Profiler.hh (9496:28d88a0fda74) Profiler.hh (9497:2759161b9d7f)
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#ifndef __MEM_RUBY_PROFILER_PROFILER_HH__
46#define __MEM_RUBY_PROFILER_PROFILER_HH__
47
48#include <iostream>
49#include <map>
50#include <string>
51#include <vector>
52
53#include "base/hashmap.hh"
54#include "mem/protocol/AccessType.hh"
55#include "mem/protocol/GenericMachineType.hh"
56#include "mem/protocol/GenericRequestType.hh"
57#include "mem/protocol/PrefetchBit.hh"
58#include "mem/protocol/RubyAccessMode.hh"
59#include "mem/protocol/RubyRequestType.hh"
60#include "mem/ruby/common/Address.hh"
61#include "mem/ruby/common/Global.hh"
62#include "mem/ruby/common/Histogram.hh"
63#include "mem/ruby/common/Set.hh"
64#include "mem/ruby/system/MachineID.hh"
65#include "mem/ruby/system/MemoryControl.hh"
66#include "params/RubyProfiler.hh"
67#include "sim/sim_object.hh"
68
69class RubyRequest;
70class AddressProfiler;
71
72class Profiler : public SimObject
73{
74 public:
75 typedef RubyProfilerParams Params;
76 Profiler(const Params *);
77 ~Profiler();
78
79 void wakeup();
80
81 void setPeriodicStatsFile(const std::string& filename);
82 void setPeriodicStatsInterval(int64_t period);
83
84 void printStats(std::ostream& out, bool short_stats=false);
85 void printShortStats(std::ostream& out) { printStats(out, true); }
86 void printTraceStats(std::ostream& out) const;
87 void clearStats();
88 void printResourceUsage(std::ostream& out) const;
89
90 AddressProfiler* getAddressProfiler() { return m_address_profiler_ptr; }
91 AddressProfiler* getInstructionProfiler() { return m_inst_profiler_ptr; }
92
93 void addAddressTraceSample(const RubyRequest& msg, NodeID id);
94
95 void profileRequest(const std::string& requestStr);
96 void profileSharing(const Address& addr, AccessType type,
97 NodeID requestor, const Set& sharers,
98 const Set& owner);
99
100 void profileMulticastRetry(const Address& addr, int count);
101
102 void profileFilterAction(int action);
103
104 void profileConflictingRequests(const Address& addr);
105
106 void
107 profileOutstandingRequest(int outstanding)
108 {
109 m_outstanding_requests.add(outstanding);
110 }
111
112 void
113 profileOutstandingPersistentRequest(int outstanding)
114 {
115 m_outstanding_persistent_requests.add(outstanding);
116 }
117
118 void
119 profileAverageLatencyEstimate(int latency)
120 {
121 m_average_latency_estimate.add(latency);
122 }
123
124 void recordPrediction(bool wasGood, bool wasPredicted);
125
126 void startTransaction(int cpu);
127 void endTransaction(int cpu);
128 void profilePFWait(Time waitTime);
129
130 void controllerBusy(MachineID machID);
131 void bankBusy();
132
133 void missLatency(Time t,
134 RubyRequestType type,
135 const GenericMachineType respondingMach);
136
137 void missLatencyWcc(Time issuedTime,
138 Time initialRequestTime,
139 Time forwardRequestTime,
140 Time firstResponseTime,
141 Time completionTime);
142
143 void missLatencyDir(Time issuedTime,
144 Time initialRequestTime,
145 Time forwardRequestTime,
146 Time firstResponseTime,
147 Time completionTime);
148
149 void swPrefetchLatency(Time t,
150 RubyRequestType type,
151 const GenericMachineType respondingMach);
152
153 void sequencerRequests(int num) { m_sequencer_requests.add(num); }
154
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#ifndef __MEM_RUBY_PROFILER_PROFILER_HH__
46#define __MEM_RUBY_PROFILER_PROFILER_HH__
47
48#include <iostream>
49#include <map>
50#include <string>
51#include <vector>
52
53#include "base/hashmap.hh"
54#include "mem/protocol/AccessType.hh"
55#include "mem/protocol/GenericMachineType.hh"
56#include "mem/protocol/GenericRequestType.hh"
57#include "mem/protocol/PrefetchBit.hh"
58#include "mem/protocol/RubyAccessMode.hh"
59#include "mem/protocol/RubyRequestType.hh"
60#include "mem/ruby/common/Address.hh"
61#include "mem/ruby/common/Global.hh"
62#include "mem/ruby/common/Histogram.hh"
63#include "mem/ruby/common/Set.hh"
64#include "mem/ruby/system/MachineID.hh"
65#include "mem/ruby/system/MemoryControl.hh"
66#include "params/RubyProfiler.hh"
67#include "sim/sim_object.hh"
68
69class RubyRequest;
70class AddressProfiler;
71
72class Profiler : public SimObject
73{
74 public:
75 typedef RubyProfilerParams Params;
76 Profiler(const Params *);
77 ~Profiler();
78
79 void wakeup();
80
81 void setPeriodicStatsFile(const std::string& filename);
82 void setPeriodicStatsInterval(int64_t period);
83
84 void printStats(std::ostream& out, bool short_stats=false);
85 void printShortStats(std::ostream& out) { printStats(out, true); }
86 void printTraceStats(std::ostream& out) const;
87 void clearStats();
88 void printResourceUsage(std::ostream& out) const;
89
90 AddressProfiler* getAddressProfiler() { return m_address_profiler_ptr; }
91 AddressProfiler* getInstructionProfiler() { return m_inst_profiler_ptr; }
92
93 void addAddressTraceSample(const RubyRequest& msg, NodeID id);
94
95 void profileRequest(const std::string& requestStr);
96 void profileSharing(const Address& addr, AccessType type,
97 NodeID requestor, const Set& sharers,
98 const Set& owner);
99
100 void profileMulticastRetry(const Address& addr, int count);
101
102 void profileFilterAction(int action);
103
104 void profileConflictingRequests(const Address& addr);
105
106 void
107 profileOutstandingRequest(int outstanding)
108 {
109 m_outstanding_requests.add(outstanding);
110 }
111
112 void
113 profileOutstandingPersistentRequest(int outstanding)
114 {
115 m_outstanding_persistent_requests.add(outstanding);
116 }
117
118 void
119 profileAverageLatencyEstimate(int latency)
120 {
121 m_average_latency_estimate.add(latency);
122 }
123
124 void recordPrediction(bool wasGood, bool wasPredicted);
125
126 void startTransaction(int cpu);
127 void endTransaction(int cpu);
128 void profilePFWait(Time waitTime);
129
130 void controllerBusy(MachineID machID);
131 void bankBusy();
132
133 void missLatency(Time t,
134 RubyRequestType type,
135 const GenericMachineType respondingMach);
136
137 void missLatencyWcc(Time issuedTime,
138 Time initialRequestTime,
139 Time forwardRequestTime,
140 Time firstResponseTime,
141 Time completionTime);
142
143 void missLatencyDir(Time issuedTime,
144 Time initialRequestTime,
145 Time forwardRequestTime,
146 Time firstResponseTime,
147 Time completionTime);
148
149 void swPrefetchLatency(Time t,
150 RubyRequestType type,
151 const GenericMachineType respondingMach);
152
153 void sequencerRequests(int num) { m_sequencer_requests.add(num); }
154
155 void profileMsgDelay(uint32_t virtualNetwork, Time delayCycles);
156
157 void print(std::ostream& out) const;
158
159 void rubyWatch(int proc);
160 bool watchAddress(Address addr);
161
162 // return Ruby's start time
163 Time
164 getRubyStartTime()
165 {
166 return m_ruby_start;
167 }
168
169 // added by SS
170 bool getHotLines() { return m_hot_lines; }
171 bool getAllInstructions() { return m_all_instructions; }
172
173 private:
174 void printRequestProfile(std::ostream &out);
155 void print(std::ostream& out) const;
156
157 void rubyWatch(int proc);
158 bool watchAddress(Address addr);
159
160 // return Ruby's start time
161 Time
162 getRubyStartTime()
163 {
164 return m_ruby_start;
165 }
166
167 // added by SS
168 bool getHotLines() { return m_hot_lines; }
169 bool getAllInstructions() { return m_all_instructions; }
170
171 private:
172 void printRequestProfile(std::ostream &out);
173 void printDelayProfile(std::ostream &out);
175
176 private:
177 // Private copy constructor and assignment operator
178 Profiler(const Profiler& obj);
179 Profiler& operator=(const Profiler& obj);
180
181 AddressProfiler* m_address_profiler_ptr;
182 AddressProfiler* m_inst_profiler_ptr;
183
184 std::vector<int64> m_instructions_executed_at_start;
185 std::vector<int64> m_cycles_executed_at_start;
186
187 std::ostream* m_periodic_output_file_ptr;
188 int64_t m_stats_period;
189
190 Time m_ruby_start;
191 time_t m_real_time_start_time;
192
193 int64_t m_busyBankCount;
194 Histogram m_multicast_retry_histogram;
195
196 Histogram m_filter_action_histogram;
197 Histogram m_tbeProfile;
198
199 Histogram m_sequencer_requests;
200 Histogram m_read_sharing_histogram;
201 Histogram m_write_sharing_histogram;
202 Histogram m_all_sharing_histogram;
203 int64 m_cache_to_cache;
204 int64 m_memory_to_cache;
205
206 Histogram m_prefetchWaitHistogram;
207
208 std::vector<Histogram> m_missLatencyHistograms;
209 std::vector<Histogram> m_machLatencyHistograms;
210 std::vector< std::vector<Histogram> > m_missMachLatencyHistograms;
211 Histogram m_wCCIssueToInitialRequestHistogram;
212 Histogram m_wCCInitialRequestToForwardRequestHistogram;
213 Histogram m_wCCForwardRequestToFirstResponseHistogram;
214 Histogram m_wCCFirstResponseToCompleteHistogram;
215 int64 m_wCCIncompleteTimes;
216 Histogram m_dirIssueToInitialRequestHistogram;
217 Histogram m_dirInitialRequestToForwardRequestHistogram;
218 Histogram m_dirForwardRequestToFirstResponseHistogram;
219 Histogram m_dirFirstResponseToCompleteHistogram;
220 int64 m_dirIncompleteTimes;
221
222 Histogram m_allMissLatencyHistogram;
223
224 Histogram m_allSWPrefetchLatencyHistogram;
225 Histogram m_SWPrefetchL2MissLatencyHistogram;
226 std::vector<Histogram> m_SWPrefetchLatencyHistograms;
227 std::vector<Histogram> m_SWPrefetchMachLatencyHistograms;
228
174
175 private:
176 // Private copy constructor and assignment operator
177 Profiler(const Profiler& obj);
178 Profiler& operator=(const Profiler& obj);
179
180 AddressProfiler* m_address_profiler_ptr;
181 AddressProfiler* m_inst_profiler_ptr;
182
183 std::vector<int64> m_instructions_executed_at_start;
184 std::vector<int64> m_cycles_executed_at_start;
185
186 std::ostream* m_periodic_output_file_ptr;
187 int64_t m_stats_period;
188
189 Time m_ruby_start;
190 time_t m_real_time_start_time;
191
192 int64_t m_busyBankCount;
193 Histogram m_multicast_retry_histogram;
194
195 Histogram m_filter_action_histogram;
196 Histogram m_tbeProfile;
197
198 Histogram m_sequencer_requests;
199 Histogram m_read_sharing_histogram;
200 Histogram m_write_sharing_histogram;
201 Histogram m_all_sharing_histogram;
202 int64 m_cache_to_cache;
203 int64 m_memory_to_cache;
204
205 Histogram m_prefetchWaitHistogram;
206
207 std::vector<Histogram> m_missLatencyHistograms;
208 std::vector<Histogram> m_machLatencyHistograms;
209 std::vector< std::vector<Histogram> > m_missMachLatencyHistograms;
210 Histogram m_wCCIssueToInitialRequestHistogram;
211 Histogram m_wCCInitialRequestToForwardRequestHistogram;
212 Histogram m_wCCForwardRequestToFirstResponseHistogram;
213 Histogram m_wCCFirstResponseToCompleteHistogram;
214 int64 m_wCCIncompleteTimes;
215 Histogram m_dirIssueToInitialRequestHistogram;
216 Histogram m_dirInitialRequestToForwardRequestHistogram;
217 Histogram m_dirForwardRequestToFirstResponseHistogram;
218 Histogram m_dirFirstResponseToCompleteHistogram;
219 int64 m_dirIncompleteTimes;
220
221 Histogram m_allMissLatencyHistogram;
222
223 Histogram m_allSWPrefetchLatencyHistogram;
224 Histogram m_SWPrefetchL2MissLatencyHistogram;
225 std::vector<Histogram> m_SWPrefetchLatencyHistograms;
226 std::vector<Histogram> m_SWPrefetchMachLatencyHistograms;
227
229 Histogram m_delayedCyclesHistogram;
230 Histogram m_delayedCyclesNonPFHistogram;
231 std::vector<Histogram> m_delayedCyclesVCHistograms;
232
233 Histogram m_outstanding_requests;
234 Histogram m_outstanding_persistent_requests;
235
236 Histogram m_average_latency_estimate;
237
238 m5::hash_set<Address> m_watch_address_set;
239
240 //added by SS
241 bool m_hot_lines;
242 bool m_all_instructions;
243
244 int m_num_of_sequencers;
245
246 protected:
247 class ProfileEvent : public Event
248 {
249 public:
250 ProfileEvent(Profiler *_profiler)
251 {
252 profiler = _profiler;
253 }
254 private:
255 void process() { profiler->wakeup(); }
256 Profiler *profiler;
257 };
258 ProfileEvent m_event;
259};
260
261inline std::ostream&
262operator<<(std::ostream& out, const Profiler& obj)
263{
264 obj.print(out);
265 out << std::flush;
266 return out;
267}
268
269#endif // __MEM_RUBY_PROFILER_PROFILER_HH__
228 Histogram m_outstanding_requests;
229 Histogram m_outstanding_persistent_requests;
230
231 Histogram m_average_latency_estimate;
232
233 m5::hash_set<Address> m_watch_address_set;
234
235 //added by SS
236 bool m_hot_lines;
237 bool m_all_instructions;
238
239 int m_num_of_sequencers;
240
241 protected:
242 class ProfileEvent : public Event
243 {
244 public:
245 ProfileEvent(Profiler *_profiler)
246 {
247 profiler = _profiler;
248 }
249 private:
250 void process() { profiler->wakeup(); }
251 Profiler *profiler;
252 };
253 ProfileEvent m_event;
254};
255
256inline std::ostream&
257operator<<(std::ostream& out, const Profiler& obj)
258{
259 obj.print(out);
260 out << std::flush;
261 return out;
262}
263
264#endif // __MEM_RUBY_PROFILER_PROFILER_HH__