src/share/vm/utilities/taskqueue.cpp

Thu, 17 Jan 2013 10:25:16 -0500

author
hseigel
date
Thu, 17 Jan 2013 10:25:16 -0500
changeset 4465
203f64878aab
parent 4299
f34d701e952e
child 6680
78bbf4d43a14
permissions
-rw-r--r--

7102489: RFE: cleanup jlong typedef on __APPLE__and _LLP64 systems.
Summary: Define jlong as long on all LP64 platforms and add JLONG_FORMAT macro.
Reviewed-by: dholmes, coleenp, mikael, kvn

duke@435 1 /*
hseigel@4465 2 * Copyright (c) 2001, 2013, 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@4299 28 #include "runtime/thread.inline.hpp"
stefank@2314 29 #include "utilities/debug.hpp"
stefank@2314 30 #include "utilities/stack.inline.hpp"
stefank@2314 31 #include "utilities/taskqueue.hpp"
duke@435 32
jmasa@981 33 #ifdef TRACESPINNING
jmasa@981 34 uint ParallelTaskTerminator::_total_yields = 0;
jmasa@981 35 uint ParallelTaskTerminator::_total_spins = 0;
jmasa@981 36 uint ParallelTaskTerminator::_total_peeks = 0;
jmasa@981 37 #endif
jmasa@981 38
jcoomes@2020 39 #if TASKQUEUE_STATS
jcoomes@2020 40 const char * const TaskQueueStats::_names[last_stat_id] = {
jcoomes@2020 41 "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
jcoomes@2020 42 };
jcoomes@2020 43
jcoomes@2064 44 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
jcoomes@2064 45 {
jcoomes@2064 46 for (unsigned int i = 0; i < last_stat_id; ++i) {
jcoomes@2064 47 _stats[i] += addend._stats[i];
jcoomes@2064 48 }
jcoomes@2064 49 return *this;
jcoomes@2064 50 }
jcoomes@2064 51
jcoomes@2020 52 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
jcoomes@2020 53 unsigned int width)
jcoomes@2020 54 {
jcoomes@2020 55 // Use a width w: 1 <= w <= max_width
jcoomes@2020 56 const unsigned int max_width = 40;
jcoomes@2020 57 const unsigned int w = MAX2(MIN2(width, max_width), 1U);
jcoomes@2020 58
jcoomes@2020 59 if (line == 0) { // spaces equal in width to the header
jcoomes@2020 60 const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
jcoomes@2020 61 stream->print("%*s", hdr_width, " ");
jcoomes@2020 62 } else if (line == 1) { // labels
jcoomes@2020 63 stream->print("%*s", w, _names[0]);
jcoomes@2020 64 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 65 stream->print(" %*s", w, _names[i]);
jcoomes@2020 66 }
jcoomes@2020 67 } else if (line == 2) { // dashed lines
jcoomes@2020 68 char dashes[max_width + 1];
jcoomes@2020 69 memset(dashes, '-', w);
jcoomes@2020 70 dashes[w] = '\0';
jcoomes@2020 71 stream->print("%s", dashes);
jcoomes@2020 72 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 73 stream->print(" %s", dashes);
jcoomes@2020 74 }
jcoomes@2020 75 }
jcoomes@2020 76 }
jcoomes@2020 77
jcoomes@2020 78 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
jcoomes@2020 79 {
jcoomes@2020 80 #define FMT SIZE_FORMAT_W(*)
jcoomes@2020 81 stream->print(FMT, width, _stats[0]);
jcoomes@2020 82 for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020 83 stream->print(" " FMT, width, _stats[i]);
jcoomes@2020 84 }
jcoomes@2020 85 #undef FMT
jcoomes@2020 86 }
jcoomes@2064 87
jcoomes@2064 88 #ifdef ASSERT
jcoomes@2064 89 // Invariants which should hold after a TaskQueue has been emptied and is
jcoomes@2064 90 // quiescent; they do not hold at arbitrary times.
jcoomes@2064 91 void TaskQueueStats::verify() const
jcoomes@2064 92 {
jcoomes@2064 93 assert(get(push) == get(pop) + get(steal),
jcoomes@2064 94 err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
jcoomes@2064 95 get(push), get(pop), get(steal)));
jcoomes@2064 96 assert(get(pop_slow) <= get(pop),
jcoomes@2064 97 err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
jcoomes@2064 98 get(pop_slow), get(pop)));
jcoomes@2064 99 assert(get(steal) <= get(steal_attempt),
jcoomes@2064 100 err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
jcoomes@2064 101 get(steal), get(steal_attempt)));
jcoomes@2064 102 assert(get(overflow) == 0 || get(push) != 0,
jcoomes@2064 103 err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
jcoomes@2064 104 get(overflow), get(push)));
jcoomes@2064 105 assert(get(overflow_max_len) == 0 || get(overflow) != 0,
jcoomes@2064 106 err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
jcoomes@2064 107 get(overflow_max_len), get(overflow)));
jcoomes@2064 108 }
jcoomes@2064 109 #endif // ASSERT
jcoomes@2020 110 #endif // TASKQUEUE_STATS
jcoomes@2020 111
duke@435 112 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
duke@435 113 const int a = 16807;
duke@435 114 const int m = 2147483647;
duke@435 115 const int q = 127773; /* m div a */
duke@435 116 const int r = 2836; /* m mod a */
duke@435 117 assert(sizeof(int) == 4, "I think this relies on that");
duke@435 118 int seed = *seed0;
duke@435 119 int hi = seed / q;
duke@435 120 int lo = seed % q;
duke@435 121 int test = a * lo - r * hi;
duke@435 122 if (test > 0)
duke@435 123 seed = test;
duke@435 124 else
duke@435 125 seed = test + m;
duke@435 126 *seed0 = seed;
duke@435 127 return seed;
duke@435 128 }
duke@435 129
duke@435 130 ParallelTaskTerminator::
duke@435 131 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
duke@435 132 _n_threads(n_threads),
duke@435 133 _queue_set(queue_set),
duke@435 134 _offered_termination(0) {}
duke@435 135
duke@435 136 bool ParallelTaskTerminator::peek_in_queue_set() {
duke@435 137 return _queue_set->peek();
duke@435 138 }
duke@435 139
duke@435 140 void ParallelTaskTerminator::yield() {
ysr@1280 141 assert(_offered_termination <= _n_threads, "Invariant");
duke@435 142 os::yield();
duke@435 143 }
duke@435 144
duke@435 145 void ParallelTaskTerminator::sleep(uint millis) {
ysr@1280 146 assert(_offered_termination <= _n_threads, "Invariant");
duke@435 147 os::sleep(Thread::current(), millis, false);
duke@435 148 }
duke@435 149
ysr@777 150 bool
ysr@777 151 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
jmasa@2188 152 assert(_n_threads > 0, "Initialization is incorrect");
ysr@1280 153 assert(_offered_termination < _n_threads, "Invariant");
duke@435 154 Atomic::inc(&_offered_termination);
duke@435 155
ysr@976 156 uint yield_count = 0;
jmasa@981 157 // Number of hard spin loops done since last yield
jmasa@981 158 uint hard_spin_count = 0;
jmasa@981 159 // Number of iterations in the hard spin loop.
jmasa@981 160 uint hard_spin_limit = WorkStealingHardSpins;
jmasa@981 161
jmasa@981 162 // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
jmasa@981 163 // If it is greater than 0, then start with a small number
jmasa@981 164 // of spins and increase number with each turn at spinning until
jmasa@981 165 // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
jmasa@981 166 // Then do a yield() call and start spinning afresh.
jmasa@981 167 if (WorkStealingSpinToYieldRatio > 0) {
jmasa@981 168 hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
jmasa@981 169 hard_spin_limit = MAX2(hard_spin_limit, 1U);
jmasa@981 170 }
jmasa@981 171 // Remember the initial spin limit.
jmasa@981 172 uint hard_spin_start = hard_spin_limit;
jmasa@981 173
jmasa@981 174 // Loop waiting for all threads to offer termination or
jmasa@981 175 // more work.
duke@435 176 while (true) {
ysr@1280 177 assert(_offered_termination <= _n_threads, "Invariant");
jmasa@981 178 // Are all threads offering termination?
duke@435 179 if (_offered_termination == _n_threads) {
duke@435 180 return true;
duke@435 181 } else {
jmasa@981 182 // Look for more work.
jmasa@981 183 // Periodically sleep() instead of yield() to give threads
jmasa@981 184 // waiting on the cores the chance to grab this code
duke@435 185 if (yield_count <= WorkStealingYieldsBeforeSleep) {
jmasa@981 186 // Do a yield or hardspin. For purposes of deciding whether
jmasa@981 187 // to sleep, count this as a yield.
duke@435 188 yield_count++;
jmasa@981 189
jmasa@981 190 // Periodically call yield() instead spinning
jmasa@981 191 // After WorkStealingSpinToYieldRatio spins, do a yield() call
jmasa@981 192 // and reset the counts and starting limit.
jmasa@981 193 if (hard_spin_count > WorkStealingSpinToYieldRatio) {
jmasa@981 194 yield();
jmasa@981 195 hard_spin_count = 0;
jmasa@981 196 hard_spin_limit = hard_spin_start;
jmasa@981 197 #ifdef TRACESPINNING
jmasa@981 198 _total_yields++;
jmasa@981 199 #endif
jmasa@981 200 } else {
jmasa@981 201 // Hard spin this time
jmasa@981 202 // Increase the hard spinning period but only up to a limit.
jmasa@981 203 hard_spin_limit = MIN2(2*hard_spin_limit,
jmasa@981 204 (uint) WorkStealingHardSpins);
jmasa@981 205 for (uint j = 0; j < hard_spin_limit; j++) {
jmasa@981 206 SpinPause();
jmasa@981 207 }
jmasa@981 208 hard_spin_count++;
jmasa@981 209 #ifdef TRACESPINNING
jmasa@981 210 _total_spins++;
jmasa@981 211 #endif
jmasa@981 212 }
duke@435 213 } else {
duke@435 214 if (PrintGCDetails && Verbose) {
duke@435 215 gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
duke@435 216 "thread %d sleeps after %d yields",
duke@435 217 Thread::current(), yield_count);
duke@435 218 }
duke@435 219 yield_count = 0;
duke@435 220 // A sleep will cause this processor to seek work on another processor's
duke@435 221 // runqueue, if it has nothing else to run (as opposed to the yield
duke@435 222 // which may only move the thread to the end of the this processor's
duke@435 223 // runqueue).
duke@435 224 sleep(WorkStealingSleepMillis);
duke@435 225 }
duke@435 226
jmasa@981 227 #ifdef TRACESPINNING
jmasa@981 228 _total_peeks++;
jmasa@981 229 #endif
ysr@777 230 if (peek_in_queue_set() ||
ysr@777 231 (terminator != NULL && terminator->should_exit_termination())) {
duke@435 232 Atomic::dec(&_offered_termination);
ysr@1280 233 assert(_offered_termination < _n_threads, "Invariant");
duke@435 234 return false;
duke@435 235 }
duke@435 236 }
duke@435 237 }
duke@435 238 }
duke@435 239
jmasa@981 240 #ifdef TRACESPINNING
jmasa@981 241 void ParallelTaskTerminator::print_termination_counts() {
hseigel@4465 242 gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: " UINT32_FORMAT
hseigel@4465 243 " Total spins: " UINT32_FORMAT " Total peeks: " UINT32_FORMAT,
jmasa@981 244 total_yields(),
jmasa@981 245 total_spins(),
jmasa@981 246 total_peeks());
jmasa@981 247 }
jmasa@981 248 #endif
jmasa@981 249
duke@435 250 void ParallelTaskTerminator::reset_for_reuse() {
duke@435 251 if (_offered_termination != 0) {
duke@435 252 assert(_offered_termination == _n_threads,
duke@435 253 "Terminator may still be in use");
duke@435 254 _offered_termination = 0;
duke@435 255 }
duke@435 256 }
duke@435 257
jcoomes@1746 258 #ifdef ASSERT
jcoomes@1746 259 bool ObjArrayTask::is_valid() const {
jcoomes@1746 260 return _obj != NULL && _obj->is_objArray() && _index > 0 &&
jcoomes@1746 261 _index < objArrayOop(_obj)->length();
jcoomes@1746 262 }
jcoomes@1746 263 #endif // ASSERT
jmasa@2188 264
jmasa@2188 265 void ParallelTaskTerminator::reset_for_reuse(int n_threads) {
jmasa@2188 266 reset_for_reuse();
jmasa@2188 267 _n_threads = n_threads;
jmasa@2188 268 }

mercurial