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/stack.inline.hpp"
  30 #include "utilities/taskqueue.hpp"
  31 #ifdef TARGET_OS_FAMILY_linux
  32 # include "thread_linux.inline.hpp"
  33 #endif
  34 #ifdef TARGET_OS_FAMILY_solaris
  35 # include "thread_solaris.inline.hpp"
  36 #endif
  37 #ifdef TARGET_OS_FAMILY_windows
  38 # include "thread_windows.inline.hpp"
  39 #endif
  40 
  41 #ifdef TRACESPINNING
  42 uint ParallelTaskTerminator::_total_yields = 0;
  43 uint ParallelTaskTerminator::_total_spins = 0;
  44 uint ParallelTaskTerminator::_total_peeks = 0;
  45 #endif
  46 
  47 #if TASKQUEUE_STATS
  48 const char * const TaskQueueStats::_names[last_stat_id] = {
  49   "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
  50 };
  51 
  52 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
  53 {
  54   for (unsigned int i = 0; i < last_stat_id; ++i) {
  55     _stats[i] += addend._stats[i];
  56   }
  57   return *this;
  58 }
  59 
  60 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
  61                                   unsigned int width)
  62 {
  63   // Use a width w: 1 <= w <= max_width
  64   const unsigned int max_width = 40;
  65   const unsigned int w = MAX2(MIN2(width, max_width), 1U);
  66 
  67   if (line == 0) { // spaces equal in width to the header
  68     const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
  69     stream->print("%*s", hdr_width, " ");
  70   } else if (line == 1) { // labels
  71     stream->print("%*s", w, _names[0]);
  72     for (unsigned int i = 1; i < last_stat_id; ++i) {
  73       stream->print(" %*s", w, _names[i]);
  74     }
  75   } else if (line == 2) { // dashed lines
  76     char dashes[max_width + 1];
  77     memset(dashes, '-', w);
  78     dashes[w] = '\0';
  79     stream->print("%s", dashes);
  80     for (unsigned int i = 1; i < last_stat_id; ++i) {
  81       stream->print(" %s", dashes);
  82     }
  83   }
  84 }
  85 
  86 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
  87 {
  88   #define FMT SIZE_FORMAT_W(*)
  89   stream->print(FMT, width, _stats[0]);
  90   for (unsigned int i = 1; i < last_stat_id; ++i) {
  91     stream->print(" " FMT, width, _stats[i]);
  92   }
  93   #undef FMT
  94 }
  95 
  96 #ifdef ASSERT
  97 // Invariants which should hold after a TaskQueue has been emptied and is
  98 // quiescent; they do not hold at arbitrary times.
  99 void TaskQueueStats::verify() const
 100 {
 101   assert(get(push) == get(pop) + get(steal),
 102          err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
 103                  get(push), get(pop), get(steal)));
 104   assert(get(pop_slow) <= get(pop),
 105          err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
 106                  get(pop_slow), get(pop)));
 107   assert(get(steal) <= get(steal_attempt),
 108          err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
 109                  get(steal), get(steal_attempt)));
 110   assert(get(overflow) == 0 || get(push) != 0,
 111          err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
 112                  get(overflow), get(push)));
 113   assert(get(overflow_max_len) == 0 || get(overflow) != 0,
 114          err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
 115                  get(overflow_max_len), get(overflow)));
 116 }
 117 #endif // ASSERT
 118 #endif // TASKQUEUE_STATS
 119 
 120 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
 121   const int a =      16807;
 122   const int m = 2147483647;
 123   const int q =     127773;  /* m div a */
 124   const int r =       2836;  /* m mod a */
 125   assert(sizeof(int) == 4, "I think this relies on that");
 126   int seed = *seed0;
 127   int hi   = seed / q;
 128   int lo   = seed % q;
 129   int test = a * lo - r * hi;
 130   if (test > 0)
 131     seed = test;
 132   else
 133     seed = test + m;
 134   *seed0 = seed;
 135   return seed;
 136 }
 137 
 138 ParallelTaskTerminator::
 139 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
 140   _n_threads(n_threads),
 141   _queue_set(queue_set),
 142   _offered_termination(0) {}
 143 
 144 bool ParallelTaskTerminator::peek_in_queue_set() {
 145   return _queue_set->peek();
 146 }
 147 
 148 void ParallelTaskTerminator::yield() {
 149   assert(_offered_termination <= _n_threads, "Invariant");
 150   os::yield();
 151 }
 152 
 153 void ParallelTaskTerminator::sleep(uint millis) {
 154   assert(_offered_termination <= _n_threads, "Invariant");
 155   os::sleep(Thread::current(), millis, false);
 156 }
 157 
 158 bool
 159 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
 160   assert(_n_threads > 0, "Initialization is incorrect");
 161   assert(_offered_termination < _n_threads, "Invariant");
 162   Atomic::inc(&_offered_termination);
 163 
 164   uint yield_count = 0;
 165   // Number of hard spin loops done since last yield
 166   uint hard_spin_count = 0;
 167   // Number of iterations in the hard spin loop.
 168   uint hard_spin_limit = WorkStealingHardSpins;
 169 
 170   // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
 171   // If it is greater than 0, then start with a small number
 172   // of spins and increase number with each turn at spinning until
 173   // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
 174   // Then do a yield() call and start spinning afresh.
 175   if (WorkStealingSpinToYieldRatio > 0) {
 176     hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
 177     hard_spin_limit = MAX2(hard_spin_limit, 1U);
 178   }
 179   // Remember the initial spin limit.
 180   uint hard_spin_start = hard_spin_limit;
 181 
 182   // Loop waiting for all threads to offer termination or
 183   // more work.
 184   while (true) {
 185     assert(_offered_termination <= _n_threads, "Invariant");
 186     // Are all threads offering termination?
 187     if (_offered_termination == _n_threads) {
 188       return true;
 189     } else {
 190       // Look for more work.
 191       // Periodically sleep() instead of yield() to give threads
 192       // waiting on the cores the chance to grab this code
 193       if (yield_count <= WorkStealingYieldsBeforeSleep) {
 194         // Do a yield or hardspin.  For purposes of deciding whether
 195         // to sleep, count this as a yield.
 196         yield_count++;
 197 
 198         // Periodically call yield() instead spinning
 199         // After WorkStealingSpinToYieldRatio spins, do a yield() call
 200         // and reset the counts and starting limit.
 201         if (hard_spin_count > WorkStealingSpinToYieldRatio) {
 202           yield();
 203           hard_spin_count = 0;
 204           hard_spin_limit = hard_spin_start;
 205 #ifdef TRACESPINNING
 206           _total_yields++;
 207 #endif
 208         } else {
 209           // Hard spin this time
 210           // Increase the hard spinning period but only up to a limit.
 211           hard_spin_limit = MIN2(2*hard_spin_limit,
 212                                  (uint) WorkStealingHardSpins);
 213           for (uint j = 0; j < hard_spin_limit; j++) {
 214             SpinPause();
 215           }
 216           hard_spin_count++;
 217 #ifdef TRACESPINNING
 218           _total_spins++;
 219 #endif
 220         }
 221       } else {
 222         if (PrintGCDetails && Verbose) {
 223          gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
 224            "thread %d sleeps after %d yields",
 225            Thread::current(), yield_count);
 226         }
 227         yield_count = 0;
 228         // A sleep will cause this processor to seek work on another processor's
 229         // runqueue, if it has nothing else to run (as opposed to the yield
 230         // which may only move the thread to the end of the this processor's
 231         // runqueue).
 232         sleep(WorkStealingSleepMillis);
 233       }
 234 
 235 #ifdef TRACESPINNING
 236       _total_peeks++;
 237 #endif
 238       if (peek_in_queue_set() ||
 239           (terminator != NULL && terminator->should_exit_termination())) {
 240         Atomic::dec(&_offered_termination);
 241         assert(_offered_termination < _n_threads, "Invariant");
 242         return false;
 243       }
 244     }
 245   }
 246 }
 247 
 248 #ifdef TRACESPINNING
 249 void ParallelTaskTerminator::print_termination_counts() {
 250   gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld  "
 251     "Total spins: %lld  Total peeks: %lld",
 252     total_yields(),
 253     total_spins(),
 254     total_peeks());
 255 }
 256 #endif
 257 
 258 void ParallelTaskTerminator::reset_for_reuse() {
 259   if (_offered_termination != 0) {
 260     assert(_offered_termination == _n_threads,
 261            "Terminator may still be in use");
 262     _offered_termination = 0;
 263   }
 264 }
 265 
 266 #ifdef ASSERT
 267 bool ObjArrayTask::is_valid() const {
 268   return _obj != NULL && _obj->is_objArray() && _index > 0 &&
 269     _index < objArrayOop(_obj)->length();
 270 }
 271 #endif // ASSERT
 272 
 273 void ParallelTaskTerminator::reset_for_reuse(int n_threads) {
 274   reset_for_reuse();
 275   _n_threads = n_threads;
 276 }
 277