src/mem/ruby/profiler/Profiler.cc

   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
  63 using namespace std;
  64 using m5::stl_helpers::operator<<;
  65
  66 static double process_memory_total();
  67 static double process_memory_resident();
  68
  69 Profiler::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
 100 Profiler::~Profiler()
 101 {
 102     if (m_periodic_output_file_ptr != &cerr) {
 103         delete m_periodic_output_file_ptr;
 104     }
 105 }
 106
 107 void
 108 Profiler::wakeup()
 109 {
 110     // FIXME - avoid the repeated code
 111
 112     vector<int64_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, g_system_ptr->clockEdge(Cycles(m_stats_period)));
 141 }
 142
 143 void
 144 Profiler::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, g_system_ptr->clockEdge(Cycles(1)));
 155 }
 156
 157 void
 158 Profiler::setPeriodicStatsInterval(int64_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, g_system_ptr->clockEdge(Cycles(1)));
 165 }
 166
 167 void
 168 Profiler::print(ostream& out) const
 169 {
 170     out << "[Profiler]";
 171 }
 172
 173 void
 174 Profiler::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<int64_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
 415 void
 416 Profiler::printResourceUsage(ostream& out) const
 417 {
 418     out << endl;
 419     out << "Resource Usage" << endl;
 420     out << "--------------" << endl;
 421
 422     int64_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
 437 void
 438 Profiler::clearStats()
 439 {
 440     m_ruby_start = g_system_ptr->getTime();
 441     m_real_time_start_time = time(NULL);
 442
 443     m_cycles_executed_at_start.resize(m_num_of_sequencers);
 444     for (int i = 0; i < m_num_of_sequencers; i++) {
 445         if (g_system_ptr == NULL) {
 446             m_cycles_executed_at_start[i] = 0;
 447         } else {
 448             m_cycles_executed_at_start[i] = g_system_ptr->getTime();
 449         }
 450     }
 451
 452     m_busyControllerCount.resize(MachineType_NUM); // all machines
 453     for (int i = 0; i < MachineType_NUM; i++) {
 454         int size = MachineType_base_count((MachineType)i);
 455         m_busyControllerCount[i].resize(size);
 456         for (int j = 0; j < size; j++) {
 457             m_busyControllerCount[i][j] = 0;
 458         }
 459     }
 460     m_busyBankCount = 0;
 461
 462     m_delayedCyclesHistogram.clear();
 463     m_delayedCyclesNonPFHistogram.clear();
 464     int size = Network::getNumberOfVirtualNetworks();
 465     m_delayedCyclesVCHistograms.resize(size);
 466     for (int i = 0; i < size; i++) {
 467         m_delayedCyclesVCHistograms[i].clear();
 468     }
 469
 470     m_missLatencyHistograms.resize(RubyRequestType_NUM);
 471     for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
 472         m_missLatencyHistograms[i].clear(200);
 473     }
 474     m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
 475     for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
 476         m_machLatencyHistograms[i].clear(200);
 477     }
 478     m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
 479     for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
 480         m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
 481         for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
 482             m_missMachLatencyHistograms[i][j].clear(200);
 483         }
 484     }
 485     m_allMissLatencyHistogram.clear(200);
 486     m_wCCIssueToInitialRequestHistogram.clear(200);
 487     m_wCCInitialRequestToForwardRequestHistogram.clear(200);
 488     m_wCCForwardRequestToFirstResponseHistogram.clear(200);
 489     m_wCCFirstResponseToCompleteHistogram.clear(200);
 490     m_wCCIncompleteTimes = 0;
 491     m_dirIssueToInitialRequestHistogram.clear(200);
 492     m_dirInitialRequestToForwardRequestHistogram.clear(200);
 493     m_dirForwardRequestToFirstResponseHistogram.clear(200);
 494     m_dirFirstResponseToCompleteHistogram.clear(200);
 495     m_dirIncompleteTimes = 0;
 496
 497     m_SWPrefetchLatencyHistograms.resize(RubyRequestType_NUM);
 498     for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
 499         m_SWPrefetchLatencyHistograms[i].clear(200);
 500     }
 501     m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
 502     for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
 503         m_SWPrefetchMachLatencyHistograms[i].clear(200);
 504     }
 505     m_allSWPrefetchLatencyHistogram.clear(200);
 506
 507     m_sequencer_requests.clear();
 508     m_read_sharing_histogram.clear();
 509     m_write_sharing_histogram.clear();
 510     m_all_sharing_histogram.clear();
 511     m_cache_to_cache = 0;
 512     m_memory_to_cache = 0;
 513
 514     // clear HashMaps
 515     m_requestProfileMap.clear();
 516
 517     // count requests profiled
 518     m_requests = 0;
 519
 520     m_outstanding_requests.clear();
 521     m_outstanding_persistent_requests.clear();
 522
 523     // Flush the prefetches through the system - used so that there
 524     // are no outstanding requests after stats are cleared
 525     //g_eventQueue_ptr->triggerAllEvents();
 526
 527     // update the start time
 528     m_ruby_start = g_system_ptr->getTime();
 529 }
 530
 531 void
 532 Profiler::addAddressTraceSample(const RubyRequest& msg, NodeID id)
 533 {
 534     if (msg.getType() != RubyRequestType_IFETCH) {
 535         // Note: The following line should be commented out if you
 536         // want to use the special profiling that is part of the GS320
 537         // protocol
 538
 539         // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
 540         // profiled by the AddressProfiler
 541         m_address_profiler_ptr->
 542             addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
 543                            msg.getType(), msg.getAccessMode(), id, false);
 544     }
 545 }
 546
 547 void
 548 Profiler::profileSharing(const Address& addr, AccessType type,
 549                          NodeID requestor, const Set& sharers,
 550                          const Set& owner)
 551 {
 552     Set set_contacted(owner);
 553     if (type == AccessType_Write) {
 554         set_contacted.addSet(sharers);
 555     }
 556     set_contacted.remove(requestor);
 557     int number_contacted = set_contacted.count();
 558
 559     if (type == AccessType_Write) {
 560         m_write_sharing_histogram.add(number_contacted);
 561     } else {
 562         m_read_sharing_histogram.add(number_contacted);
 563     }
 564     m_all_sharing_histogram.add(number_contacted);
 565
 566     if (number_contacted == 0) {
 567         m_memory_to_cache++;
 568     } else {
 569         m_cache_to_cache++;
 570     }
 571 }
 572
 573 void
 574 Profiler::profileMsgDelay(uint32_t virtualNetwork, Time delayCycles)
 575 {
 576     assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
 577     m_delayedCyclesHistogram.add(delayCycles);
 578     m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
 579     if (virtualNetwork != 0) {
 580         m_delayedCyclesNonPFHistogram.add(delayCycles);
 581     }
 582 }
 583
 584 // profiles original cache requests including PUTs
 585 void
 586 Profiler::profileRequest(const string& requestStr)
 587 {
 588     m_requests++;
 589
 590     // if it doesn't exist, conveniently, it will be created with the
 591     // default value which is 0
 592     m_requestProfileMap[requestStr]++;
 593 }
 594
 595 void
 596 Profiler::controllerBusy(MachineID machID)
 597 {
 598     m_busyControllerCount[(int)machID.type][(int)machID.num]++;
 599 }
 600
 601 void
 602 Profiler::profilePFWait(Time waitTime)
 603 {
 604     m_prefetchWaitHistogram.add(waitTime);
 605 }
 606
 607 void
 608 Profiler::bankBusy()
 609 {
 610     m_busyBankCount++;
 611 }
 612
 613 // non-zero cycle demand request
 614 void
 615 Profiler::missLatency(Time cycles,
 616                       RubyRequestType type,
 617                       const GenericMachineType respondingMach)
 618 {
 619     m_allMissLatencyHistogram.add(cycles);
 620     m_missLatencyHistograms[type].add(cycles);
 621     m_machLatencyHistograms[respondingMach].add(cycles);
 622     m_missMachLatencyHistograms[type][respondingMach].add(cycles);
 623 }
 624
 625 void
 626 Profiler::missLatencyWcc(Time issuedTime,
 627                          Time initialRequestTime,
 628                          Time forwardRequestTime,
 629                          Time firstResponseTime,
 630                          Time completionTime)
 631 {
 632     if ((issuedTime <= initialRequestTime) &&
 633         (initialRequestTime <= forwardRequestTime) &&
 634         (forwardRequestTime <= firstResponseTime) &&
 635         (firstResponseTime <= completionTime)) {
 636         m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
 637
 638         m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
 639                                                          initialRequestTime);
 640
 641         m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
 642                                                         forwardRequestTime);
 643
 644         m_wCCFirstResponseToCompleteHistogram.add(completionTime -
 645                                                   firstResponseTime);
 646     } else {
 647         m_wCCIncompleteTimes++;
 648     }
 649 }
 650
 651 void
 652 Profiler::missLatencyDir(Time issuedTime,
 653                          Time initialRequestTime,
 654                          Time forwardRequestTime,
 655                          Time firstResponseTime,
 656                          Time completionTime)
 657 {
 658     if ((issuedTime <= initialRequestTime) &&
 659         (initialRequestTime <= forwardRequestTime) &&
 660         (forwardRequestTime <= firstResponseTime) &&
 661         (firstResponseTime <= completionTime)) {
 662         m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
 663
 664         m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
 665                                                          initialRequestTime);
 666
 667         m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
 668                                                         forwardRequestTime);
 669
 670         m_dirFirstResponseToCompleteHistogram.add(completionTime -
 671                                                   firstResponseTime);
 672     } else {
 673         m_dirIncompleteTimes++;
 674     }
 675 }
 676
 677 // non-zero cycle prefetch request
 678 void
 679 Profiler::swPrefetchLatency(Time cycles,
 680                             RubyRequestType type,
 681                             const GenericMachineType respondingMach)
 682 {
 683     m_allSWPrefetchLatencyHistogram.add(cycles);
 684     m_SWPrefetchLatencyHistograms[type].add(cycles);
 685     m_SWPrefetchMachLatencyHistograms[respondingMach].add(cycles);
 686     if (respondingMach == GenericMachineType_Directory ||
 687         respondingMach == GenericMachineType_NUM) {
 688         m_SWPrefetchL2MissLatencyHistogram.add(cycles);
 689     }
 690 }
 691
 692 // Helper function
 693 static double
 694 process_memory_total()
 695 {
 696     // 4kB page size, 1024*1024 bytes per MB,
 697     const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
 698     ifstream proc_file;
 699     proc_file.open("/proc/self/statm");
 700     int total_size_in_pages = 0;
 701     int res_size_in_pages = 0;
 702     proc_file >> total_size_in_pages;
 703     proc_file >> res_size_in_pages;
 704     return double(total_size_in_pages) * MULTIPLIER; // size in megabytes
 705 }
 706
 707 static double
 708 process_memory_resident()
 709 {
 710     // 4kB page size, 1024*1024 bytes per MB,
 711     const double MULTIPLIER = 4096.0 / (1024.0 * 1024.0);
 712     ifstream proc_file;
 713     proc_file.open("/proc/self/statm");
 714     int total_size_in_pages = 0;
 715     int res_size_in_pages = 0;
 716     proc_file >> total_size_in_pages;
 717     proc_file >> res_size_in_pages;
 718     return double(res_size_in_pages) * MULTIPLIER; // size in megabytes
 719 }
 720
 721 void
 722 Profiler::rubyWatch(int id)
 723 {
 724     uint64 tr = 0;
 725     Address watch_address = Address(tr);
 726
 727     DPRINTFN("%7s %3s RUBY WATCH %d\n", g_system_ptr->getTime(), id,
 728         watch_address);
 729
 730     // don't care about success or failure
 731     m_watch_address_set.insert(watch_address);
 732 }
 733
 734 bool
 735 Profiler::watchAddress(Address addr)
 736 {
 737     return m_watch_address_set.count(addr) > 0;
 738 }
 739
 740 Profiler *
 741 RubyProfilerParams::create()
 742 {
 743     return new Profiler(this);
 744 }