BranchPredictor.py (13957:25e9c77a8a99) BranchPredictor.py (14034:937e704c6807)
1# Copyright (c) 2012 Mark D. Hill and David A. Wood
2# Copyright (c) 2015 The University of Wisconsin
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# Authors: Nilay Vaish and Dibakar Gope
29
30from m5.SimObject import SimObject
31from m5.params import *
32from m5.proxy import *
33
34class IndirectPredictor(SimObject):
35 type = 'IndirectPredictor'
36 cxx_class = 'IndirectPredictor'
37 cxx_header = "cpu/pred/indirect.hh"
38 abstract = True
39
40 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
41
42class SimpleIndirectPredictor(IndirectPredictor):
43 type = 'SimpleIndirectPredictor'
44 cxx_class = 'SimpleIndirectPredictor'
45 cxx_header = "cpu/pred/simple_indirect.hh"
46
47 indirectHashGHR = Param.Bool(True, "Hash branch predictor GHR")
48 indirectHashTargets = Param.Bool(True, "Hash path history targets")
49 indirectSets = Param.Unsigned(256, "Cache sets for indirect predictor")
50 indirectWays = Param.Unsigned(2, "Ways for indirect predictor")
51 indirectTagSize = Param.Unsigned(16, "Indirect target cache tag bits")
52 indirectPathLength = Param.Unsigned(3,
53 "Previous indirect targets to use for path history")
54 indirectGHRBits = Param.Unsigned(13, "Indirect GHR number of bits")
55 instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")
56
57class BranchPredictor(SimObject):
58 type = 'BranchPredictor'
59 cxx_class = 'BPredUnit'
60 cxx_header = "cpu/pred/bpred_unit.hh"
61 abstract = True
62
63 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
64 BTBEntries = Param.Unsigned(4096, "Number of BTB entries")
65 BTBTagSize = Param.Unsigned(16, "Size of the BTB tags, in bits")
66 RASSize = Param.Unsigned(16, "RAS size")
67 instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")
68
69 indirectBranchPred = Param.IndirectPredictor(SimpleIndirectPredictor(),
70 "Indirect branch predictor, set to NULL to disable indirect predictions")
71
72class LocalBP(BranchPredictor):
73 type = 'LocalBP'
74 cxx_class = 'LocalBP'
75 cxx_header = "cpu/pred/2bit_local.hh"
76
77 localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
78 localCtrBits = Param.Unsigned(2, "Bits per counter")
79
80
81class TournamentBP(BranchPredictor):
82 type = 'TournamentBP'
83 cxx_class = 'TournamentBP'
84 cxx_header = "cpu/pred/tournament.hh"
85
86 localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
87 localCtrBits = Param.Unsigned(2, "Bits per counter")
88 localHistoryTableSize = Param.Unsigned(2048, "size of local history table")
89 globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
90 globalCtrBits = Param.Unsigned(2, "Bits per counter")
91 choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
92 choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")
93
94
95class BiModeBP(BranchPredictor):
96 type = 'BiModeBP'
97 cxx_class = 'BiModeBP'
98 cxx_header = "cpu/pred/bi_mode.hh"
99
100 globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
101 globalCtrBits = Param.Unsigned(2, "Bits per counter")
102 choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
103 choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")
104
105class TAGEBase(SimObject):
106 type = 'TAGEBase'
107 cxx_class = 'TAGEBase'
108 cxx_header = "cpu/pred/tage_base.hh"
109
110 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
111 instShiftAmt = Param.Unsigned(Parent.instShiftAmt,
112 "Number of bits to shift instructions by")
113
114 nHistoryTables = Param.Unsigned(7, "Number of history tables")
115 minHist = Param.Unsigned(5, "Minimum history size of TAGE")
116 maxHist = Param.Unsigned(130, "Maximum history size of TAGE")
117
118 tagTableTagWidths = VectorParam.Unsigned(
119 [0, 9, 9, 10, 10, 11, 11, 12], "Tag size in TAGE tag tables")
120 logTagTableSizes = VectorParam.Int(
121 [13, 9, 9, 9, 9, 9, 9, 9], "Log2 of TAGE table sizes")
122 logRatioBiModalHystEntries = Param.Unsigned(2,
123 "Log num of prediction entries for a shared hysteresis bit " \
124 "for the Bimodal")
125
126 tagTableCounterBits = Param.Unsigned(3, "Number of tag table counter bits")
127 tagTableUBits = Param.Unsigned(2, "Number of tag table u bits")
128
129 histBufferSize = Param.Unsigned(2097152,
130 "A large number to track all branch histories(2MEntries default)")
131
132 pathHistBits = Param.Unsigned(16, "Path history size")
133 logUResetPeriod = Param.Unsigned(18,
134 "Log period in number of branches to reset TAGE useful counters")
135 numUseAltOnNa = Param.Unsigned(1, "Number of USE_ALT_ON_NA counters")
136 useAltOnNaBits = Param.Unsigned(4, "Size of the USE_ALT_ON_NA counter(s)")
137
138 maxNumAlloc = Param.Unsigned(1,
139 "Max number of TAGE entries allocted on mispredict")
140
141 # List of enabled TAGE tables. If empty, all are enabled
142 noSkip = VectorParam.Bool([], "Vector of enabled TAGE tables")
143
144 speculativeHistUpdate = Param.Bool(True,
145 "Use speculative update for histories")
146
147# TAGE branch predictor as described in https://www.jilp.org/vol8/v8paper1.pdf
148# The default sizes below are for the 8C-TAGE configuration (63.5 Kbits)
149class TAGE(BranchPredictor):
150 type = 'TAGE'
151 cxx_class = 'TAGE'
152 cxx_header = "cpu/pred/tage.hh"
153 tage = Param.TAGEBase(TAGEBase(), "Tage object")
154
155class LTAGE_TAGE(TAGEBase):
156 nHistoryTables = 12
157 minHist = 4
158 maxHist = 640
159 tagTableTagWidths = [0, 7, 7, 8, 8, 9, 10, 11, 12, 12, 13, 14, 15]
160 logTagTableSizes = [14, 10, 10, 11, 11, 11, 11, 10, 10, 10, 10, 9, 9]
161 logUResetPeriod = 19
162
163class LoopPredictor(SimObject):
164 type = 'LoopPredictor'
165 cxx_class = 'LoopPredictor'
166 cxx_header = 'cpu/pred/loop_predictor.hh'
167
168 logSizeLoopPred = Param.Unsigned(8, "Log size of the loop predictor")
169 withLoopBits = Param.Unsigned(7, "Size of the WITHLOOP counter")
170 loopTableAgeBits = Param.Unsigned(8, "Number of age bits per loop entry")
171 loopTableConfidenceBits = Param.Unsigned(2,
172 "Number of confidence bits per loop entry")
173 loopTableTagBits = Param.Unsigned(14, "Number of tag bits per loop entry")
174 loopTableIterBits = Param.Unsigned(14, "Nuber of iteration bits per loop")
175 logLoopTableAssoc = Param.Unsigned(2, "Log loop predictor associativity")
176
177 # Parameters for enabling modifications to the loop predictor
178 # They have been copied from TAGE-GSC-IMLI
179 # (http://www.irisa.fr/alf/downloads/seznec/TAGE-GSC-IMLI.tar)
180 #
181 # All of them should be disabled to match the original LTAGE implementation
182 # (http://hpca23.cse.tamu.edu/taco/camino/cbp2/cbp-src/realistic-seznec.h)
183
184 # Add speculation
185 useSpeculation = Param.Bool(False, "Use speculation")
186
187 # Add hashing for calculating the loop table index
188 useHashing = Param.Bool(False, "Use hashing")
189
190 # Add a direction bit to the loop table entries
191 useDirectionBit = Param.Bool(False, "Use direction info")
192
193 # If true, use random to decide whether to allocate or not, and only try
194 # with one entry
195 restrictAllocation = Param.Bool(False,
196 "Restrict the allocation conditions")
197
198 initialLoopIter = Param.Unsigned(1, "Initial iteration number")
199 initialLoopAge = Param.Unsigned(255, "Initial age value")
200 optionalAgeReset = Param.Bool(True,
201 "Reset age bits optionally in some cases")
202
203class TAGE_SC_L_TAGE(TAGEBase):
204 type = 'TAGE_SC_L_TAGE'
205 cxx_class = 'TAGE_SC_L_TAGE'
206 cxx_header = "cpu/pred/tage_sc_l.hh"
207 abstract = True
208 tagTableTagWidths = [0]
209 numUseAltOnNa = 16
210 pathHistBits = 27
211 maxNumAlloc = 2
212 logUResetPeriod = 10
213 useAltOnNaBits = 5
214 # TODO No speculation implemented as of now
215 speculativeHistUpdate = False
216
217 # This size does not set the final sizes of the tables (it is just used
218 # for some calculations)
219 # Instead, the number of TAGE entries comes from shortTagsTageEntries and
220 # longTagsTageEntries
221 logTagTableSize = Param.Unsigned("Log size of each tag table")
222
223 shortTagsTageFactor = Param.Unsigned(
224 "Factor for calculating the total number of short tags TAGE entries")
225
226 longTagsTageFactor = Param.Unsigned(
227 "Factor for calculating the total number of long tags TAGE entries")
228
229 shortTagsSize = Param.Unsigned(8, "Size of the short tags")
230
231 longTagsSize = Param.Unsigned("Size of the long tags")
232
233 firstLongTagTable = Param.Unsigned("First table with long tags")
234
235 truncatePathHist = Param.Bool(True,
236 "Truncate the path history to its configured size")
237
238
239class TAGE_SC_L_TAGE_64KB(TAGE_SC_L_TAGE):
240 type = 'TAGE_SC_L_TAGE_64KB'
241 cxx_class = 'TAGE_SC_L_TAGE_64KB'
242 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
243 nHistoryTables = 36
244
245 minHist = 6
246 maxHist = 3000
247
248 tagTableUBits = 1
249
250 logTagTableSizes = [13]
251
252 # This is used to handle the 2-way associativity
253 # (all odd entries are set to one, and if the corresponding even entry
254 # is set to one, then there is a 2-way associativity for this pair)
255 # Entry 0 is for the bimodal and it is ignored
256 # Note: For this implementation, some odd entries are also set to 0 to save
257 # some bits
258 noSkip = [0,0,1,0,0,0,1,0,0,1,1,1,1,1,1,1,1,1,1,
259 1,1,1,1,0,1,0,1,0,1,0,0,0,1,0,0,0,1]
260
261 logTagTableSize = 10
262 shortTagsTageFactor = 10
263 longTagsTageFactor = 20
264
265 longTagsSize = 12
266
267 firstLongTagTable = 13
268
269class TAGE_SC_L_TAGE_8KB(TAGE_SC_L_TAGE):
270 type = 'TAGE_SC_L_TAGE_8KB'
271 cxx_class = 'TAGE_SC_L_TAGE_8KB'
272 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
273
274 nHistoryTables = 30
275
276 minHist = 4
277 maxHist = 1000
278
279 logTagTableSize = 7
280 shortTagsTageFactor = 9
281 longTagsTageFactor = 17
282 longTagsSize = 12
283
284 logTagTableSizes = [12]
285
286 firstLongTagTable = 11
287
288 truncatePathHist = False
289
290 noSkip = [0,0,1,0,1,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,0,1,0,1,0,1,0,1]
291
292 tagTableUBits = 2
293
294# LTAGE branch predictor as described in
295# https://www.irisa.fr/caps/people/seznec/L-TAGE.pdf
296# It is basically a TAGE predictor plus a loop predictor
297# The differnt TAGE sizes are updated according to the paper values (256 Kbits)
298class LTAGE(TAGE):
299 type = 'LTAGE'
300 cxx_class = 'LTAGE'
301 cxx_header = "cpu/pred/ltage.hh"
302
303 tage = LTAGE_TAGE()
304
305 loop_predictor = Param.LoopPredictor(LoopPredictor(), "Loop predictor")
306
307class TAGE_SC_L_LoopPredictor(LoopPredictor):
308 type = 'TAGE_SC_L_LoopPredictor'
309 cxx_class = 'TAGE_SC_L_LoopPredictor'
310 cxx_header = "cpu/pred/tage_sc_l.hh"
311 loopTableAgeBits = 4
312 loopTableConfidenceBits = 4
313 loopTableTagBits = 10
314 loopTableIterBits = 10
315 useSpeculation = False
316 useHashing = True
317 useDirectionBit = True
318 restrictAllocation = True
319 initialLoopIter = 0
320 initialLoopAge = 7
321 optionalAgeReset = False
322
323class StatisticalCorrector(SimObject):
324 type = 'StatisticalCorrector'
325 cxx_class = 'StatisticalCorrector'
326 cxx_header = "cpu/pred/statistical_corrector.hh"
327 abstract = True
328
329 # Statistical corrector parameters
330
331 numEntriesFirstLocalHistories = Param.Unsigned(
332 "Number of entries for first local histories")
333
334 bwnb = Param.Unsigned("Num global backward branch GEHL lengths")
335 bwm = VectorParam.Int("Global backward branch GEHL lengths")
336 logBwnb = Param.Unsigned("Log num of global backward branch GEHL entries")
337
338 lnb = Param.Unsigned("Num first local history GEHL lenghts")
339 lm = VectorParam.Int("First local history GEHL lengths")
340 logLnb = Param.Unsigned("Log number of first local history GEHL entries")
341
342 inb = Param.Unsigned(1, "Num IMLI GEHL lenghts")
343 im = VectorParam.Int([8], "IMLI history GEHL lengths")
344 logInb = Param.Unsigned("Log number of IMLI GEHL entries")
345
346 logBias = Param.Unsigned("Log size of Bias tables")
347
348 logSizeUp = Param.Unsigned(6,
349 "Log size of update threshold counters tables")
350
351 chooserConfWidth = Param.Unsigned(7,
352 "Number of bits for the chooser counters")
353
354 updateThresholdWidth = Param.Unsigned(12,
355 "Number of bits for the update threshold counter")
356
357 pUpdateThresholdWidth = Param.Unsigned(8,
358 "Number of bits for the pUpdate threshold counters")
359
360 extraWeightsWidth = Param.Unsigned(6,
361 "Number of bits for the extra weights")
362
363 scCountersWidth = Param.Unsigned(6, "Statistical corrector counters width")
364
365# TAGE-SC-L branch predictor as desribed in
366# https://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
367# It is a modified LTAGE predictor plus a statistical corrector predictor
368# The TAGE modifications include bank interleaving and partial associativity
369# Two different sizes are proposed in the paper:
370# 8KB => See TAGE_SC_L_8KB below
371# 64KB => See TAGE_SC_L_64KB below
372# The TAGE_SC_L_8KB and TAGE_SC_L_64KB classes differ not only on the values
373# of some parameters, but also in some implementation details
374# Given this, the TAGE_SC_L class is left abstract
375# Note that as it is now, this branch predictor does not handle any type
376# of speculation: All the structures/histories are updated at commit time
377class TAGE_SC_L(LTAGE):
378 type = 'TAGE_SC_L'
379 cxx_class = 'TAGE_SC_L'
380 cxx_header = "cpu/pred/tage_sc_l.hh"
381 abstract = True
382
383 statistical_corrector = Param.StatisticalCorrector(
384 "Statistical Corrector")
385
386class TAGE_SC_L_64KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
387 logSizeLoopPred = 5
388
389class TAGE_SC_L_8KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
390 logSizeLoopPred = 3
391
392class TAGE_SC_L_64KB_StatisticalCorrector(StatisticalCorrector):
393 type = 'TAGE_SC_L_64KB_StatisticalCorrector'
394 cxx_class = 'TAGE_SC_L_64KB_StatisticalCorrector'
395 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
396
397 pnb = Param.Unsigned(3, "Num variation global branch GEHL lengths")
398 pm = VectorParam.Int([25, 16, 9], "Variation global branch GEHL lengths")
399 logPnb = Param.Unsigned(9,
400 "Log number of variation global branch GEHL entries")
401
402 snb = Param.Unsigned(3, "Num second local history GEHL lenghts")
403 sm = VectorParam.Int([16, 11, 6], "Second local history GEHL lengths")
404 logSnb = Param.Unsigned(9,
405 "Log number of second local history GEHL entries")
406
407 tnb = Param.Unsigned(2, "Num third local history GEHL lenghts")
408 tm = VectorParam.Int([9, 4], "Third local history GEHL lengths")
409 logTnb = Param.Unsigned(10,
410 "Log number of third local history GEHL entries")
411
412 imnb = Param.Unsigned(2, "Num second IMLI GEHL lenghts")
413 imm = VectorParam.Int([10, 4], "Second IMLI history GEHL lengths")
414 logImnb = Param.Unsigned(9, "Log number of second IMLI GEHL entries")
415
416 numEntriesSecondLocalHistories = Param.Unsigned(16,
417 "Number of entries for second local histories")
418 numEntriesThirdLocalHistories = Param.Unsigned(16,
419 "Number of entries for second local histories")
420
421 numEntriesFirstLocalHistories = 256
422
423 logBias = 8
424
425 bwnb = 3
426 bwm = [40, 24, 10]
427 logBwnb = 10
428
429 lnb = 3
430 lm = [11, 6, 3]
431 logLnb = 10
432
433 logInb = 8
434
435class TAGE_SC_L_8KB_StatisticalCorrector(StatisticalCorrector):
436 type = 'TAGE_SC_L_8KB_StatisticalCorrector'
437 cxx_class = 'TAGE_SC_L_8KB_StatisticalCorrector'
438 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
439 gnb = Param.Unsigned(2, "Num global branch GEHL lengths")
440 gm = VectorParam.Int([6, 3], "Global branch GEHL lengths")
441 logGnb = Param.Unsigned(7, "Log number of global branch GEHL entries")
442
443 numEntriesFirstLocalHistories = 64
444
445 logBias = 7
446
447 bwnb = 2
448 logBwnb = 7
449 bwm = [16, 8]
450
451 lnb = 2
452 logLnb = 7
453 lm = [6, 3]
454
455 logInb = 7
456
457# 64KB TAGE-SC-L branch predictor as described in
458# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
459class TAGE_SC_L_64KB(TAGE_SC_L):
460 type = 'TAGE_SC_L_64KB'
461 cxx_class = 'TAGE_SC_L_64KB'
462 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
463
464 tage = TAGE_SC_L_TAGE_64KB()
465 loop_predictor = TAGE_SC_L_64KB_LoopPredictor()
466 statistical_corrector = TAGE_SC_L_64KB_StatisticalCorrector()
467
468# 8KB TAGE-SC-L branch predictor as described in
469# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
470class TAGE_SC_L_8KB(TAGE_SC_L):
471 type = 'TAGE_SC_L_8KB'
472 cxx_class = 'TAGE_SC_L_8KB'
473 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
474
475 tage = TAGE_SC_L_TAGE_8KB()
476 loop_predictor = TAGE_SC_L_8KB_LoopPredictor()
477 statistical_corrector = TAGE_SC_L_8KB_StatisticalCorrector()
1# Copyright (c) 2012 Mark D. Hill and David A. Wood
2# Copyright (c) 2015 The University of Wisconsin
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# Authors: Nilay Vaish and Dibakar Gope
29
30from m5.SimObject import SimObject
31from m5.params import *
32from m5.proxy import *
33
34class IndirectPredictor(SimObject):
35 type = 'IndirectPredictor'
36 cxx_class = 'IndirectPredictor'
37 cxx_header = "cpu/pred/indirect.hh"
38 abstract = True
39
40 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
41
42class SimpleIndirectPredictor(IndirectPredictor):
43 type = 'SimpleIndirectPredictor'
44 cxx_class = 'SimpleIndirectPredictor'
45 cxx_header = "cpu/pred/simple_indirect.hh"
46
47 indirectHashGHR = Param.Bool(True, "Hash branch predictor GHR")
48 indirectHashTargets = Param.Bool(True, "Hash path history targets")
49 indirectSets = Param.Unsigned(256, "Cache sets for indirect predictor")
50 indirectWays = Param.Unsigned(2, "Ways for indirect predictor")
51 indirectTagSize = Param.Unsigned(16, "Indirect target cache tag bits")
52 indirectPathLength = Param.Unsigned(3,
53 "Previous indirect targets to use for path history")
54 indirectGHRBits = Param.Unsigned(13, "Indirect GHR number of bits")
55 instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")
56
57class BranchPredictor(SimObject):
58 type = 'BranchPredictor'
59 cxx_class = 'BPredUnit'
60 cxx_header = "cpu/pred/bpred_unit.hh"
61 abstract = True
62
63 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
64 BTBEntries = Param.Unsigned(4096, "Number of BTB entries")
65 BTBTagSize = Param.Unsigned(16, "Size of the BTB tags, in bits")
66 RASSize = Param.Unsigned(16, "RAS size")
67 instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")
68
69 indirectBranchPred = Param.IndirectPredictor(SimpleIndirectPredictor(),
70 "Indirect branch predictor, set to NULL to disable indirect predictions")
71
72class LocalBP(BranchPredictor):
73 type = 'LocalBP'
74 cxx_class = 'LocalBP'
75 cxx_header = "cpu/pred/2bit_local.hh"
76
77 localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
78 localCtrBits = Param.Unsigned(2, "Bits per counter")
79
80
81class TournamentBP(BranchPredictor):
82 type = 'TournamentBP'
83 cxx_class = 'TournamentBP'
84 cxx_header = "cpu/pred/tournament.hh"
85
86 localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
87 localCtrBits = Param.Unsigned(2, "Bits per counter")
88 localHistoryTableSize = Param.Unsigned(2048, "size of local history table")
89 globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
90 globalCtrBits = Param.Unsigned(2, "Bits per counter")
91 choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
92 choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")
93
94
95class BiModeBP(BranchPredictor):
96 type = 'BiModeBP'
97 cxx_class = 'BiModeBP'
98 cxx_header = "cpu/pred/bi_mode.hh"
99
100 globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
101 globalCtrBits = Param.Unsigned(2, "Bits per counter")
102 choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
103 choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")
104
105class TAGEBase(SimObject):
106 type = 'TAGEBase'
107 cxx_class = 'TAGEBase'
108 cxx_header = "cpu/pred/tage_base.hh"
109
110 numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
111 instShiftAmt = Param.Unsigned(Parent.instShiftAmt,
112 "Number of bits to shift instructions by")
113
114 nHistoryTables = Param.Unsigned(7, "Number of history tables")
115 minHist = Param.Unsigned(5, "Minimum history size of TAGE")
116 maxHist = Param.Unsigned(130, "Maximum history size of TAGE")
117
118 tagTableTagWidths = VectorParam.Unsigned(
119 [0, 9, 9, 10, 10, 11, 11, 12], "Tag size in TAGE tag tables")
120 logTagTableSizes = VectorParam.Int(
121 [13, 9, 9, 9, 9, 9, 9, 9], "Log2 of TAGE table sizes")
122 logRatioBiModalHystEntries = Param.Unsigned(2,
123 "Log num of prediction entries for a shared hysteresis bit " \
124 "for the Bimodal")
125
126 tagTableCounterBits = Param.Unsigned(3, "Number of tag table counter bits")
127 tagTableUBits = Param.Unsigned(2, "Number of tag table u bits")
128
129 histBufferSize = Param.Unsigned(2097152,
130 "A large number to track all branch histories(2MEntries default)")
131
132 pathHistBits = Param.Unsigned(16, "Path history size")
133 logUResetPeriod = Param.Unsigned(18,
134 "Log period in number of branches to reset TAGE useful counters")
135 numUseAltOnNa = Param.Unsigned(1, "Number of USE_ALT_ON_NA counters")
136 useAltOnNaBits = Param.Unsigned(4, "Size of the USE_ALT_ON_NA counter(s)")
137
138 maxNumAlloc = Param.Unsigned(1,
139 "Max number of TAGE entries allocted on mispredict")
140
141 # List of enabled TAGE tables. If empty, all are enabled
142 noSkip = VectorParam.Bool([], "Vector of enabled TAGE tables")
143
144 speculativeHistUpdate = Param.Bool(True,
145 "Use speculative update for histories")
146
147# TAGE branch predictor as described in https://www.jilp.org/vol8/v8paper1.pdf
148# The default sizes below are for the 8C-TAGE configuration (63.5 Kbits)
149class TAGE(BranchPredictor):
150 type = 'TAGE'
151 cxx_class = 'TAGE'
152 cxx_header = "cpu/pred/tage.hh"
153 tage = Param.TAGEBase(TAGEBase(), "Tage object")
154
155class LTAGE_TAGE(TAGEBase):
156 nHistoryTables = 12
157 minHist = 4
158 maxHist = 640
159 tagTableTagWidths = [0, 7, 7, 8, 8, 9, 10, 11, 12, 12, 13, 14, 15]
160 logTagTableSizes = [14, 10, 10, 11, 11, 11, 11, 10, 10, 10, 10, 9, 9]
161 logUResetPeriod = 19
162
163class LoopPredictor(SimObject):
164 type = 'LoopPredictor'
165 cxx_class = 'LoopPredictor'
166 cxx_header = 'cpu/pred/loop_predictor.hh'
167
168 logSizeLoopPred = Param.Unsigned(8, "Log size of the loop predictor")
169 withLoopBits = Param.Unsigned(7, "Size of the WITHLOOP counter")
170 loopTableAgeBits = Param.Unsigned(8, "Number of age bits per loop entry")
171 loopTableConfidenceBits = Param.Unsigned(2,
172 "Number of confidence bits per loop entry")
173 loopTableTagBits = Param.Unsigned(14, "Number of tag bits per loop entry")
174 loopTableIterBits = Param.Unsigned(14, "Nuber of iteration bits per loop")
175 logLoopTableAssoc = Param.Unsigned(2, "Log loop predictor associativity")
176
177 # Parameters for enabling modifications to the loop predictor
178 # They have been copied from TAGE-GSC-IMLI
179 # (http://www.irisa.fr/alf/downloads/seznec/TAGE-GSC-IMLI.tar)
180 #
181 # All of them should be disabled to match the original LTAGE implementation
182 # (http://hpca23.cse.tamu.edu/taco/camino/cbp2/cbp-src/realistic-seznec.h)
183
184 # Add speculation
185 useSpeculation = Param.Bool(False, "Use speculation")
186
187 # Add hashing for calculating the loop table index
188 useHashing = Param.Bool(False, "Use hashing")
189
190 # Add a direction bit to the loop table entries
191 useDirectionBit = Param.Bool(False, "Use direction info")
192
193 # If true, use random to decide whether to allocate or not, and only try
194 # with one entry
195 restrictAllocation = Param.Bool(False,
196 "Restrict the allocation conditions")
197
198 initialLoopIter = Param.Unsigned(1, "Initial iteration number")
199 initialLoopAge = Param.Unsigned(255, "Initial age value")
200 optionalAgeReset = Param.Bool(True,
201 "Reset age bits optionally in some cases")
202
203class TAGE_SC_L_TAGE(TAGEBase):
204 type = 'TAGE_SC_L_TAGE'
205 cxx_class = 'TAGE_SC_L_TAGE'
206 cxx_header = "cpu/pred/tage_sc_l.hh"
207 abstract = True
208 tagTableTagWidths = [0]
209 numUseAltOnNa = 16
210 pathHistBits = 27
211 maxNumAlloc = 2
212 logUResetPeriod = 10
213 useAltOnNaBits = 5
214 # TODO No speculation implemented as of now
215 speculativeHistUpdate = False
216
217 # This size does not set the final sizes of the tables (it is just used
218 # for some calculations)
219 # Instead, the number of TAGE entries comes from shortTagsTageEntries and
220 # longTagsTageEntries
221 logTagTableSize = Param.Unsigned("Log size of each tag table")
222
223 shortTagsTageFactor = Param.Unsigned(
224 "Factor for calculating the total number of short tags TAGE entries")
225
226 longTagsTageFactor = Param.Unsigned(
227 "Factor for calculating the total number of long tags TAGE entries")
228
229 shortTagsSize = Param.Unsigned(8, "Size of the short tags")
230
231 longTagsSize = Param.Unsigned("Size of the long tags")
232
233 firstLongTagTable = Param.Unsigned("First table with long tags")
234
235 truncatePathHist = Param.Bool(True,
236 "Truncate the path history to its configured size")
237
238
239class TAGE_SC_L_TAGE_64KB(TAGE_SC_L_TAGE):
240 type = 'TAGE_SC_L_TAGE_64KB'
241 cxx_class = 'TAGE_SC_L_TAGE_64KB'
242 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
243 nHistoryTables = 36
244
245 minHist = 6
246 maxHist = 3000
247
248 tagTableUBits = 1
249
250 logTagTableSizes = [13]
251
252 # This is used to handle the 2-way associativity
253 # (all odd entries are set to one, and if the corresponding even entry
254 # is set to one, then there is a 2-way associativity for this pair)
255 # Entry 0 is for the bimodal and it is ignored
256 # Note: For this implementation, some odd entries are also set to 0 to save
257 # some bits
258 noSkip = [0,0,1,0,0,0,1,0,0,1,1,1,1,1,1,1,1,1,1,
259 1,1,1,1,0,1,0,1,0,1,0,0,0,1,0,0,0,1]
260
261 logTagTableSize = 10
262 shortTagsTageFactor = 10
263 longTagsTageFactor = 20
264
265 longTagsSize = 12
266
267 firstLongTagTable = 13
268
269class TAGE_SC_L_TAGE_8KB(TAGE_SC_L_TAGE):
270 type = 'TAGE_SC_L_TAGE_8KB'
271 cxx_class = 'TAGE_SC_L_TAGE_8KB'
272 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
273
274 nHistoryTables = 30
275
276 minHist = 4
277 maxHist = 1000
278
279 logTagTableSize = 7
280 shortTagsTageFactor = 9
281 longTagsTageFactor = 17
282 longTagsSize = 12
283
284 logTagTableSizes = [12]
285
286 firstLongTagTable = 11
287
288 truncatePathHist = False
289
290 noSkip = [0,0,1,0,1,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,0,1,0,1,0,1,0,1]
291
292 tagTableUBits = 2
293
294# LTAGE branch predictor as described in
295# https://www.irisa.fr/caps/people/seznec/L-TAGE.pdf
296# It is basically a TAGE predictor plus a loop predictor
297# The differnt TAGE sizes are updated according to the paper values (256 Kbits)
298class LTAGE(TAGE):
299 type = 'LTAGE'
300 cxx_class = 'LTAGE'
301 cxx_header = "cpu/pred/ltage.hh"
302
303 tage = LTAGE_TAGE()
304
305 loop_predictor = Param.LoopPredictor(LoopPredictor(), "Loop predictor")
306
307class TAGE_SC_L_LoopPredictor(LoopPredictor):
308 type = 'TAGE_SC_L_LoopPredictor'
309 cxx_class = 'TAGE_SC_L_LoopPredictor'
310 cxx_header = "cpu/pred/tage_sc_l.hh"
311 loopTableAgeBits = 4
312 loopTableConfidenceBits = 4
313 loopTableTagBits = 10
314 loopTableIterBits = 10
315 useSpeculation = False
316 useHashing = True
317 useDirectionBit = True
318 restrictAllocation = True
319 initialLoopIter = 0
320 initialLoopAge = 7
321 optionalAgeReset = False
322
323class StatisticalCorrector(SimObject):
324 type = 'StatisticalCorrector'
325 cxx_class = 'StatisticalCorrector'
326 cxx_header = "cpu/pred/statistical_corrector.hh"
327 abstract = True
328
329 # Statistical corrector parameters
330
331 numEntriesFirstLocalHistories = Param.Unsigned(
332 "Number of entries for first local histories")
333
334 bwnb = Param.Unsigned("Num global backward branch GEHL lengths")
335 bwm = VectorParam.Int("Global backward branch GEHL lengths")
336 logBwnb = Param.Unsigned("Log num of global backward branch GEHL entries")
337
338 lnb = Param.Unsigned("Num first local history GEHL lenghts")
339 lm = VectorParam.Int("First local history GEHL lengths")
340 logLnb = Param.Unsigned("Log number of first local history GEHL entries")
341
342 inb = Param.Unsigned(1, "Num IMLI GEHL lenghts")
343 im = VectorParam.Int([8], "IMLI history GEHL lengths")
344 logInb = Param.Unsigned("Log number of IMLI GEHL entries")
345
346 logBias = Param.Unsigned("Log size of Bias tables")
347
348 logSizeUp = Param.Unsigned(6,
349 "Log size of update threshold counters tables")
350
351 chooserConfWidth = Param.Unsigned(7,
352 "Number of bits for the chooser counters")
353
354 updateThresholdWidth = Param.Unsigned(12,
355 "Number of bits for the update threshold counter")
356
357 pUpdateThresholdWidth = Param.Unsigned(8,
358 "Number of bits for the pUpdate threshold counters")
359
360 extraWeightsWidth = Param.Unsigned(6,
361 "Number of bits for the extra weights")
362
363 scCountersWidth = Param.Unsigned(6, "Statistical corrector counters width")
364
365# TAGE-SC-L branch predictor as desribed in
366# https://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
367# It is a modified LTAGE predictor plus a statistical corrector predictor
368# The TAGE modifications include bank interleaving and partial associativity
369# Two different sizes are proposed in the paper:
370# 8KB => See TAGE_SC_L_8KB below
371# 64KB => See TAGE_SC_L_64KB below
372# The TAGE_SC_L_8KB and TAGE_SC_L_64KB classes differ not only on the values
373# of some parameters, but also in some implementation details
374# Given this, the TAGE_SC_L class is left abstract
375# Note that as it is now, this branch predictor does not handle any type
376# of speculation: All the structures/histories are updated at commit time
377class TAGE_SC_L(LTAGE):
378 type = 'TAGE_SC_L'
379 cxx_class = 'TAGE_SC_L'
380 cxx_header = "cpu/pred/tage_sc_l.hh"
381 abstract = True
382
383 statistical_corrector = Param.StatisticalCorrector(
384 "Statistical Corrector")
385
386class TAGE_SC_L_64KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
387 logSizeLoopPred = 5
388
389class TAGE_SC_L_8KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
390 logSizeLoopPred = 3
391
392class TAGE_SC_L_64KB_StatisticalCorrector(StatisticalCorrector):
393 type = 'TAGE_SC_L_64KB_StatisticalCorrector'
394 cxx_class = 'TAGE_SC_L_64KB_StatisticalCorrector'
395 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
396
397 pnb = Param.Unsigned(3, "Num variation global branch GEHL lengths")
398 pm = VectorParam.Int([25, 16, 9], "Variation global branch GEHL lengths")
399 logPnb = Param.Unsigned(9,
400 "Log number of variation global branch GEHL entries")
401
402 snb = Param.Unsigned(3, "Num second local history GEHL lenghts")
403 sm = VectorParam.Int([16, 11, 6], "Second local history GEHL lengths")
404 logSnb = Param.Unsigned(9,
405 "Log number of second local history GEHL entries")
406
407 tnb = Param.Unsigned(2, "Num third local history GEHL lenghts")
408 tm = VectorParam.Int([9, 4], "Third local history GEHL lengths")
409 logTnb = Param.Unsigned(10,
410 "Log number of third local history GEHL entries")
411
412 imnb = Param.Unsigned(2, "Num second IMLI GEHL lenghts")
413 imm = VectorParam.Int([10, 4], "Second IMLI history GEHL lengths")
414 logImnb = Param.Unsigned(9, "Log number of second IMLI GEHL entries")
415
416 numEntriesSecondLocalHistories = Param.Unsigned(16,
417 "Number of entries for second local histories")
418 numEntriesThirdLocalHistories = Param.Unsigned(16,
419 "Number of entries for second local histories")
420
421 numEntriesFirstLocalHistories = 256
422
423 logBias = 8
424
425 bwnb = 3
426 bwm = [40, 24, 10]
427 logBwnb = 10
428
429 lnb = 3
430 lm = [11, 6, 3]
431 logLnb = 10
432
433 logInb = 8
434
435class TAGE_SC_L_8KB_StatisticalCorrector(StatisticalCorrector):
436 type = 'TAGE_SC_L_8KB_StatisticalCorrector'
437 cxx_class = 'TAGE_SC_L_8KB_StatisticalCorrector'
438 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
439 gnb = Param.Unsigned(2, "Num global branch GEHL lengths")
440 gm = VectorParam.Int([6, 3], "Global branch GEHL lengths")
441 logGnb = Param.Unsigned(7, "Log number of global branch GEHL entries")
442
443 numEntriesFirstLocalHistories = 64
444
445 logBias = 7
446
447 bwnb = 2
448 logBwnb = 7
449 bwm = [16, 8]
450
451 lnb = 2
452 logLnb = 7
453 lm = [6, 3]
454
455 logInb = 7
456
457# 64KB TAGE-SC-L branch predictor as described in
458# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
459class TAGE_SC_L_64KB(TAGE_SC_L):
460 type = 'TAGE_SC_L_64KB'
461 cxx_class = 'TAGE_SC_L_64KB'
462 cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
463
464 tage = TAGE_SC_L_TAGE_64KB()
465 loop_predictor = TAGE_SC_L_64KB_LoopPredictor()
466 statistical_corrector = TAGE_SC_L_64KB_StatisticalCorrector()
467
468# 8KB TAGE-SC-L branch predictor as described in
469# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
470class TAGE_SC_L_8KB(TAGE_SC_L):
471 type = 'TAGE_SC_L_8KB'
472 cxx_class = 'TAGE_SC_L_8KB'
473 cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
474
475 tage = TAGE_SC_L_TAGE_8KB()
476 loop_predictor = TAGE_SC_L_8KB_LoopPredictor()
477 statistical_corrector = TAGE_SC_L_8KB_StatisticalCorrector()
478
479class MultiperspectivePerceptron(BranchPredictor):
480 type = 'MultiperspectivePerceptron'
481 cxx_class = 'MultiperspectivePerceptron'
482 cxx_header = 'cpu/pred/multiperspective_perceptron.hh'
483 abstract = True
484
485 num_filter_entries = Param.Int("Number of filter entries")
486 num_local_histories = Param.Int("Number of local history entries")
487 local_history_length = Param.Int(11,
488 "Length in bits of each history entry")
489
490 block_size = Param.Int(21,
491 "number of ghist bits in a 'block'; this is the width of an initial "
492 "hash of ghist")
493 pcshift = Param.Int(-10, "Shift for hashing PC")
494 threshold = Param.Int(1, "Threshold for deciding low/high confidence")
495 bias0 = Param.Int(-5,
496 "Bias perceptron output this much on all-bits-zero local history")
497 bias1 = Param.Int(5,
498 "Bias perceptron output this much on all-bits-one local history")
499 biasmostly0 = Param.Int(-1,
500 "Bias perceptron output this much on almost-all-bits-zero local "
501 "history")
502 biasmostly1 = Param.Int(1,
503 "Bias perceptron output this much on almost-all-bits-one local "
504 "history")
505 nbest = Param.Int(20,
506 "Use this many of the top performing tables on a low-confidence "
507 "branch")
508 tunebits = Param.Int(24, "Number of bits in misprediction counters")
509 hshift = Param.Int(-6,
510 "How much to shift initial feauture hash before XORing with PC bits")
511 imli_mask1 = Param.UInt64(
512 "Which tables should have their indices hashed with the first IMLI "
513 "counter")
514 imli_mask4 = Param.UInt64(
515 "Which tables should have their indices hashed with the fourth IMLI "
516 "counter")
517 recencypos_mask = Param.UInt64(
518 "Which tables should have their indices hashed with the recency "
519 "position")
520 fudge = Param.Float(0.245, "Fudge factor to multiply by perceptron output")
521 n_sign_bits = Param.Int(2, "Number of sign bits per magnitude")
522 pcbit = Param.Int(2, "Bit from the PC to use for hashing global history")
523 decay = Param.Int(0, "Whether and how often to decay a random weight")
524 record_mask = Param.Int(191,
525 "Which histories are updated with filtered branch outcomes")
526 hash_taken = Param.Bool(False,
527 "Hash the taken/not taken value with a PC bit")
528 tuneonly = Param.Bool(True,
529 "If true, only count mispredictions of low-confidence branches")
530 extra_rounds = Param.Int(1,
531 "Number of extra rounds of training a single weight on a "
532 "low-confidence prediction")
533 speed = Param.Int(9, "Adaptive theta learning speed")
534 initial_theta = Param.Int(10, "Initial theta")
535 budgetbits = Param.Int("Hardware budget in bits")
536 speculative_update = Param.Bool(False,
537 "Use speculative update for histories")
538
539class MultiperspectivePerceptron8KB(MultiperspectivePerceptron):
540 type = 'MultiperspectivePerceptron8KB'
541 cxx_class = 'MultiperspectivePerceptron8KB'
542 cxx_header = 'cpu/pred/multiperspective_perceptron_8KB.hh'
543 budgetbits = 8192 * 8 + 2048
544 num_local_histories = 48
545 num_filter_entries = 0
546 imli_mask1 = 0x6
547 imli_mask4 = 0x4400
548 recencypos_mask = 0x100000090
549
550class MultiperspectivePerceptron64KB(MultiperspectivePerceptron):
551 type = 'MultiperspectivePerceptron64KB'
552 cxx_class = 'MultiperspectivePerceptron64KB'
553 cxx_header = 'cpu/pred/multiperspective_perceptron_64KB.hh'
554 budgetbits = 65536 * 8 + 2048
555 num_local_histories = 510
556 num_filter_entries = 18025
557 imli_mask1 = 0xc1000
558 imli_mask4 = 0x80008000
559 recencypos_mask = 0x100000090