src/share/vm/utilities/taskqueue.cpp

Mon, 02 Jul 2012 13:11:28 -0400

author
coleenp
date
Mon, 02 Jul 2012 13:11:28 -0400
changeset 3901
24b9c7f4cae6
parent 3156
f08d439fab8c
child 4299
f34d701e952e
permissions
-rw-r--r--

Merge

duke@435 1 /*
jcoomes@1993 2 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. 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 *
trims@1907 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907 20 * or visit www.oracle.com if you need additional information or have any
trims@1907 21 * questions.
duke@435 22 *
duke@435 23 */
duke@435 24
stefank@2314 25 #include "precompiled.hpp"
stefank@2314 26 #include "oops/oop.inline.hpp"
stefank@2314 27 #include "runtime/os.hpp"
stefank@2314 28 #include "utilities/debug.hpp"
stefank@2314 29 #include "utilities/stack.inline.hpp"
stefank@2314 30 #include "utilities/taskqueue.hpp"
stefank@2314 31 #ifdef TARGET_OS_FAMILY_linux
stefank@2314 32 # include "thread_linux.inline.hpp"
stefank@2314 33 #endif
stefank@2314 34 #ifdef TARGET_OS_FAMILY_solaris
stefank@2314 35 # include "thread_solaris.inline.hpp"
stefank@2314 36 #endif
stefank@2314 37 #ifdef TARGET_OS_FAMILY_windows
stefank@2314 38 # include "thread_windows.inline.hpp"
stefank@2314 39 #endif
never@3156 40 #ifdef TARGET_OS_FAMILY_bsd
never@3156 41 # include "thread_bsd.inline.hpp"
never@3156 42 #endif
duke@435 43
jmasa@981 44 #ifdef TRACESPINNING
jmasa@981 45 uint ParallelTaskTerminator::_total_yields = 0;
jmasa@981 46 uint ParallelTaskTerminator::_total_spins = 0;
jmasa@981 47 uint ParallelTaskTerminator::_total_peeks = 0;
jmasa@981 48 #endif
jmasa@981 49
jcoomes@2020 50 #if TASKQUEUE_STATS
jcoomes@2020 51 const char * const TaskQueueStats::_names[last_stat_id] = {
jcoomes@2020 52 "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
jcoomes@2020 53 };
jcoomes@2020 54
jcoomes@2064 55 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
jcoomes@2064 56 {
jcoomes@2064 57 for (unsigned int i = 0; i < last_stat_id; ++i) {
jcoomes@2064 58 _stats[i] += addend._stats[i];
jcoomes@2064 59 }
jcoomes@2064 60 return *this;
jcoomes@2064 61 }
jcoomes@2064 62
jcoomes@2020 63 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
jcoomes@2020 64 unsigned int width)
jcoomes@2020 65 {
jcoomes@2020 66 // Use a width w: 1 <= w <= max_width
jcoomes@2020 67 const unsigned int max_width = 40;
jcoomes@2020 68 const unsigned int w = MAX2(MIN2(width, max_width), 1U);
jcoomes@2020 69
jcoomes@2020 70 if (line == 0) { // spaces equal in width to the header
jcoomes@2020 71 const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
jcoomes@2020 72 stream->print("%*s", hdr_width, " ");
jcoomes@2020 73 } else if (line == 1) { // labels
jcoomes@2020 74 stream->print("%*s", w, _names[0]);
jcoomes@2020 75 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 76 stream->print(" %*s", w, _names[i]);
jcoomes@2020 77 }
jcoomes@2020 78 } else if (line == 2) { // dashed lines
jcoomes@2020 79 char dashes[max_width + 1];
jcoomes@2020 80 memset(dashes, '-', w);
jcoomes@2020 81 dashes[w] = '\0';
jcoomes@2020 82 stream->print("%s", dashes);
jcoomes@2020 83 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 84 stream->print(" %s", dashes);
jcoomes@2020 85 }
jcoomes@2020 86 }
jcoomes@2020 87 }
jcoomes@2020 88
jcoomes@2020 89 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
jcoomes@2020 90 {
jcoomes@2020 91 #define FMT SIZE_FORMAT_W(*)
jcoomes@2020 92 stream->print(FMT, width, _stats[0]);
jcoomes@2020 93 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 94 stream->print(" " FMT, width, _stats[i]);
jcoomes@2020 95 }
jcoomes@2020 96 #undef FMT
jcoomes@2020 97 }
jcoomes@2064 98
jcoomes@2064 99 #ifdef ASSERT
jcoomes@2064 100 // Invariants which should hold after a TaskQueue has been emptied and is
jcoomes@2064 101 // quiescent; they do not hold at arbitrary times.
jcoomes@2064 102 void TaskQueueStats::verify() const
jcoomes@2064 103 {
jcoomes@2064 104 assert(get(push) == get(pop) + get(steal),
jcoomes@2064 105 err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
jcoomes@2064 106 get(push), get(pop), get(steal)));
jcoomes@2064 107 assert(get(pop_slow) <= get(pop),
jcoomes@2064 108 err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
jcoomes@2064 109 get(pop_slow), get(pop)));
jcoomes@2064 110 assert(get(steal) <= get(steal_attempt),
jcoomes@2064 111 err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
jcoomes@2064 112 get(steal), get(steal_attempt)));
jcoomes@2064 113 assert(get(overflow) == 0 || get(push) != 0,
jcoomes@2064 114 err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
jcoomes@2064 115 get(overflow), get(push)));
jcoomes@2064 116 assert(get(overflow_max_len) == 0 || get(overflow) != 0,
jcoomes@2064 117 err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
jcoomes@2064 118 get(overflow_max_len), get(overflow)));
jcoomes@2064 119 }
jcoomes@2064 120 #endif // ASSERT
jcoomes@2020 121 #endif // TASKQUEUE_STATS
jcoomes@2020 122
duke@435 123 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
duke@435 124 const int a = 16807;
duke@435 125 const int m = 2147483647;
duke@435 126 const int q = 127773; /* m div a */
duke@435 127 const int r = 2836; /* m mod a */
duke@435 128 assert(sizeof(int) == 4, "I think this relies on that");
duke@435 129 int seed = *seed0;
duke@435 130 int hi = seed / q;
duke@435 131 int lo = seed % q;
duke@435 132 int test = a * lo - r * hi;
duke@435 133 if (test > 0)
duke@435 134 seed = test;
duke@435 135 else
duke@435 136 seed = test + m;
duke@435 137 *seed0 = seed;
duke@435 138 return seed;
duke@435 139 }
duke@435 140
duke@435 141 ParallelTaskTerminator::
duke@435 142 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
duke@435 143 _n_threads(n_threads),
duke@435 144 _queue_set(queue_set),
duke@435 145 _offered_termination(0) {}
duke@435 146
duke@435 147 bool ParallelTaskTerminator::peek_in_queue_set() {
duke@435 148 return _queue_set->peek();
duke@435 149 }
duke@435 150
duke@435 151 void ParallelTaskTerminator::yield() {
ysr@1280 152 assert(_offered_termination <= _n_threads, "Invariant");
duke@435 153 os::yield();
duke@435 154 }
duke@435 155
duke@435 156 void ParallelTaskTerminator::sleep(uint millis) {
ysr@1280 157 assert(_offered_termination <= _n_threads, "Invariant");
duke@435 158 os::sleep(Thread::current(), millis, false);
duke@435 159 }
duke@435 160
ysr@777 161 bool
ysr@777 162 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
jmasa@2188 163 assert(_n_threads > 0, "Initialization is incorrect");
ysr@1280 164 assert(_offered_termination < _n_threads, "Invariant");
duke@435 165 Atomic::inc(&_offered_termination);
duke@435 166
ysr@976 167 uint yield_count = 0;
jmasa@981 168 // Number of hard spin loops done since last yield
jmasa@981 169 uint hard_spin_count = 0;
jmasa@981 170 // Number of iterations in the hard spin loop.
jmasa@981 171 uint hard_spin_limit = WorkStealingHardSpins;
jmasa@981 172
jmasa@981 173 // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
jmasa@981 174 // If it is greater than 0, then start with a small number
jmasa@981 175 // of spins and increase number with each turn at spinning until
jmasa@981 176 // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
jmasa@981 177 // Then do a yield() call and start spinning afresh.
jmasa@981 178 if (WorkStealingSpinToYieldRatio > 0) {
jmasa@981 179 hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
jmasa@981 180 hard_spin_limit = MAX2(hard_spin_limit, 1U);
jmasa@981 181 }
jmasa@981 182 // Remember the initial spin limit.
jmasa@981 183 uint hard_spin_start = hard_spin_limit;
jmasa@981 184
jmasa@981 185 // Loop waiting for all threads to offer termination or
jmasa@981 186 // more work.
duke@435 187 while (true) {
ysr@1280 188 assert(_offered_termination <= _n_threads, "Invariant");
jmasa@981 189 // Are all threads offering termination?
duke@435 190 if (_offered_termination == _n_threads) {
duke@435 191 return true;
duke@435 192 } else {
jmasa@981 193 // Look for more work.
jmasa@981 194 // Periodically sleep() instead of yield() to give threads
jmasa@981 195 // waiting on the cores the chance to grab this code
duke@435 196 if (yield_count <= WorkStealingYieldsBeforeSleep) {
jmasa@981 197 // Do a yield or hardspin. For purposes of deciding whether
jmasa@981 198 // to sleep, count this as a yield.
duke@435 199 yield_count++;
jmasa@981 200
jmasa@981 201 // Periodically call yield() instead spinning
jmasa@981 202 // After WorkStealingSpinToYieldRatio spins, do a yield() call
jmasa@981 203 // and reset the counts and starting limit.
jmasa@981 204 if (hard_spin_count > WorkStealingSpinToYieldRatio) {
jmasa@981 205 yield();
jmasa@981 206 hard_spin_count = 0;
jmasa@981 207 hard_spin_limit = hard_spin_start;
jmasa@981 208 #ifdef TRACESPINNING
jmasa@981 209 _total_yields++;
jmasa@981 210 #endif
jmasa@981 211 } else {
jmasa@981 212 // Hard spin this time
jmasa@981 213 // Increase the hard spinning period but only up to a limit.
jmasa@981 214 hard_spin_limit = MIN2(2*hard_spin_limit,
jmasa@981 215 (uint) WorkStealingHardSpins);
jmasa@981 216 for (uint j = 0; j < hard_spin_limit; j++) {
jmasa@981 217 SpinPause();
jmasa@981 218 }
jmasa@981 219 hard_spin_count++;
jmasa@981 220 #ifdef TRACESPINNING
jmasa@981 221 _total_spins++;
jmasa@981 222 #endif
jmasa@981 223 }
duke@435 224 } else {
duke@435 225 if (PrintGCDetails && Verbose) {
duke@435 226 gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
duke@435 227 "thread %d sleeps after %d yields",
duke@435 228 Thread::current(), yield_count);
duke@435 229 }
duke@435 230 yield_count = 0;
duke@435 231 // A sleep will cause this processor to seek work on another processor's
duke@435 232 // runqueue, if it has nothing else to run (as opposed to the yield
duke@435 233 // which may only move the thread to the end of the this processor's
duke@435 234 // runqueue).
duke@435 235 sleep(WorkStealingSleepMillis);
duke@435 236 }
duke@435 237
jmasa@981 238 #ifdef TRACESPINNING
jmasa@981 239 _total_peeks++;
jmasa@981 240 #endif
ysr@777 241 if (peek_in_queue_set() ||
ysr@777 242 (terminator != NULL && terminator->should_exit_termination())) {
duke@435 243 Atomic::dec(&_offered_termination);
ysr@1280 244 assert(_offered_termination < _n_threads, "Invariant");
duke@435 245 return false;
duke@435 246 }
duke@435 247 }
duke@435 248 }
duke@435 249 }
duke@435 250
jmasa@981 251 #ifdef TRACESPINNING
jmasa@981 252 void ParallelTaskTerminator::print_termination_counts() {
jmasa@981 253 gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld "
jmasa@981 254 "Total spins: %lld Total peeks: %lld",
jmasa@981 255 total_yields(),
jmasa@981 256 total_spins(),
jmasa@981 257 total_peeks());
jmasa@981 258 }
jmasa@981 259 #endif
jmasa@981 260
duke@435 261 void ParallelTaskTerminator::reset_for_reuse() {
duke@435 262 if (_offered_termination != 0) {
duke@435 263 assert(_offered_termination == _n_threads,
duke@435 264 "Terminator may still be in use");
duke@435 265 _offered_termination = 0;
duke@435 266 }
duke@435 267 }
duke@435 268
jcoomes@1746 269 #ifdef ASSERT
jcoomes@1746 270 bool ObjArrayTask::is_valid() const {
jcoomes@1746 271 return _obj != NULL && _obj->is_objArray() && _index > 0 &&
jcoomes@1746 272 _index < objArrayOop(_obj)->length();
jcoomes@1746 273 }
jcoomes@1746 274 #endif // ASSERT
jmasa@2188 275
jmasa@2188 276 void ParallelTaskTerminator::reset_for_reuse(int n_threads) {
jmasa@2188 277 reset_for_reuse();
jmasa@2188 278 _n_threads = n_threads;
jmasa@2188 279 }

mercurial