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#include <sys/types.h>
49#include <unistd.h>
50
51#include <algorithm>
52#include <fstream>
53
54#include "base/stl_helpers.hh"
55#include "base/str.hh"
56#include "mem/protocol/MachineType.hh"
57#include "mem/protocol/RubyRequest.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
63using namespace std;
64using m5::stl_helpers::operator<<;
65
66static double process_memory_total();
67static double process_memory_resident();
68
69Profiler::Profiler(const Params *p)
70 : SimObject(p), m_event(this)
71{
72 m_inst_profiler_ptr = NULL;
73 m_address_profiler_ptr = NULL;
74
75 m_real_time_start_time = time(NULL); // Not reset in clearStats()
76 m_stats_period = 1000000; // Default
77 m_periodic_output_file_ptr = &cerr;
78
79 m_hot_lines = p->hot_lines;
80 m_all_instructions = p->all_instructions;
81
82 m_num_of_sequencers = p->num_of_sequencers;
83
84 m_hot_lines = false;
85 m_all_instructions = false;
86
87 m_address_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
88 m_address_profiler_ptr->setHotLines(m_hot_lines);
89 m_address_profiler_ptr->setAllInstructions(m_all_instructions);
90
91 if (m_all_instructions) {
92 m_inst_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
93 m_inst_profiler_ptr->setHotLines(m_hot_lines);
94 m_inst_profiler_ptr->setAllInstructions(m_all_instructions);
95 }
96
97 p->ruby_system->registerProfiler(this);
98}
99
100Profiler::~Profiler()
101{
102 if (m_periodic_output_file_ptr != &cerr) {
103 delete m_periodic_output_file_ptr;
104 }
105}
106
107void
108Profiler::wakeup()
109{
110 // FIXME - avoid the repeated code
111
112 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
113
114 for (int i = 0; i < m_num_of_sequencers; i++) {
115 perProcCycleCount[i] =
116 g_system_ptr->getTime() - m_cycles_executed_at_start[i] + 1;
117 // The +1 allows us to avoid division by zero
118 }
119
120 ostream &out = *m_periodic_output_file_ptr;
121
122 out << "ruby_cycles: " << g_system_ptr->getTime()-m_ruby_start << endl
123 << "mbytes_resident: " << process_memory_resident() << endl
124 << "mbytes_total: " << process_memory_total() << endl;
125
126 if (process_memory_total() > 0) {
127 out << "resident_ratio: "
128 << process_memory_resident() / process_memory_total() << endl;
129 }
130
131 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
132
133 out << endl;
134
135 if (m_all_instructions) {
136 m_inst_profiler_ptr->printStats(out);
137 }
138
139 //g_system_ptr->getNetwork()->printStats(out);
| 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#include <sys/types.h>
49#include <unistd.h>
50
51#include <algorithm>
52#include <fstream>
53
54#include "base/stl_helpers.hh"
55#include "base/str.hh"
56#include "mem/protocol/MachineType.hh"
57#include "mem/protocol/RubyRequest.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
63using namespace std;
64using m5::stl_helpers::operator<<;
65
66static double process_memory_total();
67static double process_memory_resident();
68
69Profiler::Profiler(const Params *p)
70 : SimObject(p), m_event(this)
71{
72 m_inst_profiler_ptr = NULL;
73 m_address_profiler_ptr = NULL;
74
75 m_real_time_start_time = time(NULL); // Not reset in clearStats()
76 m_stats_period = 1000000; // Default
77 m_periodic_output_file_ptr = &cerr;
78
79 m_hot_lines = p->hot_lines;
80 m_all_instructions = p->all_instructions;
81
82 m_num_of_sequencers = p->num_of_sequencers;
83
84 m_hot_lines = false;
85 m_all_instructions = false;
86
87 m_address_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
88 m_address_profiler_ptr->setHotLines(m_hot_lines);
89 m_address_profiler_ptr->setAllInstructions(m_all_instructions);
90
91 if (m_all_instructions) {
92 m_inst_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
93 m_inst_profiler_ptr->setHotLines(m_hot_lines);
94 m_inst_profiler_ptr->setAllInstructions(m_all_instructions);
95 }
96
97 p->ruby_system->registerProfiler(this);
98}
99
100Profiler::~Profiler()
101{
102 if (m_periodic_output_file_ptr != &cerr) {
103 delete m_periodic_output_file_ptr;
104 }
105}
106
107void
108Profiler::wakeup()
109{
110 // FIXME - avoid the repeated code
111
112 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
113
114 for (int i = 0; i < m_num_of_sequencers; i++) {
115 perProcCycleCount[i] =
116 g_system_ptr->getTime() - m_cycles_executed_at_start[i] + 1;
117 // The +1 allows us to avoid division by zero
118 }
119
120 ostream &out = *m_periodic_output_file_ptr;
121
122 out << "ruby_cycles: " << g_system_ptr->getTime()-m_ruby_start << endl
123 << "mbytes_resident: " << process_memory_resident() << endl
124 << "mbytes_total: " << process_memory_total() << endl;
125
126 if (process_memory_total() > 0) {
127 out << "resident_ratio: "
128 << process_memory_resident() / process_memory_total() << endl;
129 }
130
131 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
132
133 out << endl;
134
135 if (m_all_instructions) {
136 m_inst_profiler_ptr->printStats(out);
137 }
138
139 //g_system_ptr->getNetwork()->printStats(out);
|
140 schedule(m_event, curTick() + m_stats_period * g_system_ptr->getClock());
| 140 schedule(m_event, g_system_ptr->clockEdge(Cycles(m_stats_period )));
|
141}
142
143void
144Profiler::setPeriodicStatsFile(const string& filename)
145{
146 cout << "Recording periodic statistics to file '" << filename << "' every "
147 << m_stats_period << " Ruby cycles" << endl;
148
149 if (m_periodic_output_file_ptr != &cerr) {
150 delete m_periodic_output_file_ptr;
151 }
152
153 m_periodic_output_file_ptr = new ofstream(filename.c_str());
| 141}
142
143void
144Profiler::setPeriodicStatsFile(const string& filename)
145{
146 cout << "Recording periodic statistics to file '" << filename << "' every "
147 << m_stats_period << " Ruby cycles" << endl;
148
149 if (m_periodic_output_file_ptr != &cerr) {
150 delete m_periodic_output_file_ptr;
151 }
152
153 m_periodic_output_file_ptr = new ofstream(filename.c_str());
|
154 schedule(m_event, curTick() + g_system_ptr->getClock());
| 154 schedule(m_event, g_system_ptr->clockEdge(Cycles(1)));
|
155}
156
157void
158Profiler::setPeriodicStatsInterval(integer_t period)
159{
160 cout << "Recording periodic statistics every " << m_stats_period
161 << " Ruby cycles" << endl;
162
163 m_stats_period = period;
| 155}
156
157void
158Profiler::setPeriodicStatsInterval(integer_t period)
159{
160 cout << "Recording periodic statistics every " << m_stats_period
161 << " Ruby cycles" << endl;
162
163 m_stats_period = period;
|
164 schedule(m_event, curTick() + g_system_ptr->getClock());
| 164 schedule(m_event, g_system_ptr->clockEdge(Cycles(1)));
|
165}
166
167void
168Profiler::print(ostream& out) const
169{
170 out << "[Profiler]";
171}
172
173void
174Profiler::printStats(ostream& out, bool short_stats)
175{
176 out << endl;
177 if (short_stats) {
178 out << "SHORT ";
179 }
180 out << "Profiler Stats" << endl;
181 out << "--------------" << endl;
182
183 time_t real_time_current = time(NULL);
184 double seconds = difftime(real_time_current, m_real_time_start_time);
185 double minutes = seconds / 60.0;
186 double hours = minutes / 60.0;
187 double days = hours / 24.0;
188 Time ruby_cycles = g_system_ptr->getTime()-m_ruby_start;
189
190 if (!short_stats) {
191 out << "Elapsed_time_in_seconds: " << seconds << endl;
192 out << "Elapsed_time_in_minutes: " << minutes << endl;
193 out << "Elapsed_time_in_hours: " << hours << endl;
194 out << "Elapsed_time_in_days: " << days << endl;
195 out << endl;
196 }
197
198 // print the virtual runtimes as well
199 struct tms vtime;
200 times(&vtime);
201 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
202 minutes = seconds / 60.0;
203 hours = minutes / 60.0;
204 days = hours / 24.0;
205 out << "Virtual_time_in_seconds: " << seconds << endl;
206 out << "Virtual_time_in_minutes: " << minutes << endl;
207 out << "Virtual_time_in_hours: " << hours << endl;
208 out << "Virtual_time_in_days: " << days << endl;
209 out << endl;
210
211 out << "Ruby_current_time: " << g_system_ptr->getTime() << endl;
212 out << "Ruby_start_time: " << m_ruby_start << endl;
213 out << "Ruby_cycles: " << ruby_cycles << endl;
214 out << endl;
215
216 if (!short_stats) {
217 out << "mbytes_resident: " << process_memory_resident() << endl;
218 out << "mbytes_total: " << process_memory_total() << endl;
219 if (process_memory_total() > 0) {
220 out << "resident_ratio: "
221 << process_memory_resident()/process_memory_total() << endl;
222 }
223 out << endl;
224 }
225
226 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
227
228 for (int i = 0; i < m_num_of_sequencers; i++) {
229 perProcCycleCount[i] =
230 g_system_ptr->getTime() - m_cycles_executed_at_start[i] + 1;
231 // The +1 allows us to avoid division by zero
232 }
233
234 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
235
236 out << endl;
237
238 if (!short_stats) {
239 out << "Busy Controller Counts:" << endl;
240 for (int i = 0; i < MachineType_NUM; i++) {
241 int size = MachineType_base_count((MachineType)i);
242 for (int j = 0; j < size; j++) {
243 MachineID machID;
244 machID.type = (MachineType)i;
245 machID.num = j;
246 out << machID << ":" << m_busyControllerCount[i][j] << " ";
247 if ((j + 1) % 8 == 0) {
248 out << endl;
249 }
250 }
251 out << endl;
252 }
253 out << endl;
254
255 out << "Busy Bank Count:" << m_busyBankCount << endl;
256 out << endl;
257
258 out << "sequencer_requests_outstanding: "
259 << m_sequencer_requests << endl;
260 out << endl;
261 }
262
263 if (!short_stats) {
264 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
265 out << "----------------------------------------" << endl;
266 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
267 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
268 if (m_missLatencyHistograms[i].size() > 0) {
269 out << "miss_latency_" << RubyRequestType(i) << ": "
270 << m_missLatencyHistograms[i] << endl;
271 }
272 }
273 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
274 if (m_machLatencyHistograms[i].size() > 0) {
275 out << "miss_latency_" << GenericMachineType(i) << ": "
276 << m_machLatencyHistograms[i] << endl;
277 }
278 }
279
280 out << "miss_latency_wCC_issue_to_initial_request: "
281 << m_wCCIssueToInitialRequestHistogram << endl;
282 out << "miss_latency_wCC_initial_forward_request: "
283 << m_wCCInitialRequestToForwardRequestHistogram << endl;
284 out << "miss_latency_wCC_forward_to_first_response: "
285 << m_wCCForwardRequestToFirstResponseHistogram << endl;
286 out << "miss_latency_wCC_first_response_to_completion: "
287 << m_wCCFirstResponseToCompleteHistogram << endl;
288 out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
289 out << "miss_latency_dir_issue_to_initial_request: "
290 << m_dirIssueToInitialRequestHistogram << endl;
291 out << "miss_latency_dir_initial_forward_request: "
292 << m_dirInitialRequestToForwardRequestHistogram << endl;
293 out << "miss_latency_dir_forward_to_first_response: "
294 << m_dirForwardRequestToFirstResponseHistogram << endl;
295 out << "miss_latency_dir_first_response_to_completion: "
296 << m_dirFirstResponseToCompleteHistogram << endl;
297 out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
298
299 for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
300 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
301 if (m_missMachLatencyHistograms[i][j].size() > 0) {
302 out << "miss_latency_" << RubyRequestType(i)
303 << "_" << GenericMachineType(j) << ": "
304 << m_missMachLatencyHistograms[i][j] << endl;
305 }
306 }
307 }
308
309 out << endl;
310
311 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
312 out << "------------------------------------" << endl;
313 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
314 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
315 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
316 out << "prefetch_latency_" << RubyRequestType(i) << ": "
317 << m_SWPrefetchLatencyHistograms[i] << endl;
318 }
319 }
320 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
321 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
322 out << "prefetch_latency_" << GenericMachineType(i) << ": "
323 << m_SWPrefetchMachLatencyHistograms[i] << endl;
324 }
325 }
326 out << "prefetch_latency_L2Miss:"
327 << m_SWPrefetchL2MissLatencyHistogram << endl;
328
329 if (m_all_sharing_histogram.size() > 0) {
330 out << "all_sharing: " << m_all_sharing_histogram << endl;
331 out << "read_sharing: " << m_read_sharing_histogram << endl;
332 out << "write_sharing: " << m_write_sharing_histogram << endl;
333
334 out << "all_sharing_percent: ";
335 m_all_sharing_histogram.printPercent(out);
336 out << endl;
337
338 out << "read_sharing_percent: ";
339 m_read_sharing_histogram.printPercent(out);
340 out << endl;
341
342 out << "write_sharing_percent: ";
343 m_write_sharing_histogram.printPercent(out);
344 out << endl;
345
346 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
347 out << "all_misses: " << total_miss << endl;
348 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
349 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
350 out << "cache_to_cache_percent: "
351 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
352 << endl;
353 out << "memory_to_cache_percent: "
354 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
355 << endl;
356 out << endl;
357 }
358
359 if (m_outstanding_requests.size() > 0) {
360 out << "outstanding_requests: ";
361 m_outstanding_requests.printPercent(out);
362 out << endl;
363 out << endl;
364 }
365 }
366
367 if (!short_stats) {
368 out << "Request vs. RubySystem State Profile" << endl;
369 out << "--------------------------------" << endl;
370 out << endl;
371
372 map<string, int>::const_iterator i = m_requestProfileMap.begin();
373 map<string, int>::const_iterator end = m_requestProfileMap.end();
374 for (; i != end; ++i) {
375 const string &key = i->first;
376 int count = i->second;
377
378 double percent = (100.0 * double(count)) / double(m_requests);
379 vector<string> items;
380 tokenize(items, key, ':');
381 vector<string>::iterator j = items.begin();
382 vector<string>::iterator end = items.end();
383 for (; j != end; ++i)
384 out << setw(10) << *j;
385 out << setw(11) << count;
386 out << setw(14) << percent << endl;
387 }
388 out << endl;
389
390 out << "filter_action: " << m_filter_action_histogram << endl;
391
392 if (!m_all_instructions) {
393 m_address_profiler_ptr->printStats(out);
394 }
395
396 if (m_all_instructions) {
397 m_inst_profiler_ptr->printStats(out);
398 }
399
400 out << endl;
401 out << "Message Delayed Cycles" << endl;
402 out << "----------------------" << endl;
403 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
404 out << "Total_nonPF_delay_cycles: "
405 << m_delayedCyclesNonPFHistogram << endl;
406 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
407 out << " virtual_network_" << i << "_delay_cycles: "
408 << m_delayedCyclesVCHistograms[i] << endl;
409 }
410
411 printResourceUsage(out);
412 }
413}
414
415void
416Profiler::printResourceUsage(ostream& out) const
417{
418 out << endl;
419 out << "Resource Usage" << endl;
420 out << "--------------" << endl;
421
422 integer_t pagesize = getpagesize(); // page size in bytes
423 out << "page_size: " << pagesize << endl;
424
425 rusage usage;
426 getrusage (RUSAGE_SELF, &usage);
427
428 out << "user_time: " << usage.ru_utime.tv_sec << endl;
429 out << "system_time: " << usage.ru_stime.tv_sec << endl;
430 out << "page_reclaims: " << usage.ru_minflt << endl;
431 out << "page_faults: " << usage.ru_majflt << endl;
432 out << "swaps: " << usage.ru_nswap << endl;
433 out << "block_inputs: " << usage.ru_inblock << endl;
434 out << "block_outputs: " << usage.ru_oublock << endl;
435}
436
437void
438Profiler::clearStats()
439{
440 m_ruby_start = g_system_ptr->getTime();
441
442 m_cycles_executed_at_start.resize(m_num_of_sequencers);
443 for (int i = 0; i < m_num_of_sequencers; i++) {
444 if (g_system_ptr == NULL) {
445 m_cycles_executed_at_start[i] = 0;
446 } else {
447 m_cycles_executed_at_start[i] = g_system_ptr->getTime();
448 }
449 }
450
451 m_busyControllerCount.resize(MachineType_NUM); // all machines
452 for (int i = 0; i < MachineType_NUM; i++) {
453 int size = MachineType_base_count((MachineType)i);
454 m_busyControllerCount[i].resize(size);
455 for (int j = 0; j < size; j++) {
456 m_busyControllerCount[i][j] = 0;
457 }
458 }
459 m_busyBankCount = 0;
460
461 m_delayedCyclesHistogram.clear();
462 m_delayedCyclesNonPFHistogram.clear();
463 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
464 m_delayedCyclesVCHistograms.resize(size);
465 for (int i = 0; i < size; i++) {
466 m_delayedCyclesVCHistograms[i].clear();
467 }
468
469 m_missLatencyHistograms.resize(RubyRequestType_NUM);
470 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
471 m_missLatencyHistograms[i].clear(200);
472 }
473 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
474 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
475 m_machLatencyHistograms[i].clear(200);
476 }
477 m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
478 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
479 m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
480 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
481 m_missMachLatencyHistograms[i][j].clear(200);
482 }
483 }
484 m_allMissLatencyHistogram.clear(200);
485 m_wCCIssueToInitialRequestHistogram.clear(200);
486 m_wCCInitialRequestToForwardRequestHistogram.clear(200);
487 m_wCCForwardRequestToFirstResponseHistogram.clear(200);
488 m_wCCFirstResponseToCompleteHistogram.clear(200);
489 m_wCCIncompleteTimes = 0;
490 m_dirIssueToInitialRequestHistogram.clear(200);
491 m_dirInitialRequestToForwardRequestHistogram.clear(200);
492 m_dirForwardRequestToFirstResponseHistogram.clear(200);
493 m_dirFirstResponseToCompleteHistogram.clear(200);
494 m_dirIncompleteTimes = 0;
495
496 m_SWPrefetchLatencyHistograms.resize(RubyRequestType_NUM);
497 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
498 m_SWPrefetchLatencyHistograms[i].clear(200);
499 }
500 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
501 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
502 m_SWPrefetchMachLatencyHistograms[i].clear(200);
503 }
504 m_allSWPrefetchLatencyHistogram.clear(200);
505
506 m_sequencer_requests.clear();
507 m_read_sharing_histogram.clear();
508 m_write_sharing_histogram.clear();
509 m_all_sharing_histogram.clear();
510 m_cache_to_cache = 0;
511 m_memory_to_cache = 0;
512
513 // clear HashMaps
514 m_requestProfileMap.clear();
515
516 // count requests profiled
517 m_requests = 0;
518
519 m_outstanding_requests.clear();
520 m_outstanding_persistent_requests.clear();
521
522 // Flush the prefetches through the system - used so that there
523 // are no outstanding requests after stats are cleared
524 //g_eventQueue_ptr->triggerAllEvents();
525
526 // update the start time
527 m_ruby_start = g_system_ptr->getTime();
528}
529
530void
531Profiler::addAddressTraceSample(const RubyRequest& msg, NodeID id)
532{
533 if (msg.getType() != RubyRequestType_IFETCH) {
534 // Note: The following line should be commented out if you
535 // want to use the special profiling that is part of the GS320
536 // protocol
537
538 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
539 // profiled by the AddressProfiler
540 m_address_profiler_ptr->
541 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
542 msg.getType(), msg.getAccessMode(), id, false);
543 }
544}
545
546void
547Profiler::profileSharing(const Address& addr, AccessType type,
548 NodeID requestor, const Set& sharers,
549 const Set& owner)
550{
551 Set set_contacted(owner);
552 if (type == AccessType_Write) {
553 set_contacted.addSet(sharers);
554 }
555 set_contacted.remove(requestor);
556 int number_contacted = set_contacted.count();
557
558 if (type == AccessType_Write) {
559 m_write_sharing_histogram.add(number_contacted);
560 } else {
561 m_read_sharing_histogram.add(number_contacted);
562 }
563 m_all_sharing_histogram.add(number_contacted);
564
565 if (number_contacted == 0) {
566 m_memory_to_cache++;
567 } else {
568 m_cache_to_cache++;
569 }
570}
571
572void
573Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
574{
575 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
576 m_delayedCyclesHistogram.add(delayCycles);
577 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
578 if (virtualNetwork != 0) {
579 m_delayedCyclesNonPFHistogram.add(delayCycles);
580 }
581}
582
583// profiles original cache requests including PUTs
584void
585Profiler::profileRequest(const string& requestStr)
586{
587 m_requests++;
588
589 // if it doesn't exist, conveniently, it will be created with the
590 // default value which is 0
591 m_requestProfileMap[requestStr]++;
592}
593
594void
595Profiler::controllerBusy(MachineID machID)
596{
597 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
598}
599
600void
601Profiler::profilePFWait(Time waitTime)
602{
603 m_prefetchWaitHistogram.add(waitTime);
604}
605
606void
607Profiler::bankBusy()
608{
609 m_busyBankCount++;
610}
611
612// non-zero cycle demand request
613void
614Profiler::missLatency(Time cycles,
615 RubyRequestType type,
616 const GenericMachineType respondingMach)
617{
618 m_allMissLatencyHistogram.add(cycles);
619 m_missLatencyHistograms[type].add(cycles);
620 m_machLatencyHistograms[respondingMach].add(cycles);
621 m_missMachLatencyHistograms[type][respondingMach].add(cycles);
622}
623
624void
625Profiler::missLatencyWcc(Time issuedTime,
626 Time initialRequestTime,
627 Time forwardRequestTime,
628 Time firstResponseTime,
629 Time completionTime)
630{
631 if ((issuedTime <= initialRequestTime) &&
632 (initialRequestTime <= forwardRequestTime) &&
633 (forwardRequestTime <= firstResponseTime) &&
634 (firstResponseTime <= completionTime)) {
635 m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
636
637 m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
638 initialRequestTime);
639
640 m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
641 forwardRequestTime);
642
643 m_wCCFirstResponseToCompleteHistogram.add(completionTime -
644 firstResponseTime);
645 } else {
646 m_wCCIncompleteTimes++;
647 }
648}
649
650void
651Profiler::missLatencyDir(Time issuedTime,
652 Time initialRequestTime,
653 Time forwardRequestTime,
654 Time firstResponseTime,
655 Time completionTime)
656{
657 if ((issuedTime <= initialRequestTime) &&
658 (initialRequestTime <= forwardRequestTime) &&
659 (forwardRequestTime <= firstResponseTime) &&
660 (firstResponseTime <= completionTime)) {
661 m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
662
663 m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
664 initialRequestTime);
665
666 m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
667 forwardRequestTime);
668
669 m_dirFirstResponseToCompleteHistogram.add(completionTime -
670 firstResponseTime);
671 } else {
672 m_dirIncompleteTimes++;
673 }
674}
675
676// non-zero cycle prefetch request
677void
678Profiler::swPrefetchLatency(Time cycles,
679 RubyRequestType type,
680 const GenericMachineType respondingMach)
681{
682 m_allSWPrefetchLatencyHistogram.add(cycles);
683 m_SWPrefetchLatencyHistograms[type].add(cycles);
684 m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
685 if (respondingMach == GenericMachineType_Directory ||
686 respondingMach == GenericMachineType_NUM) {
687 m_SWPrefetchL2MissLatencyHistogram.add(cycles);
688 }
689}
690
691// Helper function
692static double
693process_memory_total()
694{
695 // 4kB page size, 1024*1024 bytes per MB,
696 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
697 ifstream proc_file;
698 proc_file.open("/proc/self/statm");
699 int total_size_in_pages = 0;
700 int res_size_in_pages = 0;
701 proc_file >> total_size_in_pages;
702 proc_file >> res_size_in_pages;
703 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
704}
705
706static double
707process_memory_resident()
708{
709 // 4kB page size, 1024*1024 bytes per MB,
710 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
711 ifstream proc_file;
712 proc_file.open("/proc/self/statm");
713 int total_size_in_pages = 0;
714 int res_size_in_pages = 0;
715 proc_file >> total_size_in_pages;
716 proc_file >> res_size_in_pages;
717 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
718}
719
720void
721Profiler::rubyWatch(int id)
722{
723 uint64 tr = 0;
724 Address watch_address = Address(tr);
725
726 DPRINTFN("%7s %3s RUBY WATCH %d\n", g_system_ptr->getTime(), id,
727 watch_address);
728
729 // don't care about success or failure
730 m_watch_address_set.insert(watch_address);
731}
732
733bool
734Profiler::watchAddress(Address addr)
735{
736 return m_watch_address_set.count(addr) > 0;
737}
738
739Profiler *
740RubyProfilerParams::create()
741{
742 return new Profiler(this);
743}
| 165}
166
167void
168Profiler::print(ostream& out) const
169{
170 out << "[Profiler]";
171}
172
173void
174Profiler::printStats(ostream& out, bool short_stats)
175{
176 out << endl;
177 if (short_stats) {
178 out << "SHORT ";
179 }
180 out << "Profiler Stats" << endl;
181 out << "--------------" << endl;
182
183 time_t real_time_current = time(NULL);
184 double seconds = difftime(real_time_current, m_real_time_start_time);
185 double minutes = seconds / 60.0;
186 double hours = minutes / 60.0;
187 double days = hours / 24.0;
188 Time ruby_cycles = g_system_ptr->getTime()-m_ruby_start;
189
190 if (!short_stats) {
191 out << "Elapsed_time_in_seconds: " << seconds << endl;
192 out << "Elapsed_time_in_minutes: " << minutes << endl;
193 out << "Elapsed_time_in_hours: " << hours << endl;
194 out << "Elapsed_time_in_days: " << days << endl;
195 out << endl;
196 }
197
198 // print the virtual runtimes as well
199 struct tms vtime;
200 times(&vtime);
201 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
202 minutes = seconds / 60.0;
203 hours = minutes / 60.0;
204 days = hours / 24.0;
205 out << "Virtual_time_in_seconds: " << seconds << endl;
206 out << "Virtual_time_in_minutes: " << minutes << endl;
207 out << "Virtual_time_in_hours: " << hours << endl;
208 out << "Virtual_time_in_days: " << days << endl;
209 out << endl;
210
211 out << "Ruby_current_time: " << g_system_ptr->getTime() << endl;
212 out << "Ruby_start_time: " << m_ruby_start << endl;
213 out << "Ruby_cycles: " << ruby_cycles << endl;
214 out << endl;
215
216 if (!short_stats) {
217 out << "mbytes_resident: " << process_memory_resident() << endl;
218 out << "mbytes_total: " << process_memory_total() << endl;
219 if (process_memory_total() > 0) {
220 out << "resident_ratio: "
221 << process_memory_resident()/process_memory_total() << endl;
222 }
223 out << endl;
224 }
225
226 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
227
228 for (int i = 0; i < m_num_of_sequencers; i++) {
229 perProcCycleCount[i] =
230 g_system_ptr->getTime() - m_cycles_executed_at_start[i] + 1;
231 // The +1 allows us to avoid division by zero
232 }
233
234 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
235
236 out << endl;
237
238 if (!short_stats) {
239 out << "Busy Controller Counts:" << endl;
240 for (int i = 0; i < MachineType_NUM; i++) {
241 int size = MachineType_base_count((MachineType)i);
242 for (int j = 0; j < size; j++) {
243 MachineID machID;
244 machID.type = (MachineType)i;
245 machID.num = j;
246 out << machID << ":" << m_busyControllerCount[i][j] << " ";
247 if ((j + 1) % 8 == 0) {
248 out << endl;
249 }
250 }
251 out << endl;
252 }
253 out << endl;
254
255 out << "Busy Bank Count:" << m_busyBankCount << endl;
256 out << endl;
257
258 out << "sequencer_requests_outstanding: "
259 << m_sequencer_requests << endl;
260 out << endl;
261 }
262
263 if (!short_stats) {
264 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
265 out << "----------------------------------------" << endl;
266 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
267 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
268 if (m_missLatencyHistograms[i].size() > 0) {
269 out << "miss_latency_" << RubyRequestType(i) << ": "
270 << m_missLatencyHistograms[i] << endl;
271 }
272 }
273 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
274 if (m_machLatencyHistograms[i].size() > 0) {
275 out << "miss_latency_" << GenericMachineType(i) << ": "
276 << m_machLatencyHistograms[i] << endl;
277 }
278 }
279
280 out << "miss_latency_wCC_issue_to_initial_request: "
281 << m_wCCIssueToInitialRequestHistogram << endl;
282 out << "miss_latency_wCC_initial_forward_request: "
283 << m_wCCInitialRequestToForwardRequestHistogram << endl;
284 out << "miss_latency_wCC_forward_to_first_response: "
285 << m_wCCForwardRequestToFirstResponseHistogram << endl;
286 out << "miss_latency_wCC_first_response_to_completion: "
287 << m_wCCFirstResponseToCompleteHistogram << endl;
288 out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
289 out << "miss_latency_dir_issue_to_initial_request: "
290 << m_dirIssueToInitialRequestHistogram << endl;
291 out << "miss_latency_dir_initial_forward_request: "
292 << m_dirInitialRequestToForwardRequestHistogram << endl;
293 out << "miss_latency_dir_forward_to_first_response: "
294 << m_dirForwardRequestToFirstResponseHistogram << endl;
295 out << "miss_latency_dir_first_response_to_completion: "
296 << m_dirFirstResponseToCompleteHistogram << endl;
297 out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
298
299 for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
300 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
301 if (m_missMachLatencyHistograms[i][j].size() > 0) {
302 out << "miss_latency_" << RubyRequestType(i)
303 << "_" << GenericMachineType(j) << ": "
304 << m_missMachLatencyHistograms[i][j] << endl;
305 }
306 }
307 }
308
309 out << endl;
310
311 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
312 out << "------------------------------------" << endl;
313 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
314 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
315 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
316 out << "prefetch_latency_" << RubyRequestType(i) << ": "
317 << m_SWPrefetchLatencyHistograms[i] << endl;
318 }
319 }
320 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
321 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
322 out << "prefetch_latency_" << GenericMachineType(i) << ": "
323 << m_SWPrefetchMachLatencyHistograms[i] << endl;
324 }
325 }
326 out << "prefetch_latency_L2Miss:"
327 << m_SWPrefetchL2MissLatencyHistogram << endl;
328
329 if (m_all_sharing_histogram.size() > 0) {
330 out << "all_sharing: " << m_all_sharing_histogram << endl;
331 out << "read_sharing: " << m_read_sharing_histogram << endl;
332 out << "write_sharing: " << m_write_sharing_histogram << endl;
333
334 out << "all_sharing_percent: ";
335 m_all_sharing_histogram.printPercent(out);
336 out << endl;
337
338 out << "read_sharing_percent: ";
339 m_read_sharing_histogram.printPercent(out);
340 out << endl;
341
342 out << "write_sharing_percent: ";
343 m_write_sharing_histogram.printPercent(out);
344 out << endl;
345
346 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
347 out << "all_misses: " << total_miss << endl;
348 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
349 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
350 out << "cache_to_cache_percent: "
351 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
352 << endl;
353 out << "memory_to_cache_percent: "
354 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
355 << endl;
356 out << endl;
357 }
358
359 if (m_outstanding_requests.size() > 0) {
360 out << "outstanding_requests: ";
361 m_outstanding_requests.printPercent(out);
362 out << endl;
363 out << endl;
364 }
365 }
366
367 if (!short_stats) {
368 out << "Request vs. RubySystem State Profile" << endl;
369 out << "--------------------------------" << endl;
370 out << endl;
371
372 map<string, int>::const_iterator i = m_requestProfileMap.begin();
373 map<string, int>::const_iterator end = m_requestProfileMap.end();
374 for (; i != end; ++i) {
375 const string &key = i->first;
376 int count = i->second;
377
378 double percent = (100.0 * double(count)) / double(m_requests);
379 vector<string> items;
380 tokenize(items, key, ':');
381 vector<string>::iterator j = items.begin();
382 vector<string>::iterator end = items.end();
383 for (; j != end; ++i)
384 out << setw(10) << *j;
385 out << setw(11) << count;
386 out << setw(14) << percent << endl;
387 }
388 out << endl;
389
390 out << "filter_action: " << m_filter_action_histogram << endl;
391
392 if (!m_all_instructions) {
393 m_address_profiler_ptr->printStats(out);
394 }
395
396 if (m_all_instructions) {
397 m_inst_profiler_ptr->printStats(out);
398 }
399
400 out << endl;
401 out << "Message Delayed Cycles" << endl;
402 out << "----------------------" << endl;
403 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
404 out << "Total_nonPF_delay_cycles: "
405 << m_delayedCyclesNonPFHistogram << endl;
406 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
407 out << " virtual_network_" << i << "_delay_cycles: "
408 << m_delayedCyclesVCHistograms[i] << endl;
409 }
410
411 printResourceUsage(out);
412 }
413}
414
415void
416Profiler::printResourceUsage(ostream& out) const
417{
418 out << endl;
419 out << "Resource Usage" << endl;
420 out << "--------------" << endl;
421
422 integer_t pagesize = getpagesize(); // page size in bytes
423 out << "page_size: " << pagesize << endl;
424
425 rusage usage;
426 getrusage (RUSAGE_SELF, &usage);
427
428 out << "user_time: " << usage.ru_utime.tv_sec << endl;
429 out << "system_time: " << usage.ru_stime.tv_sec << endl;
430 out << "page_reclaims: " << usage.ru_minflt << endl;
431 out << "page_faults: " << usage.ru_majflt << endl;
432 out << "swaps: " << usage.ru_nswap << endl;
433 out << "block_inputs: " << usage.ru_inblock << endl;
434 out << "block_outputs: " << usage.ru_oublock << endl;
435}
436
437void
438Profiler::clearStats()
439{
440 m_ruby_start = g_system_ptr->getTime();
441
442 m_cycles_executed_at_start.resize(m_num_of_sequencers);
443 for (int i = 0; i < m_num_of_sequencers; i++) {
444 if (g_system_ptr == NULL) {
445 m_cycles_executed_at_start[i] = 0;
446 } else {
447 m_cycles_executed_at_start[i] = g_system_ptr->getTime();
448 }
449 }
450
451 m_busyControllerCount.resize(MachineType_NUM); // all machines
452 for (int i = 0; i < MachineType_NUM; i++) {
453 int size = MachineType_base_count((MachineType)i);
454 m_busyControllerCount[i].resize(size);
455 for (int j = 0; j < size; j++) {
456 m_busyControllerCount[i][j] = 0;
457 }
458 }
459 m_busyBankCount = 0;
460
461 m_delayedCyclesHistogram.clear();
462 m_delayedCyclesNonPFHistogram.clear();
463 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
464 m_delayedCyclesVCHistograms.resize(size);
465 for (int i = 0; i < size; i++) {
466 m_delayedCyclesVCHistograms[i].clear();
467 }
468
469 m_missLatencyHistograms.resize(RubyRequestType_NUM);
470 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
471 m_missLatencyHistograms[i].clear(200);
472 }
473 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
474 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
475 m_machLatencyHistograms[i].clear(200);
476 }
477 m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
478 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
479 m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
480 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
481 m_missMachLatencyHistograms[i][j].clear(200);
482 }
483 }
484 m_allMissLatencyHistogram.clear(200);
485 m_wCCIssueToInitialRequestHistogram.clear(200);
486 m_wCCInitialRequestToForwardRequestHistogram.clear(200);
487 m_wCCForwardRequestToFirstResponseHistogram.clear(200);
488 m_wCCFirstResponseToCompleteHistogram.clear(200);
489 m_wCCIncompleteTimes = 0;
490 m_dirIssueToInitialRequestHistogram.clear(200);
491 m_dirInitialRequestToForwardRequestHistogram.clear(200);
492 m_dirForwardRequestToFirstResponseHistogram.clear(200);
493 m_dirFirstResponseToCompleteHistogram.clear(200);
494 m_dirIncompleteTimes = 0;
495
496 m_SWPrefetchLatencyHistograms.resize(RubyRequestType_NUM);
497 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
498 m_SWPrefetchLatencyHistograms[i].clear(200);
499 }
500 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
501 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
502 m_SWPrefetchMachLatencyHistograms[i].clear(200);
503 }
504 m_allSWPrefetchLatencyHistogram.clear(200);
505
506 m_sequencer_requests.clear();
507 m_read_sharing_histogram.clear();
508 m_write_sharing_histogram.clear();
509 m_all_sharing_histogram.clear();
510 m_cache_to_cache = 0;
511 m_memory_to_cache = 0;
512
513 // clear HashMaps
514 m_requestProfileMap.clear();
515
516 // count requests profiled
517 m_requests = 0;
518
519 m_outstanding_requests.clear();
520 m_outstanding_persistent_requests.clear();
521
522 // Flush the prefetches through the system - used so that there
523 // are no outstanding requests after stats are cleared
524 //g_eventQueue_ptr->triggerAllEvents();
525
526 // update the start time
527 m_ruby_start = g_system_ptr->getTime();
528}
529
530void
531Profiler::addAddressTraceSample(const RubyRequest& msg, NodeID id)
532{
533 if (msg.getType() != RubyRequestType_IFETCH) {
534 // Note: The following line should be commented out if you
535 // want to use the special profiling that is part of the GS320
536 // protocol
537
538 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
539 // profiled by the AddressProfiler
540 m_address_profiler_ptr->
541 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
542 msg.getType(), msg.getAccessMode(), id, false);
543 }
544}
545
546void
547Profiler::profileSharing(const Address& addr, AccessType type,
548 NodeID requestor, const Set& sharers,
549 const Set& owner)
550{
551 Set set_contacted(owner);
552 if (type == AccessType_Write) {
553 set_contacted.addSet(sharers);
554 }
555 set_contacted.remove(requestor);
556 int number_contacted = set_contacted.count();
557
558 if (type == AccessType_Write) {
559 m_write_sharing_histogram.add(number_contacted);
560 } else {
561 m_read_sharing_histogram.add(number_contacted);
562 }
563 m_all_sharing_histogram.add(number_contacted);
564
565 if (number_contacted == 0) {
566 m_memory_to_cache++;
567 } else {
568 m_cache_to_cache++;
569 }
570}
571
572void
573Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
574{
575 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
576 m_delayedCyclesHistogram.add(delayCycles);
577 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
578 if (virtualNetwork != 0) {
579 m_delayedCyclesNonPFHistogram.add(delayCycles);
580 }
581}
582
583// profiles original cache requests including PUTs
584void
585Profiler::profileRequest(const string& requestStr)
586{
587 m_requests++;
588
589 // if it doesn't exist, conveniently, it will be created with the
590 // default value which is 0
591 m_requestProfileMap[requestStr]++;
592}
593
594void
595Profiler::controllerBusy(MachineID machID)
596{
597 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
598}
599
600void
601Profiler::profilePFWait(Time waitTime)
602{
603 m_prefetchWaitHistogram.add(waitTime);
604}
605
606void
607Profiler::bankBusy()
608{
609 m_busyBankCount++;
610}
611
612// non-zero cycle demand request
613void
614Profiler::missLatency(Time cycles,
615 RubyRequestType type,
616 const GenericMachineType respondingMach)
617{
618 m_allMissLatencyHistogram.add(cycles);
619 m_missLatencyHistograms[type].add(cycles);
620 m_machLatencyHistograms[respondingMach].add(cycles);
621 m_missMachLatencyHistograms[type][respondingMach].add(cycles);
622}
623
624void
625Profiler::missLatencyWcc(Time issuedTime,
626 Time initialRequestTime,
627 Time forwardRequestTime,
628 Time firstResponseTime,
629 Time completionTime)
630{
631 if ((issuedTime <= initialRequestTime) &&
632 (initialRequestTime <= forwardRequestTime) &&
633 (forwardRequestTime <= firstResponseTime) &&
634 (firstResponseTime <= completionTime)) {
635 m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
636
637 m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
638 initialRequestTime);
639
640 m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
641 forwardRequestTime);
642
643 m_wCCFirstResponseToCompleteHistogram.add(completionTime -
644 firstResponseTime);
645 } else {
646 m_wCCIncompleteTimes++;
647 }
648}
649
650void
651Profiler::missLatencyDir(Time issuedTime,
652 Time initialRequestTime,
653 Time forwardRequestTime,
654 Time firstResponseTime,
655 Time completionTime)
656{
657 if ((issuedTime <= initialRequestTime) &&
658 (initialRequestTime <= forwardRequestTime) &&
659 (forwardRequestTime <= firstResponseTime) &&
660 (firstResponseTime <= completionTime)) {
661 m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
662
663 m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
664 initialRequestTime);
665
666 m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
667 forwardRequestTime);
668
669 m_dirFirstResponseToCompleteHistogram.add(completionTime -
670 firstResponseTime);
671 } else {
672 m_dirIncompleteTimes++;
673 }
674}
675
676// non-zero cycle prefetch request
677void
678Profiler::swPrefetchLatency(Time cycles,
679 RubyRequestType type,
680 const GenericMachineType respondingMach)
681{
682 m_allSWPrefetchLatencyHistogram.add(cycles);
683 m_SWPrefetchLatencyHistograms[type].add(cycles);
684 m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
685 if (respondingMach == GenericMachineType_Directory ||
686 respondingMach == GenericMachineType_NUM) {
687 m_SWPrefetchL2MissLatencyHistogram.add(cycles);
688 }
689}
690
691// Helper function
692static double
693process_memory_total()
694{
695 // 4kB page size, 1024*1024 bytes per MB,
696 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
697 ifstream proc_file;
698 proc_file.open("/proc/self/statm");
699 int total_size_in_pages = 0;
700 int res_size_in_pages = 0;
701 proc_file >> total_size_in_pages;
702 proc_file >> res_size_in_pages;
703 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
704}
705
706static double
707process_memory_resident()
708{
709 // 4kB page size, 1024*1024 bytes per MB,
710 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
711 ifstream proc_file;
712 proc_file.open("/proc/self/statm");
713 int total_size_in_pages = 0;
714 int res_size_in_pages = 0;
715 proc_file >> total_size_in_pages;
716 proc_file >> res_size_in_pages;
717 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
718}
719
720void
721Profiler::rubyWatch(int id)
722{
723 uint64 tr = 0;
724 Address watch_address = Address(tr);
725
726 DPRINTFN("%7s %3s RUBY WATCH %d\n", g_system_ptr->getTime(), id,
727 watch_address);
728
729 // don't care about success or failure
730 m_watch_address_set.insert(watch_address);
731}
732
733bool
734Profiler::watchAddress(Address addr)
735{
736 return m_watch_address_set.count(addr) > 0;
737}
738
739Profiler *
740RubyProfilerParams::create()
741{
742 return new Profiler(this);
743}
|