| 7 |
|
package java.util.concurrent; |
| 8 |
|
|
| 9 |
|
import java.lang.Thread.UncaughtExceptionHandler; |
| 10 |
– |
import java.lang.invoke.VarHandle; |
| 11 |
– |
import java.lang.reflect.Field; |
| 10 |
|
import java.security.AccessController; |
| 11 |
|
import java.security.AccessControlContext; |
| 12 |
|
import java.security.Permission; |
| 23 |
|
import java.util.concurrent.locks.ReentrantLock; |
| 24 |
|
import java.util.concurrent.locks.Condition; |
| 25 |
|
import jdk.internal.misc.Unsafe; |
| 26 |
+ |
//import jdk.internal.vm.SharedThreadContainer; // for loom |
| 27 |
|
|
| 28 |
|
/** |
| 29 |
|
* An {@link ExecutorService} for running {@link ForkJoinTask}s. |
| 173 |
|
* internal methods and nested classes are interrelated, their |
| 174 |
|
* main rationale and descriptions are presented here; individual |
| 175 |
|
* methods and nested classes contain only brief comments about |
| 176 |
< |
* details. |
| 176 |
> |
* details. There are a fair number of odd code constructions and |
| 177 |
> |
* design decisions for components that reside at the edge of Java |
| 178 |
> |
* vs JVM functionality. |
| 179 |
|
* |
| 180 |
|
* WorkQueues |
| 181 |
|
* ========== |
| 200 |
|
* a footprint as possible even in programs generating huge |
| 201 |
|
* numbers of tasks. To accomplish this, we shift the CAS |
| 202 |
|
* arbitrating pop vs poll (steal) from being on the indices |
| 203 |
< |
* ("base" and "top") to the slots themselves. |
| 203 |
> |
* ("base" and "top") to the slots themselves. These provide the |
| 204 |
> |
* primary required memory ordering -- see "Correct and Efficient |
| 205 |
> |
* Work-Stealing for Weak Memory Models" by Le, Pop, Cohen, and |
| 206 |
> |
* Nardelli, PPoPP 2013 |
| 207 |
> |
* (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an |
| 208 |
> |
* analysis of memory ordering requirements in work-stealing |
| 209 |
> |
* algorithms similar to the one used here. We also use ordered, |
| 210 |
> |
* moded accesses and/or fences for other control, with modes |
| 211 |
> |
* reflecting the presence or absence of other contextual sync |
| 212 |
> |
* provided by atomic and/or volatile accesses. Some methods (or |
| 213 |
> |
* their primary loops) begin with an acquire fence that amounts |
| 214 |
> |
* to an acquiring read of "this" to cover all fields (which is |
| 215 |
> |
* sometimes stronger than necessary, but less brittle). Some |
| 216 |
> |
* constructions are intentionally racy because they use read |
| 217 |
> |
* values as hints, not for correctness. |
| 218 |
> |
* |
| 219 |
> |
* We also support a user mode in which local task processing is |
| 220 |
> |
* in FIFO, not LIFO order, simply by using a local version of |
| 221 |
> |
* poll rather than pop. This can be useful in message-passing |
| 222 |
> |
* frameworks in which tasks are never joined, although with |
| 223 |
> |
* increased contention among task producers and consumers. Also, |
| 224 |
> |
* the same data structure (and class) is used for "submission |
| 225 |
> |
* queues" (described below) holding externally submitted tasks, |
| 226 |
> |
* that differ only in that a lock (field "access"; see below) is |
| 227 |
> |
* required by external callers to push and pop tasks. |
| 228 |
|
* |
| 229 |
|
* Adding tasks then takes the form of a classic array push(task) |
| 230 |
|
* in a circular buffer: |
| 233 |
|
* The actual code needs to null-check and size-check the array, |
| 234 |
|
* uses masking, not mod, for indexing a power-of-two-sized array, |
| 235 |
|
* enforces memory ordering, supports resizing, and possibly |
| 236 |
< |
* signals waiting workers to start scanning -- see below. |
| 236 |
> |
* signals waiting workers to start scanning (described below), |
| 237 |
> |
* which requires that even internal usages to strictly order |
| 238 |
> |
* accesses (using a form of lock release). |
| 239 |
|
* |
| 240 |
|
* The pop operation (always performed by owner) is of the form: |
| 241 |
|
* if ((task = getAndSet(q.array, (q.top-1) % length, null)) != null) |
| 242 |
|
* decrement top and return task; |
| 243 |
< |
* If this fails, the queue is empty. |
| 243 |
> |
* If this fails, the queue is empty. This operation is one part |
| 244 |
> |
* of the nextLocalTask method, that instead does a local-poll |
| 245 |
> |
* in FIFO mode. |
| 246 |
|
* |
| 247 |
< |
* The poll operation by another stealer thread is, basically: |
| 248 |
< |
* if (CAS nonnull task at q.array[q.base % length] to null) |
| 247 |
> |
* The poll operation is, basically: |
| 248 |
> |
* if (CAS nonnull task t = q.array[k = q.base % length] to null) |
| 249 |
|
* increment base and return task; |
| 250 |
|
* |
| 251 |
< |
* This may fail due to contention, and may be retried. |
| 252 |
< |
* Implementations must ensure a consistent snapshot of the base |
| 253 |
< |
* index and the task (by looping or trying elsewhere) before |
| 254 |
< |
* trying CAS. There isn't actually a method of this form, |
| 255 |
< |
* because failure due to inconsistency or contention is handled |
| 256 |
< |
* in different ways in different contexts, normally by first |
| 257 |
< |
* trying other queues. (For the most straightforward example, see |
| 258 |
< |
* method pollScan.) There are further variants for cases |
| 259 |
< |
* requiring inspection of elements before extracting them, so |
| 260 |
< |
* must interleave these with variants of this code. Also, a more |
| 261 |
< |
* efficient version (nextLocalTask) is used for polls by owners. |
| 262 |
< |
* It avoids some overhead because the queue cannot be growing |
| 263 |
< |
* during call. |
| 264 |
< |
* |
| 265 |
< |
* Memory ordering. See "Correct and Efficient Work-Stealing for |
| 266 |
< |
* Weak Memory Models" by Le, Pop, Cohen, and Nardelli, PPoPP 2013 |
| 267 |
< |
* (http://www.di.ens.fr/~zappa/readings/ppopp13.pdf) for an |
| 268 |
< |
* analysis of memory ordering requirements in work-stealing |
| 269 |
< |
* algorithms similar to the one used here. Inserting and |
| 270 |
< |
* extracting tasks in array slots via volatile or atomic accesses |
| 271 |
< |
* or explicit fences provides primary synchronization. |
| 272 |
< |
* |
| 273 |
< |
* Operations on deque elements require reads and writes of both |
| 274 |
< |
* indices and slots. When possible, we allow these to occur in |
| 275 |
< |
* any order. Because the base and top indices (along with other |
| 276 |
< |
* pool or array fields accessed in many methods) only imprecisely |
| 277 |
< |
* guide where to extract from, we let accesses other than the |
| 278 |
< |
* element getAndSet/CAS/setVolatile appear in any order, using |
| 279 |
< |
* plain mode. But we must still preface some methods (mainly |
| 280 |
< |
* those that may be accessed externally) with an acquireFence to |
| 281 |
< |
* avoid unbounded staleness. This is equivalent to acting as if |
| 282 |
< |
* callers use an acquiring read of the reference to the pool or |
| 283 |
< |
* queue when invoking the method, even when they do not. We use |
| 284 |
< |
* explicit acquiring reads (getSlot) rather than plain array |
| 285 |
< |
* access when acquire mode is required but not otherwise ensured |
| 286 |
< |
* by context. To reduce stalls by other stealers, we encourage |
| 287 |
< |
* timely writes to the base index by immediately following |
| 288 |
< |
* updates with a write of a volatile field that must be updated |
| 289 |
< |
* anyway, or an Opaque-mode write if there is no such |
| 290 |
< |
* opportunity. |
| 291 |
< |
* |
| 292 |
< |
* Because indices and slot contents cannot always be consistent, |
| 293 |
< |
* the emptiness check base == top is only quiescently accurate |
| 294 |
< |
* (and so used where this suffices). Otherwise, it may err on the |
| 295 |
< |
* side of possibly making the queue appear nonempty when a push, |
| 296 |
< |
* pop, or poll have not fully committed, or making it appear |
| 297 |
< |
* empty when an update of top or base has not yet been seen. |
| 298 |
< |
* Similarly, the check in push for the queue array being full may |
| 299 |
< |
* trigger when not completely full, causing a resize earlier than |
| 300 |
< |
* required. |
| 301 |
< |
* |
| 302 |
< |
* Mainly because of these potential inconsistencies among slots |
| 303 |
< |
* vs indices, the poll operation, considered individually, is not |
| 304 |
< |
* wait-free. One thief cannot successfully continue until another |
| 305 |
< |
* in-progress one (or, if previously empty, a push) visibly |
| 306 |
< |
* completes. This can stall threads when required to consume |
| 307 |
< |
* from a given queue (which may spin). However, in the |
| 308 |
< |
* aggregate, we ensure probabilistic non-blockingness at least |
| 309 |
< |
* until checking quiescence (which is intrinsically blocking): |
| 310 |
< |
* If an attempted steal fails, a scanning thief chooses a |
| 311 |
< |
* different victim target to try next. So, in order for one thief |
| 312 |
< |
* to progress, it suffices for any in-progress poll or new push |
| 313 |
< |
* on any empty queue to complete. The worst cases occur when many |
| 314 |
< |
* threads are looking for tasks being produced by a stalled |
| 315 |
< |
* producer. |
| 316 |
< |
* |
| 317 |
< |
* This approach also enables support of a user mode in which |
| 318 |
< |
* local task processing is in FIFO, not LIFO order, simply by |
| 319 |
< |
* using poll rather than pop. This can be useful in |
| 320 |
< |
* message-passing frameworks in which tasks are never joined, |
| 292 |
< |
* although with increased contention among task producers and |
| 293 |
< |
* consumers. |
| 251 |
> |
* However, there are several more cases that must be dealt with. |
| 252 |
> |
* Some of them are just due to asynchrony; others reflect |
| 253 |
> |
* contention and stealing policies. Stepping through them |
| 254 |
> |
* illustrates some of the implementation decisions in this class. |
| 255 |
> |
* |
| 256 |
> |
* * Slot k must be read with an acquiring read, which it must |
| 257 |
> |
* anyway to dereference and run the task if the (acquiring) |
| 258 |
> |
* CAS succeeds, but uses an explicit acquire fence to support |
| 259 |
> |
* the following rechecks even if the CAS is not attempted. |
| 260 |
> |
* |
| 261 |
> |
* * q.base may change between reading and using its value to |
| 262 |
> |
* index the slot. To avoid trying to use the wrong t, the |
| 263 |
> |
* index and slot must be reread (not necessarily immediately) |
| 264 |
> |
* until consistent, unless this is a local poll by owner, in |
| 265 |
> |
* which case this form of inconsistency can only appear as t |
| 266 |
> |
* being null, below. |
| 267 |
> |
* |
| 268 |
> |
* * Similarly, q.array may change (due to a resize), unless this |
| 269 |
> |
* is a local poll by owner. Otherwise, when t is present, this |
| 270 |
> |
* only needs consideration on CAS failure (since a CAS |
| 271 |
> |
* confirms the non-resized case.) |
| 272 |
> |
* |
| 273 |
> |
* * t may appear null because a previous poll operation has not |
| 274 |
> |
* yet incremented q.base, so the read is from an already-taken |
| 275 |
> |
* index. This form of stall reflects the non-lock-freedom of |
| 276 |
> |
* the poll operation. Stalls can be detected by observing that |
| 277 |
> |
* q.base doesn't change on repeated reads of null t and when |
| 278 |
> |
* no other alternatives apply, spin-wait for it to settle. To |
| 279 |
> |
* reduce producing these kinds of stalls by other stealers, we |
| 280 |
> |
* encourage timely writes to indices using store fences when |
| 281 |
> |
* memory ordering is not already constrained by context. |
| 282 |
> |
* |
| 283 |
> |
* * The CAS may fail, in which case we may want to retry unless |
| 284 |
> |
* there is too much contention. One goal is to balance and |
| 285 |
> |
* spread out the many forms of contention that may be |
| 286 |
> |
* encountered across polling and other operations to avoid |
| 287 |
> |
* sustained performance degradations. Across all cases where |
| 288 |
> |
* alternatives exist, a bounded number of CAS misses or stalls |
| 289 |
> |
* are tolerated (for slots, ctl, and elsewhere described |
| 290 |
> |
* below) before taking alternative action. These may move |
| 291 |
> |
* contention or retries elsewhere, which is still preferable |
| 292 |
> |
* to single-point bottlenecks. |
| 293 |
> |
* |
| 294 |
> |
* * Even though the check "top == base" is quiescently accurate |
| 295 |
> |
* to determine whether a queue is empty, it is not of much use |
| 296 |
> |
* when deciding whether to try to poll or repoll after a |
| 297 |
> |
* failure. Both top and base may move independently, and both |
| 298 |
> |
* lag updates to the underlying array. To reduce memory |
| 299 |
> |
* contention, when possible, non-owners avoid reading the |
| 300 |
> |
* "top" index at all, and instead use array reads, including |
| 301 |
> |
* one-ahead reads to check whether to repoll, relying on the |
| 302 |
> |
* fact that an non-empty queue does not have two null slots in |
| 303 |
> |
* a row, except in cases (resizes and shifts) that can be |
| 304 |
> |
* detected with a secondary recheck. |
| 305 |
> |
* |
| 306 |
> |
* The poll operations in q.poll(), scan(), helpJoin(), and |
| 307 |
> |
* elsewhere differ with respect to whether other queues are |
| 308 |
> |
* available to try, and the presence or nature of screening steps |
| 309 |
> |
* when only some kinds of tasks can be taken. When alternatives |
| 310 |
> |
* (or failing) is an option, they uniformly give up after |
| 311 |
> |
* boundeed numbers of stalls and/or CAS failures, which reduces |
| 312 |
> |
* contention when too many workers are polling too few tasks. |
| 313 |
> |
* Overall, in the aggregate, we ensure probabilistic |
| 314 |
> |
* non-blockingness of work-stealing at least until checking |
| 315 |
> |
* quiescence (which is intrinsically blocking): If an attempted |
| 316 |
> |
* steal fails in these ways, a scanning thief chooses a different |
| 317 |
> |
* target to try next. In contexts where alternatives aren't |
| 318 |
> |
* available, and when progress conditions can be isolated to |
| 319 |
> |
* values of a single variable, simple spinloops (using |
| 320 |
> |
* Thread.onSpinWait) are used to reduce memory traffic. |
| 321 |
|
* |
| 322 |
|
* WorkQueues are also used in a similar way for tasks submitted |
| 323 |
|
* to the pool. We cannot mix these tasks in the same queues used |
| 329 |
|
* like workers except that they are restricted to executing local |
| 330 |
|
* tasks that they submitted (or when known, subtasks thereof). |
| 331 |
|
* Insertion of tasks in shared mode requires a lock. We use only |
| 332 |
< |
* a simple spinlock (using field "source"), because submitters |
| 333 |
< |
* encountering a busy queue move to a different position to use |
| 334 |
< |
* or create other queues. They block only when registering new |
| 335 |
< |
* queues. |
| 332 |
> |
* a simple spinlock because submitters encountering a busy queue |
| 333 |
> |
* move to a different position to use or create other queues. |
| 334 |
> |
* They (spin) block only when registering new queues, and less |
| 335 |
> |
* often in tryRemove and helpComplete. The lock needed for |
| 336 |
> |
* external queues is generalized (as field "access") for |
| 337 |
> |
* operations on owned queues that require a fully-fenced write |
| 338 |
> |
* (including push, parking status, and termination) in order to |
| 339 |
> |
* deal with Dekker-like signalling constructs described below. |
| 340 |
|
* |
| 341 |
|
* Management |
| 342 |
|
* ========== |
| 352 |
|
* can globally track or maintain, so we pack them into a small |
| 353 |
|
* number of variables, often maintaining atomicity without |
| 354 |
|
* blocking or locking. Nearly all essentially atomic control |
| 355 |
< |
* state is held in a few volatile variables that are by far most |
| 356 |
< |
* often read (not written) as status and consistency checks. We |
| 357 |
< |
* pack as much information into them as we can. |
| 355 |
> |
* state is held in a few variables that are by far most often |
| 356 |
> |
* read (not written) as status and consistency checks. We pack as |
| 357 |
> |
* much information into them as we can. |
| 358 |
|
* |
| 359 |
|
* Field "ctl" contains 64 bits holding information needed to |
| 360 |
|
* atomically decide to add, enqueue (on an event queue), and |
| 362 |
|
* restrict maximum parallelism to (1<<15)-1 (which is far in |
| 363 |
|
* excess of normal operating range) to allow ids, counts, and |
| 364 |
|
* their negations (used for thresholding) to fit into 16bit |
| 365 |
< |
* subfields. |
| 366 |
< |
* |
| 367 |
< |
* Field "mode" holds configuration parameters as well as lifetime |
| 368 |
< |
* status, atomically and monotonically setting SHUTDOWN, STOP, |
| 369 |
< |
* and finally TERMINATED bits. It is updated only via bitwise |
| 370 |
< |
* atomics (getAndBitwiseOr). |
| 365 |
> |
* subfields. Field "parallelism" holds the target parallelism |
| 366 |
> |
* (normally corresponding to pool size). It is needed (nearly) |
| 367 |
> |
* only in methods updating ctl, so is packed nearby. As of the |
| 368 |
> |
* current release, users can dynamically reset target |
| 369 |
> |
* parallelism, which is read once per update, so only slowly has |
| 370 |
> |
* an effect in creating threads or letting them time out and |
| 371 |
> |
* terminate when idle. |
| 372 |
> |
* |
| 373 |
> |
* Field "mode" mainly holds lifetime status, atomically and |
| 374 |
> |
* monotonically setting SHUTDOWN, STOP, and finally TERMINATED |
| 375 |
> |
* bits. It is updated only via bitwise atomics (getAndBitwiseOr). |
| 376 |
|
* |
| 377 |
|
* Array "queues" holds references to WorkQueues. It is updated |
| 378 |
|
* (only during worker creation and termination) under the |
| 379 |
< |
* registrationLock, but is otherwise concurrently readable, and |
| 380 |
< |
* accessed directly (although always prefaced by acquireFences or |
| 381 |
< |
* other acquiring reads). To simplify index-based operations, the |
| 382 |
< |
* array size is always a power of two, and all readers must |
| 383 |
< |
* tolerate null slots. Worker queues are at odd indices. Worker |
| 384 |
< |
* ids masked with SMASK match their index. Shared (submission) |
| 385 |
< |
* queues are at even indices. Grouping them together in this way |
| 386 |
< |
* simplifies and speeds up task scanning. |
| 379 |
> |
* registrationLock, but is otherwise concurrently readable (often |
| 380 |
> |
* prefaced by a volatile read of mode to check termination, that |
| 381 |
> |
* is required anyway, and serves as an acquire fence). To |
| 382 |
> |
* simplify index-based operations, the array size is always a |
| 383 |
> |
* power of two, and all readers must tolerate null slots. Worker |
| 384 |
> |
* queues are at odd indices. Worker ids masked with SMASK match |
| 385 |
> |
* their index. Shared (submission) queues are at even |
| 386 |
> |
* indices. Grouping them together in this way simplifies and |
| 387 |
> |
* speeds up task scanning. |
| 388 |
|
* |
| 389 |
|
* All worker thread creation is on-demand, triggered by task |
| 390 |
|
* submissions, replacement of terminated workers, and/or |
| 391 |
|
* compensation for blocked workers. However, all other support |
| 392 |
|
* code is set up to work with other policies. To ensure that we |
| 393 |
|
* do not hold on to worker or task references that would prevent |
| 394 |
< |
* GC, all accesses to workQueues are via indices into the |
| 395 |
< |
* queues array (which is one source of some of the messy code |
| 396 |
< |
* constructions here). In essence, the queues array serves as |
| 397 |
< |
* a weak reference mechanism. Thus for example the stack top |
| 398 |
< |
* subfield of ctl stores indices, not references. |
| 394 |
> |
* GC, all accesses to workQueues in waiting, signalling, and |
| 395 |
> |
* control methods are via indices into the queues array (which is |
| 396 |
> |
* one source of some of the messy code constructions here). In |
| 397 |
> |
* essence, the queues array serves as a weak reference |
| 398 |
> |
* mechanism. In particular, the stack top subfield of ctl stores |
| 399 |
> |
* indices, not references. |
| 400 |
|
* |
| 401 |
|
* Queuing Idle Workers. Unlike HPC work-stealing frameworks, we |
| 402 |
|
* cannot let workers spin indefinitely scanning for tasks when |
| 441 |
|
* to some external caller. |
| 442 |
|
* |
| 443 |
|
* WorkQueue field "phase" is used by both workers and the pool to |
| 444 |
< |
* manage and track whether a worker is UNSIGNALLED (possibly |
| 445 |
< |
* blocked waiting for a signal). When a worker is enqueued its |
| 446 |
< |
* phase field is set negative. Note that phase field updates lag |
| 447 |
< |
* queue CAS releases; seeing a negative phase does not guarantee |
| 448 |
< |
* that the worker is available. When queued, the lower 16 bits of |
| 449 |
< |
* its phase must hold its pool index. So we place the index there |
| 450 |
< |
* upon initialization and never modify these bits. |
| 444 |
> |
* manage and track whether a worker is unsignalled (possibly |
| 445 |
> |
* blocked waiting for a signal), convienently using the sign bit |
| 446 |
> |
* to check. When a worker is enqueued its phase field is set |
| 447 |
> |
* negative. Note that phase field updates lag queue CAS releases; |
| 448 |
> |
* seeing a negative phase does not guarantee that the worker is |
| 449 |
> |
* available (and so is never checked in this way). When queued, |
| 450 |
> |
* the lower 16 bits of its phase must hold its pool index. So we |
| 451 |
> |
* place the index there upon initialization and never modify |
| 452 |
> |
* these bits. |
| 453 |
|
* |
| 454 |
|
* The ctl field also serves as the basis for memory |
| 455 |
|
* synchronization surrounding activation. This uses a more |
| 458 |
|
* if to its current value). However, rather than CASing ctl to |
| 459 |
|
* its current value in the common case where no action is |
| 460 |
|
* required, we reduce write contention by ensuring that |
| 461 |
< |
* signalWork invocations are prefaced with a full-volatile memory |
| 461 |
> |
* signalWork invocations are prefaced with a fully fenced memory |
| 462 |
|
* access (which is usually needed anyway). |
| 463 |
|
* |
| 464 |
|
* Signalling. Signals (in signalWork) cause new or reactivated |
| 469 |
|
* computations are purely tree structured, it suffices for every |
| 470 |
|
* worker to activate another when it pushes a task into an empty |
| 471 |
|
* queue, resulting in O(log(#threads)) steps to full activation. |
| 472 |
< |
* If instead, tasks come in serially from only a single producer, |
| 473 |
< |
* each worker taking its first (since the last quiescence) task |
| 472 |
> |
* (To reduce resource usages in some cases, at the expense of |
| 473 |
> |
* slower startup in others, activation of an idle thread is |
| 474 |
> |
* preferred over creating a new one, here and elsewhere.) If |
| 475 |
> |
* instead, tasks come in serially from only a single producer, |
| 476 |
> |
* each worker taking its first (since the last activation) task |
| 477 |
|
* from a queue should signal another if there are more tasks in |
| 478 |
|
* that queue. This is equivalent to, but generally faster than, |
| 479 |
|
* arranging the stealer take two tasks, re-pushing one on its own |
| 480 |
|
* queue, and signalling (because its queue is empty), also |
| 481 |
|
* resulting in logarithmic full activation time. Because we don't |
| 482 |
|
* know about usage patterns (or most commonly, mixtures), we use |
| 483 |
< |
* both approaches. We approximate the second rule by arranging |
| 484 |
< |
* that workers in scan() do not repeat signals when repeatedly |
| 485 |
< |
* taking tasks from any given queue, by remembering the previous |
| 486 |
< |
* one. There are narrow windows in which both rules may apply, |
| 487 |
< |
* leading to duplicate or unnecessary signals. Despite such |
| 488 |
< |
* limitations, these rules usually avoid slowdowns that otherwise |
| 489 |
< |
* occur when too many workers contend to take too few tasks, or |
| 490 |
< |
* when producers waste most of their time resignalling. However, |
| 491 |
< |
* contention and overhead effects may still occur during ramp-up, |
| 449 |
< |
* ramp-down, and small computations involving only a few workers. |
| 483 |
> |
* both approaches. Together these are minimally necessary for |
| 484 |
> |
* maintaining liveness. However, they do not account for the fact |
| 485 |
> |
* that when tasks are short-lived, signals are unnecessary |
| 486 |
> |
* because workers will already be scanning for new tasks without |
| 487 |
> |
* the need of new signals. We track these cases (variable |
| 488 |
> |
* "prevSrc" in scan() and related methods) to avoid some |
| 489 |
> |
* unnecessary signals and scans. However, signal contention and |
| 490 |
> |
* overhead effects may still occur during ramp-up, ramp-down, and |
| 491 |
> |
* small computations involving only a few workers. |
| 492 |
|
* |
| 493 |
|
* Scanning. Method scan performs top-level scanning for (and |
| 494 |
< |
* execution of) tasks. Scans by different workers and/or at |
| 495 |
< |
* different times are unlikely to poll queues in the same |
| 496 |
< |
* order. Each scan traverses and tries to poll from each queue in |
| 497 |
< |
* a pseudorandom permutation order by starting at a random index, |
| 498 |
< |
* and using a constant cyclically exhaustive stride; restarting |
| 499 |
< |
* upon contention. (Non-top-level scans; for example in |
| 500 |
< |
* helpJoin, use simpler linear probes because they do not |
| 501 |
< |
* systematically contend with top-level scans.) The pseudorandom |
| 502 |
< |
* generator need not have high-quality statistical properties in |
| 503 |
< |
* the long term. We use Marsaglia XorShifts, seeded with the Weyl |
| 504 |
< |
* sequence from ThreadLocalRandom probes, which are cheap and |
| 505 |
< |
* suffice. Scans do not otherwise explicitly take into account |
| 506 |
< |
* core affinities, loads, cache localities, etc, However, they do |
| 507 |
< |
* exploit temporal locality (which usually approximates these) by |
| 508 |
< |
* preferring to re-poll from the same queue after a successful |
| 509 |
< |
* poll before trying others (see method topLevelExec). This |
| 510 |
< |
* reduces fairness, which is partially counteracted by using a |
| 511 |
< |
* one-shot form of poll (tryPoll) that may lose to other workers. |
| 512 |
< |
* |
| 513 |
< |
* Deactivation. Method scan returns a sentinel when no tasks are |
| 514 |
< |
* found, leading to deactivation (see awaitWork). The count |
| 515 |
< |
* fields in ctl allow accurate discovery of quiescent states |
| 516 |
< |
* (i.e., when all workers are idle) after deactivation. However, |
| 517 |
< |
* this may also race with new (external) submissions, so a |
| 518 |
< |
* recheck is also needed to determine quiescence. Upon apparently |
| 519 |
< |
* triggering quiescence, awaitWork re-scans and self-signals if |
| 520 |
< |
* it may have missed a signal. In other cases, a missed signal |
| 521 |
< |
* may transiently lower parallelism because deactivation does not |
| 522 |
< |
* necessarily mean that there is no more work, only that that |
| 523 |
< |
* there were no tasks not taken by other workers. But more |
| 524 |
< |
* signals are generated (see above) to eventually reactivate if |
| 525 |
< |
* needed. |
| 526 |
< |
* |
| 527 |
< |
* Trimming workers. To release resources after periods of lack of |
| 528 |
< |
* use, a worker starting to wait when the pool is quiescent will |
| 529 |
< |
* time out and terminate if the pool has remained quiescent for |
| 530 |
< |
* period given by field keepAlive. |
| 494 |
> |
* execution of) tasks by polling a pseodo-random permutation of |
| 495 |
> |
* the array (by starting at a random index, and using a constant |
| 496 |
> |
* cyclically exhaustive stride.) It uses the same basic polling |
| 497 |
> |
* method as WorkQueue.poll(), but restarts with a different |
| 498 |
> |
* permutation on each invocation. (Non-top-level scans; for |
| 499 |
> |
* example in helpJoin, use simpler and faster linear probes |
| 500 |
> |
* because they do not systematically contend with top-level |
| 501 |
> |
* scans.) The pseudorandom generator need not have high-quality |
| 502 |
> |
* statistical properties in the long term. We use Marsaglia |
| 503 |
> |
* XorShifts, seeded with the Weyl sequence from ThreadLocalRandom |
| 504 |
> |
* probes, which are cheap and suffice. Scans do not otherwise |
| 505 |
> |
* explicitly take into account core affinities, loads, cache |
| 506 |
> |
* localities, etc, However, they do exploit temporal locality |
| 507 |
> |
* (which usually approximates these) by preferring to re-poll |
| 508 |
> |
* from the same queue (using method tryPoll()) after a successful |
| 509 |
> |
* poll before trying others (see method topLevelExec), which also |
| 510 |
> |
* reduces bookkeeping and scanning overhead. This also reduces |
| 511 |
> |
* fairness, which is partially counteracted by giving up on |
| 512 |
> |
* contention. |
| 513 |
> |
* |
| 514 |
> |
* Deactivation. When method scan indicates that no tasks are |
| 515 |
> |
* found by a worker, it deactivates (see awaitWork). Note that |
| 516 |
> |
* not finding tasks doesn't mean that there won't soon be |
| 517 |
> |
* some. Further, a scan may give up under contention, returning |
| 518 |
> |
* even without knowing whether any tasks are still present, which |
| 519 |
> |
* is OK, given the above signalling rules that will eventually |
| 520 |
> |
* maintain progress. Blocking and unblocking via park/unpark can |
| 521 |
> |
* cause serious slowdowns when tasks are rapidly but irregularly |
| 522 |
> |
* generated (which is often due to garbage collectors and other |
| 523 |
> |
* activities). One way to ameliorate is for workers to rescan |
| 524 |
> |
* multiple times, even when there are unlikely to be tasks. But |
| 525 |
> |
* this causes enough memory and CAS contention to prefer using |
| 526 |
> |
* quieter spinwaits in awaitWork; currently set to small values |
| 527 |
> |
* that only cover near-miss scenarios for deactivate vs activate |
| 528 |
> |
* races. Because idle workers are often not yet blocked (via |
| 529 |
> |
* LockSupport.park), we use the WorkQueue access field to |
| 530 |
> |
* advertise that a waiter actually needs unparking upon signal. |
| 531 |
> |
* |
| 532 |
> |
* When idle workers are not continually woken up, the count |
| 533 |
> |
* fields in ctl allow efficient and accurate discovery of |
| 534 |
> |
* quiescent states (i.e., when all workers are idle) after |
| 535 |
> |
* deactivation. However, this voting mechanism alone does not |
| 536 |
> |
* guarantee that a pool can become dormant (quiesced or |
| 537 |
> |
* terminated), because external racing producers do not vote, and |
| 538 |
> |
* can asynchronously submit new tasks. To deal with this, the |
| 539 |
> |
* final unparked thread (in awaitWork) scans external queues to |
| 540 |
> |
* check for tasks that could have been added during a race window |
| 541 |
> |
* that would not be accompanied by a signal, in which case |
| 542 |
> |
* re-activating itself (or any other worker) to recheck. The same |
| 543 |
> |
* sets of checks are used in tryTerminate, to correctly trigger |
| 544 |
> |
* delayed termination (shutDown, followed by quiescence) in the |
| 545 |
> |
* presence of racing submissions. In all cases, the notion of the |
| 546 |
> |
* "final" unparked thread is an approximation, because new |
| 547 |
> |
* workers could be in the process of being constructed, which |
| 548 |
> |
* occasionally adds some extra unnecessary processing. |
| 549 |
|
* |
| 550 |
|
* Shutdown and Termination. A call to shutdownNow invokes |
| 551 |
|
* tryTerminate to atomically set a mode bit. The calling thread, |
| 552 |
|
* as well as every other worker thereafter terminating, helps |
| 553 |
|
* terminate others by cancelling their unprocessed tasks, and |
| 554 |
< |
* waking them up. Calls to non-abrupt shutdown() preface this by |
| 555 |
< |
* checking isQuiescent before triggering the "STOP" phase of |
| 556 |
< |
* termination. To conform to ExecutorService invoke, invokeAll, |
| 557 |
< |
* and invokeAny specs, we must track pool status while waiting, |
| 558 |
< |
* and interrupt interruptible callers on termination (see |
| 559 |
< |
* ForkJoinTask.joinForPoolInvoke etc). |
| 554 |
> |
* interrupting other workers. Calls to non-abrupt shutdown() |
| 555 |
> |
* preface this by checking isQuiescent before triggering the |
| 556 |
> |
* "STOP" phase of termination. During termination, workers are |
| 557 |
> |
* stopped using all three of (often in parallel): releasing via |
| 558 |
> |
* ctl (method reactivate), interrupts, and cancelling tasks that |
| 559 |
> |
* will cause workers to not find work and exit. To support this, |
| 560 |
> |
* worker references not removed from the queues array during |
| 561 |
> |
* termination. It is possible for late thread creations to still |
| 562 |
> |
* be in progress after a quiescent termination reports terminated |
| 563 |
> |
* status, but they will also immediately terminate. To conform to |
| 564 |
> |
* ExecutorService invoke, invokeAll, and invokeAny specs, we must |
| 565 |
> |
* track pool status while waiting in ForkJoinTask.awaitDone, and |
| 566 |
> |
* interrupt interruptible callers on termination. |
| 567 |
> |
* |
| 568 |
> |
* Trimming workers. To release resources after periods of lack of |
| 569 |
> |
* use, a worker starting to wait when the pool is quiescent will |
| 570 |
> |
* time out and terminate if the pool has remained quiescent for |
| 571 |
> |
* period given by field keepAlive. |
| 572 |
|
* |
| 573 |
|
* Joining Tasks |
| 574 |
|
* ============= |
| 575 |
|
* |
| 576 |
|
* Normally, the first option when joining a task that is not done |
| 577 |
< |
* is to try to unfork it from local queue and run it. Otherwise, |
| 577 |
> |
* is to try to take it from local queue and run it. Otherwise, |
| 578 |
|
* any of several actions may be taken when one worker is waiting |
| 579 |
|
* to join a task stolen (or always held) by another. Because we |
| 580 |
|
* are multiplexing many tasks on to a pool of workers, we can't |
| 597 |
|
* possible action of a compensator is to steal and execute the |
| 598 |
|
* task being joined, the joining thread can do so directly, |
| 599 |
|
* without the need for a compensation thread; although with a |
| 600 |
< |
* (rare) possibility of reduced parallelism because of a |
| 601 |
< |
* transient gap in the queue array. |
| 600 |
> |
* possibility of reduced parallelism because of a transient gap |
| 601 |
> |
* in the queue array that stalls stealers. |
| 602 |
|
* |
| 603 |
|
* Other intermediate forms available for specific task types (for |
| 604 |
|
* example helpAsyncBlocker) often avoid or postpone the need for |
| 608 |
|
* only on compensation in method awaitBlocker. |
| 609 |
|
* |
| 610 |
|
* The algorithm in helpJoin entails a form of "linear helping". |
| 611 |
< |
* Each worker records (in field "source") the id of the queue |
| 612 |
< |
* from which it last stole a task. The scan in method helpJoin |
| 613 |
< |
* uses these markers to try to find a worker to help (i.e., steal |
| 614 |
< |
* back a task from and execute it) that could hasten completion |
| 615 |
< |
* of the actively joined task. Thus, the joiner executes a task |
| 616 |
< |
* that would be on its own local deque if the to-be-joined task |
| 617 |
< |
* had not been stolen. This is a conservative variant of the |
| 618 |
< |
* approach described in Wagner & Calder "Leapfrogging: a portable |
| 619 |
< |
* technique for implementing efficient futures" SIGPLAN Notices, |
| 620 |
< |
* 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs |
| 621 |
< |
* mainly in that we only record queue ids, not full dependency |
| 611 |
> |
* Each worker records (in field "source") a reference to the |
| 612 |
> |
* queue from which it last stole a task. The scan in method |
| 613 |
> |
* helpJoin uses these markers to try to find a worker to help |
| 614 |
> |
* (i.e., steal back a task from and execute it) that could hasten |
| 615 |
> |
* completion of the actively joined task. Thus, the joiner |
| 616 |
> |
* executes a task that would be on its own local deque if the |
| 617 |
> |
* to-be-joined task had not been stolen. This is a conservative |
| 618 |
> |
* variant of the approach described in Wagner & Calder |
| 619 |
> |
* "Leapfrogging: a portable technique for implementing efficient |
| 620 |
> |
* futures" SIGPLAN Notices, 1993 |
| 621 |
> |
* (http://portal.acm.org/citation.cfm?id=155354). It differs |
| 622 |
> |
* mainly in that we only record queues, not full dependency |
| 623 |
|
* links. This requires a linear scan of the queues array to |
| 624 |
|
* locate stealers, but isolates cost to when it is needed, rather |
| 625 |
< |
* than adding to per-task overhead. Also, searches are limited to |
| 626 |
< |
* direct and at most two levels of indirect stealers, after which |
| 627 |
< |
* there are rapidly diminishing returns on increased overhead. |
| 628 |
< |
* Searches can fail to locate stealers when stalls delay |
| 629 |
< |
* recording sources. Further, even when accurately identified, |
| 625 |
> |
* than adding to per-task overhead. For CountedCompleters, the |
| 626 |
> |
* analogous method helpComplete doesn't need stealer-tracking, |
| 627 |
> |
* but requires a similar check of completion chains. |
| 628 |
> |
* |
| 629 |
> |
* In either case, searches can fail to locate stealers when |
| 630 |
> |
* stalls delay recording sources. We avoid some of these cases by |
| 631 |
> |
* using snapshotted values of ctl as a check that the numbers of |
| 632 |
> |
* workers are not changing. But even when accurately identified, |
| 633 |
|
* stealers might not ever produce a task that the joiner can in |
| 634 |
|
* turn help with. So, compensation is tried upon failure to find |
| 635 |
|
* tasks to run. |
| 636 |
|
* |
| 561 |
– |
* Joining CountedCompleters (see helpComplete) differs from (and |
| 562 |
– |
* is generally more efficient than) other cases because task |
| 563 |
– |
* eligibility is determined by checking completion chains rather |
| 564 |
– |
* than tracking stealers. |
| 565 |
– |
* |
| 566 |
– |
* Joining under timeouts (ForkJoinTask timed get) uses a |
| 567 |
– |
* constrained mixture of helping and compensating in part because |
| 568 |
– |
* pools (actually, only the common pool) may not have any |
| 569 |
– |
* available threads: If the pool is saturated (all available |
| 570 |
– |
* workers are busy), the caller tries to remove and otherwise |
| 571 |
– |
* help; else it blocks under compensation so that it may time out |
| 572 |
– |
* independently of any tasks. |
| 573 |
– |
* |
| 637 |
|
* Compensation does not by default aim to keep exactly the target |
| 638 |
|
* parallelism number of unblocked threads running at any given |
| 639 |
|
* time. Some previous versions of this class employed immediate |
| 653 |
|
* The static common pool always exists after static |
| 654 |
|
* initialization. Since it (or any other created pool) need |
| 655 |
|
* never be used, we minimize initial construction overhead and |
| 656 |
< |
* footprint to the setup of about a dozen fields. |
| 657 |
< |
* |
| 658 |
< |
* When external threads submit to the common pool, they can |
| 659 |
< |
* perform subtask processing (see helpComplete and related |
| 660 |
< |
* methods) upon joins. This caller-helps policy makes it |
| 656 |
> |
* footprint to the setup of about a dozen fields, although with |
| 657 |
> |
* some System property parsing and with security processing that |
| 658 |
> |
* takes far longer than the actual construction when |
| 659 |
> |
* SecurityManagers are used or properties are set. The common |
| 660 |
> |
* pool is distinguished internally by having both a null |
| 661 |
> |
* workerNamePrefix and ISCOMMON config bit set, along with |
| 662 |
> |
* PRESET_SIZE set if parallelism was configured by system |
| 663 |
> |
* property. |
| 664 |
> |
* |
| 665 |
> |
* When external threads use ForkJoinTask.fork for the common |
| 666 |
> |
* pool, they can perform subtask processing (see helpComplete and |
| 667 |
> |
* related methods) upon joins. This caller-helps policy makes it |
| 668 |
|
* sensible to set common pool parallelism level to one (or more) |
| 669 |
|
* less than the total number of available cores, or even zero for |
| 670 |
< |
* pure caller-runs. We do not need to record whether external |
| 671 |
< |
* submissions are to the common pool -- if not, external help |
| 672 |
< |
* methods return quickly. These submitters would otherwise be |
| 673 |
< |
* blocked waiting for completion, so the extra effort (with |
| 674 |
< |
* liberally sprinkled task status checks) in inapplicable cases |
| 675 |
< |
* amounts to an odd form of limited spin-wait before blocking in |
| 606 |
< |
* ForkJoinTask.join. |
| 670 |
> |
* pure caller-runs. For the sake of ExecutorService specs, we can |
| 671 |
> |
* only do this for tasks entered via fork, not submit. We track |
| 672 |
> |
* this using a task status bit (markPoolSubmission). In all |
| 673 |
> |
* other cases, external threads waiting for joins first check the |
| 674 |
> |
* common pool for their task, which fails quickly if the caller |
| 675 |
> |
* did not fork to common pool. |
| 676 |
|
* |
| 677 |
|
* Guarantees for common pool parallelism zero are limited to |
| 678 |
|
* tasks that are joined by their callers in a tree-structured |
| 679 |
|
* fashion or use CountedCompleters (as is true for jdk |
| 680 |
|
* parallelStreams). Support infiltrates several methods, |
| 681 |
< |
* including those that retry helping steps until we are sure that |
| 682 |
< |
* none apply if there are no workers. |
| 681 |
> |
* including those that retry helping steps or spin until we are |
| 682 |
> |
* sure that none apply if there are no workers. |
| 683 |
|
* |
| 684 |
|
* As a more appropriate default in managed environments, unless |
| 685 |
|
* overridden by system properties, we use workers of subclass |
| 701 |
|
* should be cancelled anyway. Interrupts are cleared only when |
| 702 |
|
* necessary to ensure that calls to LockSupport.park do not loop |
| 703 |
|
* indefinitely (park returns immediately if the current thread is |
| 704 |
< |
* interrupted). If so, interruption is reinstated after blocking |
| 705 |
< |
* if status could be visible during the scope of any task. For |
| 706 |
< |
* cases in which task bodies are specified or desired to |
| 638 |
< |
* interrupt upon cancellation, ForkJoinTask.cancel can be |
| 639 |
< |
* overridden to do so (as is done for invoke{Any,All}). |
| 704 |
> |
* interrupted). For cases in which task bodies are specified or |
| 705 |
> |
* desired to interrupt upon cancellation, ForkJoinTask.cancel can |
| 706 |
> |
* be overridden to do so (as is done for invoke{Any,All}). |
| 707 |
|
* |
| 708 |
|
* Memory placement |
| 709 |
|
* ================ |
| 710 |
|
* |
| 711 |
< |
* Performance can be very sensitive to placement of instances of |
| 712 |
< |
* ForkJoinPool and WorkQueues and their queue arrays. To reduce |
| 713 |
< |
* false-sharing impact, the @Contended annotation isolates the |
| 714 |
< |
* ForkJoinPool.ctl field as well as the most heavily written |
| 715 |
< |
* WorkQueue fields. These mainly reduce cache traffic by scanners. |
| 716 |
< |
* WorkQueue arrays are presized large enough to avoid resizing |
| 717 |
< |
* (which transiently reduces throughput) in most tree-like |
| 718 |
< |
* computations, although not in some streaming usages. Initial |
| 719 |
< |
* sizes are not large enough to avoid secondary contention |
| 720 |
< |
* effects (especially for GC cardmarks) when queues are placed |
| 721 |
< |
* near each other in memory. This is common, but has different |
| 722 |
< |
* impact in different collectors and remains incompletely |
| 723 |
< |
* addressed. |
| 711 |
> |
* Performance is very sensitive to placement of instances of |
| 712 |
> |
* ForkJoinPool and WorkQueues and their queue arrays, as well the |
| 713 |
> |
* placement of their fields. Caches misses and contention due to |
| 714 |
> |
* false-sharing have been observed to slow down some programs by |
| 715 |
> |
* more than a factor of four. There is no perfect solution, in |
| 716 |
> |
* part because isolating more fields also generates more cache |
| 717 |
> |
* misses in more common cases (because some fields snd slots are |
| 718 |
> |
* usually read at the same time), and the main means of placing |
| 719 |
> |
* memory, the @Contended annotation provides only rough control |
| 720 |
> |
* (for good reason). We isolate the ForkJoinPool.ctl field as |
| 721 |
> |
* well the set of WorkQueue fields that otherwise cause the most |
| 722 |
> |
* false-sharing misses with respect to other fields. Also, |
| 723 |
> |
* ForkJoinPool fields are ordered such that fields less prone to |
| 724 |
> |
* contention effects are first, offsetting those that otherwise |
| 725 |
> |
* would be, while also reducing total footprint vs using |
| 726 |
> |
* multiple @Contended regions, which tends to slow down |
| 727 |
> |
* less-contended applications. These arrangements mainly reduce |
| 728 |
> |
* cache traffic by scanners, which speeds up finding tasks to |
| 729 |
> |
* run. Initial sizing and resizing of WorkQueue arrays is an |
| 730 |
> |
* even more delicate tradeoff because the best strategy may vary |
| 731 |
> |
* across garbage collectors. Small arrays are better for locality |
| 732 |
> |
* and reduce GC scan time, but large arrays reduce both direct |
| 733 |
> |
* false-sharing and indirect cases due to GC bookkeeping |
| 734 |
> |
* (cardmarks etc), and reduce the number of resizes, which are |
| 735 |
> |
* not especially fast because they require atomic transfers, and |
| 736 |
> |
* may cause other scanning workers to stall or give up. |
| 737 |
> |
* Currently, arrays are initialized to be fairly small but early |
| 738 |
> |
* resizes rapidly increase size by more than a factor of two |
| 739 |
> |
* until very large. (Maintenance note: any changes in fields, |
| 740 |
> |
* queues, or their uses must be accompanied by re-evaluation of |
| 741 |
> |
* these placement and sizing decisions.) |
| 742 |
|
* |
| 743 |
|
* Style notes |
| 744 |
|
* =========== |
| 745 |
|
* |
| 746 |
|
* Memory ordering relies mainly on atomic operations (CAS, |
| 747 |
< |
* getAndSet, getAndAdd) along with explicit fences. These use |
| 747 |
> |
* getAndSet, getAndAdd) along with moded accesses. These use |
| 748 |
|
* jdk-internal Unsafe for atomics and special memory modes, |
| 749 |
|
* rather than VarHandles, to avoid initialization dependencies in |
| 750 |
|
* other jdk components that require early parallelism. This can |
| 752 |
|
* outcomes across the unusual cases that arise in very racy code |
| 753 |
|
* with very few invariants. All fields are read into locals |
| 754 |
|
* before use, and null-checked if they are references, even if |
| 755 |
< |
* they can never be null under current usages. Array accesses |
| 756 |
< |
* using masked indices include checks (that are always true) that |
| 757 |
< |
* the array length is non-zero to avoid compilers inserting more |
| 758 |
< |
* expensive traps. This is usually done in a "C"-like style of |
| 759 |
< |
* listing declarations at the heads of methods or blocks, and |
| 760 |
< |
* using inline assignments on first encounter. Nearly all |
| 761 |
< |
* explicit checks lead to bypass/return, not exception throws, |
| 762 |
< |
* because they may legitimately arise during shutdown. |
| 755 |
> |
* they can never be null under current usages. Usually, |
| 756 |
> |
* computations (held in local variables) are defined as soon as |
| 757 |
> |
* logically enabled, sometimes to convince compilers that they |
| 758 |
> |
* may be performed despite memory ordering constraints. Array |
| 759 |
> |
* accesses using masked indices include checks (that are always |
| 760 |
> |
* true) that the array length is non-zero to avoid compilers |
| 761 |
> |
* inserting more expensive traps. This is usually done in a |
| 762 |
> |
* "C"-like style of listing declarations at the heads of methods |
| 763 |
> |
* or blocks, and using inline assignments on first encounter. |
| 764 |
> |
* Nearly all explicit checks lead to bypass/return, not exception |
| 765 |
> |
* throws, because they may legitimately arise during shutdown. A |
| 766 |
> |
* few unusual loop constructions encourage (with varying |
| 767 |
> |
* effectiveness) JVMs about where (not) to place safepoints. |
| 768 |
|
* |
| 769 |
|
* There is a lot of representation-level coupling among classes |
| 770 |
|
* ForkJoinPool, ForkJoinWorkerThread, and ForkJoinTask. The |
| 781 |
|
* perform reasonably even when interpreted (not compiled). |
| 782 |
|
* |
| 783 |
|
* The order of declarations in this file is (with a few exceptions): |
| 784 |
< |
* (1) Static utility functions |
| 785 |
< |
* (2) Nested (static) classes |
| 786 |
< |
* (3) Static fields |
| 784 |
> |
* (1) Static constants |
| 785 |
> |
* (2) Static utility functions |
| 786 |
> |
* (3) Nested (static) classes |
| 787 |
|
* (4) Fields, along with constants used when unpacking some of them |
| 788 |
|
* (5) Internal control methods |
| 789 |
|
* (6) Callbacks and other support for ForkJoinTask methods |
| 793 |
|
* Revision notes |
| 794 |
|
* ============== |
| 795 |
|
* |
| 796 |
< |
* The main sources of differences of January 2020 ForkJoin |
| 707 |
< |
* classes from previous version are: |
| 796 |
> |
* The main sources of differences from previous version are: |
| 797 |
|
* |
| 798 |
< |
* * ForkJoinTask now uses field "aux" to support blocking joins |
| 799 |
< |
* and/or record exceptions, replacing reliance on builtin |
| 800 |
< |
* monitors and side tables. |
| 801 |
< |
* * Scans probe slots (vs compare indices), along with related |
| 802 |
< |
* changes that reduce performance differences across most |
| 803 |
< |
* garbage collectors, and reduce contention. |
| 804 |
< |
* * Refactoring for better integration of special task types and |
| 805 |
< |
* other capabilities that had been incrementally tacked on. Plus |
| 806 |
< |
* many minor reworkings to improve consistency. |
| 798 |
> |
* * Use of Unsafe vs VarHandle, including re-instatement of some |
| 799 |
> |
* constructions from pre-VarHandle versions. |
| 800 |
> |
* * Reduced memory and signal contention, mainly by distinguishing |
| 801 |
> |
* failure cases. |
| 802 |
> |
* * Improved initialization, in part by preparing for possible |
| 803 |
> |
* removal of SecurityManager |
| 804 |
> |
* * Enable resizing (includes refactoring quiescence/termination) |
| 805 |
> |
* * Unification of most internal vs external operations; some made |
| 806 |
> |
* possible via use of WorkQueue.access, and POOLSUBMIT status in tasks. |
| 807 |
|
*/ |
| 808 |
|
|
| 809 |
+ |
// static configuration constants |
| 810 |
+ |
|
| 811 |
+ |
/** |
| 812 |
+ |
* Default idle timeout value (in milliseconds) for idle threads |
| 813 |
+ |
* to park waiting for new work before terminating. |
| 814 |
+ |
*/ |
| 815 |
+ |
static final long DEFAULT_KEEPALIVE = 60_000L; |
| 816 |
+ |
|
| 817 |
+ |
/** |
| 818 |
+ |
* Undershoot tolerance for idle timeouts |
| 819 |
+ |
*/ |
| 820 |
+ |
static final long TIMEOUT_SLOP = 20L; |
| 821 |
+ |
|
| 822 |
+ |
/** |
| 823 |
+ |
* The default value for common pool maxSpares. Overridable using |
| 824 |
+ |
* the "java.util.concurrent.ForkJoinPool.common.maximumSpares" |
| 825 |
+ |
* system property. The default value is far in excess of normal |
| 826 |
+ |
* requirements, but also far short of MAX_CAP and typical OS |
| 827 |
+ |
* thread limits, so allows JVMs to catch misuse/abuse before |
| 828 |
+ |
* running out of resources needed to do so. |
| 829 |
+ |
*/ |
| 830 |
+ |
static final int DEFAULT_COMMON_MAX_SPARES = 256; |
| 831 |
+ |
|
| 832 |
+ |
/** |
| 833 |
+ |
* Initial capacity of work-stealing queue array. Must be a power |
| 834 |
+ |
* of two, at least 2. See above. |
| 835 |
+ |
*/ |
| 836 |
+ |
static final int INITIAL_QUEUE_CAPACITY = 1 << 6; |
| 837 |
+ |
|
| 838 |
+ |
// Bounds |
| 839 |
+ |
static final int SWIDTH = 16; // width of short |
| 840 |
+ |
static final int SMASK = 0xffff; // short bits == max index |
| 841 |
+ |
static final int MAX_CAP = 0x7fff; // max #workers - 1 |
| 842 |
+ |
|
| 843 |
+ |
// pool.runState and workQueue.access bits and sentinels |
| 844 |
+ |
static final int STOP = 1 << 31; // must be negative |
| 845 |
+ |
static final int SHUTDOWN = 1; |
| 846 |
+ |
static final int TERMINATED = 2; |
| 847 |
+ |
static final int PARKED = -1; // access value when parked |
| 848 |
+ |
|
| 849 |
+ |
// {pool, workQueue}.config bits |
| 850 |
+ |
static final int FIFO = 1 << 16; // fifo queue or access mode |
| 851 |
+ |
static final int SRC = 1 << 17; // set when stealable |
| 852 |
+ |
static final int INNOCUOUS = 1 << 18; // set for Innocuous workers |
| 853 |
+ |
static final int TRIMMED = 1 << 19; // timed out while idle |
| 854 |
+ |
static final int ISCOMMON = 1 << 20; // set for common pool |
| 855 |
+ |
static final int PRESET_SIZE = 1 << 21; // size was set by property |
| 856 |
+ |
|
| 857 |
+ |
static final int UNCOMPENSATE = 1 << 16; // tryCompensate return |
| 858 |
+ |
|
| 859 |
+ |
/* |
| 860 |
+ |
* Bits and masks for ctl and bounds are packed with 4 16 bit subfields: |
| 861 |
+ |
* RC: Number of released (unqueued) workers |
| 862 |
+ |
* TC: Number of total workers |
| 863 |
+ |
* SS: version count and status of top waiting thread |
| 864 |
+ |
* ID: poolIndex of top of Treiber stack of waiters |
| 865 |
+ |
* |
| 866 |
+ |
* When convenient, we can extract the lower 32 stack top bits |
| 867 |
+ |
* (including version bits) as sp=(int)ctl. When sp is non-zero, |
| 868 |
+ |
* there are waiting workers. Count fields may be transiently |
| 869 |
+ |
* negative during termination because of out-of-order updates. |
| 870 |
+ |
* To deal with this, we use casts in and out of "short" and/or |
| 871 |
+ |
* signed shifts to maintain signedness. Because it occupies |
| 872 |
+ |
* uppermost bits, we can add one release count using getAndAdd of |
| 873 |
+ |
* RC_UNIT, rather than CAS, when returning from a blocked join. |
| 874 |
+ |
* Other updates of multiple subfields require CAS. |
| 875 |
+ |
*/ |
| 876 |
+ |
|
| 877 |
+ |
// Lower and upper word masks |
| 878 |
+ |
static final long SP_MASK = 0xffffffffL; |
| 879 |
+ |
static final long UC_MASK = ~SP_MASK; |
| 880 |
+ |
// Release counts |
| 881 |
+ |
static final int RC_SHIFT = 48; |
| 882 |
+ |
static final long RC_UNIT = 0x0001L << RC_SHIFT; |
| 883 |
+ |
static final long RC_MASK = 0xffffL << RC_SHIFT; |
| 884 |
+ |
// Total counts |
| 885 |
+ |
static final int TC_SHIFT = 32; |
| 886 |
+ |
static final long TC_UNIT = 0x0001L << TC_SHIFT; |
| 887 |
+ |
static final long TC_MASK = 0xffffL << TC_SHIFT; |
| 888 |
+ |
// sp bits |
| 889 |
+ |
static final int SS_SEQ = 1 << 16; // version count |
| 890 |
+ |
static final int INACTIVE = 1 << 31; // phase bit when idle |
| 891 |
+ |
|
| 892 |
|
// Static utilities |
| 893 |
|
|
| 894 |
|
/** |
| 895 |
|
* If there is a security manager, makes sure caller has |
| 896 |
|
* permission to modify threads. |
| 897 |
|
*/ |
| 726 |
– |
private static void checkPermission() { |
| 727 |
– |
@SuppressWarnings("removal") |
| 728 |
– |
SecurityManager security = System.getSecurityManager(); |
| 729 |
– |
if (security != null) |
| 730 |
– |
security.checkPermission(modifyThreadPermission); |
| 731 |
– |
} |
| 732 |
– |
|
| 898 |
|
@SuppressWarnings("removal") |
| 899 |
< |
static AccessControlContext contextWithPermissions(Permission ... perms) { |
| 900 |
< |
Permissions permissions = new Permissions(); |
| 901 |
< |
for (Permission perm : perms) |
| 902 |
< |
permissions.add(perm); |
| 903 |
< |
return new AccessControlContext( |
| 904 |
< |
new ProtectionDomain[] { new ProtectionDomain(null, permissions) }); |
| 899 |
> |
private static void checkPermission() { |
| 900 |
> |
SecurityManager security; RuntimePermission perm; |
| 901 |
> |
if ((security = System.getSecurityManager()) != null) { |
| 902 |
> |
if ((perm = modifyThreadPermission) == null) |
| 903 |
> |
modifyThreadPermission = perm = // races OK |
| 904 |
> |
new RuntimePermission("modifyThread"); |
| 905 |
> |
security.checkPermission(perm); |
| 906 |
> |
} |
| 907 |
|
} |
| 908 |
|
|
| 909 |
|
// Nested classes |
| 938 |
|
* new ForkJoinWorkerThread using the system class loader as the |
| 939 |
|
* thread context class loader. |
| 940 |
|
*/ |
| 774 |
– |
@SuppressWarnings("removal") |
| 941 |
|
static final class DefaultForkJoinWorkerThreadFactory |
| 942 |
|
implements ForkJoinWorkerThreadFactory { |
| 777 |
– |
// ACC for access to the factory |
| 778 |
– |
private static final AccessControlContext ACC = contextWithPermissions( |
| 779 |
– |
new RuntimePermission("getClassLoader"), |
| 780 |
– |
new RuntimePermission("setContextClassLoader")); |
| 943 |
|
public final ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
| 944 |
+ |
boolean isCommon = (pool.workerNamePrefix == null); |
| 945 |
+ |
@SuppressWarnings("removal") |
| 946 |
+ |
SecurityManager sm = System.getSecurityManager(); |
| 947 |
+ |
if (sm == null) |
| 948 |
+ |
return new ForkJoinWorkerThread(null, pool, true); |
| 949 |
+ |
else if (isCommon) |
| 950 |
+ |
return newCommonWithACC(pool); |
| 951 |
+ |
else |
| 952 |
+ |
return newRegularWithACC(pool); |
| 953 |
+ |
} |
| 954 |
+ |
|
| 955 |
+ |
/* |
| 956 |
+ |
* Create and use static AccessControlContexts only if there |
| 957 |
+ |
* is a SecurityManager. (These can be removed if/when |
| 958 |
+ |
* SecurityManagers are removed from platform.) The ACCs are |
| 959 |
+ |
* immutable and equivalent even when racily initialized, so |
| 960 |
+ |
* they don't require locking, although with the chance of |
| 961 |
+ |
* needlessly duplicate construction. |
| 962 |
+ |
*/ |
| 963 |
+ |
@SuppressWarnings("removal") |
| 964 |
+ |
static volatile AccessControlContext regularACC, commonACC; |
| 965 |
+ |
|
| 966 |
+ |
@SuppressWarnings("removal") |
| 967 |
+ |
static ForkJoinWorkerThread newRegularWithACC(ForkJoinPool pool) { |
| 968 |
+ |
AccessControlContext acc = regularACC; |
| 969 |
+ |
if (acc == null) { |
| 970 |
+ |
Permissions ps = new Permissions(); |
| 971 |
+ |
ps.add(new RuntimePermission("getClassLoader")); |
| 972 |
+ |
ps.add(new RuntimePermission("setContextClassLoader")); |
| 973 |
+ |
regularACC = acc = |
| 974 |
+ |
new AccessControlContext(new ProtectionDomain[] { |
| 975 |
+ |
new ProtectionDomain(null, ps) }); |
| 976 |
+ |
} |
| 977 |
|
return AccessController.doPrivileged( |
| 978 |
|
new PrivilegedAction<>() { |
| 979 |
|
public ForkJoinWorkerThread run() { |
| 980 |
< |
return new ForkJoinWorkerThread(null, pool, true, false); |
| 981 |
< |
}}, |
| 787 |
< |
ACC); |
| 980 |
> |
return new ForkJoinWorkerThread(null, pool, true); |
| 981 |
> |
}}, acc); |
| 982 |
|
} |
| 789 |
– |
} |
| 983 |
|
|
| 791 |
– |
/** |
| 792 |
– |
* Factory for CommonPool unless overridden by System property. |
| 793 |
– |
* Creates InnocuousForkJoinWorkerThreads if a security manager is |
| 794 |
– |
* present at time of invocation. Support requires that we break |
| 795 |
– |
* quite a lot of encapsulation (some via helper methods in |
| 796 |
– |
* ThreadLocalRandom) to access and set Thread fields. |
| 797 |
– |
*/ |
| 798 |
– |
@SuppressWarnings("removal") |
| 799 |
– |
static final class DefaultCommonPoolForkJoinWorkerThreadFactory |
| 800 |
– |
implements ForkJoinWorkerThreadFactory { |
| 984 |
|
@SuppressWarnings("removal") |
| 985 |
< |
private static final AccessControlContext ACC = contextWithPermissions( |
| 986 |
< |
modifyThreadPermission, |
| 987 |
< |
new RuntimePermission("enableContextClassLoaderOverride"), |
| 988 |
< |
new RuntimePermission("modifyThreadGroup"), |
| 989 |
< |
new RuntimePermission("getClassLoader"), |
| 990 |
< |
new RuntimePermission("setContextClassLoader")); |
| 991 |
< |
|
| 992 |
< |
public final ForkJoinWorkerThread newThread(ForkJoinPool pool) { |
| 993 |
< |
@SuppressWarnings("removal") |
| 994 |
< |
ForkJoinWorkerThread t = |
| 995 |
< |
AccessController.doPrivileged( |
| 996 |
< |
new PrivilegedAction<>() { |
| 997 |
< |
public ForkJoinWorkerThread run() { |
| 998 |
< |
return System.getSecurityManager() == null ? |
| 999 |
< |
new ForkJoinWorkerThread(null, pool, true, true): |
| 1000 |
< |
new ForkJoinWorkerThread. |
| 1001 |
< |
InnocuousForkJoinWorkerThread(pool); }}, |
| 1002 |
< |
ACC); |
| 1003 |
< |
return t; |
| 985 |
> |
static ForkJoinWorkerThread newCommonWithACC(ForkJoinPool pool) { |
| 986 |
> |
AccessControlContext acc = commonACC; |
| 987 |
> |
if (acc == null) { |
| 988 |
> |
Permissions ps = new Permissions(); |
| 989 |
> |
ps.add(new RuntimePermission("getClassLoader")); |
| 990 |
> |
ps.add(new RuntimePermission("setContextClassLoader")); |
| 991 |
> |
ps.add(new RuntimePermission("modifyThread")); |
| 992 |
> |
ps.add(new RuntimePermission("enableContextClassLoaderOverride")); |
| 993 |
> |
ps.add(new RuntimePermission("modifyThreadGroup")); |
| 994 |
> |
commonACC = acc = |
| 995 |
> |
new AccessControlContext(new ProtectionDomain[] { |
| 996 |
> |
new ProtectionDomain(null, ps) }); |
| 997 |
> |
} |
| 998 |
> |
return AccessController.doPrivileged( |
| 999 |
> |
new PrivilegedAction<>() { |
| 1000 |
> |
public ForkJoinWorkerThread run() { |
| 1001 |
> |
return new ForkJoinWorkerThread. |
| 1002 |
> |
InnocuousForkJoinWorkerThread(pool); |
| 1003 |
> |
}}, acc); |
| 1004 |
|
} |
| 1005 |
|
} |
| 1006 |
|
|
| 824 |
– |
// Constants shared across ForkJoinPool and WorkQueue |
| 825 |
– |
|
| 826 |
– |
// Bounds |
| 827 |
– |
static final int SWIDTH = 16; // width of short |
| 828 |
– |
static final int SMASK = 0xffff; // short bits == max index |
| 829 |
– |
static final int MAX_CAP = 0x7fff; // max #workers - 1 |
| 830 |
– |
|
| 831 |
– |
// Masks and units for WorkQueue.phase and ctl sp subfield |
| 832 |
– |
static final int UNSIGNALLED = 1 << 31; // must be negative |
| 833 |
– |
static final int SS_SEQ = 1 << 16; // version count |
| 834 |
– |
|
| 835 |
– |
// Mode bits and sentinels, some also used in WorkQueue fields |
| 836 |
– |
static final int FIFO = 1 << 16; // fifo queue or access mode |
| 837 |
– |
static final int SRC = 1 << 17; // set for valid queue ids |
| 838 |
– |
static final int INNOCUOUS = 1 << 18; // set for Innocuous workers |
| 839 |
– |
static final int QUIET = 1 << 19; // quiescing phase or source |
| 840 |
– |
static final int SHUTDOWN = 1 << 24; |
| 841 |
– |
static final int TERMINATED = 1 << 25; |
| 842 |
– |
static final int UNSTOPPABLE = 1 << 26; // true for common pool |
| 843 |
– |
static final int STOP = 1 << 31; // must be negative |
| 844 |
– |
static final int UNCOMPENSATE = 1 << 16; // tryCompensate return |
| 845 |
– |
|
| 846 |
– |
/** |
| 847 |
– |
* Initial capacity of work-stealing queue array. Must be a power |
| 848 |
– |
* of two, at least 2. See above. |
| 849 |
– |
*/ |
| 850 |
– |
static final int INITIAL_QUEUE_CAPACITY = 1 << 8; |
| 851 |
– |
|
| 1007 |
|
/** |
| 1008 |
|
* Queues supporting work-stealing as well as external task |
| 1009 |
|
* submission. See above for descriptions and algorithms. |
| 1010 |
|
*/ |
| 1011 |
|
static final class WorkQueue { |
| 857 |
– |
volatile int phase; // versioned, negative if inactive |
| 1012 |
|
int stackPred; // pool stack (ctl) predecessor link |
| 1013 |
|
int config; // index, mode, ORed with SRC after init |
| 1014 |
|
int base; // index of next slot for poll |
| 1015 |
|
ForkJoinTask<?>[] array; // the queued tasks; power of 2 size |
| 1016 |
|
final ForkJoinWorkerThread owner; // owning thread or null if shared |
| 1017 |
|
|
| 1018 |
< |
// segregate fields frequently updated but not read by scans or steals |
| 1018 |
> |
// fields otherwise causing more unnecessary false-sharing cache misses |
| 1019 |
|
@jdk.internal.vm.annotation.Contended("w") |
| 1020 |
|
int top; // index of next slot for push |
| 1021 |
|
@jdk.internal.vm.annotation.Contended("w") |
| 1022 |
< |
volatile int source; // source queue id, lock, or sentinel |
| 1022 |
> |
volatile int access; // values 0, 1 (locked), PARKED, STOP |
| 1023 |
> |
@jdk.internal.vm.annotation.Contended("w") |
| 1024 |
> |
volatile int phase; // versioned, negative if inactive |
| 1025 |
> |
@jdk.internal.vm.annotation.Contended("w") |
| 1026 |
> |
volatile int source; // source queue id in topLevelExec |
| 1027 |
|
@jdk.internal.vm.annotation.Contended("w") |
| 1028 |
|
int nsteals; // number of steals from other queues |
| 1029 |
|
|
| 1030 |
|
// Support for atomic operations |
| 1031 |
|
private static final Unsafe U; |
| 1032 |
< |
private static final long SOURCE; |
| 1033 |
< |
private static final long BASE; |
| 1032 |
> |
private static final long ACCESS; |
| 1033 |
> |
private static final long PHASE; |
| 1034 |
|
private static final long ABASE; |
| 1035 |
< |
private static final int ASHIFT; |
| 1036 |
< |
@SuppressWarnings("removal") |
| 1037 |
< |
static final ForkJoinTask<?> getSlot(ForkJoinTask<?>[] a, int i) { |
| 880 |
< |
return (ForkJoinTask<?>) |
| 881 |
< |
U.getObjectAcquire(a, ((long)i << ASHIFT) + ABASE); |
| 882 |
< |
} |
| 883 |
< |
@SuppressWarnings("removal") |
| 884 |
< |
static final ForkJoinTask<?> getAndClearSlot(ForkJoinTask<?>[] a, |
| 885 |
< |
int i) { |
| 1035 |
> |
private static final int ASHIFT; |
| 1036 |
> |
|
| 1037 |
> |
static ForkJoinTask<?> getAndClearSlot(ForkJoinTask<?>[] a, int i) { |
| 1038 |
|
return (ForkJoinTask<?>) |
| 1039 |
< |
U.getAndSetObject(a, ((long)i << ASHIFT) + ABASE, null); |
| 1039 |
> |
U.getAndSetReference(a, ((long)i << ASHIFT) + ABASE, null); |
| 1040 |
|
} |
| 1041 |
< |
@SuppressWarnings("removal") |
| 1042 |
< |
static final void setSlotVolatile(ForkJoinTask<?>[] a, int i, |
| 1043 |
< |
ForkJoinTask<?> v) { |
| 1044 |
< |
U.putObjectVolatile(a, ((long)i << ASHIFT) + ABASE, v); |
| 1041 |
> |
static boolean casSlotToNull(ForkJoinTask<?>[] a, int i, |
| 1042 |
> |
ForkJoinTask<?> c) { |
| 1043 |
> |
return U.compareAndSetReference(a, ((long)i << ASHIFT) + ABASE, |
| 1044 |
> |
c, null); |
| 1045 |
|
} |
| 1046 |
< |
@SuppressWarnings("removal") |
| 1047 |
< |
static final boolean casSlotToNull(ForkJoinTask<?>[] a, int i, |
| 896 |
< |
ForkJoinTask<?> c) { |
| 897 |
< |
return U.compareAndSetObject(a, ((long)i << ASHIFT) + ABASE, c, null); |
| 1046 |
> |
final void forcePhaseActive() { // clear sign bit |
| 1047 |
> |
U.getAndBitwiseAndInt(this, PHASE, 0x7fffffff); |
| 1048 |
|
} |
| 1049 |
< |
final boolean tryLock() { |
| 1050 |
< |
return U.compareAndSetInt(this, SOURCE, 0, 1); |
| 901 |
< |
} |
| 902 |
< |
final void setBaseOpaque(int b) { |
| 903 |
< |
U.putIntOpaque(this, BASE, b); |
| 1049 |
> |
final int getAndSetAccess(int v) { |
| 1050 |
> |
return U.getAndSetInt(this, ACCESS, v); |
| 1051 |
|
} |
| 1052 |
|
|
| 1053 |
|
/** |
| 1054 |
< |
* Constructor used by ForkJoinWorkerThreads. Most fields |
| 1055 |
< |
* are initialized upon thread start, in pool.registerWorker. |
| 1054 |
> |
* Constructor. For owned queues, most fields are initialized |
| 1055 |
> |
* upon thread start in pool.registerWorker. |
| 1056 |
|
*/ |
| 1057 |
< |
WorkQueue(ForkJoinWorkerThread owner, boolean isInnocuous) { |
| 911 |
< |
this.config = (isInnocuous) ? INNOCUOUS : 0; |
| 1057 |
> |
WorkQueue(ForkJoinWorkerThread owner, int config) { |
| 1058 |
|
this.owner = owner; |
| 913 |
– |
} |
| 914 |
– |
|
| 915 |
– |
/** |
| 916 |
– |
* Constructor used for external queues. |
| 917 |
– |
*/ |
| 918 |
– |
WorkQueue(int config) { |
| 919 |
– |
array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; |
| 1059 |
|
this.config = config; |
| 921 |
– |
owner = null; |
| 922 |
– |
phase = -1; |
| 1060 |
|
} |
| 1061 |
|
|
| 1062 |
|
/** |
| 1070 |
|
* Returns the approximate number of tasks in the queue. |
| 1071 |
|
*/ |
| 1072 |
|
final int queueSize() { |
| 1073 |
< |
VarHandle.acquireFence(); // ensure fresh reads by external callers |
| 937 |
< |
int n = top - base; |
| 1073 |
> |
int unused = access, n = top - base; // for ordering effect |
| 1074 |
|
return (n < 0) ? 0 : n; // ignore transient negative |
| 1075 |
|
} |
| 1076 |
|
|
| 1077 |
|
/** |
| 1078 |
< |
* Provides a more conservative estimate of whether this queue |
| 943 |
< |
* has any tasks than does queueSize. |
| 944 |
< |
*/ |
| 945 |
< |
final boolean isEmpty() { |
| 946 |
< |
return !((source != 0 && owner == null) || top - base > 0); |
| 947 |
< |
} |
| 948 |
< |
|
| 949 |
< |
/** |
| 950 |
< |
* Pushes a task. Call only by owner in unshared queues. |
| 1078 |
> |
* Pushes a task. Called only by owner or if already locked |
| 1079 |
|
* |
| 1080 |
|
* @param task the task. Caller must ensure non-null. |
| 1081 |
< |
* @param pool if nonnull, pool to signal if was empty |
| 1081 |
> |
* @param pool the pool. Must be non-null unless terminating. |
| 1082 |
> |
* @param signalIfEmpty true if signal when pushing to empty queue |
| 1083 |
|
* @throws RejectedExecutionException if array cannot be resized |
| 1084 |
|
*/ |
| 1085 |
< |
final void push(ForkJoinTask<?> task, ForkJoinPool pool) { |
| 1086 |
< |
ForkJoinTask<?>[] a = array; |
| 1087 |
< |
int s = top++, d = s - base, cap, m; // skip insert if disabled |
| 1088 |
< |
if (a != null && (cap = a.length) > 0) { |
| 1089 |
< |
setSlotVolatile(a, (m = cap - 1) & s, task); |
| 1090 |
< |
if (d == m) |
| 1091 |
< |
growArray(); |
| 1092 |
< |
if ((d == m || a[m & (s - 1)] == null) && pool != null) |
| 1093 |
< |
pool.signalWork(); // signal if was empty or resized |
| 1094 |
< |
} |
| 1095 |
< |
} |
| 1096 |
< |
|
| 1097 |
< |
/** |
| 1098 |
< |
* Pushes task to a shared queue with lock already held, and unlocks. |
| 1099 |
< |
* |
| 1100 |
< |
* @return true if caller should signal work |
| 1101 |
< |
*/ |
| 1102 |
< |
final boolean lockedPush(ForkJoinTask<?> task) { |
| 1103 |
< |
ForkJoinTask<?>[] a = array; |
| 1104 |
< |
int s = top++, d = s - base, cap, m; |
| 1105 |
< |
if (a != null && (cap = a.length) > 0) { |
| 1106 |
< |
a[(m = cap - 1) & s] = task; |
| 1107 |
< |
if (d == m) |
| 1108 |
< |
growArray(); |
| 980 |
< |
source = 0; // unlock |
| 981 |
< |
if (d == m || a[m & (s - 1)] == null) |
| 982 |
< |
return true; |
| 983 |
< |
} |
| 984 |
< |
return false; |
| 985 |
< |
} |
| 986 |
< |
|
| 987 |
< |
/** |
| 988 |
< |
* Doubles the capacity of array. Called by owner or with lock |
| 989 |
< |
* held after pre-incrementing top, which is reverted on |
| 990 |
< |
* allocation failure. |
| 991 |
< |
*/ |
| 992 |
< |
final void growArray() { |
| 993 |
< |
ForkJoinTask<?>[] oldArray = array, newArray; |
| 994 |
< |
int s = top - 1, oldCap, newCap; |
| 995 |
< |
if (oldArray != null && (oldCap = oldArray.length) > 0 && |
| 996 |
< |
(newCap = oldCap << 1) > 0) { // skip if disabled |
| 997 |
< |
try { |
| 998 |
< |
newArray = new ForkJoinTask<?>[newCap]; |
| 999 |
< |
} catch (Throwable ex) { |
| 1000 |
< |
top = s; |
| 1001 |
< |
if (owner == null) |
| 1002 |
< |
source = 0; // unlock |
| 1003 |
< |
throw new RejectedExecutionException( |
| 1004 |
< |
"Queue capacity exceeded"); |
| 1005 |
< |
} |
| 1006 |
< |
int newMask = newCap - 1, oldMask = oldCap - 1; |
| 1007 |
< |
for (int k = oldCap; k > 0; --k, --s) { |
| 1008 |
< |
ForkJoinTask<?> x; // poll old, push to new |
| 1009 |
< |
if ((x = getAndClearSlot(oldArray, s & oldMask)) == null) |
| 1010 |
< |
break; // others already taken |
| 1011 |
< |
newArray[s & newMask] = x; |
| 1085 |
> |
final void push(ForkJoinTask<?> task, ForkJoinPool pool, |
| 1086 |
> |
boolean signalIfEmpty) { |
| 1087 |
> |
boolean resize = false; |
| 1088 |
> |
int s = top++, b = base, cap, m; ForkJoinTask<?>[] a; |
| 1089 |
> |
if ((a = array) != null && (cap = a.length) > 0) { |
| 1090 |
> |
if ((m = (cap - 1)) == s - b) { |
| 1091 |
> |
resize = true; // rapidly grow until large |
| 1092 |
> |
int newCap = (cap < 1 << 24) ? cap << 2 : cap << 1; |
| 1093 |
> |
ForkJoinTask<?>[] newArray; |
| 1094 |
> |
try { |
| 1095 |
> |
newArray = new ForkJoinTask<?>[newCap]; |
| 1096 |
> |
} catch (Throwable ex) { |
| 1097 |
> |
top = s; |
| 1098 |
> |
access = 0; |
| 1099 |
> |
throw new RejectedExecutionException( |
| 1100 |
> |
"Queue capacity exceeded"); |
| 1101 |
> |
} |
| 1102 |
> |
if (newCap > 0) { // always true |
| 1103 |
> |
int newMask = newCap - 1, k = s; |
| 1104 |
> |
do { // poll old, push to new |
| 1105 |
> |
newArray[k-- & newMask] = task; |
| 1106 |
> |
} while ((task = getAndClearSlot(a, k & m)) != null); |
| 1107 |
> |
} |
| 1108 |
> |
array = newArray; |
| 1109 |
|
} |
| 1110 |
< |
VarHandle.releaseFence(); // fill before publish |
| 1111 |
< |
array = newArray; |
| 1110 |
> |
else |
| 1111 |
> |
a[m & s] = task; |
| 1112 |
> |
getAndSetAccess(0); // for memory effects if owned |
| 1113 |
> |
if ((resize || (a[m & (s - 1)] == null && signalIfEmpty)) && |
| 1114 |
> |
pool != null) |
| 1115 |
> |
pool.signalWork(); |
| 1116 |
|
} |
| 1117 |
|
} |
| 1118 |
|
|
| 1018 |
– |
// Variants of pop |
| 1019 |
– |
|
| 1119 |
|
/** |
| 1120 |
< |
* Pops and returns task, or null if empty. Called only by owner. |
| 1120 |
> |
* Takes next task, if one exists, in order specified by mode, |
| 1121 |
> |
* so acts as either local-pop or local-poll. Called only by owner. |
| 1122 |
> |
* @param fifo nonzero if FIFO mode |
| 1123 |
|
*/ |
| 1124 |
< |
private ForkJoinTask<?> pop() { |
| 1124 |
> |
final ForkJoinTask<?> nextLocalTask(int fifo) { |
| 1125 |
|
ForkJoinTask<?> t = null; |
| 1126 |
< |
int s = top, cap; ForkJoinTask<?>[] a; |
| 1127 |
< |
if ((a = array) != null && (cap = a.length) > 0 && base != s-- && |
| 1128 |
< |
(t = getAndClearSlot(a, (cap - 1) & s)) != null) |
| 1129 |
< |
top = s; |
| 1126 |
> |
ForkJoinTask<?>[] a = array; |
| 1127 |
> |
int p = top, s = p - 1, b = base, nb, cap; |
| 1128 |
> |
if (p - b > 0 && a != null && (cap = a.length) > 0) { |
| 1129 |
> |
do { |
| 1130 |
> |
if (fifo == 0 || (nb = b + 1) == p) { |
| 1131 |
> |
if ((t = getAndClearSlot(a, (cap - 1) & s)) != null) { |
| 1132 |
> |
top = s; |
| 1133 |
> |
U.storeFence(); |
| 1134 |
> |
} |
| 1135 |
> |
break; // lost race for only task |
| 1136 |
> |
} |
| 1137 |
> |
else if ((t = getAndClearSlot(a, (cap - 1) & b)) != null) { |
| 1138 |
> |
base = nb; |
| 1139 |
> |
U.storeFence(); |
| 1140 |
> |
break; |
| 1141 |
> |
} |
| 1142 |
> |
else { |
| 1143 |
> |
while (b == (b = base)) { |
| 1144 |
> |
U.loadFence(); |
| 1145 |
> |
Thread.onSpinWait(); // spin to reduce memory traffic |
| 1146 |
> |
} |
| 1147 |
> |
} |
| 1148 |
> |
} while (p - b > 0); |
| 1149 |
> |
} |
| 1150 |
|
return t; |
| 1151 |
|
} |
| 1152 |
|
|
| 1153 |
|
/** |
| 1154 |
< |
* Pops the given task for owner only if it is at the current top. |
| 1154 |
> |
* Takes next task, if one exists, using configured mode. |
| 1155 |
> |
* (Always owned, never called for Common pool.) |
| 1156 |
|
*/ |
| 1157 |
< |
final boolean tryUnpush(ForkJoinTask<?> task) { |
| 1158 |
< |
int s = top, cap; ForkJoinTask<?>[] a; |
| 1037 |
< |
if ((a = array) != null && (cap = a.length) > 0 && base != s-- && |
| 1038 |
< |
casSlotToNull(a, (cap - 1) & s, task)) { |
| 1039 |
< |
top = s; |
| 1040 |
< |
return true; |
| 1041 |
< |
} |
| 1042 |
< |
return false; |
| 1157 |
> |
final ForkJoinTask<?> nextLocalTask() { |
| 1158 |
> |
return nextLocalTask(config & FIFO); |
| 1159 |
|
} |
| 1160 |
|
|
| 1161 |
|
/** |
| 1162 |
< |
* Locking version of tryUnpush. |
| 1162 |
> |
* Pops the given task only if it is at the current top. |
| 1163 |
|
*/ |
| 1164 |
< |
final boolean externalTryUnpush(ForkJoinTask<?> task) { |
| 1164 |
> |
final boolean tryUnpush(ForkJoinTask<?> task, boolean owned) { |
| 1165 |
|
boolean taken = false; |
| 1166 |
< |
for (;;) { |
| 1167 |
< |
int s = top, cap, k; ForkJoinTask<?>[] a; |
| 1168 |
< |
if ((a = array) == null || (cap = a.length) <= 0 || |
| 1169 |
< |
a[k = (cap - 1) & (s - 1)] != task) |
| 1170 |
< |
break; |
| 1171 |
< |
if (tryLock()) { |
| 1172 |
< |
if (top == s && array == a) { |
| 1173 |
< |
if (taken = casSlotToNull(a, k, task)) { |
| 1174 |
< |
top = s - 1; |
| 1175 |
< |
source = 0; |
| 1060 |
< |
break; |
| 1061 |
< |
} |
| 1166 |
> |
ForkJoinTask<?>[] a = array; |
| 1167 |
> |
int p = top, s, cap, k; |
| 1168 |
> |
if (task != null && base != p && a != null && (cap = a.length) > 0 && |
| 1169 |
> |
a[k = (cap - 1) & (s = p - 1)] == task) { |
| 1170 |
> |
if (owned || getAndSetAccess(1) == 0) { |
| 1171 |
> |
if ((owned || (top == p && a[k] == task)) && |
| 1172 |
> |
getAndClearSlot(a, k) != null) { |
| 1173 |
> |
taken = true; |
| 1174 |
> |
top = s; |
| 1175 |
> |
U.storeFence(); |
| 1176 |
|
} |
| 1177 |
< |
source = 0; // release lock for retry |
| 1177 |
> |
if (!owned) |
| 1178 |
> |
access = 0; |
| 1179 |
|
} |
| 1065 |
– |
Thread.yield(); // trylock failure |
| 1180 |
|
} |
| 1181 |
|
return taken; |
| 1182 |
|
} |
| 1183 |
|
|
| 1184 |
|
/** |
| 1185 |
< |
* Deep form of tryUnpush: Traverses from top and removes task if |
| 1072 |
< |
* present, shifting others to fill gap. |
| 1185 |
> |
* Returns next task, if one exists, in order specified by mode. |
| 1186 |
|
*/ |
| 1187 |
< |
final boolean tryRemove(ForkJoinTask<?> task, boolean owned) { |
| 1188 |
< |
boolean taken = false; |
| 1189 |
< |
int p = top, cap; ForkJoinTask<?>[] a; ForkJoinTask<?> t; |
| 1190 |
< |
if ((a = array) != null && task != null && (cap = a.length) > 0) { |
| 1191 |
< |
int m = cap - 1, s = p - 1, d = p - base; |
| 1192 |
< |
for (int i = s, k; d > 0; --i, --d) { |
| 1193 |
< |
if ((t = a[k = i & m]) == task) { |
| 1194 |
< |
if (owned || tryLock()) { |
| 1195 |
< |
if ((owned || (array == a && top == p)) && |
| 1196 |
< |
(taken = casSlotToNull(a, k, t))) { |
| 1197 |
< |
for (int j = i; j != s; ) // shift down |
| 1085 |
< |
a[j & m] = getAndClearSlot(a, ++j & m); |
| 1086 |
< |
top = s; |
| 1087 |
< |
} |
| 1088 |
< |
if (!owned) |
| 1089 |
< |
source = 0; |
| 1090 |
< |
} |
| 1091 |
< |
break; |
| 1187 |
> |
final ForkJoinTask<?> peek() { |
| 1188 |
> |
ForkJoinTask<?>[] a = array; |
| 1189 |
> |
int cfg = config, p = top, b = base, cap; |
| 1190 |
> |
if (p != b && a != null && (cap = a.length) > 0) { |
| 1191 |
> |
if ((cfg & FIFO) == 0) |
| 1192 |
> |
return a[(cap - 1) & (p - 1)]; |
| 1193 |
> |
else { // skip over in-progress removals |
| 1194 |
> |
ForkJoinTask<?> t; |
| 1195 |
> |
for ( ; p - b > 0; ++b) { |
| 1196 |
> |
if ((t = a[(cap - 1) & b]) != null) |
| 1197 |
> |
return t; |
| 1198 |
|
} |
| 1199 |
|
} |
| 1200 |
|
} |
| 1201 |
< |
return taken; |
| 1201 |
> |
return null; |
| 1202 |
|
} |
| 1203 |
|
|
| 1098 |
– |
// variants of poll |
| 1099 |
– |
|
| 1204 |
|
/** |
| 1205 |
< |
* Tries once to poll next task in FIFO order, failing on |
| 1206 |
< |
* inconsistency or contention. |
| 1205 |
> |
* Polls for a task. Used only by non-owners in usually |
| 1206 |
> |
* uncontended contexts. |
| 1207 |
> |
* |
| 1208 |
> |
* @param pool if nonnull, pool to signal if more tasks exist |
| 1209 |
|
*/ |
| 1210 |
< |
final ForkJoinTask<?> tryPoll() { |
| 1211 |
< |
int cap, b, k; ForkJoinTask<?>[] a; |
| 1212 |
< |
if ((a = array) != null && (cap = a.length) > 0) { |
| 1213 |
< |
ForkJoinTask<?> t = getSlot(a, k = (cap - 1) & (b = base)); |
| 1214 |
< |
if (base == b++ && t != null && casSlotToNull(a, k, t)) { |
| 1215 |
< |
setBaseOpaque(b); |
| 1210 |
> |
final ForkJoinTask<?> poll(ForkJoinPool pool) { |
| 1211 |
> |
for (int b = base;;) { |
| 1212 |
> |
int cap; ForkJoinTask<?>[] a; |
| 1213 |
> |
if ((a = array) == null || (cap = a.length) <= 0) |
| 1214 |
> |
break; // currently impossible |
| 1215 |
> |
int k = (cap - 1) & b, nb = b + 1, nk = (cap - 1) & nb; |
| 1216 |
> |
ForkJoinTask<?> t = a[k]; |
| 1217 |
> |
U.loadFence(); // for re-reads |
| 1218 |
> |
if (b != (b = base)) // inconsistent |
| 1219 |
> |
; |
| 1220 |
> |
else if (t != null && casSlotToNull(a, k, t)) { |
| 1221 |
> |
base = nb; |
| 1222 |
> |
U.storeFence(); |
| 1223 |
> |
if (pool != null && a[nk] != null) |
| 1224 |
> |
pool.signalWork(); // propagate |
| 1225 |
|
return t; |
| 1226 |
|
} |
| 1227 |
+ |
else if (array != a || a[k] != null) |
| 1228 |
+ |
; // stale |
| 1229 |
+ |
else if (a[nk] == null && top - b <= 0) |
| 1230 |
+ |
break; // empty |
| 1231 |
|
} |
| 1232 |
|
return null; |
| 1233 |
|
} |
| 1234 |
|
|
| 1235 |
|
/** |
| 1236 |
< |
* Takes next task, if one exists, in order specified by mode. |
| 1236 |
> |
* Tries to poll next task in FIFO order, failing on |
| 1237 |
> |
* contention or stalls. Used only by topLevelExec to repoll |
| 1238 |
> |
* from the queue obtained from pool.scan. |
| 1239 |
|
*/ |
| 1240 |
< |
final ForkJoinTask<?> nextLocalTask(int cfg) { |
| 1241 |
< |
ForkJoinTask<?> t = null; |
| 1121 |
< |
int s = top, cap; ForkJoinTask<?>[] a; |
| 1240 |
> |
final ForkJoinTask<?> tryPoll() { |
| 1241 |
> |
int b = base, cap; ForkJoinTask<?>[] a; |
| 1242 |
|
if ((a = array) != null && (cap = a.length) > 0) { |
| 1243 |
< |
for (int b, d;;) { |
| 1244 |
< |
if ((d = s - (b = base)) <= 0) |
| 1245 |
< |
break; |
| 1246 |
< |
if (d == 1 || (cfg & FIFO) == 0) { |
| 1247 |
< |
if ((t = getAndClearSlot(a, --s & (cap - 1))) != null) |
| 1248 |
< |
top = s; |
| 1249 |
< |
break; |
| 1250 |
< |
} |
| 1251 |
< |
if ((t = getAndClearSlot(a, b++ & (cap - 1))) != null) { |
| 1252 |
< |
setBaseOpaque(b); |
| 1253 |
< |
break; |
| 1243 |
> |
for (;;) { |
| 1244 |
> |
int k = (cap - 1) & b, nb = b + 1; |
| 1245 |
> |
ForkJoinTask<?> t = a[k]; |
| 1246 |
> |
U.loadFence(); // for re-reads |
| 1247 |
> |
if (b != (b = base)) |
| 1248 |
> |
; // inconsistent |
| 1249 |
> |
else if (t != null) { |
| 1250 |
> |
if (casSlotToNull(a, k, t)) { |
| 1251 |
> |
base = nb; |
| 1252 |
> |
U.storeFence(); |
| 1253 |
> |
return t; |
| 1254 |
> |
} |
| 1255 |
> |
break; // contended |
| 1256 |
|
} |
| 1257 |
+ |
else if (a[k] == null) |
| 1258 |
+ |
break; // empty or stalled |
| 1259 |
|
} |
| 1260 |
|
} |
| 1261 |
< |
return t; |
| 1138 |
< |
} |
| 1139 |
< |
|
| 1140 |
< |
/** |
| 1141 |
< |
* Takes next task, if one exists, using configured mode. |
| 1142 |
< |
*/ |
| 1143 |
< |
final ForkJoinTask<?> nextLocalTask() { |
| 1144 |
< |
return nextLocalTask(config); |
| 1145 |
< |
} |
| 1146 |
< |
|
| 1147 |
< |
/** |
| 1148 |
< |
* Returns next task, if one exists, in order specified by mode. |
| 1149 |
< |
*/ |
| 1150 |
< |
final ForkJoinTask<?> peek() { |
| 1151 |
< |
VarHandle.acquireFence(); |
| 1152 |
< |
int cap; ForkJoinTask<?>[] a; |
| 1153 |
< |
return ((a = array) != null && (cap = a.length) > 0) ? |
| 1154 |
< |
a[(cap - 1) & ((config & FIFO) != 0 ? base : top - 1)] : null; |
| 1261 |
> |
return null; |
| 1262 |
|
} |
| 1263 |
|
|
| 1264 |
|
// specialized execution methods |
| 1265 |
|
|
| 1266 |
|
/** |
| 1267 |
|
* Runs the given (stolen) task if nonnull, as well as |
| 1268 |
< |
* remaining local tasks and/or others available from the |
| 1269 |
< |
* given queue. |
| 1268 |
> |
* remaining local tasks and/or others available from its |
| 1269 |
> |
* source queue, if any. |
| 1270 |
|
*/ |
| 1271 |
< |
final void topLevelExec(ForkJoinTask<?> task, WorkQueue q) { |
| 1272 |
< |
int cfg = config, nstolen = 1; |
| 1271 |
> |
final void topLevelExec(ForkJoinTask<?> task, WorkQueue src) { |
| 1272 |
> |
int cfg = config, fifo = cfg & FIFO, nstolen = 1; |
| 1273 |
|
while (task != null) { |
| 1274 |
|
task.doExec(); |
| 1275 |
< |
if ((task = nextLocalTask(cfg)) == null && |
| 1276 |
< |
q != null && (task = q.tryPoll()) != null) |
| 1275 |
> |
if ((task = nextLocalTask(fifo)) == null && |
| 1276 |
> |
src != null && (task = src.tryPoll()) != null) |
| 1277 |
|
++nstolen; |
| 1278 |
|
} |
| 1279 |
|
nsteals += nstolen; |
| 1283 |
|
} |
| 1284 |
|
|
| 1285 |
|
/** |
| 1286 |
+ |
* Deep form of tryUnpush: Traverses from top and removes and |
| 1287 |
+ |
* runs task if present, shifting others to fill gap. |
| 1288 |
+ |
* @return task status if removed, else 0 |
| 1289 |
+ |
*/ |
| 1290 |
+ |
final int tryRemoveAndExec(ForkJoinTask<?> task, boolean owned) { |
| 1291 |
+ |
boolean taken = false; |
| 1292 |
+ |
ForkJoinTask<?>[] a = array; |
| 1293 |
+ |
int p = top, s = p - 1, d = p - base, cap; |
| 1294 |
+ |
if (task != null && d > 0 && a != null && (cap = a.length) > 0) { |
| 1295 |
+ |
for (int m = cap - 1, i = s; ; --i) { |
| 1296 |
+ |
ForkJoinTask<?> t; int k; |
| 1297 |
+ |
if ((t = a[k = i & m]) == task) { |
| 1298 |
+ |
if (!owned && getAndSetAccess(1) != 0) |
| 1299 |
+ |
break; // fail if locked |
| 1300 |
+ |
if ((owned || (top == p && a[k] == task)) && |
| 1301 |
+ |
getAndClearSlot(a, k) != null) { |
| 1302 |
+ |
taken = true; |
| 1303 |
+ |
if (i != s && i == base) |
| 1304 |
+ |
base = i + 1; // avoid shift |
| 1305 |
+ |
else { |
| 1306 |
+ |
for (int j = i; j != s;) // shift down |
| 1307 |
+ |
a[j & m] = getAndClearSlot(a, ++j & m); |
| 1308 |
+ |
top = s; |
| 1309 |
+ |
} |
| 1310 |
+ |
U.storeFence(); |
| 1311 |
+ |
} |
| 1312 |
+ |
if (!owned) |
| 1313 |
+ |
access = 0; |
| 1314 |
+ |
break; |
| 1315 |
+ |
} |
| 1316 |
+ |
else if (t == null || --d == 0) |
| 1317 |
+ |
break; |
| 1318 |
+ |
} |
| 1319 |
+ |
} |
| 1320 |
+ |
if (!taken) |
| 1321 |
+ |
return 0; |
| 1322 |
+ |
return task.doExec(); |
| 1323 |
+ |
} |
| 1324 |
+ |
|
| 1325 |
+ |
/** |
| 1326 |
|
* Tries to pop and run tasks within the target's computation |
| 1327 |
|
* until done, not found, or limit exceeded. |
| 1328 |
|
* |
| 1329 |
< |
* @param task root of CountedCompleter computation |
| 1183 |
< |
* @param owned true if owned by a ForkJoinWorkerThread |
| 1329 |
> |
* @param task root of computation |
| 1330 |
|
* @param limit max runs, or zero for no limit |
| 1331 |
|
* @return task status on exit |
| 1332 |
|
*/ |
| 1333 |
|
final int helpComplete(ForkJoinTask<?> task, boolean owned, int limit) { |
| 1334 |
< |
int status = 0, cap, k, p, s; ForkJoinTask<?>[] a; ForkJoinTask<?> t; |
| 1335 |
< |
while (task != null && (status = task.status) >= 0 && |
| 1336 |
< |
(a = array) != null && (cap = a.length) > 0 && |
| 1337 |
< |
(t = a[k = (cap - 1) & (s = (p = top) - 1)]) |
| 1338 |
< |
instanceof CountedCompleter) { |
| 1339 |
< |
CountedCompleter<?> f = (CountedCompleter<?>)t; |
| 1340 |
< |
boolean taken = false; |
| 1341 |
< |
for (;;) { // exec if root task is a completer of t |
| 1342 |
< |
if (f == task) { |
| 1343 |
< |
if (owned) { |
| 1344 |
< |
if ((taken = casSlotToNull(a, k, t))) |
| 1199 |
< |
top = s; |
| 1200 |
< |
} |
| 1201 |
< |
else if (tryLock()) { |
| 1202 |
< |
if (top == p && array == a && |
| 1203 |
< |
(taken = casSlotToNull(a, k, t))) |
| 1204 |
< |
top = s; |
| 1205 |
< |
source = 0; |
| 1206 |
< |
} |
| 1207 |
< |
if (taken) |
| 1208 |
< |
t.doExec(); |
| 1209 |
< |
else if (!owned) |
| 1210 |
< |
Thread.yield(); // tryLock failure |
| 1334 |
> |
int status = 0; |
| 1335 |
> |
if (task != null) { |
| 1336 |
> |
outer: for (;;) { |
| 1337 |
> |
boolean taken = false; |
| 1338 |
> |
ForkJoinTask<?>[] a; ForkJoinTask<?> t; |
| 1339 |
> |
int p, s, cap, k; |
| 1340 |
> |
if ((status = task.status) < 0) |
| 1341 |
> |
return status; |
| 1342 |
> |
if ((a = array) == null || (cap = a.length) <= 0 || |
| 1343 |
> |
(t = a[k = (cap - 1) & (s = (p = top) - 1)]) == null || |
| 1344 |
> |
!(t instanceof CountedCompleter)) |
| 1345 |
|
break; |
| 1346 |
+ |
for (CountedCompleter<?> f = (CountedCompleter<?>)t;;) { |
| 1347 |
+ |
if (f == task) |
| 1348 |
+ |
break; |
| 1349 |
+ |
else if ((f = f.completer) == null) |
| 1350 |
+ |
break outer; // ineligible |
| 1351 |
+ |
} |
| 1352 |
+ |
if (!owned && getAndSetAccess(1) != 0) |
| 1353 |
+ |
break; // fail if locked |
| 1354 |
+ |
if ((owned || (top == p && a[k] == t)) && |
| 1355 |
+ |
getAndClearSlot(a, k) != null) { |
| 1356 |
+ |
taken = true; |
| 1357 |
+ |
top = s; |
| 1358 |
+ |
U.storeFence(); |
| 1359 |
+ |
} |
| 1360 |
+ |
if (!owned) |
| 1361 |
+ |
access = 0; |
| 1362 |
+ |
if (taken) { |
| 1363 |
+ |
t.doExec(); |
| 1364 |
+ |
if (limit != 0 && --limit == 0) |
| 1365 |
+ |
break; |
| 1366 |
|
} |
| 1213 |
– |
else if ((f = f.completer) == null) |
| 1214 |
– |
break; |
| 1367 |
|
} |
| 1368 |
< |
if (taken && limit != 0 && --limit == 0) |
| 1217 |
< |
break; |
| 1368 |
> |
status = task.status; |
| 1369 |
|
} |
| 1370 |
|
return status; |
| 1371 |
|
} |
| 1372 |
|
|
| 1373 |
|
/** |
| 1374 |
|
* Tries to poll and run AsynchronousCompletionTasks until |
| 1375 |
< |
* none found or blocker is released. |
| 1375 |
> |
* none found or blocker is released |
| 1376 |
|
* |
| 1377 |
|
* @param blocker the blocker |
| 1378 |
|
*/ |
| 1379 |
|
final void helpAsyncBlocker(ManagedBlocker blocker) { |
| 1380 |
< |
int cap, b, d, k; ForkJoinTask<?>[] a; ForkJoinTask<?> t; |
| 1381 |
< |
while (blocker != null && (d = top - (b = base)) > 0 && |
| 1382 |
< |
(a = array) != null && (cap = a.length) > 0 && |
| 1383 |
< |
(((t = getSlot(a, k = (cap - 1) & b)) == null && d > 1) || |
| 1384 |
< |
t instanceof |
| 1385 |
< |
CompletableFuture.AsynchronousCompletionTask) && |
| 1386 |
< |
!blocker.isReleasable()) { |
| 1387 |
< |
if (t != null && base == b++ && casSlotToNull(a, k, t)) { |
| 1388 |
< |
setBaseOpaque(b); |
| 1389 |
< |
t.doExec(); |
| 1380 |
> |
if (blocker != null) { |
| 1381 |
> |
for (;;) { |
| 1382 |
> |
int b = base, cap; ForkJoinTask<?>[] a; |
| 1383 |
> |
if ((a = array) == null || (cap = a.length) <= 0 || b == top) |
| 1384 |
> |
break; |
| 1385 |
> |
int k = (cap - 1) & b, nb = b + 1, nk = (cap - 1) & nb; |
| 1386 |
> |
ForkJoinTask<?> t = a[k]; |
| 1387 |
> |
U.loadFence(); // for re-reads |
| 1388 |
> |
if (base != b) |
| 1389 |
> |
; |
| 1390 |
> |
else if (blocker.isReleasable()) |
| 1391 |
> |
break; |
| 1392 |
> |
else if (a[k] != t) |
| 1393 |
> |
; |
| 1394 |
> |
else if (t != null) { |
| 1395 |
> |
if (!(t instanceof CompletableFuture |
| 1396 |
> |
.AsynchronousCompletionTask)) |
| 1397 |
> |
break; |
| 1398 |
> |
else if (casSlotToNull(a, k, t)) { |
| 1399 |
> |
base = nb; |
| 1400 |
> |
U.storeFence(); |
| 1401 |
> |
t.doExec(); |
| 1402 |
> |
} |
| 1403 |
> |
} |
| 1404 |
> |
else if (a[nk] == null) |
| 1405 |
> |
break; |
| 1406 |
|
} |
| 1407 |
|
} |
| 1408 |
|
} |
| 1409 |
|
|
| 1410 |
|
// misc |
| 1411 |
|
|
| 1245 |
– |
/** AccessControlContext for innocuous workers, created on 1st use. */ |
| 1246 |
– |
@SuppressWarnings("removal") |
| 1247 |
– |
private static AccessControlContext INNOCUOUS_ACC; |
| 1248 |
– |
|
| 1249 |
– |
/** |
| 1250 |
– |
* Initializes (upon registration) InnocuousForkJoinWorkerThreads. |
| 1251 |
– |
*/ |
| 1252 |
– |
@SuppressWarnings("removal") |
| 1253 |
– |
final void initializeInnocuousWorker() { |
| 1254 |
– |
AccessControlContext acc; // racy construction OK |
| 1255 |
– |
if ((acc = INNOCUOUS_ACC) == null) |
| 1256 |
– |
INNOCUOUS_ACC = acc = new AccessControlContext( |
| 1257 |
– |
new ProtectionDomain[] { new ProtectionDomain(null, null) }); |
| 1258 |
– |
Thread t = Thread.currentThread(); |
| 1259 |
– |
ThreadLocalRandom.setInheritedAccessControlContext(t, acc); |
| 1260 |
– |
ThreadLocalRandom.eraseThreadLocals(t); |
| 1261 |
– |
} |
| 1262 |
– |
|
| 1412 |
|
/** |
| 1413 |
|
* Returns true if owned by a worker thread and not known to be blocked. |
| 1414 |
|
*/ |
| 1415 |
|
final boolean isApparentlyUnblocked() { |
| 1416 |
|
Thread wt; Thread.State s; |
| 1417 |
< |
return ((wt = owner) != null && |
| 1417 |
> |
return (access != STOP && (wt = owner) != null && |
| 1418 |
|
(s = wt.getState()) != Thread.State.BLOCKED && |
| 1419 |
|
s != Thread.State.WAITING && |
| 1420 |
|
s != Thread.State.TIMED_WAITING); |
| 1421 |
|
} |
| 1422 |
|
|
| 1423 |
+ |
/** |
| 1424 |
+ |
* Callback from InnocuousForkJoinWorkerThread.onStart |
| 1425 |
+ |
*/ |
| 1426 |
+ |
final void setInnocuous() { |
| 1427 |
+ |
config |= INNOCUOUS; |
| 1428 |
+ |
} |
| 1429 |
+ |
|
| 1430 |
|
static { |
| 1431 |
|
U = Unsafe.getUnsafe(); |
| 1432 |
< |
int scale = U.arrayIndexScale(ForkJoinTask[].class); |
| 1432 |
> |
Class<WorkQueue> klass = WorkQueue.class; |
| 1433 |
> |
ACCESS = U.objectFieldOffset(klass, "access"); |
| 1434 |
> |
PHASE = U.objectFieldOffset(klass, "phase"); |
| 1435 |
> |
Class<ForkJoinTask[]> aklass = ForkJoinTask[].class; |
| 1436 |
> |
ABASE = U.arrayBaseOffset(aklass); |
| 1437 |
> |
int scale = U.arrayIndexScale(aklass); |
| 1438 |
> |
ASHIFT = 31 - Integer.numberOfLeadingZeros(scale); |
| 1439 |
|
if ((scale & (scale - 1)) != 0) |
| 1440 |
|
throw new Error("array index scale not a power of two"); |
| 1279 |
– |
ASHIFT = 31 - Integer.numberOfLeadingZeros(scale); |
| 1280 |
– |
ABASE = U.arrayBaseOffset(ForkJoinTask[].class); |
| 1281 |
– |
SOURCE = U.objectFieldOffset(WorkQueue.class, "source"); |
| 1282 |
– |
BASE = U.objectFieldOffset(WorkQueue.class, "base"); |
| 1441 |
|
} |
| 1442 |
|
} |
| 1443 |
|
|
| 1451 |
|
defaultForkJoinWorkerThreadFactory; |
| 1452 |
|
|
| 1453 |
|
/** |
| 1454 |
< |
* Permission required for callers of methods that may start or |
| 1455 |
< |
* kill threads. |
| 1454 |
> |
* Common (static) pool. Non-null for public use unless a static |
| 1455 |
> |
* construction exception, but internal usages null-check on use |
| 1456 |
> |
* to paranoically avoid potential initialization circularities |
| 1457 |
> |
* as well as to simplify generated code. |
| 1458 |
|
*/ |
| 1459 |
< |
static final RuntimePermission modifyThreadPermission; |
| 1459 |
> |
static final ForkJoinPool common; |
| 1460 |
|
|
| 1461 |
|
/** |
| 1462 |
|
* Sequence number for creating worker names |
| 1463 |
|
*/ |
| 1464 |
|
private static volatile int poolIds; |
| 1465 |
|
|
| 1306 |
– |
// static configuration constants |
| 1307 |
– |
|
| 1308 |
– |
/** |
| 1309 |
– |
* Default idle timeout value (in milliseconds) for the thread |
| 1310 |
– |
* triggering quiescence to park waiting for new work |
| 1311 |
– |
*/ |
| 1312 |
– |
private static final long DEFAULT_KEEPALIVE = 60_000L; |
| 1313 |
– |
|
| 1314 |
– |
/** |
| 1315 |
– |
* Undershoot tolerance for idle timeouts |
| 1316 |
– |
*/ |
| 1317 |
– |
private static final long TIMEOUT_SLOP = 20L; |
| 1318 |
– |
|
| 1466 |
|
/** |
| 1467 |
< |
* The default value for common pool maxSpares. Overridable using |
| 1468 |
< |
* the "java.util.concurrent.ForkJoinPool.common.maximumSpares" |
| 1322 |
< |
* system property. The default value is far in excess of normal |
| 1323 |
< |
* requirements, but also far short of MAX_CAP and typical OS |
| 1324 |
< |
* thread limits, so allows JVMs to catch misuse/abuse before |
| 1325 |
< |
* running out of resources needed to do so. |
| 1326 |
< |
*/ |
| 1327 |
< |
private static final int DEFAULT_COMMON_MAX_SPARES = 256; |
| 1328 |
< |
|
| 1329 |
< |
/* |
| 1330 |
< |
* Bits and masks for field ctl, packed with 4 16 bit subfields: |
| 1331 |
< |
* RC: Number of released (unqueued) workers minus target parallelism |
| 1332 |
< |
* TC: Number of total workers minus target parallelism |
| 1333 |
< |
* SS: version count and status of top waiting thread |
| 1334 |
< |
* ID: poolIndex of top of Treiber stack of waiters |
| 1335 |
< |
* |
| 1336 |
< |
* When convenient, we can extract the lower 32 stack top bits |
| 1337 |
< |
* (including version bits) as sp=(int)ctl. The offsets of counts |
| 1338 |
< |
* by the target parallelism and the positionings of fields makes |
| 1339 |
< |
* it possible to perform the most common checks via sign tests of |
| 1340 |
< |
* fields: When ac is negative, there are not enough unqueued |
| 1341 |
< |
* workers, when tc is negative, there are not enough total |
| 1342 |
< |
* workers. When sp is non-zero, there are waiting workers. To |
| 1343 |
< |
* deal with possibly negative fields, we use casts in and out of |
| 1344 |
< |
* "short" and/or signed shifts to maintain signedness. |
| 1345 |
< |
* |
| 1346 |
< |
* Because it occupies uppermost bits, we can add one release |
| 1347 |
< |
* count using getAndAdd of RC_UNIT, rather than CAS, when |
| 1348 |
< |
* returning from a blocked join. Other updates entail multiple |
| 1349 |
< |
* subfields and masking, requiring CAS. |
| 1350 |
< |
* |
| 1351 |
< |
* The limits packed in field "bounds" are also offset by the |
| 1352 |
< |
* parallelism level to make them comparable to the ctl rc and tc |
| 1353 |
< |
* fields. |
| 1467 |
> |
* Permission required for callers of methods that may start or |
| 1468 |
> |
* kill threads. Lazily constructed. |
| 1469 |
|
*/ |
| 1470 |
+ |
static volatile RuntimePermission modifyThreadPermission; |
| 1471 |
|
|
| 1356 |
– |
// Lower and upper word masks |
| 1357 |
– |
private static final long SP_MASK = 0xffffffffL; |
| 1358 |
– |
private static final long UC_MASK = ~SP_MASK; |
| 1359 |
– |
|
| 1360 |
– |
// Release counts |
| 1361 |
– |
private static final int RC_SHIFT = 48; |
| 1362 |
– |
private static final long RC_UNIT = 0x0001L << RC_SHIFT; |
| 1363 |
– |
private static final long RC_MASK = 0xffffL << RC_SHIFT; |
| 1364 |
– |
|
| 1365 |
– |
// Total counts |
| 1366 |
– |
private static final int TC_SHIFT = 32; |
| 1367 |
– |
private static final long TC_UNIT = 0x0001L << TC_SHIFT; |
| 1368 |
– |
private static final long TC_MASK = 0xffffL << TC_SHIFT; |
| 1369 |
– |
private static final long ADD_WORKER = 0x0001L << (TC_SHIFT + 15); // sign |
| 1472 |
|
|
| 1473 |
|
// Instance fields |
| 1372 |
– |
|
| 1373 |
– |
final long keepAlive; // milliseconds before dropping if idle |
| 1474 |
|
volatile long stealCount; // collects worker nsteals |
| 1475 |
< |
int scanRover; // advances across pollScan calls |
| 1476 |
< |
volatile int threadIds; // for worker thread names |
| 1477 |
< |
final int bounds; // min, max threads packed as shorts |
| 1478 |
< |
volatile int mode; // parallelism, runstate, queue mode |
| 1475 |
> |
volatile long threadIds; // for worker thread names |
| 1476 |
> |
final long keepAlive; // milliseconds before dropping if idle |
| 1477 |
> |
final long bounds; // min, max threads packed as shorts |
| 1478 |
> |
final int config; // static configuration bits |
| 1479 |
> |
volatile int runState; // SHUTDOWN, STOP, TERMINATED bits |
| 1480 |
|
WorkQueue[] queues; // main registry |
| 1481 |
|
final ReentrantLock registrationLock; |
| 1482 |
|
Condition termination; // lazily constructed |
| 1484 |
|
final ForkJoinWorkerThreadFactory factory; |
| 1485 |
|
final UncaughtExceptionHandler ueh; // per-worker UEH |
| 1486 |
|
final Predicate<? super ForkJoinPool> saturate; |
| 1487 |
+ |
// final SharedThreadContainer container; // for loom |
| 1488 |
|
|
| 1489 |
|
@jdk.internal.vm.annotation.Contended("fjpctl") // segregate |
| 1490 |
|
volatile long ctl; // main pool control |
| 1491 |
+ |
@jdk.internal.vm.annotation.Contended("fjpctl") // colocate |
| 1492 |
+ |
int parallelism; // target number of workers |
| 1493 |
|
|
| 1494 |
|
// Support for atomic operations |
| 1495 |
|
private static final Unsafe U; |
| 1496 |
|
private static final long CTL; |
| 1497 |
< |
private static final long MODE; |
| 1497 |
> |
private static final long RUNSTATE; |
| 1498 |
> |
private static final long PARALLELISM; |
| 1499 |
|
private static final long THREADIDS; |
| 1500 |
|
private static final long POOLIDS; |
| 1501 |
+ |
|
| 1502 |
|
private boolean compareAndSetCtl(long c, long v) { |
| 1503 |
|
return U.compareAndSetLong(this, CTL, c, v); |
| 1504 |
|
} |
| 1508 |
|
private long getAndAddCtl(long v) { |
| 1509 |
|
return U.getAndAddLong(this, CTL, v); |
| 1510 |
|
} |
| 1511 |
< |
private int getAndBitwiseOrMode(int v) { |
| 1512 |
< |
return U.getAndBitwiseOrInt(this, MODE, v); |
| 1511 |
> |
private int getAndBitwiseOrRunState(int v) { |
| 1512 |
> |
return U.getAndBitwiseOrInt(this, RUNSTATE, v); |
| 1513 |
|
} |
| 1514 |
< |
private int getAndAddThreadIds(int x) { |
| 1515 |
< |
return U.getAndAddInt(this, THREADIDS, x); |
| 1514 |
> |
private long incrementThreadIds() { |
| 1515 |
> |
return U.getAndAddLong(this, THREADIDS, 1L); |
| 1516 |
|
} |
| 1517 |
|
private static int getAndAddPoolIds(int x) { |
| 1518 |
|
return U.getAndAddInt(ForkJoinPool.class, POOLIDS, x); |
| 1519 |
|
} |
| 1520 |
+ |
private int getAndSetParallelism(int v) { |
| 1521 |
+ |
return U.getAndSetInt(this, PARALLELISM, v); |
| 1522 |
+ |
} |
| 1523 |
+ |
private int getParallelismOpaque() { |
| 1524 |
+ |
return U.getIntOpaque(this, PARALLELISM); |
| 1525 |
+ |
} |
| 1526 |
|
|
| 1527 |
< |
// Creating, registering and deregistering workers |
| 1527 |
> |
// Creating, registering, and deregistering workers |
| 1528 |
|
|
| 1529 |
|
/** |
| 1530 |
|
* Tries to construct and start one worker. Assumes that total |
| 1538 |
|
Throwable ex = null; |
| 1539 |
|
ForkJoinWorkerThread wt = null; |
| 1540 |
|
try { |
| 1541 |
< |
if (fac != null && (wt = fac.newThread(this)) != null) { |
| 1541 |
> |
if (runState >= 0 && // avoid construction if terminating |
| 1542 |
> |
fac != null && (wt = fac.newThread(this)) != null) { |
| 1543 |
|
wt.start(); |
| 1544 |
+ |
// container.start(wt); // for loom |
| 1545 |
|
return true; |
| 1546 |
|
} |
| 1547 |
|
} catch (Throwable rex) { |
| 1556 |
|
*/ |
| 1557 |
|
final String nextWorkerThreadName() { |
| 1558 |
|
String prefix = workerNamePrefix; |
| 1559 |
< |
int tid = getAndAddThreadIds(1) + 1; |
| 1559 |
> |
long tid = incrementThreadIds() + 1L; |
| 1560 |
|
if (prefix == null) // commonPool has no prefix |
| 1561 |
|
prefix = "ForkJoinPool.commonPool-worker-"; |
| 1562 |
< |
return prefix.concat(Integer.toString(tid)); |
| 1562 |
> |
return prefix.concat(Long.toString(tid)); |
| 1563 |
|
} |
| 1564 |
|
|
| 1565 |
|
/** |
| 1568 |
|
* @param w caller's WorkQueue |
| 1569 |
|
*/ |
| 1570 |
|
final void registerWorker(WorkQueue w) { |
| 1457 |
– |
ReentrantLock lock = registrationLock; |
| 1571 |
|
ThreadLocalRandom.localInit(); |
| 1572 |
|
int seed = ThreadLocalRandom.getProbe(); |
| 1573 |
+ |
ReentrantLock lock = registrationLock; |
| 1574 |
+ |
int cfg = config & FIFO; |
| 1575 |
|
if (w != null && lock != null) { |
| 1461 |
– |
int modebits = (mode & FIFO) | w.config; |
| 1576 |
|
w.array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; |
| 1577 |
< |
w.stackPred = seed; // stash for runWorker |
| 1578 |
< |
if ((modebits & INNOCUOUS) != 0) |
| 1465 |
< |
w.initializeInnocuousWorker(); |
| 1577 |
> |
cfg |= w.config | SRC; |
| 1578 |
> |
w.stackPred = seed; |
| 1579 |
|
int id = (seed << 1) | 1; // initial index guess |
| 1580 |
|
lock.lock(); |
| 1581 |
|
try { |
| 1585 |
|
for (; qs[id &= m] != null && k > 0; id -= 2, k -= 2); |
| 1586 |
|
if (k == 0) |
| 1587 |
|
id = n | 1; // resize below |
| 1588 |
< |
w.phase = w.config = id | modebits; // now publishable |
| 1588 |
> |
w.phase = w.config = id | cfg; // now publishable |
| 1589 |
|
|
| 1590 |
|
if (id < n) |
| 1591 |
|
qs[id] = w; |
| 1600 |
|
if ((q = qs[j]) != null) // shared queues may move |
| 1601 |
|
as[q.config & am] = q; |
| 1602 |
|
} |
| 1603 |
< |
VarHandle.releaseFence(); // fill before publish |
| 1603 |
> |
U.storeFence(); // fill before publish |
| 1604 |
|
queues = as; |
| 1605 |
|
} |
| 1606 |
|
} |
| 1620 |
|
* @param ex the exception causing failure, or null if none |
| 1621 |
|
*/ |
| 1622 |
|
final void deregisterWorker(ForkJoinWorkerThread wt, Throwable ex) { |
| 1623 |
< |
ReentrantLock lock = registrationLock; |
| 1624 |
< |
WorkQueue w = null; |
| 1625 |
< |
int cfg = 0; |
| 1626 |
< |
if (wt != null && (w = wt.workQueue) != null && lock != null) { |
| 1627 |
< |
WorkQueue[] qs; int n, i; |
| 1628 |
< |
cfg = w.config; |
| 1623 |
> |
WorkQueue w = (wt == null) ? null : wt.workQueue; |
| 1624 |
> |
int cfg = (w == null) ? 0 : w.config; |
| 1625 |
> |
long c = ctl; |
| 1626 |
> |
if ((cfg & TRIMMED) == 0) // decrement counts |
| 1627 |
> |
do {} while (c != (c = compareAndExchangeCtl( |
| 1628 |
> |
c, ((RC_MASK & (c - RC_UNIT)) | |
| 1629 |
> |
(TC_MASK & (c - TC_UNIT)) | |
| 1630 |
> |
(SP_MASK & c))))); |
| 1631 |
> |
else if ((int)c == 0) // was dropped on timeout |
| 1632 |
> |
cfg &= ~SRC; // suppress signal if last |
| 1633 |
> |
if (!tryTerminate(false, false) && w != null) { |
| 1634 |
> |
ReentrantLock lock; WorkQueue[] qs; int n, i; |
| 1635 |
|
long ns = w.nsteals & 0xffffffffL; |
| 1636 |
< |
lock.lock(); // remove index from array |
| 1637 |
< |
if ((qs = queues) != null && (n = qs.length) > 0 && |
| 1638 |
< |
qs[i = cfg & (n - 1)] == w) |
| 1639 |
< |
qs[i] = null; |
| 1640 |
< |
stealCount += ns; // accumulate steals |
| 1641 |
< |
lock.unlock(); |
| 1642 |
< |
long c = ctl; |
| 1643 |
< |
if ((cfg & QUIET) == 0) // unless self-signalled, decrement counts |
| 1644 |
< |
do {} while (c != (c = compareAndExchangeCtl( |
| 1645 |
< |
c, ((RC_MASK & (c - RC_UNIT)) | |
| 1527 |
< |
(TC_MASK & (c - TC_UNIT)) | |
| 1528 |
< |
(SP_MASK & c))))); |
| 1529 |
< |
else if ((int)c == 0) // was dropped on timeout |
| 1530 |
< |
cfg = 0; // suppress signal if last |
| 1531 |
< |
for (ForkJoinTask<?> t; (t = w.pop()) != null; ) |
| 1532 |
< |
ForkJoinTask.cancelIgnoringExceptions(t); // cancel tasks |
| 1636 |
> |
if ((lock = registrationLock) != null) { |
| 1637 |
> |
lock.lock(); // remove index unless terminating |
| 1638 |
> |
if ((qs = queues) != null && (n = qs.length) > 0 && |
| 1639 |
> |
qs[i = cfg & (n - 1)] == w) |
| 1640 |
> |
qs[i] = null; |
| 1641 |
> |
stealCount += ns; // accumulate steals |
| 1642 |
> |
lock.unlock(); |
| 1643 |
> |
} |
| 1644 |
> |
if ((cfg & SRC) != 0) |
| 1645 |
> |
signalWork(); // possibly replace worker |
| 1646 |
|
} |
| 1647 |
< |
|
| 1648 |
< |
if (!tryTerminate(false, false) && w != null && (cfg & SRC) != 0) |
| 1649 |
< |
signalWork(); // possibly replace worker |
| 1650 |
< |
if (ex != null) |
| 1647 |
> |
if (ex != null) { |
| 1648 |
> |
if (w != null) { |
| 1649 |
> |
w.access = STOP; // cancel tasks |
| 1650 |
> |
for (ForkJoinTask<?> t; (t = w.nextLocalTask(0)) != null; ) |
| 1651 |
> |
ForkJoinTask.cancelIgnoringExceptions(t); |
| 1652 |
> |
} |
| 1653 |
|
ForkJoinTask.rethrow(ex); |
| 1654 |
+ |
} |
| 1655 |
|
} |
| 1656 |
|
|
| 1657 |
|
/* |
| 1658 |
< |
* Tries to create or release a worker if too few are running. |
| 1658 |
> |
* Releases an idle worker, or creates one if not enough exist. |
| 1659 |
|
*/ |
| 1660 |
|
final void signalWork() { |
| 1661 |
< |
for (long c = ctl; c < 0L;) { |
| 1662 |
< |
int sp, i; WorkQueue[] qs; WorkQueue v; |
| 1663 |
< |
if ((sp = (int)c & ~UNSIGNALLED) == 0) { // no idle workers |
| 1664 |
< |
if ((c & ADD_WORKER) == 0L) // enough total workers |
| 1661 |
> |
int pc = parallelism, n; |
| 1662 |
> |
long c = ctl; |
| 1663 |
> |
WorkQueue[] qs = queues; |
| 1664 |
> |
if ((short)(c >>> RC_SHIFT) < pc && qs != null && (n = qs.length) > 0) { |
| 1665 |
> |
for (;;) { |
| 1666 |
> |
boolean create = false; |
| 1667 |
> |
int sp = (int)c & ~INACTIVE; |
| 1668 |
> |
WorkQueue v = qs[sp & (n - 1)]; |
| 1669 |
> |
int deficit = pc - (short)(c >>> TC_SHIFT); |
| 1670 |
> |
long ac = (c + RC_UNIT) & RC_MASK, nc; |
| 1671 |
> |
if (sp != 0 && v != null) |
| 1672 |
> |
nc = (v.stackPred & SP_MASK) | (c & TC_MASK); |
| 1673 |
> |
else if (deficit <= 0) |
| 1674 |
|
break; |
| 1675 |
< |
if (c == (c = compareAndExchangeCtl( |
| 1676 |
< |
c, ((RC_MASK & (c + RC_UNIT)) | |
| 1677 |
< |
(TC_MASK & (c + TC_UNIT)))))) { |
| 1678 |
< |
createWorker(); |
| 1675 |
> |
else { |
| 1676 |
> |
create = true; |
| 1677 |
> |
nc = ((c + TC_UNIT) & TC_MASK); |
| 1678 |
> |
} |
| 1679 |
> |
if (c == (c = compareAndExchangeCtl(c, nc | ac))) { |
| 1680 |
> |
if (create) |
| 1681 |
> |
createWorker(); |
| 1682 |
> |
else { |
| 1683 |
> |
Thread owner = v.owner; |
| 1684 |
> |
v.phase = sp; |
| 1685 |
> |
if (v.access == PARKED) |
| 1686 |
> |
LockSupport.unpark(owner); |
| 1687 |
> |
} |
| 1688 |
|
break; |
| 1689 |
|
} |
| 1690 |
|
} |
| 1691 |
< |
else if ((qs = queues) == null) |
| 1692 |
< |
break; // unstarted/terminated |
| 1693 |
< |
else if (qs.length <= (i = sp & SMASK)) |
| 1694 |
< |
break; // terminated |
| 1695 |
< |
else if ((v = qs[i]) == null) |
| 1696 |
< |
break; // terminating |
| 1697 |
< |
else { |
| 1698 |
< |
long nc = (v.stackPred & SP_MASK) | (UC_MASK & (c + RC_UNIT)); |
| 1699 |
< |
Thread vt = v.owner; |
| 1700 |
< |
if (c == (c = compareAndExchangeCtl(c, nc))) { |
| 1701 |
< |
v.phase = sp; |
| 1702 |
< |
LockSupport.unpark(vt); // release idle worker |
| 1691 |
> |
} |
| 1692 |
> |
} |
| 1693 |
> |
|
| 1694 |
> |
/** |
| 1695 |
> |
* Reactivates any idle worker, if one exists. |
| 1696 |
> |
* |
| 1697 |
> |
* @return the signalled worker, or null if none |
| 1698 |
> |
*/ |
| 1699 |
> |
private WorkQueue reactivate() { |
| 1700 |
> |
WorkQueue[] qs; int n; |
| 1701 |
> |
long c = ctl; |
| 1702 |
> |
if ((qs = queues) != null && (n = qs.length) > 0) { |
| 1703 |
> |
for (;;) { |
| 1704 |
> |
int sp = (int)c & ~INACTIVE; |
| 1705 |
> |
WorkQueue v = qs[sp & (n - 1)]; |
| 1706 |
> |
long ac = UC_MASK & (c + RC_UNIT); |
| 1707 |
> |
if (sp == 0 || v == null) |
| 1708 |
|
break; |
| 1709 |
+ |
if (c == (c = compareAndExchangeCtl( |
| 1710 |
+ |
c, (v.stackPred & SP_MASK) | ac))) { |
| 1711 |
+ |
Thread owner = v.owner; |
| 1712 |
+ |
v.phase = sp; |
| 1713 |
+ |
if (v.access == PARKED) |
| 1714 |
+ |
LockSupport.unpark(owner); |
| 1715 |
+ |
return v; |
| 1716 |
|
} |
| 1717 |
|
} |
| 1718 |
|
} |
| 1719 |
+ |
return null; |
| 1720 |
+ |
} |
| 1721 |
+ |
|
| 1722 |
+ |
/** |
| 1723 |
+ |
* Tries to deactivate worker w; called only on idle timeout. |
| 1724 |
+ |
*/ |
| 1725 |
+ |
private boolean tryTrim(WorkQueue w) { |
| 1726 |
+ |
if (w != null) { |
| 1727 |
+ |
int pred = w.stackPred, cfg = w.config | TRIMMED; |
| 1728 |
+ |
long c = ctl; |
| 1729 |
+ |
int sp = (int)c & ~INACTIVE; |
| 1730 |
+ |
if ((sp & SMASK) == (cfg & SMASK) && |
| 1731 |
+ |
compareAndSetCtl(c, ((pred & SP_MASK) | |
| 1732 |
+ |
(UC_MASK & (c - TC_UNIT))))) { |
| 1733 |
+ |
w.config = cfg; // add sentinel for deregisterWorker |
| 1734 |
+ |
w.phase = sp; |
| 1735 |
+ |
return true; |
| 1736 |
+ |
} |
| 1737 |
+ |
} |
| 1738 |
+ |
return false; |
| 1739 |
+ |
} |
| 1740 |
+ |
|
| 1741 |
+ |
/** |
| 1742 |
+ |
* Returns true if any submission queue is detectably nonempty. |
| 1743 |
+ |
* Accurate only when workers are quiescent; else conservatively |
| 1744 |
+ |
* approximate. |
| 1745 |
+ |
*/ |
| 1746 |
+ |
private boolean hasSubmissions() { |
| 1747 |
+ |
WorkQueue[] qs; WorkQueue q; |
| 1748 |
+ |
int n = ((qs = queues) == null) ? 0 : qs.length; |
| 1749 |
+ |
for (int i = 0; i < n; i += 2) { |
| 1750 |
+ |
if ((q = qs[i]) != null && (q.access > 0 || q.top - q.base > 0)) |
| 1751 |
+ |
return true; |
| 1752 |
+ |
} |
| 1753 |
+ |
return false; |
| 1754 |
|
} |
| 1755 |
|
|
| 1756 |
|
/** |
| 1760 |
|
* @param w caller's WorkQueue (may be null on failed initialization) |
| 1761 |
|
*/ |
| 1762 |
|
final void runWorker(WorkQueue w) { |
| 1763 |
< |
if (mode >= 0 && w != null) { // skip on failed init |
| 1583 |
< |
w.config |= SRC; // mark as valid source |
| 1763 |
> |
if (w != null) { // skip on failed init |
| 1764 |
|
int r = w.stackPred, src = 0; // use seed from registerWorker |
| 1765 |
|
do { |
| 1766 |
|
r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // xorshift |
| 1767 |
|
} while ((src = scan(w, src, r)) >= 0 || |
| 1768 |
|
(src = awaitWork(w)) == 0); |
| 1769 |
+ |
w.access = STOP; // record normal termination |
| 1770 |
|
} |
| 1771 |
|
} |
| 1772 |
|
|
| 1773 |
|
/** |
| 1774 |
|
* Scans for and if found executes top-level tasks: Tries to poll |
| 1775 |
|
* each queue starting at a random index with random stride, |
| 1776 |
< |
* returning source id or retry indicator if contended or |
| 1596 |
< |
* inconsistent. |
| 1776 |
> |
* returning source id or retry indicator. |
| 1777 |
|
* |
| 1778 |
|
* @param w caller's WorkQueue |
| 1779 |
|
* @param prevSrc the previous queue stolen from in current phase, or 0 |
| 1784 |
|
WorkQueue[] qs = queues; |
| 1785 |
|
int n = (w == null || qs == null) ? 0 : qs.length; |
| 1786 |
|
for (int step = (r >>> 16) | 1, i = n; i > 0; --i, r += step) { |
| 1787 |
< |
int j, cap, b; WorkQueue q; ForkJoinTask<?>[] a; |
| 1788 |
< |
if ((q = qs[j = r & (n - 1)]) != null && // poll at qs[j].array[k] |
| 1787 |
> |
int j, cap; WorkQueue q; ForkJoinTask<?>[] a; |
| 1788 |
> |
if ((q = qs[j = r & (n - 1)]) != null && |
| 1789 |
|
(a = q.array) != null && (cap = a.length) > 0) { |
| 1790 |
< |
int k = (cap - 1) & (b = q.base), nextBase = b + 1; |
| 1791 |
< |
int nextIndex = (cap - 1) & nextBase, src = j | SRC; |
| 1792 |
< |
ForkJoinTask<?> t = WorkQueue.getSlot(a, k); |
| 1790 |
> |
int src = j | SRC, b = q.base; |
| 1791 |
> |
int k = (cap - 1) & b, nb = b + 1, nk = (cap - 1) & nb; |
| 1792 |
> |
ForkJoinTask<?> t = a[k]; |
| 1793 |
> |
U.loadFence(); // for re-reads |
| 1794 |
|
if (q.base != b) // inconsistent |
| 1795 |
|
return prevSrc; |
| 1796 |
|
else if (t != null && WorkQueue.casSlotToNull(a, k, t)) { |
| 1797 |
< |
q.base = nextBase; |
| 1798 |
< |
ForkJoinTask<?> next = a[nextIndex]; |
| 1799 |
< |
if ((w.source = src) != prevSrc && next != null) |
| 1797 |
> |
q.base = nb; |
| 1798 |
> |
w.source = src; |
| 1799 |
> |
if (prevSrc == 0 && q.base == nb && a[nk] != null) |
| 1800 |
|
signalWork(); // propagate |
| 1801 |
|
w.topLevelExec(t, q); |
| 1802 |
|
return src; |
| 1803 |
|
} |
| 1804 |
< |
else if (a[nextIndex] != null) // revisit |
| 1805 |
< |
return prevSrc; |
| 1804 |
> |
else if (q.array != a || a[k] != null || a[nk] != null) |
| 1805 |
> |
return prevSrc; // revisit |
| 1806 |
|
} |
| 1807 |
|
} |
| 1808 |
< |
return (queues != qs) ? prevSrc: -1; // possibly resized |
| 1808 |
> |
return -1; |
| 1809 |
|
} |
| 1810 |
|
|
| 1811 |
|
/** |
| 1812 |
< |
* Advances worker phase, pushes onto ctl stack, and awaits signal |
| 1632 |
< |
* or reports termination. |
| 1812 |
> |
* Advances phase, enqueues, and awaits signal or termination. |
| 1813 |
|
* |
| 1814 |
|
* @return negative if terminated, else 0 |
| 1815 |
|
*/ |
| 1816 |
|
private int awaitWork(WorkQueue w) { |
| 1817 |
< |
int phase = (w.phase + SS_SEQ) & ~UNSIGNALLED; |
| 1818 |
< |
w.phase = phase | UNSIGNALLED; // advance phase |
| 1819 |
< |
long prevCtl = ctl, c; // enqueue |
| 1820 |
< |
do { |
| 1821 |
< |
w.stackPred = (int)prevCtl; |
| 1822 |
< |
c = ((prevCtl - RC_UNIT) & UC_MASK) | (phase & SP_MASK); |
| 1823 |
< |
} while (prevCtl != (prevCtl = compareAndExchangeCtl(prevCtl, c))); |
| 1824 |
< |
long initialCtl = c; |
| 1825 |
< |
LockSupport.setCurrentBlocker(this); // prepare to block (exit also OK) |
| 1826 |
< |
long deadline = 0L; // nonzero for timed wait |
| 1827 |
< |
boolean checkInterrupt = false; // alternate with park calls |
| 1828 |
< |
boolean terminate = false; |
| 1829 |
< |
for (int md;;) { |
| 1830 |
< |
if (terminate = ((md = mode) < 0)) |
| 1831 |
< |
break; |
| 1832 |
< |
else if (w.phase >= 0) |
| 1833 |
< |
break; |
| 1834 |
< |
else if (checkInterrupt = !checkInterrupt) |
| 1835 |
< |
Thread.interrupted(); |
| 1836 |
< |
else if ((c = ctl) == prevCtl) |
| 1837 |
< |
Thread.onSpinWait(); // signal in progress |
| 1838 |
< |
else if ((md & SMASK) + (int)(c >> RC_SHIFT) > 0) |
| 1839 |
< |
LockSupport.park(); // not the only idle worker |
| 1840 |
< |
else if (!canStop()) { // check racing submissions |
| 1841 |
< |
if (ctl == initialCtl && |
| 1842 |
< |
compareAndSetCtl(initialCtl, prevCtl)) |
| 1843 |
< |
w.phase = phase; // self-signal |
| 1844 |
< |
} |
| 1845 |
< |
else if (terminate = tryTerminate(false, false)) |
| 1846 |
< |
break; |
| 1847 |
< |
else { // timed wait or drop |
| 1848 |
< |
long now = System.currentTimeMillis(); |
| 1849 |
< |
if (deadline == 0L && (deadline = now + keepAlive) == 0L) |
| 1850 |
< |
deadline = 1L; // avoid 0 |
| 1851 |
< |
if (ctl != c) |
| 1852 |
< |
; // stale |
| 1853 |
< |
else if (deadline - now > TIMEOUT_SLOP) |
| 1674 |
< |
LockSupport.parkUntil(deadline); |
| 1675 |
< |
else if (((int)c & SMASK) == (w.config & SMASK) && |
| 1676 |
< |
compareAndSetCtl(c, ((UC_MASK & (c - TC_UNIT)) | |
| 1677 |
< |
(prevCtl & SP_MASK)))) { |
| 1678 |
< |
w.config |= QUIET; // sentinel for deregisterWorker |
| 1679 |
< |
terminate = true; // drop on timeout |
| 1817 |
> |
if (w == null) |
| 1818 |
> |
return -1; // currently impossible |
| 1819 |
> |
int p = (w.phase + SS_SEQ) & ~INACTIVE; // advance phase |
| 1820 |
> |
boolean idle = false; // true if possibly quiescent |
| 1821 |
> |
if (runState < 0) |
| 1822 |
> |
return -1; // terminating |
| 1823 |
> |
long sp = p & SP_MASK, pc = ctl, qc; |
| 1824 |
> |
w.phase = p | INACTIVE; |
| 1825 |
> |
do { // enqueue |
| 1826 |
> |
w.stackPred = (int)pc; // set ctl stack link |
| 1827 |
> |
} while (pc != (pc = compareAndExchangeCtl( |
| 1828 |
> |
pc, qc = ((pc - RC_UNIT) & UC_MASK) | sp))); |
| 1829 |
> |
if ((qc & RC_MASK) <= 0L) |
| 1830 |
> |
idle = true; |
| 1831 |
> |
WorkQueue[] qs = queues; // to spin for expected #accesses in scan+signal |
| 1832 |
> |
int spins = ((qs == null) ? 0 : ((qs.length & SMASK) << 1)) + 4, rs; |
| 1833 |
> |
if (idle && hasSubmissions() && w.phase < 0) |
| 1834 |
> |
reactivate(); // check for stragglers |
| 1835 |
> |
if ((rs = runState) < 0 || |
| 1836 |
> |
(rs != 0 && idle && tryTerminate(false, false))) |
| 1837 |
> |
return -1; // quiescent termination |
| 1838 |
> |
while ((p = w.phase) < 0 && --spins > 0) |
| 1839 |
> |
Thread.onSpinWait(); // spin before block |
| 1840 |
> |
if (p < 0) { // await signal |
| 1841 |
> |
long deadline = (idle) ? keepAlive + System.currentTimeMillis() : 0L; |
| 1842 |
> |
LockSupport.setCurrentBlocker(this); |
| 1843 |
> |
for (;;) { |
| 1844 |
> |
w.access = PARKED; // enable unpark |
| 1845 |
> |
if (w.phase < 0) { |
| 1846 |
> |
if (idle) |
| 1847 |
> |
LockSupport.parkUntil(deadline); |
| 1848 |
> |
else |
| 1849 |
> |
LockSupport.park(); |
| 1850 |
> |
} |
| 1851 |
> |
w.access = 0; // disable unpark |
| 1852 |
> |
if (w.phase >= 0) { |
| 1853 |
> |
LockSupport.setCurrentBlocker(null); |
| 1854 |
|
break; |
| 1855 |
|
} |
| 1856 |
< |
else |
| 1857 |
< |
deadline = 0L; // not at head; restart timeout |
| 1856 |
> |
Thread.interrupted(); // clear status for next park |
| 1857 |
> |
if (idle) { // check for idle timeout |
| 1858 |
> |
if (deadline - System.currentTimeMillis() < TIMEOUT_SLOP) { |
| 1859 |
> |
if (tryTrim(w)) |
| 1860 |
> |
return -1; |
| 1861 |
> |
else // not at head; restart timer |
| 1862 |
> |
deadline += keepAlive; |
| 1863 |
> |
} |
| 1864 |
> |
} |
| 1865 |
|
} |
| 1866 |
|
} |
| 1867 |
< |
LockSupport.setCurrentBlocker(null); |
| 1687 |
< |
return terminate ? -1 : 0; |
| 1867 |
> |
return (runState < 0) ? -1 : 0; |
| 1868 |
|
} |
| 1869 |
|
|
| 1870 |
|
/** |
| 1871 |
< |
* Returns true if can start terminating if enabled, or already terminated |
| 1871 |
> |
* Non-overridable version of isQuiescent. Returns true if |
| 1872 |
> |
* quiescent or already terminating. |
| 1873 |
|
*/ |
| 1874 |
|
private boolean canStop() { |
| 1875 |
< |
outer: for (long c = ctl, oldSum = 0L;;) { // repeat until stable |
| 1876 |
< |
int md; WorkQueue[] qs; |
| 1877 |
< |
if ((qs = queues) == null || ((md = mode) & STOP) != 0) |
| 1878 |
< |
return true; |
| 1879 |
< |
if ((md & SMASK) + (int)(c >> RC_SHIFT) > 0) |
| 1880 |
< |
break; // active workers |
| 1881 |
< |
long checkSum = 0L; |
| 1882 |
< |
boolean rescan = false; |
| 1883 |
< |
for (int i = 0; i < qs.length; i += 2) { // scan racing submissions |
| 1884 |
< |
WorkQueue q; ForkJoinTask<?>[] a; int s = 0, cap; |
| 1885 |
< |
if ((q = qs[i]) != null) { |
| 1886 |
< |
if ((a = q.array) != null && (cap = a.length) > 0 && |
| 1887 |
< |
((s = q.top) != q.base || a[(cap - 1) & s] != null)) |
| 1888 |
< |
break outer; |
| 1889 |
< |
else if (q.source != 0) |
| 1890 |
< |
rescan = true; // retry if locked |
| 1891 |
< |
} |
| 1892 |
< |
checkSum += (((long)i) << 32) ^ s; |
| 1875 |
> |
long c = ctl; |
| 1876 |
> |
do { |
| 1877 |
> |
if (runState < 0) |
| 1878 |
> |
break; |
| 1879 |
> |
if ((c & RC_MASK) > 0L || hasSubmissions()) |
| 1880 |
> |
return false; |
| 1881 |
> |
} while (c != (c = ctl)); // validate |
| 1882 |
> |
return true; |
| 1883 |
> |
} |
| 1884 |
> |
|
| 1885 |
> |
/** |
| 1886 |
> |
* Scans for and returns a polled task, if available. Used only |
| 1887 |
> |
* for untracked polls. Begins scan at a random index to avoid |
| 1888 |
> |
* systematic unfairness. |
| 1889 |
> |
* |
| 1890 |
> |
* @param submissionsOnly if true, only scan submission queues |
| 1891 |
> |
*/ |
| 1892 |
> |
private ForkJoinTask<?> pollScan(boolean submissionsOnly) { |
| 1893 |
> |
int r = ThreadLocalRandom.nextSecondarySeed(); |
| 1894 |
> |
if (submissionsOnly) // even indices only |
| 1895 |
> |
r &= ~1; |
| 1896 |
> |
int step = (submissionsOnly) ? 2 : 1; |
| 1897 |
> |
WorkQueue[] qs; int n; WorkQueue q; ForkJoinTask<?> t; |
| 1898 |
> |
if (runState >= 0 && (qs = queues) != null && (n = qs.length) > 0) { |
| 1899 |
> |
for (int i = n; i > 0; i -= step, r += step) { |
| 1900 |
> |
if ((q = qs[r & (n - 1)]) != null && |
| 1901 |
> |
(t = q.poll(this)) != null) |
| 1902 |
> |
return t; |
| 1903 |
|
} |
| 1713 |
– |
if (c == (c = ctl) && checkSum == oldSum && !rescan && queues == qs) |
| 1714 |
– |
return true; |
| 1715 |
– |
oldSum = checkSum; |
| 1904 |
|
} |
| 1905 |
< |
return (mode & STOP) != 0; // recheck mode on false return |
| 1905 |
> |
return null; |
| 1906 |
|
} |
| 1907 |
|
|
| 1908 |
|
/** |
| 1914 |
|
* unblocked. |
| 1915 |
|
* |
| 1916 |
|
* @param c incoming ctl value |
| 1917 |
+ |
* @param canSaturate to override saturate predicate |
| 1918 |
|
* @return UNCOMPENSATE: block then adjust, 0: block, -1 : retry |
| 1919 |
|
*/ |
| 1920 |
< |
private int tryCompensate(long c) { |
| 1920 |
> |
private int tryCompensate(long c, boolean canSaturate) { |
| 1921 |
|
Predicate<? super ForkJoinPool> sat; |
| 1922 |
< |
int md = mode, b = bounds; |
| 1923 |
< |
// counts are signed; centered at parallelism level == 0 |
| 1922 |
> |
long b = bounds; // unpack fields |
| 1923 |
> |
int pc = parallelism; |
| 1924 |
|
int minActive = (short)(b & SMASK), |
| 1925 |
< |
maxTotal = b >>> SWIDTH, |
| 1926 |
< |
active = (int)(c >> RC_SHIFT), |
| 1925 |
> |
maxTotal = (short)(b >>> SWIDTH) + pc, |
| 1926 |
> |
active = (short)(c >>> RC_SHIFT), |
| 1927 |
|
total = (short)(c >>> TC_SHIFT), |
| 1928 |
< |
sp = (int)c & ~UNSIGNALLED; |
| 1929 |
< |
if ((md & SMASK) == 0) |
| 1930 |
< |
return 0; // cannot compensate if parallelism zero |
| 1931 |
< |
else if (total >= 0) { |
| 1932 |
< |
if (sp != 0) { // activate idle worker |
| 1933 |
< |
WorkQueue[] qs; int n; WorkQueue v; |
| 1934 |
< |
if ((qs = queues) != null && (n = qs.length) > 0 && |
| 1935 |
< |
(v = qs[sp & (n - 1)]) != null) { |
| 1936 |
< |
Thread vt = v.owner; |
| 1937 |
< |
long nc = ((long)v.stackPred & SP_MASK) | (UC_MASK & c); |
| 1938 |
< |
if (compareAndSetCtl(c, nc)) { |
| 1939 |
< |
v.phase = sp; |
| 1751 |
< |
LockSupport.unpark(vt); |
| 1752 |
< |
return UNCOMPENSATE; |
| 1753 |
< |
} |
| 1928 |
> |
sp = (int)c & ~INACTIVE; |
| 1929 |
> |
if (runState < 0) // terminating |
| 1930 |
> |
return -1; |
| 1931 |
> |
else if (sp != 0 && active <= pc) { // activate idle worker |
| 1932 |
> |
WorkQueue[] qs; WorkQueue v; int i; |
| 1933 |
> |
if (ctl == c && (qs = queues) != null && |
| 1934 |
> |
qs.length > (i = sp & SMASK) && (v = qs[i]) != null) { |
| 1935 |
> |
long nc = (v.stackPred & SP_MASK) | (UC_MASK & c); |
| 1936 |
> |
if (compareAndSetCtl(c, nc)) { |
| 1937 |
> |
v.phase = sp; |
| 1938 |
> |
LockSupport.unpark(v.owner); |
| 1939 |
> |
return UNCOMPENSATE; |
| 1940 |
|
} |
| 1755 |
– |
return -1; // retry |
| 1756 |
– |
} |
| 1757 |
– |
else if (active > minActive) { // reduce parallelism |
| 1758 |
– |
long nc = ((RC_MASK & (c - RC_UNIT)) | (~RC_MASK & c)); |
| 1759 |
– |
return compareAndSetCtl(c, nc) ? UNCOMPENSATE : -1; |
| 1941 |
|
} |
| 1942 |
+ |
return -1; // retry |
| 1943 |
+ |
} |
| 1944 |
+ |
else if (active > minActive && total >= pc) { // reduce active workers |
| 1945 |
+ |
long nc = ((RC_MASK & (c - RC_UNIT)) | (~RC_MASK & c)); |
| 1946 |
+ |
return compareAndSetCtl(c, nc) ? UNCOMPENSATE : -1; |
| 1947 |
|
} |
| 1948 |
< |
if (total < maxTotal) { // expand pool |
| 1948 |
> |
else if (total < maxTotal && total < MAX_CAP) { // expand pool |
| 1949 |
|
long nc = ((c + TC_UNIT) & TC_MASK) | (c & ~TC_MASK); |
| 1950 |
|
return (!compareAndSetCtl(c, nc) ? -1 : |
| 1951 |
|
!createWorker() ? 0 : UNCOMPENSATE); |
| 1952 |
|
} |
| 1953 |
< |
else if (!compareAndSetCtl(c, c)) // validate |
| 1953 |
> |
else if (!compareAndSetCtl(c, c)) // validate |
| 1954 |
|
return -1; |
| 1955 |
< |
else if ((sat = saturate) != null && sat.test(this)) |
| 1955 |
> |
else if (canSaturate || ((sat = saturate) != null && sat.test(this))) |
| 1956 |
|
return 0; |
| 1957 |
|
else |
| 1958 |
|
throw new RejectedExecutionException( |
| 1967 |
|
} |
| 1968 |
|
|
| 1969 |
|
/** |
| 1970 |
< |
* Helps if possible until the given task is done. Scans other |
| 1971 |
< |
* queues for a task produced by one of w's stealers; returning |
| 1972 |
< |
* compensated blocking sentinel if none are found. |
| 1970 |
> |
* Helps if possible until the given task is done. Processes |
| 1971 |
> |
* compatible local tasks and scans other queues for task produced |
| 1972 |
> |
* by w's stealers; returning compensated blocking sentinel if |
| 1973 |
> |
* none are found. |
| 1974 |
|
* |
| 1975 |
|
* @param task the task |
| 1976 |
|
* @param w caller's WorkQueue |
| 1977 |
< |
* @param canHelp if false, compensate only |
| 1977 |
> |
* @param timed true if this is a timed join |
| 1978 |
|
* @return task status on exit, or UNCOMPENSATE for compensated blocking |
| 1979 |
|
*/ |
| 1980 |
< |
final int helpJoin(ForkJoinTask<?> task, WorkQueue w, boolean canHelp) { |
| 1981 |
< |
int s = 0; |
| 1982 |
< |
if (task != null && w != null) { |
| 1983 |
< |
int wsrc = w.source, wid = w.config & SMASK, r = wid + 2; |
| 1984 |
< |
boolean scan = true; |
| 1985 |
< |
long c = 0L; // track ctl stability |
| 1986 |
< |
outer: for (;;) { |
| 1987 |
< |
if ((s = task.status) < 0) |
| 1988 |
< |
break; |
| 1989 |
< |
else if (scan = !scan) { // previous scan was empty |
| 1990 |
< |
if (mode < 0) |
| 1991 |
< |
ForkJoinTask.cancelIgnoringExceptions(task); |
| 1992 |
< |
else if (c == (c = ctl) && (s = tryCompensate(c)) >= 0) |
| 1993 |
< |
break; // block |
| 1994 |
< |
} |
| 1995 |
< |
else if (canHelp) { // scan for subtasks |
| 1996 |
< |
WorkQueue[] qs = queues; |
| 1997 |
< |
int n = (qs == null) ? 0 : qs.length, m = n - 1; |
| 1998 |
< |
for (int i = n; i > 0; i -= 2, r += 2) { |
| 1999 |
< |
int j; WorkQueue q, x, y; ForkJoinTask<?>[] a; |
| 2000 |
< |
if ((q = qs[j = r & m]) != null) { |
| 2001 |
< |
int sq = q.source & SMASK, cap, b; |
| 2002 |
< |
if ((a = q.array) != null && (cap = a.length) > 0) { |
| 2003 |
< |
int k = (cap - 1) & (b = q.base); |
| 2004 |
< |
int nextBase = b + 1, src = j | SRC, sx; |
| 2005 |
< |
ForkJoinTask<?> t = WorkQueue.getSlot(a, k); |
| 2006 |
< |
boolean eligible = sq == wid || |
| 2007 |
< |
((x = qs[sq & m]) != null && // indirect |
| 2008 |
< |
((sx = (x.source & SMASK)) == wid || |
| 2009 |
< |
((y = qs[sx & m]) != null && // 2-indirect |
| 2010 |
< |
(y.source & SMASK) == wid))); |
| 2011 |
< |
if ((s = task.status) < 0) |
| 2012 |
< |
break outer; |
| 1826 |
< |
else if ((q.source & SMASK) != sq || |
| 1827 |
< |
q.base != b) |
| 1828 |
< |
scan = true; // inconsistent |
| 1829 |
< |
else if (t == null) |
| 1830 |
< |
scan |= (a[nextBase & (cap - 1)] != null || |
| 1831 |
< |
q.top != b); // lagging |
| 1832 |
< |
else if (eligible) { |
| 1833 |
< |
if (WorkQueue.casSlotToNull(a, k, t)) { |
| 1834 |
< |
q.base = nextBase; |
| 1835 |
< |
w.source = src; |
| 1836 |
< |
t.doExec(); |
| 1837 |
< |
w.source = wsrc; |
| 1838 |
< |
} |
| 1839 |
< |
scan = true; |
| 1840 |
< |
break; |
| 1841 |
< |
} |
| 1980 |
> |
final int helpJoin(ForkJoinTask<?> task, WorkQueue w, boolean timed) { |
| 1981 |
> |
if (w == null || task == null) |
| 1982 |
> |
return 0; |
| 1983 |
> |
int wsrc = w.source, wid = (w.config & SMASK) | SRC, r = wid + 2; |
| 1984 |
> |
long sctl = 0L; // track stability |
| 1985 |
> |
for (boolean rescan = true;;) { |
| 1986 |
> |
int s; WorkQueue[] qs; |
| 1987 |
> |
if ((s = task.status) < 0) |
| 1988 |
> |
return s; |
| 1989 |
> |
if (!rescan && sctl == (sctl = ctl) && |
| 1990 |
> |
(s = tryCompensate(sctl, timed)) >= 0) |
| 1991 |
> |
return s; // block |
| 1992 |
> |
rescan = false; |
| 1993 |
> |
if (runState < 0) |
| 1994 |
> |
return 0; |
| 1995 |
> |
int n = ((qs = queues) == null) ? 0 : qs.length, m = n - 1; |
| 1996 |
> |
scan: for (int i = n >>> 1; i > 0; --i, r += 2) { |
| 1997 |
> |
int j, cap; WorkQueue q; ForkJoinTask<?>[] a; |
| 1998 |
> |
if ((q = qs[j = r & m]) != null && (a = q.array) != null && |
| 1999 |
> |
(cap = a.length) > 0) { |
| 2000 |
> |
for (int src = j | SRC;;) { |
| 2001 |
> |
int sq = q.source, b = q.base; |
| 2002 |
> |
int k = (cap - 1) & b, nb = b + 1; |
| 2003 |
> |
ForkJoinTask<?> t = a[k]; |
| 2004 |
> |
U.loadFence(); // for re-reads |
| 2005 |
> |
boolean eligible = true; // check steal chain |
| 2006 |
> |
for (int d = n, v = sq;;) { // may be cyclic; bound |
| 2007 |
> |
WorkQueue p; |
| 2008 |
> |
if (v == wid) |
| 2009 |
> |
break; |
| 2010 |
> |
if (v == 0 || --d == 0 || (p = qs[v & m]) == null) { |
| 2011 |
> |
eligible = false; |
| 2012 |
> |
break; |
| 2013 |
|
} |
| 2014 |
+ |
v = p.source; |
| 2015 |
+ |
} |
| 2016 |
+ |
if (q.source != sq || q.base != b) |
| 2017 |
+ |
; // stale |
| 2018 |
+ |
else if ((s = task.status) < 0) |
| 2019 |
+ |
return s; // recheck before taking |
| 2020 |
+ |
else if (t == null) { |
| 2021 |
+ |
if (a[k] == null) { |
| 2022 |
+ |
if (!rescan && eligible && |
| 2023 |
+ |
(q.array != a || q.top != b)) |
| 2024 |
+ |
rescan = true; // resized or stalled |
| 2025 |
+ |
break; |
| 2026 |
+ |
} |
| 2027 |
+ |
} |
| 2028 |
+ |
else if (t != task && !eligible) |
| 2029 |
+ |
break; |
| 2030 |
+ |
else if (WorkQueue.casSlotToNull(a, k, t)) { |
| 2031 |
+ |
q.base = nb; |
| 2032 |
+ |
w.source = src; |
| 2033 |
+ |
t.doExec(); |
| 2034 |
+ |
w.source = wsrc; |
| 2035 |
+ |
rescan = true; |
| 2036 |
+ |
break scan; |
| 2037 |
|
} |
| 2038 |
|
} |
| 2039 |
|
} |
| 2040 |
|
} |
| 2041 |
|
} |
| 2042 |
< |
return s; |
| 1849 |
< |
} |
| 2042 |
> |
} |
| 2043 |
|
|
| 2044 |
|
/** |
| 2045 |
< |
* Extra helpJoin steps for CountedCompleters. Scans for and runs |
| 1853 |
< |
* subtasks of the given root task, returning if none are found. |
| 2045 |
> |
* Version of helpJoin for CountedCompleters. |
| 2046 |
|
* |
| 2047 |
< |
* @param task root of CountedCompleter computation |
| 2047 |
> |
* @param task the task |
| 2048 |
|
* @param w caller's WorkQueue |
| 2049 |
< |
* @param owned true if owned by a ForkJoinWorkerThread |
| 2050 |
< |
* @return task status on exit |
| 2049 |
> |
* @param owned true if w is owned by a ForkJoinWorkerThread |
| 2050 |
> |
* @param timed true if this is a timed join |
| 2051 |
> |
* @return task status on exit, or UNCOMPENSATE for compensated blocking |
| 2052 |
|
*/ |
| 2053 |
< |
final int helpComplete(ForkJoinTask<?> task, WorkQueue w, boolean owned) { |
| 2054 |
< |
int s = 0; |
| 2055 |
< |
if (task != null && w != null) { |
| 2056 |
< |
int r = w.config; |
| 2057 |
< |
boolean scan = true, locals = true; |
| 2058 |
< |
long c = 0L; |
| 2059 |
< |
outer: for (;;) { |
| 2060 |
< |
if (locals) { // try locals before scanning |
| 2061 |
< |
if ((s = w.helpComplete(task, owned, 0)) < 0) |
| 2062 |
< |
break; |
| 2063 |
< |
locals = false; |
| 2064 |
< |
} |
| 2065 |
< |
else if ((s = task.status) < 0) |
| 2066 |
< |
break; |
| 2067 |
< |
else if (scan = !scan) { |
| 2068 |
< |
if (c == (c = ctl)) |
| 2069 |
< |
break; |
| 2070 |
< |
} |
| 2071 |
< |
else { // scan for subtasks |
| 2072 |
< |
WorkQueue[] qs = queues; |
| 2073 |
< |
int n = (qs == null) ? 0 : qs.length; |
| 2074 |
< |
for (int i = n; i > 0; --i, ++r) { |
| 2075 |
< |
int j, cap, b; WorkQueue q; ForkJoinTask<?>[] a; |
| 2076 |
< |
boolean eligible = false; |
| 2077 |
< |
if ((q = qs[j = r & (n - 1)]) != null && |
| 2078 |
< |
(a = q.array) != null && (cap = a.length) > 0) { |
| 2079 |
< |
int k = (cap - 1) & (b = q.base), nextBase = b + 1; |
| 2080 |
< |
ForkJoinTask<?> t = WorkQueue.getSlot(a, k); |
| 2081 |
< |
if (t instanceof CountedCompleter) { |
| 2082 |
< |
CountedCompleter<?> f = (CountedCompleter<?>)t; |
| 2083 |
< |
do {} while (!(eligible = (f == task)) && |
| 2084 |
< |
(f = f.completer) != null); |
| 2085 |
< |
} |
| 2086 |
< |
if ((s = task.status) < 0) |
| 2087 |
< |
break outer; |
| 2088 |
< |
else if (q.base != b) |
| 2089 |
< |
scan = true; // inconsistent |
| 1897 |
< |
else if (t == null) |
| 1898 |
< |
scan |= (a[nextBase & (cap - 1)] != null || |
| 1899 |
< |
q.top != b); |
| 1900 |
< |
else if (eligible) { |
| 1901 |
< |
if (WorkQueue.casSlotToNull(a, k, t)) { |
| 1902 |
< |
q.setBaseOpaque(nextBase); |
| 1903 |
< |
t.doExec(); |
| 1904 |
< |
locals = true; |
| 1905 |
< |
} |
| 1906 |
< |
scan = true; |
| 2053 |
> |
final int helpComplete(ForkJoinTask<?> task, WorkQueue w, boolean owned, |
| 2054 |
> |
boolean timed) { |
| 2055 |
> |
if (w == null || task == null) |
| 2056 |
> |
return 0; |
| 2057 |
> |
int wsrc = w.source, r = w.config; |
| 2058 |
> |
long sctl = 0L; // track stability |
| 2059 |
> |
for (boolean rescan = true;;) { |
| 2060 |
> |
int s; WorkQueue[] qs; |
| 2061 |
> |
if ((s = w.helpComplete(task, owned, 0)) < 0) |
| 2062 |
> |
return s; |
| 2063 |
> |
if (!rescan && sctl == (sctl = ctl)) { |
| 2064 |
> |
if (!owned) |
| 2065 |
> |
return 0; |
| 2066 |
> |
if ((s = tryCompensate(sctl, timed)) >= 0) |
| 2067 |
> |
return s; |
| 2068 |
> |
} |
| 2069 |
> |
rescan = false; |
| 2070 |
> |
if (runState < 0) |
| 2071 |
> |
return 0; |
| 2072 |
> |
int n = ((qs = queues) == null) ? 0 : qs.length, m = n - 1; |
| 2073 |
> |
scan: for (int i = n; i > 0; --i, ++r) { |
| 2074 |
> |
int j, cap; WorkQueue q; ForkJoinTask<?>[] a; |
| 2075 |
> |
if ((q = qs[j = r & m]) != null && (a = q.array) != null && |
| 2076 |
> |
(cap = a.length) > 0) { |
| 2077 |
> |
poll: for (int src = j | SRC, b = q.base;;) { |
| 2078 |
> |
int k = (cap - 1) & b, nb = b + 1; |
| 2079 |
> |
ForkJoinTask<?> t = a[k]; |
| 2080 |
> |
U.loadFence(); // for re-reads |
| 2081 |
> |
if (b != (b = q.base)) |
| 2082 |
> |
; // stale |
| 2083 |
> |
else if ((s = task.status) < 0) |
| 2084 |
> |
return s; // recheck before taking |
| 2085 |
> |
else if (t == null) { |
| 2086 |
> |
if (a[k] == null) { |
| 2087 |
> |
if (!rescan && // resized or stalled |
| 2088 |
> |
(q.array != a || q.top != b)) |
| 2089 |
> |
rescan = true; |
| 2090 |
|
break; |
| 2091 |
|
} |
| 2092 |
|
} |
| 2093 |
+ |
else if (t instanceof CountedCompleter) { |
| 2094 |
+ |
CountedCompleter<?> f; |
| 2095 |
+ |
for (f = (CountedCompleter<?>)t;;) { |
| 2096 |
+ |
if (f == task) |
| 2097 |
+ |
break; |
| 2098 |
+ |
else if ((f = f.completer) == null) |
| 2099 |
+ |
break poll; // ineligible |
| 2100 |
+ |
} |
| 2101 |
+ |
if (WorkQueue.casSlotToNull(a, k, t)) { |
| 2102 |
+ |
q.base = nb; |
| 2103 |
+ |
w.source = src; |
| 2104 |
+ |
t.doExec(); |
| 2105 |
+ |
w.source = wsrc; |
| 2106 |
+ |
rescan = true; |
| 2107 |
+ |
break scan; |
| 2108 |
+ |
} |
| 2109 |
+ |
} |
| 2110 |
+ |
else |
| 2111 |
+ |
break; |
| 2112 |
|
} |
| 2113 |
|
} |
| 2114 |
|
} |
| 2115 |
|
} |
| 2116 |
< |
return s; |
| 1915 |
< |
} |
| 1916 |
< |
|
| 1917 |
< |
/** |
| 1918 |
< |
* Scans for and returns a polled task, if available. Used only |
| 1919 |
< |
* for untracked polls. Begins scan at an index (scanRover) |
| 1920 |
< |
* advanced on each call, to avoid systematic unfairness. |
| 1921 |
< |
* |
| 1922 |
< |
* @param submissionsOnly if true, only scan submission queues |
| 1923 |
< |
*/ |
| 1924 |
< |
private ForkJoinTask<?> pollScan(boolean submissionsOnly) { |
| 1925 |
< |
VarHandle.acquireFence(); |
| 1926 |
< |
int r = scanRover += 0x61c88647; // Weyl increment; raciness OK |
| 1927 |
< |
if (submissionsOnly) // even indices only |
| 1928 |
< |
r &= ~1; |
| 1929 |
< |
int step = (submissionsOnly) ? 2 : 1; |
| 1930 |
< |
WorkQueue[] qs; int n; |
| 1931 |
< |
while ((qs = queues) != null && (n = qs.length) > 0) { |
| 1932 |
< |
boolean scan = false; |
| 1933 |
< |
for (int i = 0; i < n; i += step) { |
| 1934 |
< |
int j, cap, b; WorkQueue q; ForkJoinTask<?>[] a; |
| 1935 |
< |
if ((q = qs[j = (n - 1) & (r + i)]) != null && |
| 1936 |
< |
(a = q.array) != null && (cap = a.length) > 0) { |
| 1937 |
< |
int k = (cap - 1) & (b = q.base), nextBase = b + 1; |
| 1938 |
< |
ForkJoinTask<?> t = WorkQueue.getSlot(a, k); |
| 1939 |
< |
if (q.base != b) |
| 1940 |
< |
scan = true; |
| 1941 |
< |
else if (t == null) |
| 1942 |
< |
scan |= (q.top != b || a[nextBase & (cap - 1)] != null); |
| 1943 |
< |
else if (!WorkQueue.casSlotToNull(a, k, t)) |
| 1944 |
< |
scan = true; |
| 1945 |
< |
else { |
| 1946 |
< |
q.setBaseOpaque(nextBase); |
| 1947 |
< |
return t; |
| 1948 |
< |
} |
| 1949 |
< |
} |
| 1950 |
< |
} |
| 1951 |
< |
if (!scan && queues == qs) |
| 1952 |
< |
break; |
| 1953 |
< |
} |
| 1954 |
< |
return null; |
| 1955 |
< |
} |
| 2116 |
> |
} |
| 2117 |
|
|
| 2118 |
|
/** |
| 2119 |
|
* Runs tasks until {@code isQuiescent()}. Rather than blocking |
| 2124 |
|
* @param interruptible true if return on interrupt |
| 2125 |
|
* @return positive if quiescent, negative if interrupted, else 0 |
| 2126 |
|
*/ |
| 2127 |
< |
final int helpQuiescePool(WorkQueue w, long nanos, boolean interruptible) { |
| 1967 |
< |
if (w == null) |
| 1968 |
< |
return 0; |
| 2127 |
> |
private int helpQuiesce(WorkQueue w, long nanos, boolean interruptible) { |
| 2128 |
|
long startTime = System.nanoTime(), parkTime = 0L; |
| 2129 |
< |
int prevSrc = w.source, wsrc = prevSrc, cfg = w.config, r = cfg + 1; |
| 2130 |
< |
for (boolean active = true, locals = true;;) { |
| 2131 |
< |
boolean busy = false, scan = false; |
| 2132 |
< |
if (locals) { // run local tasks before (re)polling |
| 2133 |
< |
locals = false; |
| 2134 |
< |
for (ForkJoinTask<?> u; (u = w.nextLocalTask(cfg)) != null;) |
| 2129 |
> |
int phase; // w.phase set negative when temporarily quiescent |
| 2130 |
> |
if (w == null || (phase = w.phase) < 0) |
| 2131 |
> |
return 0; |
| 2132 |
> |
int activePhase = phase, inactivePhase = phase | INACTIVE; |
| 2133 |
> |
int wsrc = w.source, r = 0; |
| 2134 |
> |
for (boolean locals = true;;) { |
| 2135 |
> |
WorkQueue[] qs; WorkQueue q; |
| 2136 |
> |
if (runState < 0) { // terminating |
| 2137 |
> |
w.phase = activePhase; |
| 2138 |
> |
return 1; |
| 2139 |
> |
} |
| 2140 |
> |
if (locals) { // run local tasks before (re)polling |
| 2141 |
> |
for (ForkJoinTask<?> u; (u = w.nextLocalTask()) != null;) |
| 2142 |
|
u.doExec(); |
| 2143 |
|
} |
| 2144 |
< |
WorkQueue[] qs = queues; |
| 2145 |
< |
int n = (qs == null) ? 0 : qs.length; |
| 2146 |
< |
for (int i = n; i > 0; --i, ++r) { |
| 2147 |
< |
int j, b, cap; WorkQueue q; ForkJoinTask<?>[] a; |
| 2148 |
< |
if ((q = qs[j = (n - 1) & r]) != null && q != w && |
| 2149 |
< |
(a = q.array) != null && (cap = a.length) > 0) { |
| 2150 |
< |
int k = (cap - 1) & (b = q.base); |
| 2151 |
< |
int nextBase = b + 1, src = j | SRC; |
| 2152 |
< |
ForkJoinTask<?> t = WorkQueue.getSlot(a, k); |
| 2153 |
< |
if (q.base != b) |
| 2154 |
< |
busy = scan = true; |
| 2155 |
< |
else if (t != null) { |
| 2156 |
< |
busy = scan = true; |
| 2157 |
< |
if (!active) { // increment before taking |
| 2158 |
< |
active = true; |
| 2159 |
< |
getAndAddCtl(RC_UNIT); |
| 2144 |
> |
boolean rescan = false, busy = locals = false, interrupted; |
| 2145 |
> |
int n = ((qs = queues) == null) ? 0 : qs.length, m = n - 1; |
| 2146 |
> |
scan: for (int i = n, j; i > 0; --i, ++r) { |
| 2147 |
> |
if ((q = qs[j = m & r]) != null && q != w) { |
| 2148 |
> |
for (int src = j | SRC;;) { |
| 2149 |
> |
ForkJoinTask<?>[] a = q.array; |
| 2150 |
> |
int b = q.base, cap; |
| 2151 |
> |
if (a == null || (cap = a.length) <= 0) |
| 2152 |
> |
break; |
| 2153 |
> |
int k = (cap - 1) & b, nb = b + 1, nk = (cap - 1) & nb; |
| 2154 |
> |
ForkJoinTask<?> t = a[k]; |
| 2155 |
> |
U.loadFence(); // for re-reads |
| 2156 |
> |
if (q.base != b || q.array != a || a[k] != t) |
| 2157 |
> |
; |
| 2158 |
> |
else if (t == null) { |
| 2159 |
> |
if (!rescan) { |
| 2160 |
> |
if (a[nk] != null || q.top - b > 0) |
| 2161 |
> |
rescan = true; |
| 2162 |
> |
else if (!busy && |
| 2163 |
> |
q.owner != null && q.phase >= 0) |
| 2164 |
> |
busy = true; |
| 2165 |
> |
} |
| 2166 |
> |
break; |
| 2167 |
|
} |
| 2168 |
< |
if (WorkQueue.casSlotToNull(a, k, t)) { |
| 2169 |
< |
q.base = nextBase; |
| 2168 |
> |
else if (phase < 0) // reactivate before taking |
| 2169 |
> |
w.phase = phase = activePhase; |
| 2170 |
> |
else if (WorkQueue.casSlotToNull(a, k, t)) { |
| 2171 |
> |
q.base = nb; |
| 2172 |
|
w.source = src; |
| 2173 |
|
t.doExec(); |
| 2174 |
< |
w.source = wsrc = prevSrc; |
| 2175 |
< |
locals = true; |
| 2174 |
> |
w.source = wsrc; |
| 2175 |
> |
rescan = locals = true; |
| 2176 |
> |
break scan; |
| 2177 |
|
} |
| 2002 |
– |
break; |
| 2003 |
– |
} |
| 2004 |
– |
else if (!busy) { |
| 2005 |
– |
if (q.top != b || a[nextBase & (cap - 1)] != null) |
| 2006 |
– |
busy = scan = true; |
| 2007 |
– |
else if (q.source != QUIET && q.phase >= 0) |
| 2008 |
– |
busy = true; |
| 2178 |
|
} |
| 2179 |
|
} |
| 2180 |
|
} |
| 2181 |
< |
VarHandle.acquireFence(); |
| 2182 |
< |
if (!scan && queues == qs) { |
| 2183 |
< |
boolean interrupted; |
| 2184 |
< |
if (!busy) { |
| 2185 |
< |
w.source = prevSrc; |
| 2186 |
< |
if (!active) |
| 2187 |
< |
getAndAddCtl(RC_UNIT); |
| 2188 |
< |
return 1; |
| 2189 |
< |
} |
| 2190 |
< |
if (wsrc != QUIET) |
| 2191 |
< |
w.source = wsrc = QUIET; |
| 2192 |
< |
if (active) { // decrement |
| 2193 |
< |
active = false; |
| 2194 |
< |
parkTime = 0L; |
| 2195 |
< |
getAndAddCtl(RC_MASK & -RC_UNIT); |
| 2196 |
< |
} |
| 2197 |
< |
else if (parkTime == 0L) { |
| 2198 |
< |
parkTime = 1L << 10; // initially about 1 usec |
| 2199 |
< |
Thread.yield(); |
| 2200 |
< |
} |
| 2201 |
< |
else if ((interrupted = interruptible && Thread.interrupted()) || |
| 2202 |
< |
System.nanoTime() - startTime > nanos) { |
| 2203 |
< |
getAndAddCtl(RC_UNIT); |
| 2035 |
< |
return interrupted ? -1 : 0; |
| 2036 |
< |
} |
| 2037 |
< |
else { |
| 2038 |
< |
LockSupport.parkNanos(this, parkTime); |
| 2039 |
< |
if (parkTime < nanos >>> 8 && parkTime < 1L << 20) |
| 2040 |
< |
parkTime <<= 1; // max sleep approx 1 sec or 1% nanos |
| 2041 |
< |
} |
| 2181 |
> |
if (rescan) |
| 2182 |
> |
; // retry |
| 2183 |
> |
else if (phase >= 0) { |
| 2184 |
> |
parkTime = 0L; |
| 2185 |
> |
w.phase = phase = inactivePhase; |
| 2186 |
> |
} |
| 2187 |
> |
else if (!busy) { |
| 2188 |
> |
w.phase = activePhase; |
| 2189 |
> |
return 1; |
| 2190 |
> |
} |
| 2191 |
> |
else if (parkTime == 0L) { |
| 2192 |
> |
parkTime = 1L << 10; // initially about 1 usec |
| 2193 |
> |
Thread.yield(); |
| 2194 |
> |
} |
| 2195 |
> |
else if ((interrupted = interruptible && Thread.interrupted()) || |
| 2196 |
> |
System.nanoTime() - startTime > nanos) { |
| 2197 |
> |
w.phase = activePhase; |
| 2198 |
> |
return interrupted ? -1 : 0; |
| 2199 |
> |
} |
| 2200 |
> |
else { |
| 2201 |
> |
LockSupport.parkNanos(this, parkTime); |
| 2202 |
> |
if (parkTime < nanos >>> 8 && parkTime < 1L << 20) |
| 2203 |
> |
parkTime <<= 1; // max sleep approx 1 sec or 1% nanos |
| 2204 |
|
} |
| 2205 |
|
} |
| 2206 |
|
} |
| 2212 |
|
* @param interruptible true if return on interrupt |
| 2213 |
|
* @return positive if quiescent, negative if interrupted, else 0 |
| 2214 |
|
*/ |
| 2215 |
< |
final int externalHelpQuiescePool(long nanos, boolean interruptible) { |
| 2215 |
> |
private int externalHelpQuiesce(long nanos, boolean interruptible) { |
| 2216 |
|
for (long startTime = System.nanoTime(), parkTime = 0L;;) { |
| 2217 |
|
ForkJoinTask<?> t; |
| 2218 |
|
if ((t = pollScan(false)) != null) { |
| 2238 |
|
} |
| 2239 |
|
|
| 2240 |
|
/** |
| 2241 |
+ |
* Helps quiesce from either internal or external caller |
| 2242 |
+ |
* |
| 2243 |
+ |
* @param pool the pool to use, or null if any |
| 2244 |
+ |
* @param nanos max wait time (Long.MAX_VALUE if effectively untimed) |
| 2245 |
+ |
* @param interruptible true if return on interrupt |
| 2246 |
+ |
* @return positive if quiescent, negative if interrupted, else 0 |
| 2247 |
+ |
*/ |
| 2248 |
+ |
final static int helpQuiescePool(ForkJoinPool pool, long nanos, |
| 2249 |
+ |
boolean interruptible) { |
| 2250 |
+ |
Thread t; ForkJoinPool p; ForkJoinWorkerThread wt; |
| 2251 |
+ |
if ((t = Thread.currentThread()) instanceof ForkJoinWorkerThread && |
| 2252 |
+ |
(p = (wt = (ForkJoinWorkerThread)t).pool) != null && |
| 2253 |
+ |
(p == pool || pool == null)) |
| 2254 |
+ |
return p.helpQuiesce(wt.workQueue, nanos, interruptible); |
| 2255 |
+ |
else if ((p = pool) != null || (p = common) != null) |
| 2256 |
+ |
return p.externalHelpQuiesce(nanos, interruptible); |
| 2257 |
+ |
else |
| 2258 |
+ |
return 0; |
| 2259 |
+ |
} |
| 2260 |
+ |
|
| 2261 |
+ |
/** |
| 2262 |
|
* Gets and removes a local or stolen task for the given worker. |
| 2263 |
|
* |
| 2264 |
|
* @return a task, if available |
| 2265 |
|
*/ |
| 2266 |
|
final ForkJoinTask<?> nextTaskFor(WorkQueue w) { |
| 2267 |
|
ForkJoinTask<?> t; |
| 2268 |
< |
if (w == null || (t = w.nextLocalTask(w.config)) == null) |
| 2268 |
> |
if (w == null || (t = w.nextLocalTask()) == null) |
| 2269 |
|
t = pollScan(false); |
| 2270 |
|
return t; |
| 2271 |
|
} |
| 2274 |
|
|
| 2275 |
|
/** |
| 2276 |
|
* Finds and locks a WorkQueue for an external submitter, or |
| 2277 |
< |
* returns null if shutdown or terminating. |
| 2277 |
> |
* throws RejectedExecutionException if shutdown or terminating. |
| 2278 |
> |
* @param isSubmit false if this is for a common pool fork |
| 2279 |
|
*/ |
| 2280 |
< |
final WorkQueue submissionQueue() { |
| 2280 |
> |
final WorkQueue submissionQueue(boolean isSubmit) { |
| 2281 |
|
int r; |
| 2282 |
+ |
ReentrantLock lock = registrationLock; |
| 2283 |
|
if ((r = ThreadLocalRandom.getProbe()) == 0) { |
| 2284 |
|
ThreadLocalRandom.localInit(); // initialize caller's probe |
| 2285 |
|
r = ThreadLocalRandom.getProbe(); |
| 2286 |
|
} |
| 2287 |
< |
for (int id = r << 1;;) { // even indices only |
| 2288 |
< |
int md = mode, n, i; WorkQueue q; ReentrantLock lock; |
| 2289 |
< |
WorkQueue[] qs = queues; |
| 2290 |
< |
if ((md & SHUTDOWN) != 0 || qs == null || (n = qs.length) <= 0) |
| 2291 |
< |
return null; |
| 2292 |
< |
else if ((q = qs[i = (n - 1) & id]) == null) { |
| 2293 |
< |
if ((lock = registrationLock) != null) { |
| 2294 |
< |
WorkQueue w = new WorkQueue(id | SRC); |
| 2295 |
< |
lock.lock(); // install under lock |
| 2296 |
< |
if (qs[i] == null) |
| 2297 |
< |
qs[i] = w; // else lost race; discard |
| 2287 |
> |
if (lock != null) { // else init error |
| 2288 |
> |
for (int id = r << 1;;) { // even indices only |
| 2289 |
> |
int n, i; WorkQueue[] qs; WorkQueue q; |
| 2290 |
> |
if ((qs = queues) == null || (n = qs.length) <= 0) |
| 2291 |
> |
break; |
| 2292 |
> |
else if ((q = qs[i = (n - 1) & id]) == null) { |
| 2293 |
> |
WorkQueue w = new WorkQueue(null, id | SRC); |
| 2294 |
> |
w.array = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; |
| 2295 |
> |
lock.lock(); // install under lock |
| 2296 |
> |
if (queues == qs && qs[i] == null) |
| 2297 |
> |
qs[i] = w; // else lost race; discard |
| 2298 |
|
lock.unlock(); |
| 2299 |
|
} |
| 2300 |
+ |
else if (q.getAndSetAccess(1) != 0) // move and restart |
| 2301 |
+ |
id = (r = ThreadLocalRandom.advanceProbe(r)) << 1; |
| 2302 |
+ |
else if (isSubmit && runState != 0) { |
| 2303 |
+ |
q.access = 0; // check while lock held |
| 2304 |
+ |
break; |
| 2305 |
+ |
} |
| 2306 |
+ |
else |
| 2307 |
+ |
return q; |
| 2308 |
|
} |
| 2116 |
– |
else if (!q.tryLock()) // move and restart |
| 2117 |
– |
id = (r = ThreadLocalRandom.advanceProbe(r)) << 1; |
| 2118 |
– |
else |
| 2119 |
– |
return q; |
| 2309 |
|
} |
| 2310 |
+ |
throw new RejectedExecutionException(); |
| 2311 |
|
} |
| 2312 |
|
|
| 2313 |
|
/** |
| 2314 |
< |
* Adds the given task to an external submission queue, or throws |
| 2315 |
< |
* exception if shutdown or terminating. |
| 2126 |
< |
* |
| 2127 |
< |
* @param task the task. Caller must ensure non-null. |
| 2128 |
< |
*/ |
| 2129 |
< |
final void externalPush(ForkJoinTask<?> task) { |
| 2130 |
< |
WorkQueue q; |
| 2131 |
< |
if ((q = submissionQueue()) == null) |
| 2132 |
< |
throw new RejectedExecutionException(); // shutdown or disabled |
| 2133 |
< |
else if (q.lockedPush(task)) |
| 2134 |
< |
signalWork(); |
| 2135 |
< |
} |
| 2136 |
< |
|
| 2137 |
< |
// relay from ForkJoinTask to avoid unnecessary initialization |
| 2138 |
< |
static final void externalPushCommon(ForkJoinTask<?> task) { |
| 2139 |
< |
Common.pool.externalPush(task); |
| 2140 |
< |
} |
| 2141 |
< |
|
| 2142 |
< |
/** |
| 2143 |
< |
* Pushes a possibly-external submission. |
| 2314 |
> |
* Pushes a submission to the pool, using internal queue if called |
| 2315 |
> |
* from ForkJoinWorkerThread, else external queue. |
| 2316 |
|
*/ |
| 2317 |
< |
private <T> ForkJoinTask<T> externalSubmit(ForkJoinTask<T> task) { |
| 2318 |
< |
Thread t; ForkJoinWorkerThread wt; WorkQueue q; |
| 2319 |
< |
if (task == null) |
| 2320 |
< |
throw new NullPointerException(); |
| 2317 |
> |
private <T> ForkJoinTask<T> poolSubmit(boolean signalIfEmpty, |
| 2318 |
> |
ForkJoinTask<T> task) { |
| 2319 |
> |
WorkQueue q; Thread t; ForkJoinWorkerThread wt; |
| 2320 |
> |
if (task == null) throw new NullPointerException(); |
| 2321 |
|
if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread) && |
| 2322 |
< |
(q = (wt = (ForkJoinWorkerThread)t).workQueue) != null && |
| 2323 |
< |
wt.pool == this) |
| 2324 |
< |
q.push(task, this); |
| 2325 |
< |
else if ((q = submissionQueue()) == null) |
| 2326 |
< |
throw new RejectedExecutionException(); // shutdown or disabled |
| 2327 |
< |
else if (q.lockedPush(task)) |
| 2328 |
< |
signalWork(); |
| 2322 |
> |
(wt = (ForkJoinWorkerThread)t).pool == this) |
| 2323 |
> |
q = wt.workQueue; |
| 2324 |
> |
else { |
| 2325 |
> |
task.markPoolSubmission(); |
| 2326 |
> |
q = submissionQueue(true); |
| 2327 |
> |
} |
| 2328 |
> |
q.push(task, this, signalIfEmpty); |
| 2329 |
|
return task; |
| 2330 |
|
} |
| 2331 |
|
|
| 2332 |
|
/** |
| 2333 |
< |
* Pushes a possibly-external task without signalling. currently unused |
| 2333 |
> |
* Returns queue for an external thread, if one exists that has |
| 2334 |
> |
* possibly ever submitted to the given pool (nonzero probe), or |
| 2335 |
> |
* null if none. |
| 2336 |
|
*/ |
| 2337 |
< |
/* |
| 2338 |
< |
private <T> ForkJoinTask<T> unsignalledSubmit(ForkJoinTask<T> task) { |
| 2339 |
< |
Thread t; ForkJoinWorkerThread wt; WorkQueue q; |
| 2340 |
< |
if (task == null) |
| 2341 |
< |
throw new NullPointerException(); |
| 2342 |
< |
if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread) && |
| 2169 |
< |
(q = (wt = (ForkJoinWorkerThread)t).workQueue) != null && |
| 2170 |
< |
wt.pool == this) |
| 2171 |
< |
q.push(task, null); |
| 2172 |
< |
else if ((q = submissionQueue()) == null) |
| 2173 |
< |
throw new RejectedExecutionException(); // shutdown or disabled |
| 2174 |
< |
else |
| 2175 |
< |
q.lockedPush(task); |
| 2176 |
< |
return task; |
| 2337 |
> |
private static WorkQueue externalQueue(ForkJoinPool p) { |
| 2338 |
> |
WorkQueue[] qs; |
| 2339 |
> |
int r = ThreadLocalRandom.getProbe(), n; |
| 2340 |
> |
return (p != null && (qs = p.queues) != null && |
| 2341 |
> |
(n = qs.length) > 0 && r != 0) ? |
| 2342 |
> |
qs[(n - 1) & (r << 1)] : null; |
| 2343 |
|
} |
| 2178 |
– |
*/ |
| 2344 |
|
|
| 2345 |
|
/** |
| 2346 |
< |
* Returns common pool queue for an external thread that has |
| 2182 |
< |
* possibly ever submitted a common pool task (nonzero probe), or |
| 2183 |
< |
* null if none. |
| 2346 |
> |
* Returns external queue for common pool. |
| 2347 |
|
*/ |
| 2348 |
|
static WorkQueue commonQueue() { |
| 2349 |
< |
ForkJoinPool p; WorkQueue[] qs; |
| 2187 |
< |
int r = ThreadLocalRandom.getProbe(), n; |
| 2188 |
< |
return ((p = Common.pool) != null && (qs = p.queues) != null && |
| 2189 |
< |
(n = qs.length) > 0 && r != 0) ? |
| 2190 |
< |
qs[(n - 1) & (r << 1)] : null; |
| 2349 |
> |
return externalQueue(common); |
| 2350 |
|
} |
| 2351 |
|
|
| 2352 |
|
/** |
| 2353 |
|
* Returns queue for an external thread, if one exists |
| 2354 |
|
*/ |
| 2355 |
|
final WorkQueue externalQueue() { |
| 2356 |
< |
WorkQueue[] qs; |
| 2198 |
< |
int r = ThreadLocalRandom.getProbe(), n; |
| 2199 |
< |
return ((qs = queues) != null && (n = qs.length) > 0 && r != 0) ? |
| 2200 |
< |
qs[(n - 1) & (r << 1)] : null; |
| 2356 |
> |
return externalQueue(this); |
| 2357 |
|
} |
| 2358 |
|
|
| 2359 |
|
/** |
| 2420 |
|
if (((t = Thread.currentThread()) instanceof ForkJoinWorkerThread) && |
| 2421 |
|
(pool = (wt = (ForkJoinWorkerThread)t).pool) != null && |
| 2422 |
|
(q = wt.workQueue) != null) { |
| 2267 |
– |
int p = pool.mode & SMASK; |
| 2268 |
– |
int a = p + (int)(pool.ctl >> RC_SHIFT); |
| 2423 |
|
int n = q.top - q.base; |
| 2424 |
+ |
int p = pool.parallelism; |
| 2425 |
+ |
int a = (short)(pool.ctl >>> RC_SHIFT); |
| 2426 |
|
return n - (a > (p >>>= 1) ? 0 : |
| 2427 |
|
a > (p >>>= 1) ? 1 : |
| 2428 |
|
a > (p >>>= 1) ? 2 : |
| 2435 |
|
// Termination |
| 2436 |
|
|
| 2437 |
|
/** |
| 2438 |
< |
* Possibly initiates and/or completes termination. |
| 2438 |
> |
* Possibly initiates and/or completes pool termination. |
| 2439 |
|
* |
| 2440 |
|
* @param now if true, unconditionally terminate, else only |
| 2441 |
|
* if no work and no active workers |
| 2443 |
|
* @return true if terminating or terminated |
| 2444 |
|
*/ |
| 2445 |
|
private boolean tryTerminate(boolean now, boolean enable) { |
| 2446 |
< |
int md; // try to set SHUTDOWN, then STOP, then help terminate |
| 2447 |
< |
if (((md = mode) & UNSTOPPABLE) != 0) |
| 2448 |
< |
return false; // for common pool |
| 2449 |
< |
if ((md & SHUTDOWN) == 0) { |
| 2450 |
< |
if (!enable) |
| 2451 |
< |
return false; |
| 2452 |
< |
md = getAndBitwiseOrMode(SHUTDOWN); |
| 2453 |
< |
} |
| 2454 |
< |
if ((md & STOP) == 0) { |
| 2455 |
< |
if (!now && !canStop()) |
| 2456 |
< |
return false; |
| 2457 |
< |
md = getAndBitwiseOrMode(STOP); |
| 2458 |
< |
} |
| 2459 |
< |
for (boolean rescan = true;;) { // repeat until no changes |
| 2460 |
< |
boolean changed = false; |
| 2461 |
< |
for (ForkJoinTask<?> t; (t = pollScan(false)) != null; ) { |
| 2462 |
< |
changed = true; |
| 2463 |
< |
ForkJoinTask.cancelIgnoringExceptions(t); // help cancel |
| 2464 |
< |
} |
| 2465 |
< |
WorkQueue[] qs; int n; WorkQueue q; Thread thread; |
| 2466 |
< |
if ((qs = queues) != null && (n = qs.length) > 0) { |
| 2467 |
< |
for (int j = 1; j < n; j += 2) { // unblock other workers |
| 2468 |
< |
if ((q = qs[j]) != null && (thread = q.owner) != null && |
| 2469 |
< |
!thread.isInterrupted()) { |
| 2470 |
< |
changed = true; |
| 2446 |
> |
int rs; ReentrantLock lock; Condition cond; |
| 2447 |
> |
if ((rs = runState) >= 0) { // set SHUTDOWN and/or STOP |
| 2448 |
> |
if ((config & ISCOMMON) != 0) |
| 2449 |
> |
return false; // cannot shutdown |
| 2450 |
> |
if (!now) { |
| 2451 |
> |
if ((rs & SHUTDOWN) == 0) { |
| 2452 |
> |
if (!enable) |
| 2453 |
> |
return false; |
| 2454 |
> |
getAndBitwiseOrRunState(SHUTDOWN); |
| 2455 |
> |
} |
| 2456 |
> |
if (!canStop()) |
| 2457 |
> |
return false; |
| 2458 |
> |
} |
| 2459 |
> |
getAndBitwiseOrRunState(SHUTDOWN | STOP); |
| 2460 |
> |
} |
| 2461 |
> |
WorkQueue released = reactivate(); // try signalling waiter |
| 2462 |
> |
int tc = (short)(ctl >>> TC_SHIFT); |
| 2463 |
> |
if (released == null && tc > 0) { // help unblock and cancel |
| 2464 |
> |
Thread current = Thread.currentThread(); |
| 2465 |
> |
WorkQueue w = ((current instanceof ForkJoinWorkerThread) ? |
| 2466 |
> |
((ForkJoinWorkerThread)current).workQueue : null); |
| 2467 |
> |
int r = (w == null) ? 0 : w.config + 1; // stagger traversals |
| 2468 |
> |
WorkQueue[] qs = queues; |
| 2469 |
> |
int n = (qs == null) ? 0 : qs.length; |
| 2470 |
> |
for (int i = 0; i < n; ++i) { |
| 2471 |
> |
WorkQueue q; Thread thread; |
| 2472 |
> |
if ((q = qs[(r + i) & (n - 1)]) != null && |
| 2473 |
> |
(thread = q.owner) != current && q.access != STOP) { |
| 2474 |
> |
if (thread != null && !thread.isInterrupted()) { |
| 2475 |
> |
q.forcePhaseActive(); // for awaitWork |
| 2476 |
|
try { |
| 2477 |
|
thread.interrupt(); |
| 2478 |
|
} catch (Throwable ignore) { |
| 2479 |
|
} |
| 2480 |
|
} |
| 2481 |
+ |
for (ForkJoinTask<?> t; (t = q.poll(null)) != null; ) |
| 2482 |
+ |
ForkJoinTask.cancelIgnoringExceptions(t); |
| 2483 |
|
} |
| 2484 |
|
} |
| 2485 |
< |
ReentrantLock lock; Condition cond; // signal when no workers |
| 2486 |
< |
if (((md = mode) & TERMINATED) == 0 && |
| 2487 |
< |
(md & SMASK) + (short)(ctl >>> TC_SHIFT) <= 0 && |
| 2488 |
< |
(getAndBitwiseOrMode(TERMINATED) & TERMINATED) == 0 && |
| 2489 |
< |
(lock = registrationLock) != null) { |
| 2490 |
< |
lock.lock(); |
| 2491 |
< |
if ((cond = termination) != null) |
| 2492 |
< |
cond.signalAll(); |
| 2493 |
< |
lock.unlock(); |
| 2331 |
< |
} |
| 2332 |
< |
if (changed) |
| 2333 |
< |
rescan = true; |
| 2334 |
< |
else if (rescan) |
| 2335 |
< |
rescan = false; |
| 2336 |
< |
else |
| 2337 |
< |
break; |
| 2485 |
> |
} |
| 2486 |
> |
if ((tc <= 0 || (short)(ctl >>> TC_SHIFT) <= 0) && |
| 2487 |
> |
(getAndBitwiseOrRunState(TERMINATED) & TERMINATED) == 0 && |
| 2488 |
> |
(lock = registrationLock) != null) { |
| 2489 |
> |
lock.lock(); // signal when no workers |
| 2490 |
> |
if ((cond = termination) != null) |
| 2491 |
> |
cond.signalAll(); |
| 2492 |
> |
lock.unlock(); |
| 2493 |
> |
// container.close(); // for loom |
| 2494 |
|
} |
| 2495 |
|
return true; |
| 2496 |
|
} |
| 2666 |
|
throw new IllegalArgumentException(); |
| 2667 |
|
if (factory == null || unit == null) |
| 2668 |
|
throw new NullPointerException(); |
| 2669 |
+ |
this.parallelism = p; |
| 2670 |
|
this.factory = factory; |
| 2671 |
|
this.ueh = handler; |
| 2672 |
|
this.saturate = saturate; |
| 2673 |
+ |
this.config = asyncMode ? FIFO : 0; |
| 2674 |
|
this.keepAlive = Math.max(unit.toMillis(keepAliveTime), TIMEOUT_SLOP); |
| 2517 |
– |
int size = 1 << (33 - Integer.numberOfLeadingZeros(p - 1)); |
| 2675 |
|
int corep = Math.min(Math.max(corePoolSize, p), MAX_CAP); |
| 2676 |
< |
int maxSpares = Math.min(maximumPoolSize, MAX_CAP) - p; |
| 2677 |
< |
int minAvail = Math.min(Math.max(minimumRunnable, 0), MAX_CAP); |
| 2678 |
< |
this.bounds = ((minAvail - p) & SMASK) | (maxSpares << SWIDTH); |
| 2679 |
< |
this.mode = p | (asyncMode ? FIFO : 0); |
| 2680 |
< |
this.ctl = ((((long)(-corep) << TC_SHIFT) & TC_MASK) | |
| 2524 |
< |
(((long)(-p) << RC_SHIFT) & RC_MASK)); |
| 2676 |
> |
int maxSpares = Math.max(0, Math.min(maximumPoolSize - p, MAX_CAP)); |
| 2677 |
> |
int minAvail = Math.max(0, Math.min(minimumRunnable, MAX_CAP)); |
| 2678 |
> |
this.bounds = (long)(minAvail & SMASK) | (long)(maxSpares << SWIDTH) | |
| 2679 |
> |
((long)corep << 32); |
| 2680 |
> |
int size = 1 << (33 - Integer.numberOfLeadingZeros(p - 1)); |
| 2681 |
|
this.registrationLock = new ReentrantLock(); |
| 2682 |
|
this.queues = new WorkQueue[size]; |
| 2683 |
|
String pid = Integer.toString(getAndAddPoolIds(1) + 1); |
| 2684 |
< |
this.workerNamePrefix = "ForkJoinPool-" + pid + "-worker-"; |
| 2685 |
< |
} |
| 2686 |
< |
|
| 2531 |
< |
// helper method for commonPool constructor |
| 2532 |
< |
private static Object newInstanceFromSystemProperty(String property) |
| 2533 |
< |
throws ReflectiveOperationException { |
| 2534 |
< |
String className = System.getProperty(property); |
| 2535 |
< |
return (className == null) |
| 2536 |
< |
? null |
| 2537 |
< |
: ClassLoader.getSystemClassLoader().loadClass(className) |
| 2538 |
< |
.getConstructor().newInstance(); |
| 2684 |
> |
String name = "ForkJoinPool-" + pid; |
| 2685 |
> |
this.workerNamePrefix = name + "-worker-"; |
| 2686 |
> |
// this.container = SharedThreadContainer.create(name); // for loom |
| 2687 |
|
} |
| 2688 |
|
|
| 2689 |
|
/** |
| 2691 |
|
* overridden by system properties |
| 2692 |
|
*/ |
| 2693 |
|
private ForkJoinPool(byte forCommonPoolOnly) { |
| 2694 |
< |
int parallelism = Math.max(1, Runtime.getRuntime().availableProcessors() - 1); |
| 2547 |
< |
int maxSpares = DEFAULT_COMMON_MAX_SPARES; |
| 2548 |
< |
ForkJoinWorkerThreadFactory fac = null; |
| 2694 |
> |
ForkJoinWorkerThreadFactory fac = defaultForkJoinWorkerThreadFactory; |
| 2695 |
|
UncaughtExceptionHandler handler = null; |
| 2696 |
+ |
int maxSpares = DEFAULT_COMMON_MAX_SPARES; |
| 2697 |
+ |
int pc = 0, preset = 0; // nonzero if size set as property |
| 2698 |
|
try { // ignore exceptions in accessing/parsing properties |
| 2551 |
– |
fac = (ForkJoinWorkerThreadFactory) newInstanceFromSystemProperty( |
| 2552 |
– |
"java.util.concurrent.ForkJoinPool.common.threadFactory"); |
| 2553 |
– |
handler = (UncaughtExceptionHandler) newInstanceFromSystemProperty( |
| 2554 |
– |
"java.util.concurrent.ForkJoinPool.common.exceptionHandler"); |
| 2699 |
|
String pp = System.getProperty |
| 2700 |
|
("java.util.concurrent.ForkJoinPool.common.parallelism"); |
| 2701 |
< |
String msp = System.getProperty |
| 2701 |
> |
if (pp != null) { |
| 2702 |
> |
pc = Math.max(0, Integer.parseInt(pp)); |
| 2703 |
> |
preset = PRESET_SIZE; |
| 2704 |
> |
} |
| 2705 |
> |
String ms = System.getProperty |
| 2706 |
|
("java.util.concurrent.ForkJoinPool.common.maximumSpares"); |
| 2707 |
< |
if (pp != null) |
| 2708 |
< |
parallelism = Integer.parseInt(pp); |
| 2709 |
< |
if (msp != null) |
| 2710 |
< |
maxSpares = Integer.parseInt(msp); |
| 2707 |
> |
if (ms != null) |
| 2708 |
> |
maxSpares = Math.max(0, Math.min(MAX_CAP, Integer.parseInt(ms))); |
| 2709 |
> |
String sf = System.getProperty |
| 2710 |
> |
("java.util.concurrent.ForkJoinPool.common.threadFactory"); |
| 2711 |
> |
String sh = System.getProperty |
| 2712 |
> |
("java.util.concurrent.ForkJoinPool.common.exceptionHandler"); |
| 2713 |
> |
if (sf != null || sh != null) { |
| 2714 |
> |
ClassLoader ldr = ClassLoader.getSystemClassLoader(); |
| 2715 |
> |
if (sf != null) |
| 2716 |
> |
fac = (ForkJoinWorkerThreadFactory) |
| 2717 |
> |
ldr.loadClass(sf).getConstructor().newInstance(); |
| 2718 |
> |
if (sh != null) |
| 2719 |
> |
handler = (UncaughtExceptionHandler) |
| 2720 |
> |
ldr.loadClass(sh).getConstructor().newInstance(); |
| 2721 |
> |
} |
| 2722 |
|
} catch (Exception ignore) { |
| 2723 |
|
} |
| 2724 |
+ |
if (preset == 0) |
| 2725 |
+ |
pc = Math.max(1, Runtime.getRuntime().availableProcessors() - 1); |
| 2726 |
+ |
int p = Math.min(pc, MAX_CAP); |
| 2727 |
+ |
int size = (p == 0) ? 1 : 1 << (33 - Integer.numberOfLeadingZeros(p-1)); |
| 2728 |
+ |
this.parallelism = p; |
| 2729 |
+ |
this.config = ISCOMMON | preset; |
| 2730 |
+ |
this.bounds = (long)(1 | (maxSpares << SWIDTH)); |
| 2731 |
+ |
this.factory = fac; |
| 2732 |
|
this.ueh = handler; |
| 2733 |
|
this.keepAlive = DEFAULT_KEEPALIVE; |
| 2734 |
|
this.saturate = null; |
| 2735 |
|
this.workerNamePrefix = null; |
| 2569 |
– |
int p = Math.min(Math.max(parallelism, 0), MAX_CAP), size; |
| 2570 |
– |
this.mode = p | UNSTOPPABLE; |
| 2571 |
– |
if (p > 0) { |
| 2572 |
– |
size = 1 << (33 - Integer.numberOfLeadingZeros(p - 1)); |
| 2573 |
– |
this.bounds = ((1 - p) & SMASK) | (maxSpares << SWIDTH); |
| 2574 |
– |
this.ctl = ((((long)(-p) << TC_SHIFT) & TC_MASK) | |
| 2575 |
– |
(((long)(-p) << RC_SHIFT) & RC_MASK)); |
| 2576 |
– |
} else { // zero min, max, spare counts, 1 slot |
| 2577 |
– |
size = 1; |
| 2578 |
– |
this.bounds = 0; |
| 2579 |
– |
this.ctl = 0L; |
| 2580 |
– |
} |
| 2581 |
– |
this.factory = (fac != null) ? fac : |
| 2582 |
– |
new DefaultCommonPoolForkJoinWorkerThreadFactory(); |
| 2583 |
– |
this.queues = new WorkQueue[size]; |
| 2736 |
|
this.registrationLock = new ReentrantLock(); |
| 2737 |
< |
} |
| 2738 |
< |
|
| 2587 |
< |
/** |
| 2588 |
< |
* Singleton class for the Common Pool to reduce static loading |
| 2589 |
< |
* dependencies |
| 2590 |
< |
*/ |
| 2591 |
< |
static final class Common { |
| 2592 |
< |
/** |
| 2593 |
< |
* Common (static) pool. Non-null for public use unless a static |
| 2594 |
< |
* construction exception, but internal usages null-check on use |
| 2595 |
< |
* to paranoically avoid potential initialization circularities |
| 2596 |
< |
* as well as to simplify generated code. |
| 2597 |
< |
*/ |
| 2598 |
< |
static final ForkJoinPool pool; |
| 2599 |
< |
|
| 2600 |
< |
/** |
| 2601 |
< |
* Common pool parallelism. To allow simpler use and management |
| 2602 |
< |
* when common pool threads are disabled, we allow the underlying |
| 2603 |
< |
* common.parallelism field to be zero, but in that case still report |
| 2604 |
< |
* parallelism as 1 to reflect resulting caller-runs mechanics. |
| 2605 |
< |
*/ |
| 2606 |
< |
static final int parallelism; |
| 2607 |
< |
|
| 2608 |
< |
static { |
| 2609 |
< |
@SuppressWarnings("removal") |
| 2610 |
< |
ForkJoinPool p = |
| 2611 |
< |
AccessController.doPrivileged(new PrivilegedAction<>() { |
| 2612 |
< |
public ForkJoinPool run() { |
| 2613 |
< |
return new ForkJoinPool((byte)0); }}); |
| 2614 |
< |
pool = p; |
| 2615 |
< |
parallelism = Math.max(p.mode & SMASK, 1); |
| 2616 |
< |
} |
| 2737 |
> |
this.queues = new WorkQueue[size]; |
| 2738 |
> |
// this.container = SharedThreadContainer.create("ForkJoinPool.commonPool"); // for loom |
| 2739 |
|
} |
| 2740 |
|
|
| 2741 |
|
/** |
| 2752 |
|
* @since 1.8 |
| 2753 |
|
*/ |
| 2754 |
|
public static ForkJoinPool commonPool() { |
| 2755 |
< |
return Common.pool; |
| 2755 |
> |
// assert common != null : "static init error"; |
| 2756 |
> |
return common; |
| 2757 |
|
} |
| 2758 |
|
|
| 2759 |
|
// Execution methods |
| 2776 |
|
* scheduled for execution |
| 2777 |
|
*/ |
| 2778 |
|
public <T> T invoke(ForkJoinTask<T> task) { |
| 2779 |
< |
externalSubmit(task); |
| 2780 |
< |
return task.joinForPoolInvoke(this); |
| 2779 |
> |
poolSubmit(true, task); |
| 2780 |
> |
return task.join(); |
| 2781 |
|
} |
| 2782 |
|
|
| 2783 |
|
/** |
| 2789 |
|
* scheduled for execution |
| 2790 |
|
*/ |
| 2791 |
|
public void execute(ForkJoinTask<?> task) { |
| 2792 |
< |
externalSubmit(task); |
| 2792 |
> |
poolSubmit(true, task); |
| 2793 |
|
} |
| 2794 |
|
|
| 2795 |
|
// AbstractExecutorService methods |
| 2802 |
|
@Override |
| 2803 |
|
@SuppressWarnings("unchecked") |
| 2804 |
|
public void execute(Runnable task) { |
| 2805 |
< |
externalSubmit((task instanceof ForkJoinTask<?>) |
| 2806 |
< |
? (ForkJoinTask<Void>) task // avoid re-wrap |
| 2807 |
< |
: new ForkJoinTask.RunnableExecuteAction(task)); |
| 2805 |
> |
poolSubmit(true, (task instanceof ForkJoinTask<?>) |
| 2806 |
> |
? (ForkJoinTask<Void>) task // avoid re-wrap |
| 2807 |
> |
: new ForkJoinTask.RunnableExecuteAction(task)); |
| 2808 |
|
} |
| 2809 |
|
|
| 2810 |
|
/** |
| 2818 |
|
* scheduled for execution |
| 2819 |
|
*/ |
| 2820 |
|
public <T> ForkJoinTask<T> submit(ForkJoinTask<T> task) { |
| 2821 |
< |
return externalSubmit(task); |
| 2821 |
> |
return poolSubmit(true, task); |
| 2822 |
|
} |
| 2823 |
|
|
| 2824 |
|
/** |
| 2828 |
|
*/ |
| 2829 |
|
@Override |
| 2830 |
|
public <T> ForkJoinTask<T> submit(Callable<T> task) { |
| 2831 |
< |
return externalSubmit(new ForkJoinTask.AdaptedCallable<T>(task)); |
| 2831 |
> |
return poolSubmit(true, new ForkJoinTask.AdaptedCallable<T>(task)); |
| 2832 |
|
} |
| 2833 |
|
|
| 2834 |
|
/** |
| 2838 |
|
*/ |
| 2839 |
|
@Override |
| 2840 |
|
public <T> ForkJoinTask<T> submit(Runnable task, T result) { |
| 2841 |
< |
return externalSubmit(new ForkJoinTask.AdaptedRunnable<T>(task, result)); |
| 2841 |
> |
return poolSubmit(true, new ForkJoinTask.AdaptedRunnable<T>(task, result)); |
| 2842 |
|
} |
| 2843 |
|
|
| 2844 |
|
/** |
| 2849 |
|
@Override |
| 2850 |
|
@SuppressWarnings("unchecked") |
| 2851 |
|
public ForkJoinTask<?> submit(Runnable task) { |
| 2852 |
< |
return externalSubmit((task instanceof ForkJoinTask<?>) |
| 2853 |
< |
? (ForkJoinTask<Void>) task // avoid re-wrap |
| 2854 |
< |
: new ForkJoinTask.AdaptedRunnableAction(task)); |
| 2852 |
> |
return poolSubmit(true, (task instanceof ForkJoinTask<?>) |
| 2853 |
> |
? (ForkJoinTask<Void>) task // avoid re-wrap |
| 2854 |
> |
: new ForkJoinTask.AdaptedRunnableAction(task)); |
| 2855 |
|
} |
| 2856 |
|
|
| 2857 |
< |
// Potential added methods |
| 2857 |
> |
// Added mainly for possible use in Loom |
| 2858 |
> |
|
| 2859 |
|
/** |
| 2860 |
|
* Submits the given task without guaranteeing that it will |
| 2861 |
< |
* eventually execute in the absence of available active threads |
| 2862 |
< |
* or invocations of {@link #activateThread activateThread}. In |
| 2863 |
< |
* some contexts, this method may reduce contention and overhead |
| 2864 |
< |
* by either deferring processing or relying on context-specific |
| 2865 |
< |
* knowledge that existing threads (possibly including the calling |
| 2742 |
< |
* thread if operating in this pool) will eventually be available |
| 2743 |
< |
* to execute the task. |
| 2861 |
> |
* eventually execute in the absence of available active threads. |
| 2862 |
> |
* In some contexts, this method may reduce contention and |
| 2863 |
> |
* overhead by relying on context-specific knowledge that existing |
| 2864 |
> |
* threads (possibly including the calling thread if operating in |
| 2865 |
> |
* this pool) will eventually be available to execute the task. |
| 2866 |
|
* |
| 2867 |
|
* @param task the task |
| 2868 |
|
* @param <T> the type of the task's result |
| 2869 |
|
* @return the task |
| 2870 |
|
*/ |
| 2871 |
< |
/* |
| 2872 |
< |
public <T> ForkJoinTask<T> lazySubmit(ForkJoinTask<T> task) { |
| 2751 |
< |
return unsignalledSubmit(task); |
| 2871 |
> |
public <T> ForkJoinTask<T> lazySubmit(ForkJoinTask<T> task) { |
| 2872 |
> |
return poolSubmit(false, task); |
| 2873 |
|
} |
| 2874 |
< |
*/ |
| 2874 |
> |
|
| 2875 |
|
/** |
| 2876 |
< |
* If there are fewer than {@link #getParallelism getParallelism} |
| 2877 |
< |
* active threads in this pool, activates one to process |
| 2878 |
< |
* tasks. This thread may in turn activate others if discovers or |
| 2879 |
< |
* generates additional tasks. This method may be used in |
| 2880 |
< |
* conjunction with {@link #lazySubmit lazySubmit} to defer task |
| 2881 |
< |
* processing. |
| 2876 |
> |
* Changes the target parallelism of this pool, controlling the |
| 2877 |
> |
* future creation, use, and termination of worker threads. |
| 2878 |
> |
* Applications include contexts in which the number of available |
| 2879 |
> |
* processors changes over time. |
| 2880 |
> |
* |
| 2881 |
> |
* @param size the target parallelism level |
| 2882 |
> |
* @return the previous parallelism level. |
| 2883 |
> |
* @throws IllegalArgumentException if size is less than 1 or |
| 2884 |
> |
* greater than the maximum supported by this |
| 2885 |
> |
* pool (currently 32767). |
| 2886 |
> |
* @throws IllegalStateException if this is the{@link #commonPool()} and |
| 2887 |
> |
* parallelism level was set by System property |
| 2888 |
> |
* {@systemProperty java.util.concurrent.ForkJoinPool.common.parallelism}. |
| 2889 |
> |
* @throws SecurityException if a security manager exists and |
| 2890 |
> |
* the caller is not permitted to modify threads |
| 2891 |
> |
* because it does not hold {@link |
| 2892 |
> |
* java.lang.RuntimePermission}{@code ("modifyThread")} |
| 2893 |
|
*/ |
| 2894 |
< |
/* |
| 2895 |
< |
public void activateThread() { |
| 2896 |
< |
signalWork(); |
| 2894 |
> |
public int setParallelism(int size) { |
| 2895 |
> |
if (size < 1 || size > MAX_CAP) |
| 2896 |
> |
throw new IllegalArgumentException(); |
| 2897 |
> |
if ((config & PRESET_SIZE) != 0) |
| 2898 |
> |
throw new IllegalStateException("Cannot override System property"); |
| 2899 |
> |
checkPermission(); |
| 2900 |
> |
return getAndSetParallelism(size); |
| 2901 |
|
} |
| 2766 |
– |
*/ |
| 2902 |
|
|
| 2903 |
|
/** |
| 2904 |
|
* @throws NullPointerException {@inheritDoc} |
| 2912 |
|
ForkJoinTask<T> f = |
| 2913 |
|
new ForkJoinTask.AdaptedInterruptibleCallable<T>(t); |
| 2914 |
|
futures.add(f); |
| 2915 |
< |
externalSubmit(f); |
| 2915 |
> |
poolSubmit(true, f); |
| 2916 |
|
} |
| 2917 |
|
for (int i = futures.size() - 1; i >= 0; --i) |
| 2918 |
< |
((ForkJoinTask<?>)futures.get(i)).awaitPoolInvoke(this); |
| 2918 |
> |
((ForkJoinTask<?>)futures.get(i)).quietlyJoin(); |
| 2919 |
|
return futures; |
| 2920 |
|
} catch (Throwable t) { |
| 2921 |
|
for (Future<T> e : futures) |
| 2935 |
|
ForkJoinTask<T> f = |
| 2936 |
|
new ForkJoinTask.AdaptedInterruptibleCallable<T>(t); |
| 2937 |
|
futures.add(f); |
| 2938 |
< |
externalSubmit(f); |
| 2938 |
> |
poolSubmit(true, f); |
| 2939 |
|
} |
| 2940 |
|
long startTime = System.nanoTime(), ns = nanos; |
| 2941 |
|
boolean timedOut = (ns < 0L); |
| 2942 |
|
for (int i = futures.size() - 1; i >= 0; --i) { |
| 2943 |
< |
Future<T> f = futures.get(i); |
| 2943 |
> |
ForkJoinTask<T> f = (ForkJoinTask<T>)futures.get(i); |
| 2944 |
|
if (!f.isDone()) { |
| 2945 |
+ |
if (!timedOut) |
| 2946 |
+ |
timedOut = !f.quietlyJoin(ns, TimeUnit.NANOSECONDS); |
| 2947 |
|
if (timedOut) |
| 2948 |
|
ForkJoinTask.cancelIgnoringExceptions(f); |
| 2949 |
< |
else { |
| 2950 |
< |
((ForkJoinTask<T>)f).awaitPoolInvoke(this, ns); |
| 2814 |
< |
if ((ns = nanos - (System.nanoTime() - startTime)) < 0L) |
| 2815 |
< |
timedOut = true; |
| 2816 |
< |
} |
| 2949 |
> |
else |
| 2950 |
> |
ns = nanos - (System.nanoTime() - startTime); |
| 2951 |
|
} |
| 2952 |
|
} |
| 2953 |
|
return futures; |
| 2974 |
|
Throwable ex = null; |
| 2975 |
|
boolean failed; |
| 2976 |
|
if (c == null || Thread.interrupted() || |
| 2977 |
< |
(pool != null && pool.mode < 0)) |
| 2977 |
> |
(pool != null && pool.runState < 0)) |
| 2978 |
|
failed = true; |
| 2979 |
|
else if (isDone()) |
| 2980 |
|
failed = false; |
| 2987 |
|
failed = true; |
| 2988 |
|
} |
| 2989 |
|
} |
| 2990 |
< |
if ((pool != null && pool.mode < 0) || |
| 2990 |
> |
if ((pool != null && pool.runState < 0) || |
| 2991 |
|
(failed && count.getAndDecrement() <= 1)) |
| 2992 |
|
trySetThrown(ex != null ? ex : new CancellationException()); |
| 2993 |
|
} |
| 3044 |
|
throw new NullPointerException(); |
| 3045 |
|
InvokeAnyTask<T> f = new InvokeAnyTask<T>(root, c); |
| 3046 |
|
fs.add(f); |
| 3047 |
< |
externalSubmit(f); |
| 3047 |
> |
poolSubmit(true, f); |
| 3048 |
|
if (root.isDone()) |
| 3049 |
|
break; |
| 3050 |
|
} |
| 3051 |
< |
return root.getForPoolInvoke(this); |
| 3051 |
> |
return root.get(); |
| 3052 |
|
} finally { |
| 3053 |
|
for (InvokeAnyTask<T> f : fs) |
| 3054 |
|
ForkJoinTask.cancelIgnoringExceptions(f); |
| 3071 |
|
throw new NullPointerException(); |
| 3072 |
|
InvokeAnyTask<T> f = new InvokeAnyTask<T>(root, c); |
| 3073 |
|
fs.add(f); |
| 3074 |
< |
externalSubmit(f); |
| 3074 |
> |
poolSubmit(true, f); |
| 3075 |
|
if (root.isDone()) |
| 3076 |
|
break; |
| 3077 |
|
} |
| 3078 |
< |
return root.getForPoolInvoke(this, nanos); |
| 3078 |
> |
return root.get(nanos, TimeUnit.NANOSECONDS); |
| 3079 |
|
} finally { |
| 3080 |
|
for (InvokeAnyTask<T> f : fs) |
| 3081 |
|
ForkJoinTask.cancelIgnoringExceptions(f); |
| 3107 |
|
* @return the targeted parallelism level of this pool |
| 3108 |
|
*/ |
| 3109 |
|
public int getParallelism() { |
| 3110 |
< |
int par = mode & SMASK; |
| 2977 |
< |
return (par > 0) ? par : 1; |
| 3110 |
> |
return Math.max(getParallelismOpaque(), 1); |
| 3111 |
|
} |
| 3112 |
|
|
| 3113 |
|
/** |
| 3117 |
|
* @since 1.8 |
| 3118 |
|
*/ |
| 3119 |
|
public static int getCommonPoolParallelism() { |
| 3120 |
< |
return Common.parallelism; |
| 3120 |
> |
return common.getParallelism(); |
| 3121 |
|
} |
| 3122 |
|
|
| 3123 |
|
/** |
| 3129 |
|
* @return the number of worker threads |
| 3130 |
|
*/ |
| 3131 |
|
public int getPoolSize() { |
| 3132 |
< |
return ((mode & SMASK) + (short)(ctl >>> TC_SHIFT)); |
| 3132 |
> |
return (short)(ctl >>> TC_SHIFT); |
| 3133 |
|
} |
| 3134 |
|
|
| 3135 |
|
/** |
| 3139 |
|
* @return {@code true} if this pool uses async mode |
| 3140 |
|
*/ |
| 3141 |
|
public boolean getAsyncMode() { |
| 3142 |
< |
return (mode & FIFO) != 0; |
| 3142 |
> |
return (config & FIFO) != 0; |
| 3143 |
|
} |
| 3144 |
|
|
| 3145 |
|
/** |
| 3151 |
|
* @return the number of worker threads |
| 3152 |
|
*/ |
| 3153 |
|
public int getRunningThreadCount() { |
| 3021 |
– |
VarHandle.acquireFence(); |
| 3154 |
|
WorkQueue[] qs; WorkQueue q; |
| 3155 |
|
int rc = 0; |
| 3156 |
< |
if ((qs = queues) != null) { |
| 3156 |
> |
if ((runState & TERMINATED) == 0 && (qs = queues) != null) { |
| 3157 |
|
for (int i = 1; i < qs.length; i += 2) { |
| 3158 |
|
if ((q = qs[i]) != null && q.isApparentlyUnblocked()) |
| 3159 |
|
++rc; |
| 3170 |
|
* @return the number of active threads |
| 3171 |
|
*/ |
| 3172 |
|
public int getActiveThreadCount() { |
| 3173 |
< |
int r = (mode & SMASK) + (int)(ctl >> RC_SHIFT); |
| 3042 |
< |
return (r <= 0) ? 0 : r; // suppress momentarily negative values |
| 3173 |
> |
return Math.max((short)(ctl >>> RC_SHIFT), 0); |
| 3174 |
|
} |
| 3175 |
|
|
| 3176 |
|
/** |
| 3205 |
|
if ((qs = queues) != null) { |
| 3206 |
|
for (int i = 1; i < qs.length; i += 2) { |
| 3207 |
|
if ((q = qs[i]) != null) |
| 3208 |
< |
count += (long)q.nsteals & 0xffffffffL; |
| 3208 |
> |
count += (long)q.nsteals & 0xffffffffL; |
| 3209 |
|
} |
| 3210 |
|
} |
| 3211 |
|
return count; |
| 3222 |
|
* @return the number of queued tasks |
| 3223 |
|
*/ |
| 3224 |
|
public long getQueuedTaskCount() { |
| 3094 |
– |
VarHandle.acquireFence(); |
| 3225 |
|
WorkQueue[] qs; WorkQueue q; |
| 3226 |
|
int count = 0; |
| 3227 |
< |
if ((qs = queues) != null) { |
| 3227 |
> |
if ((runState & TERMINATED) == 0 && (qs = queues) != null) { |
| 3228 |
|
for (int i = 1; i < qs.length; i += 2) { |
| 3229 |
|
if ((q = qs[i]) != null) |
| 3230 |
|
count += q.queueSize(); |
| 3241 |
|
* @return the number of queued submissions |
| 3242 |
|
*/ |
| 3243 |
|
public int getQueuedSubmissionCount() { |
| 3114 |
– |
VarHandle.acquireFence(); |
| 3244 |
|
WorkQueue[] qs; WorkQueue q; |
| 3245 |
|
int count = 0; |
| 3246 |
< |
if ((qs = queues) != null) { |
| 3246 |
> |
if ((runState & TERMINATED) == 0 && (qs = queues) != null) { |
| 3247 |
|
for (int i = 0; i < qs.length; i += 2) { |
| 3248 |
|
if ((q = qs[i]) != null) |
| 3249 |
|
count += q.queueSize(); |
| 3259 |
|
* @return {@code true} if there are any queued submissions |
| 3260 |
|
*/ |
| 3261 |
|
public boolean hasQueuedSubmissions() { |
| 3262 |
< |
VarHandle.acquireFence(); |
| 3134 |
< |
WorkQueue[] qs; WorkQueue q; |
| 3135 |
< |
if ((qs = queues) != null) { |
| 3136 |
< |
for (int i = 0; i < qs.length; i += 2) { |
| 3137 |
< |
if ((q = qs[i]) != null && !q.isEmpty()) |
| 3138 |
< |
return true; |
| 3139 |
< |
} |
| 3140 |
< |
} |
| 3141 |
< |
return false; |
| 3262 |
> |
return (runState & TERMINATED) == 0 && hasSubmissions(); |
| 3263 |
|
} |
| 3264 |
|
|
| 3265 |
|
/** |
| 3308 |
|
*/ |
| 3309 |
|
public String toString() { |
| 3310 |
|
// Use a single pass through queues to collect counts |
| 3190 |
– |
int md = mode; // read volatile fields first |
| 3191 |
– |
long c = ctl; |
| 3311 |
|
long st = stealCount; |
| 3312 |
|
long qt = 0L, ss = 0L; int rc = 0; |
| 3313 |
|
WorkQueue[] qs; WorkQueue q; |
| 3327 |
|
} |
| 3328 |
|
} |
| 3329 |
|
|
| 3330 |
< |
int pc = (md & SMASK); |
| 3331 |
< |
int tc = pc + (short)(c >>> TC_SHIFT); |
| 3332 |
< |
int ac = pc + (int)(c >> RC_SHIFT); |
| 3330 |
> |
int pc = parallelism; |
| 3331 |
> |
long c = ctl; |
| 3332 |
> |
int tc = (short)(c >>> TC_SHIFT); |
| 3333 |
> |
int ac = (short)(c >>> RC_SHIFT); |
| 3334 |
|
if (ac < 0) // ignore transient negative |
| 3335 |
|
ac = 0; |
| 3336 |
< |
String level = ((md & TERMINATED) != 0 ? "Terminated" : |
| 3337 |
< |
(md & STOP) != 0 ? "Terminating" : |
| 3338 |
< |
(md & SHUTDOWN) != 0 ? "Shutting down" : |
| 3336 |
> |
int rs = runState; |
| 3337 |
> |
String level = ((rs & TERMINATED) != 0 ? "Terminated" : |
| 3338 |
> |
(rs & STOP) != 0 ? "Terminating" : |
| 3339 |
> |
(rs & SHUTDOWN) != 0 ? "Shutting down" : |
| 3340 |
|
"Running"); |
| 3341 |
|
return super.toString() + |
| 3342 |
|
"[" + level + |
| 3399 |
|
* @return {@code true} if all tasks have completed following shut down |
| 3400 |
|
*/ |
| 3401 |
|
public boolean isTerminated() { |
| 3402 |
< |
return (mode & TERMINATED) != 0; |
| 3402 |
> |
return (runState & TERMINATED) != 0; |
| 3403 |
|
} |
| 3404 |
|
|
| 3405 |
|
/** |
| 3416 |
|
* @return {@code true} if terminating but not yet terminated |
| 3417 |
|
*/ |
| 3418 |
|
public boolean isTerminating() { |
| 3419 |
< |
return (mode & (STOP | TERMINATED)) == STOP; |
| 3419 |
> |
return (runState & (STOP | TERMINATED)) == STOP; |
| 3420 |
|
} |
| 3421 |
|
|
| 3422 |
|
/** |
| 3425 |
|
* @return {@code true} if this pool has been shut down |
| 3426 |
|
*/ |
| 3427 |
|
public boolean isShutdown() { |
| 3428 |
< |
return (mode & SHUTDOWN) != 0; |
| 3428 |
> |
return runState != 0; |
| 3429 |
|
} |
| 3430 |
|
|
| 3431 |
|
/** |
| 3444 |
|
*/ |
| 3445 |
|
public boolean awaitTermination(long timeout, TimeUnit unit) |
| 3446 |
|
throws InterruptedException { |
| 3447 |
< |
ReentrantLock lock; Condition cond; |
| 3447 |
> |
ReentrantLock lock; Condition cond; boolean terminated; |
| 3448 |
|
long nanos = unit.toNanos(timeout); |
| 3449 |
< |
boolean terminated = false; |
| 3450 |
< |
if ((mode & UNSTOPPABLE) != 0) { |
| 3330 |
< |
Thread t; ForkJoinWorkerThread wt; int q; |
| 3331 |
< |
if ((t = Thread.currentThread()) instanceof ForkJoinWorkerThread && |
| 3332 |
< |
(wt = (ForkJoinWorkerThread)t).pool == this) |
| 3333 |
< |
q = helpQuiescePool(wt.workQueue, nanos, true); |
| 3334 |
< |
else |
| 3335 |
< |
q = externalHelpQuiescePool(nanos, true); |
| 3336 |
< |
if (q < 0) |
| 3449 |
> |
if ((config & ISCOMMON) != 0) { |
| 3450 |
> |
if (helpQuiescePool(this, nanos, true) < 0) |
| 3451 |
|
throw new InterruptedException(); |
| 3452 |
+ |
terminated = false; |
| 3453 |
|
} |
| 3454 |
< |
else if (!(terminated = tryTerminate(false, false)) && |
| 3455 |
< |
(lock = registrationLock) != null) { |
| 3456 |
< |
lock.lock(); |
| 3457 |
< |
try { |
| 3458 |
< |
if ((cond = termination) == null) |
| 3459 |
< |
termination = cond = lock.newCondition(); |
| 3460 |
< |
while (!(terminated = ((mode & TERMINATED) != 0)) && nanos > 0L) |
| 3461 |
< |
nanos = cond.awaitNanos(nanos); |
| 3462 |
< |
} finally { |
| 3463 |
< |
lock.unlock(); |
| 3454 |
> |
else if (!(terminated = ((runState & TERMINATED) != 0))) { |
| 3455 |
> |
tryTerminate(false, false); // reduce transient blocking |
| 3456 |
> |
if ((lock = registrationLock) != null && |
| 3457 |
> |
!(terminated = (((runState & TERMINATED) != 0)))) { |
| 3458 |
> |
lock.lock(); |
| 3459 |
> |
try { |
| 3460 |
> |
if ((cond = termination) == null) |
| 3461 |
> |
termination = cond = lock.newCondition(); |
| 3462 |
> |
while (!(terminated = ((runState & TERMINATED) != 0)) && |
| 3463 |
> |
nanos > 0L) |
| 3464 |
> |
nanos = cond.awaitNanos(nanos); |
| 3465 |
> |
} finally { |
| 3466 |
> |
lock.unlock(); |
| 3467 |
> |
} |
| 3468 |
|
} |
| 3469 |
|
} |
| 3470 |
|
return terminated; |
| 3482 |
|
* timeout elapsed. |
| 3483 |
|
*/ |
| 3484 |
|
public boolean awaitQuiescence(long timeout, TimeUnit unit) { |
| 3485 |
< |
Thread t; ForkJoinWorkerThread wt; int q; |
| 3367 |
< |
long nanos = unit.toNanos(timeout); |
| 3368 |
< |
if ((t = Thread.currentThread()) instanceof ForkJoinWorkerThread && |
| 3369 |
< |
(wt = (ForkJoinWorkerThread)t).pool == this) |
| 3370 |
< |
q = helpQuiescePool(wt.workQueue, nanos, false); |
| 3371 |
< |
else |
| 3372 |
< |
q = externalHelpQuiescePool(nanos, false); |
| 3373 |
< |
return (q > 0); |
| 3485 |
> |
return (helpQuiescePool(this, unit.toNanos(timeout), false) > 0); |
| 3486 |
|
} |
| 3487 |
|
|
| 3488 |
|
/** |
| 3606 |
|
long c = ctl; |
| 3607 |
|
if (blocker.isReleasable()) |
| 3608 |
|
break; |
| 3609 |
< |
if ((comp = tryCompensate(c)) >= 0) { |
| 3609 |
> |
if ((comp = tryCompensate(c, false)) >= 0) { |
| 3610 |
|
long post = (comp == 0) ? 0L : RC_UNIT; |
| 3611 |
|
try { |
| 3612 |
|
done = blocker.block(); |
| 3640 |
|
return new ForkJoinTask.AdaptedCallable<T>(callable); |
| 3641 |
|
} |
| 3642 |
|
|
| 3531 |
– |
|
| 3643 |
|
static { |
| 3644 |
< |
Field pids; |
| 3644 |
> |
U = Unsafe.getUnsafe(); |
| 3645 |
> |
Class<ForkJoinPool> klass = ForkJoinPool.class; |
| 3646 |
|
try { |
| 3647 |
< |
pids = ForkJoinPool.class.getDeclaredField("poolIds"); |
| 3647 |
> |
POOLIDS = U.staticFieldOffset(klass.getDeclaredField("poolIds")); |
| 3648 |
|
} catch (NoSuchFieldException e) { |
| 3649 |
|
throw new ExceptionInInitializerError(e); |
| 3650 |
|
} |
| 3651 |
< |
U = Unsafe.getUnsafe(); |
| 3652 |
< |
POOLIDS = U.staticFieldOffset(pids); |
| 3653 |
< |
CTL = U.objectFieldOffset(ForkJoinPool.class, "ctl"); |
| 3654 |
< |
MODE = U.objectFieldOffset(ForkJoinPool.class, "mode"); |
| 3543 |
< |
THREADIDS = U.objectFieldOffset(ForkJoinPool.class, "threadIds"); |
| 3651 |
> |
CTL = U.objectFieldOffset(klass, "ctl"); |
| 3652 |
> |
RUNSTATE = U.objectFieldOffset(klass, "runState"); |
| 3653 |
> |
PARALLELISM = U.objectFieldOffset(klass, "parallelism"); |
| 3654 |
> |
THREADIDS = U.objectFieldOffset(klass, "threadIds"); |
| 3655 |
|
|
| 3545 |
– |
modifyThreadPermission = new RuntimePermission("modifyThread"); |
| 3656 |
|
defaultForkJoinWorkerThreadFactory = |
| 3657 |
|
new DefaultForkJoinWorkerThreadFactory(); |
| 3658 |
< |
// Reduce the risk of rare disastrous classloading in first call to |
| 3659 |
< |
// LockSupport.park: https://bugs.openjdk.java.net/browse/JDK-8074773 |
| 3660 |
< |
Class<?> ensureLoaded = LockSupport.class; |
| 3658 |
> |
@SuppressWarnings("removal") |
| 3659 |
> |
ForkJoinPool p = common = (System.getSecurityManager() == null) ? |
| 3660 |
> |
new ForkJoinPool((byte)0) : |
| 3661 |
> |
AccessController.doPrivileged(new PrivilegedAction<>() { |
| 3662 |
> |
public ForkJoinPool run() { |
| 3663 |
> |
return new ForkJoinPool((byte)0); }}); |
| 3664 |
> |
Class<?> dep = LockSupport.class; // ensure loaded |
| 3665 |
|
} |
| 3666 |
|
} |
