Mercurial > hg > truffle
annotate src/share/vm/utilities/taskqueue.hpp @ 9126:bc26f978b0ce
HotSpotResolvedObjectType: implement hasFinalizeSubclass() correctly
don't use the (wrong) cached value, but ask the runtime on each request.
Fixes regression on xml.* benchmarks @ specjvm2008. The problem was:
After the constructor of Object was deoptimized due to an assumption violation,
it was recompiled again after some time. However, on recompilation, the value
of hasFinalizeSubclass for the class was not updated and it was compiled again
with a, now wrong, assumption, which then triggers deoptimization again.
This was repeated until it hit the recompilation limit (defined by
PerMethodRecompilationCutoff), and therefore only executed by the interpreter
from now on, causing the performance regression.
| author | Bernhard Urban <bernhard.urban@jku.at> |
|---|---|
| date | Mon, 15 Apr 2013 19:54:58 +0200 |
| parents | 83f27710f5f7 |
| children | ef57c43512d6 |
| 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 |
| 3960 | 56 #ifdef TARGET_OS_ARCH_bsd_x86 |
| 57 # include "orderAccess_bsd_x86.inline.hpp" | |
| 58 #endif | |
| 59 #ifdef TARGET_OS_ARCH_bsd_zero | |
| 60 # include "orderAccess_bsd_zero.inline.hpp" | |
| 61 #endif | |
| 1972 | 62 |
| 1665 | 63 // Simple TaskQueue stats that are collected by default in debug builds. |
| 64 | |
| 65 #if !defined(TASKQUEUE_STATS) && defined(ASSERT) | |
| 66 #define TASKQUEUE_STATS 1 | |
| 67 #elif !defined(TASKQUEUE_STATS) | |
| 68 #define TASKQUEUE_STATS 0 | |
| 69 #endif | |
| 70 | |
| 71 #if TASKQUEUE_STATS | |
| 72 #define TASKQUEUE_STATS_ONLY(code) code | |
| 73 #else | |
| 74 #define TASKQUEUE_STATS_ONLY(code) | |
| 75 #endif // TASKQUEUE_STATS | |
| 76 | |
| 77 #if TASKQUEUE_STATS | |
| 78 class TaskQueueStats { | |
| 79 public: | |
| 80 enum StatId { | |
| 81 push, // number of taskqueue pushes | |
| 82 pop, // number of taskqueue pops | |
| 83 pop_slow, // subset of taskqueue pops that were done slow-path | |
| 84 steal_attempt, // number of taskqueue steal attempts | |
| 85 steal, // number of taskqueue steals | |
| 86 overflow, // number of overflow pushes | |
| 87 overflow_max_len, // max length of overflow stack | |
| 88 last_stat_id | |
| 89 }; | |
| 90 | |
| 91 public: | |
| 92 inline TaskQueueStats() { reset(); } | |
| 93 | |
| 94 inline void record_push() { ++_stats[push]; } | |
| 95 inline void record_pop() { ++_stats[pop]; } | |
| 96 inline void record_pop_slow() { record_pop(); ++_stats[pop_slow]; } | |
| 97 inline void record_steal(bool success); | |
| 98 inline void record_overflow(size_t new_length); | |
| 99 | |
| 1709 | 100 TaskQueueStats & operator +=(const TaskQueueStats & addend); |
| 101 | |
| 1665 | 102 inline size_t get(StatId id) const { return _stats[id]; } |
| 103 inline const size_t* get() const { return _stats; } | |
| 104 | |
| 105 inline void reset(); | |
| 106 | |
| 1709 | 107 // Print the specified line of the header (does not include a line separator). |
| 1665 | 108 static void print_header(unsigned int line, outputStream* const stream = tty, |
| 109 unsigned int width = 10); | |
| 1709 | 110 // Print the statistics (does not include a line separator). |
| 1665 | 111 void print(outputStream* const stream = tty, unsigned int width = 10) const; |
| 112 | |
| 1709 | 113 DEBUG_ONLY(void verify() const;) |
| 114 | |
| 1665 | 115 private: |
| 116 size_t _stats[last_stat_id]; | |
| 117 static const char * const _names[last_stat_id]; | |
| 118 }; | |
| 119 | |
| 120 void TaskQueueStats::record_steal(bool success) { | |
| 121 ++_stats[steal_attempt]; | |
| 122 if (success) ++_stats[steal]; | |
| 123 } | |
| 124 | |
| 125 void TaskQueueStats::record_overflow(size_t new_len) { | |
| 126 ++_stats[overflow]; | |
| 127 if (new_len > _stats[overflow_max_len]) _stats[overflow_max_len] = new_len; | |
| 128 } | |
| 129 | |
| 130 void TaskQueueStats::reset() { | |
| 131 memset(_stats, 0, sizeof(_stats)); | |
| 132 } | |
| 133 #endif // TASKQUEUE_STATS | |
| 134 | |
| 6197 | 135 template <unsigned int N, MEMFLAGS F> |
| 136 class TaskQueueSuper: public CHeapObj<F> { | |
| 0 | 137 protected: |
| 907 | 138 // Internal type for indexing the queue; also used for the tag. |
| 139 typedef NOT_LP64(uint16_t) LP64_ONLY(uint32_t) idx_t; | |
| 140 | |
| 141 // 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
|
142 volatile uint _bottom; |
| 0 | 143 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
144 enum { MOD_N_MASK = N - 1 }; |
| 907 | 145 |
| 146 class Age { | |
| 147 public: | |
| 148 Age(size_t data = 0) { _data = data; } | |
| 149 Age(const Age& age) { _data = age._data; } | |
| 150 Age(idx_t top, idx_t tag) { _fields._top = top; _fields._tag = tag; } | |
| 0 | 151 |
| 907 | 152 Age get() const volatile { return _data; } |
| 153 void set(Age age) volatile { _data = age._data; } | |
| 154 | |
| 155 idx_t top() const volatile { return _fields._top; } | |
| 156 idx_t tag() const volatile { return _fields._tag; } | |
| 0 | 157 |
| 907 | 158 // Increment top; if it wraps, increment tag also. |
| 159 void increment() { | |
| 160 _fields._top = increment_index(_fields._top); | |
| 161 if (_fields._top == 0) ++_fields._tag; | |
| 162 } | |
| 0 | 163 |
| 907 | 164 Age cmpxchg(const Age new_age, const Age old_age) volatile { |
| 165 return (size_t) Atomic::cmpxchg_ptr((intptr_t)new_age._data, | |
| 166 (volatile intptr_t *)&_data, | |
| 167 (intptr_t)old_age._data); | |
| 168 } | |
| 169 | |
| 170 bool operator ==(const Age& other) const { return _data == other._data; } | |
| 0 | 171 |
| 907 | 172 private: |
| 173 struct fields { | |
| 174 idx_t _top; | |
| 175 idx_t _tag; | |
| 176 }; | |
| 177 union { | |
| 178 size_t _data; | |
| 179 fields _fields; | |
| 180 }; | |
| 0 | 181 }; |
| 907 | 182 |
| 183 volatile Age _age; | |
| 184 | |
| 185 // These both operate mod N. | |
| 186 static uint increment_index(uint ind) { | |
| 187 return (ind + 1) & MOD_N_MASK; | |
| 0 | 188 } |
| 907 | 189 static uint decrement_index(uint ind) { |
| 190 return (ind - 1) & MOD_N_MASK; | |
| 0 | 191 } |
| 192 | |
| 907 | 193 // Returns a number in the range [0..N). If the result is "N-1", it should be |
| 194 // interpreted as 0. | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
195 uint dirty_size(uint bot, uint top) const { |
| 907 | 196 return (bot - top) & MOD_N_MASK; |
| 0 | 197 } |
| 198 | |
| 199 // 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
|
200 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
|
201 uint sz = dirty_size(bot, top); |
| 907 | 202 // Has the queue "wrapped", so that bottom is less than top? There's a |
| 203 // complicated special case here. A pair of threads could perform pop_local | |
| 204 // and pop_global operations concurrently, starting from a state in which | |
| 205 // _bottom == _top+1. The pop_local could succeed in decrementing _bottom, | |
| 206 // and the pop_global in incrementing _top (in which case the pop_global | |
| 207 // will be awarded the contested queue element.) The resulting state must | |
| 208 // be interpreted as an empty queue. (We only need to worry about one such | |
| 209 // event: only the queue owner performs pop_local's, and several concurrent | |
| 210 // threads attempting to perform the pop_global will all perform the same | |
| 211 // CAS, and only one can succeed.) Any stealing thread that reads after | |
| 212 // either the increment or decrement will see an empty queue, and will not | |
| 213 // join the competitors. The "sz == -1 || sz == N-1" state will not be | |
| 214 // modified by concurrent queues, so the owner thread can reset the state to | |
| 215 // _bottom == top so subsequent pushes will be performed normally. | |
| 216 return (sz == N - 1) ? 0 : sz; | |
| 0 | 217 } |
| 218 | |
| 219 public: | |
| 220 TaskQueueSuper() : _bottom(0), _age() {} | |
| 221 | |
| 1638 | 222 // Return true if the TaskQueue contains/does not contain any tasks. |
| 223 bool peek() const { return _bottom != _age.top(); } | |
| 224 bool is_empty() const { return size() == 0; } | |
| 0 | 225 |
| 226 // Return an estimate of the number of elements in the queue. | |
| 227 // The "careful" version admits the possibility of pop_local/pop_global | |
| 228 // races. | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
229 uint size() const { |
| 907 | 230 return size(_bottom, _age.top()); |
| 0 | 231 } |
| 232 | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
233 uint dirty_size() const { |
| 907 | 234 return dirty_size(_bottom, _age.top()); |
| 0 | 235 } |
| 236 | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
237 void set_empty() { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
238 _bottom = 0; |
| 907 | 239 _age.set(0); |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
240 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
241 |
| 0 | 242 // Maximum number of elements allowed in the queue. This is two less |
| 243 // 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
|
244 uint max_elems() const { return N - 2; } |
| 1284 | 245 |
| 246 // Total size of queue. | |
| 247 static const uint total_size() { return N; } | |
| 1665 | 248 |
| 249 TASKQUEUE_STATS_ONLY(TaskQueueStats stats;) | |
| 0 | 250 }; |
| 251 | |
| 6197 | 252 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
253 |
| 6197 | 254 template <class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> |
| 255 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
|
256 ArrayAllocator<E, F> _array_allocator; |
| 6197 | 257 protected: |
| 258 typedef typename TaskQueueSuper<N, F>::Age Age; | |
| 259 typedef typename TaskQueueSuper<N, F>::idx_t idx_t; | |
| 260 | |
| 261 using TaskQueueSuper<N, F>::_bottom; | |
| 262 using TaskQueueSuper<N, F>::_age; | |
| 263 using TaskQueueSuper<N, F>::increment_index; | |
| 264 using TaskQueueSuper<N, F>::decrement_index; | |
| 265 using TaskQueueSuper<N, F>::dirty_size; | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
266 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
267 public: |
| 6197 | 268 using TaskQueueSuper<N, F>::max_elems; |
| 269 using TaskQueueSuper<N, F>::size; | |
| 270 | |
| 271 #if TASKQUEUE_STATS | |
| 272 using TaskQueueSuper<N, F>::stats; | |
| 273 #endif | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
274 |
| 0 | 275 private: |
| 276 // 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
|
277 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
|
278 bool pop_local_slow(uint localBot, Age oldAge); |
| 0 | 279 |
| 280 public: | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
281 typedef E element_type; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
282 |
| 0 | 283 // Initializes the queue to empty. |
| 284 GenericTaskQueue(); | |
| 285 | |
| 286 void initialize(); | |
| 287 | |
| 1638 | 288 // Push the task "t" on the queue. Returns "false" iff the queue is full. |
| 0 | 289 inline bool push(E t); |
| 290 | |
| 1638 | 291 // Attempts to claim a task from the "local" end of the queue (the most |
| 292 // recently pushed). If successful, returns true and sets t to the task; | |
| 293 // otherwise, returns false (the queue is empty). | |
| 0 | 294 inline bool pop_local(E& t); |
| 295 | |
| 1638 | 296 // Like pop_local(), but uses the "global" end of the queue (the least |
| 297 // recently pushed). | |
| 0 | 298 bool pop_global(E& t); |
| 299 | |
| 300 // Delete any resource associated with the queue. | |
| 301 ~GenericTaskQueue(); | |
| 302 | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
303 // apply the closure to all elements in the task queue |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
304 void oops_do(OopClosure* f); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
305 |
| 0 | 306 private: |
| 307 // Element array. | |
| 308 volatile E* _elems; | |
| 309 }; | |
| 310 | |
| 6197 | 311 template<class E, MEMFLAGS F, unsigned int N> |
| 312 GenericTaskQueue<E, F, N>::GenericTaskQueue() { | |
| 907 | 313 assert(sizeof(Age) == sizeof(size_t), "Depends on this."); |
| 0 | 314 } |
| 315 | |
| 6197 | 316 template<class E, MEMFLAGS F, unsigned int N> |
| 317 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
|
318 _elems = _array_allocator.allocate(N); |
| 0 | 319 } |
| 320 | |
| 6197 | 321 template<class E, MEMFLAGS F, unsigned int N> |
| 322 void GenericTaskQueue<E, F, N>::oops_do(OopClosure* f) { | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
323 // 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
|
324 uint iters = size(); |
|
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
325 uint index = _bottom; |
|
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
326 for (uint i = 0; i < iters; ++i) { |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
327 index = decrement_index(index); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
328 // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
329 // index, &_elems[index], _elems[index]); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
330 E* t = (E*)&_elems[index]; // cast away volatility |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
331 oop* p = (oop*)t; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
332 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
|
333 f->do_oop(p); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
334 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
335 // tty->print_cr("END OopTaskQueue::oops_do"); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
336 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
337 |
| 6197 | 338 template<class E, MEMFLAGS F, unsigned int N> |
| 339 bool GenericTaskQueue<E, F, N>::push_slow(E t, uint dirty_n_elems) { | |
| 907 | 340 if (dirty_n_elems == N - 1) { |
| 0 | 341 // 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
|
342 uint localBot = _bottom; |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
343 // 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
|
344 const_cast<E&>(_elems[localBot] = t); |
|
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
345 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
| 1665 | 346 TASKQUEUE_STATS_ONLY(stats.record_push()); |
| 0 | 347 return true; |
| 907 | 348 } |
| 349 return false; | |
| 0 | 350 } |
| 351 | |
|
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
352 // 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
|
353 // 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
|
354 // 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
|
355 // 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
|
356 // 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
|
357 // and pop_global() succeeds, see pop_global()). |
| 6197 | 358 template<class E, MEMFLAGS F, unsigned int N> |
| 359 bool GenericTaskQueue<E, F, N>::pop_local_slow(uint localBot, Age oldAge) { | |
| 0 | 360 // This queue was observed to contain exactly one element; either this |
| 361 // thread will claim it, or a competing "pop_global". In either case, | |
| 362 // the queue will be logically empty afterwards. Create a new Age value | |
| 363 // that represents the empty queue for the given value of "_bottom". (We | |
| 364 // must also increment "tag" because of the case where "bottom == 1", | |
| 365 // "top == 0". A pop_global could read the queue element in that case, | |
| 366 // then have the owner thread do a pop followed by another push. Without | |
| 367 // the incrementing of "tag", the pop_global's CAS could succeed, | |
| 368 // allowing it to believe it has claimed the stale element.) | |
| 907 | 369 Age newAge((idx_t)localBot, oldAge.tag() + 1); |
| 0 | 370 // Perhaps a competing pop_global has already incremented "top", in which |
| 371 // case it wins the element. | |
| 372 if (localBot == oldAge.top()) { | |
| 373 // No competing pop_global has yet incremented "top"; we'll try to | |
| 374 // install new_age, thus claiming the element. | |
| 907 | 375 Age tempAge = _age.cmpxchg(newAge, oldAge); |
| 0 | 376 if (tempAge == oldAge) { |
| 377 // We win. | |
| 907 | 378 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); |
| 1665 | 379 TASKQUEUE_STATS_ONLY(stats.record_pop_slow()); |
| 0 | 380 return true; |
| 381 } | |
| 382 } | |
| 907 | 383 // We lose; a completing pop_global gets the element. But the queue is empty |
| 384 // and top is greater than bottom. Fix this representation of the empty queue | |
| 385 // to become the canonical one. | |
| 386 _age.set(newAge); | |
| 387 assert(dirty_size(localBot, _age.top()) != N - 1, "sanity"); | |
| 0 | 388 return false; |
| 389 } | |
| 390 | |
| 6197 | 391 template<class E, MEMFLAGS F, unsigned int N> |
| 392 bool GenericTaskQueue<E, F, N>::pop_global(E& t) { | |
| 907 | 393 Age oldAge = _age.get(); |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
394 uint localBot = _bottom; |
|
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
395 uint n_elems = size(localBot, oldAge.top()); |
| 0 | 396 if (n_elems == 0) { |
| 397 return false; | |
| 398 } | |
| 907 | 399 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
400 const_cast<E&>(t = _elems[oldAge.top()]); |
| 907 | 401 Age newAge(oldAge); |
| 402 newAge.increment(); | |
| 403 Age resAge = _age.cmpxchg(newAge, oldAge); | |
| 404 | |
| 0 | 405 // Note that using "_bottom" here might fail, since a pop_local might |
| 406 // have decremented it. | |
| 907 | 407 assert(dirty_size(localBot, newAge.top()) != N - 1, "sanity"); |
| 408 return resAge == oldAge; | |
| 0 | 409 } |
| 410 | |
| 6197 | 411 template<class E, MEMFLAGS F, unsigned int N> |
| 412 GenericTaskQueue<E, F, N>::~GenericTaskQueue() { | |
| 413 FREE_C_HEAP_ARRAY(E, _elems, F); | |
| 0 | 414 } |
| 415 | |
| 1638 | 416 // OverflowTaskQueue is a TaskQueue that also includes an overflow stack for |
| 417 // elements that do not fit in the TaskQueue. | |
| 418 // | |
|
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
419 // This class hides two methods from super classes: |
| 1638 | 420 // |
| 421 // push() - push onto the task queue or, if that fails, onto the overflow stack | |
| 422 // is_empty() - return true if both the TaskQueue and overflow stack are empty | |
| 423 // | |
|
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
424 // Note that size() is not hidden--it returns the number of elements in the |
| 1638 | 425 // TaskQueue, and does not include the size of the overflow stack. This |
| 426 // simplifies replacement of GenericTaskQueues with OverflowTaskQueues. | |
| 6197 | 427 template<class E, MEMFLAGS F, unsigned int N = TASKQUEUE_SIZE> |
| 428 class OverflowTaskQueue: public GenericTaskQueue<E, F, N> | |
| 1638 | 429 { |
| 430 public: | |
| 6197 | 431 typedef Stack<E, F> overflow_t; |
| 432 typedef GenericTaskQueue<E, F, N> taskqueue_t; | |
| 1638 | 433 |
| 1665 | 434 TASKQUEUE_STATS_ONLY(using taskqueue_t::stats;) |
| 435 | |
| 1638 | 436 // Push task t onto the queue or onto the overflow stack. Return true. |
| 437 inline bool push(E t); | |
| 438 | |
| 439 // Attempt to pop from the overflow stack; return true if anything was popped. | |
| 440 inline bool pop_overflow(E& t); | |
| 441 | |
|
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
442 inline overflow_t* overflow_stack() { return &_overflow_stack; } |
|
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
443 |
| 1638 | 444 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
|
445 inline bool overflow_empty() const { return _overflow_stack.is_empty(); } |
| 1638 | 446 inline bool is_empty() const { |
| 447 return taskqueue_empty() && overflow_empty(); | |
| 448 } | |
| 449 | |
| 450 private: | |
|
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
451 overflow_t _overflow_stack; |
| 1638 | 452 }; |
| 453 | |
| 6197 | 454 template <class E, MEMFLAGS F, unsigned int N> |
| 455 bool OverflowTaskQueue<E, F, N>::push(E t) | |
| 1638 | 456 { |
| 457 if (!taskqueue_t::push(t)) { | |
| 458 overflow_stack()->push(t); | |
|
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
459 TASKQUEUE_STATS_ONLY(stats.record_overflow(overflow_stack()->size())); |
| 1638 | 460 } |
| 461 return true; | |
| 462 } | |
| 463 | |
| 6197 | 464 template <class E, MEMFLAGS F, unsigned int N> |
| 465 bool OverflowTaskQueue<E, F, N>::pop_overflow(E& t) | |
| 1638 | 466 { |
| 467 if (overflow_empty()) return false; | |
| 468 t = overflow_stack()->pop(); | |
| 469 return true; | |
| 470 } | |
| 471 | |
| 6197 | 472 class TaskQueueSetSuper { |
| 0 | 473 protected: |
| 474 static int randomParkAndMiller(int* seed0); | |
| 475 public: | |
| 476 // Returns "true" if some TaskQueue in the set contains a task. | |
| 477 virtual bool peek() = 0; | |
| 478 }; | |
| 479 | |
| 6197 | 480 template <MEMFLAGS F> class TaskQueueSetSuperImpl: public CHeapObj<F>, public TaskQueueSetSuper { |
| 481 }; | |
| 482 | |
| 483 template<class T, MEMFLAGS F> | |
| 484 class GenericTaskQueueSet: public TaskQueueSetSuperImpl<F> { | |
| 0 | 485 private: |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
486 uint _n; |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
487 T** _queues; |
| 0 | 488 |
| 489 public: | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
490 typedef typename T::element_type E; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
491 |
| 0 | 492 GenericTaskQueueSet(int n) : _n(n) { |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
493 typedef T* GenericTaskQueuePtr; |
| 6197 | 494 _queues = NEW_C_HEAP_ARRAY(GenericTaskQueuePtr, n, F); |
| 0 | 495 for (int i = 0; i < n; i++) { |
| 496 _queues[i] = NULL; | |
| 497 } | |
| 498 } | |
| 499 | |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
500 bool steal_best_of_2(uint queue_num, int* seed, E& t); |
| 0 | 501 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
502 void register_queue(uint i, T* q); |
| 0 | 503 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
504 T* queue(uint n); |
| 0 | 505 |
| 1638 | 506 // The thread with queue number "queue_num" (and whose random number seed is |
| 507 // at "seed") is trying to steal a task from some other queue. (It may try | |
| 508 // several queues, according to some configuration parameter.) If some steal | |
| 509 // succeeds, returns "true" and sets "t" to the stolen task, otherwise returns | |
| 510 // false. | |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
511 bool steal(uint queue_num, int* seed, E& t); |
| 0 | 512 |
| 513 bool peek(); | |
| 514 }; | |
| 515 | |
| 6197 | 516 template<class T, MEMFLAGS F> void |
| 517 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
|
518 assert(i < _n, "index out of range."); |
| 0 | 519 _queues[i] = q; |
| 520 } | |
| 521 | |
| 6197 | 522 template<class T, MEMFLAGS F> T* |
| 523 GenericTaskQueueSet<T, F>::queue(uint i) { | |
| 0 | 524 return _queues[i]; |
| 525 } | |
| 526 | |
| 6197 | 527 template<class T, MEMFLAGS F> bool |
| 528 GenericTaskQueueSet<T, F>::steal(uint queue_num, int* seed, E& t) { | |
| 1665 | 529 for (uint i = 0; i < 2 * _n; i++) { |
| 530 if (steal_best_of_2(queue_num, seed, t)) { | |
| 531 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(true)); | |
| 0 | 532 return true; |
| 1665 | 533 } |
| 534 } | |
| 535 TASKQUEUE_STATS_ONLY(queue(queue_num)->stats.record_steal(false)); | |
| 0 | 536 return false; |
| 537 } | |
| 538 | |
| 6197 | 539 template<class T, MEMFLAGS F> bool |
| 540 GenericTaskQueueSet<T, F>::steal_best_of_2(uint queue_num, int* seed, E& t) { | |
| 0 | 541 if (_n > 2) { |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
542 uint k1 = queue_num; |
| 6197 | 543 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
|
544 uint k2 = queue_num; |
| 6197 | 545 while (k2 == queue_num || k2 == k1) k2 = TaskQueueSetSuper::randomParkAndMiller(seed) % _n; |
| 0 | 546 // Sample both and try the larger. |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
547 uint sz1 = _queues[k1]->size(); |
|
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
548 uint sz2 = _queues[k2]->size(); |
| 0 | 549 if (sz2 > sz1) return _queues[k2]->pop_global(t); |
| 550 else return _queues[k1]->pop_global(t); | |
| 551 } else if (_n == 2) { | |
| 552 // Just try the other one. | |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
553 uint k = (queue_num + 1) % 2; |
| 0 | 554 return _queues[k]->pop_global(t); |
| 555 } else { | |
| 556 assert(_n == 1, "can't be zero."); | |
| 557 return false; | |
| 558 } | |
| 559 } | |
| 560 | |
| 6197 | 561 template<class T, MEMFLAGS F> |
| 562 bool GenericTaskQueueSet<T, F>::peek() { | |
| 0 | 563 // Try all the queues. |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
564 for (uint j = 0; j < _n; j++) { |
| 0 | 565 if (_queues[j]->peek()) |
| 566 return true; | |
| 567 } | |
| 568 return false; | |
| 569 } | |
| 570 | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
571 // When to terminate from the termination protocol. |
| 6197 | 572 class TerminatorTerminator: public CHeapObj<mtInternal> { |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
573 public: |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
574 virtual bool should_exit_termination() = 0; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
575 }; |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
576 |
| 0 | 577 // A class to aid in the termination of a set of parallel tasks using |
| 578 // TaskQueueSet's for work stealing. | |
| 579 | |
|
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
580 #undef TRACESPINNING |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
581 |
| 0 | 582 class ParallelTaskTerminator: public StackObj { |
| 583 private: | |
| 584 int _n_threads; | |
| 585 TaskQueueSetSuper* _queue_set; | |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
586 int _offered_termination; |
| 0 | 587 |
|
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
588 #ifdef TRACESPINNING |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
589 static uint _total_yields; |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
590 static uint _total_spins; |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
591 static uint _total_peeks; |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
592 #endif |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
593 |
| 0 | 594 bool peek_in_queue_set(); |
| 595 protected: | |
| 596 virtual void yield(); | |
| 597 void sleep(uint millis); | |
| 598 | |
| 599 public: | |
| 600 | |
| 601 // "n_threads" is the number of threads to be terminated. "queue_set" is a | |
| 602 // queue sets of work queues of other threads. | |
| 603 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set); | |
| 604 | |
| 605 // The current thread has no work, and is ready to terminate if everyone | |
| 606 // else is. If returns "true", all threads are terminated. If returns | |
| 607 // "false", available work has been observed in one of the task queues, | |
| 608 // so the global task is not complete. | |
|
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
609 bool offer_termination() { |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
610 return offer_termination(NULL); |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
611 } |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
612 |
| 907 | 613 // 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
|
614 // method of the terminator parameter returns true. If terminator is |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
615 // NULL, then it is ignored. |
|
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
616 bool offer_termination(TerminatorTerminator* terminator); |
| 0 | 617 |
| 618 // Reset the terminator, so that it may be reused again. | |
| 619 // The caller is responsible for ensuring that this is done | |
| 620 // in an MT-safe manner, once the previous round of use of | |
| 621 // the terminator is finished. | |
| 622 void reset_for_reuse(); | |
|
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
623 // 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
|
624 // given number. |
|
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
625 void reset_for_reuse(int n_threads); |
| 0 | 626 |
|
546
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
627 #ifdef TRACESPINNING |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
628 static uint total_yields() { return _total_yields; } |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
629 static uint total_spins() { return _total_spins; } |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
630 static uint total_peeks() { return _total_peeks; } |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
631 static void print_termination_counts(); |
|
05c6d52fa7a9
6690928: Use spinning in combination with yields for workstealing termination.
jmasa
parents:
541
diff
changeset
|
632 #endif |
| 0 | 633 }; |
| 634 | |
| 6197 | 635 template<class E, MEMFLAGS F, unsigned int N> inline bool |
| 636 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
|
637 uint localBot = _bottom; |
| 907 | 638 assert((localBot >= 0) && (localBot < N), "_bottom out of range."); |
| 639 idx_t top = _age.top(); | |
|
541
23673011938d
6787254: Work queue capacity can be increased substantially on some platforms
ysr
parents:
375
diff
changeset
|
640 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
|
641 assert(dirty_n_elems < N, "n_elems out of range."); |
| 0 | 642 if (dirty_n_elems < max_elems()) { |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
643 // 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
|
644 const_cast<E&>(_elems[localBot] = t); |
|
1024
2c03ce058f55
6888847: TaskQueue needs release_store() for correctness on RMO machines
bobv
parents:
907
diff
changeset
|
645 OrderAccess::release_store(&_bottom, increment_index(localBot)); |
| 1665 | 646 TASKQUEUE_STATS_ONLY(stats.record_push()); |
| 0 | 647 return true; |
| 648 } else { | |
| 649 return push_slow(t, dirty_n_elems); | |
| 650 } | |
| 651 } | |
| 652 | |
| 6197 | 653 template<class E, MEMFLAGS F, unsigned int N> inline bool |
| 654 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
|
655 uint localBot = _bottom; |
| 907 | 656 // This value cannot be N-1. That can only occur as a result of |
| 0 | 657 // the assignment to bottom in this method. If it does, this method |
| 1638 | 658 // resets the size to 0 before the next call (which is sequential, |
| 0 | 659 // since this is pop_local.) |
| 907 | 660 uint dirty_n_elems = dirty_size(localBot, _age.top()); |
| 661 assert(dirty_n_elems != N - 1, "Shouldn't be possible..."); | |
| 0 | 662 if (dirty_n_elems == 0) return false; |
| 663 localBot = decrement_index(localBot); | |
| 664 _bottom = localBot; | |
| 665 // This is necessary to prevent any read below from being reordered | |
| 666 // before the store just above. | |
| 667 OrderAccess::fence(); | |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
668 const_cast<E&>(t = _elems[localBot]); |
| 0 | 669 // This is a second read of "age"; the "size()" above is the first. |
| 670 // If there's still at least one element in the queue, based on the | |
| 671 // "_bottom" and "age" we've read, then there can be no interference with | |
| 672 // a "pop_global" operation, and we're done. | |
| 907 | 673 idx_t tp = _age.top(); // XXX |
| 0 | 674 if (size(localBot, tp) > 0) { |
| 907 | 675 assert(dirty_size(localBot, tp) != N - 1, "sanity"); |
| 1665 | 676 TASKQUEUE_STATS_ONLY(stats.record_pop()); |
| 0 | 677 return true; |
| 678 } else { | |
| 679 // Otherwise, the queue contained exactly one element; we take the slow | |
| 680 // path. | |
| 907 | 681 return pop_local_slow(localBot, _age.get()); |
| 0 | 682 } |
| 683 } | |
| 684 | |
| 6197 | 685 typedef GenericTaskQueue<oop, mtGC> OopTaskQueue; |
| 686 typedef GenericTaskQueueSet<OopTaskQueue, mtGC> OopTaskQueueSet; | |
| 0 | 687 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
688 #ifdef _MSC_VER |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
689 #pragma warning(push) |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
690 // warning C4522: multiple assignment operators specified |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
691 #pragma warning(disable:4522) |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
692 #endif |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
693 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
694 // 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
|
695 // 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
|
696 // 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
|
697 class StarTask { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
698 void* _holder; // either union oop* or narrowOop* |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
699 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
700 enum { COMPRESSED_OOP_MASK = 1 }; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
701 |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
702 public: |
|
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
703 StarTask(narrowOop* p) { |
|
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
704 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
|
705 _holder = (void *)((uintptr_t)p | COMPRESSED_OOP_MASK); |
|
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
706 } |
|
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
707 StarTask(oop* p) { |
|
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
708 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
|
709 _holder = (void*)p; |
|
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
579
diff
changeset
|
710 } |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
711 StarTask() { _holder = NULL; } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
712 operator oop*() { return (oop*)_holder; } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
713 operator narrowOop*() { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
714 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
|
715 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
716 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
717 StarTask& operator=(const StarTask& t) { |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
718 _holder = t._holder; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
719 return *this; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
720 } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
721 volatile StarTask& operator=(const volatile StarTask& t) volatile { |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
722 _holder = t._holder; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
723 return *this; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
724 } |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
725 |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
726 bool is_narrow() const { |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
727 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
|
728 } |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
729 }; |
|
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
730 |
|
1311
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
731 class ObjArrayTask |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
732 { |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
733 public: |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
734 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
|
735 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
|
736 assert(idx <= size_t(max_jint), "too big"); |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
737 } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
738 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
|
739 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
740 ObjArrayTask& operator =(const ObjArrayTask& t) { |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
741 _obj = t._obj; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
742 _index = t._index; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
743 return *this; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
744 } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
745 volatile ObjArrayTask& |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
746 operator =(const volatile ObjArrayTask& t) volatile { |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
747 _obj = t._obj; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
748 _index = t._index; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
749 return *this; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
750 } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
751 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
752 inline oop obj() const { return _obj; } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
753 inline int index() const { return _index; } |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
754 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
755 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
|
756 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
757 private: |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
758 oop _obj; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
759 int _index; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
760 }; |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
761 |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
762 #ifdef _MSC_VER |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
763 #pragma warning(pop) |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
764 #endif |
|
2a1472c30599
4396719: Mark Sweep stack overflow on deeply nested Object arrays
jcoomes
parents:
1284
diff
changeset
|
765 |
| 6197 | 766 typedef OverflowTaskQueue<StarTask, mtClass> OopStarTaskQueue; |
| 767 typedef GenericTaskQueueSet<OopStarTaskQueue, mtClass> OopStarTaskQueueSet; | |
| 0 | 768 |
| 6197 | 769 typedef OverflowTaskQueue<size_t, mtInternal> RegionTaskQueue; |
| 770 typedef GenericTaskQueueSet<RegionTaskQueue, mtClass> RegionTaskQueueSet; | |
|
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1709
diff
changeset
|
771 |
| 1972 | 772 |
| 773 #endif // SHARE_VM_UTILITIES_TASKQUEUE_HPP |