Profiler.hh revision 8608:02d7ac5fb855 
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/Consumer.hh"
62#include "mem/ruby/common/Global.hh"
63#include "mem/ruby/common/Histogram.hh"
64#include "mem/ruby/common/Set.hh"
65#include "mem/ruby/system/MachineID.hh"
66#include "mem/ruby/system/MemoryControl.hh"
67#include "params/RubyProfiler.hh"
68#include "sim/sim_object.hh"
69
70class RubyRequest;
71class AddressProfiler;
72
73class Profiler : public SimObject, public Consumer
74{
75  public:
76    typedef RubyProfilerParams Params;
77    Profiler(const Params *);
78    ~Profiler();
79
80    void wakeup();
81
82    void setPeriodicStatsFile(const std::string& filename);
83    void setPeriodicStatsInterval(integer_t period);
84
85    void printStats(std::ostream& out, bool short_stats=false);
86    void printShortStats(std::ostream& out) { printStats(out, true); }
87    void printTraceStats(std::ostream& out) const;
88    void clearStats();
89    void printConfig(std::ostream& out) const;
90    void printResourceUsage(std::ostream& out) const;
91
92    AddressProfiler* getAddressProfiler() { return m_address_profiler_ptr; }
93    AddressProfiler* getInstructionProfiler() { return m_inst_profiler_ptr; }
94
95    void addAddressTraceSample(const RubyRequest& msg, NodeID id);
96
97    void profileRequest(const std::string& requestStr);
98    void profileSharing(const Address& addr, AccessType type,
99                        NodeID requestor, const Set& sharers,
100                        const Set& owner);
101
102    void profileMulticastRetry(const Address& addr, int count);
103
104    void profileFilterAction(int action);
105
106    void profileConflictingRequests(const Address& addr);
107
108    void
109    profileOutstandingRequest(int outstanding)
110    {
111        m_outstanding_requests.add(outstanding);
112    }
113
114    void
115    profileOutstandingPersistentRequest(int outstanding)
116    {
117        m_outstanding_persistent_requests.add(outstanding);
118    }
119
120    void
121    profileAverageLatencyEstimate(int latency)
122    {
123        m_average_latency_estimate.add(latency);
124    }
125
126    void recordPrediction(bool wasGood, bool wasPredicted);
127
128    void startTransaction(int cpu);
129    void endTransaction(int cpu);
130    void profilePFWait(Time waitTime);
131
132    void controllerBusy(MachineID machID);
133    void bankBusy();
134
135    void missLatency(Time t,
136                     RubyRequestType type,
137                     const GenericMachineType respondingMach);
138
139    void missLatencyWcc(Time issuedTime,
140                        Time initialRequestTime,
141                        Time forwardRequestTime,
142                        Time firstResponseTime,
143                        Time completionTime);
144
145    void missLatencyDir(Time issuedTime,
146                        Time initialRequestTime,
147                        Time forwardRequestTime,
148                        Time firstResponseTime,
149                        Time completionTime);
150
151    void swPrefetchLatency(Time t,
152                           RubyRequestType type,
153                           const GenericMachineType respondingMach);
154
155    void sequencerRequests(int num) { m_sequencer_requests.add(num); }
156
157    void profileMsgDelay(int virtualNetwork, int delayCycles);
158
159    void print(std::ostream& out) const;
160
161    void rubyWatch(int proc);
162    bool watchAddress(Address addr);
163
164    // return Ruby's start time
165    Time
166    getRubyStartTime()
167    {
168        return m_ruby_start;
169    }
170
171    // added by SS
172    bool getHotLines() { return m_hot_lines; }
173    bool getAllInstructions() { return m_all_instructions; }
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    integer_t m_stats_period;
188
189    Time m_ruby_start;
190    time_t m_real_time_start_time;
191
192    std::vector<std::vector<integer_t> > m_busyControllerCount;
193    integer_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
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    // counts all initiated cache request including PUTs
240    int m_requests;
241    std::map<std::string, int> m_requestProfileMap;
242
243    //added by SS
244    bool m_hot_lines;
245    bool m_all_instructions;
246
247    int m_num_of_sequencers;
248};
249
250inline std::ostream&
251operator<<(std::ostream& out, const Profiler& obj)
252{
253    obj.print(out);
254    out << std::flush;
255    return out;
256}
257
258#endif // __MEM_RUBY_PROFILER_PROFILER_HH__
259
260
261