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 1.8 by jsr166, Mon Jul 20 21:45:06 2009 UTC vs.
Revision 1.43 by jsr166, Wed Sep 1 20:12:39 2010 UTC

#	Line 5 | Line 5
5		*/
6
7		package jsr166y;
8	<	import java.util.*;
8	>
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.*;
10	>
11	>	import java.util.Random;
12	>	import java.util.Collection;
13	>	import java.util.concurrent.locks.LockSupport;
14
15		/**
16		* A thread managed by a {@link ForkJoinPool}.  This class is
17		* subclassable solely for the sake of adding functionality -- there
18	<	* are no overridable methods dealing with scheduling or
19	<	* execution. However, you can override initialization and termination
20	<	* methods surrounding the main task processing loop.  If you do
21	<	* create such a subclass, you will also need to supply a custom
22	<	* ForkJoinWorkerThreadFactory to use it in a ForkJoinPool.
18	>	* are no overridable methods dealing with scheduling or execution.
19	>	* However, you can override initialization and termination methods
20	>	* surrounding the main task processing loop.  If you do create such a
21	>	* subclass, you will also need to supply a custom {@link
22	>	* ForkJoinPool.ForkJoinWorkerThreadFactory} to use it in a {@code
23	>	* ForkJoinPool}.
24		*
25	+	* @since 1.7
26	+	* @author Doug Lea
27		*/
28		public class ForkJoinWorkerThread extends Thread {
29		/*
30	<	* Algorithm overview:
30	>	* Overview:
31	>	*
32	>	* ForkJoinWorkerThreads are managed by ForkJoinPools and perform
33	>	* ForkJoinTasks. This class includes bookkeeping in support of
34	>	* 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	<	* 1. Work-Stealing: Work-stealing queues are special forms of
41	<	* Deques that support only three of the four possible
42	<	* end-operations -- push, pop, and deq (aka steal), and only do
43	<	* so under the constraints that push and pop are called only from
44	<	* the owning thread, while deq may be called from other threads.
45	<	* (If you are unfamiliar with them, you probably want to read
46	<	* Herlihy and Shavit's book "The Art of Multiprocessor
47	<	* programming", chapter 16 describing these in more detail before
48	<	* proceeding.)  The main work-stealing queue design is roughly
49	<	* similar to "Dynamic Circular Work-Stealing Deque" by David
50	<	* Chase and Yossi Lev, SPAA 2005
51	<	* (http://research.sun.com/scalable/pubs/index.html).  The main
52	<	* difference ultimately stems from gc requirements that we null
53	<	* out taken slots as soon as we can, to maintain as small a
54	<	* footprint as possible even in programs generating huge numbers
55	<	* of tasks. To accomplish this, we shift the CAS arbitrating pop
56	<	* vs deq (steal) from being on the indices ("base" and "sp") to
57	<	* the slots themselves (mainly via method "casSlotNull()"). So,
58	<	* both a successful pop and deq mainly entail CAS'ing a nonnull
59	<	* slot to null.  Because we rely on CASes of references, we do
60	<	* not need tag bits on base or sp.  They are simple ints as used
61	<	* in any circular array-based queue (see for example ArrayDeque).
62	<	* Updates to the indices must still be ordered in a way that
63	<	* guarantees that (sp - base) > 0 means the queue is empty, but
64	<	* otherwise may err on the side of possibly making the queue
65	<	* appear nonempty when a push, pop, or deq have not fully
66	<	* committed. Note that this means that the deq operation,
67	<	* considered individually, is not wait-free. One thief cannot
68	<	* successfully continue until another in-progress one (or, if
69	<	* previously empty, a push) completes.  However, in the
70	<	* aggregate, we ensure at least probablistic non-blockingness. If
71	<	* an attempted steal fails, a thief always chooses a different
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 activate if necessary before
78	<	* stealing (see below).
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 this approach currently relies on
116	<	* an uncomfortable amount of "Unsafe" mechanics. To maintain
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 write
119	<	* but needs cheaper store-ordering on writes.  Because they are
120	<	* protected by volatile base reads, reads of the queue array and
121	<	* its slots do not need volatile load semantics, but writes (in
122	<	* push) require store order and CASes (in pop and deq) require
123	<	* (volatile) CAS semantics. Since these combinations aren't
124	<	* supported using ordinary volatiles, the only way to accomplish
125	<	* these effciently is to use direct Unsafe calls. (Using external
126	<	* AtomicIntegers and AtomicReferenceArrays for the indices and
127	<	* array is significantly slower because of memory locality and
128	<	* indirection effects.) Further, performance on most platforms is
129	<	* very sensitive to placement and sizing of the (resizable) queue
130	<	* array.  Even though these queues don't usually become all that
131	<	* big, the initial size must be large enough to counteract cache
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 currently initialized immediately after the thread
142	<	* gets the initial signal to start processing tasks.  However,
143	<	* all queue-related methods except pushTask are written in a way
144	<	* that allows them to instead be lazily allocated and/or disposed
145	<	* 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.
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 2, but is
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;
#	Line 137 | Line 179 | public class ForkJoinWorkerThread extend
179		private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28;
180
181		/**
182	<	* The pool this thread works in. Accessed directly by ForkJoinTask
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 when thread starts, to improve memory locality.
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, via ordered store.
201	<	* Both sp and base are allowed to wrap around on overflow, but
202	<	* (sp - base) still estimates size.
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 volatile int sp;
205	>	private int sp;
206
207		/**
208	<	* Index (mod queue.length) of least valid queue slot, which is
209	<	* always the next position to steal from if nonempty.
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 volatile int base;
214	>	private int stealHint;
215
216		/**
217	<	* Activity status. When true, this worker is considered active.
218	<	* Must be false upon construction. It must be true when executing
219	<	* tasks, and BEFORE stealing a task. It must be false before
220	<	* calling pool.sync
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	<	private boolean active;
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	<	* Run state of this worker. Supports simple versions of the usual
234	<	* shutdown/shutdownNow control.
233	>	* Number of steals, transferred and reset in pool callbacks pool
234	>	* when idle Accessed directly by pool.
235		*/
236	<	private volatile int runState;
236	>	int stealCount;
237
238		/**
239		* Seed for random number generator for choosing steal victims.
240	<	* Uses Marsaglia xorshift. Must be nonzero upon initialization.
240	>	* Uses Marsaglia xorshift. Must be initialized as nonzero.
241		*/
242		private int seed;
243
244		/**
245	<	* Number of steals, transferred to pool when idle
245	>	* Activity status. When true, this worker is considered active.
246	>	* Accessed directly by pool.  Must be false upon construction.
247		*/
248	<	private int stealCount;
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 during cleanup etc
258	>	* running, and accessed directly by pool to locate this worker in
259	>	* its workers array.
260		*/
261		int poolIndex;
262
263		/**
264	<	* The last barrier event waited for. Accessed in pool callback
265	<	* methods, but only by current thread.
264	>	* 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	<	long lastEventCount;
279	>	int spareCount;
280
281		/**
282	<	* True if use local fifo, not default lifo, for local polling
282	>	* Encoded index and count of next spare waiter. Accessed only
283	>	* by ForkJoinPool for managing spares.
284		*/
285	<	private boolean locallyFifo;
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;
300
301		/**
302		* Creates a ForkJoinWorkerThread operating in the given pool.
303	+	*
304		* @param pool the pool this thread works in
305		* @throws NullPointerException if pool is null
306		*/
307		protected ForkJoinWorkerThread(ForkJoinPool pool) {
212	–	if (pool == null) throw new NullPointerException();
308		this.pool = pool;
309	<	// Note: poolIndex is set by pool during construction
310	<	// Remaining initialization is deferred to onStart
309	>	this.locallyFifo = pool.locallyFifo;
310	>	setDaemon(true);
311	>	// To avoid exposing construction details to subclasses,
312	>	// remaining initialization is in start() and onStart()
313		}
314
315	<	// Public access methods
315	>	/**
316	>	* Performs additional initialization and starts this thread
317	>	*/
318	>	final void start(int poolIndex, UncaughtExceptionHandler ueh) {
319	>	this.poolIndex = poolIndex;
320	>	if (ueh != null)
321	>	setUncaughtExceptionHandler(ueh);
322	>	start();
323	>	}
324	>
325	>	// Public/protected methods
326
327		/**
328	<	* Returns the pool hosting this thread
328	>	* Returns the pool hosting this thread.
329	>	*
330		* @return the pool
331		*/
332		public ForkJoinPool getPool() {
#	Line 231 | Line 339 | public class ForkJoinWorkerThread extend
339		* threads (minus one) that have ever been created in the pool.
340		* This method may be useful for applications that track status or
341		* collect results per-worker rather than per-task.
342	<	* @return the index number.
342	>	*
343	>	* @return the index number
344		*/
345		public int getPoolIndex() {
346		return poolIndex;
347		}
348
349		/**
350	<	* Establishes local first-in-first-out scheduling mode for forked
351	<	* tasks that are never joined.
352	<	* @param async if true, use locally FIFO scheduling
350	>	* Initializes internal state after construction but before
351	>	* processing any tasks. If you override this method, you must
352	>	* invoke super.onStart() at the beginning of the method.
353	>	* Initialization requires care: Most fields must have legal
354	>	* default values, to ensure that attempted accesses from other
355	>	* threads work correctly even before this thread starts
356	>	* processing tasks.
357		*/
358	<	void setAsyncMode(boolean async) {
359	<	locallyFifo = async;
360	<	}
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); }
358	>	protected void onStart() {
359	>	int rs = seedGenerator.nextInt();
360	>	seed = rs == 0? 1 : rs; // seed must be nonzero
361
362	<	/**
363	<	* Transition to at least the given state. Return true if not
364	<	* already at least given state.
365	<	*/
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	<	}
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	<	/**
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;
367	>	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
368		}
369
370		/**
371	<	* Try to set status to active; fail on contention
371	>	* Performs cleanup associated with termination of this worker
372	>	* thread.  If you override this method, you must invoke
373	>	* {@code super.onTermination} at the end of the overridden method.
374	>	*
375	>	* @param exception the exception causing this thread to abort due
376	>	* to an unrecoverable error, or {@code null} if completed normally
377		*/
378	<	private boolean tryInactivate() {
379	<	if (active) {
380	<	if (!pool.tryDecrementActiveCount())
381	<	return false;
382	<	active = false;
378	>	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		}
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;
397		}
398
313	–	// Lifecycle methods
314	–
399		/**
400		* This method is required to be public, but should never be
401		* called explicitly. It performs the main run loop to execute
#	Line 321 | Line 405 | public class ForkJoinWorkerThread extend
405		Throwable exception = null;
406		try {
407		onStart();
324	–	pool.sync(this); // await first pool event
408		mainLoop();
409		} catch (Throwable ex) {
410		exception = ex;
#	Line 330 | Line 413 | public class ForkJoinWorkerThread extend
413		}
414		}
415
416	+	// helpers for run()
417	+
418		/**
419	<	* Execute tasks until shut down.
419	>	* Find and execute tasks and check status while running
420		*/
421		private void mainLoop() {
422	<	while (!isShutdown()) {
423	<	ForkJoinTask<?> t = pollTask();
424	<	if (t != null || (t = pollSubmission()) != null)
425	<	t.quietlyExec();
426	<	else if (tryInactivate())
427	<	pool.sync(this);
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	<	* Initializes internal state after construction but before
434	<	* processing any tasks. If you override this method, you must
435	<	* invoke super.onStart() at the beginning of the method.
350	<	* Initialization requires care: Most fields must have legal
351	<	* default values, to ensure that attempted accesses from other
352	<	* threads work correctly even before this thread starts
353	<	* processing tasks.
433	>	* Try to steal a task and execute it
434	>	*
435	>	* @return true if ran a task
436		*/
437	<	protected void onStart() {
438	<	// Allocate while starting to improve chances of thread-local
439	<	// isolation
440	<	queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY];
441	<	// Initial value of seed need not be especially random but
442	<	// should differ across workers and must be nonzero
443	<	int p = poolIndex + 1;
444	<	seed = p + (p << 8) + (p << 16) + (p << 24); // spread bits
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	<	* Perform cleanup associated with termination of this worker
367	<	* thread.  If you override this method, you must invoke
368	<	* super.onTermination at the end of the overridden method.
450	>	* If a submission exists, try to activate and run it;
451		*
452	<	* @param exception the exception causing this thread to abort due
371	<	* to an unrecoverable error, or null if completed normally.
452	>	* @return true if ran a task
453		*/
454	<	protected void onTermination(Throwable exception) {
455	<	// Execute remaining local tasks unless aborting or terminating
456	<	while (exception == null &&  !pool.isTerminating() && base != sp) {
457	<	try {
458	<	ForkJoinTask<?> t = popTask();
459	<	if (t != null)
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	<	} catch(Throwable ex) {
465	<	exception = ex;
464	>	UNSAFE.putOrderedObject(this, currentStealOffset, null);
465	>	if (sp != base)
466	>	execLocalTasks();
467	>	return true;
468	>	}
469		}
470		}
471	<	// 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	<	}
471	>	return false;
472		}
473
400	–	// Intrinsics-based support for queue operations.
401	–
474		/**
475	<	* Add in store-order the given task at given slot of q to
476	<	* null. Caller must ensure q is nonnull and index is in range.
475	>	* Runs local tasks until queue is empty or shut down.  Call only
476	>	* while active.
477		*/
478	<	private static void setSlot(ForkJoinTask<?>[] q, int i,
479	<	ForkJoinTask<?> t){
480	<	_unsafe.putOrderedObject(q, (i << qShift) + qBase, t);
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	<	* CAS given slot of q to null. Caller must ensure q is nonnull
507	<	* and index is in range.
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 boolean casSlotNull(ForkJoinTask<?>[] q, int i,
510	<	ForkJoinTask<?> t) {
511	<	return _unsafe.compareAndSwapObject(q, (i << qShift) + qBase, t, null);
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	<	* Sets sp in store-order.
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 void storeSp(int s) {
520	<	_unsafe.putOrderedInt(this, spOffset, s);
519	>	private static final void writeSlot(ForkJoinTask<?>[] q, int i,
520	>	ForkJoinTask<?> t) {
521	>	UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t);
522		}
523
524	<	// Main queue methods
524	>	// queue methods
525
526		/**
527	<	* Pushes a task. Called only by current thread.
528	<	* @param t the task. Caller must ensure nonnull
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;
534	<	int s = sp;
535	<	setSlot(q, s & mask, t);
536	<	storeSp(++s);
537	<	if ((s -= base) == 1)
538	<	pool.signalWork();
539	<	else if (s >= mask)
442	<	growQueue();
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	<	* either empty or contended.
545	<	* @return a task, or null if none or contended.
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 i;
454	<	int b;
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 &&
556	<	casSlotNull(q, i, t)) {
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		}
#	Line 463 | Line 561 | public class ForkJoinWorkerThread extend
561		}
562
563		/**
564	<	* Returns a popped task, or null if empty. Ensures active status
565	<	* if nonnull. Called only by current thread.
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	<	final ForkJoinTask<?> popTask() {
591	<	int s = sp;
592	<	while (s != base) {
593	<	if (tryActivate()) {
594	<	ForkJoinTask<?>[] q = queue;
595	<	int mask = q.length - 1;
596	<	int i = (s - 1) & mask;
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 || !casSlotNull(q, i, t))
598	>	if (t == null)   // lost to stealer
599		break;
600	<	storeSp(s - 1);
601	<	return t;
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
612	<	* topmost element is the given task. Called only
613	<	* by current thread while active.
614	<	* @param t the task. Caller must ensure nonnull
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	<	int mask = q.length - 1;
621	<	int s = sp - 1;
622	<	if (casSlotNull(q, s & mask, t)) {
623	<	storeSp(s);
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.
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);
638	>	int i = locallyFifo ? base : (sp - 1);
639		return q[i & mask];
640		}
641
#	Line 534 | Line 661 | public class ForkJoinWorkerThread extend
661		ForkJoinTask<?> t = oldQ[oldIndex];
662		if (t != null && !casSlotNull(oldQ, oldIndex, t))
663		t = null;
664	<	setSlot(newQ, b & newMask, t);
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 full circular traversal, which is necessary to
687	<	* accurately set active status by caller. Also restarts if pool
688	<	* events occurred since last scan, which forces refresh of
689	<	* workers array, in case barrier was associated with resize.
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. This accounts for,
694	<	* among other things, updating random seed in place without
695	<	* storing it until exit.
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	<	ForkJoinTask<?> t = null;
701	<	int r = seed;                    // extract once to keep scan quiet
702	<	ForkJoinWorkerThread[] ws;       // refreshed on outer loop
703	<	int mask;                        // must be power 2 minus 1 and > 0
704	<	outer:do {
705	<	if ((ws = pool.workers) != null && (mask = ws.length - 1) > 0) {
706	<	int idx = r;
707	<	int probes = ~mask;      // use random index while negative
708	<	for (;;) {
709	<	r = xorShift(r);     // update random seed
710	<	ForkJoinWorkerThread v = ws[mask & idx];
711	<	if (v == null || v.sp == v.base) {
712	<	if (probes <= mask)
713	<	idx = (probes++ < 0)? r : (idx + 1);
714	<	else
715	<	break;
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	<	else if (!tryActivate() || (t = v.deqTask()) == null)
735	<	continue outer;  // restart on contention
580	<	else
581	<	break outer;
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	<	} while (pool.hasNewSyncEvent(this)); // retry on pool events
745	<	seed = r;
586	<	return t;
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	<	* gets and removes a local or stolen a task
759	<	* @return a task, if available
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 ForkJoinTask<?> pollTask() {
762	<	ForkJoinTask<?> t = locallyFifo? deqTask() : popTask();
763	<	if (t == null && (t = scan()) != null)
764	<	++stealCount;
765	<	return t;
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	<	* gets a local task
602	<	* @return a task, if available
779	>	* Sets state to TERMINATED. Called only by onTermination()
780		*/
781	<	final ForkJoinTask<?> pollLocalTask() {
782	<	return locallyFifo? deqTask() : popTask();
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	<	* Returns a pool submission, if one exists, activating first.
790	<	* @return a submission, if available
789	>	* If suspended, tries to set status to unsuspended.
790	>	* Does NOT wake up if blocked.
791	>	*
792	>	* @return true if successful
793		*/
794	<	private ForkJoinTask<?> pollSubmission() {
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	<	while (p.hasQueuedSubmissions()) {
822	<	ForkJoinTask<?> t;
823	<	if (tryActivate() && (t = p.pollSubmission()) != null)
824	<	return t;
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		}
619	–	return null;
830		}
831
832	<	// Methods accessed only by Pool
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<?> t;
849	<	while (base != sp && (t = deqTask()) != null)
850	<	t.cancelIgnoringExceptions();
848	>	ForkJoinTask<?> cj = currentJoin; // try to cancel ongoing tasks
849	>	if (cj != null) {
850	>	currentJoin = null;
851	>	cj.cancelIgnoringExceptions();
852	>	try {
853	>	this.interrupt(); // awaken wait
854	>	} catch (SecurityException ignore) {
855	>	}
856	>	}
857	>	ForkJoinTask<?> cs = currentSteal;
858	>	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
870	>	* Drains tasks to given collection c.
871	>	*
872		* @return the number of tasks drained
873		*/
874	<	final int drainTasksTo(Collection<ForkJoinTask<?>> c) {
874	>	final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) {
875		int n = 0;
876	<	ForkJoinTask<?> t;
877	<	while (base != sp && (t = deqTask()) != null) {
878	<	c.add(t);
879	<	++n;
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	<	* Get and clear steal count for accumulation by pool.  Called
890	<	* only when known to be idle (in pool.sync and termination).
889	>	* Gets and removes a local task.
890	>	*
891	>	* @return a task, if available
892		*/
893	<	final int getAndClearStealCount() {
894	<	int sc = stealCount;
895	<	stealCount = 0;
896	<	return sc;
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	<	* Returns true if at least one worker in the given array appears
907	<	* to have at least one queued task.
908	<	* @param ws array of workers
909	<	*/
910	<	static boolean hasQueuedTasks(ForkJoinWorkerThread[] ws) {
911	<	if (ws != null) {
912	<	int len = ws.length;
913	<	for (int j = 0; j < 2; ++j) { // need two passes for clean sweep
914	<	for (int i = 0; i < len; ++i) {
915	<	ForkJoinWorkerThread w = ws[i];
669	<	if (w != null && w.sp != w.base)
670	<	return true;
671	<	}
672	<	}
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 false;
917	>	return t;
918		}
919
677	–	// Support methods for ForkJoinTask
678	–
920		/**
921	<	* Returns an estimate of the number of tasks in the queue.
921	>	* Possibly runs some tasks and/or blocks, until task is done.
922	>	*
923	>	* @param joinMe the task to join
924		*/
925	<	final int getQueueSize() {
926	<	int n = sp - base;
927	<	return n < 0? 0 : n; // suppress momentarily negative values
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	<	* Returns an estimate of the number of tasks, offset by a
938	<	* function of number of idle workers.
937	>	* Run tasks in local queue until given task is done.
938	>	*
939	>	* @param joinMe the task to join
940		*/
941	<	final int getEstimatedSurplusTaskCount() {
942	<	// The halving approximates weighting idle vs non-idle workers
943	<	return (sp - base) - (pool.getIdleThreadCount() >>> 1);
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	<	* Scan, returning early if joinMe done
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 ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) {
986	<	ForkJoinTask<?> t = pollTask();
987	<	if (t != null && joinMe.status < 0 && sp == base) {
988	<	pushTask(t); // unsteal if done and this task would be stealable
989	<	t = null;
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	<	return t;
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 pool isQuiescent
1109	>	* Runs tasks until {@code pool.isQuiescent()}.
1110		*/
1111		final void helpQuiescePool() {
1112	+	ForkJoinTask<?> ps = currentSteal; // to restore below
1113		for (;;) {
1114	<	ForkJoinTask<?> t = pollTask();
1115	<	if (t != null)
1114	>	ForkJoinTask<?> t = pollLocalTask();
1115	>	if (t != null || (t = scan()) != null)
1116		t.quietlyExec();
1117	<	else if (tryInactivate() && pool.isQuiescent())
1118	<	break;
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		}
719	–	do;while (!tryActivate()); // re-activate on exit
1171		}
1172
1173	<	// Temporary Unsafe mechanics for preliminary release
1174	<	private static Unsafe getUnsafe() throws Throwable {
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 Unsafe.getUnsafe();
1182	>	return sun.misc.Unsafe.getUnsafe();
1183		} catch (SecurityException se) {
1184		try {
1185		return java.security.AccessController.doPrivileged
1186	<	(new java.security.PrivilegedExceptionAction<Unsafe>() {
1187	<	public Unsafe run() throws Exception {
1188	<	return getUnsafePrivileged();
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 e.getCause();
1195	>	throw new RuntimeException("Could not initialize intrinsics",
1196	>	e.getCause());
1197		}
1198		}
1199		}
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	–	}
1200		}

Diff Legend

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