1 /* 2 * Copyright (c) 2001, 2015, Oracle and/or its affiliates. All rights reserved. 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 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "gc/shared/taskqueue.hpp" 27 #include "oops/oop.inline.hpp" 28 #include "runtime/atomic.inline.hpp" 29 #include "runtime/os.hpp" 30 #include "runtime/thread.inline.hpp" 31 #include "utilities/debug.hpp" 32 #include "utilities/stack.inline.hpp" 33 34 #ifdef TRACESPINNING 35 uint ParallelTaskTerminator::_total_yields = 0; 36 uint ParallelTaskTerminator::_total_spins = 0; 37 uint ParallelTaskTerminator::_total_peeks = 0; 38 #endif 39 40 #if TASKQUEUE_STATS 41 const char * const TaskQueueStats::_names[last_stat_id] = { 42 "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax" 43 }; 44 45 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend) 46 { 47 for (unsigned int i = 0; i < last_stat_id; ++i) { 48 _stats[i] += addend._stats[i]; 49 } 50 return *this; 51 } 52 53 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream, 54 unsigned int width) 55 { 56 // Use a width w: 1 <= w <= max_width 57 const unsigned int max_width = 40; 58 const unsigned int w = MAX2(MIN2(width, max_width), 1U); 59 60 if (line == 0) { // spaces equal in width to the header 61 const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1; 62 stream->print("%*s", hdr_width, " "); 63 } else if (line == 1) { // labels 64 stream->print("%*s", w, _names[0]); 65 for (unsigned int i = 1; i < last_stat_id; ++i) { 66 stream->print(" %*s", w, _names[i]); 67 } 68 } else if (line == 2) { // dashed lines 69 char dashes[max_width + 1]; 70 memset(dashes, '-', w); 71 dashes[w] = '\0'; 72 stream->print("%s", dashes); 73 for (unsigned int i = 1; i < last_stat_id; ++i) { 74 stream->print(" %s", dashes); 75 } 76 } 77 } 78 79 void TaskQueueStats::print(outputStream* stream, unsigned int width) const 80 { 81 #define FMT SIZE_FORMAT_W(*) 82 stream->print(FMT, width, _stats[0]); 83 for (unsigned int i = 1; i < last_stat_id; ++i) { 84 stream->print(" " FMT, width, _stats[i]); 85 } 86 #undef FMT 87 } 88 89 #ifdef ASSERT 90 // Invariants which should hold after a TaskQueue has been emptied and is 91 // quiescent; they do not hold at arbitrary times. 92 void TaskQueueStats::verify() const 93 { 94 assert(get(push) == get(pop) + get(steal), 95 err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT, 96 get(push), get(pop), get(steal))); 97 assert(get(pop_slow) <= get(pop), 98 err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT, 99 get(pop_slow), get(pop))); 100 assert(get(steal) <= get(steal_attempt), 101 err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT, 102 get(steal), get(steal_attempt))); 103 assert(get(overflow) == 0 || get(push) != 0, 104 err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT, 105 get(overflow), get(push))); 106 assert(get(overflow_max_len) == 0 || get(overflow) != 0, 107 err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT, 108 get(overflow_max_len), get(overflow))); 109 } 110 #endif // ASSERT 111 #endif // TASKQUEUE_STATS 112 113 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) { 114 const int a = 16807; 115 const int m = 2147483647; 116 const int q = 127773; /* m div a */ 117 const int r = 2836; /* m mod a */ 118 assert(sizeof(int) == 4, "I think this relies on that"); 119 int seed = *seed0; 120 int hi = seed / q; 121 int lo = seed % q; 122 int test = a * lo - r * hi; 123 if (test > 0) 124 seed = test; 125 else 126 seed = test + m; 127 *seed0 = seed; 128 return seed; 129 } 130 131 ParallelTaskTerminator:: 132 ParallelTaskTerminator(uint n_threads, TaskQueueSetSuper* queue_set) : 133 _n_threads(n_threads), 134 _queue_set(queue_set), 135 _offered_termination(0) {} 136 137 bool ParallelTaskTerminator::peek_in_queue_set() { 138 return _queue_set->peek(); 139 } 140 141 void ParallelTaskTerminator::yield() { 142 assert(_offered_termination <= _n_threads, "Invariant"); 143 os::naked_yield(); 144 } 145 146 void ParallelTaskTerminator::sleep(uint millis) { 147 assert(_offered_termination <= _n_threads, "Invariant"); 148 os::sleep(Thread::current(), millis, false); 149 } 150 151 bool 152 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) { 153 assert(_n_threads > 0, "Initialization is incorrect"); 154 assert(_offered_termination < _n_threads, "Invariant"); 155 Atomic::inc((int *)&_offered_termination); 156 157 uint yield_count = 0; 158 // Number of hard spin loops done since last yield 159 uint hard_spin_count = 0; 160 // Number of iterations in the hard spin loop. 161 uint hard_spin_limit = WorkStealingHardSpins; 162 163 // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done. 164 // If it is greater than 0, then start with a small number 165 // of spins and increase number with each turn at spinning until 166 // the count of hard spins exceeds WorkStealingSpinToYieldRatio. 167 // Then do a yield() call and start spinning afresh. 168 if (WorkStealingSpinToYieldRatio > 0) { 169 hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio; 170 hard_spin_limit = MAX2(hard_spin_limit, 1U); 171 } 172 // Remember the initial spin limit. 173 uint hard_spin_start = hard_spin_limit; 174 175 // Loop waiting for all threads to offer termination or 176 // more work. 177 while (true) { 178 assert(_offered_termination <= _n_threads, "Invariant"); 179 // Are all threads offering termination? 180 if (_offered_termination == _n_threads) { 181 return true; 182 } else { 183 // Look for more work. 184 // Periodically sleep() instead of yield() to give threads 185 // waiting on the cores the chance to grab this code 186 if (yield_count <= WorkStealingYieldsBeforeSleep) { 187 // Do a yield or hardspin. For purposes of deciding whether 188 // to sleep, count this as a yield. 189 yield_count++; 190 191 // Periodically call yield() instead spinning 192 // After WorkStealingSpinToYieldRatio spins, do a yield() call 193 // and reset the counts and starting limit. 194 if (hard_spin_count > WorkStealingSpinToYieldRatio) { 195 yield(); 196 hard_spin_count = 0; 197 hard_spin_limit = hard_spin_start; 198 #ifdef TRACESPINNING 199 _total_yields++; 200 #endif 201 } else { 202 // Hard spin this time 203 // Increase the hard spinning period but only up to a limit. 204 hard_spin_limit = MIN2(2*hard_spin_limit, 205 (uint) WorkStealingHardSpins); 206 for (uint j = 0; j < hard_spin_limit; j++) { 207 SpinPause(); 208 } 209 hard_spin_count++; 210 #ifdef TRACESPINNING 211 _total_spins++; 212 #endif 213 } 214 } else { 215 if (PrintGCDetails && Verbose) { 216 gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() " 217 "thread " PTR_FORMAT " sleeps after %u yields", 218 p2i(Thread::current()), yield_count); 219 } 220 yield_count = 0; 221 // A sleep will cause this processor to seek work on another processor's 222 // runqueue, if it has nothing else to run (as opposed to the yield 223 // which may only move the thread to the end of the this processor's 224 // runqueue). 225 sleep(WorkStealingSleepMillis); 226 } 227 228 #ifdef TRACESPINNING 229 _total_peeks++; 230 #endif 231 if (peek_in_queue_set() || 232 (terminator != NULL && terminator->should_exit_termination())) { 233 Atomic::dec((int *)&_offered_termination); 234 assert(_offered_termination < _n_threads, "Invariant"); 235 return false; 236 } 237 } 238 } 239 } 240 241 #ifdef TRACESPINNING 242 void ParallelTaskTerminator::print_termination_counts() { 243 gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %u" 244 " Total spins: %u Total peeks: %u", 245 total_yields(), 246 total_spins(), 247 total_peeks()); 248 } 249 #endif 250 251 void ParallelTaskTerminator::reset_for_reuse() { 252 if (_offered_termination != 0) { 253 assert(_offered_termination == _n_threads, 254 "Terminator may still be in use"); 255 _offered_termination = 0; 256 } 257 } 258 259 #ifdef ASSERT 260 bool ObjArrayTask::is_valid() const { 261 return _obj != NULL && _obj->is_objArray() && _index > 0 && 262 _index < objArrayOop(_obj)->length(); 263 } 264 #endif // ASSERT 265 266 void ParallelTaskTerminator::reset_for_reuse(uint n_threads) { 267 reset_for_reuse(); 268 _n_threads = n_threads; 269 }