5 |
|
*/ |
6 |
|
|
7 |
|
package jsr166y; |
8 |
< |
import java.util.*; |
9 |
< |
import java.util.concurrent.*; |
10 |
< |
import java.util.concurrent.atomic.*; |
11 |
< |
import java.util.concurrent.locks.*; |
12 |
< |
import sun.misc.Unsafe; |
13 |
< |
import java.lang.reflect.*; |
8 |
> |
|
9 |
> |
import java.util.Random; |
10 |
> |
import java.util.Collection; |
11 |
> |
import java.util.concurrent.locks.LockSupport; |
12 |
> |
import java.util.concurrent.RejectedExecutionException; |
13 |
|
|
14 |
|
/** |
15 |
< |
* A thread managed by a {@link ForkJoinPool}. This class is |
16 |
< |
* subclassable solely for the sake of adding functionality -- there |
17 |
< |
* are no overridable methods dealing with scheduling or |
18 |
< |
* execution. However, you can override initialization and termination |
19 |
< |
* methods surrounding the main task processing loop. If you do |
20 |
< |
* create such a subclass, you will also need to supply a custom |
21 |
< |
* ForkJoinWorkerThreadFactory to use it in a ForkJoinPool. |
15 |
> |
* A thread managed by a {@link ForkJoinPool}, which executes |
16 |
> |
* {@link ForkJoinTask}s. |
17 |
> |
* This class is subclassable solely for the sake of adding |
18 |
> |
* functionality -- there are no overridable methods dealing with |
19 |
> |
* scheduling or execution. However, you can override initialization |
20 |
> |
* and termination methods surrounding the main task processing loop. |
21 |
> |
* If you do create such a subclass, you will also need to supply a |
22 |
> |
* custom {@link ForkJoinPool.ForkJoinWorkerThreadFactory} to use it |
23 |
> |
* in a {@code ForkJoinPool}. |
24 |
|
* |
25 |
+ |
* @since 1.7 |
26 |
+ |
* @author Doug Lea |
27 |
|
*/ |
28 |
|
public class ForkJoinWorkerThread extends Thread { |
29 |
|
/* |
30 |
< |
* Algorithm overview: |
30 |
> |
* Overview: |
31 |
> |
* |
32 |
> |
* ForkJoinWorkerThreads are managed by ForkJoinPools and perform |
33 |
> |
* ForkJoinTasks. This class includes bookkeeping in support of |
34 |
> |
* worker activation, suspension, and lifecycle control described |
35 |
> |
* in more detail in the internal documentation of class |
36 |
> |
* ForkJoinPool. And as described further below, this class also |
37 |
> |
* includes special-cased support for some ForkJoinTask |
38 |
> |
* methods. But the main mechanics involve work-stealing: |
39 |
|
* |
40 |
< |
* 1. Work-Stealing: Work-stealing queues are special forms of |
41 |
< |
* Deques that support only three of the four possible |
42 |
< |
* end-operations -- push, pop, and deq (aka steal), and only do |
43 |
< |
* so under the constraints that push and pop are called only from |
44 |
< |
* the owning thread, while deq may be called from other threads. |
45 |
< |
* (If you are unfamiliar with them, you probably want to read |
46 |
< |
* Herlihy and Shavit's book "The Art of Multiprocessor |
47 |
< |
* programming", chapter 16 describing these in more detail before |
48 |
< |
* proceeding.) The main work-stealing queue design is roughly |
49 |
< |
* similar to "Dynamic Circular Work-Stealing Deque" by David |
50 |
< |
* Chase and Yossi Lev, SPAA 2005 |
51 |
< |
* (http://research.sun.com/scalable/pubs/index.html). The main |
52 |
< |
* difference ultimately stems from gc requirements that we null |
53 |
< |
* out taken slots as soon as we can, to maintain as small a |
54 |
< |
* footprint as possible even in programs generating huge numbers |
55 |
< |
* of tasks. To accomplish this, we shift the CAS arbitrating pop |
56 |
< |
* vs deq (steal) from being on the indices ("base" and "sp") to |
57 |
< |
* the slots themselves (mainly via method "casSlotNull()"). So, |
58 |
< |
* both a successful pop and deq mainly entail CAS'ing a nonnull |
59 |
< |
* slot to null. Because we rely on CASes of references, we do |
60 |
< |
* not need tag bits on base or sp. They are simple ints as used |
61 |
< |
* in any circular array-based queue (see for example ArrayDeque). |
62 |
< |
* Updates to the indices must still be ordered in a way that |
63 |
< |
* guarantees that (sp - base) > 0 means the queue is empty, but |
64 |
< |
* otherwise may err on the side of possibly making the queue |
65 |
< |
* appear nonempty when a push, pop, or deq have not fully |
66 |
< |
* committed. Note that this means that the deq operation, |
67 |
< |
* considered individually, is not wait-free. One thief cannot |
68 |
< |
* successfully continue until another in-progress one (or, if |
69 |
< |
* previously empty, a push) completes. However, in the |
70 |
< |
* aggregate, we ensure at least probabilistic non-blockingness. If |
71 |
< |
* an attempted steal fails, a thief always chooses a different |
40 |
> |
* Work-stealing queues are special forms of Deques that support |
41 |
> |
* only three of the four possible end-operations -- push, pop, |
42 |
> |
* and deq (aka steal), under the further constraints that push |
43 |
> |
* and pop are called only from the owning thread, while deq may |
44 |
> |
* be called from other threads. (If you are unfamiliar with |
45 |
> |
* them, you probably want to read Herlihy and Shavit's book "The |
46 |
> |
* Art of Multiprocessor programming", chapter 16 describing these |
47 |
> |
* in more detail before proceeding.) The main work-stealing |
48 |
> |
* queue design is roughly similar to those in the papers "Dynamic |
49 |
> |
* Circular Work-Stealing Deque" by Chase and Lev, SPAA 2005 |
50 |
> |
* (http://research.sun.com/scalable/pubs/index.html) and |
51 |
> |
* "Idempotent work stealing" by Michael, Saraswat, and Vechev, |
52 |
> |
* PPoPP 2009 (http://portal.acm.org/citation.cfm?id=1504186). |
53 |
> |
* The main differences ultimately stem from gc requirements that |
54 |
> |
* we null out taken slots as soon as we can, to maintain as small |
55 |
> |
* a footprint as possible even in programs generating huge |
56 |
> |
* numbers of tasks. To accomplish this, we shift the CAS |
57 |
> |
* arbitrating pop vs deq (steal) from being on the indices |
58 |
> |
* ("base" and "sp") to the slots themselves (mainly via method |
59 |
> |
* "casSlotNull()"). So, both a successful pop and deq mainly |
60 |
> |
* entail a CAS of a slot from non-null to null. Because we rely |
61 |
> |
* on CASes of references, we do not need tag bits on base or sp. |
62 |
> |
* They are simple ints as used in any circular array-based queue |
63 |
> |
* (see for example ArrayDeque). Updates to the indices must |
64 |
> |
* still be ordered in a way that guarantees that sp == base means |
65 |
> |
* the queue is empty, but otherwise may err on the side of |
66 |
> |
* possibly making the queue appear nonempty when a push, pop, or |
67 |
> |
* deq have not fully committed. Note that this means that the deq |
68 |
> |
* operation, considered individually, is not wait-free. One thief |
69 |
> |
* cannot successfully continue until another in-progress one (or, |
70 |
> |
* if previously empty, a push) completes. However, in the |
71 |
> |
* aggregate, we ensure at least probabilistic non-blockingness. |
72 |
> |
* If an attempted steal fails, a thief always chooses a different |
73 |
|
* random victim target to try next. So, in order for one thief to |
74 |
|
* progress, it suffices for any in-progress deq or new push on |
75 |
|
* any empty queue to complete. One reason this works well here is |
76 |
|
* that apparently-nonempty often means soon-to-be-stealable, |
77 |
< |
* which gives threads a chance to activate if necessary before |
78 |
< |
* stealing (see below). |
77 |
> |
* which gives threads a chance to set activation status if |
78 |
> |
* necessary before stealing. |
79 |
> |
* |
80 |
> |
* This approach also enables support for "async mode" where local |
81 |
> |
* task processing is in FIFO, not LIFO order; simply by using a |
82 |
> |
* version of deq rather than pop when locallyFifo is true (as set |
83 |
> |
* by the ForkJoinPool). This allows use in message-passing |
84 |
> |
* frameworks in which tasks are never joined. |
85 |
|
* |
86 |
< |
* Efficient implementation of this approach currently relies on |
87 |
< |
* an uncomfortable amount of "Unsafe" mechanics. To maintain |
86 |
> |
* When a worker would otherwise be blocked waiting to join a |
87 |
> |
* task, it first tries a form of linear helping: Each worker |
88 |
> |
* records (in field currentSteal) the most recent task it stole |
89 |
> |
* from some other worker. Plus, it records (in field currentJoin) |
90 |
> |
* the task it is currently actively joining. Method joinTask uses |
91 |
> |
* these markers to try to find a worker to help (i.e., steal back |
92 |
> |
* a task from and execute it) that could hasten completion of the |
93 |
> |
* actively joined task. In essence, the joiner executes a task |
94 |
> |
* that would be on its own local deque had the to-be-joined task |
95 |
> |
* not been stolen. This may be seen as a conservative variant of |
96 |
> |
* the approach in Wagner & Calder "Leapfrogging: a portable |
97 |
> |
* technique for implementing efficient futures" SIGPLAN Notices, |
98 |
> |
* 1993 (http://portal.acm.org/citation.cfm?id=155354). It differs |
99 |
> |
* in that: (1) We only maintain dependency links across workers |
100 |
> |
* upon steals, rather than use per-task bookkeeping. This may |
101 |
> |
* require a linear scan of workers array to locate stealers, but |
102 |
> |
* usually doesn't because stealers leave hints (that may become |
103 |
> |
* stale/wrong) of where to locate them. This isolates cost to |
104 |
> |
* when it is needed, rather than adding to per-task overhead. |
105 |
> |
* (2) It is "shallow", ignoring nesting and potentially cyclic |
106 |
> |
* mutual steals. (3) It is intentionally racy: field currentJoin |
107 |
> |
* is updated only while actively joining, which means that we |
108 |
> |
* miss links in the chain during long-lived tasks, GC stalls etc |
109 |
> |
* (which is OK since blocking in such cases is usually a good |
110 |
> |
* idea). (4) We bound the number of attempts to find work (see |
111 |
> |
* MAX_HELP_DEPTH) and fall back to suspending the worker and if |
112 |
> |
* necessary replacing it with a spare (see |
113 |
> |
* ForkJoinPool.awaitJoin). |
114 |
> |
* |
115 |
> |
* Efficient implementation of these algorithms currently relies |
116 |
> |
* on an uncomfortable amount of "Unsafe" mechanics. To maintain |
117 |
|
* correct orderings, reads and writes of variable base require |
118 |
< |
* volatile ordering. Variable sp does not require volatile write |
119 |
< |
* but needs cheaper store-ordering on writes. Because they are |
120 |
< |
* protected by volatile base reads, reads of the queue array and |
121 |
< |
* its slots do not need volatile load semantics, but writes (in |
122 |
< |
* push) require store order and CASes (in pop and deq) require |
123 |
< |
* (volatile) CAS semantics. Since these combinations aren't |
124 |
< |
* supported using ordinary volatiles, the only way to accomplish |
125 |
< |
* these efficiently is to use direct Unsafe calls. (Using external |
126 |
< |
* AtomicIntegers and AtomicReferenceArrays for the indices and |
127 |
< |
* array is significantly slower because of memory locality and |
128 |
< |
* indirection effects.) Further, performance on most platforms is |
129 |
< |
* very sensitive to placement and sizing of the (resizable) queue |
130 |
< |
* array. Even though these queues don't usually become all that |
131 |
< |
* big, the initial size must be large enough to counteract cache |
118 |
> |
* volatile ordering. Variable sp does not require volatile |
119 |
> |
* writes but still needs store-ordering, which we accomplish by |
120 |
> |
* pre-incrementing sp before filling the slot with an ordered |
121 |
> |
* store. (Pre-incrementing also enables backouts used in |
122 |
> |
* joinTask.) Because they are protected by volatile base reads, |
123 |
> |
* reads of the queue array and its slots by other threads do not |
124 |
> |
* need volatile load semantics, but writes (in push) require |
125 |
> |
* store order and CASes (in pop and deq) require (volatile) CAS |
126 |
> |
* semantics. (Michael, Saraswat, and Vechev's algorithm has |
127 |
> |
* similar properties, but without support for nulling slots.) |
128 |
> |
* Since these combinations aren't supported using ordinary |
129 |
> |
* volatiles, the only way to accomplish these efficiently is to |
130 |
> |
* use direct Unsafe calls. (Using external AtomicIntegers and |
131 |
> |
* AtomicReferenceArrays for the indices and array is |
132 |
> |
* significantly slower because of memory locality and indirection |
133 |
> |
* effects.) |
134 |
> |
* |
135 |
> |
* Further, performance on most platforms is very sensitive to |
136 |
> |
* placement and sizing of the (resizable) queue array. Even |
137 |
> |
* though these queues don't usually become all that big, the |
138 |
> |
* initial size must be large enough to counteract cache |
139 |
|
* contention effects across multiple queues (especially in the |
140 |
|
* presence of GC cardmarking). Also, to improve thread-locality, |
141 |
< |
* queues are currently initialized immediately after the thread |
142 |
< |
* gets the initial signal to start processing tasks. However, |
143 |
< |
* all queue-related methods except pushTask are written in a way |
144 |
< |
* that allows them to instead be lazily allocated and/or disposed |
145 |
< |
* of when empty. All together, these low-level implementation |
92 |
< |
* choices produce as much as a factor of 4 performance |
93 |
< |
* improvement compared to naive implementations, and enable the |
94 |
< |
* processing of billions of tasks per second, sometimes at the |
95 |
< |
* expense of ugliness. |
96 |
< |
* |
97 |
< |
* 2. Run control: The primary run control is based on a global |
98 |
< |
* counter (activeCount) held by the pool. It uses an algorithm |
99 |
< |
* similar to that in Herlihy and Shavit section 17.6 to cause |
100 |
< |
* threads to eventually block when all threads declare they are |
101 |
< |
* inactive. (See variable "scans".) For this to work, threads |
102 |
< |
* must be declared active when executing tasks, and before |
103 |
< |
* stealing a task. They must be inactive before blocking on the |
104 |
< |
* Pool Barrier (awaiting a new submission or other Pool |
105 |
< |
* event). In between, there is some free play which we take |
106 |
< |
* advantage of to avoid contention and rapid flickering of the |
107 |
< |
* global activeCount: If inactive, we activate only if a victim |
108 |
< |
* queue appears to be nonempty (see above). Similarly, a thread |
109 |
< |
* tries to inactivate only after a full scan of other threads. |
110 |
< |
* The net effect is that contention on activeCount is rarely a |
111 |
< |
* measurable performance issue. (There are also a few other cases |
112 |
< |
* where we scan for work rather than retry/block upon |
113 |
< |
* contention.) |
114 |
< |
* |
115 |
< |
* 3. Selection control. We maintain policy of always choosing to |
116 |
< |
* run local tasks rather than stealing, and always trying to |
117 |
< |
* steal tasks before trying to run a new submission. All steals |
118 |
< |
* are currently performed in randomly-chosen deq-order. It may be |
119 |
< |
* worthwhile to bias these with locality / anti-locality |
120 |
< |
* information, but doing this well probably requires more |
121 |
< |
* lower-level information from JVMs than currently provided. |
141 |
> |
* queues are initialized after starting. All together, these |
142 |
> |
* low-level implementation choices produce as much as a factor of |
143 |
> |
* 4 performance improvement compared to naive implementations, |
144 |
> |
* and enable the processing of billions of tasks per second, |
145 |
> |
* sometimes at the expense of ugliness. |
146 |
|
*/ |
147 |
|
|
148 |
|
/** |
149 |
+ |
* Generator for initial random seeds for random victim |
150 |
+ |
* selection. This is used only to create initial seeds. Random |
151 |
+ |
* steals use a cheaper xorshift generator per steal attempt. We |
152 |
+ |
* expect only rare contention on seedGenerator, so just use a |
153 |
+ |
* plain Random. |
154 |
+ |
*/ |
155 |
+ |
private static final Random seedGenerator = new Random(); |
156 |
+ |
|
157 |
+ |
/** |
158 |
+ |
* The maximum stolen->joining link depth allowed in helpJoinTask. |
159 |
+ |
* Depths for legitimate chains are unbounded, but we use a fixed |
160 |
+ |
* constant to avoid (otherwise unchecked) cycles and bound |
161 |
+ |
* staleness of traversal parameters at the expense of sometimes |
162 |
+ |
* blocking when we could be helping. |
163 |
+ |
*/ |
164 |
+ |
private static final int MAX_HELP_DEPTH = 8; |
165 |
+ |
|
166 |
+ |
/** |
167 |
|
* Capacity of work-stealing queue array upon initialization. |
168 |
< |
* Must be a power of two. Initial size must be at least 2, but is |
168 |
> |
* Must be a power of two. Initial size must be at least 4, but is |
169 |
|
* padded to minimize cache effects. |
170 |
|
*/ |
171 |
|
private static final int INITIAL_QUEUE_CAPACITY = 1 << 13; |
172 |
|
|
173 |
|
/** |
174 |
|
* Maximum work-stealing queue array size. Must be less than or |
175 |
< |
* equal to 1 << 28 to ensure lack of index wraparound. (This |
176 |
< |
* is less than usual bounds, because we need leftshift by 3 |
177 |
< |
* to be in int range). |
175 |
> |
* equal to 1 << (31 - width of array entry) to ensure lack of |
176 |
> |
* index wraparound. The value is set in the static block |
177 |
> |
* at the end of this file after obtaining width. |
178 |
|
*/ |
179 |
< |
private static final int MAXIMUM_QUEUE_CAPACITY = 1 << 28; |
179 |
> |
private static final int MAXIMUM_QUEUE_CAPACITY; |
180 |
|
|
181 |
|
/** |
182 |
< |
* The pool this thread works in. Accessed directly by ForkJoinTask |
182 |
> |
* The pool this thread works in. Accessed directly by ForkJoinTask. |
183 |
|
*/ |
184 |
|
final ForkJoinPool pool; |
185 |
|
|
186 |
|
/** |
187 |
|
* The work-stealing queue array. Size must be a power of two. |
188 |
< |
* Initialized when thread starts, to improve memory locality. |
188 |
> |
* Initialized in onStart, to improve memory locality. |
189 |
|
*/ |
190 |
|
private ForkJoinTask<?>[] queue; |
191 |
|
|
192 |
|
/** |
193 |
+ |
* Index (mod queue.length) of least valid queue slot, which is |
194 |
+ |
* always the next position to steal from if nonempty. |
195 |
+ |
*/ |
196 |
+ |
private volatile int base; |
197 |
+ |
|
198 |
+ |
/** |
199 |
|
* Index (mod queue.length) of next queue slot to push to or pop |
200 |
< |
* from. It is written only by owner thread, via ordered store. |
201 |
< |
* Both sp and base are allowed to wrap around on overflow, but |
202 |
< |
* (sp - base) still estimates size. |
200 |
> |
* from. It is written only by owner thread, and accessed by other |
201 |
> |
* threads only after reading (volatile) base. Both sp and base |
202 |
> |
* are allowed to wrap around on overflow, but (sp - base) still |
203 |
> |
* estimates size. |
204 |
|
*/ |
205 |
< |
private volatile int sp; |
205 |
> |
private int sp; |
206 |
|
|
207 |
|
/** |
208 |
< |
* Index (mod queue.length) of least valid queue slot, which is |
209 |
< |
* always the next position to steal from if nonempty. |
208 |
> |
* The index of most recent stealer, used as a hint to avoid |
209 |
> |
* traversal in method helpJoinTask. This is only a hint because a |
210 |
> |
* worker might have had multiple steals and this only holds one |
211 |
> |
* of them (usually the most current). Declared non-volatile, |
212 |
> |
* relying on other prevailing sync to keep reasonably current. |
213 |
|
*/ |
214 |
< |
private volatile int base; |
214 |
> |
private int stealHint; |
215 |
|
|
216 |
|
/** |
217 |
< |
* Activity status. When true, this worker is considered active. |
218 |
< |
* Must be false upon construction. It must be true when executing |
219 |
< |
* tasks, and BEFORE stealing a task. It must be false before |
220 |
< |
* calling pool.sync |
217 |
> |
* Run state of this worker. In addition to the usual run levels, |
218 |
> |
* tracks if this worker is suspended as a spare, and if it was |
219 |
> |
* killed (trimmed) while suspended. However, "active" status is |
220 |
> |
* maintained separately and modified only in conjunction with |
221 |
> |
* CASes of the pool's runState (which are currently sadly |
222 |
> |
* manually inlined for performance.) Accessed directly by pool |
223 |
> |
* to simplify checks for normal (zero) status. |
224 |
|
*/ |
225 |
< |
private boolean active; |
225 |
> |
volatile int runState; |
226 |
> |
|
227 |
> |
private static final int TERMINATING = 0x01; |
228 |
> |
private static final int TERMINATED = 0x02; |
229 |
> |
private static final int SUSPENDED = 0x04; // inactive spare |
230 |
> |
private static final int TRIMMED = 0x08; // killed while suspended |
231 |
|
|
232 |
|
/** |
233 |
< |
* Run state of this worker. Supports simple versions of the usual |
234 |
< |
* shutdown/shutdownNow control. |
233 |
> |
* Number of steals. Directly accessed (and reset) by |
234 |
> |
* pool.tryAccumulateStealCount when idle. |
235 |
|
*/ |
236 |
< |
private volatile int runState; |
236 |
> |
int stealCount; |
237 |
|
|
238 |
|
/** |
239 |
|
* Seed for random number generator for choosing steal victims. |
240 |
< |
* Uses Marsaglia xorshift. Must be nonzero upon initialization. |
240 |
> |
* Uses Marsaglia xorshift. Must be initialized as nonzero. |
241 |
|
*/ |
242 |
|
private int seed; |
243 |
|
|
244 |
|
/** |
245 |
< |
* Number of steals, transferred to pool when idle |
245 |
> |
* Activity status. When true, this worker is considered active. |
246 |
> |
* Accessed directly by pool. Must be false upon construction. |
247 |
|
*/ |
248 |
< |
private int stealCount; |
248 |
> |
boolean active; |
249 |
> |
|
250 |
> |
/** |
251 |
> |
* True if use local fifo, not default lifo, for local polling. |
252 |
> |
* Shadows value from ForkJoinPool. |
253 |
> |
*/ |
254 |
> |
private final boolean locallyFifo; |
255 |
|
|
256 |
|
/** |
257 |
|
* Index of this worker in pool array. Set once by pool before |
258 |
< |
* running, and accessed directly by pool during cleanup etc |
258 |
> |
* running, and accessed directly by pool to locate this worker in |
259 |
> |
* its workers array. |
260 |
|
*/ |
261 |
|
int poolIndex; |
262 |
|
|
263 |
|
/** |
264 |
< |
* The last barrier event waited for. Accessed in pool callback |
265 |
< |
* methods, but only by current thread. |
264 |
> |
* The last pool event waited for. Accessed only by pool in |
265 |
> |
* callback methods invoked within this thread. |
266 |
> |
*/ |
267 |
> |
int lastEventCount; |
268 |
> |
|
269 |
> |
/** |
270 |
> |
* Encoded index and event count of next event waiter. Accessed |
271 |
> |
* only by ForkJoinPool for managing event waiters. |
272 |
|
*/ |
273 |
< |
long lastEventCount; |
273 |
> |
volatile long nextWaiter; |
274 |
|
|
275 |
|
/** |
276 |
< |
* True if use local fifo, not default lifo, for local polling |
276 |
> |
* Number of times this thread suspended as spare. Accessed only |
277 |
> |
* by pool. |
278 |
|
*/ |
279 |
< |
private boolean locallyFifo; |
279 |
> |
int spareCount; |
280 |
> |
|
281 |
> |
/** |
282 |
> |
* Encoded index and count of next spare waiter. Accessed only |
283 |
> |
* by ForkJoinPool for managing spares. |
284 |
> |
*/ |
285 |
> |
volatile int nextSpare; |
286 |
> |
|
287 |
> |
/** |
288 |
> |
* The task currently being joined, set only when actively trying |
289 |
> |
* to help other stealers in helpJoinTask. Written only by this |
290 |
> |
* thread, but read by others. |
291 |
> |
*/ |
292 |
> |
private volatile ForkJoinTask<?> currentJoin; |
293 |
> |
|
294 |
> |
/** |
295 |
> |
* The task most recently stolen from another worker (or |
296 |
> |
* submission queue). Written only by this thread, but read by |
297 |
> |
* others. |
298 |
> |
*/ |
299 |
> |
private volatile ForkJoinTask<?> currentSteal; |
300 |
|
|
301 |
|
/** |
302 |
|
* Creates a ForkJoinWorkerThread operating in the given pool. |
303 |
+ |
* |
304 |
|
* @param pool the pool this thread works in |
305 |
|
* @throws NullPointerException if pool is null |
306 |
|
*/ |
307 |
|
protected ForkJoinWorkerThread(ForkJoinPool pool) { |
212 |
– |
if (pool == null) throw new NullPointerException(); |
308 |
|
this.pool = pool; |
309 |
< |
// Note: poolIndex is set by pool during construction |
310 |
< |
// Remaining initialization is deferred to onStart |
309 |
> |
this.locallyFifo = pool.locallyFifo; |
310 |
> |
setDaemon(true); |
311 |
> |
// To avoid exposing construction details to subclasses, |
312 |
> |
// remaining initialization is in start() and onStart() |
313 |
|
} |
314 |
|
|
315 |
< |
// Public access methods |
315 |
> |
/** |
316 |
> |
* Performs additional initialization and starts this thread. |
317 |
> |
*/ |
318 |
> |
final void start(int poolIndex, UncaughtExceptionHandler ueh) { |
319 |
> |
this.poolIndex = poolIndex; |
320 |
> |
if (ueh != null) |
321 |
> |
setUncaughtExceptionHandler(ueh); |
322 |
> |
start(); |
323 |
> |
} |
324 |
> |
|
325 |
> |
// Public/protected methods |
326 |
|
|
327 |
|
/** |
328 |
< |
* Returns the pool hosting this thread |
328 |
> |
* Returns the pool hosting this thread. |
329 |
> |
* |
330 |
|
* @return the pool |
331 |
|
*/ |
332 |
|
public ForkJoinPool getPool() { |
339 |
|
* threads (minus one) that have ever been created in the pool. |
340 |
|
* This method may be useful for applications that track status or |
341 |
|
* collect results per-worker rather than per-task. |
342 |
< |
* @return the index number. |
342 |
> |
* |
343 |
> |
* @return the index number |
344 |
|
*/ |
345 |
|
public int getPoolIndex() { |
346 |
|
return poolIndex; |
347 |
|
} |
348 |
|
|
349 |
|
/** |
350 |
< |
* Establishes local first-in-first-out scheduling mode for forked |
351 |
< |
* tasks that are never joined. |
352 |
< |
* @param async if true, use locally FIFO scheduling |
350 |
> |
* Initializes internal state after construction but before |
351 |
> |
* processing any tasks. If you override this method, you must |
352 |
> |
* invoke {@code super.onStart()} at the beginning of the method. |
353 |
> |
* Initialization requires care: Most fields must have legal |
354 |
> |
* default values, to ensure that attempted accesses from other |
355 |
> |
* threads work correctly even before this thread starts |
356 |
> |
* processing tasks. |
357 |
|
*/ |
358 |
< |
void setAsyncMode(boolean async) { |
359 |
< |
locallyFifo = async; |
360 |
< |
} |
248 |
< |
|
249 |
< |
// Runstate management |
250 |
< |
|
251 |
< |
// Runstate values. Order matters |
252 |
< |
private static final int RUNNING = 0; |
253 |
< |
private static final int SHUTDOWN = 1; |
254 |
< |
private static final int TERMINATING = 2; |
255 |
< |
private static final int TERMINATED = 3; |
256 |
< |
|
257 |
< |
final boolean isShutdown() { return runState >= SHUTDOWN; } |
258 |
< |
final boolean isTerminating() { return runState >= TERMINATING; } |
259 |
< |
final boolean isTerminated() { return runState == TERMINATED; } |
260 |
< |
final boolean shutdown() { return transitionRunStateTo(SHUTDOWN); } |
261 |
< |
final boolean shutdownNow() { return transitionRunStateTo(TERMINATING); } |
358 |
> |
protected void onStart() { |
359 |
> |
int rs = seedGenerator.nextInt(); |
360 |
> |
seed = rs == 0? 1 : rs; // seed must be nonzero |
361 |
|
|
362 |
< |
/** |
363 |
< |
* Transition to at least the given state. Return true if not |
364 |
< |
* already at least given state. |
365 |
< |
*/ |
267 |
< |
private boolean transitionRunStateTo(int state) { |
268 |
< |
for (;;) { |
269 |
< |
int s = runState; |
270 |
< |
if (s >= state) |
271 |
< |
return false; |
272 |
< |
if (_unsafe.compareAndSwapInt(this, runStateOffset, s, state)) |
273 |
< |
return true; |
274 |
< |
} |
275 |
< |
} |
362 |
> |
// Allocate name string and arrays in this thread |
363 |
> |
String pid = Integer.toString(pool.getPoolNumber()); |
364 |
> |
String wid = Integer.toString(poolIndex); |
365 |
> |
setName("ForkJoinPool-" + pid + "-worker-" + wid); |
366 |
|
|
367 |
< |
/** |
278 |
< |
* Try to set status to active; fail on contention |
279 |
< |
*/ |
280 |
< |
private boolean tryActivate() { |
281 |
< |
if (!active) { |
282 |
< |
if (!pool.tryIncrementActiveCount()) |
283 |
< |
return false; |
284 |
< |
active = true; |
285 |
< |
} |
286 |
< |
return true; |
367 |
> |
queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; |
368 |
|
} |
369 |
|
|
370 |
|
/** |
371 |
< |
* Try to set status to active; fail on contention |
371 |
> |
* Performs cleanup associated with termination of this worker |
372 |
> |
* thread. If you override this method, you must invoke |
373 |
> |
* {@code super.onTermination} at the end of the overridden method. |
374 |
> |
* |
375 |
> |
* @param exception the exception causing this thread to abort due |
376 |
> |
* to an unrecoverable error, or {@code null} if completed normally |
377 |
|
*/ |
378 |
< |
private boolean tryInactivate() { |
379 |
< |
if (active) { |
380 |
< |
if (!pool.tryDecrementActiveCount()) |
381 |
< |
return false; |
382 |
< |
active = false; |
378 |
> |
protected void onTermination(Throwable exception) { |
379 |
> |
try { |
380 |
> |
ForkJoinPool p = pool; |
381 |
> |
if (active) { |
382 |
> |
int a; // inline p.tryDecrementActiveCount |
383 |
> |
active = false; |
384 |
> |
do {} while (!UNSAFE.compareAndSwapInt |
385 |
> |
(p, poolRunStateOffset, a = p.runState, a - 1)); |
386 |
> |
} |
387 |
> |
cancelTasks(); |
388 |
> |
setTerminated(); |
389 |
> |
p.workerTerminated(this); |
390 |
> |
} catch (Throwable ex) { // Shouldn't ever happen |
391 |
> |
if (exception == null) // but if so, at least rethrown |
392 |
> |
exception = ex; |
393 |
> |
} finally { |
394 |
> |
if (exception != null) |
395 |
> |
UNSAFE.throwException(exception); |
396 |
|
} |
298 |
– |
return true; |
299 |
– |
} |
300 |
– |
|
301 |
– |
/** |
302 |
– |
* Computes next value for random victim probe. Scans don't |
303 |
– |
* require a very high quality generator, but also not a crummy |
304 |
– |
* one. Marsaglia xor-shift is cheap and works well. |
305 |
– |
*/ |
306 |
– |
private static int xorShift(int r) { |
307 |
– |
r ^= r << 1; |
308 |
– |
r ^= r >>> 3; |
309 |
– |
r ^= r << 10; |
310 |
– |
return r; |
397 |
|
} |
398 |
|
|
313 |
– |
// Lifecycle methods |
314 |
– |
|
399 |
|
/** |
400 |
|
* This method is required to be public, but should never be |
401 |
|
* called explicitly. It performs the main run loop to execute |
402 |
< |
* ForkJoinTasks. |
402 |
> |
* {@link ForkJoinTask}s. |
403 |
|
*/ |
404 |
|
public void run() { |
405 |
|
Throwable exception = null; |
406 |
|
try { |
407 |
|
onStart(); |
324 |
– |
pool.sync(this); // await first pool event |
408 |
|
mainLoop(); |
409 |
|
} catch (Throwable ex) { |
410 |
|
exception = ex; |
413 |
|
} |
414 |
|
} |
415 |
|
|
416 |
+ |
// helpers for run() |
417 |
+ |
|
418 |
|
/** |
419 |
< |
* Execute tasks until shut down. |
419 |
> |
* Finds and executes tasks, and checks status while running. |
420 |
|
*/ |
421 |
|
private void mainLoop() { |
422 |
< |
while (!isShutdown()) { |
423 |
< |
ForkJoinTask<?> t = pollTask(); |
424 |
< |
if (t != null || (t = pollSubmission()) != null) |
425 |
< |
t.quietlyExec(); |
426 |
< |
else if (tryInactivate()) |
427 |
< |
pool.sync(this); |
422 |
> |
boolean ran = false; // true if ran a task on last step |
423 |
> |
ForkJoinPool p = pool; |
424 |
> |
for (;;) { |
425 |
> |
p.preStep(this, ran); |
426 |
> |
if (runState != 0) |
427 |
> |
break; |
428 |
> |
ran = tryExecSteal() || tryExecSubmission(); |
429 |
|
} |
430 |
|
} |
431 |
|
|
432 |
|
/** |
433 |
< |
* Initializes internal state after construction but before |
434 |
< |
* processing any tasks. If you override this method, you must |
435 |
< |
* invoke super.onStart() at the beginning of the method. |
350 |
< |
* Initialization requires care: Most fields must have legal |
351 |
< |
* default values, to ensure that attempted accesses from other |
352 |
< |
* threads work correctly even before this thread starts |
353 |
< |
* processing tasks. |
433 |
> |
* Tries to steal a task and execute it. |
434 |
> |
* |
435 |
> |
* @return true if ran a task |
436 |
|
*/ |
437 |
< |
protected void onStart() { |
438 |
< |
// Allocate while starting to improve chances of thread-local |
439 |
< |
// isolation |
440 |
< |
queue = new ForkJoinTask<?>[INITIAL_QUEUE_CAPACITY]; |
441 |
< |
// Initial value of seed need not be especially random but |
442 |
< |
// should differ across workers and must be nonzero |
443 |
< |
int p = poolIndex + 1; |
444 |
< |
seed = p + (p << 8) + (p << 16) + (p << 24); // spread bits |
437 |
> |
private boolean tryExecSteal() { |
438 |
> |
ForkJoinTask<?> t; |
439 |
> |
if ((t = scan()) != null) { |
440 |
> |
t.quietlyExec(); |
441 |
> |
UNSAFE.putOrderedObject(this, currentStealOffset, null); |
442 |
> |
if (sp != base) |
443 |
> |
execLocalTasks(); |
444 |
> |
return true; |
445 |
> |
} |
446 |
> |
return false; |
447 |
|
} |
448 |
|
|
449 |
|
/** |
450 |
< |
* Perform cleanup associated with termination of this worker |
367 |
< |
* thread. If you override this method, you must invoke |
368 |
< |
* super.onTermination at the end of the overridden method. |
450 |
> |
* If a submission exists, try to activate and run it. |
451 |
|
* |
452 |
< |
* @param exception the exception causing this thread to abort due |
371 |
< |
* to an unrecoverable error, or null if completed normally. |
452 |
> |
* @return true if ran a task |
453 |
|
*/ |
454 |
< |
protected void onTermination(Throwable exception) { |
455 |
< |
// Execute remaining local tasks unless aborting or terminating |
456 |
< |
while (exception == null && !pool.isTerminating() && base != sp) { |
457 |
< |
try { |
458 |
< |
ForkJoinTask<?> t = popTask(); |
459 |
< |
if (t != null) |
454 |
> |
private boolean tryExecSubmission() { |
455 |
> |
ForkJoinPool p = pool; |
456 |
> |
// This loop is needed in case attempt to activate fails, in |
457 |
> |
// which case we only retry if there still appears to be a |
458 |
> |
// submission. |
459 |
> |
while (p.hasQueuedSubmissions()) { |
460 |
> |
ForkJoinTask<?> t; int a; |
461 |
> |
if (active || // inline p.tryIncrementActiveCount |
462 |
> |
(active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset, |
463 |
> |
a = p.runState, a + 1))) { |
464 |
> |
if ((t = p.pollSubmission()) != null) { |
465 |
> |
UNSAFE.putOrderedObject(this, currentStealOffset, t); |
466 |
|
t.quietlyExec(); |
467 |
< |
} catch(Throwable ex) { |
468 |
< |
exception = ex; |
467 |
> |
UNSAFE.putOrderedObject(this, currentStealOffset, null); |
468 |
> |
if (sp != base) |
469 |
> |
execLocalTasks(); |
470 |
> |
return true; |
471 |
> |
} |
472 |
|
} |
473 |
|
} |
474 |
< |
// Cancel other tasks, transition status, notify pool, and |
385 |
< |
// propagate exception to uncaught exception handler |
386 |
< |
try { |
387 |
< |
do;while (!tryInactivate()); // ensure inactive |
388 |
< |
cancelTasks(); |
389 |
< |
runState = TERMINATED; |
390 |
< |
pool.workerTerminated(this); |
391 |
< |
} catch (Throwable ex) { // Shouldn't ever happen |
392 |
< |
if (exception == null) // but if so, at least rethrown |
393 |
< |
exception = ex; |
394 |
< |
} finally { |
395 |
< |
if (exception != null) |
396 |
< |
ForkJoinTask.rethrowException(exception); |
397 |
< |
} |
474 |
> |
return false; |
475 |
|
} |
476 |
|
|
400 |
– |
// Intrinsics-based support for queue operations. |
401 |
– |
|
477 |
|
/** |
478 |
< |
* Add in store-order the given task at given slot of q to |
479 |
< |
* null. Caller must ensure q is nonnull and index is in range. |
478 |
> |
* Runs local tasks until queue is empty or shut down. Call only |
479 |
> |
* while active. |
480 |
|
*/ |
481 |
< |
private static void setSlot(ForkJoinTask<?>[] q, int i, |
482 |
< |
ForkJoinTask<?> t){ |
483 |
< |
_unsafe.putOrderedObject(q, (i << qShift) + qBase, t); |
481 |
> |
private void execLocalTasks() { |
482 |
> |
while (runState == 0) { |
483 |
> |
ForkJoinTask<?> t = locallyFifo ? locallyDeqTask() : popTask(); |
484 |
> |
if (t != null) |
485 |
> |
t.quietlyExec(); |
486 |
> |
else if (sp == base) |
487 |
> |
break; |
488 |
> |
} |
489 |
|
} |
490 |
|
|
491 |
+ |
/* |
492 |
+ |
* Intrinsics-based atomic writes for queue slots. These are |
493 |
+ |
* basically the same as methods in AtomicReferenceArray, but |
494 |
+ |
* specialized for (1) ForkJoinTask elements (2) requirement that |
495 |
+ |
* nullness and bounds checks have already been performed by |
496 |
+ |
* callers and (3) effective offsets are known not to overflow |
497 |
+ |
* from int to long (because of MAXIMUM_QUEUE_CAPACITY). We don't |
498 |
+ |
* need corresponding version for reads: plain array reads are OK |
499 |
+ |
* because they are protected by other volatile reads and are |
500 |
+ |
* confirmed by CASes. |
501 |
+ |
* |
502 |
+ |
* Most uses don't actually call these methods, but instead contain |
503 |
+ |
* inlined forms that enable more predictable optimization. We |
504 |
+ |
* don't define the version of write used in pushTask at all, but |
505 |
+ |
* instead inline there a store-fenced array slot write. |
506 |
+ |
*/ |
507 |
+ |
|
508 |
|
/** |
509 |
< |
* CAS given slot of q to null. Caller must ensure q is nonnull |
510 |
< |
* and index is in range. |
509 |
> |
* CASes slot i of array q from t to null. Caller must ensure q is |
510 |
> |
* non-null and index is in range. |
511 |
|
*/ |
512 |
< |
private static boolean casSlotNull(ForkJoinTask<?>[] q, int i, |
513 |
< |
ForkJoinTask<?> t) { |
514 |
< |
return _unsafe.compareAndSwapObject(q, (i << qShift) + qBase, t, null); |
512 |
> |
private static final boolean casSlotNull(ForkJoinTask<?>[] q, int i, |
513 |
> |
ForkJoinTask<?> t) { |
514 |
> |
return UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null); |
515 |
|
} |
516 |
|
|
517 |
|
/** |
518 |
< |
* Sets sp in store-order. |
518 |
> |
* Performs a volatile write of the given task at given slot of |
519 |
> |
* array q. Caller must ensure q is non-null and index is in |
520 |
> |
* range. This method is used only during resets and backouts. |
521 |
|
*/ |
522 |
< |
private void storeSp(int s) { |
523 |
< |
_unsafe.putOrderedInt(this, spOffset, s); |
522 |
> |
private static final void writeSlot(ForkJoinTask<?>[] q, int i, |
523 |
> |
ForkJoinTask<?> t) { |
524 |
> |
UNSAFE.putObjectVolatile(q, (i << qShift) + qBase, t); |
525 |
|
} |
526 |
|
|
527 |
< |
// Main queue methods |
527 |
> |
// queue methods |
528 |
|
|
529 |
|
/** |
530 |
< |
* Pushes a task. Called only by current thread. |
531 |
< |
* @param t the task. Caller must ensure nonnull |
530 |
> |
* Pushes a task. Call only from this thread. |
531 |
> |
* |
532 |
> |
* @param t the task. Caller must ensure non-null. |
533 |
|
*/ |
534 |
|
final void pushTask(ForkJoinTask<?> t) { |
535 |
|
ForkJoinTask<?>[] q = queue; |
536 |
< |
int mask = q.length - 1; |
537 |
< |
int s = sp; |
538 |
< |
setSlot(q, s & mask, t); |
539 |
< |
storeSp(++s); |
540 |
< |
if ((s -= base) == 1) |
541 |
< |
pool.signalWork(); |
542 |
< |
else if (s >= mask) |
442 |
< |
growQueue(); |
536 |
> |
int mask = q.length - 1; // implicit assert q != null |
537 |
> |
int s = sp++; // ok to increment sp before slot write |
538 |
> |
UNSAFE.putOrderedObject(q, ((s & mask) << qShift) + qBase, t); |
539 |
> |
if ((s -= base) == 0) |
540 |
> |
pool.signalWork(); // was empty |
541 |
> |
else if (s == mask) |
542 |
> |
growQueue(); // is full |
543 |
|
} |
544 |
|
|
545 |
|
/** |
546 |
|
* Tries to take a task from the base of the queue, failing if |
547 |
< |
* either empty or contended. |
548 |
< |
* @return a task, or null if none or contended. |
547 |
> |
* empty or contended. Note: Specializations of this code appear |
548 |
> |
* in locallyDeqTask and elsewhere. |
549 |
> |
* |
550 |
> |
* @return a task, or null if none or contended |
551 |
|
*/ |
552 |
|
final ForkJoinTask<?> deqTask() { |
553 |
|
ForkJoinTask<?> t; |
554 |
|
ForkJoinTask<?>[] q; |
555 |
< |
int i; |
454 |
< |
int b; |
555 |
> |
int b, i; |
556 |
|
if (sp != (b = base) && |
557 |
|
(q = queue) != null && // must read q after b |
558 |
< |
(t = q[i = (q.length - 1) & b]) != null && |
559 |
< |
casSlotNull(q, i, t)) { |
558 |
> |
(t = q[i = (q.length - 1) & b]) != null && base == b && |
559 |
> |
UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, t, null)) { |
560 |
|
base = b + 1; |
561 |
|
return t; |
562 |
|
} |
564 |
|
} |
565 |
|
|
566 |
|
/** |
567 |
< |
* Returns a popped task, or null if empty. Ensures active status |
568 |
< |
* if nonnull. Called only by current thread. |
567 |
> |
* Tries to take a task from the base of own queue. Assumes active |
568 |
> |
* status. Called only by this thread. |
569 |
> |
* |
570 |
> |
* @return a task, or null if none |
571 |
> |
*/ |
572 |
> |
final ForkJoinTask<?> locallyDeqTask() { |
573 |
> |
ForkJoinTask<?>[] q = queue; |
574 |
> |
if (q != null) { |
575 |
> |
ForkJoinTask<?> t; |
576 |
> |
int b, i; |
577 |
> |
while (sp != (b = base)) { |
578 |
> |
if ((t = q[i = (q.length - 1) & b]) != null && base == b && |
579 |
> |
UNSAFE.compareAndSwapObject(q, (i << qShift) + qBase, |
580 |
> |
t, null)) { |
581 |
> |
base = b + 1; |
582 |
> |
return t; |
583 |
> |
} |
584 |
> |
} |
585 |
> |
} |
586 |
> |
return null; |
587 |
> |
} |
588 |
> |
|
589 |
> |
/** |
590 |
> |
* Returns a popped task, or null if empty. Assumes active status. |
591 |
> |
* Called only by this thread. |
592 |
|
*/ |
593 |
< |
final ForkJoinTask<?> popTask() { |
594 |
< |
int s = sp; |
595 |
< |
while (s != base) { |
596 |
< |
if (tryActivate()) { |
597 |
< |
ForkJoinTask<?>[] q = queue; |
598 |
< |
int mask = q.length - 1; |
599 |
< |
int i = (s - 1) & mask; |
593 |
> |
private ForkJoinTask<?> popTask() { |
594 |
> |
ForkJoinTask<?>[] q = queue; |
595 |
> |
if (q != null) { |
596 |
> |
int s; |
597 |
> |
while ((s = sp) != base) { |
598 |
> |
int i = (q.length - 1) & --s; |
599 |
> |
long u = (i << qShift) + qBase; // raw offset |
600 |
|
ForkJoinTask<?> t = q[i]; |
601 |
< |
if (t == null || !casSlotNull(q, i, t)) |
601 |
> |
if (t == null) // lost to stealer |
602 |
|
break; |
603 |
< |
storeSp(s - 1); |
604 |
< |
return t; |
603 |
> |
if (UNSAFE.compareAndSwapObject(q, u, t, null)) { |
604 |
> |
/* |
605 |
> |
* Note: here and in related methods, as a |
606 |
> |
* performance (not correctness) issue, we'd like |
607 |
> |
* to encourage compiler not to arbitrarily |
608 |
> |
* postpone setting sp after successful CAS. |
609 |
> |
* Currently there is no intrinsic for arranging |
610 |
> |
* this, but using Unsafe putOrderedInt may be a |
611 |
> |
* preferable strategy on some compilers even |
612 |
> |
* though its main effect is a pre-, not post- |
613 |
> |
* fence. To simplify possible changes, the option |
614 |
> |
* is left in comments next to the associated |
615 |
> |
* assignments. |
616 |
> |
*/ |
617 |
> |
sp = s; // putOrderedInt may encourage more timely write |
618 |
> |
// UNSAFE.putOrderedInt(this, spOffset, s); |
619 |
> |
return t; |
620 |
> |
} |
621 |
|
} |
622 |
|
} |
623 |
|
return null; |
624 |
|
} |
625 |
|
|
626 |
|
/** |
627 |
< |
* Specialized version of popTask to pop only if |
628 |
< |
* topmost element is the given task. Called only |
629 |
< |
* by current thread while active. |
630 |
< |
* @param t the task. Caller must ensure nonnull |
627 |
> |
* Specialized version of popTask to pop only if topmost element |
628 |
> |
* is the given task. Called only by this thread while active. |
629 |
> |
* |
630 |
> |
* @param t the task. Caller must ensure non-null. |
631 |
|
*/ |
632 |
|
final boolean unpushTask(ForkJoinTask<?> t) { |
633 |
+ |
int s; |
634 |
|
ForkJoinTask<?>[] q = queue; |
635 |
< |
int mask = q.length - 1; |
636 |
< |
int s = sp - 1; |
637 |
< |
if (casSlotNull(q, s & mask, t)) { |
638 |
< |
storeSp(s); |
635 |
> |
if ((s = sp) != base && q != null && |
636 |
> |
UNSAFE.compareAndSwapObject |
637 |
> |
(q, (((q.length - 1) & --s) << qShift) + qBase, t, null)) { |
638 |
> |
sp = s; // putOrderedInt may encourage more timely write |
639 |
> |
// UNSAFE.putOrderedInt(this, spOffset, s); |
640 |
|
return true; |
641 |
|
} |
642 |
|
return false; |
643 |
|
} |
644 |
|
|
645 |
|
/** |
646 |
< |
* Returns next task. |
646 |
> |
* Returns next task, or null if empty or contended. |
647 |
|
*/ |
648 |
|
final ForkJoinTask<?> peekTask() { |
649 |
|
ForkJoinTask<?>[] q = queue; |
650 |
|
if (q == null) |
651 |
|
return null; |
652 |
|
int mask = q.length - 1; |
653 |
< |
int i = locallyFifo? base : (sp - 1); |
653 |
> |
int i = locallyFifo ? base : (sp - 1); |
654 |
|
return q[i & mask]; |
655 |
|
} |
656 |
|
|
676 |
|
ForkJoinTask<?> t = oldQ[oldIndex]; |
677 |
|
if (t != null && !casSlotNull(oldQ, oldIndex, t)) |
678 |
|
t = null; |
679 |
< |
setSlot(newQ, b & newMask, t); |
679 |
> |
writeSlot(newQ, b & newMask, t); |
680 |
|
} while (++b != bf); |
681 |
|
pool.signalWork(); |
682 |
|
} |
683 |
|
|
684 |
|
/** |
685 |
+ |
* Computes next value for random victim probe in scan(). Scans |
686 |
+ |
* don't require a very high quality generator, but also not a |
687 |
+ |
* crummy one. Marsaglia xor-shift is cheap and works well enough. |
688 |
+ |
* Note: This is manually inlined in scan(). |
689 |
+ |
*/ |
690 |
+ |
private static final int xorShift(int r) { |
691 |
+ |
r ^= r << 13; |
692 |
+ |
r ^= r >>> 17; |
693 |
+ |
return r ^ (r << 5); |
694 |
+ |
} |
695 |
+ |
|
696 |
+ |
/** |
697 |
|
* Tries to steal a task from another worker. Starts at a random |
698 |
|
* index of workers array, and probes workers until finding one |
699 |
|
* with non-empty queue or finding that all are empty. It |
700 |
|
* randomly selects the first n probes. If these are empty, it |
701 |
< |
* resorts to a full circular traversal, which is necessary to |
702 |
< |
* accurately set active status by caller. Also restarts if pool |
703 |
< |
* events occurred since last scan, which forces refresh of |
704 |
< |
* workers array, in case barrier was associated with resize. |
701 |
> |
* resorts to a circular sweep, which is necessary to accurately |
702 |
> |
* set active status. (The circular sweep uses steps of |
703 |
> |
* approximately half the array size plus 1, to avoid bias |
704 |
> |
* stemming from leftmost packing of the array in ForkJoinPool.) |
705 |
|
* |
706 |
|
* This method must be both fast and quiet -- usually avoiding |
707 |
|
* memory accesses that could disrupt cache sharing etc other than |
708 |
< |
* those needed to check for and take tasks. This accounts for, |
709 |
< |
* among other things, updating random seed in place without |
710 |
< |
* storing it until exit. |
708 |
> |
* those needed to check for and take tasks (or to activate if not |
709 |
> |
* already active). This accounts for, among other things, |
710 |
> |
* updating random seed in place without storing it until exit. |
711 |
|
* |
712 |
|
* @return a task, or null if none found |
713 |
|
*/ |
714 |
|
private ForkJoinTask<?> scan() { |
715 |
< |
ForkJoinTask<?> t = null; |
716 |
< |
int r = seed; // extract once to keep scan quiet |
717 |
< |
ForkJoinWorkerThread[] ws; // refreshed on outer loop |
718 |
< |
int mask; // must be power 2 minus 1 and > 0 |
719 |
< |
outer:do { |
720 |
< |
if ((ws = pool.workers) != null && (mask = ws.length - 1) > 0) { |
721 |
< |
int idx = r; |
722 |
< |
int probes = ~mask; // use random index while negative |
723 |
< |
for (;;) { |
724 |
< |
r = xorShift(r); // update random seed |
725 |
< |
ForkJoinWorkerThread v = ws[mask & idx]; |
726 |
< |
if (v == null || v.sp == v.base) { |
727 |
< |
if (probes <= mask) |
728 |
< |
idx = (probes++ < 0)? r : (idx + 1); |
729 |
< |
else |
730 |
< |
break; |
715 |
> |
ForkJoinPool p = pool; |
716 |
> |
ForkJoinWorkerThread[] ws; // worker array |
717 |
> |
int n; // upper bound of #workers |
718 |
> |
if ((ws = p.workers) != null && (n = ws.length) > 1) { |
719 |
> |
boolean canSteal = active; // shadow active status |
720 |
> |
int r = seed; // extract seed once |
721 |
> |
int mask = n - 1; |
722 |
> |
int j = -n; // loop counter |
723 |
> |
int k = r; // worker index, random if j < 0 |
724 |
> |
for (;;) { |
725 |
> |
ForkJoinWorkerThread v = ws[k & mask]; |
726 |
> |
r ^= r << 13; r ^= r >>> 17; r ^= r << 5; // inline xorshift |
727 |
> |
ForkJoinTask<?>[] q; ForkJoinTask<?> t; int b, a; |
728 |
> |
if (v != null && (b = v.base) != v.sp && |
729 |
> |
(q = v.queue) != null) { |
730 |
> |
int i = (q.length - 1) & b; |
731 |
> |
long u = (i << qShift) + qBase; // raw offset |
732 |
> |
int pid = poolIndex; |
733 |
> |
if ((t = q[i]) != null) { |
734 |
> |
if (!canSteal && // inline p.tryIncrementActiveCount |
735 |
> |
UNSAFE.compareAndSwapInt(p, poolRunStateOffset, |
736 |
> |
a = p.runState, a + 1)) |
737 |
> |
canSteal = active = true; |
738 |
> |
if (canSteal && v.base == b++ && |
739 |
> |
UNSAFE.compareAndSwapObject(q, u, t, null)) { |
740 |
> |
v.base = b; |
741 |
> |
v.stealHint = pid; |
742 |
> |
UNSAFE.putOrderedObject(this, |
743 |
> |
currentStealOffset, t); |
744 |
> |
seed = r; |
745 |
> |
++stealCount; |
746 |
> |
return t; |
747 |
> |
} |
748 |
|
} |
749 |
< |
else if (!tryActivate() || (t = v.deqTask()) == null) |
750 |
< |
continue outer; // restart on contention |
580 |
< |
else |
581 |
< |
break outer; |
749 |
> |
j = -n; |
750 |
> |
k = r; // restart on contention |
751 |
|
} |
752 |
+ |
else if (++j <= 0) |
753 |
+ |
k = r; |
754 |
+ |
else if (j <= n) |
755 |
+ |
k += (n >>> 1) | 1; |
756 |
+ |
else |
757 |
+ |
break; |
758 |
|
} |
759 |
< |
} while (pool.hasNewSyncEvent(this)); // retry on pool events |
760 |
< |
seed = r; |
761 |
< |
return t; |
759 |
> |
} |
760 |
> |
return null; |
761 |
> |
} |
762 |
> |
|
763 |
> |
// Run State management |
764 |
> |
|
765 |
> |
// status check methods used mainly by ForkJoinPool |
766 |
> |
final boolean isRunning() { return runState == 0; } |
767 |
> |
final boolean isTerminated() { return (runState & TERMINATED) != 0; } |
768 |
> |
final boolean isSuspended() { return (runState & SUSPENDED) != 0; } |
769 |
> |
final boolean isTrimmed() { return (runState & TRIMMED) != 0; } |
770 |
> |
|
771 |
> |
final boolean isTerminating() { |
772 |
> |
if ((runState & TERMINATING) != 0) |
773 |
> |
return true; |
774 |
> |
if (pool.isAtLeastTerminating()) { // propagate pool state |
775 |
> |
shutdown(); |
776 |
> |
return true; |
777 |
> |
} |
778 |
> |
return false; |
779 |
|
} |
780 |
|
|
781 |
|
/** |
782 |
< |
* gets and removes a local or stolen a task |
783 |
< |
* @return a task, if available |
782 |
> |
* Sets state to TERMINATING. Does NOT unpark or interrupt |
783 |
> |
* to wake up if currently blocked. Callers must do so if desired. |
784 |
|
*/ |
785 |
< |
final ForkJoinTask<?> pollTask() { |
786 |
< |
ForkJoinTask<?> t = locallyFifo? deqTask() : popTask(); |
787 |
< |
if (t == null && (t = scan()) != null) |
788 |
< |
++stealCount; |
789 |
< |
return t; |
785 |
> |
final void shutdown() { |
786 |
> |
for (;;) { |
787 |
> |
int s = runState; |
788 |
> |
if ((s & (TERMINATING|TERMINATED)) != 0) |
789 |
> |
break; |
790 |
> |
if ((s & SUSPENDED) != 0) { // kill and wakeup if suspended |
791 |
> |
if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, |
792 |
> |
(s & ~SUSPENDED) | |
793 |
> |
(TRIMMED|TERMINATING))) |
794 |
> |
break; |
795 |
> |
} |
796 |
> |
else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, |
797 |
> |
s | TERMINATING)) |
798 |
> |
break; |
799 |
> |
} |
800 |
|
} |
801 |
|
|
802 |
|
/** |
803 |
< |
* gets a local task |
602 |
< |
* @return a task, if available |
803 |
> |
* Sets state to TERMINATED. Called only by onTermination(). |
804 |
|
*/ |
805 |
< |
final ForkJoinTask<?> pollLocalTask() { |
806 |
< |
return locallyFifo? deqTask() : popTask(); |
805 |
> |
private void setTerminated() { |
806 |
> |
int s; |
807 |
> |
do {} while (!UNSAFE.compareAndSwapInt(this, runStateOffset, |
808 |
> |
s = runState, |
809 |
> |
s | (TERMINATING|TERMINATED))); |
810 |
> |
} |
811 |
> |
|
812 |
> |
/** |
813 |
> |
* If suspended, tries to set status to unsuspended. |
814 |
> |
* Does NOT wake up if blocked. |
815 |
> |
* |
816 |
> |
* @return true if successful |
817 |
> |
*/ |
818 |
> |
final boolean tryUnsuspend() { |
819 |
> |
int s; |
820 |
> |
while (((s = runState) & SUSPENDED) != 0) { |
821 |
> |
if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, |
822 |
> |
s & ~SUSPENDED)) |
823 |
> |
return true; |
824 |
> |
} |
825 |
> |
return false; |
826 |
|
} |
827 |
|
|
828 |
|
/** |
829 |
< |
* Returns a pool submission, if one exists, activating first. |
830 |
< |
* @return a submission, if available |
829 |
> |
* Sets suspended status and blocks as spare until resumed |
830 |
> |
* or shutdown. |
831 |
|
*/ |
832 |
< |
private ForkJoinTask<?> pollSubmission() { |
832 |
> |
final void suspendAsSpare() { |
833 |
> |
for (;;) { // set suspended unless terminating |
834 |
> |
int s = runState; |
835 |
> |
if ((s & TERMINATING) != 0) { // must kill |
836 |
> |
if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, |
837 |
> |
s | (TRIMMED | TERMINATING))) |
838 |
> |
return; |
839 |
> |
} |
840 |
> |
else if (UNSAFE.compareAndSwapInt(this, runStateOffset, s, |
841 |
> |
s | SUSPENDED)) |
842 |
> |
break; |
843 |
> |
} |
844 |
|
ForkJoinPool p = pool; |
845 |
< |
while (p.hasQueuedSubmissions()) { |
846 |
< |
ForkJoinTask<?> t; |
847 |
< |
if (tryActivate() && (t = p.pollSubmission()) != null) |
848 |
< |
return t; |
845 |
> |
p.pushSpare(this); |
846 |
> |
while ((runState & SUSPENDED) != 0) { |
847 |
> |
if (p.tryAccumulateStealCount(this)) { |
848 |
> |
interrupted(); // clear/ignore interrupts |
849 |
> |
if ((runState & SUSPENDED) == 0) |
850 |
> |
break; |
851 |
> |
LockSupport.park(this); |
852 |
> |
} |
853 |
|
} |
619 |
– |
return null; |
854 |
|
} |
855 |
|
|
856 |
< |
// Methods accessed only by Pool |
856 |
> |
// Misc support methods for ForkJoinPool |
857 |
> |
|
858 |
> |
/** |
859 |
> |
* Returns an estimate of the number of tasks in the queue. Also |
860 |
> |
* used by ForkJoinTask. |
861 |
> |
*/ |
862 |
> |
final int getQueueSize() { |
863 |
> |
int n; // external calls must read base first |
864 |
> |
return (n = -base + sp) <= 0 ? 0 : n; |
865 |
> |
} |
866 |
|
|
867 |
|
/** |
868 |
|
* Removes and cancels all tasks in queue. Can be called from any |
869 |
|
* thread. |
870 |
|
*/ |
871 |
|
final void cancelTasks() { |
872 |
< |
ForkJoinTask<?> t; |
873 |
< |
while (base != sp && (t = deqTask()) != null) |
874 |
< |
t.cancelIgnoringExceptions(); |
872 |
> |
ForkJoinTask<?> cj = currentJoin; // try to cancel ongoing tasks |
873 |
> |
if (cj != null && cj.status >= 0) { |
874 |
> |
cj.cancelIgnoringExceptions(); |
875 |
> |
try { |
876 |
> |
this.interrupt(); // awaken wait |
877 |
> |
} catch (SecurityException ignore) { |
878 |
> |
} |
879 |
> |
} |
880 |
> |
ForkJoinTask<?> cs = currentSteal; |
881 |
> |
if (cs != null && cs.status >= 0) |
882 |
> |
cs.cancelIgnoringExceptions(); |
883 |
> |
while (base != sp) { |
884 |
> |
ForkJoinTask<?> t = deqTask(); |
885 |
> |
if (t != null) |
886 |
> |
t.cancelIgnoringExceptions(); |
887 |
> |
} |
888 |
|
} |
889 |
|
|
890 |
|
/** |
891 |
< |
* Drains tasks to given collection c |
891 |
> |
* Drains tasks to given collection c. |
892 |
> |
* |
893 |
|
* @return the number of tasks drained |
894 |
|
*/ |
895 |
< |
final int drainTasksTo(Collection<ForkJoinTask<?>> c) { |
895 |
> |
final int drainTasksTo(Collection<? super ForkJoinTask<?>> c) { |
896 |
|
int n = 0; |
897 |
< |
ForkJoinTask<?> t; |
898 |
< |
while (base != sp && (t = deqTask()) != null) { |
899 |
< |
c.add(t); |
900 |
< |
++n; |
897 |
> |
while (base != sp) { |
898 |
> |
ForkJoinTask<?> t = deqTask(); |
899 |
> |
if (t != null) { |
900 |
> |
c.add(t); |
901 |
> |
++n; |
902 |
> |
} |
903 |
|
} |
904 |
|
return n; |
905 |
|
} |
906 |
|
|
907 |
+ |
// Support methods for ForkJoinTask |
908 |
+ |
|
909 |
|
/** |
910 |
< |
* Get and clear steal count for accumulation by pool. Called |
911 |
< |
* only when known to be idle (in pool.sync and termination). |
910 |
> |
* Gets and removes a local task. |
911 |
> |
* |
912 |
> |
* @return a task, if available |
913 |
|
*/ |
914 |
< |
final int getAndClearStealCount() { |
915 |
< |
int sc = stealCount; |
916 |
< |
stealCount = 0; |
917 |
< |
return sc; |
914 |
> |
final ForkJoinTask<?> pollLocalTask() { |
915 |
> |
ForkJoinPool p = pool; |
916 |
> |
while (sp != base) { |
917 |
> |
int a; // inline p.tryIncrementActiveCount |
918 |
> |
if (active || |
919 |
> |
(active = UNSAFE.compareAndSwapInt(p, poolRunStateOffset, |
920 |
> |
a = p.runState, a + 1))) |
921 |
> |
return locallyFifo ? locallyDeqTask() : popTask(); |
922 |
> |
} |
923 |
> |
return null; |
924 |
|
} |
925 |
|
|
926 |
|
/** |
927 |
< |
* Returns true if at least one worker in the given array appears |
928 |
< |
* to have at least one queued task. |
929 |
< |
* @param ws array of workers |
930 |
< |
*/ |
931 |
< |
static boolean hasQueuedTasks(ForkJoinWorkerThread[] ws) { |
932 |
< |
if (ws != null) { |
933 |
< |
int len = ws.length; |
934 |
< |
for (int j = 0; j < 2; ++j) { // need two passes for clean sweep |
935 |
< |
for (int i = 0; i < len; ++i) { |
936 |
< |
ForkJoinWorkerThread w = ws[i]; |
669 |
< |
if (w != null && w.sp != w.base) |
670 |
< |
return true; |
671 |
< |
} |
672 |
< |
} |
927 |
> |
* Gets and removes a local or stolen task. |
928 |
> |
* |
929 |
> |
* @return a task, if available |
930 |
> |
*/ |
931 |
> |
final ForkJoinTask<?> pollTask() { |
932 |
> |
ForkJoinTask<?> t = pollLocalTask(); |
933 |
> |
if (t == null) { |
934 |
> |
t = scan(); |
935 |
> |
// cannot retain/track/help steal |
936 |
> |
UNSAFE.putOrderedObject(this, currentStealOffset, null); |
937 |
|
} |
938 |
< |
return false; |
938 |
> |
return t; |
939 |
|
} |
940 |
|
|
941 |
< |
// Support methods for ForkJoinTask |
941 |
> |
/** |
942 |
> |
* Possibly runs some tasks and/or blocks, until task is done. |
943 |
> |
* |
944 |
> |
* @param joinMe the task to join |
945 |
> |
* @param timed true if use timed wait |
946 |
> |
* @param nanos wait time if timed |
947 |
> |
*/ |
948 |
> |
final void joinTask(ForkJoinTask<?> joinMe, boolean timed, long nanos) { |
949 |
> |
// currentJoin only written by this thread; only need ordered store |
950 |
> |
ForkJoinTask<?> prevJoin = currentJoin; |
951 |
> |
UNSAFE.putOrderedObject(this, currentJoinOffset, joinMe); |
952 |
> |
pool.awaitJoin(joinMe, this, timed, nanos); |
953 |
> |
UNSAFE.putOrderedObject(this, currentJoinOffset, prevJoin); |
954 |
> |
} |
955 |
|
|
956 |
|
/** |
957 |
< |
* Returns an estimate of the number of tasks in the queue. |
958 |
< |
*/ |
959 |
< |
final int getQueueSize() { |
960 |
< |
int n = sp - base; |
961 |
< |
return n < 0? 0 : n; // suppress momentarily negative values |
957 |
> |
* Tries to locate and help perform tasks for a stealer of the |
958 |
> |
* given task, or in turn one of its stealers. Traces |
959 |
> |
* currentSteal->currentJoin links looking for a thread working on |
960 |
> |
* a descendant of the given task and with a non-empty queue to |
961 |
> |
* steal back and execute tasks from. |
962 |
> |
* |
963 |
> |
* The implementation is very branchy to cope with potential |
964 |
> |
* inconsistencies or loops encountering chains that are stale, |
965 |
> |
* unknown, or of length greater than MAX_HELP_DEPTH links. All |
966 |
> |
* of these cases are dealt with by just returning back to the |
967 |
> |
* caller, who is expected to retry if other join mechanisms also |
968 |
> |
* don't work out. |
969 |
> |
* |
970 |
> |
* @param joinMe the task to join |
971 |
> |
* @param running if false, then must update pool count upon |
972 |
> |
* running a task |
973 |
> |
* @return value of running on exit |
974 |
> |
*/ |
975 |
> |
final boolean helpJoinTask(ForkJoinTask<?> joinMe, boolean running) { |
976 |
> |
/* |
977 |
> |
* Initial checks to (1) abort if terminating; (2) clean out |
978 |
> |
* old cancelled tasks from local queue; (3) if joinMe is next |
979 |
> |
* task, run it; (4) omit scan if local queue nonempty (since |
980 |
> |
* it may contain non-descendents of joinMe). |
981 |
> |
*/ |
982 |
> |
ForkJoinPool p = pool; |
983 |
> |
for (;;) { |
984 |
> |
ForkJoinTask<?>[] q; |
985 |
> |
int s; |
986 |
> |
if (joinMe.status < 0) |
987 |
> |
return running; |
988 |
> |
else if ((runState & TERMINATING) != 0) |
989 |
> |
joinMe.cancelIgnoringExceptions(); |
990 |
> |
else if ((s = sp) == base || (q = queue) == null) |
991 |
> |
break; // queue empty |
992 |
> |
else { |
993 |
> |
int i = (q.length - 1) & --s; |
994 |
> |
long u = (i << qShift) + qBase; // raw offset |
995 |
> |
ForkJoinTask<?> t = q[i]; |
996 |
> |
if (t == null) |
997 |
> |
break; // lost to a stealer |
998 |
> |
else if (t != joinMe && t.status >= 0) |
999 |
> |
return running; // cannot safely help |
1000 |
> |
else if ((running || |
1001 |
> |
(running = p.tryIncrementRunningCount())) && |
1002 |
> |
UNSAFE.compareAndSwapObject(q, u, t, null)) { |
1003 |
> |
sp = s; // putOrderedInt may encourage more timely write |
1004 |
> |
// UNSAFE.putOrderedInt(this, spOffset, s); |
1005 |
> |
if (t.status >= 0) |
1006 |
> |
t.quietlyExec(); |
1007 |
> |
} |
1008 |
> |
} |
1009 |
> |
} |
1010 |
> |
|
1011 |
> |
int n; // worker array size |
1012 |
> |
ForkJoinWorkerThread[] ws = p.workers; |
1013 |
> |
if (ws != null && (n = ws.length) > 1) { // need at least 2 workers |
1014 |
> |
ForkJoinTask<?> task = joinMe; // base of chain |
1015 |
> |
ForkJoinWorkerThread thread = this; // thread with stolen task |
1016 |
> |
|
1017 |
> |
outer:for (int d = 0; d < MAX_HELP_DEPTH; ++d) { // chain length |
1018 |
> |
// Try to find v, the stealer of task, by first using hint |
1019 |
> |
ForkJoinWorkerThread v = ws[thread.stealHint & (n - 1)]; |
1020 |
> |
if (v == null || v.currentSteal != task) { |
1021 |
> |
for (int j = 0; ; ++j) { // search array |
1022 |
> |
if (j < n) { |
1023 |
> |
ForkJoinTask<?> vs; |
1024 |
> |
if ((v = ws[j]) != null && |
1025 |
> |
(vs = v.currentSteal) != null) { |
1026 |
> |
if (joinMe.status < 0) |
1027 |
> |
break outer; |
1028 |
> |
if (vs == task) { |
1029 |
> |
if (task.status < 0) |
1030 |
> |
break outer; // stale |
1031 |
> |
thread.stealHint = j; |
1032 |
> |
break; // save hint for next time |
1033 |
> |
} |
1034 |
> |
} |
1035 |
> |
} |
1036 |
> |
else |
1037 |
> |
break outer; // no stealer |
1038 |
> |
} |
1039 |
> |
} |
1040 |
> |
|
1041 |
> |
// Try to help v, using specialized form of deqTask |
1042 |
> |
for (;;) { |
1043 |
> |
if (joinMe.status < 0) |
1044 |
> |
break outer; |
1045 |
> |
int b = v.base; |
1046 |
> |
ForkJoinTask<?>[] q = v.queue; |
1047 |
> |
if (b == v.sp || q == null) |
1048 |
> |
break; // empty |
1049 |
> |
int i = (q.length - 1) & b; |
1050 |
> |
long u = (i << qShift) + qBase; |
1051 |
> |
ForkJoinTask<?> t = q[i]; |
1052 |
> |
if (task.status < 0) |
1053 |
> |
break outer; // stale |
1054 |
> |
if (t != null && |
1055 |
> |
(running || |
1056 |
> |
(running = p.tryIncrementRunningCount())) && |
1057 |
> |
v.base == b++ && |
1058 |
> |
UNSAFE.compareAndSwapObject(q, u, t, null)) { |
1059 |
> |
if (t != joinMe && joinMe.status < 0) { |
1060 |
> |
UNSAFE.putObjectVolatile(q, u, t); |
1061 |
> |
break outer; // joinMe cancelled; back out |
1062 |
> |
} |
1063 |
> |
v.base = b; |
1064 |
> |
if (t.status >= 0) { |
1065 |
> |
ForkJoinTask<?> ps = currentSteal; |
1066 |
> |
int pid = poolIndex; |
1067 |
> |
v.stealHint = pid; |
1068 |
> |
UNSAFE.putOrderedObject(this, |
1069 |
> |
currentStealOffset, t); |
1070 |
> |
t.quietlyExec(); |
1071 |
> |
UNSAFE.putOrderedObject(this, |
1072 |
> |
currentStealOffset, ps); |
1073 |
> |
} |
1074 |
> |
} |
1075 |
> |
} |
1076 |
> |
|
1077 |
> |
// Try to descend to find v's stealer |
1078 |
> |
ForkJoinTask<?> next = v.currentJoin; |
1079 |
> |
if (task.status < 0 || next == null || next == task) |
1080 |
> |
break; // stale, dead-end, or cyclic |
1081 |
> |
if ((runState & TERMINATING) != 0) |
1082 |
> |
joinMe.cancelIgnoringExceptions(); |
1083 |
> |
if (joinMe.status < 0) |
1084 |
> |
break; |
1085 |
> |
task = next; |
1086 |
> |
thread = v; |
1087 |
> |
} |
1088 |
> |
} |
1089 |
> |
return running; |
1090 |
|
} |
1091 |
|
|
1092 |
|
/** |
1093 |
+ |
* Implements ForkJoinTask.getSurplusQueuedTaskCount(). |
1094 |
|
* Returns an estimate of the number of tasks, offset by a |
1095 |
|
* function of number of idle workers. |
1096 |
+ |
* |
1097 |
+ |
* This method provides a cheap heuristic guide for task |
1098 |
+ |
* partitioning when programmers, frameworks, tools, or languages |
1099 |
+ |
* have little or no idea about task granularity. In essence by |
1100 |
+ |
* offering this method, we ask users only about tradeoffs in |
1101 |
+ |
* overhead vs expected throughput and its variance, rather than |
1102 |
+ |
* how finely to partition tasks. |
1103 |
+ |
* |
1104 |
+ |
* In a steady state strict (tree-structured) computation, each |
1105 |
+ |
* thread makes available for stealing enough tasks for other |
1106 |
+ |
* threads to remain active. Inductively, if all threads play by |
1107 |
+ |
* the same rules, each thread should make available only a |
1108 |
+ |
* constant number of tasks. |
1109 |
+ |
* |
1110 |
+ |
* The minimum useful constant is just 1. But using a value of 1 |
1111 |
+ |
* would require immediate replenishment upon each steal to |
1112 |
+ |
* maintain enough tasks, which is infeasible. Further, |
1113 |
+ |
* partitionings/granularities of offered tasks should minimize |
1114 |
+ |
* steal rates, which in general means that threads nearer the top |
1115 |
+ |
* of computation tree should generate more than those nearer the |
1116 |
+ |
* bottom. In perfect steady state, each thread is at |
1117 |
+ |
* approximately the same level of computation tree. However, |
1118 |
+ |
* producing extra tasks amortizes the uncertainty of progress and |
1119 |
+ |
* diffusion assumptions. |
1120 |
+ |
* |
1121 |
+ |
* So, users will want to use values larger, but not much larger |
1122 |
+ |
* than 1 to both smooth over transient shortages and hedge |
1123 |
+ |
* against uneven progress; as traded off against the cost of |
1124 |
+ |
* extra task overhead. We leave the user to pick a threshold |
1125 |
+ |
* value to compare with the results of this call to guide |
1126 |
+ |
* decisions, but recommend values such as 3. |
1127 |
+ |
* |
1128 |
+ |
* When all threads are active, it is on average OK to estimate |
1129 |
+ |
* surplus strictly locally. In steady-state, if one thread is |
1130 |
+ |
* maintaining say 2 surplus tasks, then so are others. So we can |
1131 |
+ |
* just use estimated queue length (although note that (sp - base) |
1132 |
+ |
* can be an overestimate because of stealers lagging increments |
1133 |
+ |
* of base). However, this strategy alone leads to serious |
1134 |
+ |
* mis-estimates in some non-steady-state conditions (ramp-up, |
1135 |
+ |
* ramp-down, other stalls). We can detect many of these by |
1136 |
+ |
* further considering the number of "idle" threads, that are |
1137 |
+ |
* known to have zero queued tasks, so compensate by a factor of |
1138 |
+ |
* (#idle/#active) threads. |
1139 |
|
*/ |
1140 |
|
final int getEstimatedSurplusTaskCount() { |
1141 |
< |
// The halving approximates weighting idle vs non-idle workers |
693 |
< |
return (sp - base) - (pool.getIdleThreadCount() >>> 1); |
694 |
< |
} |
695 |
< |
|
696 |
< |
/** |
697 |
< |
* Scan, returning early if joinMe done |
698 |
< |
*/ |
699 |
< |
final ForkJoinTask<?> scanWhileJoining(ForkJoinTask<?> joinMe) { |
700 |
< |
ForkJoinTask<?> t = pollTask(); |
701 |
< |
if (t != null && joinMe.status < 0 && sp == base) { |
702 |
< |
pushTask(t); // unsteal if done and this task would be stealable |
703 |
< |
t = null; |
704 |
< |
} |
705 |
< |
return t; |
1141 |
> |
return sp - base - pool.idlePerActive(); |
1142 |
|
} |
1143 |
|
|
1144 |
|
/** |
1145 |
< |
* Runs tasks until pool isQuiescent |
1145 |
> |
* Runs tasks until {@code pool.isQuiescent()}. |
1146 |
|
*/ |
1147 |
|
final void helpQuiescePool() { |
1148 |
+ |
ForkJoinTask<?> ps = currentSteal; // to restore below |
1149 |
|
for (;;) { |
1150 |
< |
ForkJoinTask<?> t = pollTask(); |
1151 |
< |
if (t != null) |
1150 |
> |
ForkJoinTask<?> t = pollLocalTask(); |
1151 |
> |
if (t != null || (t = scan()) != null) |
1152 |
|
t.quietlyExec(); |
1153 |
< |
else if (tryInactivate() && pool.isQuiescent()) |
1154 |
< |
break; |
1153 |
> |
else { |
1154 |
> |
ForkJoinPool p = pool; |
1155 |
> |
int a; // to inline CASes |
1156 |
> |
if (active) { |
1157 |
> |
if (!UNSAFE.compareAndSwapInt |
1158 |
> |
(p, poolRunStateOffset, a = p.runState, a - 1)) |
1159 |
> |
continue; // retry later |
1160 |
> |
active = false; // inactivate |
1161 |
> |
UNSAFE.putOrderedObject(this, currentStealOffset, ps); |
1162 |
> |
} |
1163 |
> |
if (p.isQuiescent()) { |
1164 |
> |
active = true; // re-activate |
1165 |
> |
do {} while (!UNSAFE.compareAndSwapInt |
1166 |
> |
(p, poolRunStateOffset, a = p.runState, a+1)); |
1167 |
> |
return; |
1168 |
> |
} |
1169 |
> |
} |
1170 |
> |
} |
1171 |
> |
} |
1172 |
> |
|
1173 |
> |
// Unsafe mechanics |
1174 |
> |
|
1175 |
> |
private static final sun.misc.Unsafe UNSAFE = getUnsafe(); |
1176 |
> |
private static final long spOffset = |
1177 |
> |
objectFieldOffset("sp", ForkJoinWorkerThread.class); |
1178 |
> |
private static final long runStateOffset = |
1179 |
> |
objectFieldOffset("runState", ForkJoinWorkerThread.class); |
1180 |
> |
private static final long currentJoinOffset = |
1181 |
> |
objectFieldOffset("currentJoin", ForkJoinWorkerThread.class); |
1182 |
> |
private static final long currentStealOffset = |
1183 |
> |
objectFieldOffset("currentSteal", ForkJoinWorkerThread.class); |
1184 |
> |
private static final long qBase = |
1185 |
> |
UNSAFE.arrayBaseOffset(ForkJoinTask[].class); |
1186 |
> |
private static final long poolRunStateOffset = // to inline CAS |
1187 |
> |
objectFieldOffset("runState", ForkJoinPool.class); |
1188 |
> |
|
1189 |
> |
private static final int qShift; |
1190 |
> |
|
1191 |
> |
static { |
1192 |
> |
int s = UNSAFE.arrayIndexScale(ForkJoinTask[].class); |
1193 |
> |
if ((s & (s-1)) != 0) |
1194 |
> |
throw new Error("data type scale not a power of two"); |
1195 |
> |
qShift = 31 - Integer.numberOfLeadingZeros(s); |
1196 |
> |
MAXIMUM_QUEUE_CAPACITY = 1 << (31 - qShift); |
1197 |
> |
} |
1198 |
> |
|
1199 |
> |
private static long objectFieldOffset(String field, Class<?> klazz) { |
1200 |
> |
try { |
1201 |
> |
return UNSAFE.objectFieldOffset(klazz.getDeclaredField(field)); |
1202 |
> |
} catch (NoSuchFieldException e) { |
1203 |
> |
// Convert Exception to corresponding Error |
1204 |
> |
NoSuchFieldError error = new NoSuchFieldError(field); |
1205 |
> |
error.initCause(e); |
1206 |
> |
throw error; |
1207 |
|
} |
719 |
– |
do;while (!tryActivate()); // re-activate on exit |
1208 |
|
} |
1209 |
|
|
1210 |
< |
// Temporary Unsafe mechanics for preliminary release |
1211 |
< |
private static Unsafe getUnsafe() throws Throwable { |
1210 |
> |
/** |
1211 |
> |
* Returns a sun.misc.Unsafe. Suitable for use in a 3rd party package. |
1212 |
> |
* Replace with a simple call to Unsafe.getUnsafe when integrating |
1213 |
> |
* into a jdk. |
1214 |
> |
* |
1215 |
> |
* @return a sun.misc.Unsafe |
1216 |
> |
*/ |
1217 |
> |
private static sun.misc.Unsafe getUnsafe() { |
1218 |
|
try { |
1219 |
< |
return Unsafe.getUnsafe(); |
1219 |
> |
return sun.misc.Unsafe.getUnsafe(); |
1220 |
|
} catch (SecurityException se) { |
1221 |
|
try { |
1222 |
|
return java.security.AccessController.doPrivileged |
1223 |
< |
(new java.security.PrivilegedExceptionAction<Unsafe>() { |
1224 |
< |
public Unsafe run() throws Exception { |
1225 |
< |
return getUnsafePrivileged(); |
1223 |
> |
(new java.security |
1224 |
> |
.PrivilegedExceptionAction<sun.misc.Unsafe>() { |
1225 |
> |
public sun.misc.Unsafe run() throws Exception { |
1226 |
> |
java.lang.reflect.Field f = sun.misc |
1227 |
> |
.Unsafe.class.getDeclaredField("theUnsafe"); |
1228 |
> |
f.setAccessible(true); |
1229 |
> |
return (sun.misc.Unsafe) f.get(null); |
1230 |
|
}}); |
1231 |
|
} catch (java.security.PrivilegedActionException e) { |
1232 |
< |
throw e.getCause(); |
1232 |
> |
throw new RuntimeException("Could not initialize intrinsics", |
1233 |
> |
e.getCause()); |
1234 |
|
} |
1235 |
|
} |
1236 |
|
} |
738 |
– |
|
739 |
– |
private static Unsafe getUnsafePrivileged() |
740 |
– |
throws NoSuchFieldException, IllegalAccessException { |
741 |
– |
Field f = Unsafe.class.getDeclaredField("theUnsafe"); |
742 |
– |
f.setAccessible(true); |
743 |
– |
return (Unsafe) f.get(null); |
744 |
– |
} |
745 |
– |
|
746 |
– |
private static long fieldOffset(String fieldName) |
747 |
– |
throws NoSuchFieldException { |
748 |
– |
return _unsafe.objectFieldOffset |
749 |
– |
(ForkJoinWorkerThread.class.getDeclaredField(fieldName)); |
750 |
– |
} |
751 |
– |
|
752 |
– |
static final Unsafe _unsafe; |
753 |
– |
static final long baseOffset; |
754 |
– |
static final long spOffset; |
755 |
– |
static final long runStateOffset; |
756 |
– |
static final long qBase; |
757 |
– |
static final int qShift; |
758 |
– |
static { |
759 |
– |
try { |
760 |
– |
_unsafe = getUnsafe(); |
761 |
– |
baseOffset = fieldOffset("base"); |
762 |
– |
spOffset = fieldOffset("sp"); |
763 |
– |
runStateOffset = fieldOffset("runState"); |
764 |
– |
qBase = _unsafe.arrayBaseOffset(ForkJoinTask[].class); |
765 |
– |
int s = _unsafe.arrayIndexScale(ForkJoinTask[].class); |
766 |
– |
if ((s & (s-1)) != 0) |
767 |
– |
throw new Error("data type scale not a power of two"); |
768 |
– |
qShift = 31 - Integer.numberOfLeadingZeros(s); |
769 |
– |
} catch (Throwable e) { |
770 |
– |
throw new RuntimeException("Could not initialize intrinsics", e); |
771 |
– |
} |
772 |
– |
} |
1237 |
|
} |