Tue, 18 Mar 2014 19:07:22 +0100
8029075: String deduplication in G1
Summary: Implementation of JEP 192, http://openjdk.java.net/jeps/192
Reviewed-by: brutisso, tschatzl, coleenp
1 /*
2 * Copyright (c) 2013, 2014 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"
30 #include "gc_implementation/g1/g1StringDedup.hpp"
32 // Helper class for avoiding interleaved logging
33 class LineBuffer: public StackObj {
35 private:
36 static const int BUFFER_LEN = 1024;
37 static const int INDENT_CHARS = 3;
38 char _buffer[BUFFER_LEN];
39 int _indent_level;
40 int _cur;
42 void vappend(const char* format, va_list ap) {
43 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
44 if (res != -1) {
45 _cur += res;
46 } else {
47 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
48 _buffer[BUFFER_LEN -1] = 0;
49 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
50 }
51 }
53 public:
54 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
55 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
56 _buffer[_cur] = ' ';
57 }
58 }
60 #ifndef PRODUCT
61 ~LineBuffer() {
62 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
63 }
64 #endif
66 void append(const char* format, ...) {
67 va_list ap;
68 va_start(ap, format);
69 vappend(format, ap);
70 va_end(ap);
71 }
73 void append_and_print_cr(const char* format, ...) {
74 va_list ap;
75 va_start(ap, format);
76 vappend(format, ap);
77 va_end(ap);
78 gclog_or_tty->print_cr("%s", _buffer);
79 _cur = _indent_level * INDENT_CHARS;
80 }
81 };
83 template <class T>
84 void WorkerDataArray<T>::print(int level, const char* title) {
85 if (_length == 1) {
86 // No need for min, max, average and sum for only one worker
87 LineBuffer buf(level);
88 buf.append("[%s: ", title);
89 buf.append(_print_format, _data[0]);
90 buf.append_and_print_cr("]");
91 return;
92 }
94 T min = _data[0];
95 T max = _data[0];
96 T sum = 0;
98 LineBuffer buf(level);
99 buf.append("[%s:", title);
100 for (uint i = 0; i < _length; ++i) {
101 T val = _data[i];
102 min = MIN2(val, min);
103 max = MAX2(val, max);
104 sum += val;
105 if (G1Log::finest()) {
106 buf.append(" ");
107 buf.append(_print_format, val);
108 }
109 }
111 if (G1Log::finest()) {
112 buf.append_and_print_cr("");
113 }
115 double avg = (double)sum / (double)_length;
116 buf.append(" Min: ");
117 buf.append(_print_format, min);
118 buf.append(", Avg: ");
119 buf.append("%.1lf", avg); // Always print average as a double
120 buf.append(", Max: ");
121 buf.append(_print_format, max);
122 buf.append(", Diff: ");
123 buf.append(_print_format, max - min);
124 if (_print_sum) {
125 // for things like the start and end times the sum is not
126 // that relevant
127 buf.append(", Sum: ");
128 buf.append(_print_format, sum);
129 }
130 buf.append_and_print_cr("]");
131 }
133 #ifndef PRODUCT
135 template <> const int WorkerDataArray<int>::_uninitialized = -1;
136 template <> const double WorkerDataArray<double>::_uninitialized = -1.0;
137 template <> const size_t WorkerDataArray<size_t>::_uninitialized = (size_t)-1;
139 template <class T>
140 void WorkerDataArray<T>::reset() {
141 for (uint i = 0; i < _length; i++) {
142 _data[i] = (T)_uninitialized;
143 }
144 }
146 template <class T>
147 void WorkerDataArray<T>::verify() {
148 for (uint i = 0; i < _length; i++) {
149 assert(_data[i] != _uninitialized,
150 err_msg("Invalid data for worker " UINT32_FORMAT ", data: %lf, uninitialized: %lf",
151 i, (double)_data[i], (double)_uninitialized));
152 }
153 }
155 #endif
157 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
158 _max_gc_threads(max_gc_threads),
159 _last_gc_worker_start_times_ms(_max_gc_threads, "%.1lf", false),
160 _last_ext_root_scan_times_ms(_max_gc_threads, "%.1lf"),
161 _last_satb_filtering_times_ms(_max_gc_threads, "%.1lf"),
162 _last_update_rs_times_ms(_max_gc_threads, "%.1lf"),
163 _last_update_rs_processed_buffers(_max_gc_threads, "%d"),
164 _last_scan_rs_times_ms(_max_gc_threads, "%.1lf"),
165 _last_strong_code_root_scan_times_ms(_max_gc_threads, "%.1lf"),
166 _last_strong_code_root_mark_times_ms(_max_gc_threads, "%.1lf"),
167 _last_obj_copy_times_ms(_max_gc_threads, "%.1lf"),
168 _last_termination_times_ms(_max_gc_threads, "%.1lf"),
169 _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
170 _last_gc_worker_end_times_ms(_max_gc_threads, "%.1lf", false),
171 _last_gc_worker_times_ms(_max_gc_threads, "%.1lf"),
172 _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf"),
173 _cur_string_dedup_queue_fixup_worker_times_ms(_max_gc_threads, "%.1lf"),
174 _cur_string_dedup_table_fixup_worker_times_ms(_max_gc_threads, "%.1lf")
175 {
176 assert(max_gc_threads > 0, "Must have some GC threads");
177 }
179 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
180 assert(active_gc_threads > 0, "The number of threads must be > 0");
181 assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max nubmer of threads");
182 _active_gc_threads = active_gc_threads;
184 _last_gc_worker_start_times_ms.reset();
185 _last_ext_root_scan_times_ms.reset();
186 _last_satb_filtering_times_ms.reset();
187 _last_update_rs_times_ms.reset();
188 _last_update_rs_processed_buffers.reset();
189 _last_scan_rs_times_ms.reset();
190 _last_strong_code_root_scan_times_ms.reset();
191 _last_strong_code_root_mark_times_ms.reset();
192 _last_obj_copy_times_ms.reset();
193 _last_termination_times_ms.reset();
194 _last_termination_attempts.reset();
195 _last_gc_worker_end_times_ms.reset();
196 _last_gc_worker_times_ms.reset();
197 _last_gc_worker_other_times_ms.reset();
198 }
200 void G1GCPhaseTimes::note_gc_end() {
201 _last_gc_worker_start_times_ms.verify();
202 _last_ext_root_scan_times_ms.verify();
203 _last_satb_filtering_times_ms.verify();
204 _last_update_rs_times_ms.verify();
205 _last_update_rs_processed_buffers.verify();
206 _last_scan_rs_times_ms.verify();
207 _last_strong_code_root_scan_times_ms.verify();
208 _last_strong_code_root_mark_times_ms.verify();
209 _last_obj_copy_times_ms.verify();
210 _last_termination_times_ms.verify();
211 _last_termination_attempts.verify();
212 _last_gc_worker_end_times_ms.verify();
214 for (uint i = 0; i < _active_gc_threads; i++) {
215 double worker_time = _last_gc_worker_end_times_ms.get(i) - _last_gc_worker_start_times_ms.get(i);
216 _last_gc_worker_times_ms.set(i, worker_time);
218 double worker_known_time = _last_ext_root_scan_times_ms.get(i) +
219 _last_satb_filtering_times_ms.get(i) +
220 _last_update_rs_times_ms.get(i) +
221 _last_scan_rs_times_ms.get(i) +
222 _last_strong_code_root_scan_times_ms.get(i) +
223 _last_strong_code_root_mark_times_ms.get(i) +
224 _last_obj_copy_times_ms.get(i) +
225 _last_termination_times_ms.get(i);
227 double worker_other_time = worker_time - worker_known_time;
228 _last_gc_worker_other_times_ms.set(i, worker_other_time);
229 }
231 _last_gc_worker_times_ms.verify();
232 _last_gc_worker_other_times_ms.verify();
233 }
235 void G1GCPhaseTimes::note_string_dedup_fixup_start() {
236 _cur_string_dedup_queue_fixup_worker_times_ms.reset();
237 _cur_string_dedup_table_fixup_worker_times_ms.reset();
238 }
240 void G1GCPhaseTimes::note_string_dedup_fixup_end() {
241 _cur_string_dedup_queue_fixup_worker_times_ms.verify();
242 _cur_string_dedup_table_fixup_worker_times_ms.verify();
243 }
245 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
246 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
247 }
249 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, int workers) {
250 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %d]", str, value, workers);
251 }
253 double G1GCPhaseTimes::accounted_time_ms() {
254 // Subtract the root region scanning wait time. It's initialized to
255 // zero at the start of the pause.
256 double misc_time_ms = _root_region_scan_wait_time_ms;
258 misc_time_ms += _cur_collection_par_time_ms;
260 // Now subtract the time taken to fix up roots in generated code
261 misc_time_ms += _cur_collection_code_root_fixup_time_ms;
263 // Strong code root migration time
264 misc_time_ms += _cur_strong_code_root_migration_time_ms;
266 // Strong code root purge time
267 misc_time_ms += _cur_strong_code_root_purge_time_ms;
269 if (G1StringDedup::is_enabled()) {
270 // String dedup fixup time
271 misc_time_ms += _cur_string_dedup_fixup_time_ms;
272 }
274 // Subtract the time taken to clean the card table from the
275 // current value of "other time"
276 misc_time_ms += _cur_clear_ct_time_ms;
278 return misc_time_ms;
279 }
281 void G1GCPhaseTimes::print(double pause_time_sec) {
282 if (_root_region_scan_wait_time_ms > 0.0) {
283 print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
284 }
285 if (G1CollectedHeap::use_parallel_gc_threads()) {
286 print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
287 _last_gc_worker_start_times_ms.print(2, "GC Worker Start (ms)");
288 _last_ext_root_scan_times_ms.print(2, "Ext Root Scanning (ms)");
289 if (_last_satb_filtering_times_ms.sum() > 0.0) {
290 _last_satb_filtering_times_ms.print(2, "SATB Filtering (ms)");
291 }
292 if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
293 _last_strong_code_root_mark_times_ms.print(2, "Code Root Marking (ms)");
294 }
295 _last_update_rs_times_ms.print(2, "Update RS (ms)");
296 _last_update_rs_processed_buffers.print(3, "Processed Buffers");
297 _last_scan_rs_times_ms.print(2, "Scan RS (ms)");
298 _last_strong_code_root_scan_times_ms.print(2, "Code Root Scanning (ms)");
299 _last_obj_copy_times_ms.print(2, "Object Copy (ms)");
300 _last_termination_times_ms.print(2, "Termination (ms)");
301 if (G1Log::finest()) {
302 _last_termination_attempts.print(3, "Termination Attempts");
303 }
304 _last_gc_worker_other_times_ms.print(2, "GC Worker Other (ms)");
305 _last_gc_worker_times_ms.print(2, "GC Worker Total (ms)");
306 _last_gc_worker_end_times_ms.print(2, "GC Worker End (ms)");
307 } else {
308 _last_ext_root_scan_times_ms.print(1, "Ext Root Scanning (ms)");
309 if (_last_satb_filtering_times_ms.sum() > 0.0) {
310 _last_satb_filtering_times_ms.print(1, "SATB Filtering (ms)");
311 }
312 if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
313 _last_strong_code_root_mark_times_ms.print(1, "Code Root Marking (ms)");
314 }
315 _last_update_rs_times_ms.print(1, "Update RS (ms)");
316 _last_update_rs_processed_buffers.print(2, "Processed Buffers");
317 _last_scan_rs_times_ms.print(1, "Scan RS (ms)");
318 _last_strong_code_root_scan_times_ms.print(1, "Code Root Scanning (ms)");
319 _last_obj_copy_times_ms.print(1, "Object Copy (ms)");
320 }
321 print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
322 print_stats(1, "Code Root Migration", _cur_strong_code_root_migration_time_ms);
323 print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
324 if (G1StringDedup::is_enabled()) {
325 print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
326 _cur_string_dedup_queue_fixup_worker_times_ms.print(2, "Queue Fixup (ms)");
327 _cur_string_dedup_table_fixup_worker_times_ms.print(2, "Table Fixup (ms)");
328 }
329 print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
330 double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
331 print_stats(1, "Other", misc_time_ms);
332 if (_cur_verify_before_time_ms > 0.0) {
333 print_stats(2, "Verify Before", _cur_verify_before_time_ms);
334 }
335 if (G1CollectedHeap::heap()->evacuation_failed()) {
336 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
337 _cur_evac_fail_restore_remsets;
338 print_stats(2, "Evacuation Failure", evac_fail_handling);
339 if (G1Log::finest()) {
340 print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
341 print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
342 print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
343 }
344 }
345 print_stats(2, "Choose CSet",
346 (_recorded_young_cset_choice_time_ms +
347 _recorded_non_young_cset_choice_time_ms));
348 print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
349 print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
350 if (G1DeferredRSUpdate) {
351 print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
352 }
353 print_stats(2, "Free CSet",
354 (_recorded_young_free_cset_time_ms +
355 _recorded_non_young_free_cset_time_ms));
356 if (G1Log::finest()) {
357 print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
358 print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
359 }
360 if (_cur_verify_after_time_ms > 0.0) {
361 print_stats(2, "Verify After", _cur_verify_after_time_ms);
362 }
363 }