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>
| 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#include <fstream>
|
50
51#include "base/stl_helpers.hh"
52#include "base/str.hh"
53#include "mem/protocol/CacheMsg.hh"
54#include "mem/protocol/MachineType.hh"
55#include "mem/protocol/Protocol.hh"
| 51
52#include "base/stl_helpers.hh"
53#include "base/str.hh"
54#include "mem/protocol/CacheMsg.hh"
55#include "mem/protocol/MachineType.hh"
56#include "mem/protocol/Protocol.hh"
|
56#include "mem/ruby/common/Debug.hh"
| |
57#include "mem/ruby/network/Network.hh"
58#include "mem/ruby/profiler/AddressProfiler.hh"
59#include "mem/ruby/profiler/Profiler.hh"
60#include "mem/ruby/system/System.hh"
61#include "mem/ruby/system/System.hh"
62
63using namespace std;
64using m5::stl_helpers::operator<<;
65
| 57#include "mem/ruby/network/Network.hh"
58#include "mem/ruby/profiler/AddressProfiler.hh"
59#include "mem/ruby/profiler/Profiler.hh"
60#include "mem/ruby/system/System.hh"
61#include "mem/ruby/system/System.hh"
62
63using namespace std;
64using m5::stl_helpers::operator<<;
65
|
66extern ostream* debug_cout_ptr;
67
| |
68static double process_memory_total();
69static double process_memory_resident();
70
71Profiler::Profiler(const Params *p)
72 : SimObject(p)
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
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->getCycleCount(i) - 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_eventQueue_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 g_eventQueue_ptr->scheduleEvent(this, 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());
154 g_eventQueue_ptr->scheduleEvent(this, 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;
164 g_eventQueue_ptr->scheduleEvent(this, 1);
165}
166
167void
168Profiler::printConfig(ostream& out) const
169{
170 out << endl;
171 out << "Profiler Configuration" << endl;
172 out << "----------------------" << endl;
173 out << "periodic_stats_period: " << m_stats_period << endl;
174}
175
176void
177Profiler::print(ostream& out) const
178{
179 out << "[Profiler]";
180}
181
182void
183Profiler::printStats(ostream& out, bool short_stats)
184{
185 out << endl;
186 if (short_stats) {
187 out << "SHORT ";
188 }
189 out << "Profiler Stats" << endl;
190 out << "--------------" << endl;
191
192 time_t real_time_current = time(NULL);
193 double seconds = difftime(real_time_current, m_real_time_start_time);
194 double minutes = seconds / 60.0;
195 double hours = minutes / 60.0;
196 double days = hours / 24.0;
197 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
198
199 if (!short_stats) {
200 out << "Elapsed_time_in_seconds: " << seconds << endl;
201 out << "Elapsed_time_in_minutes: " << minutes << endl;
202 out << "Elapsed_time_in_hours: " << hours << endl;
203 out << "Elapsed_time_in_days: " << days << endl;
204 out << endl;
205 }
206
207 // print the virtual runtimes as well
208 struct tms vtime;
209 times(&vtime);
210 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
211 minutes = seconds / 60.0;
212 hours = minutes / 60.0;
213 days = hours / 24.0;
214 out << "Virtual_time_in_seconds: " << seconds << endl;
215 out << "Virtual_time_in_minutes: " << minutes << endl;
216 out << "Virtual_time_in_hours: " << hours << endl;
217 out << "Virtual_time_in_days: " << days << endl;
218 out << endl;
219
220 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
221 out << "Ruby_start_time: " << m_ruby_start << endl;
222 out << "Ruby_cycles: " << ruby_cycles << endl;
223 out << endl;
224
225 if (!short_stats) {
226 out << "mbytes_resident: " << process_memory_resident() << endl;
227 out << "mbytes_total: " << process_memory_total() << endl;
228 if (process_memory_total() > 0) {
229 out << "resident_ratio: "
230 << process_memory_resident()/process_memory_total() << endl;
231 }
232 out << endl;
233 }
234
235 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
236
237 for (int i = 0; i < m_num_of_sequencers; i++) {
238 perProcCycleCount[i] =
239 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
240 // The +1 allows us to avoid division by zero
241 }
242
243 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
244
245 out << endl;
246
247 if (!short_stats) {
248 out << "Busy Controller Counts:" << endl;
249 for (int i = 0; i < MachineType_NUM; i++) {
250 int size = MachineType_base_count((MachineType)i);
251 for (int j = 0; j < size; j++) {
252 MachineID machID;
253 machID.type = (MachineType)i;
254 machID.num = j;
255 out << machID << ":" << m_busyControllerCount[i][j] << " ";
256 if ((j + 1) % 8 == 0) {
257 out << endl;
258 }
259 }
260 out << endl;
261 }
262 out << endl;
263
264 out << "Busy Bank Count:" << m_busyBankCount << endl;
265 out << endl;
266
267 out << "sequencer_requests_outstanding: "
268 << m_sequencer_requests << endl;
269 out << endl;
270 }
271
272 if (!short_stats) {
273 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
274 out << "----------------------------------------" << endl;
275 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
276 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
277 if (m_missLatencyHistograms[i].size() > 0) {
278 out << "miss_latency_" << RubyRequestType(i) << ": "
279 << m_missLatencyHistograms[i] << endl;
280 }
281 }
282 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
283 if (m_machLatencyHistograms[i].size() > 0) {
284 out << "miss_latency_" << GenericMachineType(i) << ": "
285 << m_machLatencyHistograms[i] << endl;
286 }
287 }
288
289 out << "miss_latency_wCC_issue_to_initial_request: "
290 << m_wCCIssueToInitialRequestHistogram << endl;
291 out << "miss_latency_wCC_initial_forward_request: "
292 << m_wCCInitialRequestToForwardRequestHistogram << endl;
293 out << "miss_latency_wCC_forward_to_first_response: "
294 << m_wCCForwardRequestToFirstResponseHistogram << endl;
295 out << "miss_latency_wCC_first_response_to_completion: "
296 << m_wCCFirstResponseToCompleteHistogram << endl;
297 out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
298 out << "miss_latency_dir_issue_to_initial_request: "
299 << m_dirIssueToInitialRequestHistogram << endl;
300 out << "miss_latency_dir_initial_forward_request: "
301 << m_dirInitialRequestToForwardRequestHistogram << endl;
302 out << "miss_latency_dir_forward_to_first_response: "
303 << m_dirForwardRequestToFirstResponseHistogram << endl;
304 out << "miss_latency_dir_first_response_to_completion: "
305 << m_dirFirstResponseToCompleteHistogram << endl;
306 out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
307
308 for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
309 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
310 if (m_missMachLatencyHistograms[i][j].size() > 0) {
311 out << "miss_latency_" << RubyRequestType(i)
312 << "_" << GenericMachineType(j) << ": "
313 << m_missMachLatencyHistograms[i][j] << endl;
314 }
315 }
316 }
317
318 out << endl;
319
320 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
321 out << "------------------------------------" << endl;
322 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
323 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
324 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
325 out << "prefetch_latency_" << CacheRequestType(i) << ": "
326 << m_SWPrefetchLatencyHistograms[i] << endl;
327 }
328 }
329 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
330 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
331 out << "prefetch_latency_" << GenericMachineType(i) << ": "
332 << m_SWPrefetchMachLatencyHistograms[i] << endl;
333 }
334 }
335 out << "prefetch_latency_L2Miss:"
336 << m_SWPrefetchL2MissLatencyHistogram << endl;
337
338 if (m_all_sharing_histogram.size() > 0) {
339 out << "all_sharing: " << m_all_sharing_histogram << endl;
340 out << "read_sharing: " << m_read_sharing_histogram << endl;
341 out << "write_sharing: " << m_write_sharing_histogram << endl;
342
343 out << "all_sharing_percent: ";
344 m_all_sharing_histogram.printPercent(out);
345 out << endl;
346
347 out << "read_sharing_percent: ";
348 m_read_sharing_histogram.printPercent(out);
349 out << endl;
350
351 out << "write_sharing_percent: ";
352 m_write_sharing_histogram.printPercent(out);
353 out << endl;
354
355 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
356 out << "all_misses: " << total_miss << endl;
357 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
358 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
359 out << "cache_to_cache_percent: "
360 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
361 << endl;
362 out << "memory_to_cache_percent: "
363 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
364 << endl;
365 out << endl;
366 }
367
368 if (m_outstanding_requests.size() > 0) {
369 out << "outstanding_requests: ";
370 m_outstanding_requests.printPercent(out);
371 out << endl;
372 out << endl;
373 }
374 }
375
376 if (!short_stats) {
377 out << "Request vs. RubySystem State Profile" << endl;
378 out << "--------------------------------" << endl;
379 out << endl;
380
381 map<string, int>::const_iterator i = m_requestProfileMap.begin();
382 map<string, int>::const_iterator end = m_requestProfileMap.end();
383 for (; i != end; ++i) {
384 const string &key = i->first;
385 int count = i->second;
386
387 double percent = (100.0 * double(count)) / double(m_requests);
388 vector<string> items;
389 tokenize(items, key, ':');
390 vector<string>::iterator j = items.begin();
391 vector<string>::iterator end = items.end();
392 for (; j != end; ++i)
393 out << setw(10) << *j;
394 out << setw(11) << count;
395 out << setw(14) << percent << endl;
396 }
397 out << endl;
398
399 out << "filter_action: " << m_filter_action_histogram << endl;
400
401 if (!m_all_instructions) {
402 m_address_profiler_ptr->printStats(out);
403 }
404
405 if (m_all_instructions) {
406 m_inst_profiler_ptr->printStats(out);
407 }
408
409 out << endl;
410 out << "Message Delayed Cycles" << endl;
411 out << "----------------------" << endl;
412 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
413 out << "Total_nonPF_delay_cycles: "
414 << m_delayedCyclesNonPFHistogram << endl;
415 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
416 out << " virtual_network_" << i << "_delay_cycles: "
417 << m_delayedCyclesVCHistograms[i] << endl;
418 }
419
420 printResourceUsage(out);
421 }
422}
423
424void
425Profiler::printResourceUsage(ostream& out) const
426{
427 out << endl;
428 out << "Resource Usage" << endl;
429 out << "--------------" << endl;
430
431 integer_t pagesize = getpagesize(); // page size in bytes
432 out << "page_size: " << pagesize << endl;
433
434 rusage usage;
435 getrusage (RUSAGE_SELF, &usage);
436
437 out << "user_time: " << usage.ru_utime.tv_sec << endl;
438 out << "system_time: " << usage.ru_stime.tv_sec << endl;
439 out << "page_reclaims: " << usage.ru_minflt << endl;
440 out << "page_faults: " << usage.ru_majflt << endl;
441 out << "swaps: " << usage.ru_nswap << endl;
442 out << "block_inputs: " << usage.ru_inblock << endl;
443 out << "block_outputs: " << usage.ru_oublock << endl;
444}
445
446void
447Profiler::clearStats()
448{
449 m_ruby_start = g_eventQueue_ptr->getTime();
450
451 m_cycles_executed_at_start.resize(m_num_of_sequencers);
452 for (int i = 0; i < m_num_of_sequencers; i++) {
453 if (g_system_ptr == NULL) {
454 m_cycles_executed_at_start[i] = 0;
455 } else {
456 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
457 }
458 }
459
460 m_busyControllerCount.resize(MachineType_NUM); // all machines
461 for (int i = 0; i < MachineType_NUM; i++) {
462 int size = MachineType_base_count((MachineType)i);
463 m_busyControllerCount[i].resize(size);
464 for (int j = 0; j < size; j++) {
465 m_busyControllerCount[i][j] = 0;
466 }
467 }
468 m_busyBankCount = 0;
469
470 m_delayedCyclesHistogram.clear();
471 m_delayedCyclesNonPFHistogram.clear();
472 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
473 m_delayedCyclesVCHistograms.resize(size);
474 for (int i = 0; i < size; i++) {
475 m_delayedCyclesVCHistograms[i].clear();
476 }
477
478 m_missLatencyHistograms.resize(RubyRequestType_NUM);
479 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
480 m_missLatencyHistograms[i].clear(200);
481 }
482 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
483 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
484 m_machLatencyHistograms[i].clear(200);
485 }
486 m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
487 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
488 m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
489 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
490 m_missMachLatencyHistograms[i][j].clear(200);
491 }
492 }
493 m_allMissLatencyHistogram.clear(200);
494 m_wCCIssueToInitialRequestHistogram.clear(200);
495 m_wCCInitialRequestToForwardRequestHistogram.clear(200);
496 m_wCCForwardRequestToFirstResponseHistogram.clear(200);
497 m_wCCFirstResponseToCompleteHistogram.clear(200);
498 m_wCCIncompleteTimes = 0;
499 m_dirIssueToInitialRequestHistogram.clear(200);
500 m_dirInitialRequestToForwardRequestHistogram.clear(200);
501 m_dirForwardRequestToFirstResponseHistogram.clear(200);
502 m_dirFirstResponseToCompleteHistogram.clear(200);
503 m_dirIncompleteTimes = 0;
504
505 m_SWPrefetchLatencyHistograms.resize(CacheRequestType_NUM);
506 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
507 m_SWPrefetchLatencyHistograms[i].clear(200);
508 }
509 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
510 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
511 m_SWPrefetchMachLatencyHistograms[i].clear(200);
512 }
513 m_allSWPrefetchLatencyHistogram.clear(200);
514
515 m_sequencer_requests.clear();
516 m_read_sharing_histogram.clear();
517 m_write_sharing_histogram.clear();
518 m_all_sharing_histogram.clear();
519 m_cache_to_cache = 0;
520 m_memory_to_cache = 0;
521
522 // clear HashMaps
523 m_requestProfileMap.clear();
524
525 // count requests profiled
526 m_requests = 0;
527
528 m_outstanding_requests.clear();
529 m_outstanding_persistent_requests.clear();
530
531 // Flush the prefetches through the system - used so that there
532 // are no outstanding requests after stats are cleared
533 //g_eventQueue_ptr->triggerAllEvents();
534
535 // update the start time
536 m_ruby_start = g_eventQueue_ptr->getTime();
537}
538
539void
540Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id)
541{
542 if (msg.getType() != CacheRequestType_IFETCH) {
543 // Note: The following line should be commented out if you
544 // want to use the special profiling that is part of the GS320
545 // protocol
546
547 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
548 // profiled by the AddressProfiler
549 m_address_profiler_ptr->
550 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
551 msg.getType(), msg.getAccessMode(), id, false);
552 }
553}
554
555void
556Profiler::profileSharing(const Address& addr, AccessType type,
557 NodeID requestor, const Set& sharers,
558 const Set& owner)
559{
560 Set set_contacted(owner);
561 if (type == AccessType_Write) {
562 set_contacted.addSet(sharers);
563 }
564 set_contacted.remove(requestor);
565 int number_contacted = set_contacted.count();
566
567 if (type == AccessType_Write) {
568 m_write_sharing_histogram.add(number_contacted);
569 } else {
570 m_read_sharing_histogram.add(number_contacted);
571 }
572 m_all_sharing_histogram.add(number_contacted);
573
574 if (number_contacted == 0) {
575 m_memory_to_cache++;
576 } else {
577 m_cache_to_cache++;
578 }
579}
580
581void
582Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
583{
584 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
585 m_delayedCyclesHistogram.add(delayCycles);
586 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
587 if (virtualNetwork != 0) {
588 m_delayedCyclesNonPFHistogram.add(delayCycles);
589 }
590}
591
592// profiles original cache requests including PUTs
593void
594Profiler::profileRequest(const string& requestStr)
595{
596 m_requests++;
597
598 // if it doesn't exist, conveniently, it will be created with the
599 // default value which is 0
600 m_requestProfileMap[requestStr]++;
601}
602
603void
604Profiler::controllerBusy(MachineID machID)
605{
606 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
607}
608
609void
610Profiler::profilePFWait(Time waitTime)
611{
612 m_prefetchWaitHistogram.add(waitTime);
613}
614
615void
616Profiler::bankBusy()
617{
618 m_busyBankCount++;
619}
620
621// non-zero cycle demand request
622void
623Profiler::missLatency(Time cycles,
624 RubyRequestType type,
625 const GenericMachineType respondingMach)
626{
627 m_allMissLatencyHistogram.add(cycles);
628 m_missLatencyHistograms[type].add(cycles);
629 m_machLatencyHistograms[respondingMach].add(cycles);
630 m_missMachLatencyHistograms[type][respondingMach].add(cycles);
631}
632
633void
634Profiler::missLatencyWcc(Time issuedTime,
635 Time initialRequestTime,
636 Time forwardRequestTime,
637 Time firstResponseTime,
638 Time completionTime)
639{
640 if ((issuedTime <= initialRequestTime) &&
641 (initialRequestTime <= forwardRequestTime) &&
642 (forwardRequestTime <= firstResponseTime) &&
643 (firstResponseTime <= completionTime)) {
644 m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
645
646 m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
647 initialRequestTime);
648
649 m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
650 forwardRequestTime);
651
652 m_wCCFirstResponseToCompleteHistogram.add(completionTime -
653 firstResponseTime);
654 } else {
655 m_wCCIncompleteTimes++;
656 }
657}
658
659void
660Profiler::missLatencyDir(Time issuedTime,
661 Time initialRequestTime,
662 Time forwardRequestTime,
663 Time firstResponseTime,
664 Time completionTime)
665{
666 if ((issuedTime <= initialRequestTime) &&
667 (initialRequestTime <= forwardRequestTime) &&
668 (forwardRequestTime <= firstResponseTime) &&
669 (firstResponseTime <= completionTime)) {
670 m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
671
672 m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
673 initialRequestTime);
674
675 m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
676 forwardRequestTime);
677
678 m_dirFirstResponseToCompleteHistogram.add(completionTime -
679 firstResponseTime);
680 } else {
681 m_dirIncompleteTimes++;
682 }
683}
684
685// non-zero cycle prefetch request
686void
687Profiler::swPrefetchLatency(Time cycles,
688 CacheRequestType type,
689 const GenericMachineType respondingMach)
690{
691 m_allSWPrefetchLatencyHistogram.add(cycles);
692 m_SWPrefetchLatencyHistograms[type].add(cycles);
693 m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
694 if (respondingMach == GenericMachineType_Directory ||
695 respondingMach == GenericMachineType_NUM) {
696 m_SWPrefetchL2MissLatencyHistogram.add(cycles);
697 }
698}
699
| 66static double process_memory_total();
67static double process_memory_resident();
68
69Profiler::Profiler(const Params *p)
70 : SimObject(p)
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
98Profiler::~Profiler()
99{
100 if (m_periodic_output_file_ptr != &cerr) {
101 delete m_periodic_output_file_ptr;
102 }
103}
104
105void
106Profiler::wakeup()
107{
108 // FIXME - avoid the repeated code
109
110 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
111
112 for (int i = 0; i < m_num_of_sequencers; i++) {
113 perProcCycleCount[i] =
114 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
115 // The +1 allows us to avoid division by zero
116 }
117
118 ostream &out = *m_periodic_output_file_ptr;
119
120 out << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl
121 << "mbytes_resident: " << process_memory_resident() << endl
122 << "mbytes_total: " << process_memory_total() << endl;
123
124 if (process_memory_total() > 0) {
125 out << "resident_ratio: "
126 << process_memory_resident() / process_memory_total() << endl;
127 }
128
129 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
130
131 out << endl;
132
133 if (m_all_instructions) {
134 m_inst_profiler_ptr->printStats(out);
135 }
136
137 //g_system_ptr->getNetwork()->printStats(out);
138 g_eventQueue_ptr->scheduleEvent(this, m_stats_period);
139}
140
141void
142Profiler::setPeriodicStatsFile(const string& filename)
143{
144 cout << "Recording periodic statistics to file '" << filename << "' every "
145 << m_stats_period << " Ruby cycles" << endl;
146
147 if (m_periodic_output_file_ptr != &cerr) {
148 delete m_periodic_output_file_ptr;
149 }
150
151 m_periodic_output_file_ptr = new ofstream(filename.c_str());
152 g_eventQueue_ptr->scheduleEvent(this, 1);
153}
154
155void
156Profiler::setPeriodicStatsInterval(integer_t period)
157{
158 cout << "Recording periodic statistics every " << m_stats_period
159 << " Ruby cycles" << endl;
160
161 m_stats_period = period;
162 g_eventQueue_ptr->scheduleEvent(this, 1);
163}
164
165void
166Profiler::printConfig(ostream& out) const
167{
168 out << endl;
169 out << "Profiler Configuration" << endl;
170 out << "----------------------" << endl;
171 out << "periodic_stats_period: " << m_stats_period << endl;
172}
173
174void
175Profiler::print(ostream& out) const
176{
177 out << "[Profiler]";
178}
179
180void
181Profiler::printStats(ostream& out, bool short_stats)
182{
183 out << endl;
184 if (short_stats) {
185 out << "SHORT ";
186 }
187 out << "Profiler Stats" << endl;
188 out << "--------------" << endl;
189
190 time_t real_time_current = time(NULL);
191 double seconds = difftime(real_time_current, m_real_time_start_time);
192 double minutes = seconds / 60.0;
193 double hours = minutes / 60.0;
194 double days = hours / 24.0;
195 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
196
197 if (!short_stats) {
198 out << "Elapsed_time_in_seconds: " << seconds << endl;
199 out << "Elapsed_time_in_minutes: " << minutes << endl;
200 out << "Elapsed_time_in_hours: " << hours << endl;
201 out << "Elapsed_time_in_days: " << days << endl;
202 out << endl;
203 }
204
205 // print the virtual runtimes as well
206 struct tms vtime;
207 times(&vtime);
208 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
209 minutes = seconds / 60.0;
210 hours = minutes / 60.0;
211 days = hours / 24.0;
212 out << "Virtual_time_in_seconds: " << seconds << endl;
213 out << "Virtual_time_in_minutes: " << minutes << endl;
214 out << "Virtual_time_in_hours: " << hours << endl;
215 out << "Virtual_time_in_days: " << days << endl;
216 out << endl;
217
218 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
219 out << "Ruby_start_time: " << m_ruby_start << endl;
220 out << "Ruby_cycles: " << ruby_cycles << endl;
221 out << endl;
222
223 if (!short_stats) {
224 out << "mbytes_resident: " << process_memory_resident() << endl;
225 out << "mbytes_total: " << process_memory_total() << endl;
226 if (process_memory_total() > 0) {
227 out << "resident_ratio: "
228 << process_memory_resident()/process_memory_total() << endl;
229 }
230 out << endl;
231 }
232
233 vector<integer_t> perProcCycleCount(m_num_of_sequencers);
234
235 for (int i = 0; i < m_num_of_sequencers; i++) {
236 perProcCycleCount[i] =
237 g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
238 // The +1 allows us to avoid division by zero
239 }
240
241 out << "ruby_cycles_executed: " << perProcCycleCount << endl;
242
243 out << endl;
244
245 if (!short_stats) {
246 out << "Busy Controller Counts:" << endl;
247 for (int i = 0; i < MachineType_NUM; i++) {
248 int size = MachineType_base_count((MachineType)i);
249 for (int j = 0; j < size; j++) {
250 MachineID machID;
251 machID.type = (MachineType)i;
252 machID.num = j;
253 out << machID << ":" << m_busyControllerCount[i][j] << " ";
254 if ((j + 1) % 8 == 0) {
255 out << endl;
256 }
257 }
258 out << endl;
259 }
260 out << endl;
261
262 out << "Busy Bank Count:" << m_busyBankCount << endl;
263 out << endl;
264
265 out << "sequencer_requests_outstanding: "
266 << m_sequencer_requests << endl;
267 out << endl;
268 }
269
270 if (!short_stats) {
271 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
272 out << "----------------------------------------" << endl;
273 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
274 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
275 if (m_missLatencyHistograms[i].size() > 0) {
276 out << "miss_latency_" << RubyRequestType(i) << ": "
277 << m_missLatencyHistograms[i] << endl;
278 }
279 }
280 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
281 if (m_machLatencyHistograms[i].size() > 0) {
282 out << "miss_latency_" << GenericMachineType(i) << ": "
283 << m_machLatencyHistograms[i] << endl;
284 }
285 }
286
287 out << "miss_latency_wCC_issue_to_initial_request: "
288 << m_wCCIssueToInitialRequestHistogram << endl;
289 out << "miss_latency_wCC_initial_forward_request: "
290 << m_wCCInitialRequestToForwardRequestHistogram << endl;
291 out << "miss_latency_wCC_forward_to_first_response: "
292 << m_wCCForwardRequestToFirstResponseHistogram << endl;
293 out << "miss_latency_wCC_first_response_to_completion: "
294 << m_wCCFirstResponseToCompleteHistogram << endl;
295 out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
296 out << "miss_latency_dir_issue_to_initial_request: "
297 << m_dirIssueToInitialRequestHistogram << endl;
298 out << "miss_latency_dir_initial_forward_request: "
299 << m_dirInitialRequestToForwardRequestHistogram << endl;
300 out << "miss_latency_dir_forward_to_first_response: "
301 << m_dirForwardRequestToFirstResponseHistogram << endl;
302 out << "miss_latency_dir_first_response_to_completion: "
303 << m_dirFirstResponseToCompleteHistogram << endl;
304 out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
305
306 for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
307 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
308 if (m_missMachLatencyHistograms[i][j].size() > 0) {
309 out << "miss_latency_" << RubyRequestType(i)
310 << "_" << GenericMachineType(j) << ": "
311 << m_missMachLatencyHistograms[i][j] << endl;
312 }
313 }
314 }
315
316 out << endl;
317
318 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
319 out << "------------------------------------" << endl;
320 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
321 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
322 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
323 out << "prefetch_latency_" << CacheRequestType(i) << ": "
324 << m_SWPrefetchLatencyHistograms[i] << endl;
325 }
326 }
327 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
328 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
329 out << "prefetch_latency_" << GenericMachineType(i) << ": "
330 << m_SWPrefetchMachLatencyHistograms[i] << endl;
331 }
332 }
333 out << "prefetch_latency_L2Miss:"
334 << m_SWPrefetchL2MissLatencyHistogram << endl;
335
336 if (m_all_sharing_histogram.size() > 0) {
337 out << "all_sharing: " << m_all_sharing_histogram << endl;
338 out << "read_sharing: " << m_read_sharing_histogram << endl;
339 out << "write_sharing: " << m_write_sharing_histogram << endl;
340
341 out << "all_sharing_percent: ";
342 m_all_sharing_histogram.printPercent(out);
343 out << endl;
344
345 out << "read_sharing_percent: ";
346 m_read_sharing_histogram.printPercent(out);
347 out << endl;
348
349 out << "write_sharing_percent: ";
350 m_write_sharing_histogram.printPercent(out);
351 out << endl;
352
353 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
354 out << "all_misses: " << total_miss << endl;
355 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
356 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
357 out << "cache_to_cache_percent: "
358 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
359 << endl;
360 out << "memory_to_cache_percent: "
361 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
362 << endl;
363 out << endl;
364 }
365
366 if (m_outstanding_requests.size() > 0) {
367 out << "outstanding_requests: ";
368 m_outstanding_requests.printPercent(out);
369 out << endl;
370 out << endl;
371 }
372 }
373
374 if (!short_stats) {
375 out << "Request vs. RubySystem State Profile" << endl;
376 out << "--------------------------------" << endl;
377 out << endl;
378
379 map<string, int>::const_iterator i = m_requestProfileMap.begin();
380 map<string, int>::const_iterator end = m_requestProfileMap.end();
381 for (; i != end; ++i) {
382 const string &key = i->first;
383 int count = i->second;
384
385 double percent = (100.0 * double(count)) / double(m_requests);
386 vector<string> items;
387 tokenize(items, key, ':');
388 vector<string>::iterator j = items.begin();
389 vector<string>::iterator end = items.end();
390 for (; j != end; ++i)
391 out << setw(10) << *j;
392 out << setw(11) << count;
393 out << setw(14) << percent << endl;
394 }
395 out << endl;
396
397 out << "filter_action: " << m_filter_action_histogram << endl;
398
399 if (!m_all_instructions) {
400 m_address_profiler_ptr->printStats(out);
401 }
402
403 if (m_all_instructions) {
404 m_inst_profiler_ptr->printStats(out);
405 }
406
407 out << endl;
408 out << "Message Delayed Cycles" << endl;
409 out << "----------------------" << endl;
410 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
411 out << "Total_nonPF_delay_cycles: "
412 << m_delayedCyclesNonPFHistogram << endl;
413 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
414 out << " virtual_network_" << i << "_delay_cycles: "
415 << m_delayedCyclesVCHistograms[i] << endl;
416 }
417
418 printResourceUsage(out);
419 }
420}
421
422void
423Profiler::printResourceUsage(ostream& out) const
424{
425 out << endl;
426 out << "Resource Usage" << endl;
427 out << "--------------" << endl;
428
429 integer_t pagesize = getpagesize(); // page size in bytes
430 out << "page_size: " << pagesize << endl;
431
432 rusage usage;
433 getrusage (RUSAGE_SELF, &usage);
434
435 out << "user_time: " << usage.ru_utime.tv_sec << endl;
436 out << "system_time: " << usage.ru_stime.tv_sec << endl;
437 out << "page_reclaims: " << usage.ru_minflt << endl;
438 out << "page_faults: " << usage.ru_majflt << endl;
439 out << "swaps: " << usage.ru_nswap << endl;
440 out << "block_inputs: " << usage.ru_inblock << endl;
441 out << "block_outputs: " << usage.ru_oublock << endl;
442}
443
444void
445Profiler::clearStats()
446{
447 m_ruby_start = g_eventQueue_ptr->getTime();
448
449 m_cycles_executed_at_start.resize(m_num_of_sequencers);
450 for (int i = 0; i < m_num_of_sequencers; i++) {
451 if (g_system_ptr == NULL) {
452 m_cycles_executed_at_start[i] = 0;
453 } else {
454 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
455 }
456 }
457
458 m_busyControllerCount.resize(MachineType_NUM); // all machines
459 for (int i = 0; i < MachineType_NUM; i++) {
460 int size = MachineType_base_count((MachineType)i);
461 m_busyControllerCount[i].resize(size);
462 for (int j = 0; j < size; j++) {
463 m_busyControllerCount[i][j] = 0;
464 }
465 }
466 m_busyBankCount = 0;
467
468 m_delayedCyclesHistogram.clear();
469 m_delayedCyclesNonPFHistogram.clear();
470 int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
471 m_delayedCyclesVCHistograms.resize(size);
472 for (int i = 0; i < size; i++) {
473 m_delayedCyclesVCHistograms[i].clear();
474 }
475
476 m_missLatencyHistograms.resize(RubyRequestType_NUM);
477 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
478 m_missLatencyHistograms[i].clear(200);
479 }
480 m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
481 for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
482 m_machLatencyHistograms[i].clear(200);
483 }
484 m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
485 for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
486 m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
487 for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
488 m_missMachLatencyHistograms[i][j].clear(200);
489 }
490 }
491 m_allMissLatencyHistogram.clear(200);
492 m_wCCIssueToInitialRequestHistogram.clear(200);
493 m_wCCInitialRequestToForwardRequestHistogram.clear(200);
494 m_wCCForwardRequestToFirstResponseHistogram.clear(200);
495 m_wCCFirstResponseToCompleteHistogram.clear(200);
496 m_wCCIncompleteTimes = 0;
497 m_dirIssueToInitialRequestHistogram.clear(200);
498 m_dirInitialRequestToForwardRequestHistogram.clear(200);
499 m_dirForwardRequestToFirstResponseHistogram.clear(200);
500 m_dirFirstResponseToCompleteHistogram.clear(200);
501 m_dirIncompleteTimes = 0;
502
503 m_SWPrefetchLatencyHistograms.resize(CacheRequestType_NUM);
504 for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
505 m_SWPrefetchLatencyHistograms[i].clear(200);
506 }
507 m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
508 for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
509 m_SWPrefetchMachLatencyHistograms[i].clear(200);
510 }
511 m_allSWPrefetchLatencyHistogram.clear(200);
512
513 m_sequencer_requests.clear();
514 m_read_sharing_histogram.clear();
515 m_write_sharing_histogram.clear();
516 m_all_sharing_histogram.clear();
517 m_cache_to_cache = 0;
518 m_memory_to_cache = 0;
519
520 // clear HashMaps
521 m_requestProfileMap.clear();
522
523 // count requests profiled
524 m_requests = 0;
525
526 m_outstanding_requests.clear();
527 m_outstanding_persistent_requests.clear();
528
529 // Flush the prefetches through the system - used so that there
530 // are no outstanding requests after stats are cleared
531 //g_eventQueue_ptr->triggerAllEvents();
532
533 // update the start time
534 m_ruby_start = g_eventQueue_ptr->getTime();
535}
536
537void
538Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id)
539{
540 if (msg.getType() != CacheRequestType_IFETCH) {
541 // Note: The following line should be commented out if you
542 // want to use the special profiling that is part of the GS320
543 // protocol
544
545 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
546 // profiled by the AddressProfiler
547 m_address_profiler_ptr->
548 addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
549 msg.getType(), msg.getAccessMode(), id, false);
550 }
551}
552
553void
554Profiler::profileSharing(const Address& addr, AccessType type,
555 NodeID requestor, const Set& sharers,
556 const Set& owner)
557{
558 Set set_contacted(owner);
559 if (type == AccessType_Write) {
560 set_contacted.addSet(sharers);
561 }
562 set_contacted.remove(requestor);
563 int number_contacted = set_contacted.count();
564
565 if (type == AccessType_Write) {
566 m_write_sharing_histogram.add(number_contacted);
567 } else {
568 m_read_sharing_histogram.add(number_contacted);
569 }
570 m_all_sharing_histogram.add(number_contacted);
571
572 if (number_contacted == 0) {
573 m_memory_to_cache++;
574 } else {
575 m_cache_to_cache++;
576 }
577}
578
579void
580Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
581{
582 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
583 m_delayedCyclesHistogram.add(delayCycles);
584 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
585 if (virtualNetwork != 0) {
586 m_delayedCyclesNonPFHistogram.add(delayCycles);
587 }
588}
589
590// profiles original cache requests including PUTs
591void
592Profiler::profileRequest(const string& requestStr)
593{
594 m_requests++;
595
596 // if it doesn't exist, conveniently, it will be created with the
597 // default value which is 0
598 m_requestProfileMap[requestStr]++;
599}
600
601void
602Profiler::controllerBusy(MachineID machID)
603{
604 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
605}
606
607void
608Profiler::profilePFWait(Time waitTime)
609{
610 m_prefetchWaitHistogram.add(waitTime);
611}
612
613void
614Profiler::bankBusy()
615{
616 m_busyBankCount++;
617}
618
619// non-zero cycle demand request
620void
621Profiler::missLatency(Time cycles,
622 RubyRequestType type,
623 const GenericMachineType respondingMach)
624{
625 m_allMissLatencyHistogram.add(cycles);
626 m_missLatencyHistograms[type].add(cycles);
627 m_machLatencyHistograms[respondingMach].add(cycles);
628 m_missMachLatencyHistograms[type][respondingMach].add(cycles);
629}
630
631void
632Profiler::missLatencyWcc(Time issuedTime,
633 Time initialRequestTime,
634 Time forwardRequestTime,
635 Time firstResponseTime,
636 Time completionTime)
637{
638 if ((issuedTime <= initialRequestTime) &&
639 (initialRequestTime <= forwardRequestTime) &&
640 (forwardRequestTime <= firstResponseTime) &&
641 (firstResponseTime <= completionTime)) {
642 m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
643
644 m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
645 initialRequestTime);
646
647 m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
648 forwardRequestTime);
649
650 m_wCCFirstResponseToCompleteHistogram.add(completionTime -
651 firstResponseTime);
652 } else {
653 m_wCCIncompleteTimes++;
654 }
655}
656
657void
658Profiler::missLatencyDir(Time issuedTime,
659 Time initialRequestTime,
660 Time forwardRequestTime,
661 Time firstResponseTime,
662 Time completionTime)
663{
664 if ((issuedTime <= initialRequestTime) &&
665 (initialRequestTime <= forwardRequestTime) &&
666 (forwardRequestTime <= firstResponseTime) &&
667 (firstResponseTime <= completionTime)) {
668 m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
669
670 m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
671 initialRequestTime);
672
673 m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
674 forwardRequestTime);
675
676 m_dirFirstResponseToCompleteHistogram.add(completionTime -
677 firstResponseTime);
678 } else {
679 m_dirIncompleteTimes++;
680 }
681}
682
683// non-zero cycle prefetch request
684void
685Profiler::swPrefetchLatency(Time cycles,
686 CacheRequestType type,
687 const GenericMachineType respondingMach)
688{
689 m_allSWPrefetchLatencyHistogram.add(cycles);
690 m_SWPrefetchLatencyHistograms[type].add(cycles);
691 m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
692 if (respondingMach == GenericMachineType_Directory ||
693 respondingMach == GenericMachineType_NUM) {
694 m_SWPrefetchL2MissLatencyHistogram.add(cycles);
695 }
696}
697
|
700void
701Profiler::profileTransition(const string& component, NodeID version,
702 Address addr, const string& state, const string& event,
703 const string& next_state, const string& note)
704{
705 const int EVENT_SPACES = 20;
706 const int ID_SPACES = 3;
707 const int TIME_SPACES = 7;
708 const int COMP_SPACES = 10;
709 const int STATE_SPACES = 6;
710
711 if (g_debug_ptr->getDebugTime() <= 0 ||
712 g_eventQueue_ptr->getTime() < g_debug_ptr->getDebugTime())
713 return;
714
715 ostream &out = *debug_cout_ptr;
716 out.flags(ios::right);
717 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
718 out << setw(ID_SPACES) << version << " ";
719 out << setw(COMP_SPACES) << component;
720 out << setw(EVENT_SPACES) << event << " ";
721
722 out.flags(ios::right);
723 out << setw(STATE_SPACES) << state;
724 out << ">";
725 out.flags(ios::left);
726 out << setw(STATE_SPACES) << next_state;
727
728 out << " " << addr << " " << note;
729
730 out << endl;
731}
732
| |
733// Helper function
734static double
735process_memory_total()
736{
737 // 4kB page size, 1024*1024 bytes per MB,
738 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
739 ifstream proc_file;
740 proc_file.open("/proc/self/statm");
741 int total_size_in_pages = 0;
742 int res_size_in_pages = 0;
743 proc_file >> total_size_in_pages;
744 proc_file >> res_size_in_pages;
745 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
746}
747
748static double
749process_memory_resident()
750{
751 // 4kB page size, 1024*1024 bytes per MB,
752 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
753 ifstream proc_file;
754 proc_file.open("/proc/self/statm");
755 int total_size_in_pages = 0;
756 int res_size_in_pages = 0;
757 proc_file >> total_size_in_pages;
758 proc_file >> res_size_in_pages;
759 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
760}
761
762void
763Profiler::rubyWatch(int id)
764{
765 uint64 tr = 0;
766 Address watch_address = Address(tr);
| 698// Helper function
699static double
700process_memory_total()
701{
702 // 4kB page size, 1024*1024 bytes per MB,
703 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
704 ifstream proc_file;
705 proc_file.open("/proc/self/statm");
706 int total_size_in_pages = 0;
707 int res_size_in_pages = 0;
708 proc_file >> total_size_in_pages;
709 proc_file >> res_size_in_pages;
710 return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
711}
712
713static double
714process_memory_resident()
715{
716 // 4kB page size, 1024*1024 bytes per MB,
717 const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
718 ifstream proc_file;
719 proc_file.open("/proc/self/statm");
720 int total_size_in_pages = 0;
721 int res_size_in_pages = 0;
722 proc_file >> total_size_in_pages;
723 proc_file >> res_size_in_pages;
724 return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
725}
726
727void
728Profiler::rubyWatch(int id)
729{
730 uint64 tr = 0;
731 Address watch_address = Address(tr);
|
767 const int ID_SPACES = 3;
768 const int TIME_SPACES = 7;
| |
769
| 732
|
770 ostream &out = *debug_cout_ptr;
| 733 DPRINTFN("%7s %3s RUBY WATCH %d\n", g_eventQueue_ptr->getTime(), id,
734 watch_address);
|
771
| 735
|
772 out.flags(ios::right);
773 out << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
774 out << setw(ID_SPACES) << id << " "
775 << "RUBY WATCH " << watch_address << endl;
776
| |
777 // don't care about success or failure
778 m_watch_address_set.insert(watch_address);
779}
780
781bool
782Profiler::watchAddress(Address addr)
783{
784 return m_watch_address_set.count(addr) > 0;
785}
786
787Profiler *
788RubyProfilerParams::create()
789{
790 return new Profiler(this);
791}
| 736 // don't care about success or failure
737 m_watch_address_set.insert(watch_address);
738}
739
740bool
741Profiler::watchAddress(Address addr)
742{
743 return m_watch_address_set.count(addr) > 0;
744}
745
746Profiler *
747RubyProfilerParams::create()
748{
749 return new Profiler(this);
750}
|