xref: /gem5/src/cpu/pred/BranchPredictor.py

# Copyright (c) 2012 Mark D. Hill and David A. Wood
# Copyright (c) 2015 The University of Wisconsin
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Nilay Vaish and Dibakar Gope

from m5.SimObject import SimObject
from m5.params import *
from m5.proxy import *

class IndirectPredictor(SimObject):
    type = 'IndirectPredictor'
    cxx_class = 'IndirectPredictor'
    cxx_header = "cpu/pred/indirect.hh"
    abstract = True

    numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")

class SimpleIndirectPredictor(IndirectPredictor):
    type = 'SimpleIndirectPredictor'
    cxx_class = 'SimpleIndirectPredictor'
    cxx_header = "cpu/pred/simple_indirect.hh"

    indirectHashGHR = Param.Bool(True, "Hash branch predictor GHR")
    indirectHashTargets = Param.Bool(True, "Hash path history targets")
    indirectSets = Param.Unsigned(256, "Cache sets for indirect predictor")
    indirectWays = Param.Unsigned(2, "Ways for indirect predictor")
    indirectTagSize = Param.Unsigned(16, "Indirect target cache tag bits")
    indirectPathLength = Param.Unsigned(3,
        "Previous indirect targets to use for path history")
    indirectGHRBits = Param.Unsigned(13, "Indirect GHR number of bits")
    instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")

class BranchPredictor(SimObject):
    type = 'BranchPredictor'
    cxx_class = 'BPredUnit'
    cxx_header = "cpu/pred/bpred_unit.hh"
    abstract = True

    numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
    BTBEntries = Param.Unsigned(4096, "Number of BTB entries")
    BTBTagSize = Param.Unsigned(16, "Size of the BTB tags, in bits")
    RASSize = Param.Unsigned(16, "RAS size")
    instShiftAmt = Param.Unsigned(2, "Number of bits to shift instructions by")

    indirectBranchPred = Param.IndirectPredictor(SimpleIndirectPredictor(),
      "Indirect branch predictor, set to NULL to disable indirect predictions")

class LocalBP(BranchPredictor):
    type = 'LocalBP'
    cxx_class = 'LocalBP'
    cxx_header = "cpu/pred/2bit_local.hh"

    localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
    localCtrBits = Param.Unsigned(2, "Bits per counter")


class TournamentBP(BranchPredictor):
    type = 'TournamentBP'
    cxx_class = 'TournamentBP'
    cxx_header = "cpu/pred/tournament.hh"

    localPredictorSize = Param.Unsigned(2048, "Size of local predictor")
    localCtrBits = Param.Unsigned(2, "Bits per counter")
    localHistoryTableSize = Param.Unsigned(2048, "size of local history table")
    globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
    globalCtrBits = Param.Unsigned(2, "Bits per counter")
    choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
    choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")


class BiModeBP(BranchPredictor):
    type = 'BiModeBP'
    cxx_class = 'BiModeBP'
    cxx_header = "cpu/pred/bi_mode.hh"

    globalPredictorSize = Param.Unsigned(8192, "Size of global predictor")
    globalCtrBits = Param.Unsigned(2, "Bits per counter")
    choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
    choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")

class TAGEBase(SimObject):
    type = 'TAGEBase'
    cxx_class = 'TAGEBase'
    cxx_header = "cpu/pred/tage_base.hh"

    numThreads = Param.Unsigned(Parent.numThreads, "Number of threads")
    instShiftAmt = Param.Unsigned(Parent.instShiftAmt,
        "Number of bits to shift instructions by")

    nHistoryTables = Param.Unsigned(7, "Number of history tables")
    minHist = Param.Unsigned(5, "Minimum history size of TAGE")
    maxHist = Param.Unsigned(130, "Maximum history size of TAGE")

    tagTableTagWidths = VectorParam.Unsigned(
        [0, 9, 9, 10, 10, 11, 11, 12], "Tag size in TAGE tag tables")
    logTagTableSizes = VectorParam.Int(
        [13, 9, 9, 9, 9, 9, 9, 9], "Log2 of TAGE table sizes")
    logRatioBiModalHystEntries = Param.Unsigned(2,
        "Log num of prediction entries for a shared hysteresis bit " \
        "for the Bimodal")

    tagTableCounterBits = Param.Unsigned(3, "Number of tag table counter bits")
    tagTableUBits = Param.Unsigned(2, "Number of tag table u bits")

    histBufferSize = Param.Unsigned(2097152,
            "A large number to track all branch histories(2MEntries default)")

    pathHistBits = Param.Unsigned(16, "Path history size")
    logUResetPeriod = Param.Unsigned(18,
        "Log period in number of branches to reset TAGE useful counters")
    numUseAltOnNa = Param.Unsigned(1, "Number of USE_ALT_ON_NA counters")
    useAltOnNaBits = Param.Unsigned(4, "Size of the USE_ALT_ON_NA counter(s)")

    maxNumAlloc = Param.Unsigned(1,
        "Max number of TAGE entries allocted on mispredict")

    # List of enabled TAGE tables. If empty, all are enabled
    noSkip = VectorParam.Bool([], "Vector of enabled TAGE tables")

    speculativeHistUpdate = Param.Bool(True,
        "Use speculative update for histories")

# TAGE branch predictor as described in https://www.jilp.org/vol8/v8paper1.pdf
# The default sizes below are for the 8C-TAGE configuration (63.5 Kbits)
class TAGE(BranchPredictor):
    type = 'TAGE'
    cxx_class = 'TAGE'
    cxx_header = "cpu/pred/tage.hh"
    tage = Param.TAGEBase(TAGEBase(), "Tage object")

class LTAGE_TAGE(TAGEBase):
    nHistoryTables = 12
    minHist = 4
    maxHist = 640
    tagTableTagWidths = [0, 7, 7, 8, 8, 9, 10, 11, 12, 12, 13, 14, 15]
    logTagTableSizes = [14, 10, 10, 11, 11, 11, 11, 10, 10, 10, 10, 9, 9]
    logUResetPeriod = 19

class LoopPredictor(SimObject):
    type = 'LoopPredictor'
    cxx_class = 'LoopPredictor'
    cxx_header = 'cpu/pred/loop_predictor.hh'

    logSizeLoopPred = Param.Unsigned(8, "Log size of the loop predictor")
    withLoopBits = Param.Unsigned(7, "Size of the WITHLOOP counter")
    loopTableAgeBits = Param.Unsigned(8, "Number of age bits per loop entry")
    loopTableConfidenceBits = Param.Unsigned(2,
            "Number of confidence bits per loop entry")
    loopTableTagBits = Param.Unsigned(14, "Number of tag bits per loop entry")
    loopTableIterBits = Param.Unsigned(14, "Nuber of iteration bits per loop")
    logLoopTableAssoc = Param.Unsigned(2, "Log loop predictor associativity")

    # Parameters for enabling modifications to the loop predictor
    # They have been copied from TAGE-GSC-IMLI
    # (http://www.irisa.fr/alf/downloads/seznec/TAGE-GSC-IMLI.tar)
    #
    # All of them should be disabled to match the original LTAGE implementation
    # (http://hpca23.cse.tamu.edu/taco/camino/cbp2/cbp-src/realistic-seznec.h)

    # Add speculation
    useSpeculation = Param.Bool(False, "Use speculation")

    # Add hashing for calculating the loop table index
    useHashing = Param.Bool(False, "Use hashing")

    # Add a direction bit to the loop table entries
    useDirectionBit = Param.Bool(False, "Use direction info")

    # If true, use random to decide whether to allocate or not, and only try
    # with one entry
    restrictAllocation = Param.Bool(False,
        "Restrict the allocation conditions")

    initialLoopIter = Param.Unsigned(1, "Initial iteration number")
    initialLoopAge = Param.Unsigned(255, "Initial age value")
    optionalAgeReset = Param.Bool(True,
        "Reset age bits optionally in some cases")

class TAGE_SC_L_TAGE(TAGEBase):
    type = 'TAGE_SC_L_TAGE'
    cxx_class = 'TAGE_SC_L_TAGE'
    cxx_header = "cpu/pred/tage_sc_l.hh"
    abstract = True
    tagTableTagWidths = [0]
    numUseAltOnNa = 16
    pathHistBits = 27
    maxNumAlloc = 2
    logUResetPeriod = 10
    useAltOnNaBits = 5
    # TODO No speculation implemented as of now
    speculativeHistUpdate = False

    # This size does not set the final sizes of the tables (it is just used
    # for some calculations)
    # Instead, the number of TAGE entries comes from shortTagsTageEntries and
    # longTagsTageEntries
    logTagTableSize = Param.Unsigned("Log size of each tag table")

    shortTagsTageFactor = Param.Unsigned(
        "Factor for calculating the total number of short tags TAGE entries")

    longTagsTageFactor = Param.Unsigned(
        "Factor for calculating the total number of long tags TAGE entries")

    shortTagsSize = Param.Unsigned(8, "Size of the short tags")

    longTagsSize = Param.Unsigned("Size of the long tags")

    firstLongTagTable = Param.Unsigned("First table with long tags")

    truncatePathHist = Param.Bool(True,
        "Truncate the path history to its configured size")


class TAGE_SC_L_TAGE_64KB(TAGE_SC_L_TAGE):
    type = 'TAGE_SC_L_TAGE_64KB'
    cxx_class = 'TAGE_SC_L_TAGE_64KB'
    cxx_header = "cpu/pred/tage_sc_l_64KB.hh"
    nHistoryTables = 36

    minHist = 6
    maxHist = 3000

    tagTableUBits = 1

    logTagTableSizes = [13]

    # This is used to handle the 2-way associativity
    # (all odd entries are set to one, and if the corresponding even entry
    # is set to one, then there is a 2-way associativity for this pair)
    # Entry 0 is for the bimodal and it is ignored
    # Note: For this implementation, some odd entries are also set to 0 to save
    # some bits
    noSkip = [0,0,1,0,0,0,1,0,0,1,1,1,1,1,1,1,1,1,1,
                1,1,1,1,0,1,0,1,0,1,0,0,0,1,0,0,0,1]

    logTagTableSize = 10
    shortTagsTageFactor = 10
    longTagsTageFactor = 20

    longTagsSize = 12

    firstLongTagTable = 13

class TAGE_SC_L_TAGE_8KB(TAGE_SC_L_TAGE):
    type = 'TAGE_SC_L_TAGE_8KB'
    cxx_class = 'TAGE_SC_L_TAGE_8KB'
    cxx_header = "cpu/pred/tage_sc_l_8KB.hh"

    nHistoryTables = 30

    minHist = 4
    maxHist = 1000

    logTagTableSize = 7
    shortTagsTageFactor = 9
    longTagsTageFactor = 17
    longTagsSize = 12

    logTagTableSizes = [12]

    firstLongTagTable = 11

    truncatePathHist = False

    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]

    tagTableUBits = 2

# LTAGE branch predictor as described in
# https://www.irisa.fr/caps/people/seznec/L-TAGE.pdf
# It is basically a TAGE predictor plus a loop predictor
# The differnt TAGE sizes are updated according to the paper values (256 Kbits)
class LTAGE(TAGE):
    type = 'LTAGE'
    cxx_class = 'LTAGE'
    cxx_header = "cpu/pred/ltage.hh"

    tage = LTAGE_TAGE()

    loop_predictor = Param.LoopPredictor(LoopPredictor(), "Loop predictor")

class TAGE_SC_L_LoopPredictor(LoopPredictor):
    type = 'TAGE_SC_L_LoopPredictor'
    cxx_class  = 'TAGE_SC_L_LoopPredictor'
    cxx_header = "cpu/pred/tage_sc_l.hh"
    loopTableAgeBits = 4
    loopTableConfidenceBits = 4
    loopTableTagBits = 10
    loopTableIterBits = 10
    useSpeculation = False
    useHashing = True
    useDirectionBit = True
    restrictAllocation = True
    initialLoopIter = 0
    initialLoopAge = 7
    optionalAgeReset = False

class StatisticalCorrector(SimObject):
    type = 'StatisticalCorrector'
    cxx_class  = 'StatisticalCorrector'
    cxx_header = "cpu/pred/statistical_corrector.hh"
    abstract = True

    # Statistical corrector parameters

    numEntriesFirstLocalHistories = Param.Unsigned(
        "Number of entries for first local histories")

    bwnb = Param.Unsigned("Num global backward branch GEHL lengths")
    bwm = VectorParam.Int("Global backward branch GEHL lengths")
    logBwnb = Param.Unsigned("Log num of global backward branch GEHL entries")

    lnb = Param.Unsigned("Num first local history GEHL lenghts")
    lm = VectorParam.Int("First local history GEHL lengths")
    logLnb = Param.Unsigned("Log number of first local history GEHL entries")

    inb = Param.Unsigned(1, "Num IMLI GEHL lenghts")
    im = VectorParam.Int([8], "IMLI history GEHL lengths")
    logInb = Param.Unsigned("Log number of IMLI GEHL entries")

    logBias = Param.Unsigned("Log size of Bias tables")

    logSizeUp = Param.Unsigned(6,
        "Log size of update threshold counters tables")

    chooserConfWidth = Param.Unsigned(7,
        "Number of bits for the chooser counters")

    updateThresholdWidth = Param.Unsigned(12,
        "Number of bits for the update threshold counter")

    pUpdateThresholdWidth = Param.Unsigned(8,
        "Number of bits for the pUpdate threshold counters")

    extraWeightsWidth = Param.Unsigned(6,
        "Number of bits for the extra weights")

    scCountersWidth = Param.Unsigned(6, "Statistical corrector counters width")

# TAGE-SC-L branch predictor as desribed in
# https://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
# It is a modified LTAGE predictor plus a statistical corrector predictor
# The TAGE modifications include bank interleaving and partial associativity
# Two different sizes are proposed in the paper:
# 8KB => See TAGE_SC_L_8KB below
# 64KB => See TAGE_SC_L_64KB below
# The TAGE_SC_L_8KB and TAGE_SC_L_64KB classes differ not only on the values
# of some parameters, but also in some implementation details
# Given this, the TAGE_SC_L class is left abstract
# Note that as it is now, this branch predictor does not handle any type
# of speculation: All the structures/histories are updated at commit time
class TAGE_SC_L(LTAGE):
    type = 'TAGE_SC_L'
    cxx_class = 'TAGE_SC_L'
    cxx_header = "cpu/pred/tage_sc_l.hh"
    abstract = True

    statistical_corrector = Param.StatisticalCorrector(
        "Statistical Corrector")

class TAGE_SC_L_64KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
    logSizeLoopPred = 5

class TAGE_SC_L_8KB_LoopPredictor(TAGE_SC_L_LoopPredictor):
    logSizeLoopPred = 3

class TAGE_SC_L_64KB_StatisticalCorrector(StatisticalCorrector):
    type = 'TAGE_SC_L_64KB_StatisticalCorrector'
    cxx_class  = 'TAGE_SC_L_64KB_StatisticalCorrector'
    cxx_header = "cpu/pred/tage_sc_l_64KB.hh"

    pnb = Param.Unsigned(3, "Num variation global branch GEHL lengths")
    pm = VectorParam.Int([25, 16, 9], "Variation global branch GEHL lengths")
    logPnb = Param.Unsigned(9,
        "Log number of variation global branch GEHL entries")

    snb = Param.Unsigned(3, "Num second local history GEHL lenghts")
    sm = VectorParam.Int([16, 11, 6], "Second local history GEHL lengths")
    logSnb = Param.Unsigned(9,
        "Log number of second local history GEHL entries")

    tnb = Param.Unsigned(2, "Num third local history GEHL lenghts")
    tm = VectorParam.Int([9, 4], "Third local history GEHL lengths")
    logTnb = Param.Unsigned(10,
        "Log number of third local history GEHL entries")

    imnb = Param.Unsigned(2, "Num second IMLI GEHL lenghts")
    imm = VectorParam.Int([10, 4], "Second IMLI history GEHL lengths")
    logImnb = Param.Unsigned(9, "Log number of second IMLI GEHL entries")

    numEntriesSecondLocalHistories = Param.Unsigned(16,
        "Number of entries for second local histories")
    numEntriesThirdLocalHistories = Param.Unsigned(16,
        "Number of entries for second local histories")

    numEntriesFirstLocalHistories = 256

    logBias = 8

    bwnb = 3
    bwm = [40, 24, 10]
    logBwnb = 10

    lnb = 3
    lm = [11, 6, 3]
    logLnb = 10

    logInb = 8

class TAGE_SC_L_8KB_StatisticalCorrector(StatisticalCorrector):
    type = 'TAGE_SC_L_8KB_StatisticalCorrector'
    cxx_class  = 'TAGE_SC_L_8KB_StatisticalCorrector'
    cxx_header = "cpu/pred/tage_sc_l_8KB.hh"
    gnb = Param.Unsigned(2, "Num global branch GEHL lengths")
    gm = VectorParam.Int([6, 3], "Global branch GEHL lengths")
    logGnb = Param.Unsigned(7, "Log number of global branch GEHL entries")

    numEntriesFirstLocalHistories = 64

    logBias = 7

    bwnb = 2
    logBwnb = 7
    bwm = [16, 8]

    lnb = 2
    logLnb = 7
    lm = [6, 3]

    logInb = 7

# 64KB TAGE-SC-L branch predictor as described in
# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
class TAGE_SC_L_64KB(TAGE_SC_L):
    type = 'TAGE_SC_L_64KB'
    cxx_class = 'TAGE_SC_L_64KB'
    cxx_header = "cpu/pred/tage_sc_l_64KB.hh"

    tage = TAGE_SC_L_TAGE_64KB()
    loop_predictor = TAGE_SC_L_64KB_LoopPredictor()
    statistical_corrector = TAGE_SC_L_64KB_StatisticalCorrector()

# 8KB TAGE-SC-L branch predictor as described in
# http://www.jilp.org/cbp2016/paper/AndreSeznecLimited.pdf
class TAGE_SC_L_8KB(TAGE_SC_L):
    type = 'TAGE_SC_L_8KB'
    cxx_class = 'TAGE_SC_L_8KB'
    cxx_header = "cpu/pred/tage_sc_l_8KB.hh"

    tage = TAGE_SC_L_TAGE_8KB()
    loop_predictor = TAGE_SC_L_8KB_LoopPredictor()
    statistical_corrector = TAGE_SC_L_8KB_StatisticalCorrector()

class MultiperspectivePerceptron(BranchPredictor):
    type = 'MultiperspectivePerceptron'
    cxx_class = 'MultiperspectivePerceptron'
    cxx_header = 'cpu/pred/multiperspective_perceptron.hh'
    abstract = True

    num_filter_entries = Param.Int("Number of filter entries")
    num_local_histories = Param.Int("Number of local history entries")
    local_history_length = Param.Int(11,
        "Length in bits of each history entry")

    block_size = Param.Int(21,
        "number of ghist bits in a 'block'; this is the width of an initial "
        "hash of ghist")
    pcshift = Param.Int(-10, "Shift for hashing PC")
    threshold = Param.Int(1, "Threshold for deciding low/high confidence")
    bias0 = Param.Int(-5,
        "Bias perceptron output this much on all-bits-zero local history")
    bias1 = Param.Int(5,
        "Bias perceptron output this much on all-bits-one local history")
    biasmostly0 = Param.Int(-1,
        "Bias perceptron output this much on almost-all-bits-zero local "
        "history")
    biasmostly1 = Param.Int(1,
        "Bias perceptron output this much on almost-all-bits-one local "
        "history")
    nbest = Param.Int(20,
        "Use this many of the top performing tables on a low-confidence "
        "branch")
    tunebits = Param.Int(24, "Number of bits in misprediction counters")
    hshift = Param.Int(-6,
        "How much to shift initial feauture hash before XORing with PC bits")
    imli_mask1 = Param.UInt64(
        "Which tables should have their indices hashed with the first IMLI "
        "counter")
    imli_mask4 = Param.UInt64(
        "Which tables should have their indices hashed with the fourth IMLI "
        "counter")
    recencypos_mask = Param.UInt64(
        "Which tables should have their indices hashed with the recency "
        "position")
    fudge = Param.Float(0.245, "Fudge factor to multiply by perceptron output")
    n_sign_bits = Param.Int(2, "Number of sign bits per magnitude")
    pcbit = Param.Int(2, "Bit from the PC to use for hashing global history")
    decay = Param.Int(0, "Whether and how often to decay a random weight")
    record_mask = Param.Int(191,
        "Which histories are updated with filtered branch outcomes")
    hash_taken = Param.Bool(False,
        "Hash the taken/not taken value with a PC bit")
    tuneonly = Param.Bool(True,
        "If true, only count mispredictions of low-confidence branches")
    extra_rounds = Param.Int(1,
        "Number of extra rounds of training a single weight on a "
        "low-confidence prediction")
    speed = Param.Int(9, "Adaptive theta learning speed")
    initial_theta = Param.Int(10, "Initial theta")
    budgetbits = Param.Int("Hardware budget in bits")
    speculative_update = Param.Bool(False,
        "Use speculative update for histories")

class MultiperspectivePerceptron8KB(MultiperspectivePerceptron):
    type = 'MultiperspectivePerceptron8KB'
    cxx_class = 'MultiperspectivePerceptron8KB'
    cxx_header = 'cpu/pred/multiperspective_perceptron_8KB.hh'
    budgetbits = 8192 * 8 + 2048
    num_local_histories = 48
    num_filter_entries = 0
    imli_mask1 = 0x6
    imli_mask4 = 0x4400
    recencypos_mask = 0x100000090

class MultiperspectivePerceptron64KB(MultiperspectivePerceptron):
    type = 'MultiperspectivePerceptron64KB'
    cxx_class = 'MultiperspectivePerceptron64KB'
    cxx_header = 'cpu/pred/multiperspective_perceptron_64KB.hh'
    budgetbits = 65536 * 8 + 2048
    num_local_histories = 510
    num_filter_entries = 18025
    imli_mask1 = 0xc1000
    imli_mask4 = 0x80008000
    recencypos_mask = 0x100000090