Profiler.cc revision 8436:5648986156db 
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>
51
52#include "base/stl_helpers.hh"
53#include "base/str.hh"
54#include "mem/protocol/MachineType.hh"
55#include "mem/protocol/Protocol.hh"
56#include "mem/protocol/RubyRequest.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
62using namespace std;
63using m5::stl_helpers::operator<<;
64
65static double process_memory_total();
66static double process_memory_resident();
67
68Profiler::Profiler(const Params *p)
69    : SimObject(p)
70{
71    m_inst_profiler_ptr = NULL;
72    m_address_profiler_ptr = NULL;
73
74    m_real_time_start_time = time(NULL); // Not reset in clearStats()
75    m_stats_period = 1000000; // Default
76    m_periodic_output_file_ptr = &cerr;
77
78    m_hot_lines = p->hot_lines;
79    m_all_instructions = p->all_instructions;
80
81    m_num_of_sequencers = p->num_of_sequencers;
82
83    m_hot_lines = false;
84    m_all_instructions = false;
85
86    m_address_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
87    m_address_profiler_ptr->setHotLines(m_hot_lines);
88    m_address_profiler_ptr->setAllInstructions(m_all_instructions);
89
90    if (m_all_instructions) {
91        m_inst_profiler_ptr = new AddressProfiler(m_num_of_sequencers);
92        m_inst_profiler_ptr->setHotLines(m_hot_lines);
93        m_inst_profiler_ptr->setAllInstructions(m_all_instructions);
94    }
95
96    p->ruby_system->registerProfiler(this);
97}
98
99Profiler::~Profiler()
100{
101    if (m_periodic_output_file_ptr != &cerr) {
102        delete m_periodic_output_file_ptr;
103    }
104}
105
106void
107Profiler::wakeup()
108{
109    // FIXME - avoid the repeated code
110
111    vector<integer_t> perProcCycleCount(m_num_of_sequencers);
112
113    for (int i = 0; i < m_num_of_sequencers; i++) {
114        perProcCycleCount[i] =
115            g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
116        // The +1 allows us to avoid division by zero
117    }
118
119    ostream &out = *m_periodic_output_file_ptr;
120
121    out << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl
122        << "mbytes_resident: " << process_memory_resident() << endl
123        << "mbytes_total: " << process_memory_total() << endl;
124
125    if (process_memory_total() > 0) {
126        out << "resident_ratio: "
127            << process_memory_resident() / process_memory_total() << endl;
128    }
129
130    out << "miss_latency: " << m_allMissLatencyHistogram << endl;
131
132    out << endl;
133
134    if (m_all_instructions) {
135        m_inst_profiler_ptr->printStats(out);
136    }
137
138    //g_system_ptr->getNetwork()->printStats(out);
139    g_eventQueue_ptr->scheduleEvent(this, m_stats_period);
140}
141
142void
143Profiler::setPeriodicStatsFile(const string& filename)
144{
145    cout << "Recording periodic statistics to file '" << filename << "' every "
146         << m_stats_period << " Ruby cycles" << endl;
147
148    if (m_periodic_output_file_ptr != &cerr) {
149        delete m_periodic_output_file_ptr;
150    }
151
152    m_periodic_output_file_ptr = new ofstream(filename.c_str());
153    g_eventQueue_ptr->scheduleEvent(this, 1);
154}
155
156void
157Profiler::setPeriodicStatsInterval(integer_t period)
158{
159    cout << "Recording periodic statistics every " << m_stats_period
160         << " Ruby cycles" << endl;
161
162    m_stats_period = period;
163    g_eventQueue_ptr->scheduleEvent(this, 1);
164}
165
166void
167Profiler::printConfig(ostream& out) const
168{
169    out << endl;
170    out << "Profiler Configuration" << endl;
171    out << "----------------------" << endl;
172    out << "periodic_stats_period: " << m_stats_period << endl;
173}
174
175void
176Profiler::print(ostream& out) const
177{
178    out << "[Profiler]";
179}
180
181void
182Profiler::printStats(ostream& out, bool short_stats)
183{
184    out << endl;
185    if (short_stats) {
186        out << "SHORT ";
187    }
188    out << "Profiler Stats" << endl;
189    out << "--------------" << endl;
190
191    time_t real_time_current = time(NULL);
192    double seconds = difftime(real_time_current, m_real_time_start_time);
193    double minutes = seconds / 60.0;
194    double hours = minutes / 60.0;
195    double days = hours / 24.0;
196    Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
197
198    if (!short_stats) {
199        out << "Elapsed_time_in_seconds: " << seconds << endl;
200        out << "Elapsed_time_in_minutes: " << minutes << endl;
201        out << "Elapsed_time_in_hours: " << hours << endl;
202        out << "Elapsed_time_in_days: " << days << endl;
203        out << endl;
204    }
205
206    // print the virtual runtimes as well
207    struct tms vtime;
208    times(&vtime);
209    seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
210    minutes = seconds / 60.0;
211    hours = minutes / 60.0;
212    days = hours / 24.0;
213    out << "Virtual_time_in_seconds: " << seconds << endl;
214    out << "Virtual_time_in_minutes: " << minutes << endl;
215    out << "Virtual_time_in_hours:   " << hours << endl;
216    out << "Virtual_time_in_days:    " << days << endl;
217    out << endl;
218
219    out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
220    out << "Ruby_start_time: " << m_ruby_start << endl;
221    out << "Ruby_cycles: " << ruby_cycles << endl;
222    out << endl;
223
224    if (!short_stats) {
225        out << "mbytes_resident: " << process_memory_resident() << endl;
226        out << "mbytes_total: " << process_memory_total() << endl;
227        if (process_memory_total() > 0) {
228            out << "resident_ratio: "
229                << process_memory_resident()/process_memory_total() << endl;
230        }
231        out << endl;
232    }
233
234    vector<integer_t> perProcCycleCount(m_num_of_sequencers);
235
236    for (int i = 0; i < m_num_of_sequencers; i++) {
237        perProcCycleCount[i] =
238            g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
239        // The +1 allows us to avoid division by zero
240    }
241
242    out << "ruby_cycles_executed: " << perProcCycleCount << endl;
243
244    out << endl;
245
246    if (!short_stats) {
247        out << "Busy Controller Counts:" << endl;
248        for (int i = 0; i < MachineType_NUM; i++) {
249            int size = MachineType_base_count((MachineType)i);
250            for (int j = 0; j < size; j++) {
251                MachineID machID;
252                machID.type = (MachineType)i;
253                machID.num = j;
254                out << machID << ":" << m_busyControllerCount[i][j] << "  ";
255                if ((j + 1) % 8 == 0) {
256                    out << endl;
257                }
258            }
259            out << endl;
260        }
261        out << endl;
262
263        out << "Busy Bank Count:" << m_busyBankCount << endl;
264        out << endl;
265
266        out << "sequencer_requests_outstanding: "
267            << m_sequencer_requests << endl;
268        out << endl;
269    }
270
271    if (!short_stats) {
272        out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
273        out << "----------------------------------------" << endl;
274        out << "miss_latency: " << m_allMissLatencyHistogram << endl;
275        for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
276            if (m_missLatencyHistograms[i].size() > 0) {
277                out << "miss_latency_" << RubyRequestType(i) << ": "
278                    << m_missLatencyHistograms[i] << endl;
279            }
280        }
281        for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
282            if (m_machLatencyHistograms[i].size() > 0) {
283                out << "miss_latency_" << GenericMachineType(i) << ": "
284                    << m_machLatencyHistograms[i] << endl;
285            }
286        }
287
288        out << "miss_latency_wCC_issue_to_initial_request: "
289            << m_wCCIssueToInitialRequestHistogram << endl;
290        out << "miss_latency_wCC_initial_forward_request: "
291            << m_wCCInitialRequestToForwardRequestHistogram << endl;
292        out << "miss_latency_wCC_forward_to_first_response: "
293            << m_wCCForwardRequestToFirstResponseHistogram << endl;
294        out << "miss_latency_wCC_first_response_to_completion: "
295            << m_wCCFirstResponseToCompleteHistogram << endl;
296        out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
297        out << "miss_latency_dir_issue_to_initial_request: "
298            << m_dirIssueToInitialRequestHistogram << endl;
299        out << "miss_latency_dir_initial_forward_request: "
300            << m_dirInitialRequestToForwardRequestHistogram << endl;
301        out << "miss_latency_dir_forward_to_first_response: "
302            << m_dirForwardRequestToFirstResponseHistogram << endl;
303        out << "miss_latency_dir_first_response_to_completion: "
304            << m_dirFirstResponseToCompleteHistogram << endl;
305        out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
306
307        for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
308            for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
309                if (m_missMachLatencyHistograms[i][j].size() > 0) {
310                    out << "miss_latency_" << RubyRequestType(i)
311                        << "_" << GenericMachineType(j) << ": "
312                        << m_missMachLatencyHistograms[i][j] << endl;
313                }
314            }
315        }
316
317        out << endl;
318
319        out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
320        out << "------------------------------------" << endl;
321        out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
322        for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
323            if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
324                out << "prefetch_latency_" << RubyRequestType(i) << ": "
325                    << m_SWPrefetchLatencyHistograms[i] << endl;
326            }
327        }
328        for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
329            if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
330                out << "prefetch_latency_" << GenericMachineType(i) << ": "
331                    << m_SWPrefetchMachLatencyHistograms[i] << endl;
332            }
333        }
334        out << "prefetch_latency_L2Miss:"
335            << m_SWPrefetchL2MissLatencyHistogram << endl;
336
337        if (m_all_sharing_histogram.size() > 0) {
338            out << "all_sharing: " << m_all_sharing_histogram << endl;
339            out << "read_sharing: " << m_read_sharing_histogram << endl;
340            out << "write_sharing: " << m_write_sharing_histogram << endl;
341
342            out << "all_sharing_percent: ";
343            m_all_sharing_histogram.printPercent(out);
344            out << endl;
345
346            out << "read_sharing_percent: ";
347            m_read_sharing_histogram.printPercent(out);
348            out << endl;
349
350            out << "write_sharing_percent: ";
351            m_write_sharing_histogram.printPercent(out);
352            out << endl;
353
354            int64 total_miss = m_cache_to_cache +  m_memory_to_cache;
355            out << "all_misses: " << total_miss << endl;
356            out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
357            out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
358            out << "cache_to_cache_percent: "
359                << 100.0 * (double(m_cache_to_cache) / double(total_miss))
360                << endl;
361            out << "memory_to_cache_percent: "
362                << 100.0 * (double(m_memory_to_cache) / double(total_miss))
363                << endl;
364            out << endl;
365        }
366
367        if (m_outstanding_requests.size() > 0) {
368            out << "outstanding_requests: ";
369            m_outstanding_requests.printPercent(out);
370            out << endl;
371            out << endl;
372        }
373    }
374
375    if (!short_stats) {
376        out << "Request vs. RubySystem State Profile" << endl;
377        out << "--------------------------------" << endl;
378        out << endl;
379
380        map<string, int>::const_iterator i = m_requestProfileMap.begin();
381        map<string, int>::const_iterator end = m_requestProfileMap.end();
382        for (; i != end; ++i) {
383            const string &key = i->first;
384            int count = i->second;
385
386            double percent = (100.0 * double(count)) / double(m_requests);
387            vector<string> items;
388            tokenize(items, key, ':');
389            vector<string>::iterator j = items.begin();
390            vector<string>::iterator end = items.end();
391            for (; j != end; ++i)
392                out << setw(10) << *j;
393            out << setw(11) << count;
394            out << setw(14) << percent << endl;
395        }
396        out << endl;
397
398        out << "filter_action: " << m_filter_action_histogram << endl;
399
400        if (!m_all_instructions) {
401            m_address_profiler_ptr->printStats(out);
402        }
403
404        if (m_all_instructions) {
405            m_inst_profiler_ptr->printStats(out);
406        }
407
408        out << endl;
409        out << "Message Delayed Cycles" << endl;
410        out << "----------------------" << endl;
411        out << "Total_delay_cycles: " <<   m_delayedCyclesHistogram << endl;
412        out << "Total_nonPF_delay_cycles: "
413            << m_delayedCyclesNonPFHistogram << endl;
414        for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
415            out << "  virtual_network_" << i << "_delay_cycles: "
416                << m_delayedCyclesVCHistograms[i] << endl;
417        }
418
419        printResourceUsage(out);
420    }
421}
422
423void
424Profiler::printResourceUsage(ostream& out) const
425{
426    out << endl;
427    out << "Resource Usage" << endl;
428    out << "--------------" << endl;
429
430    integer_t pagesize = getpagesize(); // page size in bytes
431    out << "page_size: " << pagesize << endl;
432
433    rusage usage;
434    getrusage (RUSAGE_SELF, &usage);
435
436    out << "user_time: " << usage.ru_utime.tv_sec << endl;
437    out << "system_time: " << usage.ru_stime.tv_sec << endl;
438    out << "page_reclaims: " << usage.ru_minflt << endl;
439    out << "page_faults: " << usage.ru_majflt << endl;
440    out << "swaps: " << usage.ru_nswap << endl;
441    out << "block_inputs: " << usage.ru_inblock << endl;
442    out << "block_outputs: " << usage.ru_oublock << endl;
443}
444
445void
446Profiler::clearStats()
447{
448    m_ruby_start = g_eventQueue_ptr->getTime();
449
450    m_cycles_executed_at_start.resize(m_num_of_sequencers);
451    for (int i = 0; i < m_num_of_sequencers; i++) {
452        if (g_system_ptr == NULL) {
453            m_cycles_executed_at_start[i] = 0;
454        } else {
455            m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
456        }
457    }
458
459    m_busyControllerCount.resize(MachineType_NUM); // all machines
460    for (int i = 0; i < MachineType_NUM; i++) {
461        int size = MachineType_base_count((MachineType)i);
462        m_busyControllerCount[i].resize(size);
463        for (int j = 0; j < size; j++) {
464            m_busyControllerCount[i][j] = 0;
465        }
466    }
467    m_busyBankCount = 0;
468
469    m_delayedCyclesHistogram.clear();
470    m_delayedCyclesNonPFHistogram.clear();
471    int size = RubySystem::getNetwork()->getNumberOfVirtualNetworks();
472    m_delayedCyclesVCHistograms.resize(size);
473    for (int i = 0; i < size; i++) {
474        m_delayedCyclesVCHistograms[i].clear();
475    }
476
477    m_missLatencyHistograms.resize(RubyRequestType_NUM);
478    for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
479        m_missLatencyHistograms[i].clear(200);
480    }
481    m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
482    for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
483        m_machLatencyHistograms[i].clear(200);
484    }
485    m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
486    for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
487        m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
488        for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
489            m_missMachLatencyHistograms[i][j].clear(200);
490        }
491    }
492    m_allMissLatencyHistogram.clear(200);
493    m_wCCIssueToInitialRequestHistogram.clear(200);
494    m_wCCInitialRequestToForwardRequestHistogram.clear(200);
495    m_wCCForwardRequestToFirstResponseHistogram.clear(200);
496    m_wCCFirstResponseToCompleteHistogram.clear(200);
497    m_wCCIncompleteTimes = 0;
498    m_dirIssueToInitialRequestHistogram.clear(200);
499    m_dirInitialRequestToForwardRequestHistogram.clear(200);
500    m_dirForwardRequestToFirstResponseHistogram.clear(200);
501    m_dirFirstResponseToCompleteHistogram.clear(200);
502    m_dirIncompleteTimes = 0;
503
504    m_SWPrefetchLatencyHistograms.resize(RubyRequestType_NUM);
505    for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
506        m_SWPrefetchLatencyHistograms[i].clear(200);
507    }
508    m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
509    for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
510        m_SWPrefetchMachLatencyHistograms[i].clear(200);
511    }
512    m_allSWPrefetchLatencyHistogram.clear(200);
513
514    m_sequencer_requests.clear();
515    m_read_sharing_histogram.clear();
516    m_write_sharing_histogram.clear();
517    m_all_sharing_histogram.clear();
518    m_cache_to_cache = 0;
519    m_memory_to_cache = 0;
520
521    // clear HashMaps
522    m_requestProfileMap.clear();
523
524    // count requests profiled
525    m_requests = 0;
526
527    m_outstanding_requests.clear();
528    m_outstanding_persistent_requests.clear();
529
530    // Flush the prefetches through the system - used so that there
531    // are no outstanding requests after stats are cleared
532    //g_eventQueue_ptr->triggerAllEvents();
533
534    // update the start time
535    m_ruby_start = g_eventQueue_ptr->getTime();
536}
537
538void
539Profiler::addAddressTraceSample(const RubyRequest& msg, NodeID id)
540{
541    if (msg.getType() != RubyRequestType_IFETCH) {
542        // Note: The following line should be commented out if you
543        // want to use the special profiling that is part of the GS320
544        // protocol
545
546        // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
547        // profiled by the AddressProfiler
548        m_address_profiler_ptr->
549            addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
550                           msg.getType(), msg.getAccessMode(), id, false);
551    }
552}
553
554void
555Profiler::profileSharing(const Address& addr, AccessType type,
556                         NodeID requestor, const Set& sharers,
557                         const Set& owner)
558{
559    Set set_contacted(owner);
560    if (type == AccessType_Write) {
561        set_contacted.addSet(sharers);
562    }
563    set_contacted.remove(requestor);
564    int number_contacted = set_contacted.count();
565
566    if (type == AccessType_Write) {
567        m_write_sharing_histogram.add(number_contacted);
568    } else {
569        m_read_sharing_histogram.add(number_contacted);
570    }
571    m_all_sharing_histogram.add(number_contacted);
572
573    if (number_contacted == 0) {
574        m_memory_to_cache++;
575    } else {
576        m_cache_to_cache++;
577    }
578}
579
580void
581Profiler::profileMsgDelay(int virtualNetwork, int delayCycles)
582{
583    assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
584    m_delayedCyclesHistogram.add(delayCycles);
585    m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
586    if (virtualNetwork != 0) {
587        m_delayedCyclesNonPFHistogram.add(delayCycles);
588    }
589}
590
591// profiles original cache requests including PUTs
592void
593Profiler::profileRequest(const string& requestStr)
594{
595    m_requests++;
596
597    // if it doesn't exist, conveniently, it will be created with the
598    // default value which is 0
599    m_requestProfileMap[requestStr]++;
600}
601
602void
603Profiler::controllerBusy(MachineID machID)
604{
605    m_busyControllerCount[(int)machID.type][(int)machID.num]++;
606}
607
608void
609Profiler::profilePFWait(Time waitTime)
610{
611    m_prefetchWaitHistogram.add(waitTime);
612}
613
614void
615Profiler::bankBusy()
616{
617    m_busyBankCount++;
618}
619
620// non-zero cycle demand request
621void
622Profiler::missLatency(Time cycles,
623                      RubyRequestType type,
624                      const GenericMachineType respondingMach)
625{
626    m_allMissLatencyHistogram.add(cycles);
627    m_missLatencyHistograms[type].add(cycles);
628    m_machLatencyHistograms[respondingMach].add(cycles);
629    m_missMachLatencyHistograms[type][respondingMach].add(cycles);
630}
631
632void
633Profiler::missLatencyWcc(Time issuedTime,
634                         Time initialRequestTime,
635                         Time forwardRequestTime,
636                         Time firstResponseTime,
637                         Time completionTime)
638{
639    if ((issuedTime <= initialRequestTime) &&
640        (initialRequestTime <= forwardRequestTime) &&
641        (forwardRequestTime <= firstResponseTime) &&
642        (firstResponseTime <= completionTime)) {
643        m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
644
645        m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
646                                                         initialRequestTime);
647
648        m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
649                                                        forwardRequestTime);
650
651        m_wCCFirstResponseToCompleteHistogram.add(completionTime -
652                                                  firstResponseTime);
653    } else {
654        m_wCCIncompleteTimes++;
655    }
656}
657
658void
659Profiler::missLatencyDir(Time issuedTime,
660                         Time initialRequestTime,
661                         Time forwardRequestTime,
662                         Time firstResponseTime,
663                         Time completionTime)
664{
665    if ((issuedTime <= initialRequestTime) &&
666        (initialRequestTime <= forwardRequestTime) &&
667        (forwardRequestTime <= firstResponseTime) &&
668        (firstResponseTime <= completionTime)) {
669        m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
670
671        m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
672                                                         initialRequestTime);
673
674        m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
675                                                        forwardRequestTime);
676
677        m_dirFirstResponseToCompleteHistogram.add(completionTime -
678                                                  firstResponseTime);
679    } else {
680        m_dirIncompleteTimes++;
681    }
682}
683
684// non-zero cycle prefetch request
685void
686Profiler::swPrefetchLatency(Time cycles,
687                            RubyRequestType type,
688                            const GenericMachineType respondingMach)
689{
690    m_allSWPrefetchLatencyHistogram.add(cycles);
691    m_SWPrefetchLatencyHistograms[type].add(cycles);
692    m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
693    if (respondingMach == GenericMachineType_Directory ||
694        respondingMach == GenericMachineType_NUM) {
695        m_SWPrefetchL2MissLatencyHistogram.add(cycles);
696    }
697}
698
699// Helper function
700static double
701process_memory_total()
702{
703    // 4kB page size, 1024*1024 bytes per MB,
704    const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
705    ifstream proc_file;
706    proc_file.open("/proc/self/statm");
707    int total_size_in_pages = 0;
708    int res_size_in_pages = 0;
709    proc_file >> total_size_in_pages;
710    proc_file >> res_size_in_pages;
711    return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
712}
713
714static double
715process_memory_resident()
716{
717    // 4kB page size, 1024*1024 bytes per MB,
718    const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
719    ifstream proc_file;
720    proc_file.open("/proc/self/statm");
721    int total_size_in_pages = 0;
722    int res_size_in_pages = 0;
723    proc_file >> total_size_in_pages;
724    proc_file >> res_size_in_pages;
725    return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
726}
727
728void
729Profiler::rubyWatch(int id)
730{
731    uint64 tr = 0;
732    Address watch_address = Address(tr);
733
734    DPRINTFN("%7s %3s RUBY WATCH %d\n", g_eventQueue_ptr->getTime(), id,
735        watch_address);
736
737    // don't care about success or failure
738    m_watch_address_set.insert(watch_address);
739}
740
741bool
742Profiler::watchAddress(Address addr)
743{
744    return m_watch_address_set.count(addr) > 0;
745}
746
747Profiler *
748RubyProfilerParams::create()
749{
750    return new Profiler(this);
751}
752