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::printRequestProfile(ostream &out)
 175 {
 176     out << "Request vs. RubySystem State Profile" << endl;
 177     out << "--------------------------------" << endl;
 178     out << endl;
 179
 180     map<string, uint64_t> m_requestProfileMap;
 181     uint64_t m_requests = 0;
 182
 183     for (uint32_t i = 0; i < MachineType_NUM; i++) {
 184         for (map<uint32_t, AbstractController*>::iterator it =
 185                   g_abs_controls[i].begin();
 186              it != g_abs_controls[i].end(); ++it) {
 187
 188             AbstractController *ctr = (*it).second;
 189             map<string, uint64_t> mp = ctr->getRequestProfileMap();
 190
 191             for (map<string, uint64_t>::iterator jt = mp.begin();
 192                  jt != mp.end(); ++jt) {
 193
 194                 map<string, uint64_t>::iterator kt =
 195                     m_requestProfileMap.find((*jt).first);
 196                 if (kt != m_requestProfileMap.end()) {
 197                     (*kt).second += (*jt).second;
 198                 } else {
 199                     m_requestProfileMap[(*jt).first] = (*jt).second;
 200                 }
 201             }
 202
 203             m_requests += ctr->getRequestCount();
 204         }
 205     }
 206
 207     map<string, uint64_t>::const_iterator i = m_requestProfileMap.begin();
 208     map<string, uint64_t>::const_iterator end = m_requestProfileMap.end();
 209     for (; i != end; ++i) {
 210         const string &key = i->first;
 211         uint64_t count = i->second;
 212
 213         double percent = (100.0 * double(count)) / double(m_requests);
 214         vector<string> items;
 215         tokenize(items, key, ':');
 216         vector<string>::iterator j = items.begin();
 217         vector<string>::iterator end = items.end();
 218         for (; j != end; ++i)
 219             out << setw(10) << *j;
 220         out << setw(11) << count;
 221         out << setw(14) << percent << endl;
 222     }
 223     out << endl;
 224 }
 225
 226 void
 227 Profiler::printStats(ostream& out, bool short_stats)
 228 {
 229     out << endl;
 230     if (short_stats) {
 231         out << "SHORT ";
 232     }
 233     out << "Profiler Stats" << endl;
 234     out << "--------------" << endl;
 235
 236     time_t real_time_current = time(NULL);
 237     double seconds = difftime(real_time_current, m_real_time_start_time);
 238     double minutes = seconds / 60.0;
 239     double hours = minutes / 60.0;
 240     double days = hours / 24.0;
 241     Time ruby_cycles = g_system_ptr->getTime()-m_ruby_start;
 242
 243     if (!short_stats) {
 244         out << "Elapsed_time_in_seconds: " << seconds << endl;
 245         out << "Elapsed_time_in_minutes: " << minutes << endl;
 246         out << "Elapsed_time_in_hours: " << hours << endl;
 247         out << "Elapsed_time_in_days: " << days << endl;
 248         out << endl;
 249     }
 250
 251     // print the virtual runtimes as well
 252     struct tms vtime;
 253     times(&vtime);
 254     seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
 255     minutes = seconds / 60.0;
 256     hours = minutes / 60.0;
 257     days = hours / 24.0;
 258     out << "Virtual_time_in_seconds: " << seconds << endl;
 259     out << "Virtual_time_in_minutes: " << minutes << endl;
 260     out << "Virtual_time_in_hours:   " << hours << endl;
 261     out << "Virtual_time_in_days:    " << days << endl;
 262     out << endl;
 263
 264     out << "Ruby_current_time: " << g_system_ptr->getTime() << endl;
 265     out << "Ruby_start_time: " << m_ruby_start << endl;
 266     out << "Ruby_cycles: " << ruby_cycles << endl;
 267     out << endl;
 268
 269     if (!short_stats) {
 270         out << "mbytes_resident: " << process_memory_resident() << endl;
 271         out << "mbytes_total: " << process_memory_total() << endl;
 272         if (process_memory_total() > 0) {
 273             out << "resident_ratio: "
 274                 << process_memory_resident()/process_memory_total() << endl;
 275         }
 276         out << endl;
 277     }
 278
 279     vector<int64_t> perProcCycleCount(m_num_of_sequencers);
 280
 281     for (int i = 0; i < m_num_of_sequencers; i++) {
 282         perProcCycleCount[i] =
 283             g_system_ptr->getTime() - m_cycles_executed_at_start[i] + 1;
 284         // The +1 allows us to avoid division by zero
 285     }
 286
 287     out << "ruby_cycles_executed: " << perProcCycleCount << endl;
 288
 289     out << endl;
 290
 291     if (!short_stats) {
 292         out << "Busy Controller Counts:" << endl;
 293         for (uint32_t i = 0; i < MachineType_NUM; i++) {
 294             uint32_t size = MachineType_base_count((MachineType)i);
 295
 296             for (uint32_t j = 0; j < size; j++) {
 297                 MachineID machID;
 298                 machID.type = (MachineType)i;
 299                 machID.num = j;
 300
 301                 AbstractController *ctr =
 302                     (*(g_abs_controls[i].find(j))).second;
 303                 out << machID << ":" << ctr->getFullyBusyCycles() << "  ";
 304                 if ((j + 1) % 8 == 0) {
 305                     out << endl;
 306                 }
 307             }
 308             out << endl;
 309         }
 310         out << endl;
 311
 312         out << "Busy Bank Count:" << m_busyBankCount << endl;
 313         out << endl;
 314
 315         out << "sequencer_requests_outstanding: "
 316             << m_sequencer_requests << endl;
 317         out << endl;
 318     }
 319
 320     if (!short_stats) {
 321         out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
 322         out << "----------------------------------------" << endl;
 323         out << "miss_latency: " << m_allMissLatencyHistogram << endl;
 324         for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
 325             if (m_missLatencyHistograms[i].size() > 0) {
 326                 out << "miss_latency_" << RubyRequestType(i) << ": "
 327                     << m_missLatencyHistograms[i] << endl;
 328             }
 329         }
 330         for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
 331             if (m_machLatencyHistograms[i].size() > 0) {
 332                 out << "miss_latency_" << GenericMachineType(i) << ": "
 333                     << m_machLatencyHistograms[i] << endl;
 334             }
 335         }
 336
 337         out << "miss_latency_wCC_issue_to_initial_request: "
 338             << m_wCCIssueToInitialRequestHistogram << endl;
 339         out << "miss_latency_wCC_initial_forward_request: "
 340             << m_wCCInitialRequestToForwardRequestHistogram << endl;
 341         out << "miss_latency_wCC_forward_to_first_response: "
 342             << m_wCCForwardRequestToFirstResponseHistogram << endl;
 343         out << "miss_latency_wCC_first_response_to_completion: "
 344             << m_wCCFirstResponseToCompleteHistogram << endl;
 345         out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
 346         out << "miss_latency_dir_issue_to_initial_request: "
 347             << m_dirIssueToInitialRequestHistogram << endl;
 348         out << "miss_latency_dir_initial_forward_request: "
 349             << m_dirInitialRequestToForwardRequestHistogram << endl;
 350         out << "miss_latency_dir_forward_to_first_response: "
 351             << m_dirForwardRequestToFirstResponseHistogram << endl;
 352         out << "miss_latency_dir_first_response_to_completion: "
 353             << m_dirFirstResponseToCompleteHistogram << endl;
 354         out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
 355
 356         for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
 357             for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
 358                 if (m_missMachLatencyHistograms[i][j].size() > 0) {
 359                     out << "miss_latency_" << RubyRequestType(i)
 360                         << "_" << GenericMachineType(j) << ": "
 361                         << m_missMachLatencyHistograms[i][j] << endl;
 362                 }
 363             }
 364         }
 365
 366         out << endl;
 367
 368         out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
 369         out << "------------------------------------" << endl;
 370         out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
 371         for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
 372             if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
 373                 out << "prefetch_latency_" << RubyRequestType(i) << ": "
 374                     << m_SWPrefetchLatencyHistograms[i] << endl;
 375             }
 376         }
 377         for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
 378             if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
 379                 out << "prefetch_latency_" << GenericMachineType(i) << ": "
 380                     << m_SWPrefetchMachLatencyHistograms[i] << endl;
 381             }
 382         }
 383         out << "prefetch_latency_L2Miss:"
 384             << m_SWPrefetchL2MissLatencyHistogram << endl;
 385
 386         if (m_all_sharing_histogram.size() > 0) {
 387             out << "all_sharing: " << m_all_sharing_histogram << endl;
 388             out << "read_sharing: " << m_read_sharing_histogram << endl;
 389             out << "write_sharing: " << m_write_sharing_histogram << endl;
 390
 391             out << "all_sharing_percent: ";
 392             m_all_sharing_histogram.printPercent(out);
 393             out << endl;
 394
 395             out << "read_sharing_percent: ";
 396             m_read_sharing_histogram.printPercent(out);
 397             out << endl;
 398
 399             out << "write_sharing_percent: ";
 400             m_write_sharing_histogram.printPercent(out);
 401             out << endl;
 402
 403             int64 total_miss = m_cache_to_cache +  m_memory_to_cache;
 404             out << "all_misses: " << total_miss << endl;
 405             out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
 406             out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
 407             out << "cache_to_cache_percent: "
 408                 << 100.0 * (double(m_cache_to_cache) / double(total_miss))
 409                 << endl;
 410             out << "memory_to_cache_percent: "
 411                 << 100.0 * (double(m_memory_to_cache) / double(total_miss))
 412                 << endl;
 413             out << endl;
 414         }
 415
 416         if (m_outstanding_requests.size() > 0) {
 417             out << "outstanding_requests: ";
 418             m_outstanding_requests.printPercent(out);
 419             out << endl;
 420             out << endl;
 421         }
 422     }
 423
 424     if (!short_stats) {
 425         printRequestProfile(out);
 426
 427         out << "filter_action: " << m_filter_action_histogram << endl;
 428
 429         if (!m_all_instructions) {
 430             m_address_profiler_ptr->printStats(out);
 431         }
 432
 433         if (m_all_instructions) {
 434             m_inst_profiler_ptr->printStats(out);
 435         }
 436
 437         out << endl;
 438         out << "Message Delayed Cycles" << endl;
 439         out << "----------------------" << endl;
 440         out << "Total_delay_cycles: " <<   m_delayedCyclesHistogram << endl;
 441         out << "Total_nonPF_delay_cycles: "
 442             << m_delayedCyclesNonPFHistogram << endl;
 443         for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
 444             out << "  virtual_network_" << i << "_delay_cycles: "
 445                 << m_delayedCyclesVCHistograms[i] << endl;
 446         }
 447
 448         printResourceUsage(out);
 449     }
 450 }
 451
 452 void
 453 Profiler::printResourceUsage(ostream& out) const
 454 {
 455     out << endl;
 456     out << "Resource Usage" << endl;
 457     out << "--------------" << endl;
 458
 459     int64_t pagesize = getpagesize(); // page size in bytes
 460     out << "page_size: " << pagesize << endl;
 461
 462     rusage usage;
 463     getrusage (RUSAGE_SELF, &usage);
 464
 465     out << "user_time: " << usage.ru_utime.tv_sec << endl;
 466     out << "system_time: " << usage.ru_stime.tv_sec << endl;
 467     out << "page_reclaims: " << usage.ru_minflt << endl;
 468     out << "page_faults: " << usage.ru_majflt << endl;
 469     out << "swaps: " << usage.ru_nswap << endl;
 470     out << "block_inputs: " << usage.ru_inblock << endl;
 471     out << "block_outputs: " << usage.ru_oublock << endl;
 472 }
 473
 474 void
 475 Profiler::clearStats()
 476 {
 477     m_ruby_start = g_system_ptr->getTime();
 478     m_real_time_start_time = time(NULL);
 479
 480     m_cycles_executed_at_start.resize(m_num_of_sequencers);
 481     for (int i = 0; i < m_num_of_sequencers; i++) {
 482         if (g_system_ptr == NULL) {
 483             m_cycles_executed_at_start[i] = 0;
 484         } else {
 485             m_cycles_executed_at_start[i] = g_system_ptr->getTime();
 486         }
 487     }
 488
 489     m_busyBankCount = 0;
 490
 491     m_delayedCyclesHistogram.clear();
 492     m_delayedCyclesNonPFHistogram.clear();
 493     int size = Network::getNumberOfVirtualNetworks();
 494     m_delayedCyclesVCHistograms.resize(size);
 495     for (int i = 0; i < size; i++) {
 496         m_delayedCyclesVCHistograms[i].clear();
 497     }
 498
 499     m_missLatencyHistograms.resize(RubyRequestType_NUM);
 500     for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
 501         m_missLatencyHistograms[i].clear(200);
 502     }
 503     m_machLatencyHistograms.resize(GenericMachineType_NUM+1);
 504     for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
 505         m_machLatencyHistograms[i].clear(200);
 506     }
 507     m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
 508     for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
 509         m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
 510         for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
 511             m_missMachLatencyHistograms[i][j].clear(200);
 512         }
 513     }
 514     m_allMissLatencyHistogram.clear(200);
 515     m_wCCIssueToInitialRequestHistogram.clear(200);
 516     m_wCCInitialRequestToForwardRequestHistogram.clear(200);
 517     m_wCCForwardRequestToFirstResponseHistogram.clear(200);
 518     m_wCCFirstResponseToCompleteHistogram.clear(200);
 519     m_wCCIncompleteTimes = 0;
 520     m_dirIssueToInitialRequestHistogram.clear(200);
 521     m_dirInitialRequestToForwardRequestHistogram.clear(200);
 522     m_dirForwardRequestToFirstResponseHistogram.clear(200);
 523     m_dirFirstResponseToCompleteHistogram.clear(200);
 524     m_dirIncompleteTimes = 0;
 525
 526     m_SWPrefetchLatencyHistograms.resize(RubyRequestType_NUM);
 527     for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
 528         m_SWPrefetchLatencyHistograms[i].clear(200);
 529     }
 530     m_SWPrefetchMachLatencyHistograms.resize(GenericMachineType_NUM+1);
 531     for (int i = 0; i < m_SWPrefetchMachLatencyHistograms.size(); i++) {
 532         m_SWPrefetchMachLatencyHistograms[i].clear(200);
 533     }
 534     m_allSWPrefetchLatencyHistogram.clear(200);
 535
 536     m_sequencer_requests.clear();
 537     m_read_sharing_histogram.clear();
 538     m_write_sharing_histogram.clear();
 539     m_all_sharing_histogram.clear();
 540     m_cache_to_cache = 0;
 541     m_memory_to_cache = 0;
 542
 543     m_outstanding_requests.clear();
 544     m_outstanding_persistent_requests.clear();
 545
 546     // Flush the prefetches through the system - used so that there
 547     // are no outstanding requests after stats are cleared
 548     //g_eventQueue_ptr->triggerAllEvents();
 549
 550     // update the start time
 551     m_ruby_start = g_system_ptr->getTime();
 552 }
 553
 554 void
 555 Profiler::addAddressTraceSample(const RubyRequest& msg, NodeID id)
 556 {
 557     if (msg.getType() != RubyRequestType_IFETCH) {
 558         // Note: The following line should be commented out if you
 559         // want to use the special profiling that is part of the GS320
 560         // protocol
 561
 562         // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be
 563         // profiled by the AddressProfiler
 564         m_address_profiler_ptr->
 565             addTraceSample(msg.getLineAddress(), msg.getProgramCounter(),
 566                            msg.getType(), msg.getAccessMode(), id, false);
 567     }
 568 }
 569
 570 void
 571 Profiler::profileSharing(const Address& addr, AccessType type,
 572                          NodeID requestor, const Set& sharers,
 573                          const Set& owner)
 574 {
 575     Set set_contacted(owner);
 576     if (type == AccessType_Write) {
 577         set_contacted.addSet(sharers);
 578     }
 579     set_contacted.remove(requestor);
 580     int number_contacted = set_contacted.count();
 581
 582     if (type == AccessType_Write) {
 583         m_write_sharing_histogram.add(number_contacted);
 584     } else {
 585         m_read_sharing_histogram.add(number_contacted);
 586     }
 587     m_all_sharing_histogram.add(number_contacted);
 588
 589     if (number_contacted == 0) {
 590         m_memory_to_cache++;
 591     } else {
 592         m_cache_to_cache++;
 593     }
 594 }
 595
 596 void
 597 Profiler::profileMsgDelay(uint32_t virtualNetwork, Time delayCycles)
 598 {
 599     assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
 600     m_delayedCyclesHistogram.add(delayCycles);
 601     m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
 602     if (virtualNetwork != 0) {
 603         m_delayedCyclesNonPFHistogram.add(delayCycles);
 604     }
 605 }
 606
 607 void
 608 Profiler::profilePFWait(Time waitTime)
 609 {
 610     m_prefetchWaitHistogram.add(waitTime);
 611 }
 612
 613 void
 614 Profiler::bankBusy()
 615 {
 616     m_busyBankCount++;
 617 }
 618
 619 // non-zero cycle demand request
 620 void
 621 Profiler::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
 631 void
 632 Profiler::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
 657 void
 658 Profiler::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
 684 void
 685 Profiler::swPrefetchLatency(Time cycles,
 686                             RubyRequestType 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
 698 // Helper function
 699 static double
 700 process_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
 713 static double
 714 process_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
 727 void
 728 Profiler::rubyWatch(int id)
 729 {
 730     uint64 tr = 0;
 731     Address watch_address = Address(tr);
 732
 733     DPRINTFN("%7s %3s RUBY WATCH %d\n", g_system_ptr->getTime(), id,
 734         watch_address);
 735
 736     // don't care about success or failure
 737     m_watch_address_set.insert(watch_address);
 738 }
 739
 740 bool
 741 Profiler::watchAddress(Address addr)
 742 {
 743     return m_watch_address_set.count(addr) > 0;
 744 }
 745
 746 Profiler *
 747 RubyProfilerParams::create()
 748 {
 749     return new Profiler(this);
 750 }