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