Mercurial > hg > graal-jvmci-8
annotate src/share/vm/utilities/taskqueue.hpp @ 1836:894b1d7c7e01
6423256: GC stacks should use a better data structure
6942771: SEGV in ParScanThreadState::take_from_overflow_stack
Reviewed-by: apetrusenko, ysr, pbk
author | jcoomes |
---|---|
date | Tue, 28 Sep 2010 15:56:15 -0700 |
parents | 8b10f48633dc |
children | f95d63e2154a |
rev | line source |
---|---|
0 | 1 /* |
1638 | 2 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. |
0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1311
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1311
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1311
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1665 | 25 // Simple TaskQueue stats that are collected by default in debug builds. |
26 | |
27 #if !defined(TASKQUEUE_STATS) && defined(ASSERT) | |
28 #define TASKQUEUE_STATS 1 | |
29 #elif !defined(TASKQUEUE_STATS) | |
30 #define TASKQUEUE_STATS 0 | |
31 #endif | |
32 | |
33 #if TASKQUEUE_STATS | |
34 #define TASKQUEUE_STATS_ONLY(code) code | |
35 #else | |
36 #define TASKQUEUE_STATS_ONLY(code) | |
37 #endif // TASKQUEUE_STATS | |
38 | |
39 #if TASKQUEUE_STATS | |
40 class TaskQueueStats { | |
41 public: | |
42 enum StatId { | |
43 push, // number of taskqueue pushes | |
44 pop, // number of taskqueue pops | |
45 pop_slow, // subset of taskqueue pops that were done slow-path | |
46 steal_attempt, // number of taskqueue steal attempts | |
47 steal, // number of taskqueue steals | |
48 overflow, // number of overflow pushes | |
49 overflow_max_len, // max length of overflow stack | |
50 last_stat_id | |
51 }; | |
52 | |
53 public: | |
54 inline TaskQueueStats() { reset(); } | |
55 | |
56 inline void record_push() { ++_stats[push]; } | |
57 inline void record_pop() { ++_stats[pop]; } | |
58 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; } | |
59 inline void record_steal(bool success); | |
60 inline void record_overflow(size_t new_length); | |
61 | |
1709 | 62 TaskQueueStats & operator +=(const TaskQueueStats & addend); |
63 | |
1665 | 64 inline size_t get(StatId id) const { return _stats[id]; } |
65 inline const size_t* get() const { return _stats; } | |
66 | |
67 inline void reset(); | |
68 | |
1709 | 69 // Print the specified line of the header (does not include a line separator). |
1665 | 70 static void print_header(unsigned int line, outputStream* const stream = tty, |
71 unsigned int width = 10); | |
1709 | 72 // Print the statistics (does not include a line separator). |
1665 | 73 void print(outputStream* const stream = tty, unsigned int width = 10) const; |
74 | |
1709 | 75 DEBUG_ONLY(void verify() const;) |
76 | |
1665 | 77 private: |
78 size_t _stats[last_stat_id]; | |
79 static const char * const _names[last_stat_id]; | |
80 }; | |
81 | |
82 void TaskQueueStats::record_steal(bool success) { | |
83 ++_stats[steal_attempt]; | |
84 if (success) ++_stats[steal]; | |
85 } | |
86 | |
87 void TaskQueueStats::record_overflow(size_t new_len) { | |
88 ++_stats[overflow]; | |
89 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len; | |
90 } | |
91 | |
92 void TaskQueueStats::reset() { | |
93 memset(_stats, 0, sizeof(_stats)); | |
94 } | |
95 #endif // TASKQUEUE_STATS | |
96 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
97 template <unsigned int N> |
0 | 98 class TaskQueueSuper: public CHeapObj { |
99 protected: | |
907 | 100 // Internal type for indexing the queue; also used for the tag. |
101 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t; | |
102 | |
103 // The first free element after the last one pushed (mod N). | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
104 volatile uint _bottom; |
0 | 105 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
106 enum { MOD_N_MASK = N - 1 }; |
907 | 107 |
108 class Age { | |
109 public: | |
110 Age(size_t data = 0) { _data = data; } | |
111 Age(const Age& age) { _data = age._data; } | |
112 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; } | |
0 | 113 |
907 | 114 Age get() const volatile { return _data; } |
115 void set(Age age) volatile { _data = age._data; } | |
116 | |
117 idx_t top() const volatile { return _fields._top; } | |
118 idx_t tag() const volatile { return _fields._tag; } | |
0 | 119 |
907 | 120 // Increment top; if it wraps, increment tag also. |
121 void increment() { | |
122 _fields._top = increment_index(_fields._top); | |
123 if (_fields._top == 0) ++_fields._tag; | |
124 } | |
0 | 125 |
907 | 126 Age cmpxchg(const Age new_age, const Age old_age) volatile { |
127 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data, | |
128 (volatile intptr_t *)&_data, | |
129 (intptr_t)old_age._data); | |
130 } | |
131 | |
132 bool operator ==(const Age& other) const { return _data == other._data; } | |
0 | 133 |
907 | 134 private: |
135 struct fields { | |
136 idx_t _top; | |
137 idx_t _tag; | |
138 }; | |
139 union { | |
140 size_t _data; | |
141 fields _fields; | |
142 }; | |
0 | 143 }; |
907 | 144 |
145 volatile Age _age; | |
146 | |
147 // These both operate mod N. | |
148 static uint increment_index(uint ind) { | |
149 return (ind + 1) & MOD_N_MASK; | |
0 | 150 } |
907 | 151 static uint decrement_index(uint ind) { |
152 return (ind - 1) & MOD_N_MASK; | |
0 | 153 } |
154 | |
907 | 155 // Returns a number in the range [0..N). If the result is "N-1", it should be |
156 // interpreted as 0. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
157 uint dirty_size(uint bot, uint top) const { |
907 | 158 return (bot - top) & MOD_N_MASK; |
0 | 159 } |
160 | |
161 // Returns the size corresponding to the given "bot" and "top". | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
162 uint size(uint bot, uint top) const { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
163 uint sz = dirty_size(bot, top); |
907 | 164 // Has the queue "wrapped", so that bottom is less than top? There's a |
165 // complicated special case here. A pair of threads could perform pop_local | |
166 // and pop_global operations concurrently, starting from a state in which | |
167 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom, | |
168 // and the pop_global in incrementing _top (in which case the pop_global | |
169 // will be awarded the contested queue element.) The resulting state must | |
170 // be interpreted as an empty queue. (We only need to worry about one such | |
171 // event: only the queue owner performs pop_local's, and several concurrent | |
172 // threads attempting to perform the pop_global will all perform the same | |
173 // CAS, and only one can succeed.) Any stealing thread that reads after | |
174 // either the increment or decrement will see an empty queue, and will not | |
175 // join the competitors. The "sz == -1 || sz == N-1" state will not be | |
176 // modified by concurrent queues, so the owner thread can reset the state to | |
177 // _bottom == top so subsequent pushes will be performed normally. | |
178 return (sz == N - 1) ? 0 : sz; | |
0 | 179 } |
180 | |
181 public: | |
182 TaskQueueSuper() : _bottom(0), _age() {} | |
183 | |
1638 | 184 // Return true if the TaskQueue contains/does not contain any tasks. |
185 bool peek() const { return _bottom != _age.top(); } | |
186 bool is_empty() const { return size() == 0; } | |
0 | 187 |
188 // Return an estimate of the number of elements in the queue. | |
189 // The "careful" version admits the possibility of pop_local/pop_global | |
190 // races. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
191 uint size() const { |
907 | 192 return size(_bottom, _age.top()); |
0 | 193 } |
194 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
195 uint dirty_size() const { |
907 | 196 return dirty_size(_bottom, _age.top()); |
0 | 197 } |
198 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
199 void set_empty() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
200 _bottom = 0; |
907 | 201 _age.set(0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
202 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
203 |
0 | 204 // Maximum number of elements allowed in the queue. This is two less |
205 // than the actual queue size, for somewhat complicated reasons. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
206 uint max_elems() const { return N - 2; } |
1284 | 207 |
208 // Total size of queue. | |
209 static const uint total_size() { return N; } | |
1665 | 210 |
211 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;) | |
0 | 212 }; |
213 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
214 template<class E, unsigned int N = TASKQUEUE_SIZE> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
215 class GenericTaskQueue: public TaskQueueSuper<N> { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
216 protected: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
217 typedef typename TaskQueueSuper<N>::Age Age; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
218 typedef typename TaskQueueSuper<N>::idx_t idx_t; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
219 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
220 using TaskQueueSuper<N>::_bottom; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
221 using TaskQueueSuper<N>::_age; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
222 using TaskQueueSuper<N>::increment_index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
223 using TaskQueueSuper<N>::decrement_index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
224 using TaskQueueSuper<N>::dirty_size; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
225 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
226 public: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
227 using TaskQueueSuper<N>::max_elems; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
228 using TaskQueueSuper<N>::size; |
1665 | 229 TASKQUEUE_STATS_ONLY(using TaskQueueSuper<N>::stats;) |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
230 |
0 | 231 private: |
232 // Slow paths for push, pop_local. (pop_global has no fast path.) | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
233 bool push_slow(E t, uint dirty_n_elems); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
234 bool pop_local_slow(uint localBot, Age oldAge); |
0 | 235 |
236 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
237 typedef E element_type; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
238 |
0 | 239 // Initializes the queue to empty. |
240 GenericTaskQueue(); | |
241 | |
242 void initialize(); | |
243 | |
1638 | 244 // Push the task "t" on the queue. Returns "false" iff the queue is full. |
0 | 245 inline bool push(E t); |
246 | |
1638 | 247 // Attempts to claim a task from the "local" end of the queue (the most |
248 // recently pushed). If successful, returns true and sets t to the task; | |
249 // otherwise, returns false (the queue is empty). | |
0 | 250 inline bool pop_local(E& t); |
251 | |
1638 | 252 // Like pop_local(), but uses the "global" end of the queue (the least |
253 // recently pushed). | |
0 | 254 bool pop_global(E& t); |
255 | |
256 // Delete any resource associated with the queue. | |
257 ~GenericTaskQueue(); | |
258 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
259 // apply the closure to all elements in the task queue |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
260 void oops_do(OopClosure* f); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
261 |
0 | 262 private: |
263 // Element array. | |
264 volatile E* _elems; | |
265 }; | |
266 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
267 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
268 GenericTaskQueue<E, N>::GenericTaskQueue() { |
907 | 269 assert(sizeof(Age) == sizeof(size_t), "Depends on this."); |
0 | 270 } |
271 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
272 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
273 void GenericTaskQueue<E, N>::initialize() { |
907 | 274 _elems = NEW_C_HEAP_ARRAY(E, N); |
0 | 275 } |
276 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
277 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
278 void GenericTaskQueue<E, N>::oops_do(OopClosure* f) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
279 // tty->print_cr("START OopTaskQueue::oops_do"); |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
280 uint iters = size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
281 uint index = _bottom; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
282 for (uint i = 0; i < iters; ++i) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
283 index = decrement_index(index); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
284 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
285 // index, &_elems[index], _elems[index]); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
286 E* t = (E*)&_elems[index]; // cast away volatility |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
287 oop* p = (oop*)t; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
288 assert((*t)->is_oop_or_null(), "Not an oop or null"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
289 f->do_oop(p); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
290 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
291 // tty->print_cr("END OopTaskQueue::oops_do"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
292 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
293 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
294 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
295 bool GenericTaskQueue<E, N>::push_slow(E t, uint dirty_n_elems) { |
907 | 296 if (dirty_n_elems == N - 1) { |
0 | 297 // Actually means 0, so do the push. |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
298 uint localBot = _bottom; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
299 // g++ complains if the volatile result of the assignment is unused. |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
300 const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
301 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 302 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 303 return true; |
907 | 304 } |
305 return false; | |
0 | 306 } |
307 | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
308 // pop_local_slow() is done by the owning thread and is trying to |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
309 // get the last task in the queue. It will compete with pop_global() |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
310 // that will be used by other threads. The tag age is incremented |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
311 // whenever the queue goes empty which it will do here if this thread |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
312 // gets the last task or in pop_global() if the queue wraps (top == 0 |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
313 // and pop_global() succeeds, see pop_global()). |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
314 template<class E, unsigned int N> |
1665 | 315 bool GenericTaskQueue<E, N>::pop_local_slow(uint localBot, Age oldAge) { |
0 | 316 // This queue was observed to contain exactly one element; either this |
317 // thread will claim it, or a competing "pop_global". In either case, | |
318 // the queue will be logically empty afterwards. Create a new Age value | |
319 // that represents the empty queue for the given value of "_bottom". (We | |
320 // must also increment "tag" because of the case where "bottom == 1", | |
321 // "top == 0". A pop_global could read the queue element in that case, | |
322 // then have the owner thread do a pop followed by another push. Without | |
323 // the incrementing of "tag", the pop_global's CAS could succeed, | |
324 // allowing it to believe it has claimed the stale element.) | |
907 | 325 Age newAge((idx_t)localBot, oldAge.tag() + 1); |
0 | 326 // Perhaps a competing pop_global has already incremented "top", in which |
327 // case it wins the element. | |
328 if (localBot == oldAge.top()) { | |
329 // No competing pop_global has yet incremented "top"; we'll try to | |
330 // install new_age, thus claiming the element. | |
907 | 331 Age tempAge = _age.cmpxchg(newAge, oldAge); |
0 | 332 if (tempAge == oldAge) { |
333 // We win. | |
907 | 334 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); |
1665 | 335 TASKQUEUE_STATS_ONLY(stats.record_pop_slow()); |
0 | 336 return true; |
337 } | |
338 } | |
907 | 339 // We lose; a completing pop_global gets the element. But the queue is empty |
340 // and top is greater than bottom. Fix this representation of the empty queue | |
341 // to become the canonical one. | |
342 _age.set(newAge); | |
343 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); | |
0 | 344 return false; |
345 } | |
346 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
347 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
348 bool GenericTaskQueue<E, N>::pop_global(E& t) { |
907 | 349 Age oldAge = _age.get(); |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
350 uint localBot = _bottom; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
351 uint n_elems = size(localBot, oldAge.top()); |
0 | 352 if (n_elems == 0) { |
353 return false; | |
354 } | |
907 | 355 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
356 const_cast<E&>(t = _elems[oldAge.top()]); |
907 | 357 Age newAge(oldAge); |
358 newAge.increment(); | |
359 Age resAge = _age.cmpxchg(newAge, oldAge); | |
360 | |
0 | 361 // Note that using "_bottom" here might fail, since a pop_local might |
362 // have decremented it. | |
907 | 363 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity"); |
364 return resAge == oldAge; | |
0 | 365 } |
366 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
367 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
368 GenericTaskQueue<E, N>::~GenericTaskQueue() { |
0 | 369 FREE_C_HEAP_ARRAY(E, _elems); |
370 } | |
371 | |
1638 | 372 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for |
373 // elements that do not fit in the TaskQueue. | |
374 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
375 // This class hides two methods from super classes: |
1638 | 376 // |
377 // push() - push onto the task queue or, if that fails, onto the overflow stack | |
378 // is_empty() - return true if both the TaskQueue and overflow stack are empty | |
379 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
380 // Note that size() is not hidden--it returns the number of elements in the |
1638 | 381 // TaskQueue, and does not include the size of the overflow stack. This |
382 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues. | |
383 template<class E, unsigned int N = TASKQUEUE_SIZE> | |
384 class OverflowTaskQueue: public GenericTaskQueue<E, N> | |
385 { | |
386 public: | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
387 typedef Stack<E> overflow_t; |
1638 | 388 typedef GenericTaskQueue<E, N> taskqueue_t; |
389 | |
1665 | 390 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;) |
391 | |
1638 | 392 // Push task t onto the queue or onto the overflow stack. Return true. |
393 inline bool push(E t); | |
394 | |
395 // Attempt to pop from the overflow stack; return true if anything was popped. | |
396 inline bool pop_overflow(E& t); | |
397 | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
398 inline overflow_t* overflow_stack() { return &_overflow_stack; } |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
399 |
1638 | 400 inline bool taskqueue_empty() const { return taskqueue_t::is_empty(); } |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
401 inline bool overflow_empty() const { return _overflow_stack.is_empty(); } |
1638 | 402 inline bool is_empty() const { |
403 return taskqueue_empty() && overflow_empty(); | |
404 } | |
405 | |
406 private: | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
407 overflow_t _overflow_stack; |
1638 | 408 }; |
409 | |
410 template <class E, unsigned int N> | |
411 bool OverflowTaskQueue<E, N>::push(E t) | |
412 { | |
413 if (!taskqueue_t::push(t)) { | |
414 overflow_stack()->push(t); | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
415 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size())); |
1638 | 416 } |
417 return true; | |
418 } | |
419 | |
420 template <class E, unsigned int N> | |
421 bool OverflowTaskQueue<E, N>::pop_overflow(E& t) | |
422 { | |
423 if (overflow_empty()) return false; | |
424 t = overflow_stack()->pop(); | |
425 return true; | |
426 } | |
427 | |
0 | 428 class TaskQueueSetSuper: public CHeapObj { |
429 protected: | |
430 static int randomParkAndMiller(int* seed0); | |
431 public: | |
432 // Returns "true" if some TaskQueue in the set contains a task. | |
433 virtual bool peek() = 0; | |
434 }; | |
435 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
436 template<class T> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
437 class GenericTaskQueueSet: public TaskQueueSetSuper { |
0 | 438 private: |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
439 uint _n; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
440 T** _queues; |
0 | 441 |
442 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
443 typedef typename T::element_type E; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
444 |
0 | 445 GenericTaskQueueSet(int n) : _n(n) { |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
446 typedef T* GenericTaskQueuePtr; |
0 | 447 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n); |
448 for (int i = 0; i < n; i++) { | |
449 _queues[i] = NULL; | |
450 } | |
451 } | |
452 | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
453 bool steal_1_random(uint queue_num, int* seed, E& t); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
454 bool steal_best_of_2(uint queue_num, int* seed, E& t); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
455 bool steal_best_of_all(uint queue_num, int* seed, E& t); |
0 | 456 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
457 void register_queue(uint i, T* q); |
0 | 458 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
459 T* queue(uint n); |
0 | 460 |
1638 | 461 // The thread with queue number "queue_num" (and whose random number seed is |
462 // at "seed") is trying to steal a task from some other queue. (It may try | |
463 // several queues, according to some configuration parameter.) If some steal | |
464 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns | |
465 // false. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
466 bool steal(uint queue_num, int* seed, E& t); |
0 | 467 |
468 bool peek(); | |
469 }; | |
470 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
471 template<class T> void |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
472 GenericTaskQueueSet<T>::register_queue(uint i, T* q) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
473 assert(i < _n, "index out of range."); |
0 | 474 _queues[i] = q; |
475 } | |
476 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
477 template<class T> T* |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
478 GenericTaskQueueSet<T>::queue(uint i) { |
0 | 479 return _queues[i]; |
480 } | |
481 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
482 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
483 GenericTaskQueueSet<T>::steal(uint queue_num, int* seed, E& t) { |
1665 | 484 for (uint i = 0; i < 2 * _n; i++) { |
485 if (steal_best_of_2(queue_num, seed, t)) { | |
486 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true)); | |
0 | 487 return true; |
1665 | 488 } |
489 } | |
490 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false)); | |
0 | 491 return false; |
492 } | |
493 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
494 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
495 GenericTaskQueueSet<T>::steal_best_of_all(uint queue_num, int* seed, E& t) { |
0 | 496 if (_n > 2) { |
497 int best_k; | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
498 uint best_sz = 0; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
499 for (uint k = 0; k < _n; k++) { |
0 | 500 if (k == queue_num) continue; |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
501 uint sz = _queues[k]->size(); |
0 | 502 if (sz > best_sz) { |
503 best_sz = sz; | |
504 best_k = k; | |
505 } | |
506 } | |
507 return best_sz > 0 && _queues[best_k]->pop_global(t); | |
508 } else if (_n == 2) { | |
509 // Just try the other one. | |
510 int k = (queue_num + 1) % 2; | |
511 return _queues[k]->pop_global(t); | |
512 } else { | |
513 assert(_n == 1, "can't be zero."); | |
514 return false; | |
515 } | |
516 } | |
517 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
518 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
519 GenericTaskQueueSet<T>::steal_1_random(uint queue_num, int* seed, E& t) { |
0 | 520 if (_n > 2) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
521 uint k = queue_num; |
0 | 522 while (k == queue_num) k = randomParkAndMiller(seed) % _n; |
523 return _queues[2]->pop_global(t); | |
524 } else if (_n == 2) { | |
525 // Just try the other one. | |
526 int k = (queue_num + 1) % 2; | |
527 return _queues[k]->pop_global(t); | |
528 } else { | |
529 assert(_n == 1, "can't be zero."); | |
530 return false; | |
531 } | |
532 } | |
533 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
534 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
535 GenericTaskQueueSet<T>::steal_best_of_2(uint queue_num, int* seed, E& t) { |
0 | 536 if (_n > 2) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
537 uint k1 = queue_num; |
0 | 538 while (k1 == queue_num) k1 = randomParkAndMiller(seed) % _n; |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
539 uint k2 = queue_num; |
0 | 540 while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n; |
541 // Sample both and try the larger. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
542 uint sz1 = _queues[k1]->size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
543 uint sz2 = _queues[k2]->size(); |
0 | 544 if (sz2 > sz1) return _queues[k2]->pop_global(t); |
545 else return _queues[k1]->pop_global(t); | |
546 } else if (_n == 2) { | |
547 // Just try the other one. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
548 uint k = (queue_num + 1) % 2; |
0 | 549 return _queues[k]->pop_global(t); |
550 } else { | |
551 assert(_n == 1, "can't be zero."); | |
552 return false; | |
553 } | |
554 } | |
555 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
556 template<class T> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
557 bool GenericTaskQueueSet<T>::peek() { |
0 | 558 // Try all the queues. |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
559 for (uint j = 0; j < _n; j++) { |
0 | 560 if (_queues[j]->peek()) |
561 return true; | |
562 } | |
563 return false; | |
564 } | |
565 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
566 // When to terminate from the termination protocol. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
567 class TerminatorTerminator: public CHeapObj { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
568 public: |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
569 virtual bool should_exit_termination() = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
570 }; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
571 |
0 | 572 // A class to aid in the termination of a set of parallel tasks using |
573 // TaskQueueSet's for work stealing. | |
574 | |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
575 #undef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
576 |
0 | 577 class ParallelTaskTerminator: public StackObj { |
578 private: | |
579 int _n_threads; | |
580 TaskQueueSetSuper* _queue_set; | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
581 int _offered_termination; |
0 | 582 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
583 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
584 static uint _total_yields; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
585 static uint _total_spins; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
586 static uint _total_peeks; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
587 #endif |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
588 |
0 | 589 bool peek_in_queue_set(); |
590 protected: | |
591 virtual void yield(); | |
592 void sleep(uint millis); | |
593 | |
594 public: | |
595 | |
596 // "n_threads" is the number of threads to be terminated. "queue_set" is a | |
597 // queue sets of work queues of other threads. | |
598 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set); | |
599 | |
600 // The current thread has no work, and is ready to terminate if everyone | |
601 // else is. If returns "true", all threads are terminated. If returns | |
602 // "false", available work has been observed in one of the task queues, | |
603 // so the global task is not complete. | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
604 bool offer_termination() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
605 return offer_termination(NULL); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
606 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
607 |
907 | 608 // As above, but it also terminates if the should_exit_termination() |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
609 // method of the terminator parameter returns true. If terminator is |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
610 // NULL, then it is ignored. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
611 bool offer_termination(TerminatorTerminator* terminator); |
0 | 612 |
613 // Reset the terminator, so that it may be reused again. | |
614 // The caller is responsible for ensuring that this is done | |
615 // in an MT-safe manner, once the previous round of use of | |
616 // the terminator is finished. | |
617 void reset_for_reuse(); | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
618 // Same as above but the number of parallel threads is set to the |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
619 // given number. |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
620 void reset_for_reuse(int n_threads); |
0 | 621 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
622 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
623 static uint total_yields() { return _total_yields; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
624 static uint total_spins() { return _total_spins; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
625 static uint total_peeks() { return _total_peeks; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
626 static void print_termination_counts(); |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
627 #endif |
0 | 628 }; |
629 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
630 template<class E, unsigned int N> inline bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
631 GenericTaskQueue<E, N>::push(E t) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
632 uint localBot = _bottom; |
907 | 633 assert((localBot >= 0) && (localBot < N), "_bottom out of range."); |
634 idx_t top = _age.top(); | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
635 uint dirty_n_elems = dirty_size(localBot, top); |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
636 assert(dirty_n_elems < N, "n_elems out of range."); |
0 | 637 if (dirty_n_elems < max_elems()) { |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
638 // g++ complains if the volatile result of the assignment is unused. |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
639 const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
640 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 641 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 642 return true; |
643 } else { | |
644 return push_slow(t, dirty_n_elems); | |
645 } | |
646 } | |
647 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
648 template<class E, unsigned int N> inline bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
649 GenericTaskQueue<E, N>::pop_local(E& t) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
650 uint localBot = _bottom; |
907 | 651 // This value cannot be N-1. That can only occur as a result of |
0 | 652 // the assignment to bottom in this method. If it does, this method |
1638 | 653 // resets the size to 0 before the next call (which is sequential, |
0 | 654 // since this is pop_local.) |
907 | 655 uint dirty_n_elems = dirty_size(localBot, _age.top()); |
656 assert(dirty_n_elems != N - 1, "Shouldn't be possible..."); | |
0 | 657 if (dirty_n_elems == 0) return false; |
658 localBot = decrement_index(localBot); | |
659 _bottom = localBot; | |
660 // This is necessary to prevent any read below from being reordered | |
661 // before the store just above. | |
662 OrderAccess::fence(); | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
663 const_cast<E&>(t = _elems[localBot]); |
0 | 664 // This is a second read of "age"; the "size()" above is the first. |
665 // If there's still at least one element in the queue, based on the | |
666 // "_bottom" and "age" we've read, then there can be no interference with | |
667 // a "pop_global" operation, and we're done. | |
907 | 668 idx_t tp = _age.top(); // XXX |
0 | 669 if (size(localBot, tp) > 0) { |
907 | 670 assert(dirty_size(localBot, tp) != N - 1, "sanity"); |
1665 | 671 TASKQUEUE_STATS_ONLY(stats.record_pop()); |
0 | 672 return true; |
673 } else { | |
674 // Otherwise, the queue contained exactly one element; we take the slow | |
675 // path. | |
907 | 676 return pop_local_slow(localBot, _age.get()); |
0 | 677 } |
678 } | |
679 | |
1638 | 680 typedef GenericTaskQueue<oop> OopTaskQueue; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
681 typedef GenericTaskQueueSet<OopTaskQueue> OopTaskQueueSet; |
0 | 682 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
683 #ifdef _MSC_VER |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
684 #pragma warning(push) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
685 // warning C4522: multiple assignment operators specified |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
686 #pragma warning(disable:4522) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
687 #endif |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
688 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
689 // This is a container class for either an oop* or a narrowOop*. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
690 // Both are pushed onto a task queue and the consumer will test is_narrow() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
691 // to determine which should be processed. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
692 class StarTask { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
693 void* _holder; // either union oop* or narrowOop* |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
694 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
695 enum { COMPRESSED_OOP_MASK = 1 }; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
696 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
697 public: |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
698 StarTask(narrowOop* p) { |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
699 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!"); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
700 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
701 } |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
702 StarTask(oop* p) { |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
703 assert(((uintptr_t)p & COMPRESSED_OOP_MASK) == 0, "Information loss!"); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
704 _holder = (void*)p; |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
705 } |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
706 StarTask() { _holder = NULL; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
707 operator oop*() { return (oop*)_holder; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
708 operator narrowOop*() { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
709 return (narrowOop*)((uintptr_t)_holder & ~COMPRESSED_OOP_MASK); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
710 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
711 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
712 StarTask& operator=(const StarTask& t) { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
713 _holder = t._holder; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
714 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
715 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
716 volatile StarTask& operator=(const volatile StarTask& t) volatile { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
717 _holder = t._holder; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
718 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
719 } |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
720 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
721 bool is_narrow() const { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
722 return (((uintptr_t)_holder & COMPRESSED_OOP_MASK) != 0); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
723 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
724 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
725 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
726 class ObjArrayTask |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
727 { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
728 public: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
729 ObjArrayTask(oop o = NULL, int idx = 0): _obj(o), _index(idx) { } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
730 ObjArrayTask(oop o, size_t idx): _obj(o), _index(int(idx)) { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
731 assert(idx <= size_t(max_jint), "too big"); |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
732 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
733 ObjArrayTask(const ObjArrayTask& t): _obj(t._obj), _index(t._index) { } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
734 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
735 ObjArrayTask& operator =(const ObjArrayTask& t) { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
736 _obj = t._obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
737 _index = t._index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
738 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
739 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
740 volatile ObjArrayTask& |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
741 operator =(const volatile ObjArrayTask& t) volatile { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
742 _obj = t._obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
743 _index = t._index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
744 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
745 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
746 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
747 inline oop obj() const { return _obj; } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
748 inline int index() const { return _index; } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
749 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
750 DEBUG_ONLY(bool is_valid() const); // Tasks to be pushed/popped must be valid. |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
751 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
752 private: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
753 oop _obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
754 int _index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
755 }; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
756 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
757 #ifdef _MSC_VER |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
758 #pragma warning(pop) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
759 #endif |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
760 |
1638 | 761 typedef OverflowTaskQueue<StarTask> OopStarTaskQueue; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
762 typedef GenericTaskQueueSet<OopStarTaskQueue> OopStarTaskQueueSet; |
0 | 763 |
1638 | 764 typedef OverflowTaskQueue<size_t> RegionTaskQueue; |
765 typedef GenericTaskQueueSet<RegionTaskQueue> RegionTaskQueueSet; | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
766 |