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 <vector>
30
31#include "base/stl_helpers.hh"
32#include "mem/protocol/RubyRequest.hh"
33#include "mem/ruby/profiler/AddressProfiler.hh"
34#include "mem/ruby/profiler/Profiler.hh"
35#include "mem/ruby/system/System.hh"
36
37using namespace std;
38typedef AddressProfiler::AddressMap AddressMap;
39
40using m5::stl_helpers::operator<<;
41
42// Helper functions
43AccessTraceForAddress&
44lookupTraceForAddress(const Address& 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)
68{
69 const int records_printed = 100;
70
71 uint64 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 (g_system_ptr->getProfiler()->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> m_touched_vec;
100 std::vector<int64> 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 i = 0; i < m_touched_vec.size(); i++) {
104 m_touched_vec[i] = 0;
105 m_touched_weighted_vec[i] = 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)
147{
148 m_num_of_sequencers = num_of_sequencers;
149 clearStats();
150}
151
152AddressProfiler::~AddressProfiler()
153{
154}
155
156void
157AddressProfiler::setHotLines(bool hot_lines)
158{
159 m_hot_lines = hot_lines;
160}
161
162void
163AddressProfiler::setAllInstructions(bool all_instructions)
164{
165 m_all_instructions = all_instructions;
166}
167
168void
169AddressProfiler::printStats(ostream& out) const
170{
171 if (m_hot_lines) {
172 out << endl;
173 out << "AddressProfiler Stats" << endl;
174 out << "---------------------" << endl;
175
176 out << endl;
177 out << "sharing_misses: " << m_sharing_miss_counter << endl;
178 out << "getx_sharing_histogram: " << m_getx_sharing_histogram << endl;
179 out << "gets_sharing_histogram: " << m_gets_sharing_histogram << endl;
180
181 out << endl;
182 out << "Hot Data Blocks" << endl;
183 out << "---------------" << endl;
184 out << endl;
185 printSorted(out, m_num_of_sequencers, m_dataAccessTrace,
186 "block_address");
187
188 out << endl;
189 out << "Hot MacroData Blocks" << endl;
190 out << "--------------------" << endl;
191 out << endl;
192 printSorted(out, m_num_of_sequencers, m_macroBlockAccessTrace,
193 "macroblock_address");
194
195 out << "Hot Instructions" << endl;
196 out << "----------------" << endl;
197 out << endl;
198 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
199 "pc_address");
200 }
201
202 if (m_all_instructions) {
203 out << endl;
204 out << "All Instructions Profile:" << endl;
205 out << "-------------------------" << endl;
206 out << endl;
207 printSorted(out, m_num_of_sequencers, m_programCounterAccessTrace,
208 "pc_address");
209 out << endl;
210 }
211
212 if (m_retryProfileHisto.size() > 0) {
213 out << "Retry Profile" << endl;
214 out << "-------------" << endl;
215 out << endl;
216 out << "retry_histogram_absolute: " << m_retryProfileHisto << endl;
217 out << "retry_histogram_write: " << m_retryProfileHistoWrite << endl;
218 out << "retry_histogram_read: " << m_retryProfileHistoRead << endl;
219
220 out << "retry_histogram_percent: ";
221 m_retryProfileHisto.printPercent(out);
222 out << endl;
223
224 printSorted(out, m_num_of_sequencers, m_retryProfileMap,
225 "block_address");
226 out << endl;
227 }
228}
229
230void
231AddressProfiler::clearStats()
232{
233 // Clear the maps
234 m_sharing_miss_counter = 0;
235 m_dataAccessTrace.clear();
236 m_macroBlockAccessTrace.clear();
237 m_programCounterAccessTrace.clear();
238 m_retryProfileMap.clear();
239 m_retryProfileHisto.clear();
240 m_retryProfileHistoRead.clear();
241 m_retryProfileHistoWrite.clear();
242 m_getx_sharing_histogram.clear();
243 m_gets_sharing_histogram.clear();
244}
245
246void
247AddressProfiler::profileGetX(const Address& datablock, const Address& PC,
248 const Set& owner, const Set& sharers,
249 NodeID requestor)
250{
251 Set indirection_set;
252 indirection_set.addSet(sharers);
253 indirection_set.addSet(owner);
254 indirection_set.remove(requestor);
255 int num_indirections = indirection_set.count();
256
257 m_getx_sharing_histogram.add(num_indirections);
258 bool indirection_miss = (num_indirections > 0);
259
260 addTraceSample(datablock, PC, RubyRequestType_ST, RubyAccessMode(0),
261 requestor, indirection_miss);
262}
263
264void
265AddressProfiler::profileGetS(const Address& datablock, const Address& PC,
266 const Set& owner, const Set& sharers,
267 NodeID requestor)
268{
269 Set indirection_set;
270 indirection_set.addSet(owner);
271 indirection_set.remove(requestor);
272 int num_indirections = indirection_set.count();
273
274 m_gets_sharing_histogram.add(num_indirections);
275 bool indirection_miss = (num_indirections > 0);
276
277 addTraceSample(datablock, PC, RubyRequestType_LD, RubyAccessMode(0),
278 requestor, indirection_miss);
279}
280
281void
282AddressProfiler::addTraceSample(Address data_addr, Address pc_addr,
283 RubyRequestType type,
284 RubyAccessMode access_mode, NodeID id,
285 bool sharing_miss)
286{
287 if (m_all_instructions) {
288 if (sharing_miss) {
289 m_sharing_miss_counter++;
290 }
291
292 // record data address trace info
293 data_addr.makeLineAddress();
294 lookupTraceForAddress(data_addr, m_dataAccessTrace).
295 update(type, access_mode, id, sharing_miss);
296
297 // record macro data address trace info
298
299 // 6 for datablock, 4 to make it 16x more coarse
300 Address macro_addr(data_addr.maskLowOrderBits(10));
301 lookupTraceForAddress(macro_addr, m_macroBlockAccessTrace).
302 update(type, access_mode, id, sharing_miss);
303
304 // record program counter address trace info
305 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
306 update(type, access_mode, id, sharing_miss);
307 }
308
309 if (m_all_instructions) {
310 // This code is used if the address profiler is an
311 // all-instructions profiler record program counter address
312 // trace info
313 lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).
314 update(type, access_mode, id, sharing_miss);
315 }
316}
317
318void
319AddressProfiler::profileRetry(const Address& data_addr, AccessType type,
320 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}