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#include "mem/ruby/profiler/AddressProfiler.hh"
30
31#include <vector>
32
33#include "base/stl_helpers.hh"
34#include "mem/protocol/RubyRequest.hh"
35#include "mem/ruby/profiler/Profiler.hh"
36
37using namespace std;
38typedef AddressProfiler::AddressMap AddressMap;
39
40using m5::stl_helpers::operator<<;
41
42// Helper functions
43AccessTraceForAddress&
44lookupTraceForAddress(Addr addr, AddressMap& record_map)
45{
46 // we create a static default object here that is used to insert
47 // since the insertion will create a copy of the object in the
48 // process. Perhaps this is optimizing early, but it doesn't seem
49 // like it could hurt.
50 static const AccessTraceForAddress dflt;
51
52 pair<AddressMap::iterator, bool> r =
53 record_map.insert(make_pair(addr, dflt));
54 AddressMap::iterator i = r.first;
55 AccessTraceForAddress &access_trace = i->second;
56 if (r.second) {
57 // there was nothing there and the insert succeed, so we need
58 // to actually set the address.
59 access_trace.setAddress(addr);
60 }
61
62 return access_trace;
63}
64
65void
66printSorted(ostream& out, int num_of_sequencers, const AddressMap &record_map,
67 string description, Profiler *profiler)
68{
69 const int records_printed = 100;
70
71 uint64_t misses = 0;
72 std::vector<const AccessTraceForAddress *> sorted;
73
74 AddressMap::const_iterator i = record_map.begin();
75 AddressMap::const_iterator end = record_map.end();
76 for (; i != end; ++i) {
77 const AccessTraceForAddress* record = &i->second;
78 misses += record->getTotal();
79 sorted.push_back(record);
80 }
81 sort(sorted.begin(), sorted.end(), AccessTraceForAddress::less_equal);
82
83 out << "Total_entries_" << description << ": " << record_map.size()
84 << endl;
85 if (profiler->getAllInstructions())
86 out << "Total_Instructions_" << description << ": " << misses << endl;
87 else
88 out << "Total_data_misses_" << description << ": " << misses << endl;
89
90 out << "total | load store atomic | user supervisor | sharing | touched-by"
91 << endl;
92
93 Histogram remaining_records(1, 100);
94 Histogram all_records(1, 100);
95 Histogram remaining_records_log(-1);
96 Histogram all_records_log(-1);
97
98 // Allows us to track how many lines where touched by n processors
99 std::vector<int64_t> m_touched_vec;
100 std::vector<int64_t> m_touched_weighted_vec;
101 m_touched_vec.resize(num_of_sequencers+1);
102 m_touched_weighted_vec.resize(num_of_sequencers+1);
103 for (int j = 0; j < m_touched_vec.size(); j++) {
104 m_touched_vec[j] = 0;
105 m_touched_weighted_vec[j] = 0;
106 }
107
108 int counter = 0;
109 int max = sorted.size();
110 while (counter < max && counter < records_printed) {
111 const AccessTraceForAddress* record = sorted[counter];
112 double percent = 100.0 * (record->getTotal() / double(misses));
113 out << description << " | " << percent << " % " << *record << endl;
114 all_records.add(record->getTotal());
115 all_records_log.add(record->getTotal());
116 counter++;
117 m_touched_vec[record->getTouchedBy()]++;
118 m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
119 }
120
121 while (counter < max) {
122 const AccessTraceForAddress* record = sorted[counter];
123 all_records.add(record->getTotal());
124 remaining_records.add(record->getTotal());
125 all_records_log.add(record->getTotal());
126 remaining_records_log.add(record->getTotal());
127 m_touched_vec[record->getTouchedBy()]++;
128 m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
129 }
130 out << endl;
131 out << "all_records_" << description << ": "
132 << all_records << endl
133 << "all_records_log_" << description << ": "
134 << all_records_log << endl
135 << "remaining_records_" << description << ": "
136 << remaining_records << endl
137 << "remaining_records_log_" << description << ": "
138 << remaining_records_log << endl
139 << "touched_by_" << description << ": "
140 << m_touched_vec << endl
141 << "touched_by_weighted_" << description << ": "
142 << m_touched_weighted_vec << endl
143 << endl;
144}
145
146AddressProfiler::AddressProfiler(int num_of_sequencers, Profiler *profiler)
147 : m_profiler(profiler)
148{
149 m_num_of_sequencers = num_of_sequencers;
150 clearStats();
151}
152
153AddressProfiler::~AddressProfiler()
154{
155}
156
157void
158AddressProfiler::setHotLines(bool hot_lines)
159{
160 m_hot_lines = hot_lines;
161}
162
163void
164AddressProfiler::setAllInstructions(bool all_instructions)
165{
166 m_all_instructions = all_instructions;
167}
168
169void
170AddressProfiler::printStats(ostream& out) const
171{
172 if (m_hot_lines) {
173 out << endl;
174 out << "AddressProfiler Stats" << endl;
175 out << "---------------------" << endl;
176
177 out << endl;
178 out << "sharing_misses: " << m_sharing_miss_counter << endl;
179 out << "getx_sharing_histogram: " << m_getx_sharing_histogram << endl;
180 out << "gets_sharing_histogram: " << m_gets_sharing_histogram << endl;
181
182 out << endl;
183 out << "Hot Data Blocks" << endl;
184 out << "---------------" << endl;
185 out << endl;
186 printSorted(out, m_num_of_sequencers, m_dataAccessTrace,
187 "block_address", m_profiler);
188
189 out << endl;
190 out << "Hot MacroData Blocks" << endl;
191 out << "--------------------" << endl;
192 out << endl;
193 printSorted(out, m_num_of_sequencers, m_macroBlockAccessTrace,
194 "macroblock_address", m_profiler);
195
196 out << "Hot Instructions" << endl;
197 out << "----------------" << endl;
198 out << endl;
199 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
200 "pc_address", m_profiler);
201 }
202
203 if (m_all_instructions) {
204 out << endl;
205 out << "All Instructions Profile:" << endl;
206 out << "-------------------------" << endl;
207 out << endl;
208 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
209 "pc_address", m_profiler);
210 out << endl;
211 }
212
213 if (m_retryProfileHisto.size() > 0) {
214 out << "Retry Profile" << endl;
215 out << "-------------" << endl;
216 out << endl;
217 out << "retry_histogram_absolute: " << m_retryProfileHisto << endl;
218 out << "retry_histogram_write: " << m_retryProfileHistoWrite << endl;
219 out << "retry_histogram_read: " << m_retryProfileHistoRead << endl;
220
221 out << "retry_histogram_percent: ";
222 m_retryProfileHisto.printPercent(out);
223 out << endl;
224
225 printSorted(out, m_num_of_sequencers, m_retryProfileMap,
226 "block_address", m_profiler);
227 out << endl;
228 }
229}
230
231void
232AddressProfiler::clearStats()
233{
234 // Clear the maps
235 m_sharing_miss_counter = 0;
236 m_dataAccessTrace.clear();
237 m_macroBlockAccessTrace.clear();
238 m_programCounterAccessTrace.clear();
239 m_retryProfileMap.clear();
240 m_retryProfileHisto.clear();
241 m_retryProfileHistoRead.clear();
242 m_retryProfileHistoWrite.clear();
243 m_getx_sharing_histogram.clear();
244 m_gets_sharing_histogram.clear();
245}
246
247void
248AddressProfiler::profileGetX(Addr datablock, Addr PC,
249 const Set& owner, const Set& sharers,
250 NodeID requestor)
251{
252 Set indirection_set;
253 indirection_set.addSet(sharers);
254 indirection_set.addSet(owner);
255 indirection_set.remove(requestor);
256 int num_indirections = indirection_set.count();
257
258 m_getx_sharing_histogram.add(num_indirections);
259 bool indirection_miss = (num_indirections > 0);
260
261 addTraceSample(datablock, PC, RubyRequestType_ST, RubyAccessMode(0),
262 requestor, indirection_miss);
263}
264
265void
266AddressProfiler::profileGetS(Addr datablock, Addr PC,
267 const Set& owner, const Set& sharers,
268 NodeID requestor)
269{
270 Set indirection_set;
271 indirection_set.addSet(owner);
272 indirection_set.remove(requestor);
273 int num_indirections = indirection_set.count();
274
275 m_gets_sharing_histogram.add(num_indirections);
276 bool indirection_miss = (num_indirections > 0);
277
278 addTraceSample(datablock, PC, RubyRequestType_LD, RubyAccessMode(0),
279 requestor, indirection_miss);
280}
281
282void
283AddressProfiler::addTraceSample(Addr data_addr, Addr pc_addr,
284 RubyRequestType type,
285 RubyAccessMode access_mode, NodeID id,
286 bool sharing_miss)
287{
288 if (m_all_instructions) {
289 if (sharing_miss) {
290 m_sharing_miss_counter++;
291 }
292
293 // record data address trace info
294 data_addr = makeLineAddress(data_addr);
295 lookupTraceForAddress(data_addr, m_dataAccessTrace).
296 update(type, access_mode, id, sharing_miss);
297
298 // record macro data address trace info
299
300 // 6 for datablock, 4 to make it 16x more coarse
301 Addr macro_addr = maskLowOrderBits(data_addr, 10);
302 lookupTraceForAddress(macro_addr, m_macroBlockAccessTrace).
303 update(type, access_mode, id, sharing_miss);
304
305 // record program counter address trace info
306 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
307 update(type, access_mode, id, sharing_miss);
308 }
309
310 if (m_all_instructions) {
311 // This code is used if the address profiler is an
312 // all-instructions profiler record program counter address
313 // trace info
314 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
315 update(type, access_mode, id, sharing_miss);
316 }
317}
318
319void
320AddressProfiler::profileRetry(Addr data_addr, AccessType type, int count)
321{
322 m_retryProfileHisto.add(count);
323 if (type == AccessType_Read) {
324 m_retryProfileHistoRead.add(count);
325 } else {
326 m_retryProfileHistoWrite.add(count);
327 }
328 if (count > 1) {
329 lookupTraceForAddress(data_addr, m_retryProfileMap).addSample(count);
330 }
331}
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#include "mem/ruby/profiler/AddressProfiler.hh"
30
31#include <vector>
32
33#include "base/stl_helpers.hh"
34#include "mem/protocol/RubyRequest.hh"
35#include "mem/ruby/profiler/Profiler.hh"
36
37using namespace std;
38typedef AddressProfiler::AddressMap AddressMap;
39
40using m5::stl_helpers::operator<<;
41
42// Helper functions
43AccessTraceForAddress&
44lookupTraceForAddress(Addr addr, AddressMap& record_map)
45{
46 // we create a static default object here that is used to insert
47 // since the insertion will create a copy of the object in the
48 // process. Perhaps this is optimizing early, but it doesn't seem
49 // like it could hurt.
50 static const AccessTraceForAddress dflt;
51
52 pair<AddressMap::iterator, bool> r =
53 record_map.insert(make_pair(addr, dflt));
54 AddressMap::iterator i = r.first;
55 AccessTraceForAddress &access_trace = i->second;
56 if (r.second) {
57 // there was nothing there and the insert succeed, so we need
58 // to actually set the address.
59 access_trace.setAddress(addr);
60 }
61
62 return access_trace;
63}
64
65void
66printSorted(ostream& out, int num_of_sequencers, const AddressMap &record_map,
67 string description, Profiler *profiler)
68{
69 const int records_printed = 100;
70
71 uint64_t misses = 0;
72 std::vector<const AccessTraceForAddress *> sorted;
73
74 AddressMap::const_iterator i = record_map.begin();
75 AddressMap::const_iterator end = record_map.end();
76 for (; i != end; ++i) {
77 const AccessTraceForAddress* record = &i->second;
78 misses += record->getTotal();
79 sorted.push_back(record);
80 }
81 sort(sorted.begin(), sorted.end(), AccessTraceForAddress::less_equal);
82
83 out << "Total_entries_" << description << ": " << record_map.size()
84 << endl;
85 if (profiler->getAllInstructions())
86 out << "Total_Instructions_" << description << ": " << misses << endl;
87 else
88 out << "Total_data_misses_" << description << ": " << misses << endl;
89
90 out << "total | load store atomic | user supervisor | sharing | touched-by"
91 << endl;
92
93 Histogram remaining_records(1, 100);
94 Histogram all_records(1, 100);
95 Histogram remaining_records_log(-1);
96 Histogram all_records_log(-1);
97
98 // Allows us to track how many lines where touched by n processors
99 std::vector<int64_t> m_touched_vec;
100 std::vector<int64_t> m_touched_weighted_vec;
101 m_touched_vec.resize(num_of_sequencers+1);
102 m_touched_weighted_vec.resize(num_of_sequencers+1);
103 for (int j = 0; j < m_touched_vec.size(); j++) {
104 m_touched_vec[j] = 0;
105 m_touched_weighted_vec[j] = 0;
106 }
107
108 int counter = 0;
109 int max = sorted.size();
110 while (counter < max && counter < records_printed) {
111 const AccessTraceForAddress* record = sorted[counter];
112 double percent = 100.0 * (record->getTotal() / double(misses));
113 out << description << " | " << percent << " % " << *record << endl;
114 all_records.add(record->getTotal());
115 all_records_log.add(record->getTotal());
116 counter++;
117 m_touched_vec[record->getTouchedBy()]++;
118 m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
119 }
120
121 while (counter < max) {
122 const AccessTraceForAddress* record = sorted[counter];
123 all_records.add(record->getTotal());
124 remaining_records.add(record->getTotal());
125 all_records_log.add(record->getTotal());
126 remaining_records_log.add(record->getTotal());
127 m_touched_vec[record->getTouchedBy()]++;
128 m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
129 }
130 out << endl;
131 out << "all_records_" << description << ": "
132 << all_records << endl
133 << "all_records_log_" << description << ": "
134 << all_records_log << endl
135 << "remaining_records_" << description << ": "
136 << remaining_records << endl
137 << "remaining_records_log_" << description << ": "
138 << remaining_records_log << endl
139 << "touched_by_" << description << ": "
140 << m_touched_vec << endl
141 << "touched_by_weighted_" << description << ": "
142 << m_touched_weighted_vec << endl
143 << endl;
144}
145
146AddressProfiler::AddressProfiler(int num_of_sequencers, Profiler *profiler)
147 : m_profiler(profiler)
148{
149 m_num_of_sequencers = num_of_sequencers;
150 clearStats();
151}
152
153AddressProfiler::~AddressProfiler()
154{
155}
156
157void
158AddressProfiler::setHotLines(bool hot_lines)
159{
160 m_hot_lines = hot_lines;
161}
162
163void
164AddressProfiler::setAllInstructions(bool all_instructions)
165{
166 m_all_instructions = all_instructions;
167}
168
169void
170AddressProfiler::printStats(ostream& out) const
171{
172 if (m_hot_lines) {
173 out << endl;
174 out << "AddressProfiler Stats" << endl;
175 out << "---------------------" << endl;
176
177 out << endl;
178 out << "sharing_misses: " << m_sharing_miss_counter << endl;
179 out << "getx_sharing_histogram: " << m_getx_sharing_histogram << endl;
180 out << "gets_sharing_histogram: " << m_gets_sharing_histogram << endl;
181
182 out << endl;
183 out << "Hot Data Blocks" << endl;
184 out << "---------------" << endl;
185 out << endl;
186 printSorted(out, m_num_of_sequencers, m_dataAccessTrace,
187 "block_address", m_profiler);
188
189 out << endl;
190 out << "Hot MacroData Blocks" << endl;
191 out << "--------------------" << endl;
192 out << endl;
193 printSorted(out, m_num_of_sequencers, m_macroBlockAccessTrace,
194 "macroblock_address", m_profiler);
195
196 out << "Hot Instructions" << endl;
197 out << "----------------" << endl;
198 out << endl;
199 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
200 "pc_address", m_profiler);
201 }
202
203 if (m_all_instructions) {
204 out << endl;
205 out << "All Instructions Profile:" << endl;
206 out << "-------------------------" << endl;
207 out << endl;
208 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
209 "pc_address", m_profiler);
210 out << endl;
211 }
212
213 if (m_retryProfileHisto.size() > 0) {
214 out << "Retry Profile" << endl;
215 out << "-------------" << endl;
216 out << endl;
217 out << "retry_histogram_absolute: " << m_retryProfileHisto << endl;
218 out << "retry_histogram_write: " << m_retryProfileHistoWrite << endl;
219 out << "retry_histogram_read: " << m_retryProfileHistoRead << endl;
220
221 out << "retry_histogram_percent: ";
222 m_retryProfileHisto.printPercent(out);
223 out << endl;
224
225 printSorted(out, m_num_of_sequencers, m_retryProfileMap,
226 "block_address", m_profiler);
227 out << endl;
228 }
229}
230
231void
232AddressProfiler::clearStats()
233{
234 // Clear the maps
235 m_sharing_miss_counter = 0;
236 m_dataAccessTrace.clear();
237 m_macroBlockAccessTrace.clear();
238 m_programCounterAccessTrace.clear();
239 m_retryProfileMap.clear();
240 m_retryProfileHisto.clear();
241 m_retryProfileHistoRead.clear();
242 m_retryProfileHistoWrite.clear();
243 m_getx_sharing_histogram.clear();
244 m_gets_sharing_histogram.clear();
245}
246
247void
248AddressProfiler::profileGetX(Addr datablock, Addr PC,
249 const Set& owner, const Set& sharers,
250 NodeID requestor)
251{
252 Set indirection_set;
253 indirection_set.addSet(sharers);
254 indirection_set.addSet(owner);
255 indirection_set.remove(requestor);
256 int num_indirections = indirection_set.count();
257
258 m_getx_sharing_histogram.add(num_indirections);
259 bool indirection_miss = (num_indirections > 0);
260
261 addTraceSample(datablock, PC, RubyRequestType_ST, RubyAccessMode(0),
262 requestor, indirection_miss);
263}
264
265void
266AddressProfiler::profileGetS(Addr datablock, Addr PC,
267 const Set& owner, const Set& sharers,
268 NodeID requestor)
269{
270 Set indirection_set;
271 indirection_set.addSet(owner);
272 indirection_set.remove(requestor);
273 int num_indirections = indirection_set.count();
274
275 m_gets_sharing_histogram.add(num_indirections);
276 bool indirection_miss = (num_indirections > 0);
277
278 addTraceSample(datablock, PC, RubyRequestType_LD, RubyAccessMode(0),
279 requestor, indirection_miss);
280}
281
282void
283AddressProfiler::addTraceSample(Addr data_addr, Addr pc_addr,
284 RubyRequestType type,
285 RubyAccessMode access_mode, NodeID id,
286 bool sharing_miss)
287{
288 if (m_all_instructions) {
289 if (sharing_miss) {
290 m_sharing_miss_counter++;
291 }
292
293 // record data address trace info
294 data_addr = makeLineAddress(data_addr);
295 lookupTraceForAddress(data_addr, m_dataAccessTrace).
296 update(type, access_mode, id, sharing_miss);
297
298 // record macro data address trace info
299
300 // 6 for datablock, 4 to make it 16x more coarse
301 Addr macro_addr = maskLowOrderBits(data_addr, 10);
302 lookupTraceForAddress(macro_addr, m_macroBlockAccessTrace).
303 update(type, access_mode, id, sharing_miss);
304
305 // record program counter address trace info
306 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
307 update(type, access_mode, id, sharing_miss);
308 }
309
310 if (m_all_instructions) {
311 // This code is used if the address profiler is an
312 // all-instructions profiler record program counter address
313 // trace info
314 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
315 update(type, access_mode, id, sharing_miss);
316 }
317}
318
319void
320AddressProfiler::profileRetry(Addr data_addr, AccessType type, int count)
321{
322 m_retryProfileHisto.add(count);
323 if (type == AccessType_Read) {
324 m_retryProfileHistoRead.add(count);
325 } else {
326 m_retryProfileHistoWrite.add(count);
327 }
328 if (count > 1) {
329 lookupTraceForAddress(data_addr, m_retryProfileMap).addSample(count);
330 }
331}