Mon, 09 Mar 2009 13:28:46 -0700
6814575: Update copyright year
Summary: Update copyright for files that have been modified in 2009, up to 03/09
Reviewed-by: katleman, tbell, ohair
duke@435 | 1 | /* |
xdono@1014 | 2 | * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved. |
duke@435 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
duke@435 | 4 | * |
duke@435 | 5 | * This code is free software; you can redistribute it and/or modify it |
duke@435 | 6 | * under the terms of the GNU General Public License version 2 only, as |
duke@435 | 7 | * published by the Free Software Foundation. |
duke@435 | 8 | * |
duke@435 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
duke@435 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
duke@435 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
duke@435 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
duke@435 | 13 | * accompanied this code). |
duke@435 | 14 | * |
duke@435 | 15 | * You should have received a copy of the GNU General Public License version |
duke@435 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
duke@435 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
duke@435 | 18 | * |
duke@435 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
duke@435 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
duke@435 | 21 | * have any questions. |
duke@435 | 22 | * |
duke@435 | 23 | */ |
duke@435 | 24 | |
duke@435 | 25 | # include "incls/_precompiled.incl" |
duke@435 | 26 | # include "incls/_taskqueue.cpp.incl" |
duke@435 | 27 | |
jmasa@981 | 28 | #ifdef TRACESPINNING |
jmasa@981 | 29 | uint ParallelTaskTerminator::_total_yields = 0; |
jmasa@981 | 30 | uint ParallelTaskTerminator::_total_spins = 0; |
jmasa@981 | 31 | uint ParallelTaskTerminator::_total_peeks = 0; |
jmasa@981 | 32 | #endif |
jmasa@981 | 33 | |
duke@435 | 34 | bool TaskQueueSuper::peek() { |
duke@435 | 35 | return _bottom != _age.top(); |
duke@435 | 36 | } |
duke@435 | 37 | |
duke@435 | 38 | int TaskQueueSetSuper::randomParkAndMiller(int *seed0) { |
duke@435 | 39 | const int a = 16807; |
duke@435 | 40 | const int m = 2147483647; |
duke@435 | 41 | const int q = 127773; /* m div a */ |
duke@435 | 42 | const int r = 2836; /* m mod a */ |
duke@435 | 43 | assert(sizeof(int) == 4, "I think this relies on that"); |
duke@435 | 44 | int seed = *seed0; |
duke@435 | 45 | int hi = seed / q; |
duke@435 | 46 | int lo = seed % q; |
duke@435 | 47 | int test = a * lo - r * hi; |
duke@435 | 48 | if (test > 0) |
duke@435 | 49 | seed = test; |
duke@435 | 50 | else |
duke@435 | 51 | seed = test + m; |
duke@435 | 52 | *seed0 = seed; |
duke@435 | 53 | return seed; |
duke@435 | 54 | } |
duke@435 | 55 | |
duke@435 | 56 | ParallelTaskTerminator:: |
duke@435 | 57 | ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) : |
duke@435 | 58 | _n_threads(n_threads), |
duke@435 | 59 | _queue_set(queue_set), |
duke@435 | 60 | _offered_termination(0) {} |
duke@435 | 61 | |
duke@435 | 62 | bool ParallelTaskTerminator::peek_in_queue_set() { |
duke@435 | 63 | return _queue_set->peek(); |
duke@435 | 64 | } |
duke@435 | 65 | |
duke@435 | 66 | void ParallelTaskTerminator::yield() { |
duke@435 | 67 | os::yield(); |
duke@435 | 68 | } |
duke@435 | 69 | |
duke@435 | 70 | void ParallelTaskTerminator::sleep(uint millis) { |
duke@435 | 71 | os::sleep(Thread::current(), millis, false); |
duke@435 | 72 | } |
duke@435 | 73 | |
ysr@777 | 74 | bool |
ysr@777 | 75 | ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) { |
duke@435 | 76 | Atomic::inc(&_offered_termination); |
duke@435 | 77 | |
ysr@976 | 78 | uint yield_count = 0; |
jmasa@981 | 79 | // Number of hard spin loops done since last yield |
jmasa@981 | 80 | uint hard_spin_count = 0; |
jmasa@981 | 81 | // Number of iterations in the hard spin loop. |
jmasa@981 | 82 | uint hard_spin_limit = WorkStealingHardSpins; |
jmasa@981 | 83 | |
jmasa@981 | 84 | // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done. |
jmasa@981 | 85 | // If it is greater than 0, then start with a small number |
jmasa@981 | 86 | // of spins and increase number with each turn at spinning until |
jmasa@981 | 87 | // the count of hard spins exceeds WorkStealingSpinToYieldRatio. |
jmasa@981 | 88 | // Then do a yield() call and start spinning afresh. |
jmasa@981 | 89 | if (WorkStealingSpinToYieldRatio > 0) { |
jmasa@981 | 90 | hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio; |
jmasa@981 | 91 | hard_spin_limit = MAX2(hard_spin_limit, 1U); |
jmasa@981 | 92 | } |
jmasa@981 | 93 | // Remember the initial spin limit. |
jmasa@981 | 94 | uint hard_spin_start = hard_spin_limit; |
jmasa@981 | 95 | |
jmasa@981 | 96 | // Loop waiting for all threads to offer termination or |
jmasa@981 | 97 | // more work. |
duke@435 | 98 | while (true) { |
jmasa@981 | 99 | // Are all threads offering termination? |
duke@435 | 100 | if (_offered_termination == _n_threads) { |
duke@435 | 101 | return true; |
duke@435 | 102 | } else { |
jmasa@981 | 103 | // Look for more work. |
jmasa@981 | 104 | // Periodically sleep() instead of yield() to give threads |
jmasa@981 | 105 | // waiting on the cores the chance to grab this code |
duke@435 | 106 | if (yield_count <= WorkStealingYieldsBeforeSleep) { |
jmasa@981 | 107 | // Do a yield or hardspin. For purposes of deciding whether |
jmasa@981 | 108 | // to sleep, count this as a yield. |
duke@435 | 109 | yield_count++; |
jmasa@981 | 110 | |
jmasa@981 | 111 | // Periodically call yield() instead spinning |
jmasa@981 | 112 | // After WorkStealingSpinToYieldRatio spins, do a yield() call |
jmasa@981 | 113 | // and reset the counts and starting limit. |
jmasa@981 | 114 | if (hard_spin_count > WorkStealingSpinToYieldRatio) { |
jmasa@981 | 115 | yield(); |
jmasa@981 | 116 | hard_spin_count = 0; |
jmasa@981 | 117 | hard_spin_limit = hard_spin_start; |
jmasa@981 | 118 | #ifdef TRACESPINNING |
jmasa@981 | 119 | _total_yields++; |
jmasa@981 | 120 | #endif |
jmasa@981 | 121 | } else { |
jmasa@981 | 122 | // Hard spin this time |
jmasa@981 | 123 | // Increase the hard spinning period but only up to a limit. |
jmasa@981 | 124 | hard_spin_limit = MIN2(2*hard_spin_limit, |
jmasa@981 | 125 | (uint) WorkStealingHardSpins); |
jmasa@981 | 126 | for (uint j = 0; j < hard_spin_limit; j++) { |
jmasa@981 | 127 | SpinPause(); |
jmasa@981 | 128 | } |
jmasa@981 | 129 | hard_spin_count++; |
jmasa@981 | 130 | #ifdef TRACESPINNING |
jmasa@981 | 131 | _total_spins++; |
jmasa@981 | 132 | #endif |
jmasa@981 | 133 | } |
duke@435 | 134 | } else { |
duke@435 | 135 | if (PrintGCDetails && Verbose) { |
duke@435 | 136 | gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() " |
duke@435 | 137 | "thread %d sleeps after %d yields", |
duke@435 | 138 | Thread::current(), yield_count); |
duke@435 | 139 | } |
duke@435 | 140 | yield_count = 0; |
duke@435 | 141 | // A sleep will cause this processor to seek work on another processor's |
duke@435 | 142 | // runqueue, if it has nothing else to run (as opposed to the yield |
duke@435 | 143 | // which may only move the thread to the end of the this processor's |
duke@435 | 144 | // runqueue). |
duke@435 | 145 | sleep(WorkStealingSleepMillis); |
duke@435 | 146 | } |
duke@435 | 147 | |
jmasa@981 | 148 | #ifdef TRACESPINNING |
jmasa@981 | 149 | _total_peeks++; |
jmasa@981 | 150 | #endif |
ysr@777 | 151 | if (peek_in_queue_set() || |
ysr@777 | 152 | (terminator != NULL && terminator->should_exit_termination())) { |
duke@435 | 153 | Atomic::dec(&_offered_termination); |
duke@435 | 154 | return false; |
duke@435 | 155 | } |
duke@435 | 156 | } |
duke@435 | 157 | } |
duke@435 | 158 | } |
duke@435 | 159 | |
jmasa@981 | 160 | #ifdef TRACESPINNING |
jmasa@981 | 161 | void ParallelTaskTerminator::print_termination_counts() { |
jmasa@981 | 162 | gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld " |
jmasa@981 | 163 | "Total spins: %lld Total peeks: %lld", |
jmasa@981 | 164 | total_yields(), |
jmasa@981 | 165 | total_spins(), |
jmasa@981 | 166 | total_peeks()); |
jmasa@981 | 167 | } |
jmasa@981 | 168 | #endif |
jmasa@981 | 169 | |
duke@435 | 170 | void ParallelTaskTerminator::reset_for_reuse() { |
duke@435 | 171 | if (_offered_termination != 0) { |
duke@435 | 172 | assert(_offered_termination == _n_threads, |
duke@435 | 173 | "Terminator may still be in use"); |
duke@435 | 174 | _offered_termination = 0; |
duke@435 | 175 | } |
duke@435 | 176 | } |
duke@435 | 177 | |
jcoomes@810 | 178 | bool RegionTaskQueueWithOverflow::is_empty() { |
jcoomes@810 | 179 | return (_region_queue.size() == 0) && |
duke@435 | 180 | (_overflow_stack->length() == 0); |
duke@435 | 181 | } |
duke@435 | 182 | |
jcoomes@810 | 183 | bool RegionTaskQueueWithOverflow::stealable_is_empty() { |
jcoomes@810 | 184 | return _region_queue.size() == 0; |
duke@435 | 185 | } |
duke@435 | 186 | |
jcoomes@810 | 187 | bool RegionTaskQueueWithOverflow::overflow_is_empty() { |
duke@435 | 188 | return _overflow_stack->length() == 0; |
duke@435 | 189 | } |
duke@435 | 190 | |
jcoomes@810 | 191 | void RegionTaskQueueWithOverflow::initialize() { |
jcoomes@810 | 192 | _region_queue.initialize(); |
duke@435 | 193 | assert(_overflow_stack == 0, "Creating memory leak"); |
duke@435 | 194 | _overflow_stack = |
jcoomes@810 | 195 | new (ResourceObj::C_HEAP) GrowableArray<RegionTask>(10, true); |
duke@435 | 196 | } |
duke@435 | 197 | |
jcoomes@810 | 198 | void RegionTaskQueueWithOverflow::save(RegionTask t) { |
jcoomes@810 | 199 | if (TraceRegionTasksQueuing && Verbose) { |
duke@435 | 200 | gclog_or_tty->print_cr("CTQ: save " PTR_FORMAT, t); |
duke@435 | 201 | } |
jcoomes@810 | 202 | if(!_region_queue.push(t)) { |
duke@435 | 203 | _overflow_stack->push(t); |
duke@435 | 204 | } |
duke@435 | 205 | } |
duke@435 | 206 | |
jcoomes@810 | 207 | // Note that using this method will retrieve all regions |
duke@435 | 208 | // that have been saved but that it will always check |
duke@435 | 209 | // the overflow stack. It may be more efficient to |
duke@435 | 210 | // check the stealable queue and the overflow stack |
duke@435 | 211 | // separately. |
jcoomes@810 | 212 | bool RegionTaskQueueWithOverflow::retrieve(RegionTask& region_task) { |
jcoomes@810 | 213 | bool result = retrieve_from_overflow(region_task); |
duke@435 | 214 | if (!result) { |
jcoomes@810 | 215 | result = retrieve_from_stealable_queue(region_task); |
duke@435 | 216 | } |
jcoomes@810 | 217 | if (TraceRegionTasksQueuing && Verbose && result) { |
duke@435 | 218 | gclog_or_tty->print_cr(" CTQ: retrieve " PTR_FORMAT, result); |
duke@435 | 219 | } |
duke@435 | 220 | return result; |
duke@435 | 221 | } |
duke@435 | 222 | |
jcoomes@810 | 223 | bool RegionTaskQueueWithOverflow::retrieve_from_stealable_queue( |
jcoomes@810 | 224 | RegionTask& region_task) { |
jcoomes@810 | 225 | bool result = _region_queue.pop_local(region_task); |
jcoomes@810 | 226 | if (TraceRegionTasksQueuing && Verbose) { |
jcoomes@810 | 227 | gclog_or_tty->print_cr("CTQ: retrieve_stealable " PTR_FORMAT, region_task); |
duke@435 | 228 | } |
duke@435 | 229 | return result; |
duke@435 | 230 | } |
duke@435 | 231 | |
jcoomes@810 | 232 | bool |
jcoomes@810 | 233 | RegionTaskQueueWithOverflow::retrieve_from_overflow(RegionTask& region_task) { |
duke@435 | 234 | bool result; |
duke@435 | 235 | if (!_overflow_stack->is_empty()) { |
jcoomes@810 | 236 | region_task = _overflow_stack->pop(); |
duke@435 | 237 | result = true; |
duke@435 | 238 | } else { |
jcoomes@810 | 239 | region_task = (RegionTask) NULL; |
duke@435 | 240 | result = false; |
duke@435 | 241 | } |
jcoomes@810 | 242 | if (TraceRegionTasksQueuing && Verbose) { |
jcoomes@810 | 243 | gclog_or_tty->print_cr("CTQ: retrieve_stealable " PTR_FORMAT, region_task); |
duke@435 | 244 | } |
duke@435 | 245 | return result; |
duke@435 | 246 | } |