6 |
|
|
7 |
|
package jsr166y; |
8 |
|
|
9 |
– |
import java.util.concurrent.*; |
10 |
– |
|
9 |
|
import java.util.ArrayList; |
10 |
|
import java.util.Arrays; |
11 |
|
import java.util.Collection; |
12 |
|
import java.util.Collections; |
13 |
|
import java.util.List; |
14 |
+ |
import java.util.concurrent.AbstractExecutorService; |
15 |
+ |
import java.util.concurrent.Callable; |
16 |
+ |
import java.util.concurrent.ExecutorService; |
17 |
+ |
import java.util.concurrent.Future; |
18 |
+ |
import java.util.concurrent.RejectedExecutionException; |
19 |
+ |
import java.util.concurrent.RunnableFuture; |
20 |
+ |
import java.util.concurrent.TimeUnit; |
21 |
+ |
import java.util.concurrent.TimeoutException; |
22 |
+ |
import java.util.concurrent.atomic.AtomicInteger; |
23 |
|
import java.util.concurrent.locks.LockSupport; |
24 |
|
import java.util.concurrent.locks.ReentrantLock; |
18 |
– |
import java.util.concurrent.atomic.AtomicInteger; |
19 |
– |
import java.util.concurrent.CountDownLatch; |
25 |
|
|
26 |
|
/** |
27 |
|
* An {@link ExecutorService} for running {@link ForkJoinTask}s. |
57 |
|
* convenient form for informal monitoring. |
58 |
|
* |
59 |
|
* <p> As is the case with other ExecutorServices, there are three |
60 |
< |
* main task execution methods summarized in the follwoing |
60 |
> |
* main task execution methods summarized in the following |
61 |
|
* table. These are designed to be used by clients not already engaged |
62 |
|
* in fork/join computations in the current pool. The main forms of |
63 |
|
* these methods accept instances of {@code ForkJoinTask}, but |
65 |
|
* Runnable}- or {@code Callable}- based activities as well. However, |
66 |
|
* tasks that are already executing in a pool should normally |
67 |
|
* <em>NOT</em> use these pool execution methods, but instead use the |
68 |
< |
* within-computation forms listed in the table. To avoid inadvertant |
64 |
< |
* cyclic task dependencies and to improve performance, task |
65 |
< |
* submissions to the current pool by an ongoing fork/join |
66 |
< |
* computations may be implicitly translated to the corresponding |
67 |
< |
* ForkJoinTask forms. |
68 |
> |
* within-computation forms listed in the table. |
69 |
|
* |
70 |
|
* <table BORDER CELLPADDING=3 CELLSPACING=1> |
71 |
|
* <tr> |
74 |
|
* <td ALIGN=CENTER> <b>Call from within fork/join computations</b></td> |
75 |
|
* </tr> |
76 |
|
* <tr> |
77 |
< |
* <td> <b>Arange async execution</td> |
77 |
> |
* <td> <b>Arrange async execution</td> |
78 |
|
* <td> {@link #execute(ForkJoinTask)}</td> |
79 |
|
* <td> {@link ForkJoinTask#fork}</td> |
80 |
|
* </tr> |
89 |
|
* <td> {@link ForkJoinTask#fork} (ForkJoinTasks <em>are</em> Futures)</td> |
90 |
|
* </tr> |
91 |
|
* </table> |
92 |
< |
* |
92 |
> |
* |
93 |
|
* <p><b>Sample Usage.</b> Normally a single {@code ForkJoinPool} is |
94 |
|
* used for all parallel task execution in a program or subsystem. |
95 |
|
* Otherwise, use would not usually outweigh the construction and |
114 |
|
* {@code IllegalArgumentException}. |
115 |
|
* |
116 |
|
* <p>This implementation rejects submitted tasks (that is, by throwing |
117 |
< |
* {@link RejectedExecutionException}) only when the pool is shut down. |
117 |
> |
* {@link RejectedExecutionException}) only when the pool is shut down |
118 |
> |
* or internal resources have been exhausted. |
119 |
|
* |
120 |
|
* @since 1.7 |
121 |
|
* @author Doug Lea |
142 |
|
* of tasks profit from cache affinities, but others are harmed by |
143 |
|
* cache pollution effects.) |
144 |
|
* |
145 |
+ |
* Beyond work-stealing support and essential bookkeeping, the |
146 |
+ |
* main responsibility of this framework is to take actions when |
147 |
+ |
* one worker is waiting to join a task stolen (or always held by) |
148 |
+ |
* another. Because we are multiplexing many tasks on to a pool |
149 |
+ |
* of workers, we can't just let them block (as in Thread.join). |
150 |
+ |
* We also cannot just reassign the joiner's run-time stack with |
151 |
+ |
* another and replace it later, which would be a form of |
152 |
+ |
* "continuation", that even if possible is not necessarily a good |
153 |
+ |
* idea. Given that the creation costs of most threads on most |
154 |
+ |
* systems mainly surrounds setting up runtime stacks, thread |
155 |
+ |
* creation and switching is usually not much more expensive than |
156 |
+ |
* stack creation and switching, and is more flexible). Instead we |
157 |
+ |
* combine two tactics: |
158 |
+ |
* |
159 |
+ |
* Helping: Arranging for the joiner to execute some task that it |
160 |
+ |
* would be running if the steal had not occurred. Method |
161 |
+ |
* ForkJoinWorkerThread.helpJoinTask tracks joining->stealing |
162 |
+ |
* links to try to find such a task. |
163 |
+ |
* |
164 |
+ |
* Compensating: Unless there are already enough live threads, |
165 |
+ |
* method helpMaintainParallelism() may create or |
166 |
+ |
* re-activate a spare thread to compensate for blocked |
167 |
+ |
* joiners until they unblock. |
168 |
+ |
* |
169 |
+ |
* It is impossible to keep exactly the target (parallelism) |
170 |
+ |
* number of threads running at any given time. Determining |
171 |
+ |
* existence of conservatively safe helping targets, the |
172 |
+ |
* availability of already-created spares, and the apparent need |
173 |
+ |
* to create new spares are all racy and require heuristic |
174 |
+ |
* guidance, so we rely on multiple retries of each. Compensation |
175 |
+ |
* occurs in slow-motion. It is triggered only upon timeouts of |
176 |
+ |
* Object.wait used for joins. This reduces poor decisions that |
177 |
+ |
* would otherwise be made when threads are waiting for others |
178 |
+ |
* that are stalled because of unrelated activities such as |
179 |
+ |
* garbage collection. |
180 |
+ |
* |
181 |
+ |
* The ManagedBlocker extension API can't use helping so relies |
182 |
+ |
* only on compensation in method awaitBlocker. |
183 |
+ |
* |
184 |
|
* The main throughput advantages of work-stealing stem from |
185 |
|
* decentralized control -- workers mostly steal tasks from each |
186 |
|
* other. We do not want to negate this by creating bottlenecks |
187 |
< |
* implementing the management responsibilities of this class. So |
188 |
< |
* we use a collection of techniques that avoid, reduce, or cope |
189 |
< |
* well with contention. These entail several instances of |
190 |
< |
* bit-packing into CASable fields to maintain only the minimally |
191 |
< |
* required atomicity. To enable such packing, we restrict maximum |
192 |
< |
* parallelism to (1<<15)-1 (enabling twice this to fit into a 16 |
193 |
< |
* bit field), which is far in excess of normal operating range. |
194 |
< |
* Even though updates to some of these bookkeeping fields do |
195 |
< |
* sometimes contend with each other, they don't normally |
196 |
< |
* cache-contend with updates to others enough to warrant memory |
197 |
< |
* padding or isolation. So they are all held as fields of |
198 |
< |
* ForkJoinPool objects. The main capabilities are as follows: |
187 |
> |
* implementing other management responsibilities. So we use a |
188 |
> |
* collection of techniques that avoid, reduce, or cope well with |
189 |
> |
* contention. These entail several instances of bit-packing into |
190 |
> |
* CASable fields to maintain only the minimally required |
191 |
> |
* atomicity. To enable such packing, we restrict maximum |
192 |
> |
* parallelism to (1<<15)-1 (enabling twice this (to accommodate |
193 |
> |
* unbalanced increments and decrements) to fit into a 16 bit |
194 |
> |
* field, which is far in excess of normal operating range. Even |
195 |
> |
* though updates to some of these bookkeeping fields do sometimes |
196 |
> |
* contend with each other, they don't normally cache-contend with |
197 |
> |
* updates to others enough to warrant memory padding or |
198 |
> |
* isolation. So they are all held as fields of ForkJoinPool |
199 |
> |
* objects. The main capabilities are as follows: |
200 |
|
* |
201 |
|
* 1. Creating and removing workers. Workers are recorded in the |
202 |
|
* "workers" array. This is an array as opposed to some other data |
212 |
|
* blocked workers. However, all other support code is set up to |
213 |
|
* work with other policies. |
214 |
|
* |
215 |
+ |
* To ensure that we do not hold on to worker references that |
216 |
+ |
* would prevent GC, ALL accesses to workers are via indices into |
217 |
+ |
* the workers array (which is one source of some of the unusual |
218 |
+ |
* code constructions here). In essence, the workers array serves |
219 |
+ |
* as a WeakReference mechanism. Thus for example the event queue |
220 |
+ |
* stores worker indices, not worker references. Access to the |
221 |
+ |
* workers in associated methods (for example releaseEventWaiters) |
222 |
+ |
* must both index-check and null-check the IDs. All such accesses |
223 |
+ |
* ignore bad IDs by returning out early from what they are doing, |
224 |
+ |
* since this can only be associated with shutdown, in which case |
225 |
+ |
* it is OK to give up. On termination, we just clobber these |
226 |
+ |
* data structures without trying to use them. |
227 |
+ |
* |
228 |
|
* 2. Bookkeeping for dynamically adding and removing workers. We |
229 |
|
* aim to approximately maintain the given level of parallelism. |
230 |
|
* When some workers are known to be blocked (on joins or via |
231 |
|
* ManagedBlocker), we may create or resume others to take their |
232 |
|
* place until they unblock (see below). Implementing this |
233 |
|
* requires counts of the number of "running" threads (i.e., those |
234 |
< |
* that are neither blocked nor artifically suspended) as well as |
234 |
> |
* that are neither blocked nor artificially suspended) as well as |
235 |
|
* the total number. These two values are packed into one field, |
236 |
|
* "workerCounts" because we need accurate snapshots when deciding |
237 |
< |
* to create, resume or suspend. To support these decisions, |
238 |
< |
* updates to spare counts must be prospective (not |
239 |
< |
* retrospective). For example, the running count is decremented |
240 |
< |
* before blocking by a thread about to block as a spare, but |
186 |
< |
* incremented by the thread about to unblock it. Updates upon |
187 |
< |
* resumption ofr threads blocking in awaitJoin or awaitBlocker |
188 |
< |
* cannot usually be prospective, so the running count is in |
189 |
< |
* general an upper bound of the number of productively running |
190 |
< |
* threads Updates to the workerCounts field sometimes transiently |
191 |
< |
* encounter a fair amount of contention when join dependencies |
192 |
< |
* are such that many threads block or unblock at about the same |
193 |
< |
* time. We alleviate this by sometimes performing an alternative |
194 |
< |
* action on contention like releasing waiters or locating spares. |
237 |
> |
* to create, resume or suspend. Note however that the |
238 |
> |
* correspondence of these counts to reality is not guaranteed. In |
239 |
> |
* particular updates for unblocked threads may lag until they |
240 |
> |
* actually wake up. |
241 |
|
* |
242 |
|
* 3. Maintaining global run state. The run state of the pool |
243 |
|
* consists of a runLevel (SHUTDOWN, TERMINATING, etc) similar to |
266 |
|
* workers that previously could not find a task to now find one: |
267 |
|
* Submission of a new task to the pool, or another worker pushing |
268 |
|
* a task onto a previously empty queue. (We also use this |
269 |
< |
* mechanism for termination and reconfiguration actions that |
269 |
> |
* mechanism for configuration and termination actions that |
270 |
|
* require wakeups of idle workers). Each worker maintains its |
271 |
|
* last known event count, and blocks when a scan for work did not |
272 |
|
* find a task AND its lastEventCount matches the current |
277 |
|
* a record (field nextEventWaiter) for the next waiting worker. |
278 |
|
* In addition to allowing simpler decisions about need for |
279 |
|
* wakeup, the event count bits in eventWaiters serve the role of |
280 |
< |
* tags to avoid ABA errors in Treiber stacks. To reduce delays |
281 |
< |
* in task diffusion, workers not otherwise occupied may invoke |
282 |
< |
* method releaseWaiters, that removes and signals (unparks) |
283 |
< |
* workers not waiting on current count. To minimize task |
284 |
< |
* production stalls associate with signalling, any worker pushing |
285 |
< |
* a task on an empty queue invokes the weaker method signalWork, |
240 |
< |
* that only releases idle workers until it detects interference |
241 |
< |
* by other threads trying to release, and lets them take |
242 |
< |
* over. The net effect is a tree-like diffusion of signals, where |
243 |
< |
* released threads (and possibly others) help with unparks. To |
244 |
< |
* further reduce contention effects a bit, failed CASes to |
245 |
< |
* increment field eventCount are tolerated without retries. |
280 |
> |
* tags to avoid ABA errors in Treiber stacks. Upon any wakeup, |
281 |
> |
* released threads also try to release at most two others. The |
282 |
> |
* net effect is a tree-like diffusion of signals, where released |
283 |
> |
* threads (and possibly others) help with unparks. To further |
284 |
> |
* reduce contention effects a bit, failed CASes to increment |
285 |
> |
* field eventCount are tolerated without retries in signalWork. |
286 |
|
* Conceptually they are merged into the same event, which is OK |
287 |
|
* when their only purpose is to enable workers to scan for work. |
288 |
|
* |
289 |
< |
* 5. Managing suspension of extra workers. When a worker is about |
290 |
< |
* to block waiting for a join (or via ManagedBlockers), we may |
291 |
< |
* create a new thread to maintain parallelism level, or at least |
292 |
< |
* avoid starvation (see below). Usually, extra threads are needed |
293 |
< |
* for only very short periods, yet join dependencies are such |
294 |
< |
* that we sometimes need them in bursts. Rather than create new |
295 |
< |
* threads each time this happens, we suspend no-longer-needed |
296 |
< |
* extra ones as "spares". For most purposes, we don't distinguish |
297 |
< |
* "extra" spare threads from normal "core" threads: On each call |
298 |
< |
* to preStep (the only point at which we can do this) a worker |
299 |
< |
* checks to see if there are now too many running workers, and if |
300 |
< |
* so, suspends itself. Methods awaitJoin and awaitBlocker look |
301 |
< |
* for suspended threads to resume before considering creating a |
302 |
< |
* new replacement. We don't need a special data structure to |
303 |
< |
* maintain spares; simply scanning the workers array looking for |
304 |
< |
* worker.isSuspended() is fine because the calling thread is |
305 |
< |
* otherwise not doing anything useful anyway; we are at least as |
306 |
< |
* happy if after locating a spare, the caller doesn't actually |
307 |
< |
* block because the join is ready before we try to adjust and |
308 |
< |
* compensate. Note that this is intrinsically racy. One thread |
309 |
< |
* may become a spare at about the same time as another is |
310 |
< |
* needlessly being created. We counteract this and related slop |
311 |
< |
* in part by requiring resumed spares to immediately recheck (in |
312 |
< |
* preStep) to see whether they they should re-suspend. The only |
313 |
< |
* effective difference between "extra" and "core" threads is that |
314 |
< |
* we allow the "extra" ones to time out and die if they are not |
315 |
< |
* resumed within a keep-alive interval of a few seconds. This is |
316 |
< |
* implemented mainly within ForkJoinWorkerThread, but requires |
317 |
< |
* some coordination (isTrimmed() -- meaning killed while |
318 |
< |
* suspended) to correctly maintain pool counts. |
319 |
< |
* |
320 |
< |
* 6. Deciding when to create new workers. The main dynamic |
321 |
< |
* control in this class is deciding when to create extra threads, |
322 |
< |
* in methods awaitJoin and awaitBlocker. We always need to create |
323 |
< |
* one when the number of running threads becomes zero. But |
324 |
< |
* because blocked joins are typically dependent, we don't |
325 |
< |
* necessarily need or want one-to-one replacement. Instead, we |
326 |
< |
* use a combination of heuristics that adds threads only when the |
327 |
< |
* pool appears to be approaching starvation. These effectively |
328 |
< |
* reduce churn at the price of systematically undershooting |
329 |
< |
* target parallelism when many threads are blocked. However, |
330 |
< |
* biasing toward undeshooting partially compensates for the above |
331 |
< |
* mechanics to suspend extra threads, that normally lead to |
332 |
< |
* overshoot because we can only suspend workers in-between |
333 |
< |
* top-level actions. It also better copes with the fact that some |
334 |
< |
* of the methods in this class tend to never become compiled (but |
335 |
< |
* are interpreted), so some components of the entire set of |
336 |
< |
* controls might execute many times faster than others. And |
289 |
> |
* 5. Managing suspension of extra workers. When a worker notices |
290 |
> |
* (usually upon timeout of a wait()) that there are too few |
291 |
> |
* running threads, we may create a new thread to maintain |
292 |
> |
* parallelism level, or at least avoid starvation. Usually, extra |
293 |
> |
* threads are needed for only very short periods, yet join |
294 |
> |
* dependencies are such that we sometimes need them in |
295 |
> |
* bursts. Rather than create new threads each time this happens, |
296 |
> |
* we suspend no-longer-needed extra ones as "spares". For most |
297 |
> |
* purposes, we don't distinguish "extra" spare threads from |
298 |
> |
* normal "core" threads: On each call to preStep (the only point |
299 |
> |
* at which we can do this) a worker checks to see if there are |
300 |
> |
* now too many running workers, and if so, suspends itself. |
301 |
> |
* Method helpMaintainParallelism looks for suspended threads to |
302 |
> |
* resume before considering creating a new replacement. The |
303 |
> |
* spares themselves are encoded on another variant of a Treiber |
304 |
> |
* Stack, headed at field "spareWaiters". Note that the use of |
305 |
> |
* spares is intrinsically racy. One thread may become a spare at |
306 |
> |
* about the same time as another is needlessly being created. We |
307 |
> |
* counteract this and related slop in part by requiring resumed |
308 |
> |
* spares to immediately recheck (in preStep) to see whether they |
309 |
> |
* should re-suspend. |
310 |
> |
* |
311 |
> |
* 6. Killing off unneeded workers. A timeout mechanism is used to |
312 |
> |
* shed unused workers: The oldest (first) event queue waiter uses |
313 |
> |
* a timed rather than hard wait. When this wait times out without |
314 |
> |
* a normal wakeup, it tries to shutdown any one (for convenience |
315 |
> |
* the newest) other spare or event waiter via |
316 |
> |
* tryShutdownUnusedWorker. This eventually reduces the number of |
317 |
> |
* worker threads to a minimum of one after a long enough period |
318 |
> |
* without use. |
319 |
> |
* |
320 |
> |
* 7. Deciding when to create new workers. The main dynamic |
321 |
> |
* control in this class is deciding when to create extra threads |
322 |
> |
* in method helpMaintainParallelism. We would like to keep |
323 |
> |
* exactly #parallelism threads running, which is an impossible |
324 |
> |
* task. We always need to create one when the number of running |
325 |
> |
* threads would become zero and all workers are busy. Beyond |
326 |
> |
* this, we must rely on heuristics that work well in the |
327 |
> |
* presence of transient phenomena such as GC stalls, dynamic |
328 |
> |
* compilation, and wake-up lags. These transients are extremely |
329 |
> |
* common -- we are normally trying to fully saturate the CPUs on |
330 |
> |
* a machine, so almost any activity other than running tasks |
331 |
> |
* impedes accuracy. Our main defense is to allow parallelism to |
332 |
> |
* lapse for a while during joins, and use a timeout to see if, |
333 |
> |
* after the resulting settling, there is still a need for |
334 |
> |
* additional workers. This also better copes with the fact that |
335 |
> |
* some of the methods in this class tend to never become compiled |
336 |
> |
* (but are interpreted), so some components of the entire set of |
337 |
> |
* controls might execute 100 times faster than others. And |
338 |
|
* similarly for cases where the apparent lack of work is just due |
339 |
|
* to GC stalls and other transient system activity. |
340 |
|
* |
349 |
|
* |
350 |
|
* Style notes: There are lots of inline assignments (of form |
351 |
|
* "while ((local = field) != 0)") which are usually the simplest |
352 |
< |
* way to ensure read orderings. Also several occurrences of the |
353 |
< |
* unusual "do {} while(!cas...)" which is the simplest way to |
354 |
< |
* force an update of a CAS'ed variable. There are also a few |
355 |
< |
* other coding oddities that help some methods perform reasonably |
356 |
< |
* even when interpreted (not compiled). |
352 |
> |
* way to ensure the required read orderings (which are sometimes |
353 |
> |
* critical). Also several occurrences of the unusual "do {} |
354 |
> |
* while (!cas...)" which is the simplest way to force an update of |
355 |
> |
* a CAS'ed variable. There are also other coding oddities that |
356 |
> |
* help some methods perform reasonably even when interpreted (not |
357 |
> |
* compiled), at the expense of some messy constructions that |
358 |
> |
* reduce byte code counts. |
359 |
|
* |
360 |
|
* The order of declarations in this file is: (1) statics (2) |
361 |
|
* fields (along with constants used when unpacking some of them) |
423 |
|
new AtomicInteger(); |
424 |
|
|
425 |
|
/** |
426 |
< |
* Absolute bound for parallelism level. Twice this number must |
427 |
< |
* fit into a 16bit field to enable word-packing for some counts. |
426 |
> |
* The time to block in a join (see awaitJoin) before checking if |
427 |
> |
* a new worker should be (re)started to maintain parallelism |
428 |
> |
* level. The value should be short enough to maintain global |
429 |
> |
* responsiveness and progress but long enough to avoid |
430 |
> |
* counterproductive firings during GC stalls or unrelated system |
431 |
> |
* activity, and to not bog down systems with continual re-firings |
432 |
> |
* on GCs or legitimately long waits. |
433 |
> |
*/ |
434 |
> |
private static final long JOIN_TIMEOUT_MILLIS = 250L; // 4 per second |
435 |
> |
|
436 |
> |
/** |
437 |
> |
* The wakeup interval (in nanoseconds) for the oldest worker |
438 |
> |
* waiting for an event to invoke tryShutdownUnusedWorker to |
439 |
> |
* shrink the number of workers. The exact value does not matter |
440 |
> |
* too much. It must be short enough to release resources during |
441 |
> |
* sustained periods of idleness, but not so short that threads |
442 |
> |
* are continually re-created. |
443 |
> |
*/ |
444 |
> |
private static final long SHRINK_RATE_NANOS = |
445 |
> |
30L * 1000L * 1000L * 1000L; // 2 per minute |
446 |
> |
|
447 |
> |
/** |
448 |
> |
* Absolute bound for parallelism level. Twice this number plus |
449 |
> |
* one (i.e., 0xfff) must fit into a 16bit field to enable |
450 |
> |
* word-packing for some counts and indices. |
451 |
|
*/ |
452 |
< |
private static final int MAX_THREADS = 0x7fff; |
452 |
> |
private static final int MAX_WORKERS = 0x7fff; |
453 |
|
|
454 |
|
/** |
455 |
|
* Array holding all worker threads in the pool. Array size must |
489 |
|
private volatile long stealCount; |
490 |
|
|
491 |
|
/** |
492 |
< |
* Encoded record of top of treiber stack of threads waiting for |
492 |
> |
* Encoded record of top of Treiber stack of threads waiting for |
493 |
|
* events. The top 32 bits contain the count being waited for. The |
494 |
< |
* bottom word contains one plus the pool index of waiting worker |
495 |
< |
* thread. |
494 |
> |
* bottom 16 bits contains one plus the pool index of waiting |
495 |
> |
* worker thread. (Bits 16-31 are unused.) |
496 |
|
*/ |
497 |
|
private volatile long eventWaiters; |
498 |
|
|
499 |
< |
private static final int EVENT_COUNT_SHIFT = 32; |
500 |
< |
private static final long WAITER_INDEX_MASK = (1L << EVENT_COUNT_SHIFT)-1L; |
499 |
> |
private static final int EVENT_COUNT_SHIFT = 32; |
500 |
> |
private static final int WAITER_ID_MASK = (1 << 16) - 1; |
501 |
|
|
502 |
|
/** |
503 |
|
* A counter for events that may wake up worker threads: |
504 |
|
* - Submission of a new task to the pool |
505 |
|
* - A worker pushing a task on an empty queue |
506 |
< |
* - termination and reconfiguration |
506 |
> |
* - termination |
507 |
|
*/ |
508 |
|
private volatile int eventCount; |
509 |
|
|
510 |
|
/** |
511 |
+ |
* Encoded record of top of Treiber stack of spare threads waiting |
512 |
+ |
* for resumption. The top 16 bits contain an arbitrary count to |
513 |
+ |
* avoid ABA effects. The bottom 16bits contains one plus the pool |
514 |
+ |
* index of waiting worker thread. |
515 |
+ |
*/ |
516 |
+ |
private volatile int spareWaiters; |
517 |
+ |
|
518 |
+ |
private static final int SPARE_COUNT_SHIFT = 16; |
519 |
+ |
private static final int SPARE_ID_MASK = (1 << 16) - 1; |
520 |
+ |
|
521 |
+ |
/** |
522 |
|
* Lifecycle control. The low word contains the number of workers |
523 |
|
* that are (probably) executing tasks. This value is atomically |
524 |
|
* incremented before a worker gets a task to run, and decremented |
525 |
< |
* when worker has no tasks and cannot find any. Bits 16-18 |
525 |
> |
* when a worker has no tasks and cannot find any. Bits 16-18 |
526 |
|
* contain runLevel value. When all are zero, the pool is |
527 |
|
* running. Level transitions are monotonic (running -> shutdown |
528 |
|
* -> terminating -> terminated) so each transition adds a bit. |
529 |
|
* These are bundled together to ensure consistent read for |
530 |
|
* termination checks (i.e., that runLevel is at least SHUTDOWN |
531 |
|
* and active threads is zero). |
532 |
+ |
* |
533 |
+ |
* Notes: Most direct CASes are dependent on these bitfield |
534 |
+ |
* positions. Also, this field is non-private to enable direct |
535 |
+ |
* performance-sensitive CASes in ForkJoinWorkerThread. |
536 |
|
*/ |
537 |
< |
private volatile int runState; |
537 |
> |
volatile int runState; |
538 |
|
|
539 |
|
// Note: The order among run level values matters. |
540 |
|
private static final int RUNLEVEL_SHIFT = 16; |
542 |
|
private static final int TERMINATING = 1 << (RUNLEVEL_SHIFT + 1); |
543 |
|
private static final int TERMINATED = 1 << (RUNLEVEL_SHIFT + 2); |
544 |
|
private static final int ACTIVE_COUNT_MASK = (1 << RUNLEVEL_SHIFT) - 1; |
464 |
– |
private static final int ONE_ACTIVE = 1; // active update delta |
545 |
|
|
546 |
|
/** |
547 |
|
* Holds number of total (i.e., created and not yet terminated) |
550 |
|
* making decisions about creating and suspending spare |
551 |
|
* threads. Updated only by CAS. Note that adding a new worker |
552 |
|
* requires incrementing both counts, since workers start off in |
553 |
< |
* running state. This field is also used for memory-fencing |
474 |
< |
* configuration parameters. |
553 |
> |
* running state. |
554 |
|
*/ |
555 |
|
private volatile int workerCounts; |
556 |
|
|
582 |
|
*/ |
583 |
|
private final int poolNumber; |
584 |
|
|
585 |
< |
// utilities for updating fields |
585 |
> |
// Utilities for CASing fields. Note that most of these |
586 |
> |
// are usually manually inlined by callers |
587 |
|
|
588 |
|
/** |
589 |
< |
* Increments running count. Also used by ForkJoinTask. |
589 |
> |
* Increments running count part of workerCounts |
590 |
|
*/ |
591 |
|
final void incrementRunningCount() { |
592 |
|
int c; |
593 |
|
do {} while (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
594 |
< |
c = workerCounts, |
594 |
> |
c = workerCounts, |
595 |
|
c + ONE_RUNNING)); |
596 |
|
} |
597 |
< |
|
597 |
> |
|
598 |
> |
/** |
599 |
> |
* Tries to increment running count part of workerCounts |
600 |
> |
*/ |
601 |
> |
final boolean tryIncrementRunningCount() { |
602 |
> |
int c; |
603 |
> |
return UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
604 |
> |
c = workerCounts, |
605 |
> |
c + ONE_RUNNING); |
606 |
> |
} |
607 |
> |
|
608 |
|
/** |
609 |
|
* Tries to decrement running count unless already zero |
610 |
|
*/ |
617 |
|
} |
618 |
|
|
619 |
|
/** |
620 |
< |
* Tries incrementing active count; fails on contention. |
621 |
< |
* Called by workers before executing tasks. |
620 |
> |
* Forces decrement of encoded workerCounts, awaiting nonzero if |
621 |
> |
* (rarely) necessary when other count updates lag. |
622 |
|
* |
623 |
< |
* @return true on success |
623 |
> |
* @param dr -- either zero or ONE_RUNNING |
624 |
> |
* @param dt -- either zero or ONE_TOTAL |
625 |
|
*/ |
626 |
< |
final boolean tryIncrementActiveCount() { |
627 |
< |
int c; |
628 |
< |
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
629 |
< |
c = runState, c + ONE_ACTIVE); |
626 |
> |
private void decrementWorkerCounts(int dr, int dt) { |
627 |
> |
for (;;) { |
628 |
> |
int wc = workerCounts; |
629 |
> |
if ((wc & RUNNING_COUNT_MASK) - dr < 0 || |
630 |
> |
(wc >>> TOTAL_COUNT_SHIFT) - dt < 0) { |
631 |
> |
if ((runState & TERMINATED) != 0) |
632 |
> |
return; // lagging termination on a backout |
633 |
> |
Thread.yield(); |
634 |
> |
} |
635 |
> |
if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
636 |
> |
wc, wc - (dr + dt))) |
637 |
> |
return; |
638 |
> |
} |
639 |
|
} |
640 |
|
|
641 |
|
/** |
645 |
|
final boolean tryDecrementActiveCount() { |
646 |
|
int c; |
647 |
|
return UNSAFE.compareAndSwapInt(this, runStateOffset, |
648 |
< |
c = runState, c - ONE_ACTIVE); |
648 |
> |
c = runState, c - 1); |
649 |
|
} |
650 |
|
|
651 |
|
/** |
674 |
|
lock.lock(); |
675 |
|
try { |
676 |
|
ForkJoinWorkerThread[] ws = workers; |
677 |
< |
int nws = ws.length; |
678 |
< |
if (k < 0 || k >= nws || ws[k] != null) { |
679 |
< |
for (k = 0; k < nws && ws[k] != null; ++k) |
677 |
> |
int n = ws.length; |
678 |
> |
if (k < 0 || k >= n || ws[k] != null) { |
679 |
> |
for (k = 0; k < n && ws[k] != null; ++k) |
680 |
|
; |
681 |
< |
if (k == nws) |
682 |
< |
ws = Arrays.copyOf(ws, nws << 1); |
681 |
> |
if (k == n) |
682 |
> |
ws = workers = Arrays.copyOf(ws, n << 1); |
683 |
|
} |
684 |
|
ws[k] = w; |
685 |
< |
workers = ws; // volatile array write ensures slot visibility |
685 |
> |
int c = eventCount; // advance event count to ensure visibility |
686 |
> |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c, c+1); |
687 |
|
} finally { |
688 |
|
lock.unlock(); |
689 |
|
} |
691 |
|
} |
692 |
|
|
693 |
|
/** |
694 |
< |
* Nulls out record of worker in workers array |
694 |
> |
* Nulls out record of worker in workers array. |
695 |
|
*/ |
696 |
|
private void forgetWorker(ForkJoinWorkerThread w) { |
697 |
|
int idx = w.poolIndex; |
698 |
< |
// Locking helps method recordWorker avoid unecessary expansion |
698 |
> |
// Locking helps method recordWorker avoid unnecessary expansion |
699 |
|
final ReentrantLock lock = this.workerLock; |
700 |
|
lock.lock(); |
701 |
|
try { |
707 |
|
} |
708 |
|
} |
709 |
|
|
609 |
– |
// adding and removing workers |
610 |
– |
|
710 |
|
/** |
711 |
< |
* Tries to create and add new worker. Assumes that worker counts |
712 |
< |
* are already updated to accommodate the worker, so adjusts on |
713 |
< |
* failure. |
711 |
> |
* Final callback from terminating worker. Removes record of |
712 |
> |
* worker from array, and adjusts counts. If pool is shutting |
713 |
> |
* down, tries to complete termination. |
714 |
|
* |
715 |
< |
* @return new worker or null if creation failed |
715 |
> |
* @param w the worker |
716 |
|
*/ |
717 |
< |
private ForkJoinWorkerThread addWorker() { |
718 |
< |
ForkJoinWorkerThread w = null; |
719 |
< |
try { |
720 |
< |
w = factory.newThread(this); |
721 |
< |
} finally { // Adjust on either null or exceptional factory return |
722 |
< |
if (w == null) { |
624 |
< |
onWorkerCreationFailure(); |
625 |
< |
return null; |
626 |
< |
} |
627 |
< |
} |
628 |
< |
w.start(recordWorker(w), ueh); |
629 |
< |
return w; |
717 |
> |
final void workerTerminated(ForkJoinWorkerThread w) { |
718 |
> |
forgetWorker(w); |
719 |
> |
decrementWorkerCounts(w.isTrimmed() ? 0 : ONE_RUNNING, ONE_TOTAL); |
720 |
> |
while (w.stealCount != 0) // collect final count |
721 |
> |
tryAccumulateStealCount(w); |
722 |
> |
tryTerminate(false); |
723 |
|
} |
724 |
|
|
725 |
+ |
// Waiting for and signalling events |
726 |
+ |
|
727 |
|
/** |
728 |
< |
* Adjusts counts upon failure to create worker |
728 |
> |
* Releases workers blocked on a count not equal to current count. |
729 |
> |
* Normally called after precheck that eventWaiters isn't zero to |
730 |
> |
* avoid wasted array checks. Gives up upon a change in count or |
731 |
> |
* upon releasing two workers, letting others take over. |
732 |
|
*/ |
733 |
< |
private void onWorkerCreationFailure() { |
734 |
< |
for (;;) { |
735 |
< |
int wc = workerCounts; |
736 |
< |
if ((wc >>> TOTAL_COUNT_SHIFT) > 0 && |
737 |
< |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
738 |
< |
wc, wc - (ONE_RUNNING|ONE_TOTAL))) |
733 |
> |
private void releaseEventWaiters() { |
734 |
> |
ForkJoinWorkerThread[] ws = workers; |
735 |
> |
int n = ws.length; |
736 |
> |
long h = eventWaiters; |
737 |
> |
int ec = eventCount; |
738 |
> |
boolean releasedOne = false; |
739 |
> |
ForkJoinWorkerThread w; int id; |
740 |
> |
while ((id = (((int)h) & WAITER_ID_MASK) - 1) >= 0 && |
741 |
> |
(int)(h >>> EVENT_COUNT_SHIFT) != ec && |
742 |
> |
id < n && (w = ws[id]) != null) { |
743 |
> |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
744 |
> |
h, w.nextWaiter)) { |
745 |
> |
LockSupport.unpark(w); |
746 |
> |
if (releasedOne) // exit on second release |
747 |
> |
break; |
748 |
> |
releasedOne = true; |
749 |
> |
} |
750 |
> |
if (eventCount != ec) |
751 |
|
break; |
752 |
+ |
h = eventWaiters; |
753 |
|
} |
643 |
– |
tryTerminate(false); // in case of failure during shutdown |
754 |
|
} |
755 |
|
|
756 |
|
/** |
757 |
< |
* Create enough total workers to establish target parallelism, |
758 |
< |
* giving up if terminating or addWorker fails |
757 |
> |
* Tries to advance eventCount and releases waiters. Called only |
758 |
> |
* from workers. |
759 |
|
*/ |
760 |
< |
private void ensureEnoughTotalWorkers() { |
761 |
< |
int wc; |
762 |
< |
while (((wc = workerCounts) >>> TOTAL_COUNT_SHIFT) < parallelism && |
763 |
< |
runState < TERMINATING) { |
764 |
< |
if ((UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
655 |
< |
wc, wc + (ONE_RUNNING|ONE_TOTAL)) && |
656 |
< |
addWorker() == null)) |
657 |
< |
break; |
658 |
< |
} |
760 |
> |
final void signalWork() { |
761 |
> |
int c; // try to increment event count -- CAS failure OK |
762 |
> |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); |
763 |
> |
if (eventWaiters != 0L) |
764 |
> |
releaseEventWaiters(); |
765 |
|
} |
766 |
|
|
767 |
|
/** |
768 |
< |
* Final callback from terminating worker. Removes record of |
769 |
< |
* worker from array, and adjusts counts. If pool is shutting |
664 |
< |
* down, tries to complete terminatation, else possibly replaces |
665 |
< |
* the worker. |
768 |
> |
* Adds the given worker to event queue and blocks until |
769 |
> |
* terminating or event count advances from the given value |
770 |
|
* |
771 |
< |
* @param w the worker |
771 |
> |
* @param w the calling worker thread |
772 |
> |
* @param ec the count |
773 |
|
*/ |
774 |
< |
final void workerTerminated(ForkJoinWorkerThread w) { |
775 |
< |
if (w.active) { // force inactive |
776 |
< |
w.active = false; |
777 |
< |
do {} while (!tryDecrementActiveCount()); |
778 |
< |
} |
779 |
< |
forgetWorker(w); |
780 |
< |
|
781 |
< |
// Decrement total count, and if was running, running count |
782 |
< |
// Spin (waiting for other updates) if either would be negative |
783 |
< |
int nr = w.isTrimmed() ? 0 : ONE_RUNNING; |
679 |
< |
int unit = ONE_TOTAL + nr; |
680 |
< |
for (;;) { |
681 |
< |
int wc = workerCounts; |
682 |
< |
int rc = wc & RUNNING_COUNT_MASK; |
683 |
< |
if (rc - nr < 0 || (wc >>> TOTAL_COUNT_SHIFT) == 0) |
684 |
< |
Thread.yield(); // back off if waiting for other updates |
685 |
< |
else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
686 |
< |
wc, wc - unit)) |
774 |
> |
private void eventSync(ForkJoinWorkerThread w, int ec) { |
775 |
> |
long nh = (((long)ec) << EVENT_COUNT_SHIFT) | ((long)(w.poolIndex+1)); |
776 |
> |
long h; |
777 |
> |
while ((runState < SHUTDOWN || !tryTerminate(false)) && |
778 |
> |
(((int)(h = eventWaiters) & WAITER_ID_MASK) == 0 || |
779 |
> |
(int)(h >>> EVENT_COUNT_SHIFT) == ec) && |
780 |
> |
eventCount == ec) { |
781 |
> |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
782 |
> |
w.nextWaiter = h, nh)) { |
783 |
> |
awaitEvent(w, ec); |
784 |
|
break; |
785 |
+ |
} |
786 |
|
} |
689 |
– |
|
690 |
– |
accumulateStealCount(w); // collect final count |
691 |
– |
if (!tryTerminate(false)) |
692 |
– |
ensureEnoughTotalWorkers(); |
787 |
|
} |
788 |
|
|
695 |
– |
// Waiting for and signalling events |
696 |
– |
|
789 |
|
/** |
790 |
< |
* Releases workers blocked on a count not equal to current count. |
790 |
> |
* Blocks the given worker (that has already been entered as an |
791 |
> |
* event waiter) until terminating or event count advances from |
792 |
> |
* the given value. The oldest (first) waiter uses a timed wait to |
793 |
> |
* occasionally one-by-one shrink the number of workers (to a |
794 |
> |
* minimum of one) if the pool has not been used for extended |
795 |
> |
* periods. |
796 |
> |
* |
797 |
> |
* @param w the calling worker thread |
798 |
> |
* @param ec the count |
799 |
|
*/ |
800 |
< |
private void releaseWaiters() { |
801 |
< |
long top; |
802 |
< |
int id; |
803 |
< |
while ((id = (int)((top = eventWaiters) & WAITER_INDEX_MASK)) > 0 && |
804 |
< |
(int)(top >>> EVENT_COUNT_SHIFT) != eventCount) { |
805 |
< |
ForkJoinWorkerThread[] ws = workers; |
806 |
< |
ForkJoinWorkerThread w; |
807 |
< |
if (ws.length >= id && (w = ws[id - 1]) != null && |
808 |
< |
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
809 |
< |
top, w.nextWaiter)) |
810 |
< |
LockSupport.unpark(w); |
800 |
> |
private void awaitEvent(ForkJoinWorkerThread w, int ec) { |
801 |
> |
while (eventCount == ec) { |
802 |
> |
if (tryAccumulateStealCount(w)) { // transfer while idle |
803 |
> |
boolean untimed = (w.nextWaiter != 0L || |
804 |
> |
(workerCounts & RUNNING_COUNT_MASK) <= 1); |
805 |
> |
long startTime = untimed ? 0 : System.nanoTime(); |
806 |
> |
Thread.interrupted(); // clear/ignore interrupt |
807 |
> |
if (w.isTerminating() || eventCount != ec) |
808 |
> |
break; // recheck after clear |
809 |
> |
if (untimed) |
810 |
> |
LockSupport.park(w); |
811 |
> |
else { |
812 |
> |
LockSupport.parkNanos(w, SHRINK_RATE_NANOS); |
813 |
> |
if (eventCount != ec || w.isTerminating()) |
814 |
> |
break; |
815 |
> |
if (System.nanoTime() - startTime >= SHRINK_RATE_NANOS) |
816 |
> |
tryShutdownUnusedWorker(ec); |
817 |
> |
} |
818 |
> |
} |
819 |
|
} |
820 |
|
} |
821 |
|
|
822 |
+ |
// Maintaining parallelism |
823 |
+ |
|
824 |
|
/** |
825 |
< |
* Ensures eventCount on exit is different (mod 2^32) than on |
716 |
< |
* entry and wakes up all waiters |
825 |
> |
* Pushes worker onto the spare stack. |
826 |
|
*/ |
827 |
< |
private void signalEvent() { |
828 |
< |
int c; |
829 |
< |
do {} while (!UNSAFE.compareAndSwapInt(this, eventCountOffset, |
830 |
< |
c = eventCount, c+1)); |
722 |
< |
releaseWaiters(); |
827 |
> |
final void pushSpare(ForkJoinWorkerThread w) { |
828 |
> |
int ns = (++w.spareCount << SPARE_COUNT_SHIFT) | (w.poolIndex + 1); |
829 |
> |
do {} while (!UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
830 |
> |
w.nextSpare = spareWaiters,ns)); |
831 |
|
} |
832 |
|
|
833 |
|
/** |
834 |
< |
* Advances eventCount and releases waiters until interference by |
835 |
< |
* other releasing threads is detected. |
834 |
> |
* Tries (once) to resume a spare if the number of running |
835 |
> |
* threads is less than target. |
836 |
|
*/ |
837 |
< |
final void signalWork() { |
838 |
< |
// EventCount CAS failures are OK -- any change in count suffices. |
839 |
< |
int ec; |
840 |
< |
UNSAFE.compareAndSwapInt(this, eventCountOffset, ec=eventCount, ec+1); |
841 |
< |
outer:for (;;) { |
842 |
< |
long top = eventWaiters; |
843 |
< |
ec = eventCount; |
844 |
< |
for (;;) { |
845 |
< |
ForkJoinWorkerThread[] ws; ForkJoinWorkerThread w; |
846 |
< |
int id = (int)(top & WAITER_INDEX_MASK); |
847 |
< |
if (id <= 0 || (int)(top >>> EVENT_COUNT_SHIFT) == ec) |
848 |
< |
return; |
849 |
< |
if ((ws = workers).length < id || (w = ws[id - 1]) == null || |
850 |
< |
!UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
851 |
< |
top, top = w.nextWaiter)) |
852 |
< |
continue outer; // possibly stale; reread |
837 |
> |
private void tryResumeSpare() { |
838 |
> |
int sw, id; |
839 |
> |
ForkJoinWorkerThread[] ws = workers; |
840 |
> |
int n = ws.length; |
841 |
> |
ForkJoinWorkerThread w; |
842 |
> |
if ((sw = spareWaiters) != 0 && |
843 |
> |
(id = (sw & SPARE_ID_MASK) - 1) >= 0 && |
844 |
> |
id < n && (w = ws[id]) != null && |
845 |
> |
(runState >= TERMINATING || |
846 |
> |
(workerCounts & RUNNING_COUNT_MASK) < parallelism) && |
847 |
> |
spareWaiters == sw && |
848 |
> |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
849 |
> |
sw, w.nextSpare)) { |
850 |
> |
int c; // increment running count before resume |
851 |
> |
do {} while (!UNSAFE.compareAndSwapInt |
852 |
> |
(this, workerCountsOffset, |
853 |
> |
c = workerCounts, c + ONE_RUNNING)); |
854 |
> |
if (w.tryUnsuspend()) |
855 |
|
LockSupport.unpark(w); |
856 |
< |
if (top != eventWaiters) // let someone else take over |
857 |
< |
return; |
748 |
< |
} |
856 |
> |
else // back out if w was shutdown |
857 |
> |
decrementWorkerCounts(ONE_RUNNING, 0); |
858 |
|
} |
859 |
|
} |
860 |
|
|
861 |
|
/** |
862 |
< |
* If worker is inactive, blocks until terminating or event count |
863 |
< |
* advances from last value held by worker; in any case helps |
864 |
< |
* release others. |
865 |
< |
* |
866 |
< |
* @param w the calling worker thread |
862 |
> |
* Tries to increase the number of running workers if below target |
863 |
> |
* parallelism: If a spare exists tries to resume it via |
864 |
> |
* tryResumeSpare. Otherwise, if not enough total workers or all |
865 |
> |
* existing workers are busy, adds a new worker. In all cases also |
866 |
> |
* helps wake up releasable workers waiting for work. |
867 |
|
*/ |
868 |
< |
private void eventSync(ForkJoinWorkerThread w) { |
869 |
< |
if (!w.active) { |
870 |
< |
int prev = w.lastEventCount; |
871 |
< |
long nextTop = (((long)prev << EVENT_COUNT_SHIFT) | |
872 |
< |
((long)(w.poolIndex + 1))); |
873 |
< |
long top; |
874 |
< |
while ((runState < SHUTDOWN || !tryTerminate(false)) && |
875 |
< |
(((int)(top = eventWaiters) & WAITER_INDEX_MASK) == 0 || |
876 |
< |
(int)(top >>> EVENT_COUNT_SHIFT) == prev) && |
877 |
< |
eventCount == prev) { |
878 |
< |
if (UNSAFE.compareAndSwapLong(this, eventWaitersOffset, |
879 |
< |
w.nextWaiter = top, nextTop)) { |
880 |
< |
accumulateStealCount(w); // transfer steals while idle |
881 |
< |
Thread.interrupted(); // clear/ignore interrupt |
882 |
< |
while (eventCount == prev) |
883 |
< |
w.doPark(); |
868 |
> |
private void helpMaintainParallelism() { |
869 |
> |
int pc = parallelism; |
870 |
> |
int wc, rs, tc; |
871 |
> |
while (((wc = workerCounts) & RUNNING_COUNT_MASK) < pc && |
872 |
> |
(rs = runState) < TERMINATING) { |
873 |
> |
if (spareWaiters != 0) |
874 |
> |
tryResumeSpare(); |
875 |
> |
else if ((tc = wc >>> TOTAL_COUNT_SHIFT) >= MAX_WORKERS || |
876 |
> |
(tc >= pc && (rs & ACTIVE_COUNT_MASK) != tc)) |
877 |
> |
break; // enough total |
878 |
> |
else if (runState == rs && workerCounts == wc && |
879 |
> |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
880 |
> |
wc + (ONE_RUNNING|ONE_TOTAL))) { |
881 |
> |
ForkJoinWorkerThread w = null; |
882 |
> |
Throwable fail = null; |
883 |
> |
try { |
884 |
> |
w = factory.newThread(this); |
885 |
> |
} catch (Throwable ex) { |
886 |
> |
fail = ex; |
887 |
> |
} |
888 |
> |
if (w == null) { // null or exceptional factory return |
889 |
> |
decrementWorkerCounts(ONE_RUNNING, ONE_TOTAL); |
890 |
> |
tryTerminate(false); // handle failure during shutdown |
891 |
> |
// If originating from an external caller, |
892 |
> |
// propagate exception, else ignore |
893 |
> |
if (fail != null && runState < TERMINATING && |
894 |
> |
!(Thread.currentThread() instanceof |
895 |
> |
ForkJoinWorkerThread)) |
896 |
> |
UNSAFE.throwException(fail); |
897 |
|
break; |
898 |
|
} |
899 |
+ |
w.start(recordWorker(w), ueh); |
900 |
+ |
if ((workerCounts >>> TOTAL_COUNT_SHIFT) >= pc) |
901 |
+ |
break; // add at most one unless total below target |
902 |
+ |
} |
903 |
+ |
} |
904 |
+ |
if (eventWaiters != 0L) |
905 |
+ |
releaseEventWaiters(); |
906 |
+ |
} |
907 |
+ |
|
908 |
+ |
/** |
909 |
+ |
* Callback from the oldest waiter in awaitEvent waking up after a |
910 |
+ |
* period of non-use. If all workers are idle, tries (once) to |
911 |
+ |
* shutdown an event waiter or a spare, if one exists. Note that |
912 |
+ |
* we don't need CAS or locks here because the method is called |
913 |
+ |
* only from one thread occasionally waking (and even misfires are |
914 |
+ |
* OK). Note that until the shutdown worker fully terminates, |
915 |
+ |
* workerCounts will overestimate total count, which is tolerable. |
916 |
+ |
* |
917 |
+ |
* @param ec the event count waited on by caller (to abort |
918 |
+ |
* attempt if count has since changed). |
919 |
+ |
*/ |
920 |
+ |
private void tryShutdownUnusedWorker(int ec) { |
921 |
+ |
if (runState == 0 && eventCount == ec) { // only trigger if all idle |
922 |
+ |
ForkJoinWorkerThread[] ws = workers; |
923 |
+ |
int n = ws.length; |
924 |
+ |
ForkJoinWorkerThread w = null; |
925 |
+ |
boolean shutdown = false; |
926 |
+ |
int sw; |
927 |
+ |
long h; |
928 |
+ |
if ((sw = spareWaiters) != 0) { // prefer killing spares |
929 |
+ |
int id = (sw & SPARE_ID_MASK) - 1; |
930 |
+ |
if (id >= 0 && id < n && (w = ws[id]) != null && |
931 |
+ |
UNSAFE.compareAndSwapInt(this, spareWaitersOffset, |
932 |
+ |
sw, w.nextSpare)) |
933 |
+ |
shutdown = true; |
934 |
+ |
} |
935 |
+ |
else if ((h = eventWaiters) != 0L) { |
936 |
+ |
long nh; |
937 |
+ |
int id = (((int)h) & WAITER_ID_MASK) - 1; |
938 |
+ |
if (id >= 0 && id < n && (w = ws[id]) != null && |
939 |
+ |
(nh = w.nextWaiter) != 0L && // keep at least one worker |
940 |
+ |
UNSAFE.compareAndSwapLong(this, eventWaitersOffset, h, nh)) |
941 |
+ |
shutdown = true; |
942 |
+ |
} |
943 |
+ |
if (w != null && shutdown) { |
944 |
+ |
w.shutdown(); |
945 |
+ |
LockSupport.unpark(w); |
946 |
|
} |
778 |
– |
w.lastEventCount = eventCount; |
947 |
|
} |
948 |
< |
releaseWaiters(); |
948 |
> |
releaseEventWaiters(); // in case of interference |
949 |
|
} |
950 |
|
|
951 |
|
/** |
952 |
|
* Callback from workers invoked upon each top-level action (i.e., |
953 |
< |
* stealing a task or taking a submission and running |
954 |
< |
* it). Performs one or both of the following: |
953 |
> |
* stealing a task or taking a submission and running it). |
954 |
> |
* Performs one or more of the following: |
955 |
|
* |
956 |
< |
* * If the worker cannot find work, updates its active status to |
957 |
< |
* inactive and updates activeCount unless there is contention, in |
958 |
< |
* which case it may try again (either in this or a subsequent |
959 |
< |
* call). Additionally, awaits the next task event and/or helps |
960 |
< |
* wake up other releasable waiters. |
961 |
< |
* |
962 |
< |
* * If there are too many running threads, suspends this worker |
963 |
< |
* (first forcing inactivation if necessary). If it is not |
964 |
< |
* resumed before a keepAlive elapses, the worker may be "trimmed" |
965 |
< |
* -- killed while suspended within suspendAsSpare. Otherwise, |
966 |
< |
* upon resume it rechecks to make sure that it is still needed. |
956 |
> |
* 1. If the worker is active and either did not run a task |
957 |
> |
* or there are too many workers, try to set its active status |
958 |
> |
* to inactive and update activeCount. On contention, we may |
959 |
> |
* try again in this or a subsequent call. |
960 |
> |
* |
961 |
> |
* 2. If not enough total workers, help create some. |
962 |
> |
* |
963 |
> |
* 3. If there are too many running workers, suspend this worker |
964 |
> |
* (first forcing inactive if necessary). If it is not needed, |
965 |
> |
* it may be shutdown while suspended (via |
966 |
> |
* tryShutdownUnusedWorker). Otherwise, upon resume it |
967 |
> |
* rechecks running thread count and need for event sync. |
968 |
> |
* |
969 |
> |
* 4. If worker did not run a task, await the next task event via |
970 |
> |
* eventSync if necessary (first forcing inactivation), upon |
971 |
> |
* which the worker may be shutdown via |
972 |
> |
* tryShutdownUnusedWorker. Otherwise, help release any |
973 |
> |
* existing event waiters that are now releasable, |
974 |
|
* |
975 |
|
* @param w the worker |
976 |
< |
* @param worked false if the worker scanned for work but didn't |
802 |
< |
* find any (in which case it may block waiting for work). |
976 |
> |
* @param ran true if worker ran a task since last call to this method |
977 |
|
*/ |
978 |
< |
final void preStep(ForkJoinWorkerThread w, boolean worked) { |
978 |
> |
final void preStep(ForkJoinWorkerThread w, boolean ran) { |
979 |
> |
int wec = w.lastEventCount; |
980 |
|
boolean active = w.active; |
981 |
< |
boolean inactivate = !worked & active; |
982 |
< |
for (;;) { |
983 |
< |
if (inactivate) { |
984 |
< |
int rs = runState; |
985 |
< |
if (UNSAFE.compareAndSwapInt(this, runStateOffset, |
986 |
< |
rs, rs - ONE_ACTIVE)) |
987 |
< |
inactivate = active = w.active = false; |
981 |
> |
boolean inactivate = false; |
982 |
> |
int pc = parallelism; |
983 |
> |
while (w.runState == 0) { |
984 |
> |
int rs = runState; |
985 |
> |
if (rs >= TERMINATING) { // propagate shutdown |
986 |
> |
w.shutdown(); |
987 |
> |
break; |
988 |
|
} |
989 |
< |
int wc = workerCounts; |
990 |
< |
if ((wc & RUNNING_COUNT_MASK) <= parallelism) { |
991 |
< |
if (!worked) |
992 |
< |
eventSync(w); |
993 |
< |
return; |
989 |
> |
if ((inactivate || (active && (rs & ACTIVE_COUNT_MASK) >= pc)) && |
990 |
> |
UNSAFE.compareAndSwapInt(this, runStateOffset, rs, --rs)) { |
991 |
> |
inactivate = active = w.active = false; |
992 |
> |
if (rs == SHUTDOWN) { // all inactive and shut down |
993 |
> |
tryTerminate(false); |
994 |
> |
continue; |
995 |
> |
} |
996 |
> |
} |
997 |
> |
int wc = workerCounts; // try to suspend as spare |
998 |
> |
if ((wc & RUNNING_COUNT_MASK) > pc) { |
999 |
> |
if (!(inactivate |= active) && // must inactivate to suspend |
1000 |
> |
workerCounts == wc && |
1001 |
> |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1002 |
> |
wc, wc - ONE_RUNNING)) |
1003 |
> |
w.suspendAsSpare(); |
1004 |
> |
} |
1005 |
> |
else if ((wc >>> TOTAL_COUNT_SHIFT) < pc) |
1006 |
> |
helpMaintainParallelism(); // not enough workers |
1007 |
> |
else if (ran) |
1008 |
> |
break; |
1009 |
> |
else { |
1010 |
> |
long h = eventWaiters; |
1011 |
> |
int ec = eventCount; |
1012 |
> |
if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != ec) |
1013 |
> |
releaseEventWaiters(); // release others before waiting |
1014 |
> |
else if (ec != wec) { |
1015 |
> |
w.lastEventCount = ec; // no need to wait |
1016 |
> |
break; |
1017 |
> |
} |
1018 |
> |
else if (!(inactivate |= active)) |
1019 |
> |
eventSync(w, wec); // must inactivate before sync |
1020 |
|
} |
820 |
– |
if (!(inactivate |= active) && // must inactivate to suspend |
821 |
– |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
822 |
– |
wc, wc - ONE_RUNNING) && |
823 |
– |
!w.suspendAsSpare()) // false if trimmed |
824 |
– |
return; |
1021 |
|
} |
1022 |
|
} |
1023 |
|
|
1024 |
|
/** |
1025 |
< |
* Tries to decrement running count, and if so, possibly creates |
1026 |
< |
* or resumes compensating threads before blocking on task joinMe. |
831 |
< |
* This code is sprawled out with manual inlining to evade some |
832 |
< |
* JIT oddities. |
1025 |
> |
* Helps and/or blocks awaiting join of the given task. |
1026 |
> |
* See above for explanation. |
1027 |
|
* |
1028 |
|
* @param joinMe the task to join |
1029 |
< |
* @return task status on exit |
1030 |
< |
*/ |
1031 |
< |
final int tryAwaitJoin(ForkJoinTask<?> joinMe) { |
1032 |
< |
int cw = workerCounts; // read now to spoil CAS if counts change as ... |
1033 |
< |
releaseWaiters(); // ... a byproduct of releaseWaiters |
1034 |
< |
int stat = joinMe.status; |
1035 |
< |
if (stat >= 0 && // inline variant of tryDecrementRunningCount |
1036 |
< |
(cw & RUNNING_COUNT_MASK) > 0 && |
1037 |
< |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1038 |
< |
cw, cw - ONE_RUNNING)) { |
1039 |
< |
int pc = parallelism; |
1040 |
< |
int scans = 0; // to require confirming passes to add threads |
1041 |
< |
outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) { |
1042 |
< |
if ((stat = joinMe.status) < 0) |
1043 |
< |
break; |
1044 |
< |
ForkJoinWorkerThread spare = null; |
1045 |
< |
ForkJoinWorkerThread[] ws = workers; |
1046 |
< |
int nws = ws.length; |
1047 |
< |
for (int i = 0; i < nws; ++i) { |
1048 |
< |
ForkJoinWorkerThread w = ws[i]; |
1049 |
< |
if (w != null && w.isSuspended()) { |
1050 |
< |
spare = w; |
1051 |
< |
break; |
1052 |
< |
} |
1029 |
> |
* @param worker the current worker thread |
1030 |
> |
* @param timed true if wait should time out |
1031 |
> |
* @param nanos timeout value if timed |
1032 |
> |
*/ |
1033 |
> |
final void awaitJoin(ForkJoinTask<?> joinMe, ForkJoinWorkerThread worker, |
1034 |
> |
boolean timed, long nanos) { |
1035 |
> |
long startTime = timed? System.nanoTime() : 0L; |
1036 |
> |
int retries = 2 + (parallelism >> 2); // #helpJoins before blocking |
1037 |
> |
boolean running = true; // false when count decremented |
1038 |
> |
while (joinMe.status >= 0) { |
1039 |
> |
if (runState >= TERMINATING) { |
1040 |
> |
joinMe.cancelIgnoringExceptions(); |
1041 |
> |
break; |
1042 |
> |
} |
1043 |
> |
running = worker.helpJoinTask(joinMe, running); |
1044 |
> |
if (joinMe.status < 0) |
1045 |
> |
break; |
1046 |
> |
if (retries > 0) { |
1047 |
> |
--retries; |
1048 |
> |
continue; |
1049 |
> |
} |
1050 |
> |
int wc = workerCounts; |
1051 |
> |
if ((wc & RUNNING_COUNT_MASK) != 0) { |
1052 |
> |
if (running) { |
1053 |
> |
if (!UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1054 |
> |
wc, wc - ONE_RUNNING)) |
1055 |
> |
continue; |
1056 |
> |
running = false; |
1057 |
|
} |
1058 |
< |
if ((stat = joinMe.status) < 0) // recheck to narrow race |
1059 |
< |
break; |
1060 |
< |
int wc = workerCounts; |
1061 |
< |
int rc = wc & RUNNING_COUNT_MASK; |
864 |
< |
if (rc >= pc) |
1058 |
> |
long h = eventWaiters; |
1059 |
> |
if (h != 0L && (int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
1060 |
> |
releaseEventWaiters(); |
1061 |
> |
if (joinMe.status < 0) |
1062 |
|
break; |
1063 |
< |
if (spare != null) { |
1064 |
< |
if (spare.tryUnsuspend()) { |
1065 |
< |
int c; // inline incrementRunningCount |
1066 |
< |
do {} while (!UNSAFE.compareAndSwapInt |
1067 |
< |
(this, workerCountsOffset, |
871 |
< |
c = workerCounts, c + ONE_RUNNING)); |
872 |
< |
LockSupport.unpark(spare); |
873 |
< |
break; |
1063 |
> |
if ((workerCounts & RUNNING_COUNT_MASK) != 0) { |
1064 |
> |
long ms; int ns; |
1065 |
> |
if (!timed) { |
1066 |
> |
ms = JOIN_TIMEOUT_MILLIS; |
1067 |
> |
ns = 0; |
1068 |
|
} |
1069 |
< |
continue; |
1070 |
< |
} |
1071 |
< |
int tc = wc >>> TOTAL_COUNT_SHIFT; |
1072 |
< |
int sc = tc - pc; |
1073 |
< |
if (rc > 0) { |
1074 |
< |
int p = pc; |
1075 |
< |
int s = sc; |
1076 |
< |
while (s-- >= 0) { // try keeping 3/4 live |
1077 |
< |
if (rc > (p -= (p >>> 2) + 1)) |
1078 |
< |
break outer; |
1069 |
> |
else { // at most JOIN_TIMEOUT_MILLIS per wait |
1070 |
> |
long nt = nanos - (System.nanoTime() - startTime); |
1071 |
> |
if (nt <= 0L) |
1072 |
> |
break; |
1073 |
> |
ms = nt / 1000000; |
1074 |
> |
if (ms > JOIN_TIMEOUT_MILLIS) { |
1075 |
> |
ms = JOIN_TIMEOUT_MILLIS; |
1076 |
> |
ns = 0; |
1077 |
> |
} |
1078 |
> |
else |
1079 |
> |
ns = (int) (nt % 1000000); |
1080 |
|
} |
1081 |
< |
} |
1082 |
< |
if (scans++ > sc && tc < MAX_THREADS && |
888 |
< |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
889 |
< |
wc + (ONE_RUNNING|ONE_TOTAL))) { |
890 |
< |
addWorker(); |
891 |
< |
break; |
1081 |
> |
if (joinMe.internalAwaitDone(ms, ns) < 0) |
1082 |
> |
break; |
1083 |
|
} |
1084 |
|
} |
1085 |
< |
if (stat >= 0) |
1086 |
< |
stat = joinMe.internalAwaitDone(); |
1087 |
< |
int c; // inline incrementRunningCount |
1085 |
> |
helpMaintainParallelism(); |
1086 |
> |
} |
1087 |
> |
if (!running) { |
1088 |
> |
int c; |
1089 |
|
do {} while (!UNSAFE.compareAndSwapInt |
1090 |
|
(this, workerCountsOffset, |
1091 |
|
c = workerCounts, c + ONE_RUNNING)); |
1092 |
|
} |
901 |
– |
return stat; |
1093 |
|
} |
1094 |
|
|
1095 |
|
/** |
1096 |
< |
* Same idea as (and mostly pasted from) tryAwaitJoin, but |
906 |
< |
* self-contained |
1096 |
> |
* Same idea as awaitJoin, but no helping, retries, or timeouts. |
1097 |
|
*/ |
1098 |
|
final void awaitBlocker(ManagedBlocker blocker) |
1099 |
|
throws InterruptedException { |
1100 |
< |
for (;;) { |
911 |
< |
if (blocker.isReleasable()) |
912 |
< |
return; |
913 |
< |
int cw = workerCounts; |
914 |
< |
releaseWaiters(); |
915 |
< |
if ((cw & RUNNING_COUNT_MASK) > 0 && |
916 |
< |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
917 |
< |
cw, cw - ONE_RUNNING)) |
918 |
< |
break; |
919 |
< |
} |
920 |
< |
boolean done = false; |
921 |
< |
int pc = parallelism; |
922 |
< |
int scans = 0; |
923 |
< |
outer: while ((workerCounts & RUNNING_COUNT_MASK) < pc) { |
924 |
< |
if (done = blocker.isReleasable()) |
925 |
< |
break; |
926 |
< |
ForkJoinWorkerThread spare = null; |
927 |
< |
ForkJoinWorkerThread[] ws = workers; |
928 |
< |
int nws = ws.length; |
929 |
< |
for (int i = 0; i < nws; ++i) { |
930 |
< |
ForkJoinWorkerThread w = ws[i]; |
931 |
< |
if (w != null && w.isSuspended()) { |
932 |
< |
spare = w; |
933 |
< |
break; |
934 |
< |
} |
935 |
< |
} |
936 |
< |
if (done = blocker.isReleasable()) |
937 |
< |
break; |
1100 |
> |
while (!blocker.isReleasable()) { |
1101 |
|
int wc = workerCounts; |
1102 |
< |
int rc = wc & RUNNING_COUNT_MASK; |
1103 |
< |
if (rc >= pc) |
1104 |
< |
break; |
1105 |
< |
if (spare != null) { |
1106 |
< |
if (spare.tryUnsuspend()) { |
1102 |
> |
if ((wc & RUNNING_COUNT_MASK) == 0) |
1103 |
> |
helpMaintainParallelism(); |
1104 |
> |
else if (UNSAFE.compareAndSwapInt(this, workerCountsOffset, |
1105 |
> |
wc, wc - ONE_RUNNING)) { |
1106 |
> |
try { |
1107 |
> |
while (!blocker.isReleasable()) { |
1108 |
> |
long h = eventWaiters; |
1109 |
> |
if (h != 0L && |
1110 |
> |
(int)(h >>> EVENT_COUNT_SHIFT) != eventCount) |
1111 |
> |
releaseEventWaiters(); |
1112 |
> |
else if ((workerCounts & RUNNING_COUNT_MASK) == 0 && |
1113 |
> |
runState < TERMINATING) |
1114 |
> |
helpMaintainParallelism(); |
1115 |
> |
else if (blocker.block()) |
1116 |
> |
break; |
1117 |
> |
} |
1118 |
> |
} finally { |
1119 |
|
int c; |
1120 |
|
do {} while (!UNSAFE.compareAndSwapInt |
1121 |
|
(this, workerCountsOffset, |
1122 |
|
c = workerCounts, c + ONE_RUNNING)); |
948 |
– |
LockSupport.unpark(spare); |
949 |
– |
break; |
1123 |
|
} |
951 |
– |
continue; |
952 |
– |
} |
953 |
– |
int tc = wc >>> TOTAL_COUNT_SHIFT; |
954 |
– |
int sc = tc - pc; |
955 |
– |
if (rc > 0) { |
956 |
– |
int p = pc; |
957 |
– |
int s = sc; |
958 |
– |
while (s-- >= 0) { |
959 |
– |
if (rc > (p -= (p >>> 2) + 1)) |
960 |
– |
break outer; |
961 |
– |
} |
962 |
– |
} |
963 |
– |
if (scans++ > sc && tc < MAX_THREADS && |
964 |
– |
UNSAFE.compareAndSwapInt(this, workerCountsOffset, wc, |
965 |
– |
wc + (ONE_RUNNING|ONE_TOTAL))) { |
966 |
– |
addWorker(); |
1124 |
|
break; |
1125 |
|
} |
1126 |
|
} |
1127 |
< |
try { |
971 |
< |
if (!done) |
972 |
< |
do {} while (!blocker.isReleasable() && |
973 |
< |
!blocker.block()); |
974 |
< |
} finally { |
975 |
< |
int c; |
976 |
< |
do {} while (!UNSAFE.compareAndSwapInt |
977 |
< |
(this, workerCountsOffset, |
978 |
< |
c = workerCounts, c + ONE_RUNNING)); |
979 |
< |
} |
980 |
< |
} |
1127 |
> |
} |
1128 |
|
|
1129 |
|
/** |
1130 |
|
* Possibly initiates and/or completes termination. |
1147 |
|
// Finish now if all threads terminated; else in some subsequent call |
1148 |
|
if ((workerCounts >>> TOTAL_COUNT_SHIFT) == 0) { |
1149 |
|
advanceRunLevel(TERMINATED); |
1150 |
< |
termination.arrive(); |
1150 |
> |
termination.forceTermination(); |
1151 |
|
} |
1152 |
|
return true; |
1153 |
|
} |
1154 |
|
|
1155 |
|
/** |
1156 |
|
* Actions on transition to TERMINATING |
1157 |
+ |
* |
1158 |
+ |
* Runs up to four passes through workers: (0) shutting down each |
1159 |
+ |
* (without waking up if parked) to quickly spread notifications |
1160 |
+ |
* without unnecessary bouncing around event queues etc (1) wake |
1161 |
+ |
* up and help cancel tasks (2) interrupt (3) mop up races with |
1162 |
+ |
* interrupted workers |
1163 |
|
*/ |
1164 |
|
private void startTerminating() { |
1165 |
< |
for (int i = 0; i < 2; ++i) { // twice to mop up newly created workers |
1166 |
< |
cancelSubmissions(); |
1167 |
< |
shutdownWorkers(); |
1168 |
< |
cancelWorkerTasks(); |
1169 |
< |
signalEvent(); |
1170 |
< |
interruptWorkers(); |
1165 |
> |
cancelSubmissions(); |
1166 |
> |
for (int passes = 0; passes < 4 && workerCounts != 0; ++passes) { |
1167 |
> |
int c; // advance event count |
1168 |
> |
UNSAFE.compareAndSwapInt(this, eventCountOffset, |
1169 |
> |
c = eventCount, c+1); |
1170 |
> |
eventWaiters = 0L; // clobber lists |
1171 |
> |
spareWaiters = 0; |
1172 |
> |
for (ForkJoinWorkerThread w : workers) { |
1173 |
> |
if (w != null) { |
1174 |
> |
w.shutdown(); |
1175 |
> |
if (passes > 0 && !w.isTerminated()) { |
1176 |
> |
w.cancelTasks(); |
1177 |
> |
LockSupport.unpark(w); |
1178 |
> |
if (passes > 1 && !w.isInterrupted()) { |
1179 |
> |
try { |
1180 |
> |
w.interrupt(); |
1181 |
> |
} catch (SecurityException ignore) { |
1182 |
> |
} |
1183 |
> |
} |
1184 |
> |
} |
1185 |
> |
} |
1186 |
> |
} |
1187 |
|
} |
1188 |
|
} |
1189 |
|
|
1190 |
|
/** |
1191 |
< |
* Clear out and cancel submissions, ignoring exceptions |
1191 |
> |
* Clears out and cancels submissions, ignoring exceptions. |
1192 |
|
*/ |
1193 |
|
private void cancelSubmissions() { |
1194 |
|
ForkJoinTask<?> task; |
1200 |
|
} |
1201 |
|
} |
1202 |
|
|
1034 |
– |
/** |
1035 |
– |
* Sets all worker run states to at least shutdown, |
1036 |
– |
* also resuming suspended workers |
1037 |
– |
*/ |
1038 |
– |
private void shutdownWorkers() { |
1039 |
– |
ForkJoinWorkerThread[] ws = workers; |
1040 |
– |
int nws = ws.length; |
1041 |
– |
for (int i = 0; i < nws; ++i) { |
1042 |
– |
ForkJoinWorkerThread w = ws[i]; |
1043 |
– |
if (w != null) |
1044 |
– |
w.shutdown(); |
1045 |
– |
} |
1046 |
– |
} |
1047 |
– |
|
1048 |
– |
/** |
1049 |
– |
* Clears out and cancels all locally queued tasks |
1050 |
– |
*/ |
1051 |
– |
private void cancelWorkerTasks() { |
1052 |
– |
ForkJoinWorkerThread[] ws = workers; |
1053 |
– |
int nws = ws.length; |
1054 |
– |
for (int i = 0; i < nws; ++i) { |
1055 |
– |
ForkJoinWorkerThread w = ws[i]; |
1056 |
– |
if (w != null) |
1057 |
– |
w.cancelTasks(); |
1058 |
– |
} |
1059 |
– |
} |
1060 |
– |
|
1061 |
– |
/** |
1062 |
– |
* Unsticks all workers blocked on joins etc |
1063 |
– |
*/ |
1064 |
– |
private void interruptWorkers() { |
1065 |
– |
ForkJoinWorkerThread[] ws = workers; |
1066 |
– |
int nws = ws.length; |
1067 |
– |
for (int i = 0; i < nws; ++i) { |
1068 |
– |
ForkJoinWorkerThread w = ws[i]; |
1069 |
– |
if (w != null && !w.isTerminated()) { |
1070 |
– |
try { |
1071 |
– |
w.interrupt(); |
1072 |
– |
} catch (SecurityException ignore) { |
1073 |
– |
} |
1074 |
– |
} |
1075 |
– |
} |
1076 |
– |
} |
1077 |
– |
|
1203 |
|
// misc support for ForkJoinWorkerThread |
1204 |
|
|
1205 |
|
/** |
1206 |
< |
* Returns pool number |
1206 |
> |
* Returns pool number. |
1207 |
|
*/ |
1208 |
|
final int getPoolNumber() { |
1209 |
|
return poolNumber; |
1210 |
|
} |
1211 |
|
|
1212 |
|
/** |
1213 |
< |
* Accumulates steal count from a worker, clearing |
1214 |
< |
* the worker's value |
1213 |
> |
* Tries to accumulate steal count from a worker, clearing |
1214 |
> |
* the worker's value if successful. |
1215 |
> |
* |
1216 |
> |
* @return true if worker steal count now zero |
1217 |
|
*/ |
1218 |
< |
final void accumulateStealCount(ForkJoinWorkerThread w) { |
1218 |
> |
final boolean tryAccumulateStealCount(ForkJoinWorkerThread w) { |
1219 |
|
int sc = w.stealCount; |
1220 |
< |
if (sc != 0) { |
1221 |
< |
long c; |
1222 |
< |
w.stealCount = 0; |
1223 |
< |
do {} while (!UNSAFE.compareAndSwapLong(this, stealCountOffset, |
1224 |
< |
c = stealCount, c + sc)); |
1220 |
> |
long c = stealCount; |
1221 |
> |
// CAS even if zero, for fence effects |
1222 |
> |
if (UNSAFE.compareAndSwapLong(this, stealCountOffset, c, c + sc)) { |
1223 |
> |
if (sc != 0) |
1224 |
> |
w.stealCount = 0; |
1225 |
> |
return true; |
1226 |
|
} |
1227 |
+ |
return sc == 0; |
1228 |
|
} |
1229 |
|
|
1230 |
|
/** |
1232 |
|
* active thread. |
1233 |
|
*/ |
1234 |
|
final int idlePerActive() { |
1235 |
< |
int pc = parallelism; // use targeted parallelism, not rc |
1236 |
< |
int ac = runState; // no mask -- artifically boosts during shutdown |
1235 |
> |
int pc = parallelism; // use parallelism, not rc |
1236 |
> |
int ac = runState; // no mask -- artificially boosts during shutdown |
1237 |
|
// Use exact results for small values, saturate past 4 |
1238 |
< |
return pc <= ac? 0 : pc >>> 1 <= ac? 1 : pc >>> 2 <= ac? 3 : pc >>> 3; |
1238 |
> |
return ((pc <= ac) ? 0 : |
1239 |
> |
(pc >>> 1 <= ac) ? 1 : |
1240 |
> |
(pc >>> 2 <= ac) ? 3 : |
1241 |
> |
pc >>> 3); |
1242 |
|
} |
1243 |
|
|
1244 |
|
// Public and protected methods |
1286 |
|
* use {@link java.lang.Runtime#availableProcessors}. |
1287 |
|
* @param factory the factory for creating new threads. For default value, |
1288 |
|
* use {@link #defaultForkJoinWorkerThreadFactory}. |
1289 |
< |
* @param handler the handler for internal worker threads that |
1290 |
< |
* terminate due to unrecoverable errors encountered while executing |
1291 |
< |
* tasks. For default value, use <code>null</code>. |
1292 |
< |
* @param asyncMode if true, |
1289 |
> |
* @param handler the handler for internal worker threads that |
1290 |
> |
* terminate due to unrecoverable errors encountered while executing |
1291 |
> |
* tasks. For default value, use {@code null}. |
1292 |
> |
* @param asyncMode if true, |
1293 |
|
* establishes local first-in-first-out scheduling mode for forked |
1294 |
|
* tasks that are never joined. This mode may be more appropriate |
1295 |
|
* than default locally stack-based mode in applications in which |
1296 |
|
* worker threads only process event-style asynchronous tasks. |
1297 |
< |
* For default value, use <code>false</code>. |
1297 |
> |
* For default value, use {@code false}. |
1298 |
|
* @throws IllegalArgumentException if parallelism less than or |
1299 |
|
* equal to zero, or greater than implementation limit |
1300 |
|
* @throws NullPointerException if the factory is null |
1303 |
|
* because it does not hold {@link |
1304 |
|
* java.lang.RuntimePermission}{@code ("modifyThread")} |
1305 |
|
*/ |
1306 |
< |
public ForkJoinPool(int parallelism, |
1306 |
> |
public ForkJoinPool(int parallelism, |
1307 |
|
ForkJoinWorkerThreadFactory factory, |
1308 |
|
Thread.UncaughtExceptionHandler handler, |
1309 |
|
boolean asyncMode) { |
1310 |
|
checkPermission(); |
1311 |
|
if (factory == null) |
1312 |
|
throw new NullPointerException(); |
1313 |
< |
if (parallelism <= 0 || parallelism > MAX_THREADS) |
1313 |
> |
if (parallelism <= 0 || parallelism > MAX_WORKERS) |
1314 |
|
throw new IllegalArgumentException(); |
1315 |
|
this.parallelism = parallelism; |
1316 |
|
this.factory = factory; |
1329 |
|
* @param pc the initial parallelism level |
1330 |
|
*/ |
1331 |
|
private static int initialArraySizeFor(int pc) { |
1332 |
< |
// See Hackers Delight, sec 3.2. We know MAX_THREADS < (1 >>> 16) |
1333 |
< |
int size = pc < MAX_THREADS ? pc + 1 : MAX_THREADS; |
1332 |
> |
// If possible, initially allocate enough space for one spare |
1333 |
> |
int size = pc < MAX_WORKERS ? pc + 1 : MAX_WORKERS; |
1334 |
> |
// See Hackers Delight, sec 3.2. We know MAX_WORKERS < (1 >>> 16) |
1335 |
|
size |= size >>> 1; |
1336 |
|
size |= size >>> 2; |
1337 |
|
size |= size >>> 4; |
1342 |
|
// Execution methods |
1343 |
|
|
1344 |
|
/** |
1345 |
< |
* Common code for execute, invoke and submit |
1345 |
> |
* Submits task and creates, starts, or resumes some workers if necessary |
1346 |
|
*/ |
1347 |
|
private <T> void doSubmit(ForkJoinTask<T> task) { |
1348 |
< |
if (task == null) |
1349 |
< |
throw new NullPointerException(); |
1350 |
< |
if (runState >= SHUTDOWN) |
1351 |
< |
throw new RejectedExecutionException(); |
1219 |
< |
// Convert submissions to current pool into forks |
1220 |
< |
Thread t = Thread.currentThread(); |
1221 |
< |
ForkJoinWorkerThread w; |
1222 |
< |
if ((t instanceof ForkJoinWorkerThread) && |
1223 |
< |
(w = (ForkJoinWorkerThread) t).pool == this) |
1224 |
< |
w.pushTask(task); |
1225 |
< |
else { |
1226 |
< |
submissionQueue.offer(task); |
1227 |
< |
signalEvent(); |
1228 |
< |
ensureEnoughTotalWorkers(); |
1229 |
< |
} |
1348 |
> |
submissionQueue.offer(task); |
1349 |
> |
int c; // try to increment event count -- CAS failure OK |
1350 |
> |
UNSAFE.compareAndSwapInt(this, eventCountOffset, c = eventCount, c+1); |
1351 |
> |
helpMaintainParallelism(); |
1352 |
|
} |
1353 |
|
|
1354 |
|
/** |
1355 |
|
* Performs the given task, returning its result upon completion. |
1234 |
– |
* If the caller is already engaged in a fork/join computation in |
1235 |
– |
* the current pool, this method is equivalent in effect to |
1236 |
– |
* {@link ForkJoinTask#invoke}. |
1356 |
|
* |
1357 |
|
* @param task the task |
1358 |
|
* @return the task's result |
1361 |
|
* scheduled for execution |
1362 |
|
*/ |
1363 |
|
public <T> T invoke(ForkJoinTask<T> task) { |
1364 |
< |
doSubmit(task); |
1365 |
< |
return task.join(); |
1364 |
> |
if (task == null) |
1365 |
> |
throw new NullPointerException(); |
1366 |
> |
if (runState >= SHUTDOWN) |
1367 |
> |
throw new RejectedExecutionException(); |
1368 |
> |
Thread t = Thread.currentThread(); |
1369 |
> |
if ((t instanceof ForkJoinWorkerThread) && |
1370 |
> |
((ForkJoinWorkerThread)t).pool == this) |
1371 |
> |
return task.invoke(); // bypass submit if in same pool |
1372 |
> |
else { |
1373 |
> |
doSubmit(task); |
1374 |
> |
return task.join(); |
1375 |
> |
} |
1376 |
> |
} |
1377 |
> |
|
1378 |
> |
/** |
1379 |
> |
* Unless terminating, forks task if within an ongoing FJ |
1380 |
> |
* computation in the current pool, else submits as external task. |
1381 |
> |
*/ |
1382 |
> |
private <T> void forkOrSubmit(ForkJoinTask<T> task) { |
1383 |
> |
if (runState >= SHUTDOWN) |
1384 |
> |
throw new RejectedExecutionException(); |
1385 |
> |
Thread t = Thread.currentThread(); |
1386 |
> |
if ((t instanceof ForkJoinWorkerThread) && |
1387 |
> |
((ForkJoinWorkerThread)t).pool == this) |
1388 |
> |
task.fork(); |
1389 |
> |
else |
1390 |
> |
doSubmit(task); |
1391 |
|
} |
1392 |
|
|
1393 |
|
/** |
1394 |
|
* Arranges for (asynchronous) execution of the given task. |
1251 |
– |
* If the caller is already engaged in a fork/join computation in |
1252 |
– |
* the current pool, this method is equivalent in effect to |
1253 |
– |
* {@link ForkJoinTask#fork}. |
1395 |
|
* |
1396 |
|
* @param task the task |
1397 |
|
* @throws NullPointerException if the task is null |
1399 |
|
* scheduled for execution |
1400 |
|
*/ |
1401 |
|
public void execute(ForkJoinTask<?> task) { |
1402 |
< |
doSubmit(task); |
1402 |
> |
if (task == null) |
1403 |
> |
throw new NullPointerException(); |
1404 |
> |
forkOrSubmit(task); |
1405 |
|
} |
1406 |
|
|
1407 |
|
// AbstractExecutorService methods |
1412 |
|
* scheduled for execution |
1413 |
|
*/ |
1414 |
|
public void execute(Runnable task) { |
1415 |
+ |
if (task == null) |
1416 |
+ |
throw new NullPointerException(); |
1417 |
|
ForkJoinTask<?> job; |
1418 |
|
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1419 |
|
job = (ForkJoinTask<?>) task; |
1420 |
|
else |
1421 |
|
job = ForkJoinTask.adapt(task, null); |
1422 |
< |
doSubmit(job); |
1422 |
> |
forkOrSubmit(job); |
1423 |
|
} |
1424 |
|
|
1425 |
|
/** |
1426 |
|
* Submits a ForkJoinTask for execution. |
1282 |
– |
* If the caller is already engaged in a fork/join computation in |
1283 |
– |
* the current pool, this method is equivalent in effect to |
1284 |
– |
* {@link ForkJoinTask#fork}. |
1427 |
|
* |
1428 |
|
* @param task the task to submit |
1429 |
|
* @return the task |
1432 |
|
* scheduled for execution |
1433 |
|
*/ |
1434 |
|
public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) { |
1435 |
< |
doSubmit(task); |
1435 |
> |
if (task == null) |
1436 |
> |
throw new NullPointerException(); |
1437 |
> |
forkOrSubmit(task); |
1438 |
|
return task; |
1439 |
|
} |
1440 |
|
|
1444 |
|
* scheduled for execution |
1445 |
|
*/ |
1446 |
|
public <T> ForkJoinTask<T> submit(Callable<T> task) { |
1447 |
+ |
if (task == null) |
1448 |
+ |
throw new NullPointerException(); |
1449 |
|
ForkJoinTask<T> job = ForkJoinTask.adapt(task); |
1450 |
< |
doSubmit(job); |
1450 |
> |
forkOrSubmit(job); |
1451 |
|
return job; |
1452 |
|
} |
1453 |
|
|
1457 |
|
* scheduled for execution |
1458 |
|
*/ |
1459 |
|
public <T> ForkJoinTask<T> submit(Runnable task, T result) { |
1460 |
+ |
if (task == null) |
1461 |
+ |
throw new NullPointerException(); |
1462 |
|
ForkJoinTask<T> job = ForkJoinTask.adapt(task, result); |
1463 |
< |
doSubmit(job); |
1463 |
> |
forkOrSubmit(job); |
1464 |
|
return job; |
1465 |
|
} |
1466 |
|
|
1470 |
|
* scheduled for execution |
1471 |
|
*/ |
1472 |
|
public ForkJoinTask<?> submit(Runnable task) { |
1473 |
+ |
if (task == null) |
1474 |
+ |
throw new NullPointerException(); |
1475 |
|
ForkJoinTask<?> job; |
1476 |
|
if (task instanceof ForkJoinTask<?>) // avoid re-wrap |
1477 |
|
job = (ForkJoinTask<?>) task; |
1478 |
|
else |
1479 |
|
job = ForkJoinTask.adapt(task, null); |
1480 |
< |
doSubmit(job); |
1480 |
> |
forkOrSubmit(job); |
1481 |
|
return job; |
1482 |
|
} |
1483 |
|
|
1537 |
|
|
1538 |
|
/** |
1539 |
|
* Returns the number of worker threads that have started but not |
1540 |
< |
* yet terminated. This result returned by this method may differ |
1540 |
> |
* yet terminated. The result returned by this method may differ |
1541 |
|
* from {@link #getParallelism} when threads are created to |
1542 |
|
* maintain parallelism when others are cooperatively blocked. |
1543 |
|
* |
1622 |
|
*/ |
1623 |
|
public long getQueuedTaskCount() { |
1624 |
|
long count = 0; |
1625 |
< |
ForkJoinWorkerThread[] ws = workers; |
1476 |
< |
int nws = ws.length; |
1477 |
< |
for (int i = 0; i < nws; ++i) { |
1478 |
< |
ForkJoinWorkerThread w = ws[i]; |
1625 |
> |
for (ForkJoinWorkerThread w : workers) |
1626 |
|
if (w != null) |
1627 |
|
count += w.getQueueSize(); |
1481 |
– |
} |
1628 |
|
return count; |
1629 |
|
} |
1630 |
|
|
1678 |
|
* @return the number of elements transferred |
1679 |
|
*/ |
1680 |
|
protected int drainTasksTo(Collection<? super ForkJoinTask<?>> c) { |
1681 |
< |
int n = submissionQueue.drainTo(c); |
1682 |
< |
ForkJoinWorkerThread[] ws = workers; |
1537 |
< |
int nws = ws.length; |
1538 |
< |
for (int i = 0; i < nws; ++i) { |
1539 |
< |
ForkJoinWorkerThread w = ws[i]; |
1540 |
< |
if (w != null) |
1541 |
< |
n += w.drainTasksTo(c); |
1542 |
< |
} |
1543 |
< |
return n; |
1544 |
< |
} |
1545 |
< |
|
1546 |
< |
/** |
1547 |
< |
* Returns count of total parks by existing workers. |
1548 |
< |
* Used during development only since not meaningful to users. |
1549 |
< |
*/ |
1550 |
< |
private int collectParkCount() { |
1551 |
< |
int count = 0; |
1552 |
< |
ForkJoinWorkerThread[] ws = workers; |
1553 |
< |
int nws = ws.length; |
1554 |
< |
for (int i = 0; i < nws; ++i) { |
1555 |
< |
ForkJoinWorkerThread w = ws[i]; |
1681 |
> |
int count = submissionQueue.drainTo(c); |
1682 |
> |
for (ForkJoinWorkerThread w : workers) |
1683 |
|
if (w != null) |
1684 |
< |
count += w.parkCount; |
1558 |
< |
} |
1684 |
> |
count += w.drainTasksTo(c); |
1685 |
|
return count; |
1686 |
|
} |
1687 |
|
|
1702 |
|
int pc = parallelism; |
1703 |
|
int rs = runState; |
1704 |
|
int ac = rs & ACTIVE_COUNT_MASK; |
1579 |
– |
// int pk = collectParkCount(); |
1705 |
|
return super.toString() + |
1706 |
|
"[" + runLevelToString(rs) + |
1707 |
|
", parallelism = " + pc + |
1711 |
|
", steals = " + st + |
1712 |
|
", tasks = " + qt + |
1713 |
|
", submissions = " + qs + |
1589 |
– |
// ", parks = " + pk + |
1714 |
|
"]"; |
1715 |
|
} |
1716 |
|
|
1785 |
|
} |
1786 |
|
|
1787 |
|
/** |
1788 |
+ |
* Returns true if terminating or terminated. Used by ForkJoinWorkerThread. |
1789 |
+ |
*/ |
1790 |
+ |
final boolean isAtLeastTerminating() { |
1791 |
+ |
return runState >= TERMINATING; |
1792 |
+ |
} |
1793 |
+ |
|
1794 |
+ |
/** |
1795 |
|
* Returns {@code true} if this pool has been shut down. |
1796 |
|
* |
1797 |
|
* @return {@code true} if this pool has been shut down |
1814 |
|
public boolean awaitTermination(long timeout, TimeUnit unit) |
1815 |
|
throws InterruptedException { |
1816 |
|
try { |
1817 |
< |
return termination.awaitAdvanceInterruptibly(0, timeout, unit) > 0; |
1818 |
< |
} catch(TimeoutException ex) { |
1817 |
> |
termination.awaitAdvanceInterruptibly(0, timeout, unit); |
1818 |
> |
} catch (TimeoutException ex) { |
1819 |
|
return false; |
1820 |
|
} |
1821 |
+ |
return true; |
1822 |
|
} |
1823 |
|
|
1824 |
|
/** |
1825 |
|
* Interface for extending managed parallelism for tasks running |
1826 |
|
* in {@link ForkJoinPool}s. |
1827 |
|
* |
1828 |
< |
* <p>A {@code ManagedBlocker} provides two methods. |
1829 |
< |
* Method {@code isReleasable} must return {@code true} if |
1830 |
< |
* blocking is not necessary. Method {@code block} blocks the |
1831 |
< |
* current thread if necessary (perhaps internally invoking |
1832 |
< |
* {@code isReleasable} before actually blocking). |
1828 |
> |
* <p>A {@code ManagedBlocker} provides two methods. Method |
1829 |
> |
* {@code isReleasable} must return {@code true} if blocking is |
1830 |
> |
* not necessary. Method {@code block} blocks the current thread |
1831 |
> |
* if necessary (perhaps internally invoking {@code isReleasable} |
1832 |
> |
* before actually blocking). The unusual methods in this API |
1833 |
> |
* accommodate synchronizers that may, but don't usually, block |
1834 |
> |
* for long periods. Similarly, they allow more efficient internal |
1835 |
> |
* handling of cases in which additional workers may be, but |
1836 |
> |
* usually are not, needed to ensure sufficient parallelism. |
1837 |
> |
* Toward this end, implementations of method {@code isReleasable} |
1838 |
> |
* must be amenable to repeated invocation. |
1839 |
|
* |
1840 |
|
* <p>For example, here is a ManagedBlocker based on a |
1841 |
|
* ReentrantLock: |
1853 |
|
* return hasLock || (hasLock = lock.tryLock()); |
1854 |
|
* } |
1855 |
|
* }}</pre> |
1856 |
+ |
* |
1857 |
+ |
* <p>Here is a class that possibly blocks waiting for an |
1858 |
+ |
* item on a given queue: |
1859 |
+ |
* <pre> {@code |
1860 |
+ |
* class QueueTaker<E> implements ManagedBlocker { |
1861 |
+ |
* final BlockingQueue<E> queue; |
1862 |
+ |
* volatile E item = null; |
1863 |
+ |
* QueueTaker(BlockingQueue<E> q) { this.queue = q; } |
1864 |
+ |
* public boolean block() throws InterruptedException { |
1865 |
+ |
* if (item == null) |
1866 |
+ |
* item = queue.take(); |
1867 |
+ |
* return true; |
1868 |
+ |
* } |
1869 |
+ |
* public boolean isReleasable() { |
1870 |
+ |
* return item != null || (item = queue.poll()) != null; |
1871 |
+ |
* } |
1872 |
+ |
* public E getItem() { // call after pool.managedBlock completes |
1873 |
+ |
* return item; |
1874 |
+ |
* } |
1875 |
+ |
* }}</pre> |
1876 |
|
*/ |
1877 |
|
public static interface ManagedBlocker { |
1878 |
|
/** |
1915 |
|
public static void managedBlock(ManagedBlocker blocker) |
1916 |
|
throws InterruptedException { |
1917 |
|
Thread t = Thread.currentThread(); |
1918 |
< |
if (t instanceof ForkJoinWorkerThread) |
1919 |
< |
((ForkJoinWorkerThread) t).pool.awaitBlocker(blocker); |
1918 |
> |
if (t instanceof ForkJoinWorkerThread) { |
1919 |
> |
ForkJoinWorkerThread w = (ForkJoinWorkerThread) t; |
1920 |
> |
w.pool.awaitBlocker(blocker); |
1921 |
> |
} |
1922 |
|
else { |
1923 |
|
do {} while (!blocker.isReleasable() && !blocker.block()); |
1924 |
|
} |
1946 |
|
private static final long eventCountOffset = |
1947 |
|
objectFieldOffset("eventCount", ForkJoinPool.class); |
1948 |
|
private static final long eventWaitersOffset = |
1949 |
< |
objectFieldOffset("eventWaiters",ForkJoinPool.class); |
1949 |
> |
objectFieldOffset("eventWaiters", ForkJoinPool.class); |
1950 |
|
private static final long stealCountOffset = |
1951 |
< |
objectFieldOffset("stealCount",ForkJoinPool.class); |
1951 |
> |
objectFieldOffset("stealCount", ForkJoinPool.class); |
1952 |
> |
private static final long spareWaitersOffset = |
1953 |
> |
objectFieldOffset("spareWaiters", ForkJoinPool.class); |
1954 |
|
|
1955 |
|
private static long objectFieldOffset(String field, Class<?> klazz) { |
1956 |
|
try { |