AddressProfiler.cc revision 6433:0f0f0fbef977 
1
2/*
3 * Copyright (c) 1999-2008 Mark D. Hill and David A. Wood
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions are
8 * met: redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer;
10 * redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution;
13 * neither the name of the copyright holders nor the names of its
14 * contributors may be used to endorse or promote products derived from
15 * this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
20 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
21 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
22 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
23 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
27 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
30/*
31 * AddressProfiler.cc
32 *
33 * Description: See AddressProfiler.hh
34 *
35 * $Id$
36 *
37 */
38
39#include "mem/ruby/profiler/AddressProfiler.hh"
40#include "mem/protocol/CacheMsg.hh"
41#include "mem/ruby/profiler/AccessTraceForAddress.hh"
42#include "mem/gems_common/PrioHeap.hh"
43#include "mem/gems_common/Map.hh"
44#include "mem/ruby/system/System.hh"
45#include "mem/ruby/profiler/Profiler.hh"
46
47// Helper functions
48static AccessTraceForAddress& lookupTraceForAddress(const Address& addr, Map<Address, AccessTraceForAddress>* record_map);
49static void printSorted(ostream& out, const Map<Address, AccessTraceForAddress>* record_map, string description);
50
51AddressProfiler::AddressProfiler()
52{
53  m_dataAccessTrace = new Map<Address, AccessTraceForAddress>;
54  m_macroBlockAccessTrace = new Map<Address, AccessTraceForAddress>;
55  m_programCounterAccessTrace = new Map<Address, AccessTraceForAddress>;
56  m_retryProfileMap = new Map<Address, AccessTraceForAddress>;
57  clearStats();
58}
59
60AddressProfiler::~AddressProfiler()
61{
62  delete m_dataAccessTrace;
63  delete m_macroBlockAccessTrace;
64  delete m_programCounterAccessTrace;
65  delete m_retryProfileMap;
66}
67
68void AddressProfiler::setHotLines(bool hot_lines){
69  m_hot_lines = hot_lines;
70}
71void AddressProfiler::setAllInstructions(bool all_instructions){
72  m_all_instructions = all_instructions;
73}
74
75void AddressProfiler::printStats(ostream& out) const
76{
77  if (m_hot_lines) {
78    out << endl;
79    out << "AddressProfiler Stats" << endl;
80    out << "---------------------" << endl;
81
82    out << endl;
83    out << "sharing_misses: " << m_sharing_miss_counter << endl;
84    out << "getx_sharing_histogram: " << m_getx_sharing_histogram << endl;
85    out << "gets_sharing_histogram: " << m_gets_sharing_histogram << endl;
86
87    out << endl;
88    out << "Hot Data Blocks" << endl;
89    out << "---------------" << endl;
90    out << endl;
91    printSorted(out, m_dataAccessTrace, "block_address");
92
93    out << endl;
94    out << "Hot MacroData Blocks" << endl;
95    out << "--------------------" << endl;
96    out << endl;
97    printSorted(out, m_macroBlockAccessTrace, "macroblock_address");
98
99    out << "Hot Instructions" << endl;
100    out << "----------------" << endl;
101    out << endl;
102    printSorted(out, m_programCounterAccessTrace, "pc_address");
103  }
104
105  if (m_all_instructions){
106    out << endl;
107    out << "All Instructions Profile:" << endl;
108    out << "-------------------------" << endl;
109    out << endl;
110    printSorted(out, m_programCounterAccessTrace, "pc_address");
111    out << endl;
112  }
113
114  if (m_retryProfileHisto.size() > 0) {
115    out << "Retry Profile" << endl;
116    out << "-------------" << endl;
117    out << endl;
118    out << "retry_histogram_absolute: " << m_retryProfileHisto << endl;
119    out << "retry_histogram_write: " << m_retryProfileHistoWrite << endl;
120    out << "retry_histogram_read: " << m_retryProfileHistoRead << endl;
121
122    out << "retry_histogram_percent: ";
123    m_retryProfileHisto.printPercent(out);
124    out << endl;
125
126    printSorted(out, m_retryProfileMap, "block_address");
127    out << endl;
128  }
129
130}
131
132void AddressProfiler::clearStats()
133{
134  // Clear the maps
135  m_sharing_miss_counter = 0;
136  m_dataAccessTrace->clear();
137  m_macroBlockAccessTrace->clear();
138  m_programCounterAccessTrace->clear();
139  m_retryProfileMap->clear();
140  m_retryProfileHisto.clear();
141  m_retryProfileHistoRead.clear();
142  m_retryProfileHistoWrite.clear();
143  m_getx_sharing_histogram.clear();
144  m_gets_sharing_histogram.clear();
145}
146
147void AddressProfiler::profileGetX(const Address& datablock, const Address& PC, const Set& owner, const Set& sharers, NodeID requestor)
148{
149  Set indirection_set;
150  indirection_set.addSet(sharers);
151  indirection_set.addSet(owner);
152  indirection_set.remove(requestor);
153  int num_indirections = indirection_set.count();
154
155  m_getx_sharing_histogram.add(num_indirections);
156  bool indirection_miss = (num_indirections > 0);
157
158  addTraceSample(datablock, PC, CacheRequestType_ST, AccessModeType(0), requestor, indirection_miss);
159}
160
161void AddressProfiler::profileGetS(const Address& datablock, const Address& PC, const Set& owner, const Set& sharers, NodeID requestor)
162{
163  Set indirection_set;
164  indirection_set.addSet(owner);
165  indirection_set.remove(requestor);
166  int num_indirections = indirection_set.count();
167
168  m_gets_sharing_histogram.add(num_indirections);
169  bool indirection_miss = (num_indirections > 0);
170
171  addTraceSample(datablock, PC, CacheRequestType_LD, AccessModeType(0), requestor, indirection_miss);
172}
173
174void AddressProfiler::addTraceSample(Address data_addr, Address pc_addr, CacheRequestType type, AccessModeType access_mode, NodeID id, bool sharing_miss)
175{
176  if (m_all_instructions) {
177    if (sharing_miss) {
178      m_sharing_miss_counter++;
179    }
180
181    // record data address trace info
182    data_addr.makeLineAddress();
183    lookupTraceForAddress(data_addr, m_dataAccessTrace).update(type, access_mode, id, sharing_miss);
184
185    // record macro data address trace info
186    Address macro_addr(data_addr.maskLowOrderBits(10)); // 6 for datablock, 4 to make it 16x more coarse
187    lookupTraceForAddress(macro_addr, m_macroBlockAccessTrace).update(type, access_mode, id, sharing_miss);
188
189    // record program counter address trace info
190    lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).update(type, access_mode, id, sharing_miss);
191  }
192
193  if (m_all_instructions) {
194    // This code is used if the address profiler is an all-instructions profiler
195    // record program counter address trace info
196    lookupTraceForAddress(pc_addr, m_programCounterAccessTrace).update(type, access_mode, id, sharing_miss);
197  }
198}
199
200void AddressProfiler::profileRetry(const Address& data_addr, AccessType type, int count)
201{
202  m_retryProfileHisto.add(count);
203  if (type == AccessType_Read) {
204    m_retryProfileHistoRead.add(count);
205  } else {
206    m_retryProfileHistoWrite.add(count);
207  }
208  if (count > 1) {
209    lookupTraceForAddress(data_addr, m_retryProfileMap).addSample(count);
210  }
211}
212
213// ***** Normal Functions ******
214
215static void printSorted(ostream& out, const Map<Address, AccessTraceForAddress>* record_map, string description)
216{
217  const int records_printed = 100;
218
219  uint64 misses = 0;
220  PrioHeap<AccessTraceForAddress*> heap;
221  Vector<Address> keys = record_map->keys();
222  for(int i=0; i<keys.size(); i++){
223    AccessTraceForAddress* record = &(record_map->lookup(keys[i]));
224    misses += record->getTotal();
225    heap.insert(record);
226  }
227
228  out << "Total_entries_" << description << ": " << keys.size() << endl;
229  if (g_system_ptr->getProfiler()->getAllInstructions())
230    out << "Total_Instructions_" << description << ": " << misses << endl;
231  else
232    out << "Total_data_misses_" << description << ": " << misses << endl;
233
234  out << "total | load store atomic | user supervisor | sharing | touched-by" << endl;
235
236  Histogram remaining_records(1, 100);
237  Histogram all_records(1, 100);
238  Histogram remaining_records_log(-1);
239  Histogram all_records_log(-1);
240
241  // Allows us to track how many lines where touched by n processors
242  Vector<int64> m_touched_vec;
243  Vector<int64> m_touched_weighted_vec;
244  m_touched_vec.setSize(RubySystem::getNumberOfSequencers()+1);
245  m_touched_weighted_vec.setSize(RubySystem::getNumberOfSequencers()+1);
246  for (int i=0; i<m_touched_vec.size(); i++) {
247    m_touched_vec[i] = 0;
248    m_touched_weighted_vec[i] = 0;
249  }
250
251  int counter = 0;
252  while((heap.size() > 0) && (counter < records_printed)) {
253    AccessTraceForAddress* record = heap.extractMin();
254    double percent = 100.0*(record->getTotal()/double(misses));
255    out << description << " | " << percent << " % " << *record << endl;
256    all_records.add(record->getTotal());
257    all_records_log.add(record->getTotal());
258    counter++;
259    m_touched_vec[record->getTouchedBy()]++;
260    m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
261  }
262
263  while(heap.size() > 0) {
264    AccessTraceForAddress* record = heap.extractMin();
265    all_records.add(record->getTotal());
266    remaining_records.add(record->getTotal());
267    all_records_log.add(record->getTotal());
268    remaining_records_log.add(record->getTotal());
269    m_touched_vec[record->getTouchedBy()]++;
270    m_touched_weighted_vec[record->getTouchedBy()] += record->getTotal();
271  }
272  out << endl;
273  out << "all_records_" << description << ": " << all_records << endl;
274  out << "all_records_log_" << description << ": " << all_records_log << endl;
275  out << "remaining_records_" << description << ": " << remaining_records << endl;
276  out << "remaining_records_log_" << description << ": " << remaining_records_log << endl;
277  out << "touched_by_" << description << ": " << m_touched_vec << endl;
278  out << "touched_by_weighted_" << description << ": " << m_touched_weighted_vec << endl;
279  out << endl;
280}
281
282static AccessTraceForAddress& lookupTraceForAddress(const Address& addr, Map<Address, AccessTraceForAddress>* record_map)
283{
284  if(record_map->exist(addr) == false){
285    record_map->add(addr, AccessTraceForAddress(addr));
286  }
287  return record_map->lookup(addr);
288}
289