1 /*
2 * Copyright (c) 2001, 2010, 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 "oops/oop.inline.hpp"
27 #include "runtime/os.hpp"
28 #include "utilities/debug.hpp"
29 #include "utilities/taskqueue.hpp"
30 #ifdef TARGET_OS_FAMILY_linux
31 # include "thread_linux.inline.hpp"
32 #endif
33 #ifdef TARGET_OS_FAMILY_solaris
34 # include "thread_solaris.inline.hpp"
35 #endif
36 #ifdef TARGET_OS_FAMILY_windows
37 # include "thread_windows.inline.hpp"
38 #endif
39
40 #ifdef TRACESPINNING
41 uint ParallelTaskTerminator::_total_yields = 0;
42 uint ParallelTaskTerminator::_total_spins = 0;
43 uint ParallelTaskTerminator::_total_peeks = 0;
44 #endif
45
46 #if TASKQUEUE_STATS
47 const char * const TaskQueueStats::_names[last_stat_id] = {
48 "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
49 };
50
51 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
52 {
53 for (unsigned int i = 0; i < last_stat_id; ++i) {
54 _stats[i] += addend._stats[i];
55 }
56 return *this;
57 }
58
59 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
60 unsigned int width)
61 {
62 // Use a width w: 1 <= w <= max_width
63 const unsigned int max_width = 40;
64 const unsigned int w = MAX2(MIN2(width, max_width), 1U);
65
66 if (line == 0) { // spaces equal in width to the header
67 const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
68 stream->print("%*s", hdr_width, " ");
69 } else if (line == 1) { // labels
70 stream->print("%*s", w, _names[0]);
71 for (unsigned int i = 1; i < last_stat_id; ++i) {
72 stream->print(" %*s", w, _names[i]);
73 }
74 } else if (line == 2) { // dashed lines
75 char dashes[max_width + 1];
76 memset(dashes, '-', w);
77 dashes[w] = '\0';
78 stream->print("%s", dashes);
79 for (unsigned int i = 1; i < last_stat_id; ++i) {
80 stream->print(" %s", dashes);
81 }
82 }
83 }
84
85 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
86 {
87 #define FMT SIZE_FORMAT_W(*)
88 stream->print(FMT, width, _stats[0]);
89 for (unsigned int i = 1; i < last_stat_id; ++i) {
90 stream->print(" " FMT, width, _stats[i]);
91 }
92 #undef FMT
93 }
94
95 #ifdef ASSERT
96 // Invariants which should hold after a TaskQueue has been emptied and is
97 // quiescent; they do not hold at arbitrary times.
98 void TaskQueueStats::verify() const
99 {
100 assert(get(push) == get(pop) + get(steal),
101 err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
102 get(push), get(pop), get(steal)));
103 assert(get(pop_slow) <= get(pop),
104 err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
105 get(pop_slow), get(pop)));
106 assert(get(steal) <= get(steal_attempt),
107 err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
108 get(steal), get(steal_attempt)));
109 assert(get(overflow) == 0 || get(push) != 0,
110 err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
111 get(overflow), get(push)));
112 assert(get(overflow_max_len) == 0 || get(overflow) != 0,
113 err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
114 get(overflow_max_len), get(overflow)));
115 }
116 #endif // ASSERT
117 #endif // TASKQUEUE_STATS
118
119 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
120 const int a = 16807;
121 const int m = 2147483647;
122 const int q = 127773; /* m div a */
123 const int r = 2836; /* m mod a */
124 assert(sizeof(int) == 4, "I think this relies on that");
125 int seed = *seed0;
126 int hi = seed / q;
127 int lo = seed % q;
128 int test = a * lo - r * hi;
129 if (test > 0)
130 seed = test;
131 else
132 seed = test + m;
133 *seed0 = seed;
134 return seed;
135 }
136
137 ParallelTaskTerminator::
138 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
139 _n_threads(n_threads),
140 _queue_set(queue_set),
141 _offered_termination(0) {}
142
143 bool ParallelTaskTerminator::peek_in_queue_set() {
144 return _queue_set->peek();
145 }
146
147 void ParallelTaskTerminator::yield() {
148 assert(_offered_termination <= _n_threads, "Invariant");
149 os::yield();
150 }
151
152 void ParallelTaskTerminator::sleep(uint millis) {
153 assert(_offered_termination <= _n_threads, "Invariant");
154 os::sleep(Thread::current(), millis, false);
155 }
156
157 bool
158 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
159 assert(_offered_termination < _n_threads, "Invariant");
160 Atomic::inc(&_offered_termination);
161
162 uint yield_count = 0;
163 // Number of hard spin loops done since last yield
164 uint hard_spin_count = 0;
165 // Number of iterations in the hard spin loop.
166 uint hard_spin_limit = WorkStealingHardSpins;
167
168 // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
169 // If it is greater than 0, then start with a small number
170 // of spins and increase number with each turn at spinning until
171 // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
172 // Then do a yield() call and start spinning afresh.
173 if (WorkStealingSpinToYieldRatio > 0) {
174 hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
175 hard_spin_limit = MAX2(hard_spin_limit, 1U);
176 }
177 // Remember the initial spin limit.
178 uint hard_spin_start = hard_spin_limit;
179
180 // Loop waiting for all threads to offer termination or
181 // more work.
182 while (true) {
183 assert(_offered_termination <= _n_threads, "Invariant");
184 // Are all threads offering termination?
185 if (_offered_termination == _n_threads) {
186 return true;
187 } else {
188 // Look for more work.
189 // Periodically sleep() instead of yield() to give threads
190 // waiting on the cores the chance to grab this code
191 if (yield_count <= WorkStealingYieldsBeforeSleep) {
192 // Do a yield or hardspin. For purposes of deciding whether
193 // to sleep, count this as a yield.
194 yield_count++;
195
196 // Periodically call yield() instead spinning
197 // After WorkStealingSpinToYieldRatio spins, do a yield() call
198 // and reset the counts and starting limit.
199 if (hard_spin_count > WorkStealingSpinToYieldRatio) {
200 yield();
201 hard_spin_count = 0;
202 hard_spin_limit = hard_spin_start;
203 #ifdef TRACESPINNING
204 _total_yields++;
205 #endif
206 } else {
207 // Hard spin this time
208 // Increase the hard spinning period but only up to a limit.
209 hard_spin_limit = MIN2(2*hard_spin_limit,
210 (uint) WorkStealingHardSpins);
211 for (uint j = 0; j < hard_spin_limit; j++) {
212 SpinPause();
213 }
214 hard_spin_count++;
215 #ifdef TRACESPINNING
216 _total_spins++;
217 #endif
218 }
219 } else {
220 if (PrintGCDetails && Verbose) {
221 gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
222 "thread %d sleeps after %d yields",
223 Thread::current(), yield_count);
224 }
225 yield_count = 0;
226 // A sleep will cause this processor to seek work on another processor's
227 // runqueue, if it has nothing else to run (as opposed to the yield
228 // which may only move the thread to the end of the this processor's
229 // runqueue).
230 sleep(WorkStealingSleepMillis);
231 }
232
233 #ifdef TRACESPINNING
234 _total_peeks++;
235 #endif
236 if (peek_in_queue_set() ||
237 (terminator != NULL && terminator->should_exit_termination())) {
238 Atomic::dec(&_offered_termination);
239 assert(_offered_termination < _n_threads, "Invariant");
240 return false;
241 }
242 }
243 }
244 }
245
246 #ifdef TRACESPINNING
247 void ParallelTaskTerminator::print_termination_counts() {
248 gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld "
249 "Total spins: %lld Total peeks: %lld",
250 total_yields(),
251 total_spins(),
252 total_peeks());
253 }
254 #endif
255
256 void ParallelTaskTerminator::reset_for_reuse() {
257 if (_offered_termination != 0) {
258 assert(_offered_termination == _n_threads,
259 "Terminator may still be in use");
260 _offered_termination = 0;
261 }
262 }
263
264 #ifdef ASSERT
265 bool ObjArrayTask::is_valid() const {
266 return _obj != NULL && _obj->is_objArray() && _index > 0 &&
267 _index < objArrayOop(_obj)->length();
268 }
269 #endif // ASSERT