	<	while (s != base) {
472	<	if (tryActivate()) {
473	<	ForkJoinTask<?>[] q = queue;
474	<	int mask = q.length - 1;
475	<	int i = (s - 1) & mask;
476	<	ForkJoinTask<?> t = q[i];
477	<	if (t == null || !casSlotNull(q, i, t))
478	<	break;
479	<	storeSp(s - 1);
480	<	return t;
481	<	}
482	<	}
483	<	return null;
484	<	}
485	<
486	<	/**
487	<	* Specialized version of popTask to pop only if
488	<	* topmost element is the given task. Called only
489	<	* by current thread while active.
490	<	* @param t the task. Caller must ensure nonnull
491	<	*/
492	<	final boolean unpushTask(ForkJoinTask<?> t) {
493	<	ForkJoinTask<?>[] q = queue;
494	<	int mask = q.length - 1;
495	<	int s = sp - 1;
496	<	if (casSlotNull(q, s & mask, t)) {
497	<	storeSp(s);
498	<	return true;
499	<	}
500	<	return false;
501	<	}
502	<
503	<	/**
504	<	* Returns next task.
505	<	*/
506	<	final ForkJoinTask<?> peekTask() {
507	<	ForkJoinTask<?>[] q = queue;
508	<	if (q == null)
509	<	return null;
510	<	int mask = q.length - 1;
511	<	int i = locallyFifo? base : (sp - 1);
512	<	return q[i & mask];
513	<	}
514	<
515	<	/**
516	<	* Doubles queue array size. Transfers elements by emulating
517	<	* steals (deqs) from old array and placing, oldest first, into
518	<	* new array.
519	<	*/
520	<	private void growQueue() {
521	<	ForkJoinTask<?>[] oldQ = queue;
522	<	int oldSize = oldQ.length;
523	<	int newSize = oldSize << 1;
524	<	if (newSize > MAXIMUM_QUEUE_CAPACITY)
525	<	throw new RejectedExecutionException("Queue capacity exceeded");
526	<	ForkJoinTask<?>[] newQ = queue = new ForkJoinTask<?>[newSize];
527	<
528	<	int b = base;
529	<	int bf = b + oldSize;
530	<	int oldMask = oldSize - 1;
531	<	int newMask = newSize - 1;
532	<	do {
533	<	int oldIndex = b & oldMask;
534	<	ForkJoinTask<?> t = oldQ[oldIndex];
535	<	if (t != null && !casSlotNull(oldQ, oldIndex, t))
536	<	t = null;
537	<	setSlot(newQ, b & newMask, t);
538	<	} while (++b != bf);
539	<	pool.signalWork();
540	<	}
541	<
542	<	/**
543	<	* Tries to steal a task from another worker. Starts at a random
544	<	* index of workers array, and probes workers until finding one
545	<	* with non-empty queue or finding that all are empty.  It
546	<	* randomly selects the first n probes. If these are empty, it
547	<	* resorts to a full circular traversal, which is necessary to
548	<	* accurately set active status by caller. Also restarts if pool
549	<	* events occurred since last scan, which forces refresh of
550	<	* workers array, in case barrier was associated with resize.
551	<	*
552	<	* This method must be both fast and quiet -- usually avoiding
553	<	* memory accesses that could disrupt cache sharing etc other than
554	<	* those needed to check for and take tasks. This accounts for,
555	<	* among other things, updating random seed in place without
556	<	* storing it until exit.
557	<	*
558	<	* @return a task, or null if none found
559	<	*/
560	<	private ForkJoinTask<?> scan() {
561	<	ForkJoinTask<?> t = null;
562	<	int r = seed;                    // extract once to keep scan quiet
563	<	ForkJoinWorkerThread[] ws;       // refreshed on outer loop
564	<	int mask;                        // must be power 2 minus 1 and > 0
565	<	outer:do {
566	<	if ((ws = pool.workers) != null && (mask = ws.length - 1) > 0) {
567	<	int idx = r;
568	<	int probes = ~mask;      // use random index while negative
569	<	for (;;) {
570	<	r = xorShift(r);     // update random seed
571	<	ForkJoinWorkerThread v = ws[mask & idx];
572	<	if (v == null || v.sp == v.base) {
573	<	if (probes <= mask)
574	<	idx = (probes++ < 0)? r : (idx + 1);
575	<	else
576	<	break;
577	<	}
578	<	else if (!tryActivate() || (t = v.deqTask()) == null)
579	<	continue outer;  // restart on contention
580	<	else
581	<	break outer;
582	<	}
583	<	}
584	<	} while (pool.hasNewSyncEvent(this)); // retry on pool events
585	<	seed = r;
586	<	return t;
587	<	}
588	<
589	<	/**
590	<	* gets and removes a local or stolen a task
591	<	* @return a task, if available
592	<	*/
593	<	final ForkJoinTask<?> pollTask() {
594	<	ForkJoinTask<?> t = locallyFifo? deqTask() : popTask();
595	<	if (t == null && (t = scan()) != null)
596	<	++stealCount;
597	<	return t;
598	<	}
599	<
600	<	/**
601	<	* gets a local task
602	<	* @return a task, if available
603	<	*/
604	<	final ForkJoinTask<?> pollLocalTask() {
605	<	return locallyFifo? deqTask() : popTask();
606	<	}
607	<
608	<	/**
609	<	* Returns a pool submission, if one exists, activating first.
610	<	* @return a submission, if available
611	<	*/
612	<	private ForkJoinTask<?> pollSubmission() {
613	<	ForkJoinPool p = pool;
614	<	while (p.hasQueuedSubmissions()) {
615	<	ForkJoinTask<?> t;
616	<	if (tryActivate() && (t = p.pollSubmission()) != null)
617	<	return t;
618	<	}
619	<	return null;
620	<	}
621	<
622	<	// Methods accessed only by Pool
623	<
624	<	/**
625	<	* Removes and cancels all tasks in queue.  Can be called from any
626	<	* thread.
627	<	*/
628	<	final void cancelTasks() {
629	<	ForkJoinTask<?> t;
630	<	while (base != sp && (t = deqTask()) != null)
631	<	t.cancelIgnoringExceptions();
632	<	}
633	<
634	<	/**
635	<	* Drains tasks to given collection c
636	<	* @return the number of tasks drained
637	<	*/
638	<	final int drainTasksTo(Collection<ForkJoinTask<?>> c) {
639	<	int n = 0;
640	<	ForkJoinTask<?> t;
641	<	while (base != sp && (t = deqTask()) != null) {
642	<	c.add(t);
643	<	++n;
644	<	}
645	<	return n;
646	<	}
647	<
648	<	/**
649	<	* Get and clear steal count for accumulation by pool.  Called
650	<	* only when known to be idle (in pool.sync and termination).
651	<	*/
652	<	final int getAndClearStealCount() {
653	<	int sc = stealCount;
654	<	stealCount = 0;
655	<	return sc;
656	<	}
657	<
658	<	/**
659	<	* Returns true if at least one worker in the given array appears
660	<	* to have at least one queued task.
661	<	* @param ws array of workers
662	<	*/
663	<	static boolean hasQueuedTasks(ForkJoinWorkerThread[] ws) {
664	<	if (ws != null) {
665	<	int len = ws.length;
666	<	for (int j = 0; j < 2; ++j) { // need two passes for clean sweep
667	<	for (int i = 0; i < len; ++i) {
668	<	ForkJoinWorkerThread w = ws[i];
669	<	if (w != null && w.sp != w.base)
670	<	return true;
671	<	}
672	<	}
673	<	}
674	<	return false;
675	<	}
676	<
677	<	// Support methods for ForkJoinTask
678	<
679	<	/**
680	<	* Returns an estimate of the number of tasks in the queue.
681	<	*/
682	<	final int getQueueSize() {
683	<	int n = sp - base;
684	<	return n < 0? 0 : n; // suppress momentarily negative values
685	<	}
686	<
687	<	/**
688	<	* Returns an estimate of the number of tasks, offset by a
689	<	* function of number of idle workers.
690	<	*/
691	<	final int getEstimatedSurplusTaskCount() {
692	<	// The halving approximates weighting idle vs non-idle workers
693	<	return (sp - base) - (pool.getIdleThreadCount() >>> 1);
694	<	}
695	<
696	<	/**
697	<	* Scan, returning early if joinMe done
698	<	*/
699	<	final ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
700	<	ForkJoinTask<?> t = pollTask();
701	<	if (t != null && joinMe.status < 0 && sp == base) {
702	<	pushTask(t); // unsteal if done and this task would be stealable
703	<	t = null;
704	<	}
705	<	return t;
706	<	}
707	<
708	<	/**
709	<	* Runs tasks until pool isQuiescent
99	>	* This method is required to be public, but should never be
100	>	* called explicitly. It performs the main run loop to execute
101	>	* {@link ForkJoinTask}s.
102		*/
103	<	final void helpQuiescePool() {
104	<	for (;;) {
713	<	ForkJoinTask<?> t = pollTask();
714	<	if (t != null)
715	<	t.quietlyExec();
716	<	else if (tryInactivate() && pool.isQuiescent())
717	<	break;
718	<	}
719	<	do;while (!tryActivate()); // re-activate on exit
720	<	}
721	<
722	<	// Temporary Unsafe mechanics for preliminary release
723	<	private static Unsafe getUnsafe() throws Throwable {
103	>	public void run() {
104	>	Throwable exception = null;
105		try {
106	<	return Unsafe.getUnsafe();
107	<	} catch (SecurityException se) {
106	>	onStart();
107	>	pool.runWorker(workQueue);
108	>	} catch (Throwable ex) {
109	>	exception = ex;
110	>	} finally {
111		try {
112	<	return java.security.AccessController.doPrivileged
113	<	(new java.security.PrivilegedExceptionAction<Unsafe>() {
114	<	public Unsafe run() throws Exception {
115	<	return getUnsafePrivileged();
116	<	}});
117	<	} catch (java.security.PrivilegedActionException e) {
734	<	throw e.getCause();
112	>	onTermination(exception);
113	>	} catch (Throwable ex) {
114	>	if (exception == null)
115	>	exception = ex;
116	>	} finally {
117	>	pool.deregisterWorker(this, exception);
118		}
119		}
120		}
738	–
739	–	private static Unsafe getUnsafePrivileged()
740	–	throws NoSuchFieldException, IllegalAccessException {
741	–	Field f = Unsafe.class.getDeclaredField("theUnsafe");
742	–	f.setAccessible(true);
743	–	return (Unsafe) f.get(null);
744	–	}
745	–
746	–	private static long fieldOffset(String fieldName)
747	–	throws NoSuchFieldException {
748	–	return _unsafe.objectFieldOffset
749	–	(ForkJoinWorkerThread.class.getDeclaredField(fieldName));
750	–	}
751	–
752	–	static final Unsafe _unsafe;
753	–	static final long baseOffset;
754	–	static final long spOffset;
755	–	static final long runStateOffset;
756	–	static final long qBase;
757	–	static final int qShift;
758	–	static {
759	–	try {
760	–	_unsafe = getUnsafe();
761	–	baseOffset = fieldOffset("base");
762	–	spOffset = fieldOffset("sp");
763	–	runStateOffset = fieldOffset("runState");
764	–	qBase = _unsafe.arrayBaseOffset(ForkJoinTask[].class);
765	–	int s = _unsafe.arrayIndexScale(ForkJoinTask[].class);
766	–	if ((s & (s-1)) != 0)
767	–	throw new Error("data type scale not a power of two");
768	–	qShift = 31 - Integer.numberOfLeadingZeros(s);
769	–	} catch (Throwable e) {
770	–	throw new RuntimeException("Could not initialize intrinsics", e);
771	–	}
772	–	}
121		}

Diff Legend

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