ruby: Import ruby and slicc from GEMS
[gem5.git] / src / mem / ruby / profiler / Profiler.cc
1 /*
2 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 /*
30 This file has been modified by Kevin Moore and Dan Nussbaum of the
31 Scalable Systems Research Group at Sun Microsystems Laboratories
32 (http://research.sun.com/scalable/) to support the Adaptive
33 Transactional Memory Test Platform (ATMTP).
34
35 Please send email to atmtp-interest@sun.com with feedback, questions, or
36 to request future announcements about ATMTP.
37
38 ----------------------------------------------------------------------
39
40 File modification date: 2008-02-23
41
42 ----------------------------------------------------------------------
43 */
44
45 /*
46 * Profiler.C
47 *
48 * Description: See Profiler.h
49 *
50 * $Id$
51 *
52 */
53
54 #include "Profiler.hh"
55 #include "CacheProfiler.hh"
56 #include "AddressProfiler.hh"
57 #include "System.hh"
58 #include "Network.hh"
59 #include "PrioHeap.hh"
60 #include "CacheMsg.hh"
61 #include "Driver.hh"
62 #include "Protocol.hh"
63 #include "util.hh"
64 #include "Map.hh"
65 #include "Debug.hh"
66 #include "MachineType.hh"
67 // #include "TransactionInterfaceManager.hh"
68 #include "interface.hh"
69 //#include "XactVisualizer.hh" //gem5:Arka for decomissioning log_tm
70 //#include "XactProfiler.hh" //gem5:Arka for decomissioning log_tm
71
72 // extern "C" {
73 // #include "Rock.hh"
74 // }
75
76 // Allows use of times() library call, which determines virtual runtime
77 #include <sys/times.h>
78
79 extern std::ostream * debug_cout_ptr;
80 extern std::ostream * xact_cout_ptr;
81
82 static double process_memory_total();
83 static double process_memory_resident();
84
85 Profiler::Profiler()
86 : m_conflicting_histogram(-1)
87 {
88 m_requestProfileMap_ptr = new Map<string, int>;
89 m_L1D_cache_profiler_ptr = new CacheProfiler("L1D_cache");
90 m_L1I_cache_profiler_ptr = new CacheProfiler("L1I_cache");
91
92 m_L2_cache_profiler_ptr = new CacheProfiler("L2_cache");
93
94 m_address_profiler_ptr = new AddressProfiler;
95 m_inst_profiler_ptr = NULL;
96 if (PROFILE_ALL_INSTRUCTIONS) {
97 m_inst_profiler_ptr = new AddressProfiler;
98 }
99
100 //m_xact_profiler_ptr = new XactProfiler; //gem5:Arka for decomissioning og log_tm
101
102 m_conflicting_map_ptr = new Map<Address, Time>;
103
104 m_real_time_start_time = time(NULL); // Not reset in clearStats()
105 m_stats_period = 1000000; // Default
106 m_periodic_output_file_ptr = &cerr;
107 m_xact_visualizer_ptr = &cout;
108
109 //---- begin XACT_MEM code
110 m_xactExceptionMap_ptr = new Map<int, int>;
111 m_procsInXactMap_ptr = new Map<int, int>;
112 m_abortIDMap_ptr = new Map<int, int>;
113 m_commitIDMap_ptr = new Map<int, int>;
114 m_xactRetryIDMap_ptr = new Map<int, int>;
115 m_xactCyclesIDMap_ptr = new Map<int, int>;
116 m_xactReadSetIDMap_ptr = new Map<int, int>;
117 m_xactWriteSetIDMap_ptr = new Map<int, int>;
118 m_xactLoadMissIDMap_ptr = new Map<int, int>;
119 m_xactStoreMissIDMap_ptr = new Map<int, int>;
120 m_xactInstrCountIDMap_ptr = new Map<int, integer_t>;
121 m_abortPCMap_ptr = new Map<Address, int>;
122 m_abortAddressMap_ptr = new Map<Address, int>;
123 m_nackXIDMap_ptr = new Map<int, int>;
124 m_nackXIDPairMap_ptr = new Map<int, Map<int, int> * >;
125 m_nackPCMap_ptr = new Map<Address, int>;
126 m_watch_address_list_ptr = new Map<Address, int>;
127 m_readSetMatch_ptr = new Map<Address, int>;
128 m_readSetNoMatch_ptr = new Map<Address, int>;
129 m_writeSetMatch_ptr = new Map<Address, int>;
130 m_writeSetNoMatch_ptr = new Map<Address, int>;
131 m_xactReadFilterBitsSetOnCommit = new Map<int, Histogram>;
132 m_xactReadFilterBitsSetOnAbort = new Map<int, Histogram>;
133 m_xactWriteFilterBitsSetOnCommit = new Map<int, Histogram>;
134 m_xactWriteFilterBitsSetOnAbort = new Map<int, Histogram>;
135 //---- end XACT_MEM code
136
137 // for MemoryControl:
138 m_memReq = 0;
139 m_memBankBusy = 0;
140 m_memBusBusy = 0;
141 m_memReadWriteBusy = 0;
142 m_memDataBusBusy = 0;
143 m_memTfawBusy = 0;
144 m_memRefresh = 0;
145 m_memRead = 0;
146 m_memWrite = 0;
147 m_memWaitCycles = 0;
148 m_memInputQ = 0;
149 m_memBankQ = 0;
150 m_memArbWait = 0;
151 m_memRandBusy = 0;
152 m_memNotOld = 0;
153
154
155 int totalBanks = RubyConfig::banksPerRank()
156 * RubyConfig::ranksPerDimm()
157 * RubyConfig::dimmsPerChannel();
158 m_memBankCount.setSize(totalBanks);
159
160 clearStats();
161 }
162
163 Profiler::~Profiler()
164 {
165 if (m_periodic_output_file_ptr != &cerr) {
166 delete m_periodic_output_file_ptr;
167 }
168 delete m_address_profiler_ptr;
169 delete m_L1D_cache_profiler_ptr;
170 delete m_L1I_cache_profiler_ptr;
171 delete m_L2_cache_profiler_ptr;
172 //delete m_xact_profiler_ptr; //gem5:Arka for decomissioning of log_tm
173 delete m_requestProfileMap_ptr;
174 delete m_conflicting_map_ptr;
175 }
176
177 void Profiler::wakeup()
178 {
179 // FIXME - avoid the repeated code
180
181 Vector<integer_t> perProcInstructionCount;
182 perProcInstructionCount.setSize(RubyConfig::numberOfProcessors());
183
184 Vector<integer_t> perProcCycleCount;
185 perProcCycleCount.setSize(RubyConfig::numberOfProcessors());
186
187 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) {
188 perProcInstructionCount[i] = g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1;
189 perProcCycleCount[i] = g_system_ptr->getDriver()->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
190 // The +1 allows us to avoid division by zero
191 }
192
193 integer_t total_misses = m_perProcTotalMisses.sum();
194 integer_t instruction_executed = perProcInstructionCount.sum();
195 integer_t simics_cycles_executed = perProcCycleCount.sum();
196 integer_t transactions_started = m_perProcStartTransaction.sum();
197 integer_t transactions_ended = m_perProcEndTransaction.sum();
198
199 (*m_periodic_output_file_ptr) << "ruby_cycles: " << g_eventQueue_ptr->getTime()-m_ruby_start << endl;
200 (*m_periodic_output_file_ptr) << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl;
201 (*m_periodic_output_file_ptr) << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
202 (*m_periodic_output_file_ptr) << "simics_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl;
203 (*m_periodic_output_file_ptr) << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl;
204 (*m_periodic_output_file_ptr) << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
205 (*m_periodic_output_file_ptr) << "L1TBE_usage: " << m_L1tbeProfile << endl;
206 (*m_periodic_output_file_ptr) << "L2TBE_usage: " << m_L2tbeProfile << endl;
207 (*m_periodic_output_file_ptr) << "mbytes_resident: " << process_memory_resident() << endl;
208 (*m_periodic_output_file_ptr) << "mbytes_total: " << process_memory_total() << endl;
209 if (process_memory_total() > 0) {
210 (*m_periodic_output_file_ptr) << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl;
211 }
212 (*m_periodic_output_file_ptr) << "miss_latency: " << m_allMissLatencyHistogram << endl;
213
214 *m_periodic_output_file_ptr << endl;
215
216 if (PROFILE_ALL_INSTRUCTIONS) {
217 m_inst_profiler_ptr->printStats(*m_periodic_output_file_ptr);
218 }
219
220 //g_system_ptr->getNetwork()->printStats(*m_periodic_output_file_ptr);
221 g_eventQueue_ptr->scheduleEvent(this, m_stats_period);
222 }
223
224 void Profiler::setPeriodicStatsFile(const string& filename)
225 {
226 cout << "Recording periodic statistics to file '" << filename << "' every "
227 << m_stats_period << " Ruby cycles" << endl;
228
229 if (m_periodic_output_file_ptr != &cerr) {
230 delete m_periodic_output_file_ptr;
231 }
232
233 m_periodic_output_file_ptr = new ofstream(filename.c_str());
234 g_eventQueue_ptr->scheduleEvent(this, 1);
235 }
236
237 void Profiler::setPeriodicStatsInterval(integer_t period)
238 {
239 cout << "Recording periodic statistics every " << m_stats_period << " Ruby cycles" << endl;
240 m_stats_period = period;
241 g_eventQueue_ptr->scheduleEvent(this, 1);
242 }
243
244 void Profiler::printConfig(ostream& out) const
245 {
246 out << endl;
247 out << "Profiler Configuration" << endl;
248 out << "----------------------" << endl;
249 out << "periodic_stats_period: " << m_stats_period << endl;
250 }
251
252 void Profiler::print(ostream& out) const
253 {
254 out << "[Profiler]";
255 }
256
257 void Profiler::printStats(ostream& out, bool short_stats)
258 {
259 out << endl;
260 if (short_stats) {
261 out << "SHORT ";
262 }
263 out << "Profiler Stats" << endl;
264 out << "--------------" << endl;
265
266 time_t real_time_current = time(NULL);
267 double seconds = difftime(real_time_current, m_real_time_start_time);
268 double minutes = seconds/60.0;
269 double hours = minutes/60.0;
270 double days = hours/24.0;
271 Time ruby_cycles = g_eventQueue_ptr->getTime()-m_ruby_start;
272
273 if (!short_stats) {
274 out << "Elapsed_time_in_seconds: " << seconds << endl;
275 out << "Elapsed_time_in_minutes: " << minutes << endl;
276 out << "Elapsed_time_in_hours: " << hours << endl;
277 out << "Elapsed_time_in_days: " << days << endl;
278 out << endl;
279 }
280
281 // print the virtual runtimes as well
282 struct tms vtime;
283 times(&vtime);
284 seconds = (vtime.tms_utime + vtime.tms_stime) / 100.0;
285 minutes = seconds / 60.0;
286 hours = minutes / 60.0;
287 days = hours / 24.0;
288 out << "Virtual_time_in_seconds: " << seconds << endl;
289 out << "Virtual_time_in_minutes: " << minutes << endl;
290 out << "Virtual_time_in_hours: " << hours << endl;
291 out << "Virtual_time_in_days: " << hours << endl;
292 out << endl;
293
294 out << "Ruby_current_time: " << g_eventQueue_ptr->getTime() << endl;
295 out << "Ruby_start_time: " << m_ruby_start << endl;
296 out << "Ruby_cycles: " << ruby_cycles << endl;
297 out << endl;
298
299 if (!short_stats) {
300 out << "mbytes_resident: " << process_memory_resident() << endl;
301 out << "mbytes_total: " << process_memory_total() << endl;
302 if (process_memory_total() > 0) {
303 out << "resident_ratio: " << process_memory_resident()/process_memory_total() << endl;
304 }
305 out << endl;
306
307 if(m_num_BA_broadcasts + m_num_BA_unicasts != 0){
308 out << endl;
309 out << "Broadcast_percent: " << (float)m_num_BA_broadcasts/(m_num_BA_broadcasts+m_num_BA_unicasts) << endl;
310 }
311 }
312
313 Vector<integer_t> perProcInstructionCount;
314 Vector<integer_t> perProcCycleCount;
315 Vector<double> perProcCPI;
316 Vector<double> perProcMissesPerInsn;
317 Vector<double> perProcInsnPerTrans;
318 Vector<double> perProcCyclesPerTrans;
319 Vector<double> perProcMissesPerTrans;
320
321 perProcInstructionCount.setSize(RubyConfig::numberOfProcessors());
322 perProcCycleCount.setSize(RubyConfig::numberOfProcessors());
323 perProcCPI.setSize(RubyConfig::numberOfProcessors());
324 perProcMissesPerInsn.setSize(RubyConfig::numberOfProcessors());
325
326 perProcInsnPerTrans.setSize(RubyConfig::numberOfProcessors());
327 perProcCyclesPerTrans.setSize(RubyConfig::numberOfProcessors());
328 perProcMissesPerTrans.setSize(RubyConfig::numberOfProcessors());
329
330 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) {
331 perProcInstructionCount[i] = g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i] + 1;
332 perProcCycleCount[i] = g_system_ptr->getDriver()->getCycleCount(i) - m_cycles_executed_at_start[i] + 1;
333 // The +1 allows us to avoid division by zero
334 perProcCPI[i] = double(ruby_cycles)/perProcInstructionCount[i];
335 perProcMissesPerInsn[i] = 1000.0 * (double(m_perProcTotalMisses[i]) / double(perProcInstructionCount[i]));
336
337 int trans = m_perProcEndTransaction[i];
338 if (trans == 0) {
339 perProcInsnPerTrans[i] = 0;
340 perProcCyclesPerTrans[i] = 0;
341 perProcMissesPerTrans[i] = 0;
342 } else {
343 perProcInsnPerTrans[i] = perProcInstructionCount[i] / double(trans);
344 perProcCyclesPerTrans[i] = ruby_cycles / double(trans);
345 perProcMissesPerTrans[i] = m_perProcTotalMisses[i] / double(trans);
346 }
347 }
348
349 integer_t total_misses = m_perProcTotalMisses.sum();
350 integer_t user_misses = m_perProcUserMisses.sum();
351 integer_t supervisor_misses = m_perProcSupervisorMisses.sum();
352 integer_t instruction_executed = perProcInstructionCount.sum();
353 integer_t simics_cycles_executed = perProcCycleCount.sum();
354 integer_t transactions_started = m_perProcStartTransaction.sum();
355 integer_t transactions_ended = m_perProcEndTransaction.sum();
356
357 double instructions_per_transaction = (transactions_ended != 0) ? double(instruction_executed) / double(transactions_ended) : 0;
358 double cycles_per_transaction = (transactions_ended != 0) ? (RubyConfig::numberOfProcessors() * double(ruby_cycles)) / double(transactions_ended) : 0;
359 double misses_per_transaction = (transactions_ended != 0) ? double(total_misses) / double(transactions_ended) : 0;
360
361 out << "Total_misses: " << total_misses << endl;
362 out << "total_misses: " << total_misses << " " << m_perProcTotalMisses << endl;
363 out << "user_misses: " << user_misses << " " << m_perProcUserMisses << endl;
364 out << "supervisor_misses: " << supervisor_misses << " " << m_perProcSupervisorMisses << endl;
365 out << endl;
366 out << "instruction_executed: " << instruction_executed << " " << perProcInstructionCount << endl;
367 out << "simics_cycles_executed: " << simics_cycles_executed << " " << perProcCycleCount << endl;
368 out << "cycles_per_instruction: " << (RubyConfig::numberOfProcessors()*double(ruby_cycles))/double(instruction_executed) << " " << perProcCPI << endl;
369 out << "misses_per_thousand_instructions: " << 1000.0 * (double(total_misses) / double(instruction_executed)) << " " << perProcMissesPerInsn << endl;
370 out << endl;
371 out << "transactions_started: " << transactions_started << " " << m_perProcStartTransaction << endl;
372 out << "transactions_ended: " << transactions_ended << " " << m_perProcEndTransaction << endl;
373 out << "instructions_per_transaction: " << instructions_per_transaction << " " << perProcInsnPerTrans << endl;
374 out << "cycles_per_transaction: " << cycles_per_transaction << " " << perProcCyclesPerTrans << endl;
375 out << "misses_per_transaction: " << misses_per_transaction << " " << perProcMissesPerTrans << endl;
376
377 out << endl;
378
379 m_L1D_cache_profiler_ptr->printStats(out);
380 m_L1I_cache_profiler_ptr->printStats(out);
381 m_L2_cache_profiler_ptr->printStats(out);
382
383 out << endl;
384
385 if (m_memReq || m_memRefresh) { // if there's a memory controller at all
386 long long int total_stalls = m_memInputQ + m_memBankQ + m_memWaitCycles;
387 double stallsPerReq = total_stalls * 1.0 / m_memReq;
388 out << "Memory control:" << endl;
389 out << " memory_total_requests: " << m_memReq << endl; // does not include refreshes
390 out << " memory_reads: " << m_memRead << endl;
391 out << " memory_writes: " << m_memWrite << endl;
392 out << " memory_refreshes: " << m_memRefresh << endl;
393 out << " memory_total_request_delays: " << total_stalls << endl;
394 out << " memory_delays_per_request: " << stallsPerReq << endl;
395 out << " memory_delays_in_input_queue: " << m_memInputQ << endl;
396 out << " memory_delays_behind_head_of_bank_queue: " << m_memBankQ << endl;
397 out << " memory_delays_stalled_at_head_of_bank_queue: " << m_memWaitCycles << endl;
398 // Note: The following "memory stalls" entries are a breakdown of the
399 // cycles which already showed up in m_memWaitCycles. The order is
400 // significant; it is the priority of attributing the cycles.
401 // For example, bank_busy is before arbitration because if the bank was
402 // busy, we didn't even check arbitration.
403 // Note: "not old enough" means that since we grouped waiting heads-of-queues
404 // into batches to avoid starvation, a request in a newer batch
405 // didn't try to arbitrate yet because there are older requests waiting.
406 out << " memory_stalls_for_bank_busy: " << m_memBankBusy << endl;
407 out << " memory_stalls_for_random_busy: " << m_memRandBusy << endl;
408 out << " memory_stalls_for_anti_starvation: " << m_memNotOld << endl;
409 out << " memory_stalls_for_arbitration: " << m_memArbWait << endl;
410 out << " memory_stalls_for_bus: " << m_memBusBusy << endl;
411 out << " memory_stalls_for_tfaw: " << m_memTfawBusy << endl;
412 out << " memory_stalls_for_read_write_turnaround: " << m_memReadWriteBusy << endl;
413 out << " memory_stalls_for_read_read_turnaround: " << m_memDataBusBusy << endl;
414 out << " accesses_per_bank: ";
415 for (int bank=0; bank < m_memBankCount.size(); bank++) {
416 out << m_memBankCount[bank] << " ";
417 //if ((bank % 8) == 7) out << " " << endl;
418 }
419 out << endl;
420 out << endl;
421 }
422
423 if (!short_stats) {
424 out << "Busy Controller Counts:" << endl;
425 for(int i=0; i < MachineType_NUM; i++) {
426 for(int j=0; j < MachineType_base_count((MachineType)i); j++) {
427 MachineID machID;
428 machID.type = (MachineType)i;
429 machID.num = j;
430 out << machID << ":" << m_busyControllerCount[i][j] << " ";
431 if ((j+1)%8 == 0) {
432 out << endl;
433 }
434 }
435 out << endl;
436 }
437 out << endl;
438
439 out << "Busy Bank Count:" << m_busyBankCount << endl;
440 out << endl;
441
442 out << "L1TBE_usage: " << m_L1tbeProfile << endl;
443 out << "L2TBE_usage: " << m_L2tbeProfile << endl;
444 out << "StopTable_usage: " << m_stopTableProfile << endl;
445 out << "sequencer_requests_outstanding: " << m_sequencer_requests << endl;
446 out << "store_buffer_size: " << m_store_buffer_size << endl;
447 out << "unique_blocks_in_store_buffer: " << m_store_buffer_blocks << endl;
448 out << endl;
449 }
450
451 if (!short_stats) {
452 out << "All Non-Zero Cycle Demand Cache Accesses" << endl;
453 out << "----------------------------------------" << endl;
454 out << "miss_latency: " << m_allMissLatencyHistogram << endl;
455 for(int i=0; i<m_missLatencyHistograms.size(); i++) {
456 if (m_missLatencyHistograms[i].size() > 0) {
457 out << "miss_latency_" << CacheRequestType(i) << ": " << m_missLatencyHistograms[i] << endl;
458 }
459 }
460 for(int i=0; i<m_machLatencyHistograms.size(); i++) {
461 if (m_machLatencyHistograms[i].size() > 0) {
462 out << "miss_latency_" << GenericMachineType(i) << ": " << m_machLatencyHistograms[i] << endl;
463 }
464 }
465 out << "miss_latency_L2Miss: " << m_L2MissLatencyHistogram << endl;
466
467 out << endl;
468
469 out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
470 out << "------------------------------------" << endl;
471 out << "prefetch_latency: " << m_allSWPrefetchLatencyHistogram << endl;
472 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) {
473 if (m_SWPrefetchLatencyHistograms[i].size() > 0) {
474 out << "prefetch_latency_" << CacheRequestType(i) << ": " << m_SWPrefetchLatencyHistograms[i] << endl;
475 }
476 }
477 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) {
478 if (m_SWPrefetchMachLatencyHistograms[i].size() > 0) {
479 out << "prefetch_latency_" << GenericMachineType(i) << ": " << m_SWPrefetchMachLatencyHistograms[i] << endl;
480 }
481 }
482 out << "prefetch_latency_L2Miss:" << m_SWPrefetchL2MissLatencyHistogram << endl;
483
484 out << "multicast_retries: " << m_multicast_retry_histogram << endl;
485 out << "gets_mask_prediction_count: " << m_gets_mask_prediction << endl;
486 out << "getx_mask_prediction_count: " << m_getx_mask_prediction << endl;
487 out << "explicit_training_mask: " << m_explicit_training_mask << endl;
488 out << endl;
489
490 if (m_all_sharing_histogram.size() > 0) {
491 out << "all_sharing: " << m_all_sharing_histogram << endl;
492 out << "read_sharing: " << m_read_sharing_histogram << endl;
493 out << "write_sharing: " << m_write_sharing_histogram << endl;
494
495 out << "all_sharing_percent: "; m_all_sharing_histogram.printPercent(out); out << endl;
496 out << "read_sharing_percent: "; m_read_sharing_histogram.printPercent(out); out << endl;
497 out << "write_sharing_percent: "; m_write_sharing_histogram.printPercent(out); out << endl;
498
499 int64 total_miss = m_cache_to_cache + m_memory_to_cache;
500 out << "all_misses: " << total_miss << endl;
501 out << "cache_to_cache_misses: " << m_cache_to_cache << endl;
502 out << "memory_to_cache_misses: " << m_memory_to_cache << endl;
503 out << "cache_to_cache_percent: " << 100.0 * (double(m_cache_to_cache) / double(total_miss)) << endl;
504 out << "memory_to_cache_percent: " << 100.0 * (double(m_memory_to_cache) / double(total_miss)) << endl;
505 out << endl;
506 }
507
508 if (m_conflicting_histogram.size() > 0) {
509 out << "conflicting_histogram: " << m_conflicting_histogram << endl;
510 out << "conflicting_histogram_percent: "; m_conflicting_histogram.printPercent(out); out << endl;
511 out << endl;
512 }
513
514 if (m_outstanding_requests.size() > 0) {
515 out << "outstanding_requests: "; m_outstanding_requests.printPercent(out); out << endl;
516 if (m_outstanding_persistent_requests.size() > 0) {
517 out << "outstanding_persistent_requests: "; m_outstanding_persistent_requests.printPercent(out); out << endl;
518 }
519 out << endl;
520 }
521 }
522
523 if (XACT_MEMORY){
524 // Transactional Memory stats
525 out << "Transactional Memory Stats:" << endl;
526 out << "------- xact --------" << endl;
527 out << "xact_size_dist: " << m_xactSizes << endl;
528 out << "xact_instr_count: " << m_xactInstrCount << endl;
529 out << "xact_time_dist: " << m_xactCycles << endl;
530 out << "xact_log_size_dist: " << m_xactLogs << endl;
531 out << "xact_read_set_size_dist: " << m_xactReads << endl;
532 out << "xact_write_set_size_dist: " << m_xactWrites << endl;
533 out << "xact_overflow_read_lines_dist: " << m_xactOverflowReads << endl;
534 out << "xact_overflow_write_lines_dist: " << m_xactOverflowWrites << endl;
535 out << "xact_overflow_read_set_size_dist: " << m_xactOverflowTotalReads << endl;
536 out << "xact_overflow_write_set_size_dist: " << m_xactOverflowTotalWrites << endl;
537 out << "xact_miss_load_dist: " << m_xactLoadMisses << endl;
538 out << "xact_miss_store_dist: " << m_xactStoreMisses << endl;
539 out << "xact_nacked: " << m_xactNacked << endl;
540 out << "xact_retries: " << m_xactRetries << endl;
541 out << "xact_abort_delays: " << m_abortDelays << endl;
542 out << "xact_aborts: " << m_transactionAborts << endl;
543 if (ATMTP_ENABLED) {
544 out << "xact_log_overflows: " << m_transactionLogOverflows << endl;
545 out << "xact_cache_overflows: " << m_transactionCacheOverflows << endl;
546 out << "xact_unsup_inst_aborts: " << m_transactionUnsupInsts << endl;
547 out << "xact_save_rest_aborts: " << m_transactionSaveRestAborts << endl;
548 }
549 out << "xact_writebacks: " << m_transWBs << endl;
550 out << "xact_extra_wbs: " << m_extraWBs << endl;
551 out << "xact_handler_startup_delay: " << m_abortStarupDelay << endl;
552 out << "xact_handler_per_block_delay: " << m_abortPerBlockDelay << endl;
553 out << "xact_inferred_aborts: " << m_inferredAborts << endl;
554 //out << "xact_histogram: " << m_procsInXact << endl;
555
556 if (!short_stats) {
557 Vector<int> nackedXIDKeys = m_nackXIDMap_ptr->keys();
558 nackedXIDKeys.sortVector();
559 out << endl;
560 int total_nacks = 0;
561 out << "------- xact Nacks by XID --------" << endl;
562 for(int i=0; i<nackedXIDKeys.size(); i++) {
563 int key = nackedXIDKeys[i];
564 int count = m_nackXIDMap_ptr->lookup(key);
565 total_nacks += count;
566 out << "xact " << key << " "
567 << setw(6) << dec << count
568 << endl;
569 }
570 out << "Total Nacks: " << total_nacks << endl;
571 out << "---------------" << endl;
572 out << endl;
573
574 // Print XID Nack Pairs
575 Vector<int> nackedXIDPairKeys = m_nackXIDPairMap_ptr->keys();
576 nackedXIDPairKeys.sortVector();
577 out << endl;
578 total_nacks = 0;
579 out << "------- xact Nacks by XID Pairs --------" << endl;
580 for(int i=0; i<nackedXIDPairKeys.size(); i++) {
581 int key = nackedXIDPairKeys[i];
582 Map<int, int> * my_map = m_nackXIDPairMap_ptr->lookup(key);
583 Vector<int> my_keys = my_map->keys();
584 my_keys.sortVector();
585 for(int j=0; j<my_keys.size(); j++){
586 int nid = my_keys[j];
587 int count = my_map->lookup(nid);
588 total_nacks += count;
589 out << "xact " << key << " nacked by xact " << nid << " "
590 << setw(6) << dec << count
591 << endl;
592 }
593 }
594 out << "Total Nacks: " << total_nacks << endl;
595 out << "---------------" << endl;
596 out << endl;
597
598
599 Vector<Address> nackedPCKeys = m_nackPCMap_ptr->keys();
600 nackedPCKeys.sortVector();
601 out << endl;
602 out << "------- xact Nacks by PC --------" << endl;
603 for(int i=0; i<nackedPCKeys.size(); i++) {
604 Address key = nackedPCKeys[i];
605 int count = m_nackPCMap_ptr->lookup(key);
606 out << "xact_Nack " << key << " "
607 << setw(4) << dec << count
608 << endl;
609 }
610 out << "---------------" << endl;
611 out << endl;
612
613
614 Vector<int> xactExceptionKeys = m_xactExceptionMap_ptr->keys();
615 xactExceptionKeys.sortVector();
616 out << "------- xact exceptions --------" << endl;
617 for(int i=0; i<xactExceptionKeys.size(); i++) {
618 int key = xactExceptionKeys[i];
619 int count = m_xactExceptionMap_ptr->lookup(key);
620 out << "xact_exception("
621 << hex << key << "):"
622 << setw(4) << dec << count
623 << endl;
624 }
625 out << endl;
626 out << "---------------" << endl;
627 out << endl;
628
629 Vector<int> abortIDKeys = m_abortIDMap_ptr->keys();
630 abortIDKeys.sortVector();
631 out << "------- xact abort by XID --------" << endl;
632 for(int i=0; i<abortIDKeys.size(); i++) {
633 int count = m_abortIDMap_ptr->lookup(abortIDKeys[i]);
634 out << "xact_aborts("
635 << dec << abortIDKeys[i] << "):"
636 << setw(7) << count
637 << endl;
638 }
639 out << endl;
640 out << "---------------" << endl;
641 out << endl;
642
643 Vector<Address> abortedPCKeys = m_abortPCMap_ptr->keys();
644 abortedPCKeys.sortVector();
645 out << endl;
646 out << "------- xact Aborts by PC --------" << endl;
647 for(int i=0; i<abortedPCKeys.size(); i++) {
648 Address key = abortedPCKeys[i];
649 int count = m_abortPCMap_ptr->lookup(key);
650 out << "xact_abort_pc " << key
651 << setw(4) << dec << count
652 << endl;
653 }
654 out << "---------------" << endl;
655 out << endl;
656
657 Vector<Address> abortedAddrKeys = m_abortAddressMap_ptr->keys();
658 abortedAddrKeys.sortVector();
659 out << endl;
660 out << "------- xact Aborts by Address --------" << endl;
661 for(int i=0; i<abortedAddrKeys.size(); i++) {
662 Address key = abortedAddrKeys[i];
663 int count = m_abortAddressMap_ptr->lookup(key);
664 out << "xact_abort_address " << key
665 << setw(4) << dec << count
666 << endl;
667 }
668 out << "---------------" << endl;
669 out << endl;
670 } // !short_stats
671
672 Vector<int> commitIDKeys = m_commitIDMap_ptr->keys();
673 commitIDKeys.sortVector();
674 out << "------- xact Commit Stats by XID --------" << endl;
675 for(int i=0; i<commitIDKeys.size(); i++) {
676 int count = m_commitIDMap_ptr->lookup(commitIDKeys[i]);
677 double retry_count = (double)m_xactRetryIDMap_ptr->lookup(commitIDKeys[i]) / count;
678 double cycles_count = (double)m_xactCyclesIDMap_ptr->lookup(commitIDKeys[i]) / count;
679 double readset_count = (double)m_xactReadSetIDMap_ptr->lookup(commitIDKeys[i]) / count;
680 double writeset_count = (double)m_xactWriteSetIDMap_ptr->lookup(commitIDKeys[i]) / count;
681 double loadmiss_count = (double)m_xactLoadMissIDMap_ptr->lookup(commitIDKeys[i]) / count;
682 double storemiss_count = (double)m_xactStoreMissIDMap_ptr->lookup(commitIDKeys[i]) / count;
683 double instr_count = (double)m_xactInstrCountIDMap_ptr->lookup(commitIDKeys[i]) / count;
684 out << "xact_stats id: "
685 << dec << commitIDKeys[i]
686 << " count: " << setw(7) << count
687 << " Cycles: " << setw(7) << cycles_count
688 << " Instr: " << setw(7) << instr_count
689 << " ReadSet: " << setw(7) << readset_count
690 << " WriteSet: " << setw(7) << writeset_count
691 << " LoadMiss: " << setw(7) << loadmiss_count
692 << " StoreMiss: " << setw(7) << storemiss_count
693 << " Retry Count: " << setw(7) << retry_count
694 << endl;
695 }
696 out << endl;
697 out << "---------------" << endl;
698 out << endl;
699
700 if (!short_stats) {
701 Vector<int> procsInXactKeys = m_procsInXactMap_ptr->keys();
702 procsInXactKeys.sortVector();
703 out << "------- xact histogram --------" << endl;
704 for(int i=0; i<procsInXactKeys.size(); i++) {
705 int count = m_procsInXactMap_ptr->lookup(procsInXactKeys[i]);
706 int key = procsInXactKeys[i];
707 out << "xact_histogram("
708 << dec << key << "):"
709 << setw(8) << count
710 << endl;
711 }
712 out << endl;
713 out << "---------------" << endl;
714 out << endl;
715
716 // Read/Write set Bloom filter stats
717 //int false_reads = 0;
718 long long int false_reads = m_readSetNoMatch;
719 Vector<Address> fp_read_keys = m_readSetNoMatch_ptr->keys();
720 out << "------- xact read set false positives -------" << endl;
721 for(int i=0; i < fp_read_keys.size(); ++i){
722 int count = m_readSetNoMatch_ptr->lookup(fp_read_keys[i]);
723 //out << "read_false_positive( " << fp_read_keys[i] << " ): "
724 // << setw(8) << dec << count << endl;
725 false_reads += count;
726 }
727 out << "Total read set false positives : " << setw(8) << false_reads << endl;
728 out << "-----------------------" << endl;
729 out << endl;
730
731 //int matching_reads = 0;
732 long long int matching_reads = m_readSetMatch;
733 long long int empty_checks = m_readSetEmptyChecks;
734 Vector<Address> read_keys = m_readSetMatch_ptr->keys();
735 out << "------- xact read set matches -------" << endl;
736 for(int i=0; i < read_keys.size(); ++i){
737 int count = m_readSetMatch_ptr->lookup(read_keys[i]);
738 //out << "read_match( " << read_keys[i] << " ): "
739 // << setw(8) << dec << count << endl;
740 matching_reads += count;
741 }
742 out << "Total read set matches : " << setw(8) << matching_reads << endl;
743 out << "Total read set empty checks : " << setw(8) << empty_checks << endl;
744 double false_positive_pct = 0.0;
745 if((false_reads + matching_reads)> 0){
746 false_positive_pct = (1.0*false_reads)/(false_reads+matching_reads)*100.0;
747 }
748 out << "Read set false positives rate : " << false_positive_pct << "%" << endl;
749 out << "-----------------------" << endl;
750 out << endl;
751
752 // for write set
753 //int false_writes = 0;
754 long long int false_writes = m_writeSetNoMatch;
755 Vector<Address> fp_write_keys = m_writeSetNoMatch_ptr->keys();
756 out << "------- xact write set false positives -------" << endl;
757 for(int i=0; i < fp_write_keys.size(); ++i){
758 int count = m_writeSetNoMatch_ptr->lookup(fp_write_keys[i]);
759 //out << "write_false_positive( " << fp_write_keys[i] << " ): "
760 // << setw(8) << dec << count << endl;
761 false_writes += count;
762 }
763 out << "Total write set false positives : " << setw(8) << false_writes << endl;
764 out << "-----------------------" << endl;
765 out << endl;
766
767 //int matching_writes = 0;
768 long long int matching_writes = m_writeSetMatch;
769 empty_checks = m_writeSetEmptyChecks;
770 Vector<Address> write_keys = m_writeSetMatch_ptr->keys();
771 out << "------- xact write set matches -------" << endl;
772 for(int i=0; i < write_keys.size(); ++i){
773 int count = m_writeSetMatch_ptr->lookup(write_keys[i]);
774 //out << "write_match( " << write_keys[i] << " ): "
775 // << setw(8) << dec << count << endl;
776 matching_writes += count;
777 }
778 out << "Total write set matches : " << setw(8) << matching_writes << endl;
779 out << "Total write set empty checks : " << setw(8) << empty_checks << endl;
780 false_positive_pct = 0.0;
781 if((matching_writes+false_writes) > 0){
782 false_positive_pct = (1.0*false_writes)/(false_writes+matching_writes)*100.0;
783 }
784 out << "Write set false positives rate : " << false_positive_pct << "%" << endl;
785 out << "-----------------------" << endl;
786 out << endl;
787
788 out << "----- Xact Signature Stats ------" << endl;
789 Vector<int> xids = m_xactReadFilterBitsSetOnCommit->keys();
790 for(int i=0; i < xids.size(); ++i){
791 int xid = xids[i];
792 out << "xid " << xid << " Read set bits set on commit: " << (m_xactReadFilterBitsSetOnCommit->lookup(xid)) << endl;
793 }
794 xids = m_xactWriteFilterBitsSetOnCommit->keys();
795 for(int i=0; i < xids.size(); ++i){
796 int xid = xids[i];
797 out << "xid " << xid << " Write set bits set on commit: " << (m_xactWriteFilterBitsSetOnCommit->lookup(xid)) << endl;
798 }
799 xids = m_xactReadFilterBitsSetOnAbort->keys();
800 for(int i=0; i < xids.size(); ++i){
801 int xid = xids[i];
802 out << "xid " << xid << " Read set bits set on abort: " << (m_xactReadFilterBitsSetOnAbort->lookup(xid)) << endl;
803 }
804 xids = m_xactWriteFilterBitsSetOnAbort->keys();
805 for(int i=0; i < xids.size(); ++i){
806 int xid = xids[i];
807 out << "xid " << xid << " Write set bits set on abort: " << (m_xactWriteFilterBitsSetOnAbort->lookup(xid)) << endl;
808 }
809 out << endl;
810
811 cout << "------- WATCHPOINTS --------" << endl;
812 cout << "False Triggers : " << m_watchpointsFalsePositiveTrigger << endl;
813 cout << "True Triggers : " << m_watchpointsTrueTrigger << endl;
814 cout << "Total Triggers : " << m_watchpointsTrueTrigger + m_watchpointsFalsePositiveTrigger << endl;
815 cout << "---------------" << endl;
816 cout << endl;
817 } // !short_stats
818 //m_xact_profiler_ptr->printStats(out, short_stats); // gem5:Arka for decomissioning of log_tm
819 } // XACT_MEMORY
820
821 if (!short_stats) {
822 out << "Request vs. System State Profile" << endl;
823 out << "--------------------------------" << endl;
824 out << endl;
825
826 Vector<string> requestProfileKeys = m_requestProfileMap_ptr->keys();
827 requestProfileKeys.sortVector();
828
829 for(int i=0; i<requestProfileKeys.size(); i++) {
830 int temp_int = m_requestProfileMap_ptr->lookup(requestProfileKeys[i]);
831 double percent = (100.0*double(temp_int))/double(m_requests);
832 while (requestProfileKeys[i] != "") {
833 out << setw(10) << string_split(requestProfileKeys[i], ':');
834 }
835 out << setw(11) << temp_int;
836 out << setw(14) << percent << endl;
837 }
838 out << endl;
839
840 out << "filter_action: " << m_filter_action_histogram << endl;
841
842 if (!PROFILE_ALL_INSTRUCTIONS) {
843 m_address_profiler_ptr->printStats(out);
844 }
845
846 if (PROFILE_ALL_INSTRUCTIONS) {
847 m_inst_profiler_ptr->printStats(out);
848 }
849
850 out << endl;
851 out << "Message Delayed Cycles" << endl;
852 out << "----------------------" << endl;
853 out << "Total_delay_cycles: " << m_delayedCyclesHistogram << endl;
854 out << "Total_nonPF_delay_cycles: " << m_delayedCyclesNonPFHistogram << endl;
855 for (int i = 0; i < m_delayedCyclesVCHistograms.size(); i++) {
856 out << " virtual_network_" << i << "_delay_cycles: " << m_delayedCyclesVCHistograms[i] << endl;
857 }
858
859 printResourceUsage(out);
860 }
861
862 }
863
864 void Profiler::printResourceUsage(ostream& out) const
865 {
866 out << endl;
867 out << "Resource Usage" << endl;
868 out << "--------------" << endl;
869
870 integer_t pagesize = getpagesize(); // page size in bytes
871 out << "page_size: " << pagesize << endl;
872
873 rusage usage;
874 getrusage (RUSAGE_SELF, &usage);
875
876 out << "user_time: " << usage.ru_utime.tv_sec << endl;
877 out << "system_time: " << usage.ru_stime.tv_sec << endl;
878 out << "page_reclaims: " << usage.ru_minflt << endl;
879 out << "page_faults: " << usage.ru_majflt << endl;
880 out << "swaps: " << usage.ru_nswap << endl;
881 out << "block_inputs: " << usage.ru_inblock << endl;
882 out << "block_outputs: " << usage.ru_oublock << endl;
883 }
884
885 void Profiler::clearStats()
886 {
887 m_num_BA_unicasts = 0;
888 m_num_BA_broadcasts = 0;
889
890 m_ruby_start = g_eventQueue_ptr->getTime();
891
892 m_instructions_executed_at_start.setSize(RubyConfig::numberOfProcessors());
893 m_cycles_executed_at_start.setSize(RubyConfig::numberOfProcessors());
894 for (int i=0; i < RubyConfig::numberOfProcessors(); i++) {
895 if (g_system_ptr == NULL) {
896 m_instructions_executed_at_start[i] = 0;
897 m_cycles_executed_at_start[i] = 0;
898 } else {
899 m_instructions_executed_at_start[i] = g_system_ptr->getDriver()->getInstructionCount(i);
900 m_cycles_executed_at_start[i] = g_system_ptr->getDriver()->getCycleCount(i);
901 }
902 }
903
904 m_perProcTotalMisses.setSize(RubyConfig::numberOfProcessors());
905 m_perProcUserMisses.setSize(RubyConfig::numberOfProcessors());
906 m_perProcSupervisorMisses.setSize(RubyConfig::numberOfProcessors());
907 m_perProcStartTransaction.setSize(RubyConfig::numberOfProcessors());
908 m_perProcEndTransaction.setSize(RubyConfig::numberOfProcessors());
909
910 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) {
911 m_perProcTotalMisses[i] = 0;
912 m_perProcUserMisses[i] = 0;
913 m_perProcSupervisorMisses[i] = 0;
914 m_perProcStartTransaction[i] = 0;
915 m_perProcEndTransaction[i] = 0;
916 }
917
918 m_busyControllerCount.setSize(MachineType_NUM); // all machines
919 for(int i=0; i < MachineType_NUM; i++) {
920 m_busyControllerCount[i].setSize(MachineType_base_count((MachineType)i));
921 for(int j=0; j < MachineType_base_count((MachineType)i); j++) {
922 m_busyControllerCount[i][j] = 0;
923 }
924 }
925 m_busyBankCount = 0;
926
927 m_delayedCyclesHistogram.clear();
928 m_delayedCyclesNonPFHistogram.clear();
929 m_delayedCyclesVCHistograms.setSize(NUMBER_OF_VIRTUAL_NETWORKS);
930 for (int i = 0; i < NUMBER_OF_VIRTUAL_NETWORKS; i++) {
931 m_delayedCyclesVCHistograms[i].clear();
932 }
933
934 m_gets_mask_prediction.clear();
935 m_getx_mask_prediction.clear();
936 m_explicit_training_mask.clear();
937
938 m_missLatencyHistograms.setSize(CacheRequestType_NUM);
939 for(int i=0; i<m_missLatencyHistograms.size(); i++) {
940 m_missLatencyHistograms[i].clear(200);
941 }
942 m_machLatencyHistograms.setSize(GenericMachineType_NUM+1);
943 for(int i=0; i<m_machLatencyHistograms.size(); i++) {
944 m_machLatencyHistograms[i].clear(200);
945 }
946 m_allMissLatencyHistogram.clear(200);
947 m_L2MissLatencyHistogram.clear(200);
948
949 m_SWPrefetchLatencyHistograms.setSize(CacheRequestType_NUM);
950 for(int i=0; i<m_SWPrefetchLatencyHistograms.size(); i++) {
951 m_SWPrefetchLatencyHistograms[i].clear(200);
952 }
953 m_SWPrefetchMachLatencyHistograms.setSize(GenericMachineType_NUM+1);
954 for(int i=0; i<m_SWPrefetchMachLatencyHistograms.size(); i++) {
955 m_SWPrefetchMachLatencyHistograms[i].clear(200);
956 }
957 m_allSWPrefetchLatencyHistogram.clear(200);
958 m_SWPrefetchL2MissLatencyHistogram.clear(200);
959
960 m_multicast_retry_histogram.clear();
961
962 m_L1tbeProfile.clear();
963 m_L2tbeProfile.clear();
964 m_stopTableProfile.clear();
965 m_filter_action_histogram.clear();
966
967 m_sequencer_requests.clear();
968 m_store_buffer_size.clear();
969 m_store_buffer_blocks.clear();
970 m_read_sharing_histogram.clear();
971 m_write_sharing_histogram.clear();
972 m_all_sharing_histogram.clear();
973 m_cache_to_cache = 0;
974 m_memory_to_cache = 0;
975
976 m_predictions = 0;
977 m_predictionOpportunities = 0;
978 m_goodPredictions = 0;
979
980 // clear HashMaps
981 m_requestProfileMap_ptr->clear();
982
983 // count requests profiled
984 m_requests = 0;
985
986 // Conflicting requests
987 m_conflicting_map_ptr->clear();
988 m_conflicting_histogram.clear();
989
990 m_outstanding_requests.clear();
991 m_outstanding_persistent_requests.clear();
992
993 m_L1D_cache_profiler_ptr->clearStats();
994 m_L1I_cache_profiler_ptr->clearStats();
995 m_L2_cache_profiler_ptr->clearStats();
996 //m_xact_profiler_ptr->clearStats(); //gem5:Arka for decomissiong of log_tm
997
998 //---- begin XACT_MEM code
999 ASSERT(m_xactExceptionMap_ptr != NULL);
1000 ASSERT(m_procsInXactMap_ptr != NULL);
1001 ASSERT(m_abortIDMap_ptr != NULL);
1002 ASSERT(m_abortPCMap_ptr != NULL);
1003 ASSERT( m_nackXIDMap_ptr != NULL);
1004 ASSERT(m_nackPCMap_ptr != NULL);
1005
1006 m_abortStarupDelay = -1;
1007 m_abortPerBlockDelay = -1;
1008 m_transWBs = 0;
1009 m_extraWBs = 0;
1010 m_transactionAborts = 0;
1011 m_transactionLogOverflows = 0;
1012 m_transactionCacheOverflows = 0;
1013 m_transactionUnsupInsts = 0;
1014 m_transactionSaveRestAborts = 0;
1015 m_inferredAborts = 0;
1016 m_xactNacked = 0;
1017
1018 m_xactLogs.clear();
1019 m_xactCycles.clear();
1020 m_xactReads.clear();
1021 m_xactWrites.clear();
1022 m_xactSizes.clear();
1023 m_abortDelays.clear();
1024 m_xactRetries.clear();
1025 m_xactOverflowReads.clear();
1026 m_xactOverflowWrites.clear();
1027 m_xactLoadMisses.clear();
1028 m_xactStoreMisses.clear();
1029 m_xactOverflowTotalReads.clear();
1030 m_xactOverflowTotalWrites.clear();
1031
1032 m_xactExceptionMap_ptr->clear();
1033 m_procsInXactMap_ptr->clear();
1034 m_abortIDMap_ptr->clear();
1035 m_commitIDMap_ptr->clear();
1036 m_xactRetryIDMap_ptr->clear();
1037 m_xactCyclesIDMap_ptr->clear();
1038 m_xactReadSetIDMap_ptr->clear();
1039 m_xactWriteSetIDMap_ptr->clear();
1040 m_xactLoadMissIDMap_ptr->clear();
1041 m_xactStoreMissIDMap_ptr->clear();
1042 m_xactInstrCountIDMap_ptr->clear();
1043 m_abortPCMap_ptr->clear();
1044 m_abortAddressMap_ptr->clear();
1045 m_nackXIDMap_ptr->clear();
1046 m_nackXIDPairMap_ptr->clear();
1047 m_nackPCMap_ptr->clear();
1048
1049 m_xactReadFilterBitsSetOnCommit->clear();
1050 m_xactReadFilterBitsSetOnAbort->clear();
1051 m_xactWriteFilterBitsSetOnCommit->clear();
1052 m_xactWriteFilterBitsSetOnAbort->clear();
1053
1054 m_readSetEmptyChecks = 0;
1055 m_readSetMatch = 0;
1056 m_readSetNoMatch = 0;
1057 m_writeSetEmptyChecks = 0;
1058 m_writeSetMatch = 0;
1059 m_writeSetNoMatch = 0;
1060
1061 m_xact_visualizer_last = 0;
1062 m_watchpointsFalsePositiveTrigger = 0;
1063 m_watchpointsTrueTrigger = 0;
1064 //---- end XACT_MEM code
1065
1066 // for MemoryControl:
1067 m_memReq = 0;
1068 m_memBankBusy = 0;
1069 m_memBusBusy = 0;
1070 m_memTfawBusy = 0;
1071 m_memReadWriteBusy = 0;
1072 m_memDataBusBusy = 0;
1073 m_memRefresh = 0;
1074 m_memRead = 0;
1075 m_memWrite = 0;
1076 m_memWaitCycles = 0;
1077 m_memInputQ = 0;
1078 m_memBankQ = 0;
1079 m_memArbWait = 0;
1080 m_memRandBusy = 0;
1081 m_memNotOld = 0;
1082
1083 for (int bank=0; bank < m_memBankCount.size(); bank++) {
1084 m_memBankCount[bank] = 0;
1085 }
1086
1087 // Flush the prefetches through the system - used so that there are no outstanding requests after stats are cleared
1088 //g_eventQueue_ptr->triggerAllEvents();
1089
1090 // update the start time
1091 m_ruby_start = g_eventQueue_ptr->getTime();
1092 }
1093
1094 void Profiler::addPrimaryStatSample(const CacheMsg& msg, NodeID id)
1095 {
1096 if (Protocol::m_TwoLevelCache) {
1097 if (msg.getType() == CacheRequestType_IFETCH) {
1098 addL1IStatSample(msg, id);
1099 } else {
1100 addL1DStatSample(msg, id);
1101 }
1102 // profile the address after an L1 miss (outside of the processor for CMP)
1103 if (Protocol::m_CMP) {
1104 addAddressTraceSample(msg, id);
1105 }
1106 } else {
1107 addL2StatSample(CacheRequestType_to_GenericRequestType(msg.getType()),
1108 msg.getAccessMode(), msg.getSize(), msg.getPrefetch(), id);
1109 addAddressTraceSample(msg, id);
1110 }
1111 }
1112
1113 void Profiler::profileConflictingRequests(const Address& addr)
1114 {
1115 assert(addr == line_address(addr));
1116 Time last_time = m_ruby_start;
1117 if (m_conflicting_map_ptr->exist(addr)) {
1118 Time last_time = m_conflicting_map_ptr->lookup(addr);
1119 }
1120 Time current_time = g_eventQueue_ptr->getTime();
1121 assert (current_time - last_time > 0);
1122 m_conflicting_histogram.add(current_time - last_time);
1123 m_conflicting_map_ptr->add(addr, current_time);
1124 }
1125
1126 void Profiler::addSecondaryStatSample(CacheRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id)
1127 {
1128 addSecondaryStatSample(CacheRequestType_to_GenericRequestType(requestType), type, msgSize, pfBit, id);
1129 }
1130
1131 void Profiler::addSecondaryStatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id)
1132 {
1133 addL2StatSample(requestType, type, msgSize, pfBit, id);
1134 }
1135
1136 void Profiler::addL2StatSample(GenericRequestType requestType, AccessModeType type, int msgSize, PrefetchBit pfBit, NodeID id)
1137 {
1138 m_perProcTotalMisses[id]++;
1139 if (type == AccessModeType_SupervisorMode) {
1140 m_perProcSupervisorMisses[id]++;
1141 } else {
1142 m_perProcUserMisses[id]++;
1143 }
1144 m_L2_cache_profiler_ptr->addStatSample(requestType, type, msgSize, pfBit);
1145 }
1146
1147 void Profiler::addL1DStatSample(const CacheMsg& msg, NodeID id)
1148 {
1149 m_L1D_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()),
1150 msg.getAccessMode(), msg.getSize(), msg.getPrefetch());
1151 }
1152
1153 void Profiler::addL1IStatSample(const CacheMsg& msg, NodeID id)
1154 {
1155 m_L1I_cache_profiler_ptr->addStatSample(CacheRequestType_to_GenericRequestType(msg.getType()),
1156 msg.getAccessMode(), msg.getSize(), msg.getPrefetch());
1157 }
1158
1159 void Profiler::addAddressTraceSample(const CacheMsg& msg, NodeID id)
1160 {
1161 if (msg.getType() != CacheRequestType_IFETCH) {
1162
1163 // Note: The following line should be commented out if you want to
1164 // use the special profiling that is part of the GS320 protocol
1165
1166 // NOTE: Unless PROFILE_HOT_LINES or PROFILE_ALL_INSTRUCTIONS are enabled, nothing will be profiled by the AddressProfiler
1167 m_address_profiler_ptr->addTraceSample(msg.getAddress(), msg.getProgramCounter(), msg.getType(), msg.getAccessMode(), id, false);
1168 }
1169 }
1170
1171 void Profiler::profileSharing(const Address& addr, AccessType type, NodeID requestor, const Set& sharers, const Set& owner)
1172 {
1173 Set set_contacted(owner);
1174 if (type == AccessType_Write) {
1175 set_contacted.addSet(sharers);
1176 }
1177 set_contacted.remove(requestor);
1178 int number_contacted = set_contacted.count();
1179
1180 if (type == AccessType_Write) {
1181 m_write_sharing_histogram.add(number_contacted);
1182 } else {
1183 m_read_sharing_histogram.add(number_contacted);
1184 }
1185 m_all_sharing_histogram.add(number_contacted);
1186
1187 if (number_contacted == 0) {
1188 m_memory_to_cache++;
1189 } else {
1190 m_cache_to_cache++;
1191 }
1192
1193 }
1194
1195 void Profiler::profileMsgDelay(int virtualNetwork, int delayCycles) {
1196 assert(virtualNetwork < m_delayedCyclesVCHistograms.size());
1197 m_delayedCyclesHistogram.add(delayCycles);
1198 m_delayedCyclesVCHistograms[virtualNetwork].add(delayCycles);
1199 if (virtualNetwork != 0) {
1200 m_delayedCyclesNonPFHistogram.add(delayCycles);
1201 }
1202 }
1203
1204 // profiles original cache requests including PUTs
1205 void Profiler::profileRequest(const string& requestStr)
1206 {
1207 m_requests++;
1208
1209 if (m_requestProfileMap_ptr->exist(requestStr)) {
1210 (m_requestProfileMap_ptr->lookup(requestStr))++;
1211 } else {
1212 m_requestProfileMap_ptr->add(requestStr, 1);
1213 }
1214 }
1215
1216 void Profiler::recordPrediction(bool wasGood, bool wasPredicted)
1217 {
1218 m_predictionOpportunities++;
1219 if(wasPredicted){
1220 m_predictions++;
1221 if(wasGood){
1222 m_goodPredictions++;
1223 }
1224 }
1225 }
1226
1227 void Profiler::profileFilterAction(int action)
1228 {
1229 m_filter_action_histogram.add(action);
1230 }
1231
1232 void Profiler::profileMulticastRetry(const Address& addr, int count)
1233 {
1234 m_multicast_retry_histogram.add(count);
1235 }
1236
1237 void Profiler::startTransaction(int cpu)
1238 {
1239 m_perProcStartTransaction[cpu]++;
1240 }
1241
1242 void Profiler::endTransaction(int cpu)
1243 {
1244 m_perProcEndTransaction[cpu]++;
1245 }
1246
1247 void Profiler::controllerBusy(MachineID machID)
1248 {
1249 m_busyControllerCount[(int)machID.type][(int)machID.num]++;
1250 }
1251
1252 void Profiler::profilePFWait(Time waitTime)
1253 {
1254 m_prefetchWaitHistogram.add(waitTime);
1255 }
1256
1257 void Profiler::bankBusy()
1258 {
1259 m_busyBankCount++;
1260 }
1261
1262 // non-zero cycle demand request
1263 void Profiler::missLatency(Time t, CacheRequestType type, GenericMachineType respondingMach)
1264 {
1265 m_allMissLatencyHistogram.add(t);
1266 m_missLatencyHistograms[type].add(t);
1267 m_machLatencyHistograms[respondingMach].add(t);
1268 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) {
1269 m_L2MissLatencyHistogram.add(t);
1270 }
1271 }
1272
1273 // non-zero cycle prefetch request
1274 void Profiler::swPrefetchLatency(Time t, CacheRequestType type, GenericMachineType respondingMach)
1275 {
1276 m_allSWPrefetchLatencyHistogram.add(t);
1277 m_SWPrefetchLatencyHistograms[type].add(t);
1278 m_SWPrefetchMachLatencyHistograms[respondingMach].add(t);
1279 if(respondingMach == GenericMachineType_Directory || respondingMach == GenericMachineType_NUM) {
1280 m_SWPrefetchL2MissLatencyHistogram.add(t);
1281 }
1282 }
1283
1284 void Profiler::profileTransition(const string& component, NodeID id, NodeID version, Address addr,
1285 const string& state, const string& event,
1286 const string& next_state, const string& note)
1287 {
1288 const int EVENT_SPACES = 20;
1289 const int ID_SPACES = 3;
1290 const int TIME_SPACES = 7;
1291 const int COMP_SPACES = 10;
1292 const int STATE_SPACES = 6;
1293
1294 if ((g_debug_ptr->getDebugTime() > 0) &&
1295 (g_eventQueue_ptr->getTime() >= g_debug_ptr->getDebugTime())) {
1296 (* debug_cout_ptr).flags(ios::right);
1297 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1298 (* debug_cout_ptr) << setw(ID_SPACES) << id << " ";
1299 (* debug_cout_ptr) << setw(ID_SPACES) << version << " ";
1300 (* debug_cout_ptr) << setw(COMP_SPACES) << component;
1301 (* debug_cout_ptr) << setw(EVENT_SPACES) << event << " ";
1302 for (int i=0; i < RubyConfig::numberOfProcessors(); i++) {
1303
1304 if (i == id) {
1305 (* debug_cout_ptr).flags(ios::right);
1306 (* debug_cout_ptr) << setw(STATE_SPACES) << state;
1307 (* debug_cout_ptr) << ">";
1308 (* debug_cout_ptr).flags(ios::left);
1309 (* debug_cout_ptr) << setw(STATE_SPACES) << next_state;
1310 } else {
1311 // cout << setw(STATE_SPACES) << " " << " " << setw(STATE_SPACES) << " ";
1312 }
1313 }
1314 (* debug_cout_ptr) << " " << addr << " " << note;
1315
1316 (* debug_cout_ptr) << endl;
1317 }
1318 }
1319
1320 // Helper function
1321 static double process_memory_total()
1322 {
1323 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB,
1324 ifstream proc_file;
1325 proc_file.open("/proc/self/statm");
1326 int total_size_in_pages = 0;
1327 int res_size_in_pages = 0;
1328 proc_file >> total_size_in_pages;
1329 proc_file >> res_size_in_pages;
1330 return double(total_size_in_pages)*MULTIPLIER; // size in megabytes
1331 }
1332
1333 static double process_memory_resident()
1334 {
1335 const double MULTIPLIER = 4096.0/(1024.0*1024.0); // 4kB page size, 1024*1024 bytes per MB,
1336 ifstream proc_file;
1337 proc_file.open("/proc/self/statm");
1338 int total_size_in_pages = 0;
1339 int res_size_in_pages = 0;
1340 proc_file >> total_size_in_pages;
1341 proc_file >> res_size_in_pages;
1342 return double(res_size_in_pages)*MULTIPLIER; // size in megabytes
1343 }
1344
1345 void Profiler::profileGetXMaskPrediction(const Set& pred_set)
1346 {
1347 m_getx_mask_prediction.add(pred_set.count());
1348 }
1349
1350 void Profiler::profileGetSMaskPrediction(const Set& pred_set)
1351 {
1352 m_gets_mask_prediction.add(pred_set.count());
1353 }
1354
1355 void Profiler::profileTrainingMask(const Set& pred_set)
1356 {
1357 m_explicit_training_mask.add(pred_set.count());
1358 }
1359
1360 int64 Profiler::getTotalInstructionsExecuted() const
1361 {
1362 int64 sum = 1; // Starting at 1 allows us to avoid division by zero
1363 for(int i=0; i < RubyConfig::numberOfProcessors(); i++) {
1364 sum += (g_system_ptr->getDriver()->getInstructionCount(i) - m_instructions_executed_at_start[i]);
1365 }
1366 return sum;
1367 }
1368
1369 int64 Profiler::getTotalTransactionsExecuted() const
1370 {
1371 int64 sum = m_perProcEndTransaction.sum();
1372 if (sum > 0) {
1373 return sum;
1374 } else {
1375 return 1; // Avoid division by zero errors
1376 }
1377 }
1378
1379
1380 // The following case statement converts CacheRequestTypes to GenericRequestTypes
1381 // allowing all profiling to be done with a single enum type instead of slow strings
1382 GenericRequestType Profiler::CacheRequestType_to_GenericRequestType(const CacheRequestType& type) {
1383 switch (type) {
1384 case CacheRequestType_LD:
1385 return GenericRequestType_LD;
1386 break;
1387 case CacheRequestType_ST:
1388 return GenericRequestType_ST;
1389 break;
1390 case CacheRequestType_ATOMIC:
1391 return GenericRequestType_ATOMIC;
1392 break;
1393 case CacheRequestType_IFETCH:
1394 return GenericRequestType_IFETCH;
1395 break;
1396 case CacheRequestType_LD_XACT:
1397 return GenericRequestType_LD_XACT;
1398 break;
1399 case CacheRequestType_LDX_XACT:
1400 return GenericRequestType_LDX_XACT;
1401 break;
1402 case CacheRequestType_ST_XACT:
1403 return GenericRequestType_ST_XACT;
1404 break;
1405 case CacheRequestType_NULL:
1406 return GenericRequestType_NULL;
1407 break;
1408 default:
1409 ERROR_MSG("Unexpected cache request type");
1410 }
1411 }
1412
1413 //---- begin Transactional Memory CODE
1414 void Profiler::profileTransaction(int size, int logSize, int readS, int writeS, int overflow_readS, int overflow_writeS, int retries, int useful_cycles, bool nacked, int loadMisses, int storeMisses, int instrCount, int xid){
1415 m_xactLogs.add(logSize);
1416 m_xactSizes.add(size);
1417 m_xactReads.add(readS);
1418 m_xactWrites.add(writeS);
1419 m_xactRetries.add(retries);
1420 m_xactCycles.add(useful_cycles);
1421 m_xactLoadMisses.add(loadMisses);
1422 m_xactStoreMisses.add(storeMisses);
1423 m_xactInstrCount.add(instrCount);
1424
1425 // was this transaction nacked?
1426 if(nacked){
1427 m_xactNacked++;
1428 }
1429
1430 // for overflowed transactions
1431 if(overflow_readS > 0 || overflow_writeS > 0){
1432 m_xactOverflowReads.add(overflow_readS);
1433 m_xactOverflowWrites.add(overflow_writeS);
1434 m_xactOverflowTotalReads.add(readS);
1435 m_xactOverflowTotalWrites.add(writeS);
1436 }
1437
1438 // Record commits by xid
1439 if(!m_commitIDMap_ptr->exist(xid)){
1440 m_commitIDMap_ptr->add(xid, 1);
1441 m_xactRetryIDMap_ptr->add(xid, retries);
1442 m_xactCyclesIDMap_ptr->add(xid, useful_cycles);
1443 m_xactReadSetIDMap_ptr->add(xid, readS);
1444 m_xactWriteSetIDMap_ptr->add(xid, writeS);
1445 m_xactLoadMissIDMap_ptr->add(xid, loadMisses);
1446 m_xactStoreMissIDMap_ptr->add(xid, storeMisses);
1447 m_xactInstrCountIDMap_ptr->add(xid, instrCount);
1448 } else {
1449 (m_commitIDMap_ptr->lookup(xid))++;
1450 (m_xactRetryIDMap_ptr->lookup(xid)) += retries;
1451 (m_xactCyclesIDMap_ptr->lookup(xid)) += useful_cycles;
1452 (m_xactReadSetIDMap_ptr->lookup(xid)) += readS;
1453 (m_xactWriteSetIDMap_ptr->lookup(xid)) += writeS;
1454 (m_xactLoadMissIDMap_ptr->lookup(xid)) += loadMisses;
1455 (m_xactStoreMissIDMap_ptr->lookup(xid)) += storeMisses;
1456 (m_xactInstrCountIDMap_ptr->lookup(xid)) += instrCount;
1457 }
1458 }
1459
1460 void Profiler::profileBeginTransaction(NodeID id, int tid, int xid, int thread, Address pc, bool isOpen){
1461 //- if(PROFILE_XACT){
1462 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 2)){
1463 const char* openStr = isOpen ? " OPEN" : " CLOSED";
1464 const int ID_SPACES = 3;
1465 const int TIME_SPACES = 7;
1466 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1467 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1468 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1469 // The actual processor number
1470 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1471 (* debug_cout_ptr).flags(ios::right);
1472 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1473 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1474 << " XACT BEGIN " << xid
1475 << " PC 0x" << hex << pc.getAddress()
1476 << dec
1477 << " *PC 0x" << hex << myInst << dec
1478 << " '" << myInstStr << "'"
1479 << openStr
1480 << endl;
1481 }
1482 }
1483
1484 void Profiler::profileCommitTransaction(NodeID id, int tid, int xid, int thread, Address pc, bool isOpen){
1485 //- if(PROFILE_XACT){
1486 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 2)){
1487 const char* openStr = isOpen ? " OPEN" : " CLOSED";
1488 const int ID_SPACES = 3;
1489 const int TIME_SPACES = 7;
1490 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1491 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1492 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1493 // The actual processor number
1494 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1495 (* debug_cout_ptr).flags(ios::right);
1496 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1497 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1498 << " XACT COMMIT " << xid
1499 << " PC 0x" << hex << pc.getAddress()
1500 << dec
1501 << " *PC 0x" << hex << myInst << dec
1502 << " '" << myInstStr << "'"
1503 << openStr
1504 << endl;
1505 }
1506
1507 }
1508
1509 // for profiling overflows
1510 void Profiler::profileLoadOverflow(NodeID id, int tid, int xid, int thread, Address addr, bool l1_overflow){
1511 //- if(PROFILE_XACT){
1512 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1513 const int ID_SPACES = 3;
1514 const int TIME_SPACES = 7;
1515 string overflow_str = " XACT LOAD L1 OVERFLOW ";
1516 if(!l1_overflow){
1517 overflow_str = " XACT LOAD L2 OVERFLOW ";
1518 }
1519 // The actual processor number
1520 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1521 (* debug_cout_ptr).flags(ios::right);
1522 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1523 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1524 << overflow_str << xid
1525 << " ADDR " << addr
1526 << endl;
1527 }
1528 }
1529
1530 // for profiling overflows
1531 void Profiler::profileStoreOverflow(NodeID id, int tid, int xid, int thread, Address addr, bool l1_overflow){
1532 //- if(PROFILE_XACT){
1533 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1534 const int ID_SPACES = 3;
1535 const int TIME_SPACES = 7;
1536 string overflow_str = " XACT STORE L1 OVERFLOW ";
1537 if(!l1_overflow){
1538 overflow_str = " XACT STORE L2 OVERFLOW ";
1539 }
1540 // The actual processor number
1541 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1542 (* debug_cout_ptr).flags(ios::right);
1543 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1544 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1545 << overflow_str << xid
1546 << " ADDR " << addr
1547 << endl;
1548 }
1549 }
1550
1551 void Profiler::profileLoadTransaction(NodeID id, int tid, int xid, int thread, Address addr, Address logicalAddress, Address pc){
1552 //- if(PROFILE_XACT){
1553 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 3)){
1554 const int ID_SPACES = 3;
1555 const int TIME_SPACES = 7;
1556 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1557 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1558 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1559 // The actual processor number
1560 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1561 (* debug_cout_ptr).flags(ios::right);
1562 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1563 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1564 << " XACT LOAD " << xid
1565 << " " << addr
1566 << " VA " << logicalAddress
1567 << " PC " << pc
1568 << " *PC 0x" << hex << myInst << dec
1569 << " '" << myInstStr << "'"
1570 //<< " VAL 0x" << hex << SIMICS_read_physical_memory(proc_no, SIMICS_translate_data_address(proc_no, logicalAddress), 4) << dec
1571 << " VAL 0x" << hex << g_system_ptr->getDriver()->readPhysicalMemory(proc_no, addr.getAddress(), 4) << dec
1572 << endl;
1573 }
1574 }
1575
1576 void Profiler::profileLoad(NodeID id, int tid, int xid, int thread, Address addr, Address logicalAddress, Address pc){
1577 if(PROFILE_NONXACT){
1578 const int ID_SPACES = 3;
1579 const int TIME_SPACES = 7;
1580 // The actual processor number
1581 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1582 (* debug_cout_ptr).flags(ios::right);
1583 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1584 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1585 << " LOAD " << xid
1586 << " " << addr
1587 << " VA " << logicalAddress
1588 << " PC " << pc
1589 //<< " VAL 0x" << hex << SIMICS_read_physical_memory(proc_no, SIMICS_translate_data_address(proc_no, logicalAddress), 4) << dec
1590 << " VAL 0x" << hex << g_system_ptr->getDriver()->readPhysicalMemory(proc_no, addr.getAddress(), 4) << dec
1591 << endl;
1592 }
1593 }
1594
1595 void Profiler::profileStoreTransaction(NodeID id, int tid, int xid, int thread, Address addr, Address logicalAddress, Address pc){
1596 //- if(PROFILE_XACT){
1597 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 3)){
1598 const int ID_SPACES = 3;
1599 const int TIME_SPACES = 7;
1600 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1601 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1602 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1603 // The actual processor number
1604 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1605 (* debug_cout_ptr).flags(ios::right);
1606 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1607 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1608 << " XACT STORE " << xid
1609 << " " << addr
1610 << " VA " << logicalAddress
1611 << " PC " << pc
1612 << " *PC 0x" << hex << myInst << dec
1613 << " '" << myInstStr << "'"
1614 << endl;
1615 }
1616 }
1617
1618 void Profiler::profileStore(NodeID id, int tid, int xid, int thread, Address addr, Address logicalAddress, Address pc){
1619 if(PROFILE_NONXACT){
1620 const int ID_SPACES = 3;
1621 const int TIME_SPACES = 7;
1622 // The actual processor number
1623 int proc_no = id*RubyConfig::numberofSMTThreads() + thread;
1624 (* debug_cout_ptr).flags(ios::right);
1625 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1626 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1627 << " STORE " << xid
1628 << " " << addr
1629 << " VA " << logicalAddress
1630 << " PC " << pc
1631 << endl;
1632 }
1633 }
1634
1635 void Profiler::profileNack(NodeID id, int tid, int xid, int thread, int nacking_thread, NodeID nackedBy, Address addr, Address logicalAddress, Address pc, uint64 seq_ts, uint64 nack_ts, bool possibleCycle){
1636 int nid = 0; // g_system_ptr->getChip(nackedBy/RubyConfig::numberOfProcsPerChip())->getTransactionInterfaceManager(nackedBy%RubyConfig::numberOfProcsPerChip())->getXID(nacking_thread);
1637 assert(0);
1638 //- if(PROFILE_XACT){
1639 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1640 const int ID_SPACES = 3;
1641 const int TIME_SPACES = 7;
1642 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1643 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1644 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1645 // The actual processor number
1646 int proc_no = id*g_NUM_SMT_THREADS + thread;
1647 int nack_proc_no = nackedBy*g_NUM_SMT_THREADS + nacking_thread;
1648 Address nack_pc = SIMICS_get_program_counter(nack_proc_no);
1649 (* debug_cout_ptr).flags(ios::right);
1650 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1651 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1652 << " XACT NACK " << xid
1653 << " by " << nack_proc_no
1654 << " [ " << nackedBy
1655 << ", " << nacking_thread
1656 << " ]"
1657 << " NID: " << nid
1658 << " " << addr
1659 << " VA " << logicalAddress
1660 << " PC " << pc
1661 << " *PC 0x" << hex << myInst << dec
1662 << " '" << myInstStr << "'"
1663 << " NackerPC " << nack_pc
1664 << " my_ts " << seq_ts
1665 << " nack_ts " << nack_ts
1666 << " possible_cycle " << possibleCycle
1667 << endl;
1668 }
1669
1670 // Record nacks by xid
1671 if(!m_nackXIDMap_ptr->exist(xid)){
1672 m_nackXIDMap_ptr->add(xid, 1);
1673 } else {
1674 (m_nackXIDMap_ptr->lookup(xid))++;
1675 }
1676
1677 // Record nack ID pairs by xid
1678 if(!m_nackXIDPairMap_ptr->exist(xid)){
1679 Map<int, int> * new_map = new Map<int, int>;
1680 new_map->add(nid, 1);
1681 m_nackXIDPairMap_ptr->add(xid, new_map);
1682 }
1683 else{
1684 // retrieve existing map
1685 Map<int, int> * my_map = m_nackXIDPairMap_ptr->lookup(xid);
1686 if(!my_map->exist(nid)){
1687 my_map->add(nid, 1);
1688 }
1689 else{
1690 (my_map->lookup(nid))++;
1691 }
1692 }
1693
1694 // Record nacks by pc
1695 if(!m_nackPCMap_ptr->exist(pc)){
1696 m_nackPCMap_ptr->add(pc, 1);
1697 } else {
1698 (m_nackPCMap_ptr->lookup(pc))++;
1699 }
1700 }
1701
1702 void Profiler::profileExposedConflict(NodeID id, int xid, int thread, Address addr, Address pc){
1703 //if(PROFILE_XACT){
1704 const int ID_SPACES = 3;
1705 const int TIME_SPACES = 7;
1706 // The actual processor number
1707 int proc_no = id*g_NUM_SMT_THREADS + thread;
1708 (* debug_cout_ptr).flags(ios::right);
1709 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1710 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " "
1711 << " EXPOSED ACTION CONFLICT " << xid
1712 << " ADDR " << addr
1713 << " PC " << pc
1714 << endl;
1715 //}
1716 }
1717
1718 void Profiler::profileInferredAbort(){
1719 m_inferredAborts++;
1720 }
1721
1722 void Profiler::profileAbortDelayConstants(int startupDelay, int perBlock){
1723 m_abortStarupDelay = startupDelay;
1724 m_abortPerBlockDelay = perBlock;
1725 }
1726
1727 void Profiler::profileAbortTransaction(NodeID id, int tid, int xid, int thread, int delay, int abortingThread, int abortingProc, Address addr, Address pc){
1728 const int ID_SPACES = 3;
1729 const int TIME_SPACES = 7;
1730 int abortingXID = -1;
1731 // The actual processor number
1732 int proc_no = id*g_NUM_SMT_THREADS + thread;
1733 // we are passed in physical proc number. Compute logical abort proc_no
1734 int logical_abort_proc_no = abortingProc/g_NUM_SMT_THREADS;
1735 if(abortingProc >= 0){
1736 AbstractChip * c = g_system_ptr->getChip(logical_abort_proc_no/RubyConfig::numberOfProcsPerChip());
1737 abortingXID = 0; // c->getTransactionInterfaceManager(logical_abort_proc_no%RubyConfig::numberOfProcsPerChip())->getXID(abortingThread);
1738 assert(0);
1739 }
1740 //- if(PROFILE_XACT){
1741 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1742 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1743 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1744 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1745 (* debug_cout_ptr).flags(ios::right);
1746 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1747 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << "]" << " TID " << tid
1748 << " XACT ABORT " << xid
1749 << " caused by " << abortingProc
1750 << " [ " << logical_abort_proc_no
1751 << ", " << abortingThread
1752 << " ]"
1753 << " xid: " << abortingXID << " "
1754 << " address: " << addr
1755 << " delay: " << delay
1756 << " PC " << pc
1757 << " *PC 0x" << hex << myInst << dec
1758 << " '" << myInstStr << "'"
1759 << endl;
1760 }
1761 m_transactionAborts++;
1762
1763 // Record aborts by xid
1764 if(!m_abortIDMap_ptr->exist(xid)){
1765 m_abortIDMap_ptr->add(xid, 1);
1766 } else {
1767 (m_abortIDMap_ptr->lookup(xid))++;
1768 }
1769 m_abortDelays.add(delay);
1770
1771 // Record aborts by pc
1772 if(!m_abortPCMap_ptr->exist(pc)){
1773 m_abortPCMap_ptr->add(pc, 1);
1774 } else {
1775 (m_abortPCMap_ptr->lookup(pc))++;
1776 }
1777
1778 // Record aborts by address
1779 if(!m_abortAddressMap_ptr->exist(addr)){
1780 m_abortAddressMap_ptr->add(addr, 1);
1781 } else {
1782 (m_abortAddressMap_ptr->lookup(addr))++;
1783 }
1784 }
1785
1786 void Profiler::profileTransWB(){
1787 m_transWBs++;
1788 }
1789
1790 void Profiler::profileExtraWB(){
1791 m_extraWBs++;
1792 }
1793
1794 void Profiler::profileXactChange(int procs, int cycles){
1795 if(!m_procsInXactMap_ptr->exist(procs)){
1796 m_procsInXactMap_ptr->add(procs, cycles);
1797 } else {
1798 (m_procsInXactMap_ptr->lookup(procs)) += cycles;
1799 }
1800 }
1801
1802 void Profiler::profileReadSet(Address addr, bool bf_filter_result, bool perfect_filter_result, NodeID id, int thread){
1803 // do NOT count instances when signature is empty!
1804 if(!bf_filter_result && !perfect_filter_result){
1805 m_readSetEmptyChecks++;
1806 return;
1807 }
1808
1809 if(bf_filter_result != perfect_filter_result){
1810 m_readSetNoMatch++;
1811 /*
1812 // we have a false positive
1813 if(!m_readSetNoMatch_ptr->exist(addr)){
1814 m_readSetNoMatch_ptr->add(addr, 1);
1815 }
1816 else{
1817 (m_readSetNoMatch_ptr->lookup(addr))++;
1818 }
1819 */
1820 }
1821 else{
1822 m_readSetMatch++;
1823 /*
1824 // Bloom filter agrees with perfect filter
1825 if(!m_readSetMatch_ptr->exist(addr)){
1826 m_readSetMatch_ptr->add(addr, 1);
1827 }
1828 else{
1829 (m_readSetMatch_ptr->lookup(addr))++;
1830 }
1831 */
1832 }
1833 }
1834
1835
1836 void Profiler::profileRemoteReadSet(Address addr, bool bf_filter_result, bool perfect_filter_result, NodeID id, int thread){
1837 if(bf_filter_result != perfect_filter_result){
1838 // we have a false positive
1839 if(!m_remoteReadSetNoMatch_ptr->exist(addr)){
1840 m_remoteReadSetNoMatch_ptr->add(addr, 1);
1841 }
1842 else{
1843 (m_remoteReadSetNoMatch_ptr->lookup(addr))++;
1844 }
1845 }
1846 else{
1847 // Bloom filter agrees with perfect filter
1848 if(!m_remoteReadSetMatch_ptr->exist(addr)){
1849 m_remoteReadSetMatch_ptr->add(addr, 1);
1850 }
1851 else{
1852 (m_remoteReadSetMatch_ptr->lookup(addr))++;
1853 }
1854 }
1855 }
1856
1857 void Profiler::profileWriteSet(Address addr, bool bf_filter_result, bool perfect_filter_result, NodeID id, int thread){
1858 // do NOT count instances when signature is empty!
1859 if(!bf_filter_result && !perfect_filter_result){
1860 m_writeSetEmptyChecks++;
1861 return;
1862 }
1863
1864 if(bf_filter_result != perfect_filter_result){
1865 m_writeSetNoMatch++;
1866 /*
1867 // we have a false positive
1868 if(!m_writeSetNoMatch_ptr->exist(addr)){
1869 m_writeSetNoMatch_ptr->add(addr, 1);
1870 }
1871 else{
1872 (m_writeSetNoMatch_ptr->lookup(addr))++;
1873 }
1874 */
1875 }
1876 else{
1877 m_writeSetMatch++;
1878 /*
1879 // Bloom filter agrees with perfect filter
1880 if(!m_writeSetMatch_ptr->exist(addr)){
1881 m_writeSetMatch_ptr->add(addr, 1);
1882 }
1883 else{
1884 (m_writeSetMatch_ptr->lookup(addr))++;
1885 }
1886 */
1887 }
1888 }
1889
1890
1891 void Profiler::profileRemoteWriteSet(Address addr, bool bf_filter_result, bool perfect_filter_result, NodeID id, int thread){
1892 if(bf_filter_result != perfect_filter_result){
1893 // we have a false positive
1894 if(!m_remoteWriteSetNoMatch_ptr->exist(addr)){
1895 m_remoteWriteSetNoMatch_ptr->add(addr, 1);
1896 }
1897 else{
1898 (m_remoteWriteSetNoMatch_ptr->lookup(addr))++;
1899 }
1900 }
1901 else{
1902 // Bloom filter agrees with perfect filter
1903 if(!m_remoteWriteSetMatch_ptr->exist(addr)){
1904 m_remoteWriteSetMatch_ptr->add(addr, 1);
1905 }
1906 else{
1907 (m_remoteWriteSetMatch_ptr->lookup(addr))++;
1908 }
1909 }
1910 }
1911
1912 void Profiler::profileTransactionLogOverflow(NodeID id, Address addr, Address pc){
1913 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1914 const int ID_SPACES = 3;
1915 const int TIME_SPACES = 7;
1916 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1917 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1918 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1919 (* debug_cout_ptr).flags(ios::right);
1920 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1921 (* debug_cout_ptr) << setw(ID_SPACES) << id << " "
1922 << " XACT LOG OVERFLOW"
1923 << " ADDR " << addr
1924 << " PC " << pc
1925 << " *PC 0x" << hex << myInst << dec
1926 << " '" << myInstStr << "'"
1927 << endl;
1928
1929 }
1930 m_transactionLogOverflows++;
1931 }
1932
1933 void Profiler::profileTransactionCacheOverflow(NodeID id, Address addr, Address pc){
1934 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1935 const int ID_SPACES = 3;
1936 const int TIME_SPACES = 7;
1937 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1938 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1939 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1940 (* debug_cout_ptr).flags(ios::right);
1941 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1942 (* debug_cout_ptr) << setw(ID_SPACES) << id << " "
1943 << " XACT CACHE OVERFLOW "
1944 << " ADDR " << addr
1945 << " PC " << pc
1946 << " *PC 0x" << hex << myInst << dec
1947 << " '" << myInstStr << "'"
1948 << endl;
1949
1950 }
1951 m_transactionCacheOverflows++;
1952 }
1953
1954 void Profiler::profileGetCPS(NodeID id, uint32 cps, Address pc){
1955 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
1956 const int ID_SPACES = 3;
1957 const int TIME_SPACES = 7;
1958 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
1959 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
1960 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
1961
1962 (* debug_cout_ptr).flags(ios::right);
1963 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
1964 (* debug_cout_ptr) << setw(ID_SPACES) << id << " "
1965 << " XACT GET CPS"
1966 << " PC " << pc
1967 << " *PC 0x" << hex << myInst << dec
1968 << " '" << myInstStr << "'"
1969 << " CPS 0x" << hex << cps << dec
1970 << endl;
1971 }
1972 }
1973 //---- end Transactional Memory CODE
1974
1975
1976 void Profiler::profileExceptionStart(bool xact, NodeID id, int thread, int val, int trap_level, uinteger_t pc, uinteger_t npc){
1977 if(xact){
1978 if(!m_xactExceptionMap_ptr->exist(val)){
1979 m_xactExceptionMap_ptr->add(val, 1);
1980 } else {
1981 (m_xactExceptionMap_ptr->lookup(val))++;
1982 }
1983 }
1984
1985 if (!xact && !PROFILE_NONXACT) return;
1986
1987 if(PROFILE_EXCEPTIONS){
1988 const int ID_SPACES = 3;
1989 const int TIME_SPACES = 7;
1990 // The actual processor number
1991 int proc_no = id*g_NUM_SMT_THREADS + thread;
1992
1993 // get the excepting instruction
1994 const char * instruction;
1995 physical_address_t addr = SIMICS_translate_address( proc_no, Address(pc));
1996 if(val != 0x64 && addr != 0x0){
1997 // ignore instruction TLB miss
1998 instruction = SIMICS_disassemble_physical( proc_no, addr );
1999 }
2000
2001 (* debug_cout_ptr).flags(ios::right);
2002 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2003 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << " ]" << " ";
2004 if (xact)
2005 (* debug_cout_ptr) << " XACT Exception(";
2006 else
2007 (* debug_cout_ptr) << " Exception(";
2008
2009 (* debug_cout_ptr) << hex << val << dec << ")_START--Trap Level " << trap_level
2010 << "--(PC=0x" << hex << pc << ", " << npc << ")"
2011 << dec;
2012
2013 if(val != 0x64 && addr != 0x0){
2014 (* debug_cout_ptr) << " instruction = " << instruction;
2015 }
2016 else{
2017 (* debug_cout_ptr) << " instruction = INSTRUCTION TLB MISS";
2018 }
2019 (* debug_cout_ptr) << dec << endl;
2020 }
2021 }
2022
2023 void Profiler::profileExceptionDone(bool xact, NodeID id, int thread, int val, int trap_level, uinteger_t pc, uinteger_t npc, uinteger_t tpc, uinteger_t tnpc){
2024 if (!xact && !PROFILE_NONXACT) return;
2025
2026 if (PROFILE_EXCEPTIONS){
2027 const int ID_SPACES = 3;
2028 const int TIME_SPACES = 7;
2029 // The actual processor number
2030 int proc_no = id*g_NUM_SMT_THREADS + thread;
2031
2032 // get the excepting instruction
2033 const char * instruction;
2034 instruction = SIMICS_disassemble_physical( proc_no, SIMICS_translate_address( proc_no, Address(pc) ) );
2035
2036
2037 (* debug_cout_ptr).flags(ios::right);
2038 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2039 (* debug_cout_ptr) << setw(ID_SPACES) << proc_no << " [" << id << "," << thread << " ]" << " ";
2040 if (xact)
2041 (* debug_cout_ptr) << " XACT Exception(";
2042 else
2043 (* debug_cout_ptr) << " Exception(";
2044
2045 (* debug_cout_ptr) << hex << val << dec << ")_DONE--Trap Level " << trap_level
2046 << "--(PC=0x" << hex << pc << ", " << npc << dec << ")"
2047 << "--(TPC=0x" << hex << tpc << ", " << tnpc << dec << ")"
2048 << endl;
2049 }
2050 }
2051
2052 void Profiler::rubyWatch(int id){
2053 int rn_g1 = SIMICS_get_register_number(id, "g1");
2054 uint64 tr = SIMICS_read_register(id, rn_g1);
2055 Address watch_address = Address(tr);
2056 const int ID_SPACES = 3;
2057 const int TIME_SPACES = 7;
2058
2059 (* debug_cout_ptr).flags(ios::right);
2060 (* debug_cout_ptr) << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2061 (* debug_cout_ptr) << setw(ID_SPACES) << id << " "
2062 << "RUBY WATCH "
2063 << watch_address
2064 << endl;
2065
2066 if(!m_watch_address_list_ptr->exist(watch_address)){
2067 m_watch_address_list_ptr->add(watch_address, 1);
2068 }
2069 }
2070
2071 bool Profiler::watchAddress(Address addr){
2072 if (m_watch_address_list_ptr->exist(addr))
2073 return true;
2074 else
2075 return false;
2076 }
2077
2078 void Profiler::profileReadFilterBitsSet(int xid, int bits, bool isCommit) {
2079 if (isCommit) {
2080 if(!m_xactReadFilterBitsSetOnCommit->exist(xid)){
2081 Histogram hist;
2082 hist.add(bits);
2083 m_xactReadFilterBitsSetOnCommit->add(xid, hist);
2084 }
2085 else{
2086 (m_xactReadFilterBitsSetOnCommit->lookup(xid)).add(bits);
2087 }
2088 } else {
2089 if(!m_xactReadFilterBitsSetOnAbort->exist(xid)){
2090 Histogram hist;
2091 hist.add(bits);
2092 m_xactReadFilterBitsSetOnAbort->add(xid, hist);
2093 }
2094 else{
2095 (m_xactReadFilterBitsSetOnAbort->lookup(xid)).add(bits);
2096 }
2097 }
2098 }
2099
2100 void Profiler::profileWriteFilterBitsSet(int xid, int bits, bool isCommit) {
2101 if (isCommit) {
2102 if(!m_xactWriteFilterBitsSetOnCommit->exist(xid)){
2103 Histogram hist;
2104 hist.add(bits);
2105 m_xactWriteFilterBitsSetOnCommit->add(xid, hist);
2106 }
2107 else{
2108 (m_xactWriteFilterBitsSetOnCommit->lookup(xid)).add(bits);
2109 }
2110 } else {
2111 if(!m_xactWriteFilterBitsSetOnAbort->exist(xid)){
2112 Histogram hist;
2113 hist.add(bits);
2114 m_xactWriteFilterBitsSetOnAbort->add(xid, hist);
2115 }
2116 else{
2117 (m_xactWriteFilterBitsSetOnAbort->lookup(xid)).add(bits);
2118 }
2119 }
2120 }
2121 /*
2122 //gem5:Arka for decomissioning log_tm
2123
2124 void Profiler::setXactVisualizerFile(char * filename){
2125 if ( (filename == NULL) ||
2126 (!strcmp(filename, "none")) ) {
2127 m_xact_visualizer_ptr = &cout;
2128 return;
2129 }
2130
2131 if (m_xact_visualizer.is_open() ) {
2132 m_xact_visualizer.close ();
2133 }
2134 m_xact_visualizer.open (filename, std::ios::out);
2135 if (! m_xact_visualizer.is_open() ) {
2136 cerr << "setXactVisualizer: can't open file " << filename << endl;
2137 }
2138 else {
2139 m_xact_visualizer_ptr = &m_xact_visualizer;
2140 }
2141 cout << "setXactVisualizer file " << filename << endl;
2142 }
2143
2144 void Profiler::printTransactionState(bool can_skip){
2145 if (!XACT_VISUALIZER) return;
2146 int num_processors = RubyConfig::numberOfProcessors() * RubyConfig::numberofSMTThreads();
2147
2148 if (!g_system_ptr->getXactVisualizer()->existXactActivity() && can_skip)
2149 return;
2150
2151 if (can_skip && ((g_eventQueue_ptr->getTime()/10000) <= m_xact_visualizer_last))
2152 return;
2153
2154 Vector<char> xactStateVector = g_system_ptr->getXactVisualizer()->getTransactionStateVector();
2155 for (int i = 0 ; i < num_processors; i++){
2156 (* m_xact_visualizer_ptr) << xactStateVector[i] << " ";
2157 }
2158 (* m_xact_visualizer_ptr) << " " << g_eventQueue_ptr->getTime() << endl;
2159 m_xact_visualizer_last = g_eventQueue_ptr->getTime() / 10000;
2160 }
2161 */
2162 void Profiler::watchpointsFalsePositiveTrigger()
2163 {
2164 m_watchpointsFalsePositiveTrigger++;
2165 }
2166
2167 void Profiler::watchpointsTrueTrigger()
2168 {
2169 m_watchpointsTrueTrigger++;
2170 }
2171
2172 // For MemoryControl:
2173 void Profiler::profileMemReq(int bank) {
2174 m_memReq++;
2175 m_memBankCount[bank]++;
2176 }
2177 void Profiler::profileMemBankBusy() { m_memBankBusy++; }
2178 void Profiler::profileMemBusBusy() { m_memBusBusy++; }
2179 void Profiler::profileMemReadWriteBusy() { m_memReadWriteBusy++; }
2180 void Profiler::profileMemDataBusBusy() { m_memDataBusBusy++; }
2181 void Profiler::profileMemTfawBusy() { m_memTfawBusy++; }
2182 void Profiler::profileMemRefresh() { m_memRefresh++; }
2183 void Profiler::profileMemRead() { m_memRead++; }
2184 void Profiler::profileMemWrite() { m_memWrite++; }
2185 void Profiler::profileMemWaitCycles(int cycles) { m_memWaitCycles += cycles; }
2186 void Profiler::profileMemInputQ(int cycles) { m_memInputQ += cycles; }
2187 void Profiler::profileMemBankQ(int cycles) { m_memBankQ += cycles; }
2188 void Profiler::profileMemArbWait(int cycles) { m_memArbWait += cycles; }
2189 void Profiler::profileMemRandBusy() { m_memRandBusy++; }
2190 void Profiler::profileMemNotOld() { m_memNotOld++; }
2191
2192
2193 //----------- ATMTP -------------------//
2194
2195 void Profiler::profileTransactionTCC(NodeID id, Address pc){
2196 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
2197 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
2198 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
2199 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
2200
2201 const int ID_SPACES = 3;
2202 const int TIME_SPACES = 7;
2203 cout.flags(ios::right);
2204 cout << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2205 cout << setw(ID_SPACES) << id << " "
2206 << " XACT Aborting! Executed TCC "
2207 << " PC: " << pc
2208 << " *PC: 0x" << hex << myInst << dec
2209 << " '" << myInstStr << "'"
2210 << endl;
2211 }
2212 m_transactionUnsupInsts++;
2213 }
2214
2215 void Profiler::profileTransactionUnsupInst(NodeID id, Address pc){
2216 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
2217 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
2218 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
2219 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
2220
2221 const int ID_SPACES = 3;
2222 const int TIME_SPACES = 7;
2223 cout.flags(ios::right);
2224 cout << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2225 cout << setw(ID_SPACES) << id << " "
2226 << " XACT Aborting! Executed Unsupported Instruction "
2227 << " PC: " << pc
2228 << " *PC: 0x" << hex << myInst << dec
2229 << " '" << myInstStr << "'"
2230 << endl;
2231 }
2232 m_transactionUnsupInsts++;
2233 }
2234
2235 void Profiler::profileTransactionSaveInst(NodeID id, Address pc){
2236 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
2237 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
2238 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
2239 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
2240
2241 const int ID_SPACES = 3;
2242 const int TIME_SPACES = 7;
2243 cout.flags(ios::right);
2244 cout << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2245 cout << setw(ID_SPACES) << id << " "
2246 << " XACT Aborting! Executed Save Instruction "
2247 << " PC: " << pc
2248 << " *PC: 0x" << hex << myInst << dec
2249 << " '" << myInstStr << "'"
2250 << endl;
2251 }
2252 m_transactionSaveRestAborts++;
2253 }
2254
2255 void Profiler::profileTransactionRestoreInst(NodeID id, Address pc){
2256 if(PROFILE_XACT || (ATMTP_DEBUG_LEVEL >= 1)){
2257 physical_address_t myPhysPC = SIMICS_translate_address(id, pc);
2258 integer_t myInst = SIMICS_read_physical_memory(id, myPhysPC, 4);
2259 const char *myInstStr = SIMICS_disassemble_physical(id, myPhysPC);
2260
2261 const int ID_SPACES = 3;
2262 const int TIME_SPACES = 7;
2263 cout.flags(ios::right);
2264 cout << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2265 cout << setw(ID_SPACES) << id << " "
2266 << " XACT Aborting! Executed Restore Instruction "
2267 << " PC: " << pc
2268 << " *PC: 0x" << hex << myInst << dec
2269 << " '" << myInstStr << "'"
2270 << endl;
2271 }
2272 m_transactionSaveRestAborts++;
2273 }
2274
2275 void Profiler::profileTimerInterrupt(NodeID id,
2276 uinteger_t tick, uinteger_t tick_cmpr,
2277 uinteger_t stick, uinteger_t stick_cmpr,
2278 int trap_level,
2279 uinteger_t pc, uinteger_t npc,
2280 uinteger_t pstate, int pil){
2281 if (PROFILE_EXCEPTIONS) {
2282 const int ID_SPACES = 3;
2283 const int TIME_SPACES = 7;
2284 cout.flags(ios::right);
2285 cout << setw(TIME_SPACES) << g_eventQueue_ptr->getTime() << " ";
2286 cout << setw(ID_SPACES) << id << " ";
2287 cout << hex << "Timer--(Tick=0x" << tick << ", TckCmp=0x" << tick_cmpr
2288 << ", STick=0x" << stick << ", STickCmp=0x" << stick_cmpr
2289 << ")--(PC=" << pc << ", " << npc
2290 << dec << ")--(TL=" << trap_level << ", pil=" << pil
2291 << hex << ", pstate=0x" << pstate
2292 << dec << ")" << endl;
2293 }
2294 }