src/share/vm/gc_implementation/g1/g1GCPhaseTimes.cpp

Tue, 28 May 2013 09:32:06 +0200

author
tschatzl
date
Tue, 28 May 2013 09:32:06 +0200
changeset 5204
e72f7eecc96d
parent 5078
194f52aa2f23
child 5548
5888334c9c24
permissions
-rw-r--r--

8013895: G1: G1SummarizeRSetStats output on Linux needs improvemen
Summary: Fixed the output of G1SummarizeRSetStats: too small datatype for the number of concurrently processed cards, added concurrent remembered set thread time retrieval for Linux and Windows (BSD uses os::elapsedTime() now), and other cleanup. The information presented during VM operation is now relative to the previous output, not always cumulative if G1SummarizeRSetStatsPeriod > 0. At VM exit, the code prints a cumulative summary.
Reviewed-by: johnc, jwilhelm

     1 /*
     2  * Copyright (c) 2013, 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  */
    26 #include "precompiled.hpp"
    27 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
    28 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
    29 #include "gc_implementation/g1/g1Log.hpp"
    31 // Helper class for avoiding interleaved logging
    32 class LineBuffer: public StackObj {
    34 private:
    35   static const int BUFFER_LEN = 1024;
    36   static const int INDENT_CHARS = 3;
    37   char _buffer[BUFFER_LEN];
    38   int _indent_level;
    39   int _cur;
    41   void vappend(const char* format, va_list ap) {
    42     int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
    43     if (res != -1) {
    44       _cur += res;
    45     } else {
    46       DEBUG_ONLY(warning("buffer too small in LineBuffer");)
    47       _buffer[BUFFER_LEN -1] = 0;
    48       _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
    49     }
    50   }
    52 public:
    53   explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
    54     for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
    55       _buffer[_cur] = ' ';
    56     }
    57   }
    59 #ifndef PRODUCT
    60   ~LineBuffer() {
    61     assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
    62   }
    63 #endif
    65   void append(const char* format, ...) {
    66     va_list ap;
    67     va_start(ap, format);
    68     vappend(format, ap);
    69     va_end(ap);
    70   }
    72   void append_and_print_cr(const char* format, ...) {
    73     va_list ap;
    74     va_start(ap, format);
    75     vappend(format, ap);
    76     va_end(ap);
    77     gclog_or_tty->print_cr("%s", _buffer);
    78     _cur = _indent_level * INDENT_CHARS;
    79   }
    80 };
    82 template <class T>
    83 void WorkerDataArray<T>::print(int level, const char* title) {
    84   if (_length == 1) {
    85     // No need for min, max, average and sum for only one worker
    86     LineBuffer buf(level);
    87     buf.append("[%s:  ", title);
    88     buf.append(_print_format, _data[0]);
    89     buf.append_and_print_cr("]");
    90     return;
    91   }
    93   T min = _data[0];
    94   T max = _data[0];
    95   T sum = 0;
    97   LineBuffer buf(level);
    98   buf.append("[%s:", title);
    99   for (uint i = 0; i < _length; ++i) {
   100     T val = _data[i];
   101     min = MIN2(val, min);
   102     max = MAX2(val, max);
   103     sum += val;
   104     if (G1Log::finest()) {
   105       buf.append("  ");
   106       buf.append(_print_format, val);
   107     }
   108   }
   110   if (G1Log::finest()) {
   111     buf.append_and_print_cr("");
   112   }
   114   double avg = (double)sum / (double)_length;
   115   buf.append(" Min: ");
   116   buf.append(_print_format, min);
   117   buf.append(", Avg: ");
   118   buf.append("%.1lf", avg); // Always print average as a double
   119   buf.append(", Max: ");
   120   buf.append(_print_format, max);
   121   buf.append(", Diff: ");
   122   buf.append(_print_format, max - min);
   123   if (_print_sum) {
   124     // for things like the start and end times the sum is not
   125     // that relevant
   126     buf.append(", Sum: ");
   127     buf.append(_print_format, sum);
   128   }
   129   buf.append_and_print_cr("]");
   130 }
   132 #ifndef PRODUCT
   134 template <> const int WorkerDataArray<int>::_uninitialized = -1;
   135 template <> const double WorkerDataArray<double>::_uninitialized = -1.0;
   136 template <> const size_t WorkerDataArray<size_t>::_uninitialized = (size_t)-1;
   138 template <class T>
   139 void WorkerDataArray<T>::reset() {
   140   for (uint i = 0; i < _length; i++) {
   141     _data[i] = (T)_uninitialized;
   142   }
   143 }
   145 template <class T>
   146 void WorkerDataArray<T>::verify() {
   147   for (uint i = 0; i < _length; i++) {
   148     assert(_data[i] != _uninitialized,
   149         err_msg("Invalid data for worker " UINT32_FORMAT ", data: %lf, uninitialized: %lf",
   150             i, (double)_data[i], (double)_uninitialized));
   151   }
   152 }
   154 #endif
   156 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
   157   _max_gc_threads(max_gc_threads),
   158   _last_gc_worker_start_times_ms(_max_gc_threads, "%.1lf", false),
   159   _last_ext_root_scan_times_ms(_max_gc_threads, "%.1lf"),
   160   _last_satb_filtering_times_ms(_max_gc_threads, "%.1lf"),
   161   _last_update_rs_times_ms(_max_gc_threads, "%.1lf"),
   162   _last_update_rs_processed_buffers(_max_gc_threads, "%d"),
   163   _last_scan_rs_times_ms(_max_gc_threads, "%.1lf"),
   164   _last_obj_copy_times_ms(_max_gc_threads, "%.1lf"),
   165   _last_termination_times_ms(_max_gc_threads, "%.1lf"),
   166   _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
   167   _last_gc_worker_end_times_ms(_max_gc_threads, "%.1lf", false),
   168   _last_gc_worker_times_ms(_max_gc_threads, "%.1lf"),
   169   _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf")
   170 {
   171   assert(max_gc_threads > 0, "Must have some GC threads");
   172 }
   174 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
   175   assert(active_gc_threads > 0, "The number of threads must be > 0");
   176   assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max nubmer of threads");
   177   _active_gc_threads = active_gc_threads;
   179   _last_gc_worker_start_times_ms.reset();
   180   _last_ext_root_scan_times_ms.reset();
   181   _last_satb_filtering_times_ms.reset();
   182   _last_update_rs_times_ms.reset();
   183   _last_update_rs_processed_buffers.reset();
   184   _last_scan_rs_times_ms.reset();
   185   _last_obj_copy_times_ms.reset();
   186   _last_termination_times_ms.reset();
   187   _last_termination_attempts.reset();
   188   _last_gc_worker_end_times_ms.reset();
   189   _last_gc_worker_times_ms.reset();
   190   _last_gc_worker_other_times_ms.reset();
   191 }
   193 void G1GCPhaseTimes::note_gc_end() {
   194   _last_gc_worker_start_times_ms.verify();
   195   _last_ext_root_scan_times_ms.verify();
   196   _last_satb_filtering_times_ms.verify();
   197   _last_update_rs_times_ms.verify();
   198   _last_update_rs_processed_buffers.verify();
   199   _last_scan_rs_times_ms.verify();
   200   _last_obj_copy_times_ms.verify();
   201   _last_termination_times_ms.verify();
   202   _last_termination_attempts.verify();
   203   _last_gc_worker_end_times_ms.verify();
   205   for (uint i = 0; i < _active_gc_threads; i++) {
   206     double worker_time = _last_gc_worker_end_times_ms.get(i) - _last_gc_worker_start_times_ms.get(i);
   207     _last_gc_worker_times_ms.set(i, worker_time);
   209     double worker_known_time = _last_ext_root_scan_times_ms.get(i) +
   210                                _last_satb_filtering_times_ms.get(i) +
   211                                _last_update_rs_times_ms.get(i) +
   212                                _last_scan_rs_times_ms.get(i) +
   213                                _last_obj_copy_times_ms.get(i) +
   214                                _last_termination_times_ms.get(i);
   216     double worker_other_time = worker_time - worker_known_time;
   217     _last_gc_worker_other_times_ms.set(i, worker_other_time);
   218   }
   220   _last_gc_worker_times_ms.verify();
   221   _last_gc_worker_other_times_ms.verify();
   222 }
   224 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
   225   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
   226 }
   228 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, int workers) {
   229   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %d]", str, value, workers);
   230 }
   232 double G1GCPhaseTimes::accounted_time_ms() {
   233     // Subtract the root region scanning wait time. It's initialized to
   234     // zero at the start of the pause.
   235     double misc_time_ms = _root_region_scan_wait_time_ms;
   237     misc_time_ms += _cur_collection_par_time_ms;
   239     // Now subtract the time taken to fix up roots in generated code
   240     misc_time_ms += _cur_collection_code_root_fixup_time_ms;
   242     // Subtract the time taken to clean the card table from the
   243     // current value of "other time"
   244     misc_time_ms += _cur_clear_ct_time_ms;
   246     return misc_time_ms;
   247 }
   249 void G1GCPhaseTimes::print(double pause_time_sec) {
   250   if (_root_region_scan_wait_time_ms > 0.0) {
   251     print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
   252   }
   253   if (G1CollectedHeap::use_parallel_gc_threads()) {
   254     print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
   255     _last_gc_worker_start_times_ms.print(2, "GC Worker Start (ms)");
   256     _last_ext_root_scan_times_ms.print(2, "Ext Root Scanning (ms)");
   257     if (_last_satb_filtering_times_ms.sum() > 0.0) {
   258       _last_satb_filtering_times_ms.print(2, "SATB Filtering (ms)");
   259     }
   260     _last_update_rs_times_ms.print(2, "Update RS (ms)");
   261       _last_update_rs_processed_buffers.print(3, "Processed Buffers");
   262     _last_scan_rs_times_ms.print(2, "Scan RS (ms)");
   263     _last_obj_copy_times_ms.print(2, "Object Copy (ms)");
   264     _last_termination_times_ms.print(2, "Termination (ms)");
   265     if (G1Log::finest()) {
   266       _last_termination_attempts.print(3, "Termination Attempts");
   267     }
   268     _last_gc_worker_other_times_ms.print(2, "GC Worker Other (ms)");
   269     _last_gc_worker_times_ms.print(2, "GC Worker Total (ms)");
   270     _last_gc_worker_end_times_ms.print(2, "GC Worker End (ms)");
   271   } else {
   272     _last_ext_root_scan_times_ms.print(1, "Ext Root Scanning (ms)");
   273     if (_last_satb_filtering_times_ms.sum() > 0.0) {
   274       _last_satb_filtering_times_ms.print(1, "SATB Filtering (ms)");
   275     }
   276     _last_update_rs_times_ms.print(1, "Update RS (ms)");
   277       _last_update_rs_processed_buffers.print(2, "Processed Buffers");
   278     _last_scan_rs_times_ms.print(1, "Scan RS (ms)");
   279     _last_obj_copy_times_ms.print(1, "Object Copy (ms)");
   280   }
   281   print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
   282   print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
   283   double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
   284   print_stats(1, "Other", misc_time_ms);
   285   if (_cur_verify_before_time_ms > 0.0) {
   286     print_stats(2, "Verify Before", _cur_verify_before_time_ms);
   287   }
   288   print_stats(2, "Choose CSet",
   289     (_recorded_young_cset_choice_time_ms +
   290     _recorded_non_young_cset_choice_time_ms));
   291   print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
   292   print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
   293   print_stats(2, "Free CSet",
   294     (_recorded_young_free_cset_time_ms +
   295     _recorded_non_young_free_cset_time_ms));
   296   if (_cur_verify_after_time_ms > 0.0) {
   297     print_stats(2, "Verify After", _cur_verify_after_time_ms);
   298   }
   299 }

mercurial