duke@435: /*
jcoomes@1993:  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: # include "incls/_precompiled.incl"
duke@435: # include "incls/_taskqueue.cpp.incl"
duke@435: 
jmasa@981: #ifdef TRACESPINNING
jmasa@981: uint ParallelTaskTerminator::_total_yields = 0;
jmasa@981: uint ParallelTaskTerminator::_total_spins = 0;
jmasa@981: uint ParallelTaskTerminator::_total_peeks = 0;
jmasa@981: #endif
jmasa@981: 
jcoomes@2020: #if TASKQUEUE_STATS
jcoomes@2020: const char * const TaskQueueStats::_names[last_stat_id] = {
jcoomes@2020:   "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
jcoomes@2020: };
jcoomes@2020: 
jcoomes@2064: TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
jcoomes@2064: {
jcoomes@2064:   for (unsigned int i = 0; i < last_stat_id; ++i) {
jcoomes@2064:     _stats[i] += addend._stats[i];
jcoomes@2064:   }
jcoomes@2064:   return *this;
jcoomes@2064: }
jcoomes@2064: 
jcoomes@2020: void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
jcoomes@2020:                                   unsigned int width)
jcoomes@2020: {
jcoomes@2020:   // Use a width w: 1 <= w <= max_width
jcoomes@2020:   const unsigned int max_width = 40;
jcoomes@2020:   const unsigned int w = MAX2(MIN2(width, max_width), 1U);
jcoomes@2020: 
jcoomes@2020:   if (line == 0) { // spaces equal in width to the header
jcoomes@2020:     const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
jcoomes@2020:     stream->print("%*s", hdr_width, " ");
jcoomes@2020:   } else if (line == 1) { // labels
jcoomes@2020:     stream->print("%*s", w, _names[0]);
jcoomes@2020:     for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020:       stream->print(" %*s", w, _names[i]);
jcoomes@2020:     }
jcoomes@2020:   } else if (line == 2) { // dashed lines
jcoomes@2020:     char dashes[max_width + 1];
jcoomes@2020:     memset(dashes, '-', w);
jcoomes@2020:     dashes[w] = '\0';
jcoomes@2020:     stream->print("%s", dashes);
jcoomes@2020:     for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020:       stream->print(" %s", dashes);
jcoomes@2020:     }
jcoomes@2020:   }
jcoomes@2020: }
jcoomes@2020: 
jcoomes@2020: void TaskQueueStats::print(outputStream* stream, unsigned int width) const
jcoomes@2020: {
jcoomes@2020:   #define FMT SIZE_FORMAT_W(*)
jcoomes@2020:   stream->print(FMT, width, _stats[0]);
jcoomes@2020:   for (unsigned int i = 1; i < last_stat_id; ++i) {
jcoomes@2020:     stream->print(" " FMT, width, _stats[i]);
jcoomes@2020:   }
jcoomes@2020:   #undef FMT
jcoomes@2020: }
jcoomes@2064: 
jcoomes@2064: #ifdef ASSERT
jcoomes@2064: // Invariants which should hold after a TaskQueue has been emptied and is
jcoomes@2064: // quiescent; they do not hold at arbitrary times.
jcoomes@2064: void TaskQueueStats::verify() const
jcoomes@2064: {
jcoomes@2064:   assert(get(push) == get(pop) + get(steal),
jcoomes@2064:          err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
jcoomes@2064:                  get(push), get(pop), get(steal)));
jcoomes@2064:   assert(get(pop_slow) <= get(pop),
jcoomes@2064:          err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
jcoomes@2064:                  get(pop_slow), get(pop)));
jcoomes@2064:   assert(get(steal) <= get(steal_attempt),
jcoomes@2064:          err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
jcoomes@2064:                  get(steal), get(steal_attempt)));
jcoomes@2064:   assert(get(overflow) == 0 || get(push) != 0,
jcoomes@2064:          err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
jcoomes@2064:                  get(overflow), get(push)));
jcoomes@2064:   assert(get(overflow_max_len) == 0 || get(overflow) != 0,
jcoomes@2064:          err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
jcoomes@2064:                  get(overflow_max_len), get(overflow)));
jcoomes@2064: }
jcoomes@2064: #endif // ASSERT
jcoomes@2020: #endif // TASKQUEUE_STATS
jcoomes@2020: 
duke@435: int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
duke@435:   const int a =      16807;
duke@435:   const int m = 2147483647;
duke@435:   const int q =     127773;  /* m div a */
duke@435:   const int r =       2836;  /* m mod a */
duke@435:   assert(sizeof(int) == 4, "I think this relies on that");
duke@435:   int seed = *seed0;
duke@435:   int hi   = seed / q;
duke@435:   int lo   = seed % q;
duke@435:   int test = a * lo - r * hi;
duke@435:   if (test > 0)
duke@435:     seed = test;
duke@435:   else
duke@435:     seed = test + m;
duke@435:   *seed0 = seed;
duke@435:   return seed;
duke@435: }
duke@435: 
duke@435: ParallelTaskTerminator::
duke@435: ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
duke@435:   _n_threads(n_threads),
duke@435:   _queue_set(queue_set),
duke@435:   _offered_termination(0) {}
duke@435: 
duke@435: bool ParallelTaskTerminator::peek_in_queue_set() {
duke@435:   return _queue_set->peek();
duke@435: }
duke@435: 
duke@435: void ParallelTaskTerminator::yield() {
ysr@1280:   assert(_offered_termination <= _n_threads, "Invariant");
duke@435:   os::yield();
duke@435: }
duke@435: 
duke@435: void ParallelTaskTerminator::sleep(uint millis) {
ysr@1280:   assert(_offered_termination <= _n_threads, "Invariant");
duke@435:   os::sleep(Thread::current(), millis, false);
duke@435: }
duke@435: 
ysr@777: bool
ysr@777: ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
ysr@1280:   assert(_offered_termination < _n_threads, "Invariant");
duke@435:   Atomic::inc(&_offered_termination);
duke@435: 
ysr@976:   uint yield_count = 0;
jmasa@981:   // Number of hard spin loops done since last yield
jmasa@981:   uint hard_spin_count = 0;
jmasa@981:   // Number of iterations in the hard spin loop.
jmasa@981:   uint hard_spin_limit = WorkStealingHardSpins;
jmasa@981: 
jmasa@981:   // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
jmasa@981:   // If it is greater than 0, then start with a small number
jmasa@981:   // of spins and increase number with each turn at spinning until
jmasa@981:   // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
jmasa@981:   // Then do a yield() call and start spinning afresh.
jmasa@981:   if (WorkStealingSpinToYieldRatio > 0) {
jmasa@981:     hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
jmasa@981:     hard_spin_limit = MAX2(hard_spin_limit, 1U);
jmasa@981:   }
jmasa@981:   // Remember the initial spin limit.
jmasa@981:   uint hard_spin_start = hard_spin_limit;
jmasa@981: 
jmasa@981:   // Loop waiting for all threads to offer termination or
jmasa@981:   // more work.
duke@435:   while (true) {
ysr@1280:     assert(_offered_termination <= _n_threads, "Invariant");
jmasa@981:     // Are all threads offering termination?
duke@435:     if (_offered_termination == _n_threads) {
duke@435:       return true;
duke@435:     } else {
jmasa@981:       // Look for more work.
jmasa@981:       // Periodically sleep() instead of yield() to give threads
jmasa@981:       // waiting on the cores the chance to grab this code
duke@435:       if (yield_count <= WorkStealingYieldsBeforeSleep) {
jmasa@981:         // Do a yield or hardspin.  For purposes of deciding whether
jmasa@981:         // to sleep, count this as a yield.
duke@435:         yield_count++;
jmasa@981: 
jmasa@981:         // Periodically call yield() instead spinning
jmasa@981:         // After WorkStealingSpinToYieldRatio spins, do a yield() call
jmasa@981:         // and reset the counts and starting limit.
jmasa@981:         if (hard_spin_count > WorkStealingSpinToYieldRatio) {
jmasa@981:           yield();
jmasa@981:           hard_spin_count = 0;
jmasa@981:           hard_spin_limit = hard_spin_start;
jmasa@981: #ifdef TRACESPINNING
jmasa@981:           _total_yields++;
jmasa@981: #endif
jmasa@981:         } else {
jmasa@981:           // Hard spin this time
jmasa@981:           // Increase the hard spinning period but only up to a limit.
jmasa@981:           hard_spin_limit = MIN2(2*hard_spin_limit,
jmasa@981:                                  (uint) WorkStealingHardSpins);
jmasa@981:           for (uint j = 0; j < hard_spin_limit; j++) {
jmasa@981:             SpinPause();
jmasa@981:           }
jmasa@981:           hard_spin_count++;
jmasa@981: #ifdef TRACESPINNING
jmasa@981:           _total_spins++;
jmasa@981: #endif
jmasa@981:         }
duke@435:       } else {
duke@435:         if (PrintGCDetails && Verbose) {
duke@435:          gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
duke@435:            "thread %d sleeps after %d yields",
duke@435:            Thread::current(), yield_count);
duke@435:         }
duke@435:         yield_count = 0;
duke@435:         // A sleep will cause this processor to seek work on another processor's
duke@435:         // runqueue, if it has nothing else to run (as opposed to the yield
duke@435:         // which may only move the thread to the end of the this processor's
duke@435:         // runqueue).
duke@435:         sleep(WorkStealingSleepMillis);
duke@435:       }
duke@435: 
jmasa@981: #ifdef TRACESPINNING
jmasa@981:       _total_peeks++;
jmasa@981: #endif
ysr@777:       if (peek_in_queue_set() ||
ysr@777:           (terminator != NULL && terminator->should_exit_termination())) {
duke@435:         Atomic::dec(&_offered_termination);
ysr@1280:         assert(_offered_termination < _n_threads, "Invariant");
duke@435:         return false;
duke@435:       }
duke@435:     }
duke@435:   }
duke@435: }
duke@435: 
jmasa@981: #ifdef TRACESPINNING
jmasa@981: void ParallelTaskTerminator::print_termination_counts() {
jmasa@981:   gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld  "
jmasa@981:     "Total spins: %lld  Total peeks: %lld",
jmasa@981:     total_yields(),
jmasa@981:     total_spins(),
jmasa@981:     total_peeks());
jmasa@981: }
jmasa@981: #endif
jmasa@981: 
duke@435: void ParallelTaskTerminator::reset_for_reuse() {
duke@435:   if (_offered_termination != 0) {
duke@435:     assert(_offered_termination == _n_threads,
duke@435:            "Terminator may still be in use");
duke@435:     _offered_termination = 0;
duke@435:   }
duke@435: }
duke@435: 
jcoomes@1746: #ifdef ASSERT
jcoomes@1746: bool ObjArrayTask::is_valid() const {
jcoomes@1746:   return _obj != NULL && _obj->is_objArray() && _index > 0 &&
jcoomes@1746:     _index < objArrayOop(_obj)->length();
jcoomes@1746: }
jcoomes@1746: #endif // ASSERT