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"
| 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/ruby/common/Driver.hh"
| |
62#include "mem/protocol/Protocol.hh" 63#include "mem/gems_common/util.hh" 64#include "mem/gems_common/Map.hh" 65#include "mem/ruby/common/Debug.hh" 66#include "mem/protocol/MachineType.hh" 67 68// Allows use of times() library call, which determines virtual runtime 69#include <sys/times.h> 70 71extern std::ostream * debug_cout_ptr; 72 73static double process_memory_total(); 74static double process_memory_resident(); 75
| 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
|
76Profiler::Profiler()
| 75Profiler::Profiler(const string & name)
|
77 : m_conflicting_histogram(-1) 78{
| 76 : m_conflicting_histogram(-1) 77{
|
| 78 m_name = name;
|
79 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 83 m_L2_cache_profiler_ptr = new CacheProfiler("L2_cache"); 84
| 79 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 83 m_L2_cache_profiler_ptr = new CacheProfiler("L2_cache"); 84
|
| 85 m_inst_profiler_ptr = NULL; 86 m_address_profiler_ptr = NULL; 87 88/*
|
85 m_address_profiler_ptr = new AddressProfiler; 86 m_inst_profiler_ptr = NULL;
| 89 m_address_profiler_ptr = new AddressProfiler; 90 m_inst_profiler_ptr = NULL;
|
87 if (PROFILE_ALL_INSTRUCTIONS) {
| 91 if (m_all_instructions) {
|
88 m_inst_profiler_ptr = new AddressProfiler; 89 }
| 92 m_inst_profiler_ptr = new AddressProfiler; 93 }
|
90
| 94*/
|
91 m_conflicting_map_ptr = new Map<Address, Time>; 92 93 m_real_time_start_time = time(NULL); // Not reset in clearStats() 94 m_stats_period = 1000000; // Default 95 m_periodic_output_file_ptr = &cerr; 96
| 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//changed by SS 102/*
|
97 // for MemoryControl: 98 m_memReq = 0; 99 m_memBankBusy = 0; 100 m_memBusBusy = 0; 101 m_memReadWriteBusy = 0; 102 m_memDataBusBusy = 0; 103 m_memTfawBusy = 0; 104 m_memRefresh = 0; 105 m_memRead = 0; 106 m_memWrite = 0; 107 m_memWaitCycles = 0; 108 m_memInputQ = 0; 109 m_memBankQ = 0; 110 m_memArbWait = 0; 111 m_memRandBusy = 0; 112 m_memNotOld = 0; 113 114 115 int totalBanks = RubyConfig::banksPerRank() 116 * RubyConfig::ranksPerDimm() 117 * RubyConfig::dimmsPerChannel(); 118 m_memBankCount.setSize(totalBanks);
| 103 // for MemoryControl: 104 m_memReq = 0; 105 m_memBankBusy = 0; 106 m_memBusBusy = 0; 107 m_memReadWriteBusy = 0; 108 m_memDataBusBusy = 0; 109 m_memTfawBusy = 0; 110 m_memRefresh = 0; 111 m_memRead = 0; 112 m_memWrite = 0; 113 m_memWaitCycles = 0; 114 m_memInputQ = 0; 115 m_memBankQ = 0; 116 m_memArbWait = 0; 117 m_memRandBusy = 0; 118 m_memNotOld = 0; 119 120 121 int totalBanks = RubyConfig::banksPerRank() 122 * RubyConfig::ranksPerDimm() 123 * RubyConfig::dimmsPerChannel(); 124 m_memBankCount.setSize(totalBanks);
|
119 120 clearStats();
| 125*/
|
121} 122 123Profiler::~Profiler() 124{ 125 if (m_periodic_output_file_ptr != &cerr) { 126 delete m_periodic_output_file_ptr; 127 } 128 delete m_address_profiler_ptr; 129 delete m_L1D_cache_profiler_ptr; 130 delete m_L1I_cache_profiler_ptr; 131 delete m_L2_cache_profiler_ptr; 132 delete m_requestProfileMap_ptr; 133 delete m_conflicting_map_ptr; 134} 135
| 126} 127 128Profiler::~Profiler() 129{ 130 if (m_periodic_output_file_ptr != &cerr) { 131 delete m_periodic_output_file_ptr; 132 } 133 delete m_address_profiler_ptr; 134 delete m_L1D_cache_profiler_ptr; 135 delete m_L1I_cache_profiler_ptr; 136 delete m_L2_cache_profiler_ptr; 137 delete m_requestProfileMap_ptr; 138 delete m_conflicting_map_ptr; 139} 140
|
| 141void Profiler::init(const vector<string> & argv, vector<string> memory_control_names) 142{ 143 // added by SS 144 vector<string>::iterator it; 145 memory_control_profiler* mcp; 146 m_memory_control_names = memory_control_names; 147// printf ( "Here in Profiler::init \n"); 148 for ( it=memory_control_names.begin() ; it < memory_control_names.end(); it++ ){ 149// printf ( "Here in Profiler::init memory control name %s \n", (*it).c_str()); 150 mcp = new memory_control_profiler; 151 mcp->m_memReq = 0; 152 mcp->m_memBankBusy = 0; 153 mcp->m_memBusBusy = 0; 154 mcp->m_memReadWriteBusy = 0; 155 mcp->m_memDataBusBusy = 0; 156 mcp->m_memTfawBusy = 0; 157 mcp->m_memRefresh = 0; 158 mcp->m_memRead = 0; 159 mcp->m_memWrite = 0; 160 mcp->m_memWaitCycles = 0; 161 mcp->m_memInputQ = 0; 162 mcp->m_memBankQ = 0; 163 mcp->m_memArbWait = 0; 164 mcp->m_memRandBusy = 0; 165 mcp->m_memNotOld = 0; 166 167 mcp->m_banks_per_rank = RubySystem::getMemoryControl((*it).c_str())->getBanksPerRank(); 168 mcp->m_ranks_per_dimm = RubySystem::getMemoryControl((*it).c_str())->getRanksPerDimm(); 169 mcp->m_dimms_per_channel = RubySystem::getMemoryControl((*it).c_str())->getDimmsPerChannel(); 170 171 int totalBanks = mcp->m_banks_per_rank 172 * mcp->m_ranks_per_dimm 173 * mcp->m_dimms_per_channel; 174 175 mcp->m_memBankCount.setSize(totalBanks); 176 177 m_memory_control_profilers [(*it).c_str()] = mcp; 178 } 179 180 clearStats(); 181 m_hot_lines = false; 182 m_all_instructions = false; 183 184 for (size_t i=0; i<argv.size(); i+=2) { 185 if ( argv[i] == "hot_lines") { 186 m_hot_lines = (argv[i+1]=="true"); 187 } else if ( argv[i] == "all_instructions") { 188 m_all_instructions = (argv[i+1]=="true"); 189 }else { 190 cerr << "WARNING: Profiler: Unkown configuration parameter: " << argv[i] << endl; 191 assert(false); 192 } 193 } 194 195 m_address_profiler_ptr = new AddressProfiler; 196 m_address_profiler_ptr -> setHotLines(m_hot_lines); 197 m_address_profiler_ptr -> setAllInstructions(m_all_instructions); 198 199 if (m_all_instructions) { 200 m_inst_profiler_ptr = new AddressProfiler; 201 m_inst_profiler_ptr -> setHotLines(m_hot_lines); 202 m_inst_profiler_ptr -> setAllInstructions(m_all_instructions); 203 } 204} 205
|
136void Profiler::wakeup() 137{ 138 // FIXME - avoid the repeated code 139 140 Vector<integer_t> perProcInstructionCount;
| 206void Profiler::wakeup() 207{ 208 // FIXME - avoid the repeated code 209 210 Vector<integer_t> perProcInstructionCount;
|
141 perProcInstructionCount.setSize(RubyConfig::numberOfProcessors());
| 211 perProcInstructionCount.setSize(RubySystem::getNumberOfSequencers());
|
142 143 Vector<integer_t> perProcCycleCount;
| 212 213 Vector<integer_t> perProcCycleCount;
|
144 perProcCycleCount.setSize(RubyConfig::numberOfProcessors());
| 214 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers());
|
145
| 215
|
146 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) { 147 perProcInstructionCount[i] = g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1; 148 perProcCycleCount[i] = g_system_ptr->getDriver()->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
| 216 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 217 perProcInstructionCount[i] = g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1; 218 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
|
149 // The +1 allows us to avoid division by zero 150 } 151 152 integer_t total_misses = m_perProcTotalMisses.sum(); 153 integer_t instruction_executed = perProcInstructionCount.sum();
| 219 // The +1 allows us to avoid division by zero 220 } 221 222 integer_t total_misses = m_perProcTotalMisses.sum(); 223 integer_t instruction_executed = perProcInstructionCount.sum();
|
154 integer_t cycles_executed = perProcCycleCount.sum();
| 224 integer_t simics_cycles_executed = perProcCycleCount.sum();
|
155 integer_t transactions_started = m_perProcStartTransaction.sum(); 156 integer_t transactions_ended = m_perProcEndTransaction.sum(); 157 158 (*m_periodic_output_file_ptr) << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl; 159 (*m_periodic_output_file_ptr) << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 160 (*m_periodic_output_file_ptr) << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
| 225 integer_t transactions_started = m_perProcStartTransaction.sum(); 226 integer_t transactions_ended = m_perProcEndTransaction.sum(); 227 228 (*m_periodic_output_file_ptr) << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl; 229 (*m_periodic_output_file_ptr) << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 230 (*m_periodic_output_file_ptr) << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
|
161 (*m_periodic_output_file_ptr) << "cycles_executed: " << cycles_executed << " " << perProcCycleCount << endl;
| 231 (*m_periodic_output_file_ptr) << "simics_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl;
|
162 (*m_periodic_output_file_ptr) << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 163 (*m_periodic_output_file_ptr) << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl; 164 (*m_periodic_output_file_ptr) << "L1TBE_usage: " << m_L1tbeProfile << endl; 165 (*m_periodic_output_file_ptr) << "L2TBE_usage: " << m_L2tbeProfile << endl; 166 (*m_periodic_output_file_ptr) << "mbytes_resident: " << process_memory_resident() << endl; 167 (*m_periodic_output_file_ptr) << "mbytes_total: " << process_memory_total() << endl; 168 if (process_memory_total() > 0) { 169 (*m_periodic_output_file_ptr) << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 170 } 171 (*m_periodic_output_file_ptr) << "miss_latency: " << m_allMissLatencyHistogram << endl; 172 173 *m_periodic_output_file_ptr << endl; 174
| 232 (*m_periodic_output_file_ptr) << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 233 (*m_periodic_output_file_ptr) << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl; 234 (*m_periodic_output_file_ptr) << "L1TBE_usage: " << m_L1tbeProfile << endl; 235 (*m_periodic_output_file_ptr) << "L2TBE_usage: " << m_L2tbeProfile << endl; 236 (*m_periodic_output_file_ptr) << "mbytes_resident: " << process_memory_resident() << endl; 237 (*m_periodic_output_file_ptr) << "mbytes_total: " << process_memory_total() << endl; 238 if (process_memory_total() > 0) { 239 (*m_periodic_output_file_ptr) << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 240 } 241 (*m_periodic_output_file_ptr) << "miss_latency: " << m_allMissLatencyHistogram << endl; 242 243 *m_periodic_output_file_ptr << endl; 244
|
175 if (PROFILE_ALL_INSTRUCTIONS) {
| 245 if (m_all_instructions) {
|
176 m_inst_profiler_ptr->printStats(*m_periodic_output_file_ptr); 177 } 178 179 //g_system_ptr->getNetwork()->printStats(*m_periodic_output_file_ptr); 180 g_eventQueue_ptr->scheduleEvent(this, m_stats_period); 181} 182 183void Profiler::setPeriodicStatsFile(const string& filename) 184{ 185 cout << "Recording periodic statistics to file '" << filename << "' every " 186 << m_stats_period << " Ruby cycles" << endl; 187 188 if (m_periodic_output_file_ptr != &cerr) { 189 delete m_periodic_output_file_ptr; 190 } 191 192 m_periodic_output_file_ptr = new ofstream(filename.c_str()); 193 g_eventQueue_ptr->scheduleEvent(this, 1); 194} 195 196void Profiler::setPeriodicStatsInterval(integer_t period) 197{ 198 cout << "Recording periodic statistics every " << m_stats_period << " Ruby cycles" << endl; 199 m_stats_period = period; 200 g_eventQueue_ptr->scheduleEvent(this, 1); 201} 202 203void Profiler::printConfig(ostream& out) const 204{ 205 out << endl; 206 out << "Profiler Configuration" << endl; 207 out << "----------------------" << endl; 208 out << "periodic_stats_period: " << m_stats_period << endl; 209} 210 211void Profiler::print(ostream& out) const 212{ 213 out << "[Profiler]"; 214} 215 216void Profiler::printStats(ostream& out, bool short_stats) 217{ 218 out << endl; 219 if (short_stats) { 220 out << "SHORT "; 221 } 222 out << "Profiler Stats" << endl; 223 out << "--------------" << endl; 224 225 time_t real_time_current = time(NULL); 226 double seconds = difftime(real_time_current, m_real_time_start_time); 227 double minutes = seconds/60.0; 228 double hours = minutes/60.0; 229 double days = hours/24.0; 230 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start; 231 232 if (!short_stats) { 233 out << "Elapsed_time_in_seconds: " << seconds << endl; 234 out << "Elapsed_time_in_minutes: " << minutes << endl; 235 out << "Elapsed_time_in_hours: " << hours << endl; 236 out << "Elapsed_time_in_days: " << days << endl; 237 out << endl; 238 } 239 240 // print the virtual runtimes as well 241 struct tms vtime; 242 times(&vtime); 243 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0; 244 minutes = seconds / 60.0; 245 hours = minutes / 60.0; 246 days = hours / 24.0; 247 out << "Virtual_time_in_seconds: " << seconds << endl; 248 out << "Virtual_time_in_minutes: " << minutes << endl; 249 out << "Virtual_time_in_hours: " << hours << endl; 250 out << "Virtual_time_in_days: " << hours << endl; 251 out << endl; 252 253 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl; 254 out << "Ruby_start_time: " << m_ruby_start << endl; 255 out << "Ruby_cycles: " << ruby_cycles << endl; 256 out << endl; 257 258 if (!short_stats) { 259 out << "mbytes_resident: " << process_memory_resident() << endl; 260 out << "mbytes_total: " << process_memory_total() << endl; 261 if (process_memory_total() > 0) { 262 out << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 263 } 264 out << endl; 265 266 if(m_num_BA_broadcasts + m_num_BA_unicasts != 0){ 267 out << endl; 268 out << "Broadcast_percent: " << (float)m_num_BA_broadcasts/(m_num_BA_broadcasts+m_num_BA_unicasts) << endl; 269 } 270 } 271 272 Vector<integer_t> perProcInstructionCount; 273 Vector<integer_t> perProcCycleCount; 274 Vector<double> perProcCPI; 275 Vector<double> perProcMissesPerInsn; 276 Vector<double> perProcInsnPerTrans; 277 Vector<double> perProcCyclesPerTrans; 278 Vector<double> perProcMissesPerTrans; 279
| 246 m_inst_profiler_ptr->printStats(*m_periodic_output_file_ptr); 247 } 248 249 //g_system_ptr->getNetwork()->printStats(*m_periodic_output_file_ptr); 250 g_eventQueue_ptr->scheduleEvent(this, m_stats_period); 251} 252 253void Profiler::setPeriodicStatsFile(const string& filename) 254{ 255 cout << "Recording periodic statistics to file '" << filename << "' every " 256 << m_stats_period << " Ruby cycles" << endl; 257 258 if (m_periodic_output_file_ptr != &cerr) { 259 delete m_periodic_output_file_ptr; 260 } 261 262 m_periodic_output_file_ptr = new ofstream(filename.c_str()); 263 g_eventQueue_ptr->scheduleEvent(this, 1); 264} 265 266void Profiler::setPeriodicStatsInterval(integer_t period) 267{ 268 cout << "Recording periodic statistics every " << m_stats_period << " Ruby cycles" << endl; 269 m_stats_period = period; 270 g_eventQueue_ptr->scheduleEvent(this, 1); 271} 272 273void Profiler::printConfig(ostream& out) const 274{ 275 out << endl; 276 out << "Profiler Configuration" << endl; 277 out << "----------------------" << endl; 278 out << "periodic_stats_period: " << m_stats_period << endl; 279} 280 281void Profiler::print(ostream& out) const 282{ 283 out << "[Profiler]"; 284} 285 286void Profiler::printStats(ostream& out, bool short_stats) 287{ 288 out << endl; 289 if (short_stats) { 290 out << "SHORT "; 291 } 292 out << "Profiler Stats" << endl; 293 out << "--------------" << endl; 294 295 time_t real_time_current = time(NULL); 296 double seconds = difftime(real_time_current, m_real_time_start_time); 297 double minutes = seconds/60.0; 298 double hours = minutes/60.0; 299 double days = hours/24.0; 300 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start; 301 302 if (!short_stats) { 303 out << "Elapsed_time_in_seconds: " << seconds << endl; 304 out << "Elapsed_time_in_minutes: " << minutes << endl; 305 out << "Elapsed_time_in_hours: " << hours << endl; 306 out << "Elapsed_time_in_days: " << days << endl; 307 out << endl; 308 } 309 310 // print the virtual runtimes as well 311 struct tms vtime; 312 times(&vtime); 313 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0; 314 minutes = seconds / 60.0; 315 hours = minutes / 60.0; 316 days = hours / 24.0; 317 out << "Virtual_time_in_seconds: " << seconds << endl; 318 out << "Virtual_time_in_minutes: " << minutes << endl; 319 out << "Virtual_time_in_hours: " << hours << endl; 320 out << "Virtual_time_in_days: " << hours << endl; 321 out << endl; 322 323 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl; 324 out << "Ruby_start_time: " << m_ruby_start << endl; 325 out << "Ruby_cycles: " << ruby_cycles << endl; 326 out << endl; 327 328 if (!short_stats) { 329 out << "mbytes_resident: " << process_memory_resident() << endl; 330 out << "mbytes_total: " << process_memory_total() << endl; 331 if (process_memory_total() > 0) { 332 out << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl; 333 } 334 out << endl; 335 336 if(m_num_BA_broadcasts + m_num_BA_unicasts != 0){ 337 out << endl; 338 out << "Broadcast_percent: " << (float)m_num_BA_broadcasts/(m_num_BA_broadcasts+m_num_BA_unicasts) << endl; 339 } 340 } 341 342 Vector<integer_t> perProcInstructionCount; 343 Vector<integer_t> perProcCycleCount; 344 Vector<double> perProcCPI; 345 Vector<double> perProcMissesPerInsn; 346 Vector<double> perProcInsnPerTrans; 347 Vector<double> perProcCyclesPerTrans; 348 Vector<double> perProcMissesPerTrans; 349
|
280 perProcInstructionCount.setSize(RubyConfig::numberOfProcessors()); 281 perProcCycleCount.setSize(RubyConfig::numberOfProcessors()); 282 perProcCPI.setSize(RubyConfig::numberOfProcessors()); 283 perProcMissesPerInsn.setSize(RubyConfig::numberOfProcessors());
| 350 perProcInstructionCount.setSize(RubySystem::getNumberOfSequencers()); 351 perProcCycleCount.setSize(RubySystem::getNumberOfSequencers()); 352 perProcCPI.setSize(RubySystem::getNumberOfSequencers()); 353 perProcMissesPerInsn.setSize(RubySystem::getNumberOfSequencers());
|
284
| 354
|
285 perProcInsnPerTrans.setSize(RubyConfig::numberOfProcessors()); 286 perProcCyclesPerTrans.setSize(RubyConfig::numberOfProcessors()); 287 perProcMissesPerTrans.setSize(RubyConfig::numberOfProcessors());
| 355 perProcInsnPerTrans.setSize(RubySystem::getNumberOfSequencers()); 356 perProcCyclesPerTrans.setSize(RubySystem::getNumberOfSequencers()); 357 perProcMissesPerTrans.setSize(RubySystem::getNumberOfSequencers());
|
288
| 358
|
289 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) { 290 perProcInstructionCount[i] = g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1; 291 perProcCycleCount[i] = g_system_ptr->getDriver()->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
| 359 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 360 perProcInstructionCount[i] = g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1; 361 perProcCycleCount[i] = g_system_ptr->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
|
292 // The +1 allows us to avoid division by zero 293 perProcCPI[i] = double(ruby_cycles)/perProcInstructionCount[i]; 294 perProcMissesPerInsn[i] = 1000.0 * (double(m_perProcTotalMisses[i]) / double(perProcInstructionCount[i])); 295 296 int trans = m_perProcEndTransaction[i]; 297 if (trans == 0) { 298 perProcInsnPerTrans[i] = 0; 299 perProcCyclesPerTrans[i] = 0; 300 perProcMissesPerTrans[i] = 0; 301 } else { 302 perProcInsnPerTrans[i] = perProcInstructionCount[i] / double(trans); 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(); 311 integer_t instruction_executed = perProcInstructionCount.sum();
| 362 // The +1 allows us to avoid division by zero 363 perProcCPI[i] = double(ruby_cycles)/perProcInstructionCount[i]; 364 perProcMissesPerInsn[i] = 1000.0 * (double(m_perProcTotalMisses[i]) / double(perProcInstructionCount[i])); 365 366 int trans = m_perProcEndTransaction[i]; 367 if (trans == 0) { 368 perProcInsnPerTrans[i] = 0; 369 perProcCyclesPerTrans[i] = 0; 370 perProcMissesPerTrans[i] = 0; 371 } else { 372 perProcInsnPerTrans[i] = perProcInstructionCount[i] / double(trans); 373 perProcCyclesPerTrans[i] = ruby_cycles / double(trans); 374 perProcMissesPerTrans[i] = m_perProcTotalMisses[i] / double(trans); 375 } 376 } 377 378 integer_t total_misses = m_perProcTotalMisses.sum(); 379 integer_t user_misses = m_perProcUserMisses.sum(); 380 integer_t supervisor_misses = m_perProcSupervisorMisses.sum(); 381 integer_t instruction_executed = perProcInstructionCount.sum();
|
312 integer_t cycles_executed = perProcCycleCount.sum();
| 382 integer_t simics_cycles_executed = perProcCycleCount.sum();
|
313 integer_t transactions_started = m_perProcStartTransaction.sum(); 314 integer_t transactions_ended = m_perProcEndTransaction.sum(); 315 316 double instructions_per_transaction = (transactions_ended != 0) ? double(instruction_executed) / double(transactions_ended) : 0;
| 383 integer_t transactions_started = m_perProcStartTransaction.sum(); 384 integer_t transactions_ended = m_perProcEndTransaction.sum(); 385 386 double instructions_per_transaction = (transactions_ended != 0) ? double(instruction_executed) / double(transactions_ended) : 0;
|
317 double cycles_per_transaction = (transactions_ended != 0) ? (RubyConfig::numberOfProcessors() * double(ruby_cycles)) / double(transactions_ended) : 0;
| 387 double cycles_per_transaction = (transactions_ended != 0) ? (RubySystem::getNumberOfSequencers() * double(ruby_cycles)) / double(transactions_ended) : 0;
|
318 double misses_per_transaction = (transactions_ended != 0) ? double(total_misses) / double(transactions_ended) : 0; 319 320 out << "Total_misses: " << total_misses << endl; 321 out << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 322 out << "user_misses: " << user_misses << " " << m_perProcUserMisses << endl; 323 out << "supervisor_misses: " << supervisor_misses << " " << m_perProcSupervisorMisses << endl; 324 out << endl; 325 out << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
| 388 double misses_per_transaction = (transactions_ended != 0) ? double(total_misses) / double(transactions_ended) : 0; 389 390 out << "Total_misses: " << total_misses << endl; 391 out << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl; 392 out << "user_misses: " << user_misses << " " << m_perProcUserMisses << endl; 393 out << "supervisor_misses: " << supervisor_misses << " " << m_perProcSupervisorMisses << endl; 394 out << endl; 395 out << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
|
326 out << "cycles_executed: " << cycles_executed << " " << perProcCycleCount << endl; 327 out << "cycles_per_instruction: " << (RubyConfig::numberOfProcessors()*double(ruby_cycles))/double(instruction_executed) << " " << perProcCPI << endl;
| 396 out << "ruby_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl; 397 out << "cycles_per_instruction: " << (RubySystem::getNumberOfSequencers()*double(ruby_cycles))/double(instruction_executed) << " " << perProcCPI << endl;
|
328 out << "misses_per_thousand_instructions: " << 1000.0 * (double(total_misses) / double(instruction_executed)) << " " << perProcMissesPerInsn << endl; 329 out << endl; 330 out << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 331 out << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl; 332 out << "instructions_per_transaction: " << instructions_per_transaction << " " << perProcInsnPerTrans << endl; 333 out << "cycles_per_transaction: " << cycles_per_transaction << " " << perProcCyclesPerTrans << endl; 334 out << "misses_per_transaction: " << misses_per_transaction << " " << perProcMissesPerTrans << endl; 335 336 out << endl; 337 338 m_L1D_cache_profiler_ptr->printStats(out); 339 m_L1I_cache_profiler_ptr->printStats(out); 340 m_L2_cache_profiler_ptr->printStats(out); 341 342 out << endl; 343
| 398 out << "misses_per_thousand_instructions: " << 1000.0 * (double(total_misses) / double(instruction_executed)) << " " << perProcMissesPerInsn << endl; 399 out << endl; 400 out << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl; 401 out << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl; 402 out << "instructions_per_transaction: " << instructions_per_transaction << " " << perProcInsnPerTrans << endl; 403 out << "cycles_per_transaction: " << cycles_per_transaction << " " << perProcCyclesPerTrans << endl; 404 out << "misses_per_transaction: " << misses_per_transaction << " " << perProcMissesPerTrans << endl; 405 406 out << endl; 407 408 m_L1D_cache_profiler_ptr->printStats(out); 409 m_L1I_cache_profiler_ptr->printStats(out); 410 m_L2_cache_profiler_ptr->printStats(out); 411 412 out << endl; 413
|
344 if (m_memReq || m_memRefresh) { // if there's a memory controller at all 345 long long int total_stalls = m_memInputQ + m_memBankQ + m_memWaitCycles; 346 double stallsPerReq = total_stalls * 1.0 / m_memReq; 347 out << "Memory control:" << endl; 348 out << " memory_total_requests: " << m_memReq << endl; // does not include refreshes 349 out << " memory_reads: " << m_memRead << endl; 350 out << " memory_writes: " << m_memWrite << endl; 351 out << " memory_refreshes: " << m_memRefresh << endl; 352 out << " memory_total_request_delays: " << total_stalls << endl; 353 out << " memory_delays_per_request: " << stallsPerReq << endl; 354 out << " memory_delays_in_input_queue: " << m_memInputQ << endl; 355 out << " memory_delays_behind_head_of_bank_queue: " << m_memBankQ << endl; 356 out << " memory_delays_stalled_at_head_of_bank_queue: " << m_memWaitCycles << endl; 357 // Note: The following "memory stalls" entries are a breakdown of the 358 // cycles which already showed up in m_memWaitCycles. The order is 359 // significant; it is the priority of attributing the cycles. 360 // For example, bank_busy is before arbitration because if the bank was 361 // busy, we didn't even check arbitration. 362 // Note: "not old enough" means that since we grouped waiting heads-of-queues 363 // into batches to avoid starvation, a request in a newer batch 364 // didn't try to arbitrate yet because there are older requests waiting. 365 out << " memory_stalls_for_bank_busy: " << m_memBankBusy << endl; 366 out << " memory_stalls_for_random_busy: " << m_memRandBusy << endl; 367 out << " memory_stalls_for_anti_starvation: " << m_memNotOld << endl; 368 out << " memory_stalls_for_arbitration: " << m_memArbWait << endl; 369 out << " memory_stalls_for_bus: " << m_memBusBusy << endl; 370 out << " memory_stalls_for_tfaw: " << m_memTfawBusy << endl; 371 out << " memory_stalls_for_read_write_turnaround: " << m_memReadWriteBusy << endl; 372 out << " memory_stalls_for_read_read_turnaround: " << m_memDataBusBusy << endl; 373 out << " accesses_per_bank: "; 374 for (int bank=0; bank < m_memBankCount.size(); bank++) { 375 out << m_memBankCount[bank] << " "; 376 //if ((bank % 8) == 7) out << " " << endl;
| 414 vector<string>::iterator it; 415 416 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 417 long long int m_memReq = m_memory_control_profilers[(*it).c_str()] -> m_memReq; 418 long long int m_memRefresh = m_memory_control_profilers[(*it).c_str()] -> m_memRefresh; 419 long long int m_memInputQ = m_memory_control_profilers[(*it).c_str()] -> m_memInputQ; 420 long long int m_memBankQ = m_memory_control_profilers[(*it).c_str()] -> m_memBankQ; 421 long long int m_memWaitCycles = m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles; 422 long long int m_memRead = m_memory_control_profilers[(*it).c_str()] -> m_memRead; 423 long long int m_memWrite = m_memory_control_profilers[(*it).c_str()] -> m_memWrite; 424 long long int m_memBankBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy; 425 long long int m_memRandBusy = m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy; 426 long long int m_memNotOld = m_memory_control_profilers[(*it).c_str()] -> m_memNotOld; 427 long long int m_memArbWait = m_memory_control_profilers[(*it).c_str()] -> m_memArbWait; 428 long long int m_memBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy; 429 long long int m_memTfawBusy = m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy; 430 long long int m_memReadWriteBusy = m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy; 431 long long int m_memDataBusBusy = m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy; 432 Vector<long long int> m_memBankCount = m_memory_control_profilers[(*it).c_str()] -> m_memBankCount; 433 434 if (m_memReq || m_memRefresh) { // if there's a memory controller at all 435 long long int total_stalls = m_memInputQ + m_memBankQ + m_memWaitCycles; 436 double stallsPerReq = total_stalls * 1.0 / m_memReq; 437 out << "Memory control:" << endl; 438 out << " memory_total_requests: " << m_memReq << endl; // does not include refreshes 439 out << " memory_reads: " << m_memRead << endl; 440 out << " memory_writes: " << m_memWrite << endl; 441 out << " memory_refreshes: " << m_memRefresh << endl; 442 out << " memory_total_request_delays: " << total_stalls << endl; 443 out << " memory_delays_per_request: " << stallsPerReq << endl; 444 out << " memory_delays_in_input_queue: " << m_memInputQ << endl; 445 out << " memory_delays_behind_head_of_bank_queue: " << m_memBankQ << endl; 446 out << " memory_delays_stalled_at_head_of_bank_queue: " << m_memWaitCycles << endl; 447 // Note: The following "memory stalls" entries are a breakdown of the 448 // cycles which already showed up in m_memWaitCycles. The order is 449 // significant; it is the priority of attributing the cycles. 450 // For example, bank_busy is before arbitration because if the bank was 451 // busy, we didn't even check arbitration. 452 // Note: "not old enough" means that since we grouped waiting heads-of-queues 453 // into batches to avoid starvation, a request in a newer batch 454 // didn't try to arbitrate yet because there are older requests waiting. 455 out << " memory_stalls_for_bank_busy: " << m_memBankBusy << endl; 456 out << " memory_stalls_for_random_busy: " << m_memRandBusy << endl; 457 out << " memory_stalls_for_anti_starvation: " << m_memNotOld << endl; 458 out << " memory_stalls_for_arbitration: " << m_memArbWait << endl; 459 out << " memory_stalls_for_bus: " << m_memBusBusy << endl; 460 out << " memory_stalls_for_tfaw: " << m_memTfawBusy << endl; 461 out << " memory_stalls_for_read_write_turnaround: " << m_memReadWriteBusy << endl; 462 out << " memory_stalls_for_read_read_turnaround: " << m_memDataBusBusy << endl; 463 out << " accesses_per_bank: "; 464 for (int bank=0; bank < m_memBankCount.size(); bank++) { 465 out << m_memBankCount[bank] << " "; 466 //if ((bank % 8) == 7) out << " " << endl; 467 } 468 out << endl; 469 out << endl;
|
377 }
| 470 }
|
378 out << endl; 379 out << endl;
| |
380 }
| 471 }
|
381
| |
382 if (!short_stats) { 383 out << "Busy Controller Counts:" << endl; 384 for(int i=0; i < MachineType_NUM; i++) { 385 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 386 MachineID machID; 387 machID.type = (MachineType)i; 388 machID.num = j; 389 out << machID << ":" << m_busyControllerCount[i][j] << " "; 390 if ((j+1)%8 == 0) { 391 out << endl; 392 } 393 } 394 out << endl; 395 } 396 out << endl; 397 398 out << "Busy Bank Count:" << m_busyBankCount << endl; 399 out << endl; 400 401 out << "L1TBE_usage: " << m_L1tbeProfile << endl; 402 out << "L2TBE_usage: " << m_L2tbeProfile << endl; 403 out << "StopTable_usage: " << m_stopTableProfile << endl; 404 out << "sequencer_requests_outstanding: " << m_sequencer_requests << endl; 405 out << "store_buffer_size: " << m_store_buffer_size << endl; 406 out << "unique_blocks_in_store_buffer: " << m_store_buffer_blocks << endl; 407 out << endl; 408 } 409 410 if (!short_stats) { 411 out << "All Non-Zero Cycle Demand Cache Accesses" << endl; 412 out << "----------------------------------------" << endl; 413 out << "miss_latency: " << m_allMissLatencyHistogram << endl; 414 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 415 if (m_missLatencyHistograms[i].size() > 0) {
| 472 if (!short_stats) { 473 out << "Busy Controller Counts:" << endl; 474 for(int i=0; i < MachineType_NUM; i++) { 475 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 476 MachineID machID; 477 machID.type = (MachineType)i; 478 machID.num = j; 479 out << machID << ":" << m_busyControllerCount[i][j] << " "; 480 if ((j+1)%8 == 0) { 481 out << endl; 482 } 483 } 484 out << endl; 485 } 486 out << endl; 487 488 out << "Busy Bank Count:" << m_busyBankCount << endl; 489 out << endl; 490 491 out << "L1TBE_usage: " << m_L1tbeProfile << endl; 492 out << "L2TBE_usage: " << m_L2tbeProfile << endl; 493 out << "StopTable_usage: " << m_stopTableProfile << endl; 494 out << "sequencer_requests_outstanding: " << m_sequencer_requests << endl; 495 out << "store_buffer_size: " << m_store_buffer_size << endl; 496 out << "unique_blocks_in_store_buffer: " << m_store_buffer_blocks << endl; 497 out << endl; 498 } 499 500 if (!short_stats) { 501 out << "All Non-Zero Cycle Demand Cache Accesses" << endl; 502 out << "----------------------------------------" << endl; 503 out << "miss_latency: " << m_allMissLatencyHistogram << endl; 504 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 505 if (m_missLatencyHistograms[i].size() > 0) {
|
416 out << "miss_latency_" << CacheRequestType(i) << ": " << m_missLatencyHistograms[i] << endl;
| 506 out << "miss_latency_" << RubyRequestType(i) << ": " << m_missLatencyHistograms[i] << endl;
|
417 } 418 } 419 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 420 if (m_machLatencyHistograms[i].size() > 0) { 421 out << "miss_latency_" << GenericMachineType(i) << ": " << m_machLatencyHistograms[i] << endl; 422 } 423 } 424 out << "miss_latency_L2Miss: " << m_L2MissLatencyHistogram << endl; 425 426 out << endl; 427 428 out << "All Non-Zero Cycle SW Prefetch Requests" << endl; 429 out << "------------------------------------" << endl; 430 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl; 431 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 432 if (m_SWPrefetchLatencyHistograms[i].size() > 0) { 433 out << "prefetch_latency_" << CacheRequestType(i) << ": " << m_SWPrefetchLatencyHistograms[i] << endl; 434 } 435 } 436 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 437 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) { 438 out << "prefetch_latency_" << GenericMachineType(i) << ": " << m_SWPrefetchMachLatencyHistograms[i] << endl; 439 } 440 } 441 out << "prefetch_latency_L2Miss:" << m_SWPrefetchL2MissLatencyHistogram << endl; 442 443 out << "multicast_retries: " << m_multicast_retry_histogram << endl; 444 out << "gets_mask_prediction_count: " << m_gets_mask_prediction << endl; 445 out << "getx_mask_prediction_count: " << m_getx_mask_prediction << endl; 446 out << "explicit_training_mask: " << m_explicit_training_mask << endl; 447 out << endl; 448 449 if (m_all_sharing_histogram.size() > 0) { 450 out << "all_sharing: " << m_all_sharing_histogram << endl; 451 out << "read_sharing: " << m_read_sharing_histogram << endl; 452 out << "write_sharing: " << m_write_sharing_histogram << endl; 453 454 out << "all_sharing_percent: "; m_all_sharing_histogram.printPercent(out); out << endl; 455 out << "read_sharing_percent: "; m_read_sharing_histogram.printPercent(out); out << endl; 456 out << "write_sharing_percent: "; m_write_sharing_histogram.printPercent(out); out << endl; 457 458 int64 total_miss = m_cache_to_cache + m_memory_to_cache; 459 out << "all_misses: " << total_miss << endl; 460 out << "cache_to_cache_misses: " << m_cache_to_cache << endl; 461 out << "memory_to_cache_misses: " << m_memory_to_cache << endl; 462 out << "cache_to_cache_percent: " << 100.0 * (double(m_cache_to_cache) / double(total_miss)) << endl; 463 out << "memory_to_cache_percent: " << 100.0 * (double(m_memory_to_cache) / double(total_miss)) << endl; 464 out << endl; 465 } 466 467 if (m_conflicting_histogram.size() > 0) { 468 out << "conflicting_histogram: " << m_conflicting_histogram << endl; 469 out << "conflicting_histogram_percent: "; m_conflicting_histogram.printPercent(out); out << endl; 470 out << endl; 471 } 472 473 if (m_outstanding_requests.size() > 0) { 474 out << "outstanding_requests: "; m_outstanding_requests.printPercent(out); out << endl; 475 if (m_outstanding_persistent_requests.size() > 0) { 476 out << "outstanding_persistent_requests: "; m_outstanding_persistent_requests.printPercent(out); out << endl; 477 } 478 out << endl; 479 } 480 } 481 482 if (!short_stats) { 483 out << "Request vs. RubySystem State Profile" << endl; 484 out << "--------------------------------" << endl; 485 out << endl; 486 487 Vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys(); 488 requestProfileKeys.sortVector(); 489 490 for(int i=0; i<requestProfileKeys.size(); i++) { 491 int temp_int = m_requestProfileMap_ptr->lookup(requestProfileKeys[i]); 492 double percent = (100.0*double(temp_int))/double(m_requests); 493 while (requestProfileKeys[i] != "") { 494 out << setw(10) << string_split(requestProfileKeys[i], ':'); 495 } 496 out << setw(11) << temp_int; 497 out << setw(14) << percent << endl; 498 } 499 out << endl; 500 501 out << "filter_action: " << m_filter_action_histogram << endl; 502
| 507 } 508 } 509 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 510 if (m_machLatencyHistograms[i].size() > 0) { 511 out << "miss_latency_" << GenericMachineType(i) << ": " << m_machLatencyHistograms[i] << endl; 512 } 513 } 514 out << "miss_latency_L2Miss: " << m_L2MissLatencyHistogram << endl; 515 516 out << endl; 517 518 out << "All Non-Zero Cycle SW Prefetch Requests" << endl; 519 out << "------------------------------------" << endl; 520 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl; 521 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 522 if (m_SWPrefetchLatencyHistograms[i].size() > 0) { 523 out << "prefetch_latency_" << CacheRequestType(i) << ": " << m_SWPrefetchLatencyHistograms[i] << endl; 524 } 525 } 526 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 527 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) { 528 out << "prefetch_latency_" << GenericMachineType(i) << ": " << m_SWPrefetchMachLatencyHistograms[i] << endl; 529 } 530 } 531 out << "prefetch_latency_L2Miss:" << m_SWPrefetchL2MissLatencyHistogram << endl; 532 533 out << "multicast_retries: " << m_multicast_retry_histogram << endl; 534 out << "gets_mask_prediction_count: " << m_gets_mask_prediction << endl; 535 out << "getx_mask_prediction_count: " << m_getx_mask_prediction << endl; 536 out << "explicit_training_mask: " << m_explicit_training_mask << endl; 537 out << endl; 538 539 if (m_all_sharing_histogram.size() > 0) { 540 out << "all_sharing: " << m_all_sharing_histogram << endl; 541 out << "read_sharing: " << m_read_sharing_histogram << endl; 542 out << "write_sharing: " << m_write_sharing_histogram << endl; 543 544 out << "all_sharing_percent: "; m_all_sharing_histogram.printPercent(out); out << endl; 545 out << "read_sharing_percent: "; m_read_sharing_histogram.printPercent(out); out << endl; 546 out << "write_sharing_percent: "; m_write_sharing_histogram.printPercent(out); out << endl; 547 548 int64 total_miss = m_cache_to_cache + m_memory_to_cache; 549 out << "all_misses: " << total_miss << endl; 550 out << "cache_to_cache_misses: " << m_cache_to_cache << endl; 551 out << "memory_to_cache_misses: " << m_memory_to_cache << endl; 552 out << "cache_to_cache_percent: " << 100.0 * (double(m_cache_to_cache) / double(total_miss)) << endl; 553 out << "memory_to_cache_percent: " << 100.0 * (double(m_memory_to_cache) / double(total_miss)) << endl; 554 out << endl; 555 } 556 557 if (m_conflicting_histogram.size() > 0) { 558 out << "conflicting_histogram: " << m_conflicting_histogram << endl; 559 out << "conflicting_histogram_percent: "; m_conflicting_histogram.printPercent(out); out << endl; 560 out << endl; 561 } 562 563 if (m_outstanding_requests.size() > 0) { 564 out << "outstanding_requests: "; m_outstanding_requests.printPercent(out); out << endl; 565 if (m_outstanding_persistent_requests.size() > 0) { 566 out << "outstanding_persistent_requests: "; m_outstanding_persistent_requests.printPercent(out); out << endl; 567 } 568 out << endl; 569 } 570 } 571 572 if (!short_stats) { 573 out << "Request vs. RubySystem State Profile" << endl; 574 out << "--------------------------------" << endl; 575 out << endl; 576 577 Vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys(); 578 requestProfileKeys.sortVector(); 579 580 for(int i=0; i<requestProfileKeys.size(); i++) { 581 int temp_int = m_requestProfileMap_ptr->lookup(requestProfileKeys[i]); 582 double percent = (100.0*double(temp_int))/double(m_requests); 583 while (requestProfileKeys[i] != "") { 584 out << setw(10) << string_split(requestProfileKeys[i], ':'); 585 } 586 out << setw(11) << temp_int; 587 out << setw(14) << percent << endl; 588 } 589 out << endl; 590 591 out << "filter_action: " << m_filter_action_histogram << endl; 592
|
503 if (!PROFILE_ALL_INSTRUCTIONS) {
| 593 if (!m_all_instructions) {
|
504 m_address_profiler_ptr->printStats(out); 505 } 506
| 594 m_address_profiler_ptr->printStats(out); 595 } 596
|
507 if (PROFILE_ALL_INSTRUCTIONS) {
| 597 if (m_all_instructions) {
|
508 m_inst_profiler_ptr->printStats(out); 509 } 510 511 out << endl; 512 out << "Message Delayed Cycles" << endl; 513 out << "----------------------" << endl; 514 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl; 515 out << "Total_nonPF_delay_cycles: " << m_delayedCyclesNonPFHistogram << endl; 516 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) { 517 out << " virtual_network_" << i << "_delay_cycles: " << m_delayedCyclesVCHistograms[i] << endl; 518 } 519 520 printResourceUsage(out); 521 } 522 523} 524 525void Profiler::printResourceUsage(ostream& out) const 526{ 527 out << endl; 528 out << "Resource Usage" << endl; 529 out << "--------------" << endl; 530 531 integer_t pagesize = getpagesize(); // page size in bytes 532 out << "page_size: " << pagesize << endl; 533 534 rusage usage; 535 getrusage (RUSAGE_SELF, &usage); 536 537 out << "user_time: " << usage.ru_utime.tv_sec << endl; 538 out << "system_time: " << usage.ru_stime.tv_sec << endl; 539 out << "page_reclaims: " << usage.ru_minflt << endl; 540 out << "page_faults: " << usage.ru_majflt << endl; 541 out << "swaps: " << usage.ru_nswap << endl; 542 out << "block_inputs: " << usage.ru_inblock << endl; 543 out << "block_outputs: " << usage.ru_oublock << endl; 544} 545 546void Profiler::clearStats() 547{ 548 m_num_BA_unicasts = 0; 549 m_num_BA_broadcasts = 0; 550 551 m_ruby_start = g_eventQueue_ptr->getTime(); 552
| 598 m_inst_profiler_ptr->printStats(out); 599 } 600 601 out << endl; 602 out << "Message Delayed Cycles" << endl; 603 out << "----------------------" << endl; 604 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl; 605 out << "Total_nonPF_delay_cycles: " << m_delayedCyclesNonPFHistogram << endl; 606 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) { 607 out << " virtual_network_" << i << "_delay_cycles: " << m_delayedCyclesVCHistograms[i] << endl; 608 } 609 610 printResourceUsage(out); 611 } 612 613} 614 615void Profiler::printResourceUsage(ostream& out) const 616{ 617 out << endl; 618 out << "Resource Usage" << endl; 619 out << "--------------" << endl; 620 621 integer_t pagesize = getpagesize(); // page size in bytes 622 out << "page_size: " << pagesize << endl; 623 624 rusage usage; 625 getrusage (RUSAGE_SELF, &usage); 626 627 out << "user_time: " << usage.ru_utime.tv_sec << endl; 628 out << "system_time: " << usage.ru_stime.tv_sec << endl; 629 out << "page_reclaims: " << usage.ru_minflt << endl; 630 out << "page_faults: " << usage.ru_majflt << endl; 631 out << "swaps: " << usage.ru_nswap << endl; 632 out << "block_inputs: " << usage.ru_inblock << endl; 633 out << "block_outputs: " << usage.ru_oublock << endl; 634} 635 636void Profiler::clearStats() 637{ 638 m_num_BA_unicasts = 0; 639 m_num_BA_broadcasts = 0; 640 641 m_ruby_start = g_eventQueue_ptr->getTime(); 642
|
553 m_instructions_executed_at_start.setSize(RubyConfig::numberOfProcessors()); 554 m_cycles_executed_at_start.setSize(RubyConfig::numberOfProcessors()); 555 for (int i=0; i < RubyConfig::numberOfProcessors(); i++) {
| 643 m_instructions_executed_at_start.setSize(RubySystem::getNumberOfSequencers()); 644 m_cycles_executed_at_start.setSize(RubySystem::getNumberOfSequencers()); 645 for (int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
|
556 if (g_system_ptr == NULL) { 557 m_instructions_executed_at_start[i] = 0; 558 m_cycles_executed_at_start[i] = 0; 559 } else {
| 646 if (g_system_ptr == NULL) { 647 m_instructions_executed_at_start[i] = 0; 648 m_cycles_executed_at_start[i] = 0; 649 } else {
|
560 m_instructions_executed_at_start[i] = g_system_ptr->getDriver()->getInstructionCount(i); 561 m_cycles_executed_at_start[i] = g_system_ptr->getDriver()->getCycleCount(i);
| 650 m_instructions_executed_at_start[i] = g_system_ptr->getInstructionCount(i); 651 m_cycles_executed_at_start[i] = g_system_ptr->getCycleCount(i);
|
562 } 563 } 564
| 652 } 653 } 654
|
565 m_perProcTotalMisses.setSize(RubyConfig::numberOfProcessors()); 566 m_perProcUserMisses.setSize(RubyConfig::numberOfProcessors()); 567 m_perProcSupervisorMisses.setSize(RubyConfig::numberOfProcessors()); 568 m_perProcStartTransaction.setSize(RubyConfig::numberOfProcessors()); 569 m_perProcEndTransaction.setSize(RubyConfig::numberOfProcessors());
| 655 m_perProcTotalMisses.setSize(RubySystem::getNumberOfSequencers()); 656 m_perProcUserMisses.setSize(RubySystem::getNumberOfSequencers()); 657 m_perProcSupervisorMisses.setSize(RubySystem::getNumberOfSequencers()); 658 m_perProcStartTransaction.setSize(RubySystem::getNumberOfSequencers()); 659 m_perProcEndTransaction.setSize(RubySystem::getNumberOfSequencers());
|
570
| 660
|
571 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) {
| 661 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) {
|
572 m_perProcTotalMisses[i] = 0; 573 m_perProcUserMisses[i] = 0; 574 m_perProcSupervisorMisses[i] = 0; 575 m_perProcStartTransaction[i] = 0; 576 m_perProcEndTransaction[i] = 0; 577 } 578 579 m_busyControllerCount.setSize(MachineType_NUM); // all machines 580 for(int i=0; i < MachineType_NUM; i++) { 581 m_busyControllerCount[i].setSize(MachineType_base_count((MachineType)i)); 582 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 583 m_busyControllerCount[i][j] = 0; 584 } 585 } 586 m_busyBankCount = 0; 587 588 m_delayedCyclesHistogram.clear(); 589 m_delayedCyclesNonPFHistogram.clear();
| 662 m_perProcTotalMisses[i] = 0; 663 m_perProcUserMisses[i] = 0; 664 m_perProcSupervisorMisses[i] = 0; 665 m_perProcStartTransaction[i] = 0; 666 m_perProcEndTransaction[i] = 0; 667 } 668 669 m_busyControllerCount.setSize(MachineType_NUM); // all machines 670 for(int i=0; i < MachineType_NUM; i++) { 671 m_busyControllerCount[i].setSize(MachineType_base_count((MachineType)i)); 672 for(int j=0; j < MachineType_base_count((MachineType)i); j++) { 673 m_busyControllerCount[i][j] = 0; 674 } 675 } 676 m_busyBankCount = 0; 677 678 m_delayedCyclesHistogram.clear(); 679 m_delayedCyclesNonPFHistogram.clear();
|
590 m_delayedCyclesVCHistograms.setSize(NUMBER_OF_VIRTUAL_NETWORKS); 591 for (int i = 0; i < NUMBER_OF_VIRTUAL_NETWORKS; i++) {
| 680 m_delayedCyclesVCHistograms.setSize(RubySystem::getNetwork()->getNumberOfVirtualNetworks()); 681 for (int i = 0; i < RubySystem::getNetwork()->getNumberOfVirtualNetworks(); i++) {
|
592 m_delayedCyclesVCHistograms[i].clear(); 593 } 594 595 m_gets_mask_prediction.clear(); 596 m_getx_mask_prediction.clear(); 597 m_explicit_training_mask.clear(); 598 599 m_missLatencyHistograms.setSize(CacheRequestType_NUM); 600 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 601 m_missLatencyHistograms[i].clear(200); 602 } 603 m_machLatencyHistograms.setSize(GenericMachineType_NUM+1); 604 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 605 m_machLatencyHistograms[i].clear(200); 606 } 607 m_allMissLatencyHistogram.clear(200); 608 m_L2MissLatencyHistogram.clear(200); 609 610 m_SWPrefetchLatencyHistograms.setSize(CacheRequestType_NUM); 611 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 612 m_SWPrefetchLatencyHistograms[i].clear(200); 613 } 614 m_SWPrefetchMachLatencyHistograms.setSize(GenericMachineType_NUM+1); 615 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 616 m_SWPrefetchMachLatencyHistograms[i].clear(200); 617 } 618 m_allSWPrefetchLatencyHistogram.clear(200); 619 m_SWPrefetchL2MissLatencyHistogram.clear(200); 620 621 m_multicast_retry_histogram.clear(); 622 623 m_L1tbeProfile.clear(); 624 m_L2tbeProfile.clear(); 625 m_stopTableProfile.clear(); 626 m_filter_action_histogram.clear(); 627 628 m_sequencer_requests.clear(); 629 m_store_buffer_size.clear(); 630 m_store_buffer_blocks.clear(); 631 m_read_sharing_histogram.clear(); 632 m_write_sharing_histogram.clear(); 633 m_all_sharing_histogram.clear(); 634 m_cache_to_cache = 0; 635 m_memory_to_cache = 0; 636 637 m_predictions = 0; 638 m_predictionOpportunities = 0; 639 m_goodPredictions = 0; 640 641 // clear HashMaps 642 m_requestProfileMap_ptr->clear(); 643 644 // count requests profiled 645 m_requests = 0; 646 647 // Conflicting requests 648 m_conflicting_map_ptr->clear(); 649 m_conflicting_histogram.clear(); 650 651 m_outstanding_requests.clear(); 652 m_outstanding_persistent_requests.clear(); 653 654 m_L1D_cache_profiler_ptr->clearStats(); 655 m_L1I_cache_profiler_ptr->clearStats(); 656 m_L2_cache_profiler_ptr->clearStats(); 657 658 // for MemoryControl:
| 682 m_delayedCyclesVCHistograms[i].clear(); 683 } 684 685 m_gets_mask_prediction.clear(); 686 m_getx_mask_prediction.clear(); 687 m_explicit_training_mask.clear(); 688 689 m_missLatencyHistograms.setSize(CacheRequestType_NUM); 690 for(int i=0; i<m_missLatencyHistograms.size(); i++) { 691 m_missLatencyHistograms[i].clear(200); 692 } 693 m_machLatencyHistograms.setSize(GenericMachineType_NUM+1); 694 for(int i=0; i<m_machLatencyHistograms.size(); i++) { 695 m_machLatencyHistograms[i].clear(200); 696 } 697 m_allMissLatencyHistogram.clear(200); 698 m_L2MissLatencyHistogram.clear(200); 699 700 m_SWPrefetchLatencyHistograms.setSize(CacheRequestType_NUM); 701 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) { 702 m_SWPrefetchLatencyHistograms[i].clear(200); 703 } 704 m_SWPrefetchMachLatencyHistograms.setSize(GenericMachineType_NUM+1); 705 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) { 706 m_SWPrefetchMachLatencyHistograms[i].clear(200); 707 } 708 m_allSWPrefetchLatencyHistogram.clear(200); 709 m_SWPrefetchL2MissLatencyHistogram.clear(200); 710 711 m_multicast_retry_histogram.clear(); 712 713 m_L1tbeProfile.clear(); 714 m_L2tbeProfile.clear(); 715 m_stopTableProfile.clear(); 716 m_filter_action_histogram.clear(); 717 718 m_sequencer_requests.clear(); 719 m_store_buffer_size.clear(); 720 m_store_buffer_blocks.clear(); 721 m_read_sharing_histogram.clear(); 722 m_write_sharing_histogram.clear(); 723 m_all_sharing_histogram.clear(); 724 m_cache_to_cache = 0; 725 m_memory_to_cache = 0; 726 727 m_predictions = 0; 728 m_predictionOpportunities = 0; 729 m_goodPredictions = 0; 730 731 // clear HashMaps 732 m_requestProfileMap_ptr->clear(); 733 734 // count requests profiled 735 m_requests = 0; 736 737 // Conflicting requests 738 m_conflicting_map_ptr->clear(); 739 m_conflicting_histogram.clear(); 740 741 m_outstanding_requests.clear(); 742 m_outstanding_persistent_requests.clear(); 743 744 m_L1D_cache_profiler_ptr->clearStats(); 745 m_L1I_cache_profiler_ptr->clearStats(); 746 m_L2_cache_profiler_ptr->clearStats(); 747 748 // for MemoryControl:
|
| 749/*
|
659 m_memReq = 0; 660 m_memBankBusy = 0; 661 m_memBusBusy = 0; 662 m_memTfawBusy = 0; 663 m_memReadWriteBusy = 0; 664 m_memDataBusBusy = 0; 665 m_memRefresh = 0; 666 m_memRead = 0; 667 m_memWrite = 0; 668 m_memWaitCycles = 0; 669 m_memInputQ = 0; 670 m_memBankQ = 0; 671 m_memArbWait = 0; 672 m_memRandBusy = 0; 673 m_memNotOld = 0; 674 675 for (int bank=0; bank < m_memBankCount.size(); bank++) { 676 m_memBankCount[bank] = 0; 677 }
| 750 m_memReq = 0; 751 m_memBankBusy = 0; 752 m_memBusBusy = 0; 753 m_memTfawBusy = 0; 754 m_memReadWriteBusy = 0; 755 m_memDataBusBusy = 0; 756 m_memRefresh = 0; 757 m_memRead = 0; 758 m_memWrite = 0; 759 m_memWaitCycles = 0; 760 m_memInputQ = 0; 761 m_memBankQ = 0; 762 m_memArbWait = 0; 763 m_memRandBusy = 0; 764 m_memNotOld = 0; 765 766 for (int bank=0; bank < m_memBankCount.size(); bank++) { 767 m_memBankCount[bank] = 0; 768 }
|
| 769*/ 770//added by SS 771 vector<string>::iterator it;
|
678
| 772
|
| 773 for ( it=m_memory_control_names.begin() ; it < m_memory_control_names.end(); it++ ){ 774 m_memory_control_profilers[(*it).c_str()] -> m_memReq = 0; 775 m_memory_control_profilers[(*it).c_str()] -> m_memBankBusy = 0; 776 m_memory_control_profilers[(*it).c_str()] -> m_memBusBusy = 0; 777 m_memory_control_profilers[(*it).c_str()] -> m_memTfawBusy = 0; 778 m_memory_control_profilers[(*it).c_str()] -> m_memReadWriteBusy = 0; 779 m_memory_control_profilers[(*it).c_str()] -> m_memDataBusBusy = 0; 780 m_memory_control_profilers[(*it).c_str()] -> m_memRefresh = 0; 781 m_memory_control_profilers[(*it).c_str()] -> m_memRead = 0; 782 m_memory_control_profilers[(*it).c_str()] -> m_memWrite = 0; 783 m_memory_control_profilers[(*it).c_str()] -> m_memWaitCycles = 0; 784 m_memory_control_profilers[(*it).c_str()] -> m_memInputQ = 0; 785 m_memory_control_profilers[(*it).c_str()] -> m_memBankQ = 0; 786 m_memory_control_profilers[(*it).c_str()] -> m_memArbWait = 0; 787 m_memory_control_profilers[(*it).c_str()] -> m_memRandBusy = 0; 788 m_memory_control_profilers[(*it).c_str()] -> m_memNotOld = 0; 789 790 for (int bank=0; bank < m_memory_control_profilers[(*it).c_str()] -> m_memBankCount.size(); bank++) { 791 m_memory_control_profilers[(*it).c_str()] -> m_memBankCount[bank] = 0; 792 } 793 }
|
679 // Flush the prefetches through the system - used so that there are no outstanding requests after stats are cleared 680 //g_eventQueue_ptr->triggerAllEvents(); 681 682 // update the start time 683 m_ruby_start = g_eventQueue_ptr->getTime(); 684} 685 686void Profiler::addPrimaryStatSample(const CacheMsg& msg, NodeID id) 687{ 688 if (Protocol::m_TwoLevelCache) { 689 if (msg.getType() == CacheRequestType_IFETCH) { 690 addL1IStatSample(msg, id); 691 } else { 692 addL1DStatSample(msg, id); 693 } 694 // profile the address after an L1 miss (outside of the processor for CMP) 695 if (Protocol::m_CMP) { 696 addAddressTraceSample(msg, id); 697 } 698 } else { 699 addL2StatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 700 msg.getAccessMode(), msg.getSize(), msg.getPrefetch(), id); 701 addAddressTraceSample(msg, id); 702 } 703} 704 705void Profiler::profileConflictingRequests(const Address& addr) 706{ 707 assert(addr == line_address(addr)); 708 Time last_time = m_ruby_start; 709 if (m_conflicting_map_ptr->exist(addr)) {
| 794 // Flush the prefetches through the system - used so that there are no outstanding requests after stats are cleared 795 //g_eventQueue_ptr->triggerAllEvents(); 796 797 // update the start time 798 m_ruby_start = g_eventQueue_ptr->getTime(); 799} 800 801void Profiler::addPrimaryStatSample(const CacheMsg& msg, NodeID id) 802{ 803 if (Protocol::m_TwoLevelCache) { 804 if (msg.getType() == CacheRequestType_IFETCH) { 805 addL1IStatSample(msg, id); 806 } else { 807 addL1DStatSample(msg, id); 808 } 809 // profile the address after an L1 miss (outside of the processor for CMP) 810 if (Protocol::m_CMP) { 811 addAddressTraceSample(msg, id); 812 } 813 } else { 814 addL2StatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 815 msg.getAccessMode(), msg.getSize(), msg.getPrefetch(), id); 816 addAddressTraceSample(msg, id); 817 } 818} 819 820void Profiler::profileConflictingRequests(const Address& addr) 821{ 822 assert(addr == line_address(addr)); 823 Time last_time = m_ruby_start; 824 if (m_conflicting_map_ptr->exist(addr)) {
|
710 last_time = m_conflicting_map_ptr->lookup(addr);
| 825 Time last_time = m_conflicting_map_ptr->lookup(addr);
|
711 } 712 Time current_time = g_eventQueue_ptr->getTime(); 713 assert (current_time - last_time > 0); 714 m_conflicting_histogram.add(current_time - last_time); 715 m_conflicting_map_ptr->add(addr, current_time); 716} 717 718void Profiler::addSecondaryStatSample(CacheRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 719{ 720 addSecondaryStatSample(CacheRequestType_to_GenericRequestType(requestType), type, msgSize, pfBit, id); 721} 722 723void Profiler::addSecondaryStatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 724{ 725 addL2StatSample(requestType, type, msgSize, pfBit, id); 726} 727 728void Profiler::addL2StatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 729{ 730 m_perProcTotalMisses[id]++; 731 if (type == AccessModeType_SupervisorMode) { 732 m_perProcSupervisorMisses[id]++; 733 } else { 734 m_perProcUserMisses[id]++; 735 } 736 m_L2_cache_profiler_ptr->addStatSample(requestType, type, msgSize, pfBit); 737} 738 739void Profiler::addL1DStatSample(const CacheMsg& msg, NodeID id) 740{ 741 m_L1D_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 742 msg.getAccessMode(), msg.getSize(), msg.getPrefetch()); 743} 744 745void Profiler::addL1IStatSample(const CacheMsg& msg, NodeID id) 746{ 747 m_L1I_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 748 msg.getAccessMode(), msg.getSize(), msg.getPrefetch()); 749} 750 751void Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id) 752{ 753 if (msg.getType() != CacheRequestType_IFETCH) { 754 755 // Note: The following line should be commented out if you want to 756 // use the special profiling that is part of the GS320 protocol 757
| 826 } 827 Time current_time = g_eventQueue_ptr->getTime(); 828 assert (current_time - last_time > 0); 829 m_conflicting_histogram.add(current_time - last_time); 830 m_conflicting_map_ptr->add(addr, current_time); 831} 832 833void Profiler::addSecondaryStatSample(CacheRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 834{ 835 addSecondaryStatSample(CacheRequestType_to_GenericRequestType(requestType), type, msgSize, pfBit, id); 836} 837 838void Profiler::addSecondaryStatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 839{ 840 addL2StatSample(requestType, type, msgSize, pfBit, id); 841} 842 843void Profiler::addL2StatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id) 844{ 845 m_perProcTotalMisses[id]++; 846 if (type == AccessModeType_SupervisorMode) { 847 m_perProcSupervisorMisses[id]++; 848 } else { 849 m_perProcUserMisses[id]++; 850 } 851 m_L2_cache_profiler_ptr->addStatSample(requestType, type, msgSize, pfBit); 852} 853 854void Profiler::addL1DStatSample(const CacheMsg& msg, NodeID id) 855{ 856 m_L1D_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 857 msg.getAccessMode(), msg.getSize(), msg.getPrefetch()); 858} 859 860void Profiler::addL1IStatSample(const CacheMsg& msg, NodeID id) 861{ 862 m_L1I_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()), 863 msg.getAccessMode(), msg.getSize(), msg.getPrefetch()); 864} 865 866void Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id) 867{ 868 if (msg.getType() != CacheRequestType_IFETCH) { 869 870 // Note: The following line should be commented out if you want to 871 // use the special profiling that is part of the GS320 protocol 872
|
758 // NOTE: Unless PROFILE_HOT_LINES or PROFILE_ALL_INSTRUCTIONS are enabled, nothing will be profiled by the AddressProfiler 759 m_address_profiler_ptr->addTraceSample(msg.getAddress(), msg.getProgramCounter(), msg.getType(), msg.getAccessMode(), id, false);
| 873 // NOTE: Unless PROFILE_HOT_LINES or RubyConfig::getProfileAllInstructions() are enabled, nothing will be profiled by the AddressProfiler 874 m_address_profiler_ptr->addTraceSample(msg.getLineAddress(), msg.getProgramCounter(), msg.getType(), msg.getAccessMode(), id, false);
|
760 } 761} 762 763void Profiler::profileSharing(const Address& addr, AccessType type, NodeID requestor, const Set& sharers, const Set& owner) 764{ 765 Set set_contacted(owner); 766 if (type == AccessType_Write) { 767 set_contacted.addSet(sharers); 768 } 769 set_contacted.remove(requestor); 770 int number_contacted = set_contacted.count(); 771 772 if (type == AccessType_Write) { 773 m_write_sharing_histogram.add(number_contacted); 774 } else { 775 m_read_sharing_histogram.add(number_contacted); 776 } 777 m_all_sharing_histogram.add(number_contacted); 778 779 if (number_contacted == 0) { 780 m_memory_to_cache++; 781 } else { 782 m_cache_to_cache++; 783 } 784 785} 786 787void Profiler::profileMsgDelay(int virtualNetwork, int delayCycles) { 788 assert(virtualNetwork < m_delayedCyclesVCHistograms.size()); 789 m_delayedCyclesHistogram.add(delayCycles); 790 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles); 791 if (virtualNetwork != 0) { 792 m_delayedCyclesNonPFHistogram.add(delayCycles); 793 } 794} 795 796// profiles original cache requests including PUTs 797void Profiler::profileRequest(const string& requestStr) 798{ 799 m_requests++; 800 801 if (m_requestProfileMap_ptr->exist(requestStr)) { 802 (m_requestProfileMap_ptr->lookup(requestStr))++; 803 } else { 804 m_requestProfileMap_ptr->add(requestStr, 1); 805 } 806} 807 808void Profiler::recordPrediction(bool wasGood, bool wasPredicted) 809{ 810 m_predictionOpportunities++; 811 if(wasPredicted){ 812 m_predictions++; 813 if(wasGood){ 814 m_goodPredictions++; 815 } 816 } 817} 818 819void Profiler::profileFilterAction(int action) 820{ 821 m_filter_action_histogram.add(action); 822} 823 824void Profiler::profileMulticastRetry(const Address& addr, int count) 825{ 826 m_multicast_retry_histogram.add(count); 827} 828 829void Profiler::startTransaction(int cpu) 830{ 831 m_perProcStartTransaction[cpu]++; 832} 833 834void Profiler::endTransaction(int cpu) 835{ 836 m_perProcEndTransaction[cpu]++; 837} 838 839void Profiler::controllerBusy(MachineID machID) 840{ 841 m_busyControllerCount[(int)machID.type][(int)machID.num]++; 842} 843 844void Profiler::profilePFWait(Time waitTime) 845{ 846 m_prefetchWaitHistogram.add(waitTime); 847} 848 849void Profiler::bankBusy() 850{ 851 m_busyBankCount++; 852} 853 854// non-zero cycle demand request
| 875 } 876} 877 878void Profiler::profileSharing(const Address& addr, AccessType type, NodeID requestor, const Set& sharers, const Set& owner) 879{ 880 Set set_contacted(owner); 881 if (type == AccessType_Write) { 882 set_contacted.addSet(sharers); 883 } 884 set_contacted.remove(requestor); 885 int number_contacted = set_contacted.count(); 886 887 if (type == AccessType_Write) { 888 m_write_sharing_histogram.add(number_contacted); 889 } else { 890 m_read_sharing_histogram.add(number_contacted); 891 } 892 m_all_sharing_histogram.add(number_contacted); 893 894 if (number_contacted == 0) { 895 m_memory_to_cache++; 896 } else { 897 m_cache_to_cache++; 898 } 899 900} 901 902void Profiler::profileMsgDelay(int virtualNetwork, int delayCycles) { 903 assert(virtualNetwork < m_delayedCyclesVCHistograms.size()); 904 m_delayedCyclesHistogram.add(delayCycles); 905 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles); 906 if (virtualNetwork != 0) { 907 m_delayedCyclesNonPFHistogram.add(delayCycles); 908 } 909} 910 911// profiles original cache requests including PUTs 912void Profiler::profileRequest(const string& requestStr) 913{ 914 m_requests++; 915 916 if (m_requestProfileMap_ptr->exist(requestStr)) { 917 (m_requestProfileMap_ptr->lookup(requestStr))++; 918 } else { 919 m_requestProfileMap_ptr->add(requestStr, 1); 920 } 921} 922 923void Profiler::recordPrediction(bool wasGood, bool wasPredicted) 924{ 925 m_predictionOpportunities++; 926 if(wasPredicted){ 927 m_predictions++; 928 if(wasGood){ 929 m_goodPredictions++; 930 } 931 } 932} 933 934void Profiler::profileFilterAction(int action) 935{ 936 m_filter_action_histogram.add(action); 937} 938 939void Profiler::profileMulticastRetry(const Address& addr, int count) 940{ 941 m_multicast_retry_histogram.add(count); 942} 943 944void Profiler::startTransaction(int cpu) 945{ 946 m_perProcStartTransaction[cpu]++; 947} 948 949void Profiler::endTransaction(int cpu) 950{ 951 m_perProcEndTransaction[cpu]++; 952} 953 954void Profiler::controllerBusy(MachineID machID) 955{ 956 m_busyControllerCount[(int)machID.type][(int)machID.num]++; 957} 958 959void Profiler::profilePFWait(Time waitTime) 960{ 961 m_prefetchWaitHistogram.add(waitTime); 962} 963 964void Profiler::bankBusy() 965{ 966 m_busyBankCount++; 967} 968 969// non-zero cycle demand request
|
855void Profiler::missLatency(Time t, CacheRequestType type, GenericMachineType respondingMach)
| 970void Profiler::missLatency(Time t, RubyRequestType type)
|
856{ 857 m_allMissLatencyHistogram.add(t); 858 m_missLatencyHistograms[type].add(t);
| 971{ 972 m_allMissLatencyHistogram.add(t); 973 m_missLatencyHistograms[type].add(t);
|
| 974 /*
|
859 m_machLatencyHistograms[respondingMach].add(t); 860 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 861 m_L2MissLatencyHistogram.add(t); 862 }
| 975 m_machLatencyHistograms[respondingMach].add(t); 976 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 977 m_L2MissLatencyHistogram.add(t); 978 }
|
| 979 */
|
863} 864 865// non-zero cycle prefetch request 866void Profiler::swPrefetchLatency(Time t, CacheRequestType type, GenericMachineType respondingMach) 867{ 868 m_allSWPrefetchLatencyHistogram.add(t); 869 m_SWPrefetchLatencyHistograms[type].add(t); 870 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t); 871 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 872 m_SWPrefetchL2MissLatencyHistogram.add(t); 873 } 874} 875
| 980} 981 982// non-zero cycle prefetch request 983void Profiler::swPrefetchLatency(Time t, CacheRequestType type, GenericMachineType respondingMach) 984{ 985 m_allSWPrefetchLatencyHistogram.add(t); 986 m_SWPrefetchLatencyHistograms[type].add(t); 987 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t); 988 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) { 989 m_SWPrefetchL2MissLatencyHistogram.add(t); 990 } 991} 992
|
876void Profiler::profileTransition(const string& component, NodeID id, NodeID version, Address addr,
| 993void Profiler::profileTransition(const string& component, NodeID version, Address addr,
|
877 const string& state, const string& event, 878 const string& next_state, const string& note) 879{ 880 const int EVENT_SPACES = 20; 881 const int ID_SPACES = 3; 882 const int TIME_SPACES = 7; 883 const int COMP_SPACES = 10; 884 const int STATE_SPACES = 6; 885 886 if ((g_debug_ptr->getDebugTime() > 0) && 887 (g_eventQueue_ptr->getTime() >= g_debug_ptr->getDebugTime())) { 888 (* debug_cout_ptr).flags(ios::right); 889 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
| 994 const string& state, const string& event, 995 const string& next_state, const string& note) 996{ 997 const int EVENT_SPACES = 20; 998 const int ID_SPACES = 3; 999 const int TIME_SPACES = 7; 1000 const int COMP_SPACES = 10; 1001 const int STATE_SPACES = 6; 1002 1003 if ((g_debug_ptr->getDebugTime() > 0) && 1004 (g_eventQueue_ptr->getTime() >= g_debug_ptr->getDebugTime())) { 1005 (* debug_cout_ptr).flags(ios::right); 1006 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
|
890 (* debug_cout_ptr) << setw(ID_SPACES) << id << " ";
| |
891 (* debug_cout_ptr) << setw(ID_SPACES) << version << " "; 892 (* debug_cout_ptr) << setw(COMP_SPACES) << component; 893 (* debug_cout_ptr) << setw(EVENT_SPACES) << event << " ";
| 1007 (* debug_cout_ptr) << setw(ID_SPACES) << version << " "; 1008 (* debug_cout_ptr) << setw(COMP_SPACES) << component; 1009 (* debug_cout_ptr) << setw(EVENT_SPACES) << event << " ";
|
894 for (int i=0; i < RubyConfig::numberOfProcessors(); i++) {
| |
895
| 1010
|
896 if (i == id) { 897 (* debug_cout_ptr).flags(ios::right); 898 (* debug_cout_ptr) << setw(STATE_SPACES) << state; 899 (* debug_cout_ptr) << ">"; 900 (* debug_cout_ptr).flags(ios::left); 901 (* debug_cout_ptr) << setw(STATE_SPACES) << next_state; 902 } else { 903 // cout << setw(STATE_SPACES) << " " << " " << setw(STATE_SPACES) << " "; 904 } 905 }
| 1011 (* debug_cout_ptr).flags(ios::right); 1012 (* debug_cout_ptr) << setw(STATE_SPACES) << state; 1013 (* debug_cout_ptr) << ">"; 1014 (* debug_cout_ptr).flags(ios::left); 1015 (* debug_cout_ptr) << setw(STATE_SPACES) << next_state; 1016
|
906 (* debug_cout_ptr) << " " << addr << " " << note; 907 908 (* debug_cout_ptr) << endl; 909 } 910} 911 912// Helper function 913static double process_memory_total() 914{ 915 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 916 ifstream proc_file; 917 proc_file.open("/proc/self/statm"); 918 int total_size_in_pages = 0; 919 int res_size_in_pages = 0; 920 proc_file >> total_size_in_pages; 921 proc_file >> res_size_in_pages; 922 return double(total_size_in_pages)*MULTIPLIER; // size in megabytes 923} 924 925static double process_memory_resident() 926{ 927 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 928 ifstream proc_file; 929 proc_file.open("/proc/self/statm"); 930 int total_size_in_pages = 0; 931 int res_size_in_pages = 0; 932 proc_file >> total_size_in_pages; 933 proc_file >> res_size_in_pages; 934 return double(res_size_in_pages)*MULTIPLIER; // size in megabytes 935} 936 937void Profiler::profileGetXMaskPrediction(const Set& pred_set) 938{ 939 m_getx_mask_prediction.add(pred_set.count()); 940} 941 942void Profiler::profileGetSMaskPrediction(const Set& pred_set) 943{ 944 m_gets_mask_prediction.add(pred_set.count()); 945} 946 947void Profiler::profileTrainingMask(const Set& pred_set) 948{ 949 m_explicit_training_mask.add(pred_set.count()); 950} 951
| 1017 (* debug_cout_ptr) << " " << addr << " " << note; 1018 1019 (* debug_cout_ptr) << endl; 1020 } 1021} 1022 1023// Helper function 1024static double process_memory_total() 1025{ 1026 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 1027 ifstream proc_file; 1028 proc_file.open("/proc/self/statm"); 1029 int total_size_in_pages = 0; 1030 int res_size_in_pages = 0; 1031 proc_file >> total_size_in_pages; 1032 proc_file >> res_size_in_pages; 1033 return double(total_size_in_pages)*MULTIPLIER; // size in megabytes 1034} 1035 1036static double process_memory_resident() 1037{ 1038 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB, 1039 ifstream proc_file; 1040 proc_file.open("/proc/self/statm"); 1041 int total_size_in_pages = 0; 1042 int res_size_in_pages = 0; 1043 proc_file >> total_size_in_pages; 1044 proc_file >> res_size_in_pages; 1045 return double(res_size_in_pages)*MULTIPLIER; // size in megabytes 1046} 1047 1048void Profiler::profileGetXMaskPrediction(const Set& pred_set) 1049{ 1050 m_getx_mask_prediction.add(pred_set.count()); 1051} 1052 1053void Profiler::profileGetSMaskPrediction(const Set& pred_set) 1054{ 1055 m_gets_mask_prediction.add(pred_set.count()); 1056} 1057 1058void Profiler::profileTrainingMask(const Set& pred_set) 1059{ 1060 m_explicit_training_mask.add(pred_set.count()); 1061} 1062
|
952// For MemoryControl: 953void Profiler::profileMemReq(int bank) { 954 m_memReq++; 955 m_memBankCount[bank]++; 956} 957 958void Profiler::profileMemBankBusy() { m_memBankBusy++; } 959void Profiler::profileMemBusBusy() { m_memBusBusy++; } 960void Profiler::profileMemReadWriteBusy() { m_memReadWriteBusy++; } 961void Profiler::profileMemDataBusBusy() { m_memDataBusBusy++; } 962void Profiler::profileMemTfawBusy() { m_memTfawBusy++; } 963void Profiler::profileMemRefresh() { m_memRefresh++; } 964void Profiler::profileMemRead() { m_memRead++; } 965void Profiler::profileMemWrite() { m_memWrite++; } 966void Profiler::profileMemWaitCycles(int cycles) { m_memWaitCycles += cycles; } 967void Profiler::profileMemInputQ(int cycles) { m_memInputQ += cycles; } 968void Profiler::profileMemBankQ(int cycles) { m_memBankQ += cycles; } 969void Profiler::profileMemArbWait(int cycles) { m_memArbWait += cycles; } 970void Profiler::profileMemRandBusy() { m_memRandBusy++; } 971void Profiler::profileMemNotOld() { m_memNotOld++; } 972
| |
973int64 Profiler::getTotalInstructionsExecuted() const 974{ 975 int64 sum = 1; // Starting at 1 allows us to avoid division by zero
| 1063int64 Profiler::getTotalInstructionsExecuted() const 1064{ 1065 int64 sum = 1; // Starting at 1 allows us to avoid division by zero
|
976 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) { 977 sum += (g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i]);
| 1066 for(int i=0; i < RubySystem::getNumberOfSequencers(); i++) { 1067 sum += (g_system_ptr->getInstructionCount(i) - m_instructions_executed_at_start[i]);
|
978 } 979 return sum; 980} 981 982int64 Profiler::getTotalTransactionsExecuted() const 983{ 984 int64 sum = m_perProcEndTransaction.sum(); 985 if (sum > 0) { 986 return sum; 987 } else { 988 return 1; // Avoid division by zero errors 989 } 990} 991 992 993// The following case statement converts CacheRequestTypes to GenericRequestTypes 994// allowing all profiling to be done with a single enum type instead of slow strings 995GenericRequestType Profiler::CacheRequestType_to_GenericRequestType(const CacheRequestType& type) { 996 switch (type) { 997 case CacheRequestType_LD: 998 return GenericRequestType_LD; 999 break; 1000 case CacheRequestType_ST: 1001 return GenericRequestType_ST; 1002 break; 1003 case CacheRequestType_ATOMIC: 1004 return GenericRequestType_ATOMIC; 1005 break; 1006 case CacheRequestType_IFETCH: 1007 return GenericRequestType_IFETCH; 1008 break; 1009 case CacheRequestType_NULL: 1010 return GenericRequestType_NULL; 1011 break; 1012 default: 1013 ERROR_MSG("Unexpected cache request type"); 1014 } 1015} 1016
| 1068 } 1069 return sum; 1070} 1071 1072int64 Profiler::getTotalTransactionsExecuted() const 1073{ 1074 int64 sum = m_perProcEndTransaction.sum(); 1075 if (sum > 0) { 1076 return sum; 1077 } else { 1078 return 1; // Avoid division by zero errors 1079 } 1080} 1081 1082 1083// The following case statement converts CacheRequestTypes to GenericRequestTypes 1084// allowing all profiling to be done with a single enum type instead of slow strings 1085GenericRequestType Profiler::CacheRequestType_to_GenericRequestType(const CacheRequestType& type) { 1086 switch (type) { 1087 case CacheRequestType_LD: 1088 return GenericRequestType_LD; 1089 break; 1090 case CacheRequestType_ST: 1091 return GenericRequestType_ST; 1092 break; 1093 case CacheRequestType_ATOMIC: 1094 return GenericRequestType_ATOMIC; 1095 break; 1096 case CacheRequestType_IFETCH: 1097 return GenericRequestType_IFETCH; 1098 break; 1099 case CacheRequestType_NULL: 1100 return GenericRequestType_NULL; 1101 break; 1102 default: 1103 ERROR_MSG("Unexpected cache request type"); 1104 } 1105} 1106
|
| 1107void Profiler::rubyWatch(int id){ 1108 int rn_g1 = 0;//SIMICS_get_register_number(id, "g1"); 1109 uint64 tr = 0;//SIMICS_read_register(id, rn_g1); 1110 Address watch_address = Address(tr); 1111 const int ID_SPACES = 3; 1112 const int TIME_SPACES = 7; 1113 1114 (* debug_cout_ptr).flags(ios::right); 1115 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " "; 1116 (* debug_cout_ptr) << setw(ID_SPACES) << id << " " 1117 << "RUBY WATCH " 1118 << watch_address 1119 << endl; 1120 1121 if(!m_watch_address_list_ptr->exist(watch_address)){ 1122 m_watch_address_list_ptr->add(watch_address, 1); 1123 } 1124} 1125 1126bool Profiler::watchAddress(Address addr){ 1127 if (m_watch_address_list_ptr->exist(addr)) 1128 return true; 1129 else 1130 return false; 1131} 1132 1133// For MemoryControl: 1134void Profiler::profileMemReq(string name, int bank) { 1135// printf("name is %s", name.c_str()); 1136 assert(m_memory_control_profilers.count(name) == 1); 1137 m_memory_control_profilers[name] -> m_memReq++; 1138 m_memory_control_profilers[name] -> m_memBankCount[bank]++; 1139} 1140void Profiler::profileMemBankBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankBusy++; } 1141void Profiler::profileMemBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBusBusy++; } 1142void Profiler::profileMemReadWriteBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memReadWriteBusy++; } 1143void Profiler::profileMemDataBusBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memDataBusBusy++; } 1144void Profiler::profileMemTfawBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memTfawBusy++; } 1145void Profiler::profileMemRefresh(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRefresh++; } 1146void Profiler::profileMemRead(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRead++; } 1147void Profiler::profileMemWrite(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWrite++; } 1148void Profiler::profileMemWaitCycles(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memWaitCycles += cycles; } 1149void Profiler::profileMemInputQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memInputQ += cycles; } 1150void Profiler::profileMemBankQ(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memBankQ += cycles; } 1151void Profiler::profileMemArbWait(string name, int cycles) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memArbWait += cycles; } 1152void Profiler::profileMemRandBusy(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memRandBusy++; } 1153void Profiler::profileMemNotOld(string name) { assert(m_memory_control_profilers.count(name) == 1); m_memory_control_profilers[name] -> m_memNotOld++; } 1154
|
| |