Mercurial > hg > truffle
annotate src/share/vm/utilities/taskqueue.hpp @ 3187:6ff76a1b8339
The benchmark tool should now print zero values to the csv file, if a benchmark fails
author | Josef Haider <josef.haider@khg.jku.at> |
---|---|
date | Thu, 07 Jul 2011 19:43:17 +0200 |
parents | 1d1603768966 |
children | f08d439fab8c |
rev | line source |
---|---|
0 | 1 /* |
2426
1d1603768966
7010070: Update all 2010 Oracle-changed OpenJDK files to have the proper copyright dates - second pass
trims
parents:
2192
diff
changeset
|
2 * Copyright (c) 2001, 2011, 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 | |
1972 | 25 #ifndef SHARE_VM_UTILITIES_TASKQUEUE_HPP |
26 #define SHARE_VM_UTILITIES_TASKQUEUE_HPP | |
27 | |
28 #include "memory/allocation.hpp" | |
29 #include "memory/allocation.inline.hpp" | |
30 #include "runtime/mutex.hpp" | |
31 #include "utilities/stack.hpp" | |
32 #ifdef TARGET_OS_ARCH_linux_x86 | |
33 # include "orderAccess_linux_x86.inline.hpp" | |
34 #endif | |
35 #ifdef TARGET_OS_ARCH_linux_sparc | |
36 # include "orderAccess_linux_sparc.inline.hpp" | |
37 #endif | |
38 #ifdef TARGET_OS_ARCH_linux_zero | |
39 # include "orderAccess_linux_zero.inline.hpp" | |
40 #endif | |
41 #ifdef TARGET_OS_ARCH_solaris_x86 | |
42 # include "orderAccess_solaris_x86.inline.hpp" | |
43 #endif | |
44 #ifdef TARGET_OS_ARCH_solaris_sparc | |
45 # include "orderAccess_solaris_sparc.inline.hpp" | |
46 #endif | |
47 #ifdef TARGET_OS_ARCH_windows_x86 | |
48 # include "orderAccess_windows_x86.inline.hpp" | |
49 #endif | |
2192
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
50 #ifdef TARGET_OS_ARCH_linux_arm |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
51 # include "orderAccess_linux_arm.inline.hpp" |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
52 #endif |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
53 #ifdef TARGET_OS_ARCH_linux_ppc |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
54 # include "orderAccess_linux_ppc.inline.hpp" |
b92c45f2bc75
7016023: Enable building ARM and PPC from src/closed repository
bobv
parents:
1972
diff
changeset
|
55 #endif |
1972 | 56 |
1665 | 57 // Simple TaskQueue stats that are collected by default in debug builds. |
58 | |
59 #if !defined(TASKQUEUE_STATS) && defined(ASSERT) | |
60 #define TASKQUEUE_STATS 1 | |
61 #elif !defined(TASKQUEUE_STATS) | |
62 #define TASKQUEUE_STATS 0 | |
63 #endif | |
64 | |
65 #if TASKQUEUE_STATS | |
66 #define TASKQUEUE_STATS_ONLY(code) code | |
67 #else | |
68 #define TASKQUEUE_STATS_ONLY(code) | |
69 #endif // TASKQUEUE_STATS | |
70 | |
71 #if TASKQUEUE_STATS | |
72 class TaskQueueStats { | |
73 public: | |
74 enum StatId { | |
75 push, // number of taskqueue pushes | |
76 pop, // number of taskqueue pops | |
77 pop_slow, // subset of taskqueue pops that were done slow-path | |
78 steal_attempt, // number of taskqueue steal attempts | |
79 steal, // number of taskqueue steals | |
80 overflow, // number of overflow pushes | |
81 overflow_max_len, // max length of overflow stack | |
82 last_stat_id | |
83 }; | |
84 | |
85 public: | |
86 inline TaskQueueStats() { reset(); } | |
87 | |
88 inline void record_push() { ++_stats[push]; } | |
89 inline void record_pop() { ++_stats[pop]; } | |
90 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; } | |
91 inline void record_steal(bool success); | |
92 inline void record_overflow(size_t new_length); | |
93 | |
1709 | 94 TaskQueueStats & operator +=(const TaskQueueStats & addend); |
95 | |
1665 | 96 inline size_t get(StatId id) const { return _stats[id]; } |
97 inline const size_t* get() const { return _stats; } | |
98 | |
99 inline void reset(); | |
100 | |
1709 | 101 // Print the specified line of the header (does not include a line separator). |
1665 | 102 static void print_header(unsigned int line, outputStream* const stream = tty, |
103 unsigned int width = 10); | |
1709 | 104 // Print the statistics (does not include a line separator). |
1665 | 105 void print(outputStream* const stream = tty, unsigned int width = 10) const; |
106 | |
1709 | 107 DEBUG_ONLY(void verify() const;) |
108 | |
1665 | 109 private: |
110 size_t _stats[last_stat_id]; | |
111 static const char * const _names[last_stat_id]; | |
112 }; | |
113 | |
114 void TaskQueueStats::record_steal(bool success) { | |
115 ++_stats[steal_attempt]; | |
116 if (success) ++_stats[steal]; | |
117 } | |
118 | |
119 void TaskQueueStats::record_overflow(size_t new_len) { | |
120 ++_stats[overflow]; | |
121 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len; | |
122 } | |
123 | |
124 void TaskQueueStats::reset() { | |
125 memset(_stats, 0, sizeof(_stats)); | |
126 } | |
127 #endif // TASKQUEUE_STATS | |
128 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
129 template <unsigned int N> |
0 | 130 class TaskQueueSuper: public CHeapObj { |
131 protected: | |
907 | 132 // Internal type for indexing the queue; also used for the tag. |
133 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t; | |
134 | |
135 // 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
|
136 volatile uint _bottom; |
0 | 137 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
138 enum { MOD_N_MASK = N - 1 }; |
907 | 139 |
140 class Age { | |
141 public: | |
142 Age(size_t data = 0) { _data = data; } | |
143 Age(const Age& age) { _data = age._data; } | |
144 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; } | |
0 | 145 |
907 | 146 Age get() const volatile { return _data; } |
147 void set(Age age) volatile { _data = age._data; } | |
148 | |
149 idx_t top() const volatile { return _fields._top; } | |
150 idx_t tag() const volatile { return _fields._tag; } | |
0 | 151 |
907 | 152 // Increment top; if it wraps, increment tag also. |
153 void increment() { | |
154 _fields._top = increment_index(_fields._top); | |
155 if (_fields._top == 0) ++_fields._tag; | |
156 } | |
0 | 157 |
907 | 158 Age cmpxchg(const Age new_age, const Age old_age) volatile { |
159 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data, | |
160 (volatile intptr_t *)&_data, | |
161 (intptr_t)old_age._data); | |
162 } | |
163 | |
164 bool operator ==(const Age& other) const { return _data == other._data; } | |
0 | 165 |
907 | 166 private: |
167 struct fields { | |
168 idx_t _top; | |
169 idx_t _tag; | |
170 }; | |
171 union { | |
172 size_t _data; | |
173 fields _fields; | |
174 }; | |
0 | 175 }; |
907 | 176 |
177 volatile Age _age; | |
178 | |
179 // These both operate mod N. | |
180 static uint increment_index(uint ind) { | |
181 return (ind + 1) & MOD_N_MASK; | |
0 | 182 } |
907 | 183 static uint decrement_index(uint ind) { |
184 return (ind - 1) & MOD_N_MASK; | |
0 | 185 } |
186 | |
907 | 187 // Returns a number in the range [0..N). If the result is "N-1", it should be |
188 // interpreted as 0. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
189 uint dirty_size(uint bot, uint top) const { |
907 | 190 return (bot - top) & MOD_N_MASK; |
0 | 191 } |
192 | |
193 // 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
|
194 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
|
195 uint sz = dirty_size(bot, top); |
907 | 196 // Has the queue "wrapped", so that bottom is less than top? There's a |
197 // complicated special case here. A pair of threads could perform pop_local | |
198 // and pop_global operations concurrently, starting from a state in which | |
199 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom, | |
200 // and the pop_global in incrementing _top (in which case the pop_global | |
201 // will be awarded the contested queue element.) The resulting state must | |
202 // be interpreted as an empty queue. (We only need to worry about one such | |
203 // event: only the queue owner performs pop_local's, and several concurrent | |
204 // threads attempting to perform the pop_global will all perform the same | |
205 // CAS, and only one can succeed.) Any stealing thread that reads after | |
206 // either the increment or decrement will see an empty queue, and will not | |
207 // join the competitors. The "sz == -1 || sz == N-1" state will not be | |
208 // modified by concurrent queues, so the owner thread can reset the state to | |
209 // _bottom == top so subsequent pushes will be performed normally. | |
210 return (sz == N - 1) ? 0 : sz; | |
0 | 211 } |
212 | |
213 public: | |
214 TaskQueueSuper() : _bottom(0), _age() {} | |
215 | |
1638 | 216 // Return true if the TaskQueue contains/does not contain any tasks. |
217 bool peek() const { return _bottom != _age.top(); } | |
218 bool is_empty() const { return size() == 0; } | |
0 | 219 |
220 // Return an estimate of the number of elements in the queue. | |
221 // The "careful" version admits the possibility of pop_local/pop_global | |
222 // races. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
223 uint size() const { |
907 | 224 return size(_bottom, _age.top()); |
0 | 225 } |
226 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
227 uint dirty_size() const { |
907 | 228 return dirty_size(_bottom, _age.top()); |
0 | 229 } |
230 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
231 void set_empty() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
232 _bottom = 0; |
907 | 233 _age.set(0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
234 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
235 |
0 | 236 // Maximum number of elements allowed in the queue. This is two less |
237 // 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
|
238 uint max_elems() const { return N - 2; } |
1284 | 239 |
240 // Total size of queue. | |
241 static const uint total_size() { return N; } | |
1665 | 242 |
243 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;) | |
0 | 244 }; |
245 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
246 template<class E, unsigned int N = TASKQUEUE_SIZE> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
247 class GenericTaskQueue: public TaskQueueSuper<N> { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
248 protected: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
249 typedef typename TaskQueueSuper<N>::Age Age; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
250 typedef typename TaskQueueSuper<N>::idx_t idx_t; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
251 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
252 using TaskQueueSuper<N>::_bottom; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
253 using TaskQueueSuper<N>::_age; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
254 using TaskQueueSuper<N>::increment_index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
255 using TaskQueueSuper<N>::decrement_index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
256 using TaskQueueSuper<N>::dirty_size; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
257 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
258 public: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
259 using TaskQueueSuper<N>::max_elems; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
260 using TaskQueueSuper<N>::size; |
1665 | 261 TASKQUEUE_STATS_ONLY(using TaskQueueSuper<N>::stats;) |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
262 |
0 | 263 private: |
264 // 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
|
265 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
|
266 bool pop_local_slow(uint localBot, Age oldAge); |
0 | 267 |
268 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
269 typedef E element_type; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
270 |
0 | 271 // Initializes the queue to empty. |
272 GenericTaskQueue(); | |
273 | |
274 void initialize(); | |
275 | |
1638 | 276 // Push the task "t" on the queue. Returns "false" iff the queue is full. |
0 | 277 inline bool push(E t); |
278 | |
1638 | 279 // Attempts to claim a task from the "local" end of the queue (the most |
280 // recently pushed). If successful, returns true and sets t to the task; | |
281 // otherwise, returns false (the queue is empty). | |
0 | 282 inline bool pop_local(E& t); |
283 | |
1638 | 284 // Like pop_local(), but uses the "global" end of the queue (the least |
285 // recently pushed). | |
0 | 286 bool pop_global(E& t); |
287 | |
288 // Delete any resource associated with the queue. | |
289 ~GenericTaskQueue(); | |
290 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
291 // apply the closure to all elements in the task queue |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
292 void oops_do(OopClosure* f); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
293 |
0 | 294 private: |
295 // Element array. | |
296 volatile E* _elems; | |
297 }; | |
298 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
299 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
300 GenericTaskQueue<E, N>::GenericTaskQueue() { |
907 | 301 assert(sizeof(Age) == sizeof(size_t), "Depends on this."); |
0 | 302 } |
303 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
304 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
305 void GenericTaskQueue<E, N>::initialize() { |
907 | 306 _elems = NEW_C_HEAP_ARRAY(E, N); |
0 | 307 } |
308 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
309 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
310 void GenericTaskQueue<E, N>::oops_do(OopClosure* f) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
311 // 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
|
312 uint iters = size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
313 uint index = _bottom; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
314 for (uint i = 0; i < iters; ++i) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
315 index = decrement_index(index); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
316 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
317 // index, &_elems[index], _elems[index]); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
318 E* t = (E*)&_elems[index]; // cast away volatility |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
319 oop* p = (oop*)t; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
320 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
|
321 f->do_oop(p); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
322 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
323 // tty->print_cr("END OopTaskQueue::oops_do"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
324 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
325 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
326 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
327 bool GenericTaskQueue<E, N>::push_slow(E t, uint dirty_n_elems) { |
907 | 328 if (dirty_n_elems == N - 1) { |
0 | 329 // 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
|
330 uint localBot = _bottom; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
331 // 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
|
332 const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
333 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 334 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 335 return true; |
907 | 336 } |
337 return false; | |
0 | 338 } |
339 | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
340 // 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
|
341 // 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
|
342 // 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
|
343 // 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
|
344 // 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
|
345 // and pop_global() succeeds, see pop_global()). |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
346 template<class E, unsigned int N> |
1665 | 347 bool GenericTaskQueue<E, N>::pop_local_slow(uint localBot, Age oldAge) { |
0 | 348 // This queue was observed to contain exactly one element; either this |
349 // thread will claim it, or a competing "pop_global". In either case, | |
350 // the queue will be logically empty afterwards. Create a new Age value | |
351 // that represents the empty queue for the given value of "_bottom". (We | |
352 // must also increment "tag" because of the case where "bottom == 1", | |
353 // "top == 0". A pop_global could read the queue element in that case, | |
354 // then have the owner thread do a pop followed by another push. Without | |
355 // the incrementing of "tag", the pop_global's CAS could succeed, | |
356 // allowing it to believe it has claimed the stale element.) | |
907 | 357 Age newAge((idx_t)localBot, oldAge.tag() + 1); |
0 | 358 // Perhaps a competing pop_global has already incremented "top", in which |
359 // case it wins the element. | |
360 if (localBot == oldAge.top()) { | |
361 // No competing pop_global has yet incremented "top"; we'll try to | |
362 // install new_age, thus claiming the element. | |
907 | 363 Age tempAge = _age.cmpxchg(newAge, oldAge); |
0 | 364 if (tempAge == oldAge) { |
365 // We win. | |
907 | 366 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); |
1665 | 367 TASKQUEUE_STATS_ONLY(stats.record_pop_slow()); |
0 | 368 return true; |
369 } | |
370 } | |
907 | 371 // We lose; a completing pop_global gets the element. But the queue is empty |
372 // and top is greater than bottom. Fix this representation of the empty queue | |
373 // to become the canonical one. | |
374 _age.set(newAge); | |
375 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); | |
0 | 376 return false; |
377 } | |
378 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
379 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
380 bool GenericTaskQueue<E, N>::pop_global(E& t) { |
907 | 381 Age oldAge = _age.get(); |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
382 uint localBot = _bottom; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
383 uint n_elems = size(localBot, oldAge.top()); |
0 | 384 if (n_elems == 0) { |
385 return false; | |
386 } | |
907 | 387 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
388 const_cast<E&>(t = _elems[oldAge.top()]); |
907 | 389 Age newAge(oldAge); |
390 newAge.increment(); | |
391 Age resAge = _age.cmpxchg(newAge, oldAge); | |
392 | |
0 | 393 // Note that using "_bottom" here might fail, since a pop_local might |
394 // have decremented it. | |
907 | 395 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity"); |
396 return resAge == oldAge; | |
0 | 397 } |
398 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
399 template<class E, unsigned int N> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
400 GenericTaskQueue<E, N>::~GenericTaskQueue() { |
0 | 401 FREE_C_HEAP_ARRAY(E, _elems); |
402 } | |
403 | |
1638 | 404 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for |
405 // elements that do not fit in the TaskQueue. | |
406 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
407 // This class hides two methods from super classes: |
1638 | 408 // |
409 // push() - push onto the task queue or, if that fails, onto the overflow stack | |
410 // is_empty() - return true if both the TaskQueue and overflow stack are empty | |
411 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
412 // Note that size() is not hidden--it returns the number of elements in the |
1638 | 413 // TaskQueue, and does not include the size of the overflow stack. This |
414 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues. | |
415 template<class E, unsigned int N = TASKQUEUE_SIZE> | |
416 class OverflowTaskQueue: public GenericTaskQueue<E, N> | |
417 { | |
418 public: | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
419 typedef Stack<E> overflow_t; |
1638 | 420 typedef GenericTaskQueue<E, N> taskqueue_t; |
421 | |
1665 | 422 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;) |
423 | |
1638 | 424 // Push task t onto the queue or onto the overflow stack. Return true. |
425 inline bool push(E t); | |
426 | |
427 // Attempt to pop from the overflow stack; return true if anything was popped. | |
428 inline bool pop_overflow(E& t); | |
429 | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
430 inline overflow_t* overflow_stack() { return &_overflow_stack; } |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
431 |
1638 | 432 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
|
433 inline bool overflow_empty() const { return _overflow_stack.is_empty(); } |
1638 | 434 inline bool is_empty() const { |
435 return taskqueue_empty() && overflow_empty(); | |
436 } | |
437 | |
438 private: | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
439 overflow_t _overflow_stack; |
1638 | 440 }; |
441 | |
442 template <class E, unsigned int N> | |
443 bool OverflowTaskQueue<E, N>::push(E t) | |
444 { | |
445 if (!taskqueue_t::push(t)) { | |
446 overflow_stack()->push(t); | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
447 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size())); |
1638 | 448 } |
449 return true; | |
450 } | |
451 | |
452 template <class E, unsigned int N> | |
453 bool OverflowTaskQueue<E, N>::pop_overflow(E& t) | |
454 { | |
455 if (overflow_empty()) return false; | |
456 t = overflow_stack()->pop(); | |
457 return true; | |
458 } | |
459 | |
0 | 460 class TaskQueueSetSuper: public CHeapObj { |
461 protected: | |
462 static int randomParkAndMiller(int* seed0); | |
463 public: | |
464 // Returns "true" if some TaskQueue in the set contains a task. | |
465 virtual bool peek() = 0; | |
466 }; | |
467 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
468 template<class T> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
469 class GenericTaskQueueSet: public TaskQueueSetSuper { |
0 | 470 private: |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
471 uint _n; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
472 T** _queues; |
0 | 473 |
474 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
475 typedef typename T::element_type E; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
476 |
0 | 477 GenericTaskQueueSet(int n) : _n(n) { |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
478 typedef T* GenericTaskQueuePtr; |
0 | 479 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n); |
480 for (int i = 0; i < n; i++) { | |
481 _queues[i] = NULL; | |
482 } | |
483 } | |
484 | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
485 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
|
486 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
|
487 bool steal_best_of_all(uint queue_num, int* seed, E& t); |
0 | 488 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
489 void register_queue(uint i, T* q); |
0 | 490 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
491 T* queue(uint n); |
0 | 492 |
1638 | 493 // The thread with queue number "queue_num" (and whose random number seed is |
494 // at "seed") is trying to steal a task from some other queue. (It may try | |
495 // several queues, according to some configuration parameter.) If some steal | |
496 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns | |
497 // false. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
498 bool steal(uint queue_num, int* seed, E& t); |
0 | 499 |
500 bool peek(); | |
501 }; | |
502 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
503 template<class T> void |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
504 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
|
505 assert(i < _n, "index out of range."); |
0 | 506 _queues[i] = q; |
507 } | |
508 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
509 template<class T> T* |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
510 GenericTaskQueueSet<T>::queue(uint i) { |
0 | 511 return _queues[i]; |
512 } | |
513 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
514 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
515 GenericTaskQueueSet<T>::steal(uint queue_num, int* seed, E& t) { |
1665 | 516 for (uint i = 0; i < 2 * _n; i++) { |
517 if (steal_best_of_2(queue_num, seed, t)) { | |
518 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true)); | |
0 | 519 return true; |
1665 | 520 } |
521 } | |
522 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false)); | |
0 | 523 return false; |
524 } | |
525 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
526 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
527 GenericTaskQueueSet<T>::steal_best_of_all(uint queue_num, int* seed, E& t) { |
0 | 528 if (_n > 2) { |
529 int best_k; | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
530 uint best_sz = 0; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
531 for (uint k = 0; k < _n; k++) { |
0 | 532 if (k == queue_num) continue; |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
533 uint sz = _queues[k]->size(); |
0 | 534 if (sz > best_sz) { |
535 best_sz = sz; | |
536 best_k = k; | |
537 } | |
538 } | |
539 return best_sz > 0 && _queues[best_k]->pop_global(t); | |
540 } else if (_n == 2) { | |
541 // Just try the other one. | |
542 int k = (queue_num + 1) % 2; | |
543 return _queues[k]->pop_global(t); | |
544 } else { | |
545 assert(_n == 1, "can't be zero."); | |
546 return false; | |
547 } | |
548 } | |
549 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
550 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
551 GenericTaskQueueSet<T>::steal_1_random(uint queue_num, int* seed, E& t) { |
0 | 552 if (_n > 2) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
553 uint k = queue_num; |
0 | 554 while (k == queue_num) k = randomParkAndMiller(seed) % _n; |
555 return _queues[2]->pop_global(t); | |
556 } else if (_n == 2) { | |
557 // Just try the other one. | |
558 int k = (queue_num + 1) % 2; | |
559 return _queues[k]->pop_global(t); | |
560 } else { | |
561 assert(_n == 1, "can't be zero."); | |
562 return false; | |
563 } | |
564 } | |
565 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
566 template<class T> bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
567 GenericTaskQueueSet<T>::steal_best_of_2(uint queue_num, int* seed, E& t) { |
0 | 568 if (_n > 2) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
569 uint k1 = queue_num; |
0 | 570 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
|
571 uint k2 = queue_num; |
0 | 572 while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n; |
573 // Sample both and try the larger. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
574 uint sz1 = _queues[k1]->size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
575 uint sz2 = _queues[k2]->size(); |
0 | 576 if (sz2 > sz1) return _queues[k2]->pop_global(t); |
577 else return _queues[k1]->pop_global(t); | |
578 } else if (_n == 2) { | |
579 // Just try the other one. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
580 uint k = (queue_num + 1) % 2; |
0 | 581 return _queues[k]->pop_global(t); |
582 } else { | |
583 assert(_n == 1, "can't be zero."); | |
584 return false; | |
585 } | |
586 } | |
587 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
588 template<class T> |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
589 bool GenericTaskQueueSet<T>::peek() { |
0 | 590 // Try all the queues. |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
591 for (uint j = 0; j < _n; j++) { |
0 | 592 if (_queues[j]->peek()) |
593 return true; | |
594 } | |
595 return false; | |
596 } | |
597 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
598 // When to terminate from the termination protocol. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
599 class TerminatorTerminator: public CHeapObj { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
600 public: |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
601 virtual bool should_exit_termination() = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
602 }; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
603 |
0 | 604 // A class to aid in the termination of a set of parallel tasks using |
605 // TaskQueueSet's for work stealing. | |
606 | |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
607 #undef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
608 |
0 | 609 class ParallelTaskTerminator: public StackObj { |
610 private: | |
611 int _n_threads; | |
612 TaskQueueSetSuper* _queue_set; | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
613 int _offered_termination; |
0 | 614 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
615 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
616 static uint _total_yields; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
617 static uint _total_spins; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
618 static uint _total_peeks; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
619 #endif |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
620 |
0 | 621 bool peek_in_queue_set(); |
622 protected: | |
623 virtual void yield(); | |
624 void sleep(uint millis); | |
625 | |
626 public: | |
627 | |
628 // "n_threads" is the number of threads to be terminated. "queue_set" is a | |
629 // queue sets of work queues of other threads. | |
630 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set); | |
631 | |
632 // The current thread has no work, and is ready to terminate if everyone | |
633 // else is. If returns "true", all threads are terminated. If returns | |
634 // "false", available work has been observed in one of the task queues, | |
635 // so the global task is not complete. | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
636 bool offer_termination() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
637 return offer_termination(NULL); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
638 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
639 |
907 | 640 // 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
|
641 // method of the terminator parameter returns true. If terminator is |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
642 // NULL, then it is ignored. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
643 bool offer_termination(TerminatorTerminator* terminator); |
0 | 644 |
645 // Reset the terminator, so that it may be reused again. | |
646 // The caller is responsible for ensuring that this is done | |
647 // in an MT-safe manner, once the previous round of use of | |
648 // the terminator is finished. | |
649 void reset_for_reuse(); | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
650 // 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
|
651 // given number. |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
652 void reset_for_reuse(int n_threads); |
0 | 653 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
654 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
655 static uint total_yields() { return _total_yields; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
656 static uint total_spins() { return _total_spins; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
657 static uint total_peeks() { return _total_peeks; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
658 static void print_termination_counts(); |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
659 #endif |
0 | 660 }; |
661 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
662 template<class E, unsigned int N> inline bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
663 GenericTaskQueue<E, N>::push(E t) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
664 uint localBot = _bottom; |
907 | 665 assert((localBot >= 0) && (localBot < N), "_bottom out of range."); |
666 idx_t top = _age.top(); | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
667 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
|
668 assert(dirty_n_elems < N, "n_elems out of range."); |
0 | 669 if (dirty_n_elems < max_elems()) { |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
670 // 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
|
671 const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
672 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 673 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 674 return true; |
675 } else { | |
676 return push_slow(t, dirty_n_elems); | |
677 } | |
678 } | |
679 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
680 template<class E, unsigned int N> inline bool |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
681 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
|
682 uint localBot = _bottom; |
907 | 683 // This value cannot be N-1. That can only occur as a result of |
0 | 684 // the assignment to bottom in this method. If it does, this method |
1638 | 685 // resets the size to 0 before the next call (which is sequential, |
0 | 686 // since this is pop_local.) |
907 | 687 uint dirty_n_elems = dirty_size(localBot, _age.top()); |
688 assert(dirty_n_elems != N - 1, "Shouldn't be possible..."); | |
0 | 689 if (dirty_n_elems == 0) return false; |
690 localBot = decrement_index(localBot); | |
691 _bottom = localBot; | |
692 // This is necessary to prevent any read below from being reordered | |
693 // before the store just above. | |
694 OrderAccess::fence(); | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
695 const_cast<E&>(t = _elems[localBot]); |
0 | 696 // This is a second read of "age"; the "size()" above is the first. |
697 // If there's still at least one element in the queue, based on the | |
698 // "_bottom" and "age" we've read, then there can be no interference with | |
699 // a "pop_global" operation, and we're done. | |
907 | 700 idx_t tp = _age.top(); // XXX |
0 | 701 if (size(localBot, tp) > 0) { |
907 | 702 assert(dirty_size(localBot, tp) != N - 1, "sanity"); |
1665 | 703 TASKQUEUE_STATS_ONLY(stats.record_pop()); |
0 | 704 return true; |
705 } else { | |
706 // Otherwise, the queue contained exactly one element; we take the slow | |
707 // path. | |
907 | 708 return pop_local_slow(localBot, _age.get()); |
0 | 709 } |
710 } | |
711 | |
1638 | 712 typedef GenericTaskQueue<oop> OopTaskQueue; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
713 typedef GenericTaskQueueSet<OopTaskQueue> OopTaskQueueSet; |
0 | 714 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
715 #ifdef _MSC_VER |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
716 #pragma warning(push) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
717 // warning C4522: multiple assignment operators specified |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
718 #pragma warning(disable:4522) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
719 #endif |
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 // 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
|
722 // 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
|
723 // 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
|
724 class StarTask { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
725 void* _holder; // either union oop* or narrowOop* |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
726 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
727 enum { COMPRESSED_OOP_MASK = 1 }; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
728 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
729 public: |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
730 StarTask(narrowOop* p) { |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
731 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
|
732 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
733 } |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
734 StarTask(oop* p) { |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
735 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
|
736 _holder = (void*)p; |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
737 } |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
738 StarTask() { _holder = NULL; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
739 operator oop*() { return (oop*)_holder; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
740 operator narrowOop*() { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
741 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
|
742 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
743 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
744 StarTask& operator=(const StarTask& t) { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
745 _holder = t._holder; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
746 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
747 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
748 volatile StarTask& operator=(const volatile StarTask& t) volatile { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
749 _holder = t._holder; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
750 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
751 } |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
752 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
753 bool is_narrow() const { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
754 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
|
755 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
756 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
757 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
758 class ObjArrayTask |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
759 { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
760 public: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
761 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
|
762 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
|
763 assert(idx <= size_t(max_jint), "too big"); |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
764 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
765 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
|
766 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
767 ObjArrayTask& operator =(const ObjArrayTask& t) { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
768 _obj = t._obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
769 _index = t._index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
770 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
771 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
772 volatile ObjArrayTask& |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
773 operator =(const volatile ObjArrayTask& t) volatile { |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
774 _obj = t._obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
775 _index = t._index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
776 return *this; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
777 } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
778 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
779 inline oop obj() const { return _obj; } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
780 inline int index() const { return _index; } |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
781 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
782 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
|
783 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
784 private: |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
785 oop _obj; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
786 int _index; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
787 }; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
788 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
789 #ifdef _MSC_VER |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
790 #pragma warning(pop) |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
791 #endif |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
792 |
1638 | 793 typedef OverflowTaskQueue<StarTask> OopStarTaskQueue; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
794 typedef GenericTaskQueueSet<OopStarTaskQueue> OopStarTaskQueueSet; |
0 | 795 |
1638 | 796 typedef OverflowTaskQueue<size_t> RegionTaskQueue; |
797 typedef GenericTaskQueueSet<RegionTaskQueue> RegionTaskQueueSet; | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
798 |
1972 | 799 |
800 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP |