1/*
2 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/*
30 This file has been modified by Kevin Moore and Dan Nussbaum of the
31 Scalable Systems Research Group at Sun Microsystems Laboratories
32 (http://research.sun.com/scalable/) to support the Adaptive
33 Transactional Memory Test Platform (ATMTP).
34
35 Please send email to atmtp-interest@sun.com with feedback, questions, or
36 to request future announcements about ATMTP.
37
38 ----------------------------------------------------------------------
39
40 File modification date: 2008-02-23
41
42 ----------------------------------------------------------------------
43*/
44
45// Allows use of times() library call, which determines virtual runtime
46#include <sys/resource.h>
47#include <sys/times.h>
48
49#include <algorithm>
50
51#include "base/stl_helpers.hh"
52#include "base/str.hh"
| 1/*
2 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/*
30 This file has been modified by Kevin Moore and Dan Nussbaum of the
31 Scalable Systems Research Group at Sun Microsystems Laboratories
32 (http://research.sun.com/scalable/) to support the Adaptive
33 Transactional Memory Test Platform (ATMTP).
34
35 Please send email to atmtp-interest@sun.com with feedback, questions, or
36 to request future announcements about ATMTP.
37
38 ----------------------------------------------------------------------
39
40 File modification date: 2008-02-23
41
42 ----------------------------------------------------------------------
43*/
44
45// Allows use of times() library call, which determines virtual runtime
46#include <sys/resource.h>
47#include <sys/times.h>
48
49#include <algorithm>
50
51#include "base/stl_helpers.hh"
52#include "base/str.hh"
|
53#include "mem/gems_common/Map.hh"
| |
54#include "mem/gems_common/PrioHeap.hh"
55#include "mem/protocol/CacheMsg.hh"
56#include "mem/protocol/MachineType.hh"
57#include "mem/protocol/Protocol.hh"
58#include "mem/ruby/common/Debug.hh"
59#include "mem/ruby/network/Network.hh"
60#include "mem/ruby/profiler/AddressProfiler.hh"
61#include "mem/ruby/profiler/Profiler.hh"
62#include "mem/ruby/system/System.hh"
63#include "mem/ruby/system/System.hh"
64
65using namespace std;
66using m5::stl_helpers::operator<<;
67
68extern ostream* debug_cout_ptr;
69
70static double process_memory_total();
71static double process_memory_resident();
72
73Profiler::Profiler(const Params *p)
74 : SimObject(p)
75{
| 53#include "mem/gems_common/PrioHeap.hh"
54#include "mem/protocol/CacheMsg.hh"
55#include "mem/protocol/MachineType.hh"
56#include "mem/protocol/Protocol.hh"
57#include "mem/ruby/common/Debug.hh"
58#include "mem/ruby/network/Network.hh"
59#include "mem/ruby/profiler/AddressProfiler.hh"
60#include "mem/ruby/profiler/Profiler.hh"
61#include "mem/ruby/system/System.hh"
62#include "mem/ruby/system/System.hh"
63
64using namespace std;
65using m5::stl_helpers::operator<<;
66
67extern ostream* debug_cout_ptr;
68
69static double process_memory_total();
70static double process_memory_resident();
71
72Profiler::Profiler(const Params *p)
73 : SimObject(p)
74{
|
76 m_requestProfileMap_ptr = new Map<string, int>;
77
| |
78 m_inst_profiler_ptr = NULL;
79 m_address_profiler_ptr = NULL;
80
81 m_real_time_start_time = time(NULL); // Not reset in clearStats()
82 m_stats_period = 1000000; // Default
83 m_periodic_output_file_ptr = &cerr;
84
85 m_hot_lines = p->hot_lines;
86 m_all_instructions = p->all_instructions;
87
88 m_num_of_sequencers = p->num_of_sequencers;
89
90 m_hot_lines = false;
91 m_all_instructions = false;
92
93 m_address_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
94 m_address_profiler_ptr->setHotLines(m_hot_lines);
95 m_address_profiler_ptr->setAllInstructions(m_all_instructions);
96
97 if (m_all_instructions) {
98 m_inst_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
99 m_inst_profiler_ptr->setHotLines(m_hot_lines);
100 m_inst_profiler_ptr->setAllInstructions(m_all_instructions);
101 }
102}
103
104Profiler::~Profiler()
105{
106 if (m_periodic_output_file_ptr != &cerr) {
107 delete m_periodic_output_file_ptr;
108 }
| 75 m_inst_profiler_ptr = NULL;
76 m_address_profiler_ptr = NULL;
77
78 m_real_time_start_time = time(NULL); // Not reset in clearStats()
79 m_stats_period = 1000000; // Default
80 m_periodic_output_file_ptr = &cerr;
81
82 m_hot_lines = p->hot_lines;
83 m_all_instructions = p->all_instructions;
84
85 m_num_of_sequencers = p->num_of_sequencers;
86
87 m_hot_lines = false;
88 m_all_instructions = false;
89
90 m_address_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
91 m_address_profiler_ptr->setHotLines(m_hot_lines);
92 m_address_profiler_ptr->setAllInstructions(m_all_instructions);
93
94 if (m_all_instructions) {
95 m_inst_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
96 m_inst_profiler_ptr->setHotLines(m_hot_lines);
97 m_inst_profiler_ptr->setAllInstructions(m_all_instructions);
98 }
99}
100
101Profiler::~Profiler()
102{
103 if (m_periodic_output_file_ptr != &cerr) {
104 delete m_periodic_output_file_ptr;
105 }
|
109
110 delete m_requestProfileMap_ptr;
| |
111}
112
113void
114Profiler::wakeup()
115{
116 // FIXME - avoid the repeated code
117
118 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
119
120 for (int i = 0; i < m_num_of_sequencers; i++) {
121 perProcCycleCount[i] =
122 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
123 // The +1 allows us to avoid division by zero
124 }
125
126 ostream &out = *m_periodic_output_file_ptr;
127
128 out << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl
129 << "mbytes_resident: " << process_memory_resident() << endl
130 << "mbytes_total: " << process_memory_total() << endl;
131
132 if (process_memory_total() > 0) {
133 out << "resident_ratio: "
134 << process_memory_resident() / process_memory_total() << endl;
135 }
136
137 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
138
139 out << endl;
140
141 if (m_all_instructions) {
142 m_inst_profiler_ptr->printStats(out);
143 }
144
145 //g_system_ptr->getNetwork()->printStats(out);
146 g_eventQueue_ptr->scheduleEvent(this, m_stats_period);
147}
148
149void
150Profiler::setPeriodicStatsFile(const string& filename)
151{
152 cout << "Recording periodic statistics to file '" << filename << "' every "
153 << m_stats_period << " Ruby cycles" << endl;
154
155 if (m_periodic_output_file_ptr != &cerr) {
156 delete m_periodic_output_file_ptr;
157 }
158
159 m_periodic_output_file_ptr = new ofstream(filename.c_str());
160 g_eventQueue_ptr->scheduleEvent(this, 1);
161}
162
163void
164Profiler::setPeriodicStatsInterval(integer_t period)
165{
166 cout << "Recording periodic statistics every " << m_stats_period
167 << " Ruby cycles" << endl;
168
169 m_stats_period = period;
170 g_eventQueue_ptr->scheduleEvent(this, 1);
171}
172
173void
174Profiler::printConfig(ostream& out) const
175{
176 out << endl;
177 out << "Profiler Configuration" << endl;
178 out << "----------------------" << endl;
179 out << "periodic_stats_period: " << m_stats_period << endl;
180}
181
182void
183Profiler::print(ostream& out) const
184{
185 out << "[Profiler]";
186}
187
188void
189Profiler::printStats(ostream& out, bool short_stats)
190{
191 out << endl;
192 if (short_stats) {
193 out << "SHORT ";
194 }
195 out << "Profiler Stats" << endl;
196 out << "--------------" << endl;
197
198 time_t real_time_current = time(NULL);
199 double seconds = difftime(real_time_current, m_real_time_start_time);
200 double minutes = seconds / 60.0;
201 double hours = minutes / 60.0;
202 double days = hours / 24.0;
203 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
204
205 if (!short_stats) {
206 out << "Elapsed_time_in_seconds: " << seconds << endl;
207 out << "Elapsed_time_in_minutes: " << minutes << endl;
208 out << "Elapsed_time_in_hours: " << hours << endl;
209 out << "Elapsed_time_in_days: " << days << endl;
210 out << endl;
211 }
212
213 // print the virtual runtimes as well
214 struct tms vtime;
215 times(&vtime);
216 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
217 minutes = seconds / 60.0;
218 hours = minutes / 60.0;
219 days = hours / 24.0;
220 out << "Virtual_time_in_seconds: " << seconds << endl;
221 out << "Virtual_time_in_minutes: " << minutes << endl;
222 out << "Virtual_time_in_hours: " << hours << endl;
223 out << "Virtual_time_in_days: " << days << endl;
224 out << endl;
225
226 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
227 out << "Ruby_start_time: " << m_ruby_start << endl;
228 out << "Ruby_cycles: " << ruby_cycles << endl;
229 out << endl;
230
231 if (!short_stats) {
232 out << "mbytes_resident: " << process_memory_resident() << endl;
233 out << "mbytes_total: " << process_memory_total() << endl;
234 if (process_memory_total() > 0) {
235 out << "resident_ratio: "
236 << process_memory_resident()/process_memory_total() << endl;
237 }
238 out << endl;
239 }
240
241 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
242
243 for (int i = 0; i < m_num_of_sequencers; i++) {
244 perProcCycleCount[i] =
245 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
246 // The +1 allows us to avoid division by zero
247 }
248
249 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
250
251 out << endl;
252
253 if (!short_stats) {
254 out << "Busy Controller Counts:" << endl;
255 for (int i = 0; i < MachineType_NUM; i++) {
256 int size = MachineType_base_count((MachineType)i);
257 for (int j = 0; j < size; j++) {
258 MachineID machID;
259 machID.type = (MachineType)i;
260 machID.num = j;
261 out << machID << ":" << m_busyControllerCount[i][j] << " ";
262 if ((j + 1) % 8 == 0) {
263 out << endl;
264 }
265 }
266 out << endl;
267 }
268 out << endl;
269
270 out << "Busy Bank Count:" << m_busyBankCount << endl;
271 out << endl;
272
273 out << "sequencer_requests_outstanding: "
274 << m_sequencer_requests << endl;
275 out << endl;
276 }
277
278 if (!short_stats) {
279 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
280 out << "----------------------------------------" << endl;
281 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
282 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
283 if (m_missLatencyHistograms[i].size() > 0) {
284 out << "miss_latency_" << RubyRequestType(i) << ": "
285 << m_missLatencyHistograms[i] << endl;
286 }
287 }
288 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
289 if (m_machLatencyHistograms[i].size() > 0) {
290 out << "miss_latency_" << GenericMachineType(i) << ": "
291 << m_machLatencyHistograms[i] << endl;
292 }
293 }
294
295 out << endl;
296
297 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
298 out << "------------------------------------" << endl;
299 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
300 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
301 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
302 out << "prefetch_latency_" << CacheRequestType(i) << ": "
303 << m_SWPrefetchLatencyHistograms[i] << endl;
304 }
305 }
306 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
307 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
308 out << "prefetch_latency_" << GenericMachineType(i) << ": "
309 << m_SWPrefetchMachLatencyHistograms[i] << endl;
310 }
311 }
312 out << "prefetch_latency_L2Miss:"
313 << m_SWPrefetchL2MissLatencyHistogram << endl;
314
315 if (m_all_sharing_histogram.size() > 0) {
316 out << "all_sharing: " << m_all_sharing_histogram << endl;
317 out << "read_sharing: " << m_read_sharing_histogram << endl;
318 out << "write_sharing: " << m_write_sharing_histogram << endl;
319
320 out << "all_sharing_percent: ";
321 m_all_sharing_histogram.printPercent(out);
322 out << endl;
323
324 out << "read_sharing_percent: ";
325 m_read_sharing_histogram.printPercent(out);
326 out << endl;
327
328 out << "write_sharing_percent: ";
329 m_write_sharing_histogram.printPercent(out);
330 out << endl;
331
332 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
333 out << "all_misses: " << total_miss << endl;
334 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
335 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
336 out << "cache_to_cache_percent: "
337 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
338 << endl;
339 out << "memory_to_cache_percent: "
340 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
341 << endl;
342 out << endl;
343 }
344
345 if (m_outstanding_requests.size() > 0) {
346 out << "outstanding_requests: ";
347 m_outstanding_requests.printPercent(out);
348 out << endl;
349 out << endl;
350 }
351 }
352
353 if (!short_stats) {
354 out << "Request vs. RubySystem State Profile" << endl;
355 out << "--------------------------------" << endl;
356 out << endl;
357
| 106}
107
108void
109Profiler::wakeup()
110{
111 // FIXME - avoid the repeated code
112
113 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
114
115 for (int i = 0; i < m_num_of_sequencers; i++) {
116 perProcCycleCount[i] =
117 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
118 // The +1 allows us to avoid division by zero
119 }
120
121 ostream &out = *m_periodic_output_file_ptr;
122
123 out << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl
124 << "mbytes_resident: " << process_memory_resident() << endl
125 << "mbytes_total: " << process_memory_total() << endl;
126
127 if (process_memory_total() > 0) {
128 out << "resident_ratio: "
129 << process_memory_resident() / process_memory_total() << endl;
130 }
131
132 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
133
134 out << endl;
135
136 if (m_all_instructions) {
137 m_inst_profiler_ptr->printStats(out);
138 }
139
140 //g_system_ptr->getNetwork()->printStats(out);
141 g_eventQueue_ptr->scheduleEvent(this, m_stats_period);
142}
143
144void
145Profiler::setPeriodicStatsFile(const string& filename)
146{
147 cout << "Recording periodic statistics to file '" << filename << "' every "
148 << m_stats_period << " Ruby cycles" << endl;
149
150 if (m_periodic_output_file_ptr != &cerr) {
151 delete m_periodic_output_file_ptr;
152 }
153
154 m_periodic_output_file_ptr = new ofstream(filename.c_str());
155 g_eventQueue_ptr->scheduleEvent(this, 1);
156}
157
158void
159Profiler::setPeriodicStatsInterval(integer_t period)
160{
161 cout << "Recording periodic statistics every " << m_stats_period
162 << " Ruby cycles" << endl;
163
164 m_stats_period = period;
165 g_eventQueue_ptr->scheduleEvent(this, 1);
166}
167
168void
169Profiler::printConfig(ostream& out) const
170{
171 out << endl;
172 out << "Profiler Configuration" << endl;
173 out << "----------------------" << endl;
174 out << "periodic_stats_period: " << m_stats_period << endl;
175}
176
177void
178Profiler::print(ostream& out) const
179{
180 out << "[Profiler]";
181}
182
183void
184Profiler::printStats(ostream& out, bool short_stats)
185{
186 out << endl;
187 if (short_stats) {
188 out << "SHORT ";
189 }
190 out << "Profiler Stats" << endl;
191 out << "--------------" << endl;
192
193 time_t real_time_current = time(NULL);
194 double seconds = difftime(real_time_current, m_real_time_start_time);
195 double minutes = seconds / 60.0;
196 double hours = minutes / 60.0;
197 double days = hours / 24.0;
198 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
199
200 if (!short_stats) {
201 out << "Elapsed_time_in_seconds: " << seconds << endl;
202 out << "Elapsed_time_in_minutes: " << minutes << endl;
203 out << "Elapsed_time_in_hours: " << hours << endl;
204 out << "Elapsed_time_in_days: " << days << endl;
205 out << endl;
206 }
207
208 // print the virtual runtimes as well
209 struct tms vtime;
210 times(&vtime);
211 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
212 minutes = seconds / 60.0;
213 hours = minutes / 60.0;
214 days = hours / 24.0;
215 out << "Virtual_time_in_seconds: " << seconds << endl;
216 out << "Virtual_time_in_minutes: " << minutes << endl;
217 out << "Virtual_time_in_hours: " << hours << endl;
218 out << "Virtual_time_in_days: " << days << endl;
219 out << endl;
220
221 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
222 out << "Ruby_start_time: " << m_ruby_start << endl;
223 out << "Ruby_cycles: " << ruby_cycles << endl;
224 out << endl;
225
226 if (!short_stats) {
227 out << "mbytes_resident: " << process_memory_resident() << endl;
228 out << "mbytes_total: " << process_memory_total() << endl;
229 if (process_memory_total() > 0) {
230 out << "resident_ratio: "
231 << process_memory_resident()/process_memory_total() << endl;
232 }
233 out << endl;
234 }
235
236 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
237
238 for (int i = 0; i < m_num_of_sequencers; i++) {
239 perProcCycleCount[i] =
240 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
241 // The +1 allows us to avoid division by zero
242 }
243
244 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
245
246 out << endl;
247
248 if (!short_stats) {
249 out << "Busy Controller Counts:" << endl;
250 for (int i = 0; i < MachineType_NUM; i++) {
251 int size = MachineType_base_count((MachineType)i);
252 for (int j = 0; j < size; j++) {
253 MachineID machID;
254 machID.type = (MachineType)i;
255 machID.num = j;
256 out << machID << ":" << m_busyControllerCount[i][j] << " ";
257 if ((j + 1) % 8 == 0) {
258 out << endl;
259 }
260 }
261 out << endl;
262 }
263 out << endl;
264
265 out << "Busy Bank Count:" << m_busyBankCount << endl;
266 out << endl;
267
268 out << "sequencer_requests_outstanding: "
269 << m_sequencer_requests << endl;
270 out << endl;
271 }
272
273 if (!short_stats) {
274 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
275 out << "----------------------------------------" << endl;
276 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
277 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
278 if (m_missLatencyHistograms[i].size() > 0) {
279 out << "miss_latency_" << RubyRequestType(i) << ": "
280 << m_missLatencyHistograms[i] << endl;
281 }
282 }
283 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
284 if (m_machLatencyHistograms[i].size() > 0) {
285 out << "miss_latency_" << GenericMachineType(i) << ": "
286 << m_machLatencyHistograms[i] << endl;
287 }
288 }
289
290 out << endl;
291
292 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
293 out << "------------------------------------" << endl;
294 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
295 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
296 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
297 out << "prefetch_latency_" << CacheRequestType(i) << ": "
298 << m_SWPrefetchLatencyHistograms[i] << endl;
299 }
300 }
301 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
302 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
303 out << "prefetch_latency_" << GenericMachineType(i) << ": "
304 << m_SWPrefetchMachLatencyHistograms[i] << endl;
305 }
306 }
307 out << "prefetch_latency_L2Miss:"
308 << m_SWPrefetchL2MissLatencyHistogram << endl;
309
310 if (m_all_sharing_histogram.size() > 0) {
311 out << "all_sharing: " << m_all_sharing_histogram << endl;
312 out << "read_sharing: " << m_read_sharing_histogram << endl;
313 out << "write_sharing: " << m_write_sharing_histogram << endl;
314
315 out << "all_sharing_percent: ";
316 m_all_sharing_histogram.printPercent(out);
317 out << endl;
318
319 out << "read_sharing_percent: ";
320 m_read_sharing_histogram.printPercent(out);
321 out << endl;
322
323 out << "write_sharing_percent: ";
324 m_write_sharing_histogram.printPercent(out);
325 out << endl;
326
327 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
328 out << "all_misses: " << total_miss << endl;
329 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
330 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
331 out << "cache_to_cache_percent: "
332 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
333 << endl;
334 out << "memory_to_cache_percent: "
335 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
336 << endl;
337 out << endl;
338 }
339
340 if (m_outstanding_requests.size() > 0) {
341 out << "outstanding_requests: ";
342 m_outstanding_requests.printPercent(out);
343 out << endl;
344 out << endl;
345 }
346 }
347
348 if (!short_stats) {
349 out << "Request vs. RubySystem State Profile" << endl;
350 out << "--------------------------------" << endl;
351 out << endl;
352
|
358 vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys();
359 sort(requestProfileKeys.begin(), requestProfileKeys.end());
| 353 map<string, int>::const_iterator i = m_requestProfileMap.begin();
354 map<string, int>::const_iterator end = m_requestProfileMap.end();
355 for (; i != end; ++i) {
356 const string &key = i->first;
357 int count = i->second;
|
360
| 358
|
361 for (int i = 0; i < requestProfileKeys.size(); i++) {
362 int temp_int =
363 m_requestProfileMap_ptr->lookup(requestProfileKeys[i]);
364 double percent = (100.0 * double(temp_int)) / double(m_requests);
| 359 double percent = (100.0 * double(count)) / double(m_requests);
|
365 vector<string> items;
| 360 vector<string> items;
|
366 tokenize(items, requestProfileKeys[i], ':');
367 vector<string>::iterator i = items.begin();
| 361 tokenize(items, key, ':');
362 vector<string>::iterator j = items.begin();
|
368 vector<string>::iterator end = items.end();
| 363 vector<string>::iterator end = items.end();
|
369 for (; i != end; ++i)
370 out << setw(10) << *i;
371 out << setw(11) << temp_int;
| 364 for (; j != end; ++i)
365 out << setw(10) << *j;
366 out << setw(11) << count;
|
372 out << setw(14) << percent << endl;
373 }
374 out << endl;
375
376 out << "filter_action: " << m_filter_action_histogram << endl;
377
378 if (!m_all_instructions) {
379 m_address_profiler_ptr->printStats(out);
380 }
381
382 if (m_all_instructions) {
383 m_inst_profiler_ptr->printStats(out);
384 }
385
386 out << endl;
387 out << "Message Delayed Cycles" << endl;
388 out << "----------------------" << endl;
389 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
390 out << "Total_nonPF_delay_cycles: "
391 << m_delayedCyclesNonPFHistogram << endl;
392 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
393 out << " virtual_network_" << i << "_delay_cycles: "
394 << m_delayedCyclesVCHistograms[i] << endl;
395 }
396
397 printResourceUsage(out);
398 }
399}
400
401void
402Profiler::printResourceUsage(ostream& out) const
403{
404 out << endl;
405 out << "Resource Usage" << endl;
406 out << "--------------" << endl;
407
408 integer_t pagesize = getpagesize(); // page size in bytes
409 out << "page_size: " << pagesize << endl;
410
411 rusage usage;
412 getrusage (RUSAGE_SELF, &usage);
413
414 out << "user_time: " << usage.ru_utime.tv_sec << endl;
415 out << "system_time: " << usage.ru_stime.tv_sec << endl;
416 out << "page_reclaims: " << usage.ru_minflt << endl;
417 out << "page_faults: " << usage.ru_majflt << endl;
418 out << "swaps: " << usage.ru_nswap << endl;
419 out << "block_inputs: " << usage.ru_inblock << endl;
420 out << "block_outputs: " << usage.ru_oublock << endl;
421}
422
423void
424Profiler::clearStats()
425{
426 m_ruby_start = g_eventQueue_ptr->getTime();
427
428 m_cycles_executed_at_start.resize(m_num_of_sequencers);
429 for (int i = 0; i < m_num_of_sequencers; i++) {
430 if (g_system_ptr == NULL) {
431 m_cycles_executed_at_start[i] = 0;
432 } else {
433 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
434 }
435 }
436
437 m_busyControllerCount.resize(MachineType_NUM); // all machines
438 for (int i = 0; i < MachineType_NUM; i++) {
439 int size = MachineType_base_count((MachineType)i);
440 m_busyControllerCount[i].resize(size);
441 for (int j = 0; j < size; j++) {
442 m_busyControllerCount[i][j] = 0;
443 }
444 }
445 m_busyBankCount = 0;
446
447 m_delayedCyclesHistogram.clear();
448 m_delayedCyclesNonPFHistogram.clear();
449 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
450 m_delayedCyclesVCHistograms.resize(size);
451 for (int i = 0; i < size; i++) {
452 m_delayedCyclesVCHistograms[i].clear();
453 }
454
455 m_missLatencyHistograms.resize(RubyRequestType_NUM);
456 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
457 m_missLatencyHistograms[i].clear(200);
458 }
459 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
460 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
461 m_machLatencyHistograms[i].clear(200);
462 }
463 m_allMissLatencyHistogram.clear(200);
464
465 m_SWPrefetchLatencyHistograms.resize(CacheRequestType_NUM);
466 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
467 m_SWPrefetchLatencyHistograms[i].clear(200);
468 }
469 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
470 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
471 m_SWPrefetchMachLatencyHistograms[i].clear(200);
472 }
473 m_allSWPrefetchLatencyHistogram.clear(200);
474
475 m_sequencer_requests.clear();
476 m_read_sharing_histogram.clear();
477 m_write_sharing_histogram.clear();
478 m_all_sharing_histogram.clear();
479 m_cache_to_cache = 0;
480 m_memory_to_cache = 0;
481
482 // clear HashMaps
| 367 out << setw(14) << percent << endl;
368 }
369 out << endl;
370
371 out << "filter_action: " << m_filter_action_histogram << endl;
372
373 if (!m_all_instructions) {
374 m_address_profiler_ptr->printStats(out);
375 }
376
377 if (m_all_instructions) {
378 m_inst_profiler_ptr->printStats(out);
379 }
380
381 out << endl;
382 out << "Message Delayed Cycles" << endl;
383 out << "----------------------" << endl;
384 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
385 out << "Total_nonPF_delay_cycles: "
386 << m_delayedCyclesNonPFHistogram << endl;
387 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
388 out << " virtual_network_" << i << "_delay_cycles: "
389 << m_delayedCyclesVCHistograms[i] << endl;
390 }
391
392 printResourceUsage(out);
393 }
394}
395
396void
397Profiler::printResourceUsage(ostream& out) const
398{
399 out << endl;
400 out << "Resource Usage" << endl;
401 out << "--------------" << endl;
402
403 integer_t pagesize = getpagesize(); // page size in bytes
404 out << "page_size: " << pagesize << endl;
405
406 rusage usage;
407 getrusage (RUSAGE_SELF, &usage);
408
409 out << "user_time: " << usage.ru_utime.tv_sec << endl;
410 out << "system_time: " << usage.ru_stime.tv_sec << endl;
411 out << "page_reclaims: " << usage.ru_minflt << endl;
412 out << "page_faults: " << usage.ru_majflt << endl;
413 out << "swaps: " << usage.ru_nswap << endl;
414 out << "block_inputs: " << usage.ru_inblock << endl;
415 out << "block_outputs: " << usage.ru_oublock << endl;
416}
417
418void
419Profiler::clearStats()
420{
421 m_ruby_start = g_eventQueue_ptr->getTime();
422
423 m_cycles_executed_at_start.resize(m_num_of_sequencers);
424 for (int i = 0; i < m_num_of_sequencers; i++) {
425 if (g_system_ptr == NULL) {
426 m_cycles_executed_at_start[i] = 0;
427 } else {
428 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
429 }
430 }
431
432 m_busyControllerCount.resize(MachineType_NUM); // all machines
433 for (int i = 0; i < MachineType_NUM; i++) {
434 int size = MachineType_base_count((MachineType)i);
435 m_busyControllerCount[i].resize(size);
436 for (int j = 0; j < size; j++) {
437 m_busyControllerCount[i][j] = 0;
438 }
439 }
440 m_busyBankCount = 0;
441
442 m_delayedCyclesHistogram.clear();
443 m_delayedCyclesNonPFHistogram.clear();
444 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
445 m_delayedCyclesVCHistograms.resize(size);
446 for (int i = 0; i < size; i++) {
447 m_delayedCyclesVCHistograms[i].clear();
448 }
449
450 m_missLatencyHistograms.resize(RubyRequestType_NUM);
451 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
452 m_missLatencyHistograms[i].clear(200);
453 }
454 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
455 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
456 m_machLatencyHistograms[i].clear(200);
457 }
458 m_allMissLatencyHistogram.clear(200);
459
460 m_SWPrefetchLatencyHistograms.resize(CacheRequestType_NUM);
461 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
462 m_SWPrefetchLatencyHistograms[i].clear(200);
463 }
464 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
465 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
466 m_SWPrefetchMachLatencyHistograms[i].clear(200);
467 }
468 m_allSWPrefetchLatencyHistogram.clear(200);
469
470 m_sequencer_requests.clear();
471 m_read_sharing_histogram.clear();
472 m_write_sharing_histogram.clear();
473 m_all_sharing_histogram.clear();
474 m_cache_to_cache = 0;
475 m_memory_to_cache = 0;
476
477 // clear HashMaps
|
483 m_requestProfileMap_ptr->clear();
| 478 m_requestProfileMap.clear();
|
484
485 // count requests profiled
486 m_requests = 0;
487
488 m_outstanding_requests.clear();
489 m_outstanding_persistent_requests.clear();
490
491 // Flush the prefetches through the system - used so that there
492 // are no outstanding requests after stats are cleared
493 //g_eventQueue_ptr->triggerAllEvents();
494
495 // update the start time
496 m_ruby_start = g_eventQueue_ptr->getTime();
497}
498
499void
500Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id)
501{
502 if (msg.getType() != CacheRequestType_IFETCH) {
503 // Note: The following line should be commented out if you
504 // want to use the special profiling that is part of the GS320
505 // protocol
506
507 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
508 // profiled by the AddressProfiler
509 m_address_profiler_ptr->
510 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
511 msg.getType(), msg.getAccessMode(), id, false);
512 }
513}
514
515void
516Profiler::profileSharing(const Address& addr, AccessType type,
517 NodeID requestor, const Set& sharers,
518 const Set& owner)
519{
520 Set set_contacted(owner);
521 if (type == AccessType_Write) {
522 set_contacted.addSet(sharers);
523 }
524 set_contacted.remove(requestor);
525 int number_contacted = set_contacted.count();
526
527 if (type == AccessType_Write) {
528 m_write_sharing_histogram.add(number_contacted);
529 } else {
530 m_read_sharing_histogram.add(number_contacted);
531 }
532 m_all_sharing_histogram.add(number_contacted);
533
534 if (number_contacted == 0) {
535 m_memory_to_cache++;
536 } else {
537 m_cache_to_cache++;
538 }
539}
540
541void
542Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
543{
544 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
545 m_delayedCyclesHistogram.add(delayCycles);
546 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
547 if (virtualNetwork != 0) {
548 m_delayedCyclesNonPFHistogram.add(delayCycles);
549 }
550}
551
552// profiles original cache requests including PUTs
553void
554Profiler::profileRequest(const string& requestStr)
555{
556 m_requests++;
557
| 479
480 // count requests profiled
481 m_requests = 0;
482
483 m_outstanding_requests.clear();
484 m_outstanding_persistent_requests.clear();
485
486 // Flush the prefetches through the system - used so that there
487 // are no outstanding requests after stats are cleared
488 //g_eventQueue_ptr->triggerAllEvents();
489
490 // update the start time
491 m_ruby_start = g_eventQueue_ptr->getTime();
492}
493
494void
495Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id)
496{
497 if (msg.getType() != CacheRequestType_IFETCH) {
498 // Note: The following line should be commented out if you
499 // want to use the special profiling that is part of the GS320
500 // protocol
501
502 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
503 // profiled by the AddressProfiler
504 m_address_profiler_ptr->
505 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
506 msg.getType(), msg.getAccessMode(), id, false);
507 }
508}
509
510void
511Profiler::profileSharing(const Address& addr, AccessType type,
512 NodeID requestor, const Set& sharers,
513 const Set& owner)
514{
515 Set set_contacted(owner);
516 if (type == AccessType_Write) {
517 set_contacted.addSet(sharers);
518 }
519 set_contacted.remove(requestor);
520 int number_contacted = set_contacted.count();
521
522 if (type == AccessType_Write) {
523 m_write_sharing_histogram.add(number_contacted);
524 } else {
525 m_read_sharing_histogram.add(number_contacted);
526 }
527 m_all_sharing_histogram.add(number_contacted);
528
529 if (number_contacted == 0) {
530 m_memory_to_cache++;
531 } else {
532 m_cache_to_cache++;
533 }
534}
535
536void
537Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
538{
539 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
540 m_delayedCyclesHistogram.add(delayCycles);
541 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
542 if (virtualNetwork != 0) {
543 m_delayedCyclesNonPFHistogram.add(delayCycles);
544 }
545}
546
547// profiles original cache requests including PUTs
548void
549Profiler::profileRequest(const string& requestStr)
550{
551 m_requests++;
552
|
558 if (m_requestProfileMap_ptr->exist(requestStr)) {
559 (m_requestProfileMap_ptr->lookup(requestStr))++;
560 } else {
561 m_requestProfileMap_ptr->add(requestStr, 1);
562 }
| 553 // if it doesn't exist, conveniently, it will be created with the
554 // default value which is 0
555 m_requestProfileMap[requestStr]++;
|
563}
564
565void
566Profiler::controllerBusy(MachineID machID)
567{
568 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
569}
570
571void
572Profiler::profilePFWait(Time waitTime)
573{
574 m_prefetchWaitHistogram.add(waitTime);
575}
576
577void
578Profiler::bankBusy()
579{
580 m_busyBankCount++;
581}
582
583// non-zero cycle demand request
584void
585Profiler::missLatency(Time t, RubyRequestType type)
586{
587 m_allMissLatencyHistogram.add(t);
588 m_missLatencyHistograms[type].add(t);
589}
590
591// non-zero cycle prefetch request
592void
593Profiler::swPrefetchLatency(Time t, CacheRequestType type,
594 GenericMachineType respondingMach)
595{
596 m_allSWPrefetchLatencyHistogram.add(t);
597 m_SWPrefetchLatencyHistograms[type].add(t);
598 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t);
599 if (respondingMach == GenericMachineType_Directory ||
600 respondingMach == GenericMachineType_NUM) {
601 m_SWPrefetchL2MissLatencyHistogram.add(t);
602 }
603}
604
605void
606Profiler::profileTransition(const string& component, NodeID version,
607 Address addr, const string& state, const string& event,
608 const string& next_state, const string& note)
609{
610 const int EVENT_SPACES = 20;
611 const int ID_SPACES = 3;
612 const int TIME_SPACES = 7;
613 const int COMP_SPACES = 10;
614 const int STATE_SPACES = 6;
615
616 if (g_debug_ptr->getDebugTime() <= 0 ||
617 g_eventQueue_ptr->getTime() < g_debug_ptr->getDebugTime())
618 return;
619
620 ostream &out = *debug_cout_ptr;
621 out.flags(ios::right);
622 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
623 out << setw(ID_SPACES) << version << " ";
624 out << setw(COMP_SPACES) << component;
625 out << setw(EVENT_SPACES) << event << " ";
626
627 out.flags(ios::right);
628 out << setw(STATE_SPACES) << state;
629 out << ">";
630 out.flags(ios::left);
631 out << setw(STATE_SPACES) << next_state;
632
633 out << " " << addr << " " << note;
634
635 out << endl;
636}
637
638// Helper function
639static double
640process_memory_total()
641{
642 // 4kB page size, 1024*1024 bytes per MB,
643 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
644 ifstream proc_file;
645 proc_file.open("/proc/self/statm");
646 int total_size_in_pages = 0;
647 int res_size_in_pages = 0;
648 proc_file >> total_size_in_pages;
649 proc_file >> res_size_in_pages;
650 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
651}
652
653static double
654process_memory_resident()
655{
656 // 4kB page size, 1024*1024 bytes per MB,
657 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
658 ifstream proc_file;
659 proc_file.open("/proc/self/statm");
660 int total_size_in_pages = 0;
661 int res_size_in_pages = 0;
662 proc_file >> total_size_in_pages;
663 proc_file >> res_size_in_pages;
664 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
665}
666
667void
668Profiler::rubyWatch(int id)
669{
670 uint64 tr = 0;
671 Address watch_address = Address(tr);
672 const int ID_SPACES = 3;
673 const int TIME_SPACES = 7;
674
675 ostream &out = *debug_cout_ptr;
676
677 out.flags(ios::right);
678 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
679 out << setw(ID_SPACES) << id << " "
680 << "RUBY WATCH " << watch_address << endl;
681
| 556}
557
558void
559Profiler::controllerBusy(MachineID machID)
560{
561 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
562}
563
564void
565Profiler::profilePFWait(Time waitTime)
566{
567 m_prefetchWaitHistogram.add(waitTime);
568}
569
570void
571Profiler::bankBusy()
572{
573 m_busyBankCount++;
574}
575
576// non-zero cycle demand request
577void
578Profiler::missLatency(Time t, RubyRequestType type)
579{
580 m_allMissLatencyHistogram.add(t);
581 m_missLatencyHistograms[type].add(t);
582}
583
584// non-zero cycle prefetch request
585void
586Profiler::swPrefetchLatency(Time t, CacheRequestType type,
587 GenericMachineType respondingMach)
588{
589 m_allSWPrefetchLatencyHistogram.add(t);
590 m_SWPrefetchLatencyHistograms[type].add(t);
591 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t);
592 if (respondingMach == GenericMachineType_Directory ||
593 respondingMach == GenericMachineType_NUM) {
594 m_SWPrefetchL2MissLatencyHistogram.add(t);
595 }
596}
597
598void
599Profiler::profileTransition(const string& component, NodeID version,
600 Address addr, const string& state, const string& event,
601 const string& next_state, const string& note)
602{
603 const int EVENT_SPACES = 20;
604 const int ID_SPACES = 3;
605 const int TIME_SPACES = 7;
606 const int COMP_SPACES = 10;
607 const int STATE_SPACES = 6;
608
609 if (g_debug_ptr->getDebugTime() <= 0 ||
610 g_eventQueue_ptr->getTime() < g_debug_ptr->getDebugTime())
611 return;
612
613 ostream &out = *debug_cout_ptr;
614 out.flags(ios::right);
615 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
616 out << setw(ID_SPACES) << version << " ";
617 out << setw(COMP_SPACES) << component;
618 out << setw(EVENT_SPACES) << event << " ";
619
620 out.flags(ios::right);
621 out << setw(STATE_SPACES) << state;
622 out << ">";
623 out.flags(ios::left);
624 out << setw(STATE_SPACES) << next_state;
625
626 out << " " << addr << " " << note;
627
628 out << endl;
629}
630
631// Helper function
632static double
633process_memory_total()
634{
635 // 4kB page size, 1024*1024 bytes per MB,
636 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
637 ifstream proc_file;
638 proc_file.open("/proc/self/statm");
639 int total_size_in_pages = 0;
640 int res_size_in_pages = 0;
641 proc_file >> total_size_in_pages;
642 proc_file >> res_size_in_pages;
643 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
644}
645
646static double
647process_memory_resident()
648{
649 // 4kB page size, 1024*1024 bytes per MB,
650 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
651 ifstream proc_file;
652 proc_file.open("/proc/self/statm");
653 int total_size_in_pages = 0;
654 int res_size_in_pages = 0;
655 proc_file >> total_size_in_pages;
656 proc_file >> res_size_in_pages;
657 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
658}
659
660void
661Profiler::rubyWatch(int id)
662{
663 uint64 tr = 0;
664 Address watch_address = Address(tr);
665 const int ID_SPACES = 3;
666 const int TIME_SPACES = 7;
667
668 ostream &out = *debug_cout_ptr;
669
670 out.flags(ios::right);
671 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
672 out << setw(ID_SPACES) << id << " "
673 << "RUBY WATCH " << watch_address << endl;
674
|
682 if (!m_watch_address_list_ptr->exist(watch_address)) {
683 m_watch_address_list_ptr->add(watch_address, 1);
684 }
| 675 // don't care about success or failure
676 m_watch_address_set.insert(watch_address);
|
685}
686
687bool
688Profiler::watchAddress(Address addr)
689{
| 677}
678
679bool
680Profiler::watchAddress(Address addr)
681{
|
690 if (m_watch_address_list_ptr->exist(addr))
691 return true;
692 else
693 return false;
| 682 return m_watch_address_set.count(addr) > 0;
|
694}
695
696Profiler *
697RubyProfilerParams::create()
698{
699 return new Profiler(this);
700}
| 683}
684
685Profiler *
686RubyProfilerParams::create()
687{
688 return new Profiler(this);
689}
|