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/* 46 * Profiler.cc 47 * 48 * Description: See Profiler.hh 49 * 50 * $Id$ 51 * 52 */ 53 54#include "mem/ruby/profiler/Profiler.hh"
| 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/* 46 * Profiler.cc 47 * 48 * Description: See Profiler.hh 49 * 50 * $Id$ 51 * 52 */ 53 54#include "mem/ruby/profiler/Profiler.hh"
|
55#include "mem/ruby/profiler/CacheProfiler.hh"
| |
56#include "mem/ruby/profiler/AddressProfiler.hh" 57#include "mem/ruby/system/System.hh" 58#include "mem/ruby/network/Network.hh" 59#include "mem/gems_common/PrioHeap.hh" 60#include "mem/protocol/CacheMsg.hh" 61#include "mem/protocol/Protocol.hh" 62#include "mem/gems_common/util.hh" 63#include "mem/gems_common/Map.hh" 64#include "mem/ruby/common/Debug.hh" 65#include "mem/protocol/MachineType.hh" 66 67// Allows use of times() library call, which determines virtual runtime 68#include <sys/times.h> 69 70extern std::ostream * debug_cout_ptr; 71 72static double process_memory_total(); 73static double process_memory_resident(); 74 75Profiler::Profiler(const string & name)
| 55#include "mem/ruby/profiler/AddressProfiler.hh" 56#include "mem/ruby/system/System.hh" 57#include "mem/ruby/network/Network.hh" 58#include "mem/gems_common/PrioHeap.hh" 59#include "mem/protocol/CacheMsg.hh" 60#include "mem/protocol/Protocol.hh" 61#include "mem/gems_common/util.hh" 62#include "mem/gems_common/Map.hh" 63#include "mem/ruby/common/Debug.hh" 64#include "mem/protocol/MachineType.hh" 65 66// Allows use of times() library call, which determines virtual runtime 67#include <sys/times.h> 68 69extern std::ostream * debug_cout_ptr; 70 71static double process_memory_total(); 72static double process_memory_resident(); 73 74Profiler::Profiler(const string & name)
|
76 : m_conflicting_histogram(-1)
| |
77{ 78 m_name = name; 79 m_requestProfileMap_ptr = new Map<string, int>;
| 75{ 76 m_name = name; 77 m_requestProfileMap_ptr = new Map<string, int>;
|
80 m_L1D_cache_profiler_ptr = new CacheProfiler("L1D_cache"); 81 m_L1I_cache_profiler_ptr = new CacheProfiler("L1I_cache");
| |
82
| 78
|
83 m_L2_cache_profiler_ptr = new CacheProfiler("L2_cache"); 84
| |
85 m_inst_profiler_ptr = NULL; 86 m_address_profiler_ptr = NULL; 87
| 79 m_inst_profiler_ptr = NULL; 80 m_address_profiler_ptr = NULL; 81
|
88/* 89 m_address_profiler_ptr = new AddressProfiler; 90 m_inst_profiler_ptr = NULL; 91 if (m_all_instructions) { 92 m_inst_profiler_ptr = new AddressProfiler; 93 } 94*/ 95 m_conflicting_map_ptr = new Map<Address, Time>; 96
| |
97 m_real_time_start_time = time(NULL); // Not reset in clearStats() 98 m_stats_period = 1000000; // Default 99 m_periodic_output_file_ptr = &cerr; 100 101} 102 103Profiler::~Profiler() 104{ 105 if (m_periodic_output_file_ptr != &cerr) { 106 delete m_periodic_output_file_ptr; 107 }
| 82 m_real_time_start_time = time(NULL); // Not reset in clearStats() 83 m_stats_period = 1000000; // Default 84 m_periodic_output_file_ptr = &cerr; 85 86} 87 88Profiler::~Profiler() 89{ 90 if (m_periodic_output_file_ptr != &cerr) { 91 delete m_periodic_output_file_ptr; 92 }
|
108 delete m_address_profiler_ptr; 109 delete m_L1D_cache_profiler_ptr; 110 delete m_L1I_cache_profiler_ptr; 111 delete m_L2_cache_profiler_ptr;
| |
112 delete m_requestProfileMap_ptr;
| 93 delete m_requestProfileMap_ptr;
|
113 delete m_conflicting_map_ptr;
| |
114} 115 116void Profiler::init(const vector<string> & argv, vector<string> memory_control_names) 117{ 118 // added by SS 119 vector<string>::iterator it; 120 memory_control_profiler* mcp; 121 m_memory_control_names = memory_control_names; 122// printf ( "Here in Profiler::init \n"); 123 for ( it=memory_control_names.begin() ; it < memory_control_names.end(); it++ ){ 124// printf ( "Here in Profiler::init memory control name %s \n", (*it).c_str()); 125 mcp = new memory_control_profiler; 126 mcp->m_memReq = 0; 127 mcp->m_memBankBusy = 0; 128 mcp->m_memBusBusy = 0; 129 mcp->m_memReadWriteBusy = 0; 130 mcp->m_memDataBusBusy = 0; 131 mcp->m_memTfawBusy = 0; 132 mcp->m_memRefresh = 0; 133 mcp->m_memRead = 0; 134 mcp->m_memWrite = 0; 135 mcp->m_memWaitCycles = 0; 136 mcp->m_memInputQ = 0; 137 mcp->m_memBankQ = 0; 138 mcp->m_memArbWait = 0; 139 mcp->m_memRandBusy = 0; 140 mcp->m_memNotOld = 0; 141 142 mcp->m_banks_per_rank = RubySystem::getMemoryControl((*it).c_str())->getBanksPerRank(); 143 mcp->m_ranks_per_dimm = RubySystem::getMemoryControl((*it).c_str())->getRanksPerDimm(); 144 mcp->m_dimms_per_channel = RubySystem::getMemoryControl((*it).c_str())->getDimmsPerChannel(); 145 146 int totalBanks = mcp->m_banks_per_rank 147 * mcp->m_ranks_per_dimm 148 * mcp->m_dimms_per_channel; 149 150 mcp->m_memBankCount.setSize(totalBanks); 151 152 m_memory_control_profilers [(*it).c_str()] = mcp; 153 } 154 155 clearStats(); 156 m_hot_lines = false; 157 m_all_instructions = false; 158 159 for (size_t i=0; i<argv.size(); i+=2) { 160 if ( argv[i] == "hot_lines") { 161 m_hot_lines = (argv[i+1]=="true"); 162 } else if ( argv[i] == "all_instructions") { 163 m_all_instructions = (argv[i+1]=="true"); 164 }else { 165 cerr << "WARNING: Profiler: Unkown configuration parameter: " << argv[i] << endl; 166 assert(false); 167 } 168 } 169 170 m_address_profiler_ptr = new AddressProfiler; 171 m_address_profiler_ptr -> setHotLines(m_hot_lines); 172 m_address_profiler_ptr -> setAllInstructions(m_all_instructions); 173 174 if (m_all_instructions) { 175 m_inst_profiler_ptr = new AddressProfiler; 176 m_inst_profiler_ptr -> setHotLines(m_hot_lines); 177 m_inst_profiler_ptr -> setAllInstructions(m_all_instructions); 178 } 179} 180 181void Profiler::wakeup() 182{ 183 // FIXME - avoid the repeated code 184
| 94} 95 96void Profiler::init(const vector<string> & argv, vector<string> memory_control_names) 97{ 98 // added by SS 99 vector<string>::iterator it; 100 memory_control_profiler* mcp; 101 m_memory_control_names = memory_control_names; 102// printf ( "Here in Profiler::init \n"); 103 for ( it=memory_control_names.begin() ; it < memory_control_names.end(); it++ ){ 104// printf ( "Here in Profiler::init memory control name %s \n", (*it).c_str()); 105 mcp = new memory_control_profiler; 106 mcp->m_memReq = 0; 107 mcp->m_memBankBusy = 0; 108 mcp->m_memBusBusy = 0; 109 mcp->m_memReadWriteBusy = 0; 110 mcp->m_memDataBusBusy = 0; 111 mcp->m_memTfawBusy = 0; 112 mcp->m_memRefresh = 0; 113 mcp->m_memRead = 0; 114 mcp->m_memWrite = 0; 115 mcp->m_memWaitCycles = 0; 116 mcp->m_memInputQ = 0; 117 mcp->m_memBankQ = 0; 118 mcp->m_memArbWait = 0; 119 mcp->m_memRandBusy = 0; 120 mcp->m_memNotOld = 0; 121 122 mcp->m_banks_per_rank = RubySystem::getMemoryControl((*it).c_str())->getBanksPerRank(); 123 mcp->m_ranks_per_dimm = RubySystem::getMemoryControl((*it).c_str())->getRanksPerDimm(); 124 mcp->m_dimms_per_channel = RubySystem::getMemoryControl((*it).c_str())->getDimmsPerChannel(); 125 126 int totalBanks = mcp->m_banks_per_rank 127 * mcp->m_ranks_per_dimm 128 * mcp->m_dimms_per_channel; 129 130 mcp->m_memBankCount.setSize(totalBanks); 131 132 m_memory_control_profilers [(*it).c_str()] = mcp; 133 } 134 135 clearStats(); 136 m_hot_lines = false; 137 m_all_instructions = false; 138 139 for (size_t i=0; i<argv.size(); i+=2) { 140 if ( argv[i] == "hot_lines") { 141 m_hot_lines = (argv[i+1]=="true"); 142 } else if ( argv[i] == "all_instructions") { 143 m_all_instructions = (argv[i+1]=="true"); 144 }else { 145 cerr << "WARNING: Profiler: Unkown configuration parameter: " << argv[i] << endl; 146 assert(false); 147 } 148 } 149 150 m_address_profiler_ptr = new AddressProfiler; 151 m_address_profiler_ptr -> setHotLines(m_hot_lines); 152 m_address_profiler_ptr -> setAllInstructions(m_all_instructions); 153 154 if (m_all_instructions) { 155 m_inst_profiler_ptr = new AddressProfiler; 156 m_inst_profiler_ptr -> setHotLines(m_hot_lines); 157 m_inst_profiler_ptr -> setAllInstructions(m_all_instructions); 158 } 159} 160 161void Profiler::wakeup() 162{ 163 // FIXME - avoid the repeated code 164
|
185 Vector<integer_t> perProcInstructionCount; 186 perProcInstructionCount.setSize(RubySystem::getNumberOfSequencers()); 187
| |
188 Vector<integer_t> perProcCycleCount; 189 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers()); 190 191 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
| 165 Vector<integer_t> perProcCycleCount; 166 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers()); 167 168 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
|
192 perProcInstructionCount[i] = g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1;
| |
193 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1; 194 // The +1 allows us to avoid division by zero 195 } 196 197 integer_t total_misses = m_perProcTotalMisses.sum();
| 169 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1; 170 // The +1 allows us to avoid division by zero 171 } 172 173 integer_t total_misses = m_perProcTotalMisses.sum();
|
198 integer_t instruction_executed = perProcInstructionCount.sum();
| |
199 integer_t simics_cycles_executed = perProcCycleCount.sum(); 200 integer_t transactions_started = m_perProcStartTransaction.sum(); 201 integer_t transactions_ended = m_perProcEndTransaction.sum(); 202 203 (*m_periodic_output_file_ptr) << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl; 204 (*m_periodic_output_file_ptr) << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl;
| 174 integer_t simics_cycles_executed = perProcCycleCount.sum(); 175 integer_t transactions_started = m_perProcStartTransaction.sum(); 176 integer_t transactions_ended = m_perProcEndTransaction.sum(); 177 178 (*m_periodic_output_file_ptr) << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl; 179 (*m_periodic_output_file_ptr) << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl;
|
205 (*m_periodic_output_file_ptr) << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
| |
206 (*m_periodic_output_file_ptr) << "simics_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl; 207 (*m_periodic_output_file_ptr) << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 208 (*m_periodic_output_file_ptr) << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
| 180 (*m_periodic_output_file_ptr) << "simics_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl; 181 (*m_periodic_output_file_ptr) << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 182 (*m_periodic_output_file_ptr) << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
|
209 (*m_periodic_output_file_ptr) << "L1TBE_usage: " << m_L1tbeProfile << endl; 210 (*m_periodic_output_file_ptr) << "L2TBE_usage: " << m_L2tbeProfile << endl;
| |
211 (*m_periodic_output_file_ptr) << "mbytes_resident: " << process_memory_resident() << endl; 212 (*m_periodic_output_file_ptr) << "mbytes_total: " << process_memory_total() << endl; 213 if (process_memory_total() > 0) { 214 (*m_periodic_output_file_ptr) << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 215 } 216 (*m_periodic_output_file_ptr) << "miss_latency: " << m_allMissLatencyHistogram << endl; 217 218 *m_periodic_output_file_ptr << endl; 219 220 if (m_all_instructions) { 221 m_inst_profiler_ptr->printStats(*m_periodic_output_file_ptr); 222 } 223 224 //g_system_ptr->getNetwork()->printStats(*m_periodic_output_file_ptr); 225 g_eventQueue_ptr->scheduleEvent(this, m_stats_period); 226} 227 228void Profiler::setPeriodicStatsFile(const string& filename) 229{ 230 cout << "Recording periodic statistics to file '" << filename << "' every " 231 << m_stats_period << " Ruby cycles" << endl; 232 233 if (m_periodic_output_file_ptr != &cerr) { 234 delete m_periodic_output_file_ptr; 235 } 236 237 m_periodic_output_file_ptr = new ofstream(filename.c_str()); 238 g_eventQueue_ptr->scheduleEvent(this, 1); 239} 240 241void Profiler::setPeriodicStatsInterval(integer_t period) 242{ 243 cout << "Recording periodic statistics every " << m_stats_period << " Ruby cycles" << endl; 244 m_stats_period = period; 245 g_eventQueue_ptr->scheduleEvent(this, 1); 246} 247 248void Profiler::printConfig(ostream& out) const 249{ 250 out << endl; 251 out << "Profiler Configuration" << endl; 252 out << "----------------------" << endl; 253 out << "periodic_stats_period: " << m_stats_period << endl; 254} 255 256void Profiler::print(ostream& out) const 257{ 258 out << "[Profiler]"; 259} 260 261void Profiler::printStats(ostream& out, bool short_stats) 262{ 263 out << endl; 264 if (short_stats) { 265 out << "SHORT "; 266 } 267 out << "Profiler Stats" << endl; 268 out << "--------------" << endl; 269 270 time_t real_time_current = time(NULL); 271 double seconds = difftime(real_time_current, m_real_time_start_time); 272 double minutes = seconds/60.0; 273 double hours = minutes/60.0; 274 double days = hours/24.0; 275 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start; 276 277 if (!short_stats) { 278 out << "Elapsed_time_in_seconds: " << seconds << endl; 279 out << "Elapsed_time_in_minutes: " << minutes << endl; 280 out << "Elapsed_time_in_hours: " << hours << endl; 281 out << "Elapsed_time_in_days: " << days << endl; 282 out << endl; 283 } 284 285 // print the virtual runtimes as well 286 struct tms vtime; 287 times(&vtime); 288 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0; 289 minutes = seconds / 60.0; 290 hours = minutes / 60.0; 291 days = hours / 24.0; 292 out << "Virtual_time_in_seconds: " << seconds << endl; 293 out << "Virtual_time_in_minutes: " << minutes << endl; 294 out << "Virtual_time_in_hours: " << hours << endl;
| 183 (*m_periodic_output_file_ptr) << "mbytes_resident: " << process_memory_resident() << endl; 184 (*m_periodic_output_file_ptr) << "mbytes_total: " << process_memory_total() << endl; 185 if (process_memory_total() > 0) { 186 (*m_periodic_output_file_ptr) << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 187 } 188 (*m_periodic_output_file_ptr) << "miss_latency: " << m_allMissLatencyHistogram << endl; 189 190 *m_periodic_output_file_ptr << endl; 191 192 if (m_all_instructions) { 193 m_inst_profiler_ptr->printStats(*m_periodic_output_file_ptr); 194 } 195 196 //g_system_ptr->getNetwork()->printStats(*m_periodic_output_file_ptr); 197 g_eventQueue_ptr->scheduleEvent(this, m_stats_period); 198} 199 200void Profiler::setPeriodicStatsFile(const string& filename) 201{ 202 cout << "Recording periodic statistics to file '" << filename << "' every " 203 << m_stats_period << " Ruby cycles" << endl; 204 205 if (m_periodic_output_file_ptr != &cerr) { 206 delete m_periodic_output_file_ptr; 207 } 208 209 m_periodic_output_file_ptr = new ofstream(filename.c_str()); 210 g_eventQueue_ptr->scheduleEvent(this, 1); 211} 212 213void Profiler::setPeriodicStatsInterval(integer_t period) 214{ 215 cout << "Recording periodic statistics every " << m_stats_period << " Ruby cycles" << endl; 216 m_stats_period = period; 217 g_eventQueue_ptr->scheduleEvent(this, 1); 218} 219 220void Profiler::printConfig(ostream& out) const 221{ 222 out << endl; 223 out << "Profiler Configuration" << endl; 224 out << "----------------------" << endl; 225 out << "periodic_stats_period: " << m_stats_period << endl; 226} 227 228void Profiler::print(ostream& out) const 229{ 230 out << "[Profiler]"; 231} 232 233void Profiler::printStats(ostream& out, bool short_stats) 234{ 235 out << endl; 236 if (short_stats) { 237 out << "SHORT "; 238 } 239 out << "Profiler Stats" << endl; 240 out << "--------------" << endl; 241 242 time_t real_time_current = time(NULL); 243 double seconds = difftime(real_time_current, m_real_time_start_time); 244 double minutes = seconds/60.0; 245 double hours = minutes/60.0; 246 double days = hours/24.0; 247 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start; 248 249 if (!short_stats) { 250 out << "Elapsed_time_in_seconds: " << seconds << endl; 251 out << "Elapsed_time_in_minutes: " << minutes << endl; 252 out << "Elapsed_time_in_hours: " << hours << endl; 253 out << "Elapsed_time_in_days: " << days << endl; 254 out << endl; 255 } 256 257 // print the virtual runtimes as well 258 struct tms vtime; 259 times(&vtime); 260 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0; 261 minutes = seconds / 60.0; 262 hours = minutes / 60.0; 263 days = hours / 24.0; 264 out << "Virtual_time_in_seconds: " << seconds << endl; 265 out << "Virtual_time_in_minutes: " << minutes << endl; 266 out << "Virtual_time_in_hours: " << hours << endl;
|
295 out << "Virtual_time_in_days: " << hours << endl;
| 267 out << "Virtual_time_in_days: " << days << endl;
|
296 out << endl; 297 298 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl; 299 out << "Ruby_start_time: " << m_ruby_start << endl; 300 out << "Ruby_cycles: " << ruby_cycles << endl; 301 out << endl; 302 303 if (!short_stats) { 304 out << "mbytes_resident: " << process_memory_resident() << endl; 305 out << "mbytes_total: " << process_memory_total() << endl; 306 if (process_memory_total() > 0) { 307 out << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 308 } 309 out << endl; 310
| 268 out << endl; 269 270 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl; 271 out << "Ruby_start_time: " << m_ruby_start << endl; 272 out << "Ruby_cycles: " << ruby_cycles << endl; 273 out << endl; 274 275 if (!short_stats) { 276 out << "mbytes_resident: " << process_memory_resident() << endl; 277 out << "mbytes_total: " << process_memory_total() << endl; 278 if (process_memory_total() > 0) { 279 out << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 280 } 281 out << endl; 282
|
311 if(m_num_BA_broadcasts + m_num_BA_unicasts != 0){ 312 out << endl; 313 out << "Broadcast_percent: " << (float)m_num_BA_broadcasts/(m_num_BA_broadcasts+m_num_BA_unicasts) << endl; 314 }
| |
315 } 316
| 283 } 284
|
317 Vector<integer_t> perProcInstructionCount;
| |
318 Vector<integer_t> perProcCycleCount;
| 285 Vector<integer_t> perProcCycleCount;
|
319 Vector<double> perProcCPI; 320 Vector<double> perProcMissesPerInsn; 321 Vector<double> perProcInsnPerTrans;
| |
322 Vector<double> perProcCyclesPerTrans; 323 Vector<double> perProcMissesPerTrans; 324
| 286 Vector<double> perProcCyclesPerTrans; 287 Vector<double> perProcMissesPerTrans; 288
|
325 perProcInstructionCount.setSize(RubySystem::getNumberOfSequencers()); 326 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers()); 327 perProcCPI.setSize(RubySystem::getNumberOfSequencers()); 328 perProcMissesPerInsn.setSize(RubySystem::getNumberOfSequencers());
| |
329
| 289
|
330 perProcInsnPerTrans.setSize(RubySystem::getNumberOfSequencers());
| 290 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers());
|
331 perProcCyclesPerTrans.setSize(RubySystem::getNumberOfSequencers()); 332 perProcMissesPerTrans.setSize(RubySystem::getNumberOfSequencers()); 333 334 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
| 291 perProcCyclesPerTrans.setSize(RubySystem::getNumberOfSequencers()); 292 perProcMissesPerTrans.setSize(RubySystem::getNumberOfSequencers()); 293 294 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
|
335 perProcInstructionCount[i] = g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1;
| |
336 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1; 337 // The +1 allows us to avoid division by zero
| 295 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1; 296 // The +1 allows us to avoid division by zero
|
338 perProcCPI[i] = double(ruby_cycles)/perProcInstructionCount[i]; 339 perProcMissesPerInsn[i] = 1000.0 * (double(m_perProcTotalMisses[i]) / double(perProcInstructionCount[i]));
| |
340 341 int trans = m_perProcEndTransaction[i]; 342 if (trans == 0) {
| 297 298 int trans = m_perProcEndTransaction[i]; 299 if (trans == 0) {
|
343 perProcInsnPerTrans[i] = 0;
| |
344 perProcCyclesPerTrans[i] = 0; 345 perProcMissesPerTrans[i] = 0; 346 } else {
| 300 perProcCyclesPerTrans[i] = 0; 301 perProcMissesPerTrans[i] = 0; 302 } else {
|
347 perProcInsnPerTrans[i] = perProcInstructionCount[i] / double(trans);
| |
348 perProcCyclesPerTrans[i] = ruby_cycles / double(trans); 349 perProcMissesPerTrans[i] = m_perProcTotalMisses[i] / double(trans); 350 } 351 } 352 353 integer_t total_misses = m_perProcTotalMisses.sum(); 354 integer_t user_misses = m_perProcUserMisses.sum(); 355 integer_t supervisor_misses = m_perProcSupervisorMisses.sum();
| 303 perProcCyclesPerTrans[i] = ruby_cycles / double(trans); 304 perProcMissesPerTrans[i] = m_perProcTotalMisses[i] / double(trans); 305 } 306 } 307 308 integer_t total_misses = m_perProcTotalMisses.sum(); 309 integer_t user_misses = m_perProcUserMisses.sum(); 310 integer_t supervisor_misses = m_perProcSupervisorMisses.sum();
|
356 integer_t instruction_executed = perProcInstructionCount.sum();
| |
357 integer_t simics_cycles_executed = perProcCycleCount.sum(); 358 integer_t transactions_started = m_perProcStartTransaction.sum(); 359 integer_t transactions_ended = m_perProcEndTransaction.sum(); 360
| 311 integer_t simics_cycles_executed = perProcCycleCount.sum(); 312 integer_t transactions_started = m_perProcStartTransaction.sum(); 313 integer_t transactions_ended = m_perProcEndTransaction.sum(); 314
|
361 double instructions_per_transaction = (transactions_ended != 0) ? double(instruction_executed) / double(transactions_ended) : 0;
| |
362 double cycles_per_transaction = (transactions_ended != 0) ? (RubySystem::getNumberOfSequencers() * double(ruby_cycles)) / double(transactions_ended) : 0; 363 double misses_per_transaction = (transactions_ended != 0) ? double(total_misses) / double(transactions_ended) : 0; 364 365 out << "Total_misses: " << total_misses << endl; 366 out << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 367 out << "user_misses: " << user_misses << " " << m_perProcUserMisses << endl; 368 out << "supervisor_misses: " << supervisor_misses << " " << m_perProcSupervisorMisses << endl; 369 out << endl;
| 315 double cycles_per_transaction = (transactions_ended != 0) ? (RubySystem::getNumberOfSequencers() * double(ruby_cycles)) / double(transactions_ended) : 0; 316 double misses_per_transaction = (transactions_ended != 0) ? double(total_misses) / double(transactions_ended) : 0; 317 318 out << "Total_misses: " << total_misses << endl; 319 out << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 320 out << "user_misses: " << user_misses << " " << m_perProcUserMisses << endl; 321 out << "supervisor_misses: " << supervisor_misses << " " << m_perProcSupervisorMisses << endl; 322 out << endl;
|
370 out << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
| |
371 out << "ruby_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl;
| 323 out << "ruby_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl;
|
372 out << "cycles_per_instruction: " << (RubySystem::getNumberOfSequencers()*double(ruby_cycles))/double(instruction_executed) << " " << perProcCPI << endl; 373 out << "misses_per_thousand_instructions: " << 1000.0 * (double(total_misses) / double(instruction_executed)) << " " << perProcMissesPerInsn << endl;
| |
374 out << endl; 375 out << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 376 out << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
| 324 out << endl; 325 out << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 326 out << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
|
377 out << "instructions_per_transaction: " << instructions_per_transaction << " " << perProcInsnPerTrans << endl;
| |
378 out << "cycles_per_transaction: " << cycles_per_transaction << " " << perProcCyclesPerTrans << endl; 379 out << "misses_per_transaction: " << misses_per_transaction << " " << perProcMissesPerTrans << endl; 380 381 out << endl; 382
| 327 out << "cycles_per_transaction: " << cycles_per_transaction << " " << perProcCyclesPerTrans << endl; 328 out << "misses_per_transaction: " << misses_per_transaction << " " << perProcMissesPerTrans << endl; 329 330 out << endl; 331
|
383 // m_L1D_cache_profiler_ptr->printStats(out); 384 // m_L1I_cache_profiler_ptr->printStats(out); 385 // m_L2_cache_profiler_ptr->printStats(out); 386
| |
387 out << endl; 388 389 vector<string>::iterator it; 390 391 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 392 long long int m_memReq = m_memory_control_profilers[(*it).c_str()] -> m_memReq; 393 long long int m_memRefresh = m_memory_control_profilers[(*it).c_str()] -> m_memRefresh; 394 long long int m_memInputQ = m_memory_control_profilers[(*it).c_str()] -> m_memInputQ; 395 long long int m_memBankQ = m_memory_control_profilers[(*it).c_str()] -> m_memBankQ; 396 long long int m_memWaitCycles = m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles; 397 long long int m_memRead = m_memory_control_profilers[(*it).c_str()] -> m_memRead; 398 long long int m_memWrite = m_memory_control_profilers[(*it).c_str()] -> m_memWrite; 399 long long int m_memBankBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy; 400 long long int m_memRandBusy = m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy; 401 long long int m_memNotOld = m_memory_control_profilers[(*it).c_str()] -> m_memNotOld; 402 long long int m_memArbWait = m_memory_control_profilers[(*it).c_str()] -> m_memArbWait; 403 long long int m_memBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy; 404 long long int m_memTfawBusy = m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy; 405 long long int m_memReadWriteBusy = m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy; 406 long long int m_memDataBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy; 407 Vector<long long int> m_memBankCount = m_memory_control_profilers[(*it).c_str()] -> m_memBankCount; 408 409 if (m_memReq || m_memRefresh) { // if there's a memory controller at all 410 long long int total_stalls = m_memInputQ + m_memBankQ + m_memWaitCycles; 411 double stallsPerReq = total_stalls * 1.0 / m_memReq;
| 332 out << endl; 333 334 vector<string>::iterator it; 335 336 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 337 long long int m_memReq = m_memory_control_profilers[(*it).c_str()] -> m_memReq; 338 long long int m_memRefresh = m_memory_control_profilers[(*it).c_str()] -> m_memRefresh; 339 long long int m_memInputQ = m_memory_control_profilers[(*it).c_str()] -> m_memInputQ; 340 long long int m_memBankQ = m_memory_control_profilers[(*it).c_str()] -> m_memBankQ; 341 long long int m_memWaitCycles = m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles; 342 long long int m_memRead = m_memory_control_profilers[(*it).c_str()] -> m_memRead; 343 long long int m_memWrite = m_memory_control_profilers[(*it).c_str()] -> m_memWrite; 344 long long int m_memBankBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy; 345 long long int m_memRandBusy = m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy; 346 long long int m_memNotOld = m_memory_control_profilers[(*it).c_str()] -> m_memNotOld; 347 long long int m_memArbWait = m_memory_control_profilers[(*it).c_str()] -> m_memArbWait; 348 long long int m_memBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy; 349 long long int m_memTfawBusy = m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy; 350 long long int m_memReadWriteBusy = m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy; 351 long long int m_memDataBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy; 352 Vector<long long int> m_memBankCount = m_memory_control_profilers[(*it).c_str()] -> m_memBankCount; 353 354 if (m_memReq || m_memRefresh) { // if there's a memory controller at all 355 long long int total_stalls = m_memInputQ + m_memBankQ + m_memWaitCycles; 356 double stallsPerReq = total_stalls * 1.0 / m_memReq;
|
412 out << "Memory control:" << endl;
| 357 out << "Memory control " << (*it) << ":" << endl;
|
413 out << " memory_total_requests: " << m_memReq << endl; // does not include refreshes 414 out << " memory_reads: " << m_memRead << endl; 415 out << " memory_writes: " << m_memWrite << endl; 416 out << " memory_refreshes: " << m_memRefresh << endl; 417 out << " memory_total_request_delays: " << total_stalls << endl; 418 out << " memory_delays_per_request: " << stallsPerReq << endl; 419 out << " memory_delays_in_input_queue: " << m_memInputQ << endl; 420 out << " memory_delays_behind_head_of_bank_queue: " << m_memBankQ << endl; 421 out << " memory_delays_stalled_at_head_of_bank_queue: " << m_memWaitCycles << endl; 422 // Note: The following "memory stalls" entries are a breakdown of the 423 // cycles which already showed up in m_memWaitCycles. The order is 424 // significant; it is the priority of attributing the cycles. 425 // For example, bank_busy is before arbitration because if the bank was 426 // busy, we didn't even check arbitration. 427 // Note: "not old enough" means that since we grouped waiting heads-of-queues 428 // into batches to avoid starvation, a request in a newer batch 429 // didn't try to arbitrate yet because there are older requests waiting. 430 out << " memory_stalls_for_bank_busy: " << m_memBankBusy << endl; 431 out << " memory_stalls_for_random_busy: " << m_memRandBusy << endl; 432 out << " memory_stalls_for_anti_starvation: " << m_memNotOld << endl; 433 out << " memory_stalls_for_arbitration: " << m_memArbWait << endl; 434 out << " memory_stalls_for_bus: " << m_memBusBusy << endl; 435 out << " memory_stalls_for_tfaw: " << m_memTfawBusy << endl; 436 out << " memory_stalls_for_read_write_turnaround: " << m_memReadWriteBusy << endl; 437 out << " memory_stalls_for_read_read_turnaround: " << m_memDataBusBusy << endl; 438 out << " accesses_per_bank: "; 439 for (int bank=0; bank < m_memBankCount.size(); bank++) { 440 out << m_memBankCount[bank] << " "; 441 //if ((bank % 8) == 7) out << " " << endl; 442 } 443 out << endl; 444 out << endl; 445 } 446 } 447 if (!short_stats) { 448 out << "Busy Controller Counts:" << endl; 449 for(int i=0; i < MachineType_NUM; i++) { 450 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 451 MachineID machID; 452 machID.type = (MachineType)i; 453 machID.num = j; 454 out << machID << ":" << m_busyControllerCount[i][j] << " "; 455 if ((j+1)%8 == 0) { 456 out << endl; 457 } 458 } 459 out << endl; 460 } 461 out << endl; 462 463 out << "Busy Bank Count:" << m_busyBankCount << endl; 464 out << endl; 465
| 358 out << " memory_total_requests: " << m_memReq << endl; // does not include refreshes 359 out << " memory_reads: " << m_memRead << endl; 360 out << " memory_writes: " << m_memWrite << endl; 361 out << " memory_refreshes: " << m_memRefresh << endl; 362 out << " memory_total_request_delays: " << total_stalls << endl; 363 out << " memory_delays_per_request: " << stallsPerReq << endl; 364 out << " memory_delays_in_input_queue: " << m_memInputQ << endl; 365 out << " memory_delays_behind_head_of_bank_queue: " << m_memBankQ << endl; 366 out << " memory_delays_stalled_at_head_of_bank_queue: " << m_memWaitCycles << endl; 367 // Note: The following "memory stalls" entries are a breakdown of the 368 // cycles which already showed up in m_memWaitCycles. The order is 369 // significant; it is the priority of attributing the cycles. 370 // For example, bank_busy is before arbitration because if the bank was 371 // busy, we didn't even check arbitration. 372 // Note: "not old enough" means that since we grouped waiting heads-of-queues 373 // into batches to avoid starvation, a request in a newer batch 374 // didn't try to arbitrate yet because there are older requests waiting. 375 out << " memory_stalls_for_bank_busy: " << m_memBankBusy << endl; 376 out << " memory_stalls_for_random_busy: " << m_memRandBusy << endl; 377 out << " memory_stalls_for_anti_starvation: " << m_memNotOld << endl; 378 out << " memory_stalls_for_arbitration: " << m_memArbWait << endl; 379 out << " memory_stalls_for_bus: " << m_memBusBusy << endl; 380 out << " memory_stalls_for_tfaw: " << m_memTfawBusy << endl; 381 out << " memory_stalls_for_read_write_turnaround: " << m_memReadWriteBusy << endl; 382 out << " memory_stalls_for_read_read_turnaround: " << m_memDataBusBusy << endl; 383 out << " accesses_per_bank: "; 384 for (int bank=0; bank < m_memBankCount.size(); bank++) { 385 out << m_memBankCount[bank] << " "; 386 //if ((bank % 8) == 7) out << " " << endl; 387 } 388 out << endl; 389 out << endl; 390 } 391 } 392 if (!short_stats) { 393 out << "Busy Controller Counts:" << endl; 394 for(int i=0; i < MachineType_NUM; i++) { 395 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 396 MachineID machID; 397 machID.type = (MachineType)i; 398 machID.num = j; 399 out << machID << ":" << m_busyControllerCount[i][j] << " "; 400 if ((j+1)%8 == 0) { 401 out << endl; 402 } 403 } 404 out << endl; 405 } 406 out << endl; 407 408 out << "Busy Bank Count:" << m_busyBankCount << endl; 409 out << endl; 410
|
466 out << "L1TBE_usage: " << m_L1tbeProfile << endl; 467 out << "L2TBE_usage: " << m_L2tbeProfile << endl; 468 out << "StopTable_usage: " << m_stopTableProfile << endl;
| |
469 out << "sequencer_requests_outstanding: " << m_sequencer_requests << endl;
| 411 out << "sequencer_requests_outstanding: " << m_sequencer_requests << endl;
|
470 out << "store_buffer_size: " << m_store_buffer_size << endl; 471 out << "unique_blocks_in_store_buffer: " << m_store_buffer_blocks << endl;
| |
472 out << endl; 473 } 474 475 if (!short_stats) { 476 out << "All Non-Zero Cycle Demand Cache Accesses" << endl; 477 out << "----------------------------------------" << endl; 478 out << "miss_latency: " << m_allMissLatencyHistogram << endl; 479 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 480 if (m_missLatencyHistograms[i].size() > 0) { 481 out << "miss_latency_" << RubyRequestType(i) << ": " << m_missLatencyHistograms[i] << endl; 482 } 483 } 484 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 485 if (m_machLatencyHistograms[i].size() > 0) { 486 out << "miss_latency_" << GenericMachineType(i) << ": " << m_machLatencyHistograms[i] << endl; 487 } 488 }
| 412 out << endl; 413 } 414 415 if (!short_stats) { 416 out << "All Non-Zero Cycle Demand Cache Accesses" << endl; 417 out << "----------------------------------------" << endl; 418 out << "miss_latency: " << m_allMissLatencyHistogram << endl; 419 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 420 if (m_missLatencyHistograms[i].size() > 0) { 421 out << "miss_latency_" << RubyRequestType(i) << ": " << m_missLatencyHistograms[i] << endl; 422 } 423 } 424 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 425 if (m_machLatencyHistograms[i].size() > 0) { 426 out << "miss_latency_" << GenericMachineType(i) << ": " << m_machLatencyHistograms[i] << endl; 427 } 428 }
|
489 out << "miss_latency_L2Miss: " << m_L2MissLatencyHistogram << endl;
| |
490 491 out << endl; 492 493 out << "All Non-Zero Cycle SW Prefetch Requests" << endl; 494 out << "------------------------------------" << endl; 495 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl; 496 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 497 if (m_SWPrefetchLatencyHistograms[i].size() > 0) { 498 out << "prefetch_latency_" << CacheRequestType(i) << ": " << m_SWPrefetchLatencyHistograms[i] << endl; 499 } 500 } 501 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 502 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) { 503 out << "prefetch_latency_" << GenericMachineType(i) << ": " << m_SWPrefetchMachLatencyHistograms[i] << endl; 504 } 505 } 506 out << "prefetch_latency_L2Miss:" << m_SWPrefetchL2MissLatencyHistogram << endl; 507
| 429 430 out << endl; 431 432 out << "All Non-Zero Cycle SW Prefetch Requests" << endl; 433 out << "------------------------------------" << endl; 434 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl; 435 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 436 if (m_SWPrefetchLatencyHistograms[i].size() > 0) { 437 out << "prefetch_latency_" << CacheRequestType(i) << ": " << m_SWPrefetchLatencyHistograms[i] << endl; 438 } 439 } 440 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 441 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) { 442 out << "prefetch_latency_" << GenericMachineType(i) << ": " << m_SWPrefetchMachLatencyHistograms[i] << endl; 443 } 444 } 445 out << "prefetch_latency_L2Miss:" << m_SWPrefetchL2MissLatencyHistogram << endl; 446
|
508 out << "multicast_retries: " << m_multicast_retry_histogram << endl; 509 out << "gets_mask_prediction_count: " << m_gets_mask_prediction << endl; 510 out << "getx_mask_prediction_count: " << m_getx_mask_prediction << endl; 511 out << "explicit_training_mask: " << m_explicit_training_mask << endl; 512 out << endl; 513
| |
514 if (m_all_sharing_histogram.size() > 0) { 515 out << "all_sharing: " << m_all_sharing_histogram << endl; 516 out << "read_sharing: " << m_read_sharing_histogram << endl; 517 out << "write_sharing: " << m_write_sharing_histogram << endl; 518 519 out << "all_sharing_percent: "; m_all_sharing_histogram.printPercent(out); out << endl; 520 out << "read_sharing_percent: "; m_read_sharing_histogram.printPercent(out); out << endl; 521 out << "write_sharing_percent: "; m_write_sharing_histogram.printPercent(out); out << endl; 522 523 int64 total_miss = m_cache_to_cache + m_memory_to_cache; 524 out << "all_misses: " << total_miss << endl; 525 out << "cache_to_cache_misses: " << m_cache_to_cache << endl; 526 out << "memory_to_cache_misses: " << m_memory_to_cache << endl; 527 out << "cache_to_cache_percent: " << 100.0 * (double(m_cache_to_cache) / double(total_miss)) << endl; 528 out << "memory_to_cache_percent: " << 100.0 * (double(m_memory_to_cache) / double(total_miss)) << endl; 529 out << endl; 530 } 531
| 447 if (m_all_sharing_histogram.size() > 0) { 448 out << "all_sharing: " << m_all_sharing_histogram << endl; 449 out << "read_sharing: " << m_read_sharing_histogram << endl; 450 out << "write_sharing: " << m_write_sharing_histogram << endl; 451 452 out << "all_sharing_percent: "; m_all_sharing_histogram.printPercent(out); out << endl; 453 out << "read_sharing_percent: "; m_read_sharing_histogram.printPercent(out); out << endl; 454 out << "write_sharing_percent: "; m_write_sharing_histogram.printPercent(out); out << endl; 455 456 int64 total_miss = m_cache_to_cache + m_memory_to_cache; 457 out << "all_misses: " << total_miss << endl; 458 out << "cache_to_cache_misses: " << m_cache_to_cache << endl; 459 out << "memory_to_cache_misses: " << m_memory_to_cache << endl; 460 out << "cache_to_cache_percent: " << 100.0 * (double(m_cache_to_cache) / double(total_miss)) << endl; 461 out << "memory_to_cache_percent: " << 100.0 * (double(m_memory_to_cache) / double(total_miss)) << endl; 462 out << endl; 463 } 464
|
532 if (m_conflicting_histogram.size() > 0) { 533 out << "conflicting_histogram: " << m_conflicting_histogram << endl; 534 out << "conflicting_histogram_percent: "; m_conflicting_histogram.printPercent(out); out << endl; 535 out << endl; 536 } 537
| |
538 if (m_outstanding_requests.size() > 0) { 539 out << "outstanding_requests: "; m_outstanding_requests.printPercent(out); out << endl;
| 465 if (m_outstanding_requests.size() > 0) { 466 out << "outstanding_requests: "; m_outstanding_requests.printPercent(out); out << endl;
|
540 if (m_outstanding_persistent_requests.size() > 0) { 541 out << "outstanding_persistent_requests: "; m_outstanding_persistent_requests.printPercent(out); out << endl; 542 }
| |
543 out << endl; 544 } 545 } 546 547 if (!short_stats) { 548 out << "Request vs. RubySystem State Profile" << endl; 549 out << "--------------------------------" << endl; 550 out << endl; 551 552 Vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys(); 553 requestProfileKeys.sortVector(); 554 555 for(int i=0; i<requestProfileKeys.size(); i++) { 556 int temp_int = m_requestProfileMap_ptr->lookup(requestProfileKeys[i]); 557 double percent = (100.0*double(temp_int))/double(m_requests); 558 while (requestProfileKeys[i] != "") { 559 out << setw(10) << string_split(requestProfileKeys[i], ':'); 560 } 561 out << setw(11) << temp_int; 562 out << setw(14) << percent << endl; 563 } 564 out << endl; 565 566 out << "filter_action: " << m_filter_action_histogram << endl; 567 568 if (!m_all_instructions) { 569 m_address_profiler_ptr->printStats(out); 570 } 571 572 if (m_all_instructions) { 573 m_inst_profiler_ptr->printStats(out); 574 } 575 576 out << endl; 577 out << "Message Delayed Cycles" << endl; 578 out << "----------------------" << endl; 579 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl; 580 out << "Total_nonPF_delay_cycles: " << m_delayedCyclesNonPFHistogram << endl; 581 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) { 582 out << " virtual_network_" << i << "_delay_cycles: " << m_delayedCyclesVCHistograms[i] << endl; 583 } 584 585 printResourceUsage(out); 586 } 587 588} 589 590void Profiler::printResourceUsage(ostream& out) const 591{ 592 out << endl; 593 out << "Resource Usage" << endl; 594 out << "--------------" << endl; 595 596 integer_t pagesize = getpagesize(); // page size in bytes 597 out << "page_size: " << pagesize << endl; 598 599 rusage usage; 600 getrusage (RUSAGE_SELF, &usage); 601 602 out << "user_time: " << usage.ru_utime.tv_sec << endl; 603 out << "system_time: " << usage.ru_stime.tv_sec << endl; 604 out << "page_reclaims: " << usage.ru_minflt << endl; 605 out << "page_faults: " << usage.ru_majflt << endl; 606 out << "swaps: " << usage.ru_nswap << endl; 607 out << "block_inputs: " << usage.ru_inblock << endl; 608 out << "block_outputs: " << usage.ru_oublock << endl; 609} 610 611void Profiler::clearStats() 612{
| 467 out << endl; 468 } 469 } 470 471 if (!short_stats) { 472 out << "Request vs. RubySystem State Profile" << endl; 473 out << "--------------------------------" << endl; 474 out << endl; 475 476 Vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys(); 477 requestProfileKeys.sortVector(); 478 479 for(int i=0; i<requestProfileKeys.size(); i++) { 480 int temp_int = m_requestProfileMap_ptr->lookup(requestProfileKeys[i]); 481 double percent = (100.0*double(temp_int))/double(m_requests); 482 while (requestProfileKeys[i] != "") { 483 out << setw(10) << string_split(requestProfileKeys[i], ':'); 484 } 485 out << setw(11) << temp_int; 486 out << setw(14) << percent << endl; 487 } 488 out << endl; 489 490 out << "filter_action: " << m_filter_action_histogram << endl; 491 492 if (!m_all_instructions) { 493 m_address_profiler_ptr->printStats(out); 494 } 495 496 if (m_all_instructions) { 497 m_inst_profiler_ptr->printStats(out); 498 } 499 500 out << endl; 501 out << "Message Delayed Cycles" << endl; 502 out << "----------------------" << endl; 503 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl; 504 out << "Total_nonPF_delay_cycles: " << m_delayedCyclesNonPFHistogram << endl; 505 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) { 506 out << " virtual_network_" << i << "_delay_cycles: " << m_delayedCyclesVCHistograms[i] << endl; 507 } 508 509 printResourceUsage(out); 510 } 511 512} 513 514void Profiler::printResourceUsage(ostream& out) const 515{ 516 out << endl; 517 out << "Resource Usage" << endl; 518 out << "--------------" << endl; 519 520 integer_t pagesize = getpagesize(); // page size in bytes 521 out << "page_size: " << pagesize << endl; 522 523 rusage usage; 524 getrusage (RUSAGE_SELF, &usage); 525 526 out << "user_time: " << usage.ru_utime.tv_sec << endl; 527 out << "system_time: " << usage.ru_stime.tv_sec << endl; 528 out << "page_reclaims: " << usage.ru_minflt << endl; 529 out << "page_faults: " << usage.ru_majflt << endl; 530 out << "swaps: " << usage.ru_nswap << endl; 531 out << "block_inputs: " << usage.ru_inblock << endl; 532 out << "block_outputs: " << usage.ru_oublock << endl; 533} 534 535void Profiler::clearStats() 536{
|
613 m_num_BA_unicasts = 0; 614 m_num_BA_broadcasts = 0; 615
| |
616 m_ruby_start = g_eventQueue_ptr->getTime(); 617
| 537 m_ruby_start = g_eventQueue_ptr->getTime(); 538
|
618 m_instructions_executed_at_start.setSize(RubySystem::getNumberOfSequencers());
| |
619 m_cycles_executed_at_start.setSize(RubySystem::getNumberOfSequencers()); 620 for (int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 621 if (g_system_ptr == NULL) {
| 539 m_cycles_executed_at_start.setSize(RubySystem::getNumberOfSequencers()); 540 for (int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 541 if (g_system_ptr == NULL) {
|
622 m_instructions_executed_at_start[i] = 0;
| |
623 m_cycles_executed_at_start[i] = 0; 624 } else {
| 542 m_cycles_executed_at_start[i] = 0; 543 } else {
|
625 m_instructions_executed_at_start[i] = g_system_ptr->getInstructionCount(i);
| |
626 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i); 627 } 628 } 629 630 m_perProcTotalMisses.setSize(RubySystem::getNumberOfSequencers()); 631 m_perProcUserMisses.setSize(RubySystem::getNumberOfSequencers()); 632 m_perProcSupervisorMisses.setSize(RubySystem::getNumberOfSequencers()); 633 m_perProcStartTransaction.setSize(RubySystem::getNumberOfSequencers()); 634 m_perProcEndTransaction.setSize(RubySystem::getNumberOfSequencers()); 635 636 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 637 m_perProcTotalMisses[i] = 0; 638 m_perProcUserMisses[i] = 0; 639 m_perProcSupervisorMisses[i] = 0; 640 m_perProcStartTransaction[i] = 0; 641 m_perProcEndTransaction[i] = 0; 642 } 643 644 m_busyControllerCount.setSize(MachineType_NUM); // all machines 645 for(int i=0; i < MachineType_NUM; i++) { 646 m_busyControllerCount[i].setSize(MachineType_base_count((MachineType)i)); 647 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 648 m_busyControllerCount[i][j] = 0; 649 } 650 } 651 m_busyBankCount = 0; 652 653 m_delayedCyclesHistogram.clear(); 654 m_delayedCyclesNonPFHistogram.clear(); 655 m_delayedCyclesVCHistograms.setSize(RubySystem::getNetwork()->getNumberOfVirtualNetworks()); 656 for (int i = 0; i < RubySystem::getNetwork()->getNumberOfVirtualNetworks(); i++) { 657 m_delayedCyclesVCHistograms[i].clear(); 658 } 659
| 544 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i); 545 } 546 } 547 548 m_perProcTotalMisses.setSize(RubySystem::getNumberOfSequencers()); 549 m_perProcUserMisses.setSize(RubySystem::getNumberOfSequencers()); 550 m_perProcSupervisorMisses.setSize(RubySystem::getNumberOfSequencers()); 551 m_perProcStartTransaction.setSize(RubySystem::getNumberOfSequencers()); 552 m_perProcEndTransaction.setSize(RubySystem::getNumberOfSequencers()); 553 554 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 555 m_perProcTotalMisses[i] = 0; 556 m_perProcUserMisses[i] = 0; 557 m_perProcSupervisorMisses[i] = 0; 558 m_perProcStartTransaction[i] = 0; 559 m_perProcEndTransaction[i] = 0; 560 } 561 562 m_busyControllerCount.setSize(MachineType_NUM); // all machines 563 for(int i=0; i < MachineType_NUM; i++) { 564 m_busyControllerCount[i].setSize(MachineType_base_count((MachineType)i)); 565 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 566 m_busyControllerCount[i][j] = 0; 567 } 568 } 569 m_busyBankCount = 0; 570 571 m_delayedCyclesHistogram.clear(); 572 m_delayedCyclesNonPFHistogram.clear(); 573 m_delayedCyclesVCHistograms.setSize(RubySystem::getNetwork()->getNumberOfVirtualNetworks()); 574 for (int i = 0; i < RubySystem::getNetwork()->getNumberOfVirtualNetworks(); i++) { 575 m_delayedCyclesVCHistograms[i].clear(); 576 } 577
|
660 m_gets_mask_prediction.clear(); 661 m_getx_mask_prediction.clear(); 662 m_explicit_training_mask.clear(); 663 664 m_missLatencyHistograms.setSize(CacheRequestType_NUM);
| 578 m_missLatencyHistograms.setSize(RubyRequestType_NUM);
|
665 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 666 m_missLatencyHistograms[i].clear(200); 667 } 668 m_machLatencyHistograms.setSize(GenericMachineType_NUM+1); 669 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 670 m_machLatencyHistograms[i].clear(200); 671 } 672 m_allMissLatencyHistogram.clear(200);
| 579 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 580 m_missLatencyHistograms[i].clear(200); 581 } 582 m_machLatencyHistograms.setSize(GenericMachineType_NUM+1); 583 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 584 m_machLatencyHistograms[i].clear(200); 585 } 586 m_allMissLatencyHistogram.clear(200);
|
673 m_L2MissLatencyHistogram.clear(200);
| |
674 675 m_SWPrefetchLatencyHistograms.setSize(CacheRequestType_NUM); 676 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 677 m_SWPrefetchLatencyHistograms[i].clear(200); 678 } 679 m_SWPrefetchMachLatencyHistograms.setSize(GenericMachineType_NUM+1); 680 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 681 m_SWPrefetchMachLatencyHistograms[i].clear(200); 682 } 683 m_allSWPrefetchLatencyHistogram.clear(200);
| 587 588 m_SWPrefetchLatencyHistograms.setSize(CacheRequestType_NUM); 589 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 590 m_SWPrefetchLatencyHistograms[i].clear(200); 591 } 592 m_SWPrefetchMachLatencyHistograms.setSize(GenericMachineType_NUM+1); 593 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 594 m_SWPrefetchMachLatencyHistograms[i].clear(200); 595 } 596 m_allSWPrefetchLatencyHistogram.clear(200);
|
684 m_SWPrefetchL2MissLatencyHistogram.clear(200);
| |
685
| 597
|
686 m_multicast_retry_histogram.clear(); 687 688 m_L1tbeProfile.clear(); 689 m_L2tbeProfile.clear(); 690 m_stopTableProfile.clear(); 691 m_filter_action_histogram.clear(); 692
| |
693 m_sequencer_requests.clear();
| 598 m_sequencer_requests.clear();
|
694 m_store_buffer_size.clear(); 695 m_store_buffer_blocks.clear();
| |
696 m_read_sharing_histogram.clear(); 697 m_write_sharing_histogram.clear(); 698 m_all_sharing_histogram.clear(); 699 m_cache_to_cache = 0; 700 m_memory_to_cache = 0; 701
| 599 m_read_sharing_histogram.clear(); 600 m_write_sharing_histogram.clear(); 601 m_all_sharing_histogram.clear(); 602 m_cache_to_cache = 0; 603 m_memory_to_cache = 0; 604
|
702 m_predictions = 0; 703 m_predictionOpportunities = 0; 704 m_goodPredictions = 0; 705
| |
706 // clear HashMaps 707 m_requestProfileMap_ptr->clear(); 708 709 // count requests profiled 710 m_requests = 0; 711
| 605 // clear HashMaps 606 m_requestProfileMap_ptr->clear(); 607 608 // count requests profiled 609 m_requests = 0; 610
|
712 // Conflicting requests 713 m_conflicting_map_ptr->clear(); 714 m_conflicting_histogram.clear(); 715
| |
716 m_outstanding_requests.clear(); 717 m_outstanding_persistent_requests.clear(); 718
| 611 m_outstanding_requests.clear(); 612 m_outstanding_persistent_requests.clear(); 613
|
719 m_L1D_cache_profiler_ptr->clearStats(); 720 m_L1I_cache_profiler_ptr->clearStats(); 721 m_L2_cache_profiler_ptr->clearStats(); 722 723 // for MemoryControl: 724/* 725 m_memReq = 0; 726 m_memBankBusy = 0; 727 m_memBusBusy = 0; 728 m_memTfawBusy = 0; 729 m_memReadWriteBusy = 0; 730 m_memDataBusBusy = 0; 731 m_memRefresh = 0; 732 m_memRead = 0; 733 m_memWrite = 0; 734 m_memWaitCycles = 0; 735 m_memInputQ = 0; 736 m_memBankQ = 0; 737 m_memArbWait = 0; 738 m_memRandBusy = 0; 739 m_memNotOld = 0; 740 741 for (int bank=0; bank < m_memBankCount.size(); bank++) { 742 m_memBankCount[bank] = 0; 743 } 744*/
| |
745//added by SS 746 vector<string>::iterator it; 747 748 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 749 m_memory_control_profilers[(*it).c_str()] -> m_memReq = 0; 750 m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy = 0; 751 m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy = 0; 752 m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy = 0; 753 m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy = 0; 754 m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy = 0; 755 m_memory_control_profilers[(*it).c_str()] -> m_memRefresh = 0; 756 m_memory_control_profilers[(*it).c_str()] -> m_memRead = 0; 757 m_memory_control_profilers[(*it).c_str()] -> m_memWrite = 0; 758 m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles = 0; 759 m_memory_control_profilers[(*it).c_str()] -> m_memInputQ = 0; 760 m_memory_control_profilers[(*it).c_str()] -> m_memBankQ = 0; 761 m_memory_control_profilers[(*it).c_str()] -> m_memArbWait = 0; 762 m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy = 0; 763 m_memory_control_profilers[(*it).c_str()] -> m_memNotOld = 0; 764 765 for (int bank=0; bank < m_memory_control_profilers[(*it).c_str()] -> m_memBankCount.size(); bank++) { 766 m_memory_control_profilers[(*it).c_str()] -> m_memBankCount[bank] = 0; 767 } 768 } 769 // Flush the prefetches through the system - used so that there are no outstanding requests after stats are cleared 770 //g_eventQueue_ptr->triggerAllEvents(); 771 772 // update the start time 773 m_ruby_start = g_eventQueue_ptr->getTime(); 774} 775
| 614//added by SS 615 vector<string>::iterator it; 616 617 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 618 m_memory_control_profilers[(*it).c_str()] -> m_memReq = 0; 619 m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy = 0; 620 m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy = 0; 621 m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy = 0; 622 m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy = 0; 623 m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy = 0; 624 m_memory_control_profilers[(*it).c_str()] -> m_memRefresh = 0; 625 m_memory_control_profilers[(*it).c_str()] -> m_memRead = 0; 626 m_memory_control_profilers[(*it).c_str()] -> m_memWrite = 0; 627 m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles = 0; 628 m_memory_control_profilers[(*it).c_str()] -> m_memInputQ = 0; 629 m_memory_control_profilers[(*it).c_str()] -> m_memBankQ = 0; 630 m_memory_control_profilers[(*it).c_str()] -> m_memArbWait = 0; 631 m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy = 0; 632 m_memory_control_profilers[(*it).c_str()] -> m_memNotOld = 0; 633 634 for (int bank=0; bank < m_memory_control_profilers[(*it).c_str()] -> m_memBankCount.size(); bank++) { 635 m_memory_control_profilers[(*it).c_str()] -> m_memBankCount[bank] = 0; 636 } 637 } 638 // Flush the prefetches through the system - used so that there are no outstanding requests after stats are cleared 639 //g_eventQueue_ptr->triggerAllEvents(); 640 641 // update the start time 642 m_ruby_start = g_eventQueue_ptr->getTime(); 643} 644
|
776void Profiler::profileConflictingRequests(const Address& addr) 777{ 778 assert(addr == line_address(addr)); 779 Time last_time = m_ruby_start; 780 if (m_conflicting_map_ptr->exist(addr)) { 781 last_time = m_conflicting_map_ptr->lookup(addr); 782 } 783 Time current_time = g_eventQueue_ptr->getTime(); 784 assert (current_time - last_time > 0); 785 m_conflicting_histogram.add(current_time - last_time); 786 m_conflicting_map_ptr->add(addr, current_time); 787} 788
| |
789void Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id) 790{ 791 if (msg.getType() != CacheRequestType_IFETCH) { 792 793 // Note: The following line should be commented out if you want to 794 // use the special profiling that is part of the GS320 protocol 795 796 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be profiled by the AddressProfiler 797 m_address_profiler_ptr->addTraceSample(msg.getLineAddress(), msg.getProgramCounter(), msg.getType(), msg.getAccessMode(), id, false); 798 } 799} 800 801void Profiler::profileSharing(const Address& addr, AccessType type, NodeID requestor, const Set& sharers, const Set& owner) 802{ 803 Set set_contacted(owner); 804 if (type == AccessType_Write) { 805 set_contacted.addSet(sharers); 806 } 807 set_contacted.remove(requestor); 808 int number_contacted = set_contacted.count(); 809 810 if (type == AccessType_Write) { 811 m_write_sharing_histogram.add(number_contacted); 812 } else { 813 m_read_sharing_histogram.add(number_contacted); 814 } 815 m_all_sharing_histogram.add(number_contacted); 816 817 if (number_contacted == 0) { 818 m_memory_to_cache++; 819 } else { 820 m_cache_to_cache++; 821 } 822 823} 824 825void Profiler::profileMsgDelay(int virtualNetwork, int delayCycles) { 826 assert(virtualNetwork < m_delayedCyclesVCHistograms.size()); 827 m_delayedCyclesHistogram.add(delayCycles); 828 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles); 829 if (virtualNetwork != 0) { 830 m_delayedCyclesNonPFHistogram.add(delayCycles); 831 } 832} 833 834// profiles original cache requests including PUTs 835void Profiler::profileRequest(const string& requestStr) 836{ 837 m_requests++; 838 839 if (m_requestProfileMap_ptr->exist(requestStr)) { 840 (m_requestProfileMap_ptr->lookup(requestStr))++; 841 } else { 842 m_requestProfileMap_ptr->add(requestStr, 1); 843 } 844} 845
| 645void Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id) 646{ 647 if (msg.getType() != CacheRequestType_IFETCH) { 648 649 // Note: The following line should be commented out if you want to 650 // use the special profiling that is part of the GS320 protocol 651 652 // NOTE: Unless PROFILE_HOT_LINES is enabled, nothing will be profiled by the AddressProfiler 653 m_address_profiler_ptr->addTraceSample(msg.getLineAddress(), msg.getProgramCounter(), msg.getType(), msg.getAccessMode(), id, false); 654 } 655} 656 657void Profiler::profileSharing(const Address& addr, AccessType type, NodeID requestor, const Set& sharers, const Set& owner) 658{ 659 Set set_contacted(owner); 660 if (type == AccessType_Write) { 661 set_contacted.addSet(sharers); 662 } 663 set_contacted.remove(requestor); 664 int number_contacted = set_contacted.count(); 665 666 if (type == AccessType_Write) { 667 m_write_sharing_histogram.add(number_contacted); 668 } else { 669 m_read_sharing_histogram.add(number_contacted); 670 } 671 m_all_sharing_histogram.add(number_contacted); 672 673 if (number_contacted == 0) { 674 m_memory_to_cache++; 675 } else { 676 m_cache_to_cache++; 677 } 678 679} 680 681void Profiler::profileMsgDelay(int virtualNetwork, int delayCycles) { 682 assert(virtualNetwork < m_delayedCyclesVCHistograms.size()); 683 m_delayedCyclesHistogram.add(delayCycles); 684 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles); 685 if (virtualNetwork != 0) { 686 m_delayedCyclesNonPFHistogram.add(delayCycles); 687 } 688} 689 690// profiles original cache requests including PUTs 691void Profiler::profileRequest(const string& requestStr) 692{ 693 m_requests++; 694 695 if (m_requestProfileMap_ptr->exist(requestStr)) { 696 (m_requestProfileMap_ptr->lookup(requestStr))++; 697 } else { 698 m_requestProfileMap_ptr->add(requestStr, 1); 699 } 700} 701
|
846void Profiler::recordPrediction(bool wasGood, bool wasPredicted) 847{ 848 m_predictionOpportunities++; 849 if(wasPredicted){ 850 m_predictions++; 851 if(wasGood){ 852 m_goodPredictions++; 853 } 854 } 855} 856 857void Profiler::profileFilterAction(int action) 858{ 859 m_filter_action_histogram.add(action); 860} 861 862void Profiler::profileMulticastRetry(const Address& addr, int count) 863{ 864 m_multicast_retry_histogram.add(count); 865} 866
| |
867void Profiler::startTransaction(int cpu) 868{ 869 m_perProcStartTransaction[cpu]++; 870} 871 872void Profiler::endTransaction(int cpu) 873{ 874 m_perProcEndTransaction[cpu]++; 875} 876 877void Profiler::controllerBusy(MachineID machID) 878{ 879 m_busyControllerCount[(int)machID.type][(int)machID.num]++; 880} 881 882void Profiler::profilePFWait(Time waitTime) 883{ 884 m_prefetchWaitHistogram.add(waitTime); 885} 886 887void Profiler::bankBusy() 888{ 889 m_busyBankCount++; 890} 891 892// non-zero cycle demand request 893void Profiler::missLatency(Time t, RubyRequestType type) 894{ 895 m_allMissLatencyHistogram.add(t); 896 m_missLatencyHistograms[type].add(t);
| 702void Profiler::startTransaction(int cpu) 703{ 704 m_perProcStartTransaction[cpu]++; 705} 706 707void Profiler::endTransaction(int cpu) 708{ 709 m_perProcEndTransaction[cpu]++; 710} 711 712void Profiler::controllerBusy(MachineID machID) 713{ 714 m_busyControllerCount[(int)machID.type][(int)machID.num]++; 715} 716 717void Profiler::profilePFWait(Time waitTime) 718{ 719 m_prefetchWaitHistogram.add(waitTime); 720} 721 722void Profiler::bankBusy() 723{ 724 m_busyBankCount++; 725} 726 727// non-zero cycle demand request 728void Profiler::missLatency(Time t, RubyRequestType type) 729{ 730 m_allMissLatencyHistogram.add(t); 731 m_missLatencyHistograms[type].add(t);
|
897 /* 898 m_machLatencyHistograms[respondingMach].add(t); 899 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 900 m_L2MissLatencyHistogram.add(t); 901 } 902 */
| |
903} 904 905// non-zero cycle prefetch request 906void Profiler::swPrefetchLatency(Time t, CacheRequestType type, GenericMachineType respondingMach) 907{ 908 m_allSWPrefetchLatencyHistogram.add(t); 909 m_SWPrefetchLatencyHistograms[type].add(t); 910 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t); 911 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 912 m_SWPrefetchL2MissLatencyHistogram.add(t); 913 } 914} 915 916void Profiler::profileTransition(const string& component, NodeID version, Address addr, 917 const string& state, const string& event, 918 const string& next_state, const string& note) 919{ 920 const int EVENT_SPACES = 20; 921 const int ID_SPACES = 3; 922 const int TIME_SPACES = 7; 923 const int COMP_SPACES = 10; 924 const int STATE_SPACES = 6; 925 926 if ((g_debug_ptr->getDebugTime() > 0) && 927 (g_eventQueue_ptr->getTime() >= g_debug_ptr->getDebugTime())) { 928 (* debug_cout_ptr).flags(ios::right); 929 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " "; 930 (* debug_cout_ptr) << setw(ID_SPACES) << version << " "; 931 (* debug_cout_ptr) << setw(COMP_SPACES) << component; 932 (* debug_cout_ptr) << setw(EVENT_SPACES) << event << " "; 933 934 (* debug_cout_ptr).flags(ios::right); 935 (* debug_cout_ptr) << setw(STATE_SPACES) << state; 936 (* debug_cout_ptr) << ">"; 937 (* debug_cout_ptr).flags(ios::left); 938 (* debug_cout_ptr) << setw(STATE_SPACES) << next_state; 939 940 (* debug_cout_ptr) << " " << addr << " " << note; 941 942 (* debug_cout_ptr) << endl; 943 } 944} 945 946// Helper function 947static double process_memory_total() 948{ 949 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 950 ifstream proc_file; 951 proc_file.open("/proc/self/statm"); 952 int total_size_in_pages = 0; 953 int res_size_in_pages = 0; 954 proc_file >> total_size_in_pages; 955 proc_file >> res_size_in_pages; 956 return double(total_size_in_pages)*MULTIPLIER; // size in megabytes 957} 958 959static double process_memory_resident() 960{ 961 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 962 ifstream proc_file; 963 proc_file.open("/proc/self/statm"); 964 int total_size_in_pages = 0; 965 int res_size_in_pages = 0; 966 proc_file >> total_size_in_pages; 967 proc_file >> res_size_in_pages; 968 return double(res_size_in_pages)*MULTIPLIER; // size in megabytes 969} 970
| 732} 733 734// non-zero cycle prefetch request 735void Profiler::swPrefetchLatency(Time t, CacheRequestType type, GenericMachineType respondingMach) 736{ 737 m_allSWPrefetchLatencyHistogram.add(t); 738 m_SWPrefetchLatencyHistograms[type].add(t); 739 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t); 740 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 741 m_SWPrefetchL2MissLatencyHistogram.add(t); 742 } 743} 744 745void Profiler::profileTransition(const string& component, NodeID version, Address addr, 746 const string& state, const string& event, 747 const string& next_state, const string& note) 748{ 749 const int EVENT_SPACES = 20; 750 const int ID_SPACES = 3; 751 const int TIME_SPACES = 7; 752 const int COMP_SPACES = 10; 753 const int STATE_SPACES = 6; 754 755 if ((g_debug_ptr->getDebugTime() > 0) && 756 (g_eventQueue_ptr->getTime() >= g_debug_ptr->getDebugTime())) { 757 (* debug_cout_ptr).flags(ios::right); 758 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " "; 759 (* debug_cout_ptr) << setw(ID_SPACES) << version << " "; 760 (* debug_cout_ptr) << setw(COMP_SPACES) << component; 761 (* debug_cout_ptr) << setw(EVENT_SPACES) << event << " "; 762 763 (* debug_cout_ptr).flags(ios::right); 764 (* debug_cout_ptr) << setw(STATE_SPACES) << state; 765 (* debug_cout_ptr) << ">"; 766 (* debug_cout_ptr).flags(ios::left); 767 (* debug_cout_ptr) << setw(STATE_SPACES) << next_state; 768 769 (* debug_cout_ptr) << " " << addr << " " << note; 770 771 (* debug_cout_ptr) << endl; 772 } 773} 774 775// Helper function 776static double process_memory_total() 777{ 778 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 779 ifstream proc_file; 780 proc_file.open("/proc/self/statm"); 781 int total_size_in_pages = 0; 782 int res_size_in_pages = 0; 783 proc_file >> total_size_in_pages; 784 proc_file >> res_size_in_pages; 785 return double(total_size_in_pages)*MULTIPLIER; // size in megabytes 786} 787 788static double process_memory_resident() 789{ 790 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 791 ifstream proc_file; 792 proc_file.open("/proc/self/statm"); 793 int total_size_in_pages = 0; 794 int res_size_in_pages = 0; 795 proc_file >> total_size_in_pages; 796 proc_file >> res_size_in_pages; 797 return double(res_size_in_pages)*MULTIPLIER; // size in megabytes 798} 799
|
971void Profiler::profileGetXMaskPrediction(const Set& pred_set) 972{ 973 m_getx_mask_prediction.add(pred_set.count()); 974} 975 976void Profiler::profileGetSMaskPrediction(const Set& pred_set) 977{ 978 m_gets_mask_prediction.add(pred_set.count()); 979} 980 981void Profiler::profileTrainingMask(const Set& pred_set) 982{ 983 m_explicit_training_mask.add(pred_set.count()); 984} 985 986int64 Profiler::getTotalInstructionsExecuted() const 987{ 988 int64 sum = 1; // Starting at 1 allows us to avoid division by zero 989 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 990 sum += (g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i]); 991 } 992 return sum; 993} 994 995int64 Profiler::getTotalTransactionsExecuted() const 996{ 997 int64 sum = m_perProcEndTransaction.sum(); 998 if (sum > 0) { 999 return sum; 1000 } else { 1001 return 1; // Avoid division by zero errors 1002 } 1003} 1004 1005
| |
1006void Profiler::rubyWatch(int id){ 1007 //int rn_g1 = 0;//SIMICS_get_register_number(id, "g1"); 1008 uint64 tr = 0;//SIMICS_read_register(id, rn_g1); 1009 Address watch_address = Address(tr); 1010 const int ID_SPACES = 3; 1011 const int TIME_SPACES = 7; 1012 1013 (* debug_cout_ptr).flags(ios::right); 1014 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " "; 1015 (* debug_cout_ptr) << setw(ID_SPACES) << id << " " 1016 << "RUBY WATCH " 1017 << watch_address 1018 << endl; 1019 1020 if(!m_watch_address_list_ptr->exist(watch_address)){ 1021 m_watch_address_list_ptr->add(watch_address, 1); 1022 } 1023} 1024 1025bool Profiler::watchAddress(Address addr){ 1026 if (m_watch_address_list_ptr->exist(addr)) 1027 return true; 1028 else 1029 return false; 1030} 1031
| 800void Profiler::rubyWatch(int id){ 801 //int rn_g1 = 0;//SIMICS_get_register_number(id, "g1"); 802 uint64 tr = 0;//SIMICS_read_register(id, rn_g1); 803 Address watch_address = Address(tr); 804 const int ID_SPACES = 3; 805 const int TIME_SPACES = 7; 806 807 (* debug_cout_ptr).flags(ios::right); 808 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " "; 809 (* debug_cout_ptr) << setw(ID_SPACES) << id << " " 810 << "RUBY WATCH " 811 << watch_address 812 << endl; 813 814 if(!m_watch_address_list_ptr->exist(watch_address)){ 815 m_watch_address_list_ptr->add(watch_address, 1); 816 } 817} 818 819bool Profiler::watchAddress(Address addr){ 820 if (m_watch_address_list_ptr->exist(addr)) 821 return true; 822 else 823 return false; 824} 825
|
| 826int64 Profiler::getTotalTransactionsExecuted() const { 827 return m_perProcEndTransaction.sum(); 828} 829
|
1032// For MemoryControl: 1033void Profiler::profileMemReq(string name, int bank) { 1034// printf("name is %s", name.c_str()); 1035 assert(m_memory_control_profilers.count(name) == 1); 1036 m_memory_control_profilers[name] -> m_memReq++; 1037 m_memory_control_profilers[name] -> m_memBankCount[bank]++; 1038} 1039void Profiler::profileMemBankBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankBusy++; } 1040void Profiler::profileMemBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBusBusy++; } 1041void Profiler::profileMemReadWriteBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memReadWriteBusy++; } 1042void Profiler::profileMemDataBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memDataBusBusy++; } 1043void Profiler::profileMemTfawBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memTfawBusy++; } 1044void Profiler::profileMemRefresh(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRefresh++; } 1045void Profiler::profileMemRead(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRead++; } 1046void Profiler::profileMemWrite(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWrite++; } 1047void Profiler::profileMemWaitCycles(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWaitCycles += cycles; } 1048void Profiler::profileMemInputQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memInputQ += cycles; } 1049void Profiler::profileMemBankQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankQ += cycles; } 1050void Profiler::profileMemArbWait(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memArbWait += cycles; } 1051void Profiler::profileMemRandBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRandBusy++; } 1052void Profiler::profileMemNotOld(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memNotOld++; } 1053
| 830// For MemoryControl: 831void Profiler::profileMemReq(string name, int bank) { 832// printf("name is %s", name.c_str()); 833 assert(m_memory_control_profilers.count(name) == 1); 834 m_memory_control_profilers[name] -> m_memReq++; 835 m_memory_control_profilers[name] -> m_memBankCount[bank]++; 836} 837void Profiler::profileMemBankBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankBusy++; } 838void Profiler::profileMemBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBusBusy++; } 839void Profiler::profileMemReadWriteBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memReadWriteBusy++; } 840void Profiler::profileMemDataBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memDataBusBusy++; } 841void Profiler::profileMemTfawBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memTfawBusy++; } 842void Profiler::profileMemRefresh(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRefresh++; } 843void Profiler::profileMemRead(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRead++; } 844void Profiler::profileMemWrite(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWrite++; } 845void Profiler::profileMemWaitCycles(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWaitCycles += cycles; } 846void Profiler::profileMemInputQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memInputQ += cycles; } 847void Profiler::profileMemBankQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankQ += cycles; } 848void Profiler::profileMemArbWait(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memArbWait += cycles; } 849void Profiler::profileMemRandBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRandBusy++; } 850void Profiler::profileMemNotOld(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memNotOld++; } 851
|