Mercurial > hg > graal-compiler
annotate src/share/vm/utilities/taskqueue.hpp @ 14411:bdd155477289
8023033: PPC64 (part 13): basic changes for AIX
Summary: Added AIX includes alpha-sorted before BSD. Fix compilation issues with xlC in shared code. Basic shared platform dependend adaption (vm_version etc.).
Reviewed-by: kvn, dholmes, stefank
author | goetz |
---|---|
date | Thu, 22 Aug 2013 09:39:54 -0700 |
parents | cd25d3be91c5 |
children | e2722a66aba7 |
rev | line source |
---|---|
0 | 1 /* |
6842
b9a9ed0f8eeb
7197424: update copyright year to match last edit in jdk8 hotspot repository
mikael
parents:
6197
diff
changeset
|
2 * Copyright (c) 2001, 2012, 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 |
14411 | 56 #ifdef TARGET_OS_ARCH_aix_ppc |
57 # include "orderAccess_aix_ppc.inline.hpp" | |
58 #endif | |
3960 | 59 #ifdef TARGET_OS_ARCH_bsd_x86 |
60 # include "orderAccess_bsd_x86.inline.hpp" | |
61 #endif | |
62 #ifdef TARGET_OS_ARCH_bsd_zero | |
63 # include "orderAccess_bsd_zero.inline.hpp" | |
64 #endif | |
1972 | 65 |
1665 | 66 // Simple TaskQueue stats that are collected by default in debug builds. |
67 | |
68 #if !defined(TASKQUEUE_STATS) && defined(ASSERT) | |
69 #define TASKQUEUE_STATS 1 | |
70 #elif !defined(TASKQUEUE_STATS) | |
71 #define TASKQUEUE_STATS 0 | |
72 #endif | |
73 | |
74 #if TASKQUEUE_STATS | |
75 #define TASKQUEUE_STATS_ONLY(code) code | |
76 #else | |
77 #define TASKQUEUE_STATS_ONLY(code) | |
78 #endif // TASKQUEUE_STATS | |
79 | |
80 #if TASKQUEUE_STATS | |
81 class TaskQueueStats { | |
82 public: | |
83 enum StatId { | |
84 push, // number of taskqueue pushes | |
85 pop, // number of taskqueue pops | |
86 pop_slow, // subset of taskqueue pops that were done slow-path | |
87 steal_attempt, // number of taskqueue steal attempts | |
88 steal, // number of taskqueue steals | |
89 overflow, // number of overflow pushes | |
90 overflow_max_len, // max length of overflow stack | |
91 last_stat_id | |
92 }; | |
93 | |
94 public: | |
95 inline TaskQueueStats() { reset(); } | |
96 | |
97 inline void record_push() { ++_stats[push]; } | |
98 inline void record_pop() { ++_stats[pop]; } | |
99 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; } | |
100 inline void record_steal(bool success); | |
101 inline void record_overflow(size_t new_length); | |
102 | |
1709 | 103 TaskQueueStats & operator +=(const TaskQueueStats & addend); |
104 | |
1665 | 105 inline size_t get(StatId id) const { return _stats[id]; } |
106 inline const size_t* get() const { return _stats; } | |
107 | |
108 inline void reset(); | |
109 | |
1709 | 110 // Print the specified line of the header (does not include a line separator). |
1665 | 111 static void print_header(unsigned int line, outputStream* const stream = tty, |
112 unsigned int width = 10); | |
1709 | 113 // Print the statistics (does not include a line separator). |
1665 | 114 void print(outputStream* const stream = tty, unsigned int width = 10) const; |
115 | |
1709 | 116 DEBUG_ONLY(void verify() const;) |
117 | |
1665 | 118 private: |
119 size_t _stats[last_stat_id]; | |
120 static const char * const _names[last_stat_id]; | |
121 }; | |
122 | |
123 void TaskQueueStats::record_steal(bool success) { | |
124 ++_stats[steal_attempt]; | |
125 if (success) ++_stats[steal]; | |
126 } | |
127 | |
128 void TaskQueueStats::record_overflow(size_t new_len) { | |
129 ++_stats[overflow]; | |
130 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len; | |
131 } | |
132 | |
133 void TaskQueueStats::reset() { | |
134 memset(_stats, 0, sizeof(_stats)); | |
135 } | |
136 #endif // TASKQUEUE_STATS | |
137 | |
6197 | 138 template <unsigned int N, MEMFLAGS F> |
139 class TaskQueueSuper: public CHeapObj<F> { | |
0 | 140 protected: |
907 | 141 // Internal type for indexing the queue; also used for the tag. |
142 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t; | |
143 | |
144 // 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
|
145 volatile uint _bottom; |
0 | 146 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
147 enum { MOD_N_MASK = N - 1 }; |
907 | 148 |
149 class Age { | |
150 public: | |
151 Age(size_t data = 0) { _data = data; } | |
152 Age(const Age& age) { _data = age._data; } | |
153 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; } | |
0 | 154 |
907 | 155 Age get() const volatile { return _data; } |
156 void set(Age age) volatile { _data = age._data; } | |
157 | |
158 idx_t top() const volatile { return _fields._top; } | |
159 idx_t tag() const volatile { return _fields._tag; } | |
0 | 160 |
907 | 161 // Increment top; if it wraps, increment tag also. |
162 void increment() { | |
163 _fields._top = increment_index(_fields._top); | |
164 if (_fields._top == 0) ++_fields._tag; | |
165 } | |
0 | 166 |
907 | 167 Age cmpxchg(const Age new_age, const Age old_age) volatile { |
168 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data, | |
169 (volatile intptr_t *)&_data, | |
170 (intptr_t)old_age._data); | |
171 } | |
172 | |
173 bool operator ==(const Age& other) const { return _data == other._data; } | |
0 | 174 |
907 | 175 private: |
176 struct fields { | |
177 idx_t _top; | |
178 idx_t _tag; | |
179 }; | |
180 union { | |
181 size_t _data; | |
182 fields _fields; | |
183 }; | |
0 | 184 }; |
907 | 185 |
186 volatile Age _age; | |
187 | |
188 // These both operate mod N. | |
189 static uint increment_index(uint ind) { | |
190 return (ind + 1) & MOD_N_MASK; | |
0 | 191 } |
907 | 192 static uint decrement_index(uint ind) { |
193 return (ind - 1) & MOD_N_MASK; | |
0 | 194 } |
195 | |
907 | 196 // Returns a number in the range [0..N). If the result is "N-1", it should be |
197 // interpreted as 0. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
198 uint dirty_size(uint bot, uint top) const { |
907 | 199 return (bot - top) & MOD_N_MASK; |
0 | 200 } |
201 | |
202 // 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
|
203 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
|
204 uint sz = dirty_size(bot, top); |
907 | 205 // Has the queue "wrapped", so that bottom is less than top? There's a |
206 // complicated special case here. A pair of threads could perform pop_local | |
207 // and pop_global operations concurrently, starting from a state in which | |
208 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom, | |
209 // and the pop_global in incrementing _top (in which case the pop_global | |
210 // will be awarded the contested queue element.) The resulting state must | |
211 // be interpreted as an empty queue. (We only need to worry about one such | |
212 // event: only the queue owner performs pop_local's, and several concurrent | |
213 // threads attempting to perform the pop_global will all perform the same | |
214 // CAS, and only one can succeed.) Any stealing thread that reads after | |
215 // either the increment or decrement will see an empty queue, and will not | |
216 // join the competitors. The "sz == -1 || sz == N-1" state will not be | |
217 // modified by concurrent queues, so the owner thread can reset the state to | |
218 // _bottom == top so subsequent pushes will be performed normally. | |
219 return (sz == N - 1) ? 0 : sz; | |
0 | 220 } |
221 | |
222 public: | |
223 TaskQueueSuper() : _bottom(0), _age() {} | |
224 | |
1638 | 225 // Return true if the TaskQueue contains/does not contain any tasks. |
226 bool peek() const { return _bottom != _age.top(); } | |
227 bool is_empty() const { return size() == 0; } | |
0 | 228 |
229 // Return an estimate of the number of elements in the queue. | |
230 // The "careful" version admits the possibility of pop_local/pop_global | |
231 // races. | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
232 uint size() const { |
907 | 233 return size(_bottom, _age.top()); |
0 | 234 } |
235 | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
236 uint dirty_size() const { |
907 | 237 return dirty_size(_bottom, _age.top()); |
0 | 238 } |
239 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
240 void set_empty() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
241 _bottom = 0; |
907 | 242 _age.set(0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
243 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
244 |
0 | 245 // Maximum number of elements allowed in the queue. This is two less |
246 // 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
|
247 uint max_elems() const { return N - 2; } |
1284 | 248 |
249 // Total size of queue. | |
250 static const uint total_size() { return N; } | |
1665 | 251 |
252 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;) | |
0 | 253 }; |
254 | |
6197 | 255 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
256 |
6197 | 257 template <class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> |
258 class GenericTaskQueue: public TaskQueueSuper<N, F> { | |
9073
83f27710f5f7
7197666: java -d64 -version core dumps in a box with lots of memory
brutisso
parents:
6924
diff
changeset
|
259 ArrayAllocator<E, F> _array_allocator; |
6197 | 260 protected: |
261 typedef typename TaskQueueSuper<N, F>::Age Age; | |
262 typedef typename TaskQueueSuper<N, F>::idx_t idx_t; | |
263 | |
264 using TaskQueueSuper<N, F>::_bottom; | |
265 using TaskQueueSuper<N, F>::_age; | |
266 using TaskQueueSuper<N, F>::increment_index; | |
267 using TaskQueueSuper<N, F>::decrement_index; | |
268 using TaskQueueSuper<N, F>::dirty_size; | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
269 |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
270 public: |
6197 | 271 using TaskQueueSuper<N, F>::max_elems; |
272 using TaskQueueSuper<N, F>::size; | |
273 | |
274 #if TASKQUEUE_STATS | |
275 using TaskQueueSuper<N, F>::stats; | |
276 #endif | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
277 |
0 | 278 private: |
279 // 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
|
280 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
|
281 bool pop_local_slow(uint localBot, Age oldAge); |
0 | 282 |
283 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
284 typedef E element_type; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
285 |
0 | 286 // Initializes the queue to empty. |
287 GenericTaskQueue(); | |
288 | |
289 void initialize(); | |
290 | |
1638 | 291 // Push the task "t" on the queue. Returns "false" iff the queue is full. |
0 | 292 inline bool push(E t); |
293 | |
1638 | 294 // Attempts to claim a task from the "local" end of the queue (the most |
295 // recently pushed). If successful, returns true and sets t to the task; | |
296 // otherwise, returns false (the queue is empty). | |
0 | 297 inline bool pop_local(E& t); |
298 | |
1638 | 299 // Like pop_local(), but uses the "global" end of the queue (the least |
300 // recently pushed). | |
0 | 301 bool pop_global(E& t); |
302 | |
303 // Delete any resource associated with the queue. | |
304 ~GenericTaskQueue(); | |
305 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
306 // apply the closure to all elements in the task queue |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
307 void oops_do(OopClosure* f); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
308 |
0 | 309 private: |
310 // Element array. | |
311 volatile E* _elems; | |
312 }; | |
313 | |
6197 | 314 template<class E, MEMFLAGS F, unsigned int N> |
315 GenericTaskQueue<E, F, N>::GenericTaskQueue() { | |
907 | 316 assert(sizeof(Age) == sizeof(size_t), "Depends on this."); |
0 | 317 } |
318 | |
6197 | 319 template<class E, MEMFLAGS F, unsigned int N> |
320 void GenericTaskQueue<E, F, N>::initialize() { | |
9073
83f27710f5f7
7197666: java -d64 -version core dumps in a box with lots of memory
brutisso
parents:
6924
diff
changeset
|
321 _elems = _array_allocator.allocate(N); |
0 | 322 } |
323 | |
6197 | 324 template<class E, MEMFLAGS F, unsigned int N> |
325 void GenericTaskQueue<E, F, N>::oops_do(OopClosure* f) { | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
326 // 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
|
327 uint iters = size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
328 uint index = _bottom; |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
329 for (uint i = 0; i < iters; ++i) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
330 index = decrement_index(index); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
331 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
332 // index, &_elems[index], _elems[index]); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
333 E* t = (E*)&_elems[index]; // cast away volatility |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
334 oop* p = (oop*)t; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
335 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
|
336 f->do_oop(p); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
337 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
338 // tty->print_cr("END OopTaskQueue::oops_do"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
339 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
340 |
6197 | 341 template<class E, MEMFLAGS F, unsigned int N> |
342 bool GenericTaskQueue<E, F, N>::push_slow(E t, uint dirty_n_elems) { | |
907 | 343 if (dirty_n_elems == N - 1) { |
0 | 344 // 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
|
345 uint localBot = _bottom; |
10973
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
346 // g++ complains if the volatile result of the assignment is |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
347 // unused, so we cast the volatile away. We cannot cast directly |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
348 // to void, because gcc treats that as not using the result of the |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
349 // assignment. However, casting to E& means that we trigger an |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
350 // unused-value warning. So, we cast the E& to void. |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
351 (void)const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
352 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 353 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 354 return true; |
907 | 355 } |
356 return false; | |
0 | 357 } |
358 | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
359 // 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
|
360 // 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
|
361 // 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
|
362 // 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
|
363 // 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
|
364 // and pop_global() succeeds, see pop_global()). |
6197 | 365 template<class E, MEMFLAGS F, unsigned int N> |
366 bool GenericTaskQueue<E, F, N>::pop_local_slow(uint localBot, Age oldAge) { | |
0 | 367 // This queue was observed to contain exactly one element; either this |
368 // thread will claim it, or a competing "pop_global". In either case, | |
369 // the queue will be logically empty afterwards. Create a new Age value | |
370 // that represents the empty queue for the given value of "_bottom". (We | |
371 // must also increment "tag" because of the case where "bottom == 1", | |
372 // "top == 0". A pop_global could read the queue element in that case, | |
373 // then have the owner thread do a pop followed by another push. Without | |
374 // the incrementing of "tag", the pop_global's CAS could succeed, | |
375 // allowing it to believe it has claimed the stale element.) | |
907 | 376 Age newAge((idx_t)localBot, oldAge.tag() + 1); |
0 | 377 // Perhaps a competing pop_global has already incremented "top", in which |
378 // case it wins the element. | |
379 if (localBot == oldAge.top()) { | |
380 // No competing pop_global has yet incremented "top"; we'll try to | |
381 // install new_age, thus claiming the element. | |
907 | 382 Age tempAge = _age.cmpxchg(newAge, oldAge); |
0 | 383 if (tempAge == oldAge) { |
384 // We win. | |
907 | 385 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); |
1665 | 386 TASKQUEUE_STATS_ONLY(stats.record_pop_slow()); |
0 | 387 return true; |
388 } | |
389 } | |
907 | 390 // We lose; a completing pop_global gets the element. But the queue is empty |
391 // and top is greater than bottom. Fix this representation of the empty queue | |
392 // to become the canonical one. | |
393 _age.set(newAge); | |
394 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); | |
0 | 395 return false; |
396 } | |
397 | |
6197 | 398 template<class E, MEMFLAGS F, unsigned int N> |
399 bool GenericTaskQueue<E, F, N>::pop_global(E& t) { | |
907 | 400 Age oldAge = _age.get(); |
11997 | 401 // Architectures with weak memory model require a barrier here |
402 // to guarantee that bottom is not older than age, | |
403 // which is crucial for the correctness of the algorithm. | |
404 #if !(defined SPARC || defined IA32 || defined AMD64) | |
405 OrderAccess::fence(); | |
406 #endif | |
407 uint localBot = OrderAccess::load_acquire((volatile juint*)&_bottom); | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
408 uint n_elems = size(localBot, oldAge.top()); |
0 | 409 if (n_elems == 0) { |
410 return false; | |
411 } | |
907 | 412 |
10973
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
413 // g++ complains if the volatile result of the assignment is |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
414 // unused, so we cast the volatile away. We cannot cast directly |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
415 // to void, because gcc treats that as not using the result of the |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
416 // assignment. However, casting to E& means that we trigger an |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
417 // unused-value warning. So, we cast the E& to void. |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
418 (void) const_cast<E&>(t = _elems[oldAge.top()]); |
907 | 419 Age newAge(oldAge); |
420 newAge.increment(); | |
421 Age resAge = _age.cmpxchg(newAge, oldAge); | |
422 | |
0 | 423 // Note that using "_bottom" here might fail, since a pop_local might |
424 // have decremented it. | |
907 | 425 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity"); |
426 return resAge == oldAge; | |
0 | 427 } |
428 | |
6197 | 429 template<class E, MEMFLAGS F, unsigned int N> |
430 GenericTaskQueue<E, F, N>::~GenericTaskQueue() { | |
431 FREE_C_HEAP_ARRAY(E, _elems, F); | |
0 | 432 } |
433 | |
1638 | 434 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for |
435 // elements that do not fit in the TaskQueue. | |
436 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
437 // This class hides two methods from super classes: |
1638 | 438 // |
439 // push() - push onto the task queue or, if that fails, onto the overflow stack | |
440 // is_empty() - return true if both the TaskQueue and overflow stack are empty | |
441 // | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
442 // Note that size() is not hidden--it returns the number of elements in the |
1638 | 443 // TaskQueue, and does not include the size of the overflow stack. This |
444 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues. | |
6197 | 445 template<class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> |
446 class OverflowTaskQueue: public GenericTaskQueue<E, F, N> | |
1638 | 447 { |
448 public: | |
6197 | 449 typedef Stack<E, F> overflow_t; |
450 typedef GenericTaskQueue<E, F, N> taskqueue_t; | |
1638 | 451 |
1665 | 452 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;) |
453 | |
1638 | 454 // Push task t onto the queue or onto the overflow stack. Return true. |
455 inline bool push(E t); | |
456 | |
457 // Attempt to pop from the overflow stack; return true if anything was popped. | |
458 inline bool pop_overflow(E& t); | |
459 | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
460 inline overflow_t* overflow_stack() { return &_overflow_stack; } |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
461 |
1638 | 462 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
|
463 inline bool overflow_empty() const { return _overflow_stack.is_empty(); } |
1638 | 464 inline bool is_empty() const { |
465 return taskqueue_empty() && overflow_empty(); | |
466 } | |
467 | |
468 private: | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
469 overflow_t _overflow_stack; |
1638 | 470 }; |
471 | |
6197 | 472 template <class E, MEMFLAGS F, unsigned int N> |
473 bool OverflowTaskQueue<E, F, N>::push(E t) | |
1638 | 474 { |
475 if (!taskqueue_t::push(t)) { | |
476 overflow_stack()->push(t); | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
477 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size())); |
1638 | 478 } |
479 return true; | |
480 } | |
481 | |
6197 | 482 template <class E, MEMFLAGS F, unsigned int N> |
483 bool OverflowTaskQueue<E, F, N>::pop_overflow(E& t) | |
1638 | 484 { |
485 if (overflow_empty()) return false; | |
486 t = overflow_stack()->pop(); | |
487 return true; | |
488 } | |
489 | |
6197 | 490 class TaskQueueSetSuper { |
0 | 491 protected: |
492 static int randomParkAndMiller(int* seed0); | |
493 public: | |
494 // Returns "true" if some TaskQueue in the set contains a task. | |
495 virtual bool peek() = 0; | |
496 }; | |
497 | |
6197 | 498 template <MEMFLAGS F> class TaskQueueSetSuperImpl: public CHeapObj<F>, public TaskQueueSetSuper { |
499 }; | |
500 | |
501 template<class T, MEMFLAGS F> | |
502 class GenericTaskQueueSet: public TaskQueueSetSuperImpl<F> { | |
0 | 503 private: |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
504 uint _n; |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
505 T** _queues; |
0 | 506 |
507 public: | |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
508 typedef typename T::element_type E; |
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
509 |
0 | 510 GenericTaskQueueSet(int n) : _n(n) { |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
511 typedef T* GenericTaskQueuePtr; |
6197 | 512 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n, F); |
0 | 513 for (int i = 0; i < n; i++) { |
514 _queues[i] = NULL; | |
515 } | |
516 } | |
517 | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
518 bool steal_best_of_2(uint queue_num, int* seed, E& t); |
0 | 519 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
520 void register_queue(uint i, T* q); |
0 | 521 |
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
522 T* queue(uint n); |
0 | 523 |
1638 | 524 // The thread with queue number "queue_num" (and whose random number seed is |
525 // at "seed") is trying to steal a task from some other queue. (It may try | |
526 // several queues, according to some configuration parameter.) If some steal | |
527 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns | |
528 // false. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
529 bool steal(uint queue_num, int* seed, E& t); |
0 | 530 |
531 bool peek(); | |
532 }; | |
533 | |
6197 | 534 template<class T, MEMFLAGS F> void |
535 GenericTaskQueueSet<T, F>::register_queue(uint i, T* q) { | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
536 assert(i < _n, "index out of range."); |
0 | 537 _queues[i] = q; |
538 } | |
539 | |
6197 | 540 template<class T, MEMFLAGS F> T* |
541 GenericTaskQueueSet<T, F>::queue(uint i) { | |
0 | 542 return _queues[i]; |
543 } | |
544 | |
6197 | 545 template<class T, MEMFLAGS F> bool |
546 GenericTaskQueueSet<T, F>::steal(uint queue_num, int* seed, E& t) { | |
1665 | 547 for (uint i = 0; i < 2 * _n; i++) { |
548 if (steal_best_of_2(queue_num, seed, t)) { | |
549 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true)); | |
0 | 550 return true; |
1665 | 551 } |
552 } | |
553 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false)); | |
0 | 554 return false; |
555 } | |
556 | |
6197 | 557 template<class T, MEMFLAGS F> bool |
558 GenericTaskQueueSet<T, F>::steal_best_of_2(uint queue_num, int* seed, E& t) { | |
0 | 559 if (_n > 2) { |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
560 uint k1 = queue_num; |
6197 | 561 while (k1 == queue_num) k1 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n; |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
562 uint k2 = queue_num; |
6197 | 563 while (k2 == queue_num || k2 == k1) k2 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n; |
0 | 564 // Sample both and try the larger. |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
565 uint sz1 = _queues[k1]->size(); |
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
566 uint sz2 = _queues[k2]->size(); |
0 | 567 if (sz2 > sz1) return _queues[k2]->pop_global(t); |
568 else return _queues[k1]->pop_global(t); | |
569 } else if (_n == 2) { | |
570 // Just try the other one. | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
571 uint k = (queue_num + 1) % 2; |
0 | 572 return _queues[k]->pop_global(t); |
573 } else { | |
574 assert(_n == 1, "can't be zero."); | |
575 return false; | |
576 } | |
577 } | |
578 | |
6197 | 579 template<class T, MEMFLAGS F> |
580 bool GenericTaskQueueSet<T, F>::peek() { | |
0 | 581 // Try all the queues. |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
582 for (uint j = 0; j < _n; j++) { |
0 | 583 if (_queues[j]->peek()) |
584 return true; | |
585 } | |
586 return false; | |
587 } | |
588 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
589 // When to terminate from the termination protocol. |
6197 | 590 class TerminatorTerminator: public CHeapObj<mtInternal> { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
591 public: |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
592 virtual bool should_exit_termination() = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
593 }; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
594 |
0 | 595 // A class to aid in the termination of a set of parallel tasks using |
596 // TaskQueueSet's for work stealing. | |
597 | |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
598 #undef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
599 |
0 | 600 class ParallelTaskTerminator: public StackObj { |
601 private: | |
602 int _n_threads; | |
603 TaskQueueSetSuper* _queue_set; | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
604 int _offered_termination; |
0 | 605 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
606 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
607 static uint _total_yields; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
608 static uint _total_spins; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
609 static uint _total_peeks; |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
610 #endif |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
611 |
0 | 612 bool peek_in_queue_set(); |
613 protected: | |
614 virtual void yield(); | |
615 void sleep(uint millis); | |
616 | |
617 public: | |
618 | |
619 // "n_threads" is the number of threads to be terminated. "queue_set" is a | |
620 // queue sets of work queues of other threads. | |
621 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set); | |
622 | |
623 // The current thread has no work, and is ready to terminate if everyone | |
624 // else is. If returns "true", all threads are terminated. If returns | |
625 // "false", available work has been observed in one of the task queues, | |
626 // so the global task is not complete. | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
627 bool offer_termination() { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
628 return offer_termination(NULL); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
629 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
630 |
907 | 631 // 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
|
632 // method of the terminator parameter returns true. If terminator is |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
633 // NULL, then it is ignored. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
634 bool offer_termination(TerminatorTerminator* terminator); |
0 | 635 |
636 // Reset the terminator, so that it may be reused again. | |
637 // The caller is responsible for ensuring that this is done | |
638 // in an MT-safe manner, once the previous round of use of | |
639 // the terminator is finished. | |
640 void reset_for_reuse(); | |
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
641 // 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
|
642 // given number. |
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
643 void reset_for_reuse(int n_threads); |
0 | 644 |
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
645 #ifdef TRACESPINNING |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
646 static uint total_yields() { return _total_yields; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
647 static uint total_spins() { return _total_spins; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
648 static uint total_peeks() { return _total_peeks; } |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
649 static void print_termination_counts(); |
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
650 #endif |
0 | 651 }; |
652 | |
6197 | 653 template<class E, MEMFLAGS F, unsigned int N> inline bool |
654 GenericTaskQueue<E, F, N>::push(E t) { | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
655 uint localBot = _bottom; |
11997 | 656 assert(localBot < N, "_bottom out of range."); |
907 | 657 idx_t top = _age.top(); |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
658 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
|
659 assert(dirty_n_elems < N, "n_elems out of range."); |
0 | 660 if (dirty_n_elems < max_elems()) { |
10973
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
661 // g++ complains if the volatile result of the assignment is |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
662 // unused, so we cast the volatile away. We cannot cast directly |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
663 // to void, because gcc treats that as not using the result of the |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
664 // assignment. However, casting to E& means that we trigger an |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
665 // unused-value warning. So, we cast the E& to void. |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
666 (void) const_cast<E&>(_elems[localBot] = t); |
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
667 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
1665 | 668 TASKQUEUE_STATS_ONLY(stats.record_push()); |
0 | 669 return true; |
670 } else { | |
671 return push_slow(t, dirty_n_elems); | |
672 } | |
673 } | |
674 | |
6197 | 675 template<class E, MEMFLAGS F, unsigned int N> inline bool |
676 GenericTaskQueue<E, F, N>::pop_local(E& t) { | |
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
677 uint localBot = _bottom; |
907 | 678 // This value cannot be N-1. That can only occur as a result of |
0 | 679 // the assignment to bottom in this method. If it does, this method |
1638 | 680 // resets the size to 0 before the next call (which is sequential, |
0 | 681 // since this is pop_local.) |
907 | 682 uint dirty_n_elems = dirty_size(localBot, _age.top()); |
683 assert(dirty_n_elems != N - 1, "Shouldn't be possible..."); | |
0 | 684 if (dirty_n_elems == 0) return false; |
685 localBot = decrement_index(localBot); | |
686 _bottom = localBot; | |
687 // This is necessary to prevent any read below from being reordered | |
688 // before the store just above. | |
689 OrderAccess::fence(); | |
10973
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
690 // g++ complains if the volatile result of the assignment is |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
691 // unused, so we cast the volatile away. We cannot cast directly |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
692 // to void, because gcc treats that as not using the result of the |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
693 // assignment. However, casting to E& means that we trigger an |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
694 // unused-value warning. So, we cast the E& to void. |
ef57c43512d6
8014431: cleanup warnings indicated by the -Wunused-value compiler option on linux
ccheung
parents:
9073
diff
changeset
|
695 (void) 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 | |
6197 | 712 typedef GenericTaskQueue<oop, mtGC> OopTaskQueue; |
713 typedef GenericTaskQueueSet<OopTaskQueue, mtGC> 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 |
6197 | 793 typedef OverflowTaskQueue<StarTask, mtClass> OopStarTaskQueue; |
794 typedef GenericTaskQueueSet<OopStarTaskQueue, mtClass> OopStarTaskQueueSet; | |
0 | 795 |
6197 | 796 typedef OverflowTaskQueue<size_t, mtInternal> RegionTaskQueue; |
797 typedef GenericTaskQueueSet<RegionTaskQueue, mtClass> 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 |