derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

cpu: Made TAGE a SimObject that can be used by other predictors

Browse Source 
The TAGE implementation is now a SimObject so that other branch predictors
can easily use it. It has also been updated with the latest available TAGE
implementation from Andre Seznec:

http://www.irisa.fr/alf/downloads/seznec/TAGE-GSC-IMLI.tar

Change-Id: I2251b8b2d7f94124f9955f52b917dc3b064f090e
Reviewed-on: https://gem5-review.googlesource.com/c/15317
Reviewed-by: Giacomo Travaglini <giacomo.travaglini@arm.com>
Maintainer: Andreas Sandberg <andreas.sandberg@arm.com>
This commit is contained in:
Jairo Balart
2019-01-05 10:24:17 +01:00
committed by Pau Cabre
parent f0e2caf84f
commit 4ba89236f0
16 changed files with 1504 additions and 1062 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/cpu/pred/2bit_local.cc
View File

@@ -119,7 +119,7 @@ LocalBP::lookup(ThreadID tid, Addr branch_addr, void * &bp_history)
void
LocalBP::update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed)
bool squashed, const StaticInstPtr & inst, Addr corrTarget)
{
assert(bp_history == NULL);
unsigned local_predictor_idx;

2
src/cpu/pred/2bit_local.hh
View File

@@ -92,7 +92,7 @@ class LocalBP : public BPredUnit
* @param taken Whether or not the branch was taken.
*/
void update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed);
bool squashed, const StaticInstPtr & inst, Addr corrTarget);
void squash(ThreadID tid, void *bp_history)
{ assert(bp_history == NULL); }

46
src/cpu/pred/BranchPredictor.py
View File

@@ -87,12 +87,14 @@ class BiModeBP(BranchPredictor):
choicePredictorSize = Param.Unsigned(8192, "Size of choice predictor")
choiceCtrBits = Param.Unsigned(2, "Bits of choice counters")
# 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"
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")
@@ -115,8 +117,33 @@ class TAGE(BranchPredictor):
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")
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
# LTAGE branch predictor as described in
# https://www.irisa.fr/caps/people/seznec/L-TAGE.pdf
@@ -127,12 +154,7 @@ class LTAGE(TAGE):
cxx_class = 'LTAGE'
cxx_header = "cpu/pred/ltage.hh"
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
tage = LTAGE_TAGE()
logSizeLoopPred = Param.Unsigned(8, "Log size of the loop predictor")
withLoopBits = Param.Unsigned(7, "Size of the WITHLOOP counter")

1
src/cpu/pred/SConscript
View File

@@ -43,6 +43,7 @@ Source('indirect.cc')
Source('ras.cc')
Source('tournament.cc')
Source ('bi_mode.cc')
Source('tage_base.cc')
Source('tage.cc')
Source('ltage.cc')
DebugFlag('FreeList')

2
src/cpu/pred/bi_mode.cc
View File

@@ -161,7 +161,7 @@ BiModeBP::btbUpdate(ThreadID tid, Addr branchAddr, void * &bpHistory)
*/
void
BiModeBP::update(ThreadID tid, Addr branchAddr, bool taken, void *bpHistory,
bool squashed)
bool squashed, const StaticInstPtr & inst, Addr corrTarget)
{
assert(bpHistory);

2
src/cpu/pred/bi_mode.hh
View File

@@ -62,7 +62,7 @@ class BiModeBP : public BPredUnit
bool lookup(ThreadID tid, Addr branch_addr, void * &bp_history);
void btbUpdate(ThreadID tid, Addr branch_addr, void * &bp_history);
void update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed);
bool squashed, const StaticInstPtr & inst, Addr corrTarget);
unsigned getGHR(ThreadID tid, void *bp_history) const;
private:

11
src/cpu/pred/bpred_unit.cc
View File

@@ -210,7 +210,7 @@ BPredUnit::predict(const StaticInstPtr &inst, const InstSeqNum &seqNum,
"for PC %s\n", tid, seqNum, pc);
PredictorHistory predict_record(seqNum, pc.instAddr(),
pred_taken, bp_history, tid);
pred_taken, bp_history, tid, inst);
// Now lookup in the BTB or RAS.
if (pred_taken) {
@@ -309,6 +309,7 @@ BPredUnit::predict(const StaticInstPtr &inst, const InstSeqNum &seqNum,
}
TheISA::advancePC(target, inst);
}
predict_record.target = target.instAddr();
pc = target;
@@ -332,7 +333,9 @@ BPredUnit::update(const InstSeqNum &done_sn, ThreadID tid)
// Update the branch predictor with the correct results.
update(tid, predHist[tid].back().pc,
predHist[tid].back().predTaken,
predHist[tid].back().bpHistory, false);
predHist[tid].back().bpHistory, false,
predHist[tid].back().inst,
predHist[tid].back().target);
predHist[tid].pop_back();
}
@@ -440,9 +443,11 @@ BPredUnit::squash(const InstSeqNum &squashed_sn,
// Remember the correct direction for the update at commit.
pred_hist.front().predTaken = actually_taken;
pred_hist.front().target = corrTarget.instAddr();
update(tid, (*hist_it).pc, actually_taken,
pred_hist.front().bpHistory, true);
pred_hist.front().bpHistory, true, pred_hist.front().inst,
corrTarget.instAddr());
if (actually_taken) {
if (hist_it->wasReturn && !hist_it->usedRAS) {

20
src/cpu/pred/bpred_unit.hh
View File

@@ -175,10 +175,15 @@ class BPredUnit : public SimObject
* associated with the branch lookup that is being updated.
* @param squashed Set to true when this function is called during a
* squash operation.
* @param inst Static instruction information
* @param corrTarget The resolved target of the branch (only needed
* for squashed branches)
* @todo Make this update flexible enough to handle a global predictor.
*/
virtual void update(ThreadID tid, Addr instPC, bool taken,
void *bp_history, bool squashed) = 0;
void *bp_history, bool squashed,
const StaticInstPtr & inst = StaticInst::nullStaticInstPtr,
Addr corrTarget = MaxAddr) = 0;
/**
* Updates the BTB with the target of a branch.
* @param inst_PC The branch's PC that will be updated.
@@ -200,10 +205,11 @@ class BPredUnit : public SimObject
*/
PredictorHistory(const InstSeqNum &seq_num, Addr instPC,
bool pred_taken, void *bp_history,
ThreadID _tid)
ThreadID _tid, const StaticInstPtr & inst)
: seqNum(seq_num), pc(instPC), bpHistory(bp_history), RASTarget(0),
RASIndex(0), tid(_tid), predTaken(pred_taken), usedRAS(0), pushedRAS(0),
wasCall(0), wasReturn(0), wasIndirect(0)
wasCall(0), wasReturn(0), wasIndirect(0),
target(MaxAddr), inst(inst)
{}
bool operator==(const PredictorHistory &entry) const {
@@ -248,6 +254,14 @@ class BPredUnit : public SimObject
/** Wether this instruction was an indirect branch */
bool wasIndirect;
/** Target of the branch. First it is predicted, and fixed later
* if necessary
*/
Addr target;
/** The branch instrction */
const StaticInstPtr inst;
};
typedef std::deque<PredictorHistory> History;

125
src/cpu/pred/ltage.cc
View File

@@ -168,9 +168,10 @@ LTAGE::loopUpdate(Addr pc, bool taken, LTageBranchInfo* bi)
ltable[idx].confidence = 0;
ltable[idx].currentIter = 0;
return;
} else if (bi->loopPred != bi->tagePred) {
} else if (bi->loopPred != bi->tageBranchInfo->tagePred) {
DPRINTF(LTage, "Loop Prediction success:%lx\n",pc);
unsignedCtrUpdate(ltable[idx].age, true, loopTableAgeBits);
TAGEBase::unsignedCtrUpdate(ltable[idx].age, true,
loopTableAgeBits);
}
}
@@ -190,7 +191,7 @@ LTAGE::loopUpdate(Addr pc, bool taken, LTageBranchInfo* bi)
if (ltable[idx].currentIter == ltable[idx].numIter) {
DPRINTF(LTage, "Loop End predicted successfully:%lx\n", pc);
unsignedCtrUpdate(ltable[idx].confidence, true,
TAGEBase::unsignedCtrUpdate(ltable[idx].confidence, true,
loopTableConfidenceBits);
//just do not predict when the loop count is 1 or 2
if (ltable[idx].numIter < 3) {
@@ -218,10 +219,11 @@ LTAGE::loopUpdate(Addr pc, bool taken, LTageBranchInfo* bi)
}
} else if (useDirectionBit ?
((bi->loopPredValid ? bi->loopPred : bi->tagePred) != taken) :
((bi->loopPredValid ?
bi->loopPred : bi->tageBranchInfo->tagePred) != taken) :
taken) {
//try to allocate an entry on taken branch
int nrand = random_mt.random<int>();
int nrand = TAGEBase::getRandom();
for (int i = 0; i < (1 << logLoopTableAssoc); i++) {
int loop_hit = (nrand + i) & ((1 << logLoopTableAssoc) - 1);
idx = bi->loopIndex + loop_hit;
@@ -248,10 +250,11 @@ LTAGE::loopUpdate(Addr pc, bool taken, LTageBranchInfo* bi)
bool
LTAGE::predict(ThreadID tid, Addr branch_pc, bool cond_branch, void* &b)
{
LTageBranchInfo *bi = new LTageBranchInfo(nHistoryTables+1);
LTageBranchInfo *bi = new LTageBranchInfo(*tage);
b = (void*)(bi);
bool pred_taken = tagePredict(tid, branch_pc, cond_branch, bi);
bool pred_taken = tage->tagePredict(tid, branch_pc, cond_branch,
bi->tageBranchInfo);
if (cond_branch) {
// loop prediction
@@ -259,12 +262,13 @@ LTAGE::predict(ThreadID tid, Addr branch_pc, bool cond_branch, void* &b)
if ((loopUseCounter >= 0) && bi->loopPredValid) {
pred_taken = bi->loopPred;
bi->provider = LOOP;
bi->tageBranchInfo->provider = LOOP;
}
DPRINTF(LTage, "Predict for %lx: taken?:%d, loopTaken?:%d, "
"loopValid?:%d, loopUseCounter:%d, tagePred:%d, altPred:%d\n",
branch_pc, pred_taken, bi->loopPred, bi->loopPredValid,
loopUseCounter, bi->tagePred, bi->altTaken);
loopUseCounter, bi->tageBranchInfo->tagePred,
bi->tageBranchInfo->altTaken);
if (useSpeculation) {
specLoopUpdate(pred_taken, bi);
@@ -275,40 +279,68 @@ LTAGE::predict(ThreadID tid, Addr branch_pc, bool cond_branch, void* &b)
}
void
LTAGE::condBranchUpdate(Addr branch_pc, bool taken,
TageBranchInfo* tage_bi, int nrand)
LTAGE::update(ThreadID tid, Addr branch_pc, bool taken, void* bp_history,
bool squashed, const StaticInstPtr & inst, Addr corrTarget)
{
LTageBranchInfo* bi = static_cast<LTageBranchInfo*>(tage_bi);
assert(bp_history);
if (useSpeculation) {
// recalculate loop prediction without speculation
// It is ok to overwrite the loop prediction fields in bi
// as the stats have already been updated with the previous
// values
bi->loopPred = getLoop(branch_pc, bi, false);
}
LTageBranchInfo* bi = static_cast<LTageBranchInfo*>(bp_history);
if (bi->loopPredValid) {
if (bi->tagePred != bi->loopPred) {
ctrUpdate(loopUseCounter,
(bi->loopPred == taken),
withLoopBits);
if (squashed) {
if (tage->isSpeculativeUpdateEnabled()) {
// This restores the global history, then update it
// and recomputes the folded histories.
tage->squash(tid, taken, bi->tageBranchInfo, corrTarget);
squashLoop(bi);
}
return;
}
loopUpdate(branch_pc, taken, bi);
int nrand = TAGEBase::getRandom() & 3;
if (bi->tageBranchInfo->condBranch) {
DPRINTF(LTage, "Updating tables for branch:%lx; taken?:%d\n",
branch_pc, taken);
tage->updateStats(taken, bi->tageBranchInfo);
// update stats
if (bi->tageBranchInfo->provider == LOOP) {
if (taken == bi->loopPred) {
loopPredictorCorrect++;
} else {
loopPredictorWrong++;
}
}
// cond Branch Update
if (useSpeculation) {
// recalculate loop prediction without speculation
// It is ok to overwrite the loop prediction fields in bi
// as the stats have already been updated with the previous
// values
bi->loopPred = getLoop(branch_pc, bi, false);
}
if (bi->loopPredValid) {
if (bi->tageBranchInfo->tagePred != bi->loopPred) {
TAGEBase::ctrUpdate(loopUseCounter,
(bi->loopPred == taken),
withLoopBits);
}
}
TAGE::condBranchUpdate(branch_pc, taken, bi, nrand);
loopUpdate(branch_pc, taken, bi);
tage->condBranchUpdate(tid, branch_pc, taken, bi->tageBranchInfo,
nrand, corrTarget);
}
tage->updateHistories(tid, branch_pc, taken, bi->tageBranchInfo, false,
inst, corrTarget);
delete bi;
}
void
LTAGE::squash(ThreadID tid, bool taken, void *bp_history)
LTAGE::squashLoop(LTageBranchInfo* bi)
{
TAGE::squash(tid, taken, bp_history);
LTageBranchInfo* bi = (LTageBranchInfo*)(bp_history);
if (bi->condBranch) {
if (bi->tageBranchInfo->condBranch) {
if (bi->loopHit >= 0) {
int idx = finallindex(bi->loopIndex,
bi->loopLowPcBits,
@@ -322,37 +354,14 @@ void
LTAGE::squash(ThreadID tid, void *bp_history)
{
LTageBranchInfo* bi = (LTageBranchInfo*)(bp_history);
if (bi->condBranch) {
if (bi->loopHit >= 0) {
int idx = finallindex(bi->loopIndex,
bi->loopLowPcBits,
bi->loopHit);
ltable[idx].currentIterSpec = bi->currentIter;
}
if (bi->tageBranchInfo->condBranch) {
squashLoop(bi);
}
TAGE::squash(tid, bp_history);
}
void
LTAGE::updateStats(bool taken, TageBranchInfo* bi)
{
TAGE::updateStats(taken, bi);
LTageBranchInfo * ltage_bi = static_cast<LTageBranchInfo *>(bi);
if (ltage_bi->provider == LOOP) {
if (taken == ltage_bi->loopPred) {
loopPredictorCorrect++;
} else {
loopPredictorWrong++;
}
}
}
void
LTAGE::regStats()
{

42
src/cpu/pred/ltage.hh
View File

@@ -66,6 +66,9 @@ class LTAGE: public TAGE
// Base class methods.
void squash(ThreadID tid, void *bp_history) override;
void update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed, const StaticInstPtr & inst,
Addr corrTarget) override;
void regStats() override;
@@ -88,7 +91,7 @@ class LTAGE: public TAGE
// more provider types
enum {
LOOP = LAST_TAGE_PROVIDER_TYPE + 1
LOOP = TAGEBase::LAST_TAGE_PROVIDER_TYPE + 1
};
// Primary branch history entry
@@ -103,12 +106,15 @@ class LTAGE: public TAGE
int loopLowPcBits; // only for useHashing
int loopHit;
LTageBranchInfo(int sz)
: TageBranchInfo(sz),
LTageBranchInfo(TAGEBase &tage)
: TageBranchInfo(tage),
loopTag(0), currentIter(0),
loopPred(false),
loopPredValid(false), loopIndex(0), loopLowPcBits(0), loopHit(0)
{}
virtual ~LTageBranchInfo()
{}
};
/**
@@ -156,17 +162,6 @@ class LTAGE: public TAGE
*/
void specLoopUpdate(bool taken, LTageBranchInfo* bi);
/**
* Update LTAGE for conditional branches.
* @param branch_pc The unshifted branch PC.
* @param taken Actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
* @nrand Random int number from 0 to 3
*/
void condBranchUpdate(
Addr branch_pc, bool taken, TageBranchInfo* bi, int nrand) override;
/**
* Get a branch prediction from LTAGE. *NOT* an override of
* BpredUnit::predict().
@@ -184,24 +179,9 @@ class LTAGE: public TAGE
* Restores speculatively updated path and direction histories.
* Also recomputes compressed (folded) histories based on the
* correct branch outcome.
* This version of squash() is called once on a branch misprediction.
* @param tid The Thread ID to select the histories to rollback.
* @param taken The correct branch outcome.
* @param bp_history Wrapping pointer to TageBranchInfo (to allow
* storing derived class prediction information in the
* base class).
* @post bp_history points to valid memory.
* @param bi Branch information.
*/
void squash(
ThreadID tid, bool taken, void *bp_history) override;
/**
* Update the stats
* @param taken Actual branch outcome
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
void updateStats(bool taken, TageBranchInfo* bi) override;
void squashLoop(LTageBranchInfo * bi);
const unsigned logSizeLoopPred;
const unsigned loopTableAgeBits;

643
src/cpu/pred/tage.cc
View File

@@ -47,222 +47,8 @@
#include "debug/Fetch.hh"
#include "debug/Tage.hh"
TAGE::TAGE(const TAGEParams *params)
: BPredUnit(params),
logRatioBiModalHystEntries(params->logRatioBiModalHystEntries),
nHistoryTables(params->nHistoryTables),
tagTableCounterBits(params->tagTableCounterBits),
tagTableUBits(params->tagTableUBits),
histBufferSize(params->histBufferSize),
minHist(params->minHist),
maxHist(params->maxHist),
pathHistBits(params->pathHistBits),
tagTableTagWidths(params->tagTableTagWidths),
logTagTableSizes(params->logTagTableSizes),
threadHistory(params->numThreads),
logUResetPeriod(params->logUResetPeriod),
useAltOnNaBits(params->useAltOnNaBits)
TAGE::TAGE(const TAGEParams *params) : BPredUnit(params), tage(params->tage)
{
// Current method for periodically resetting the u counter bits only
// works for 1 or 2 bits
// Also make sure that it is not 0
assert(tagTableUBits <= 2 && (tagTableUBits > 0));
// we use int type for the path history, so it cannot be more than
// its size
assert(pathHistBits <= (sizeof(int)*8));
// initialize the counter to half of the period
assert(logUResetPeriod != 0);
tCounter = ULL(1) << (logUResetPeriod - 1);
assert(params->histBufferSize > params->maxHist * 2);
useAltPredForNewlyAllocated = 0;
for (auto& history : threadHistory) {
history.pathHist = 0;
history.globalHistory = new uint8_t[histBufferSize];
history.gHist = history.globalHistory;
memset(history.gHist, 0, histBufferSize);
history.ptGhist = 0;
}
histLengths = new int [nHistoryTables+1];
histLengths[1] = minHist;
histLengths[nHistoryTables] = maxHist;
for (int i = 2; i <= nHistoryTables; i++) {
histLengths[i] = (int) (((double) minHist *
pow ((double) (maxHist) / (double) minHist,
(double) (i - 1) / (double) ((nHistoryTables- 1))))
+ 0.5);
}
assert(tagTableTagWidths.size() == (nHistoryTables+1));
assert(logTagTableSizes.size() == (nHistoryTables+1));
// First entry is for the Bimodal table and it is untagged in this
// implementation
assert(tagTableTagWidths[0] == 0);
for (auto& history : threadHistory) {
history.computeIndices = new FoldedHistory[nHistoryTables+1];
history.computeTags[0] = new FoldedHistory[nHistoryTables+1];
history.computeTags[1] = new FoldedHistory[nHistoryTables+1];
for (int i = 1; i <= nHistoryTables; i++) {
history.computeIndices[i].init(
histLengths[i], (logTagTableSizes[i]));
history.computeTags[0][i].init(
history.computeIndices[i].origLength, tagTableTagWidths[i]);
history.computeTags[1][i].init(
history.computeIndices[i].origLength, tagTableTagWidths[i]-1);
DPRINTF(Tage, "HistLength:%d, TTSize:%d, TTTWidth:%d\n",
histLengths[i], logTagTableSizes[i], tagTableTagWidths[i]);
}
}
const uint64_t bimodalTableSize = ULL(1) << logTagTableSizes[0];
btablePrediction.resize(bimodalTableSize, false);
btableHysteresis.resize(bimodalTableSize >> logRatioBiModalHystEntries,
true);
gtable = new TageEntry*[nHistoryTables + 1];
for (int i = 1; i <= nHistoryTables; i++) {
gtable[i] = new TageEntry[1<<(logTagTableSizes[i])];
}
tableIndices = new int [nHistoryTables+1];
tableTags = new int [nHistoryTables+1];
}
int
TAGE::bindex(Addr pc_in) const
{
return ((pc_in >> instShiftAmt) & ((ULL(1) << (logTagTableSizes[0])) - 1));
}
int
TAGE::F(int A, int size, int bank) const
{
int A1, A2;
A = A & ((ULL(1) << size) - 1);
A1 = (A & ((ULL(1) << logTagTableSizes[bank]) - 1));
A2 = (A >> logTagTableSizes[bank]);
A2 = ((A2 << bank) & ((ULL(1) << logTagTableSizes[bank]) - 1))
+ (A2 >> (logTagTableSizes[bank] - bank));
A = A1 ^ A2;
A = ((A << bank) & ((ULL(1) << logTagTableSizes[bank]) - 1))
+ (A >> (logTagTableSizes[bank] - bank));
return (A);
}
// gindex computes a full hash of pc, ghist and pathHist
int
TAGE::gindex(ThreadID tid, Addr pc, int bank) const
{
int index;
int hlen = (histLengths[bank] > pathHistBits) ? pathHistBits :
histLengths[bank];
const Addr shiftedPc = pc >> instShiftAmt;
index =
shiftedPc ^
(shiftedPc >> ((int) abs(logTagTableSizes[bank] - bank) + 1)) ^
threadHistory[tid].computeIndices[bank].comp ^
F(threadHistory[tid].pathHist, hlen, bank);
return (index & ((ULL(1) << (logTagTableSizes[bank])) - 1));
}
// Tag computation
uint16_t
TAGE::gtag(ThreadID tid, Addr pc, int bank) const
{
int tag = (pc >> instShiftAmt) ^
threadHistory[tid].computeTags[0][bank].comp ^
(threadHistory[tid].computeTags[1][bank].comp << 1);
return (tag & ((ULL(1) << tagTableTagWidths[bank]) - 1));
}
// Up-down saturating counter
void
TAGE::ctrUpdate(int8_t & ctr, bool taken, int nbits)
{
assert(nbits <= sizeof(int8_t) << 3);
if (taken) {
if (ctr < ((1 << (nbits - 1)) - 1))
ctr++;
} else {
if (ctr > -(1 << (nbits - 1)))
ctr--;
}
}
// Up-down unsigned saturating counter
void
TAGE::unsignedCtrUpdate(uint8_t & ctr, bool up, unsigned nbits)
{
assert(nbits <= sizeof(uint8_t) << 3);
if (up) {
if (ctr < ((1 << nbits) - 1))
ctr++;
} else {
if (ctr)
ctr--;
}
}
// Bimodal prediction
bool
TAGE::getBimodePred(Addr pc, TageBranchInfo* bi) const
{
return btablePrediction[bi->bimodalIndex];
}
// Update the bimodal predictor: a hysteresis bit is shared among N prediction
// bits (N = 2 ^ logRatioBiModalHystEntries)
void
TAGE::baseUpdate(Addr pc, bool taken, TageBranchInfo* bi)
{
int inter = (btablePrediction[bi->bimodalIndex] << 1)
+ btableHysteresis[bi->bimodalIndex >> logRatioBiModalHystEntries];
if (taken) {
if (inter < 3)
inter++;
} else if (inter > 0) {
inter--;
}
const bool pred = inter >> 1;
const bool hyst = inter & 1;
btablePrediction[bi->bimodalIndex] = pred;
btableHysteresis[bi->bimodalIndex >> logRatioBiModalHystEntries] = hyst;
DPRINTF(Tage, "Updating branch %lx, pred:%d, hyst:%d\n", pc, pred, hyst);
}
// shifting the global history: we manage the history in a big table in order
// to reduce simulation time
void
TAGE::updateGHist(uint8_t * &h, bool dir, uint8_t * tab, int &pt)
{
if (pt == 0) {
DPRINTF(Tage, "Rolling over the histories\n");
// Copy beginning of globalHistoryBuffer to end, such that
// the last maxHist outcomes are still reachable
// through pt[0 .. maxHist - 1].
for (int i = 0; i < maxHist; i++)
tab[histBufferSize - maxHist + i] = tab[i];
pt = histBufferSize - maxHist;
h = &tab[pt];
}
pt--;
h--;
h[0] = (dir) ? 1 : 0;
}
// Get GHR for hashing indirect predictor
@@ -272,308 +58,68 @@ unsigned
TAGE::getGHR(ThreadID tid, void *bp_history) const
{
TageBranchInfo* bi = static_cast<TageBranchInfo*>(bp_history);
unsigned val = 0;
for (unsigned i = 0; i < 32; i++) {
// Make sure we don't go out of bounds
int gh_offset = bi->ptGhist + i;
assert(&(threadHistory[tid].globalHistory[gh_offset]) <
threadHistory[tid].globalHistory + histBufferSize);
val |= ((threadHistory[tid].globalHistory[gh_offset] & 0x1) << i);
}
return val;
}
//prediction
bool
TAGE::predict(ThreadID tid, Addr branch_pc, bool cond_branch, void* &b)
{
TageBranchInfo *bi = new TageBranchInfo(nHistoryTables+1);
b = (void*)(bi);
return tagePredict(tid, branch_pc, cond_branch, bi);
}
bool
TAGE::tagePredict(ThreadID tid, Addr branch_pc,
bool cond_branch, TageBranchInfo* bi)
{
Addr pc = branch_pc;
bool pred_taken = true;
if (cond_branch) {
// TAGE prediction
// computes the table addresses and the partial tags
for (int i = 1; i <= nHistoryTables; i++) {
tableIndices[i] = gindex(tid, pc, i);
bi->tableIndices[i] = tableIndices[i];
tableTags[i] = gtag(tid, pc, i);
bi->tableTags[i] = tableTags[i];
}
bi->bimodalIndex = bindex(pc);
bi->hitBank = 0;
bi->altBank = 0;
//Look for the bank with longest matching history
for (int i = nHistoryTables; i > 0; i--) {
if (gtable[i][tableIndices[i]].tag == tableTags[i]) {
bi->hitBank = i;
bi->hitBankIndex = tableIndices[bi->hitBank];
break;
}
}
//Look for the alternate bank
for (int i = bi->hitBank - 1; i > 0; i--) {
if (gtable[i][tableIndices[i]].tag == tableTags[i]) {
bi->altBank = i;
bi->altBankIndex = tableIndices[bi->altBank];
break;
}
}
//computes the prediction and the alternate prediction
if (bi->hitBank > 0) {
if (bi->altBank > 0) {
bi->altTaken =
gtable[bi->altBank][tableIndices[bi->altBank]].ctr >= 0;
}else {
bi->altTaken = getBimodePred(pc, bi);
}
bi->longestMatchPred =
gtable[bi->hitBank][tableIndices[bi->hitBank]].ctr >= 0;
bi->pseudoNewAlloc =
abs(2 * gtable[bi->hitBank][bi->hitBankIndex].ctr + 1) <= 1;
//if the entry is recognized as a newly allocated entry and
//useAltPredForNewlyAllocated is positive use the alternate
//prediction
if ((useAltPredForNewlyAllocated < 0) || ! bi->pseudoNewAlloc) {
bi->tagePred = bi->longestMatchPred;
bi->provider = TAGE_LONGEST_MATCH;
} else {
bi->tagePred = bi->altTaken;
bi->provider = bi->altBank ? TAGE_ALT_MATCH
: BIMODAL_ALT_MATCH;
}
} else {
bi->altTaken = getBimodePred(pc, bi);
bi->tagePred = bi->altTaken;
bi->longestMatchPred = bi->altTaken;
bi->provider = BIMODAL_ONLY;
}
//end TAGE prediction
pred_taken = (bi->tagePred);
DPRINTF(Tage, "Predict for %lx: taken?:%d, tagePred:%d, altPred:%d\n",
branch_pc, pred_taken, bi->tagePred, bi->altTaken);
}
bi->branchPC = branch_pc;
bi->condBranch = cond_branch;
return pred_taken;
return tage->getGHR(tid, bi->tageBranchInfo);
}
// PREDICTOR UPDATE
void
TAGE::update(ThreadID tid, Addr branch_pc, bool taken, void* bp_history,
bool squashed)
bool squashed, const StaticInstPtr & inst, Addr corrTarget)
{
assert(bp_history);
TageBranchInfo *bi = static_cast<TageBranchInfo*>(bp_history);
assert(corrTarget != MaxAddr);
if (squashed) {
// This restores the global history, then update it
// and recomputes the folded histories.
squash(tid, taken, bp_history);
tage->squash(tid, taken, bi->tageBranchInfo, corrTarget);
return;
}
int nrand = random_mt.random<int>(0,3);
if (bi->condBranch) {
int nrand = TAGEBase::getRandom() & 3;
if (bi->tageBranchInfo->condBranch) {
DPRINTF(Tage, "Updating tables for branch:%lx; taken?:%d\n",
branch_pc, taken);
updateStats(taken, bi);
condBranchUpdate(branch_pc, taken, bi, nrand);
}
if (!squashed) {
delete bi;
}
}
void
TAGE::condBranchUpdate(Addr branch_pc, bool taken,
TageBranchInfo* bi, int nrand)
{
// TAGE UPDATE
// try to allocate a new entries only if prediction was wrong
bool longest_match_pred = false;
bool alloc = (bi->tagePred != taken) && (bi->hitBank < nHistoryTables);
if (bi->hitBank > 0) {
// Manage the selection between longest matching and alternate
// matching for "pseudo"-newly allocated longest matching entry
longest_match_pred = bi->longestMatchPred;
bool PseudoNewAlloc = bi->pseudoNewAlloc;
// an entry is considered as newly allocated if its prediction
// counter is weak
if (PseudoNewAlloc) {
if (longest_match_pred == taken) {
alloc = false;
}
// if it was delivering the correct prediction, no need to
// allocate new entry even if the overall prediction was false
if (longest_match_pred != bi->altTaken) {
ctrUpdate(useAltPredForNewlyAllocated,
bi->altTaken == taken, useAltOnNaBits);
}
}
tage->updateStats(taken, bi->tageBranchInfo);
tage->condBranchUpdate(tid, branch_pc, taken, bi->tageBranchInfo,
nrand, corrTarget);
}
if (alloc) {
// is there some "unuseful" entry to allocate
uint8_t min = 1;
for (int i = nHistoryTables; i > bi->hitBank; i--) {
if (gtable[i][bi->tableIndices[i]].u < min) {
min = gtable[i][bi->tableIndices[i]].u;
}
}
tage->updateHistories(tid, branch_pc, taken, bi->tageBranchInfo, false,
inst, corrTarget);
// we allocate an entry with a longer history
// to avoid ping-pong, we do not choose systematically the next
// entry, but among the 3 next entries
int Y = nrand &
((ULL(1) << (nHistoryTables - bi->hitBank - 1)) - 1);
int X = bi->hitBank + 1;
if (Y & 1) {
X++;
if (Y & 2)
X++;
}
// No entry available, forces one to be available
if (min > 0) {
gtable[X][bi->tableIndices[X]].u = 0;
}
//Allocate only one entry
for (int i = X; i <= nHistoryTables; i++) {
if ((gtable[i][bi->tableIndices[i]].u == 0)) {
gtable[i][bi->tableIndices[i]].tag = bi->tableTags[i];
gtable[i][bi->tableIndices[i]].ctr = (taken) ? 0 : -1;
break;
}
}
}
//periodic reset of u: reset is not complete but bit by bit
tCounter++;
if ((tCounter & ((ULL(1) << logUResetPeriod) - 1)) == 0) {
// reset least significant bit
// most significant bit becomes least significant bit
for (int i = 1; i <= nHistoryTables; i++) {
for (int j = 0; j < (ULL(1) << logTagTableSizes[i]); j++) {
gtable[i][j].u = gtable[i][j].u >> 1;
}
}
}
if (bi->hitBank > 0) {
DPRINTF(Tage, "Updating tag table entry (%d,%d) for branch %lx\n",
bi->hitBank, bi->hitBankIndex, branch_pc);
ctrUpdate(gtable[bi->hitBank][bi->hitBankIndex].ctr, taken,
tagTableCounterBits);
// if the provider entry is not certified to be useful also update
// the alternate prediction
if (gtable[bi->hitBank][bi->hitBankIndex].u == 0) {
if (bi->altBank > 0) {
ctrUpdate(gtable[bi->altBank][bi->altBankIndex].ctr, taken,
tagTableCounterBits);
DPRINTF(Tage, "Updating tag table entry (%d,%d) for"
" branch %lx\n", bi->hitBank, bi->hitBankIndex,
branch_pc);
}
if (bi->altBank == 0) {
baseUpdate(branch_pc, taken, bi);
}
}
// update the u counter
if (bi->tagePred != bi->altTaken) {
unsignedCtrUpdate(gtable[bi->hitBank][bi->hitBankIndex].u,
bi->tagePred == taken, tagTableUBits);
}
} else {
baseUpdate(branch_pc, taken, bi);
}
}
void
TAGE::updateHistories(ThreadID tid, Addr branch_pc, bool taken, void* b)
{
TageBranchInfo* bi = (TageBranchInfo*)(b);
ThreadHistory& tHist = threadHistory[tid];
// UPDATE HISTORIES
bool pathbit = ((branch_pc >> instShiftAmt) & 1);
//on a squash, return pointers to this and recompute indices.
//update user history
updateGHist(tHist.gHist, taken, tHist.globalHistory, tHist.ptGhist);
tHist.pathHist = (tHist.pathHist << 1) + pathbit;
tHist.pathHist = (tHist.pathHist & ((ULL(1) << pathHistBits) - 1));
bi->ptGhist = tHist.ptGhist;
bi->pathHist = tHist.pathHist;
//prepare next index and tag computations for user branchs
for (int i = 1; i <= nHistoryTables; i++)
{
bi->ci[i] = tHist.computeIndices[i].comp;
bi->ct0[i] = tHist.computeTags[0][i].comp;
bi->ct1[i] = tHist.computeTags[1][i].comp;
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
DPRINTF(Tage, "Updating global histories with branch:%lx; taken?:%d, "
"path Hist: %x; pointer:%d\n", branch_pc, taken, tHist.pathHist,
tHist.ptGhist);
}
void
TAGE::squash(ThreadID tid, bool taken, void *bp_history)
{
TageBranchInfo* bi = (TageBranchInfo*)(bp_history);
ThreadHistory& tHist = threadHistory[tid];
DPRINTF(Tage, "Restoring branch info: %lx; taken? %d; PathHistory:%x, "
"pointer:%d\n", bi->branchPC,taken, bi->pathHist, bi->ptGhist);
tHist.pathHist = bi->pathHist;
tHist.ptGhist = bi->ptGhist;
tHist.gHist = &(tHist.globalHistory[tHist.ptGhist]);
tHist.gHist[0] = (taken ? 1 : 0);
for (int i = 1; i <= nHistoryTables; i++) {
tHist.computeIndices[i].comp = bi->ci[i];
tHist.computeTags[0][i].comp = bi->ct0[i];
tHist.computeTags[1][i].comp = bi->ct1[i];
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
delete bi;
}
void
TAGE::squash(ThreadID tid, void *bp_history)
{
TageBranchInfo* bi = (TageBranchInfo*)(bp_history);
DPRINTF(Tage, "Deleting branch info: %lx\n", bi->branchPC);
TageBranchInfo *bi = static_cast<TageBranchInfo*>(bp_history);
DPRINTF(Tage, "Deleting branch info: %lx\n", bi->tageBranchInfo->branchPC);
delete bi;
}
bool
TAGE::predict(ThreadID tid, Addr branch_pc, bool cond_branch, void* &b)
{
TageBranchInfo *bi = new TageBranchInfo(*tage);
b = (void*)(bi);
return tage->tagePredict(tid, branch_pc, cond_branch, bi->tageBranchInfo);
}
bool
TAGE::lookup(ThreadID tid, Addr branch_pc, void* &bp_history)
{
bool retval = predict(tid, branch_pc, true, bp_history);
TageBranchInfo *bi = static_cast<TageBranchInfo*>(bp_history);
DPRINTF(Tage, "Lookup branch: %lx; predict:%d\n", branch_pc, retval);
updateHistories(tid, branch_pc, retval, bp_history);
assert(threadHistory[tid].gHist ==
&threadHistory[tid].globalHistory[threadHistory[tid].ptGhist]);
tage->updateHistories(tid, branch_pc, retval, bi->tageBranchInfo, true);
return retval;
}
@@ -581,19 +127,8 @@ TAGE::lookup(ThreadID tid, Addr branch_pc, void* &bp_history)
void
TAGE::btbUpdate(ThreadID tid, Addr branch_pc, void* &bp_history)
{
TageBranchInfo* bi = (TageBranchInfo*) bp_history;
ThreadHistory& tHist = threadHistory[tid];
DPRINTF(Tage, "BTB miss resets prediction: %lx\n", branch_pc);
assert(tHist.gHist == &tHist.globalHistory[tHist.ptGhist]);
tHist.gHist[0] = 0;
for (int i = 1; i <= nHistoryTables; i++) {
tHist.computeIndices[i].comp = bi->ci[i];
tHist.computeTags[0][i].comp = bi->ct0[i];
tHist.computeTags[1][i].comp = bi->ct1[i];
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
TageBranchInfo *bi = static_cast<TageBranchInfo*>(bp_history);
tage->btbUpdate(tid, branch_pc, bi->tageBranchInfo);
}
void
@@ -601,122 +136,8 @@ TAGE::uncondBranch(ThreadID tid, Addr br_pc, void* &bp_history)
{
DPRINTF(Tage, "UnConditionalBranch: %lx\n", br_pc);
predict(tid, br_pc, false, bp_history);
updateHistories(tid, br_pc, true, bp_history);
assert(threadHistory[tid].gHist ==
&threadHistory[tid].globalHistory[threadHistory[tid].ptGhist]);
}
void
TAGE::updateStats(bool taken, TageBranchInfo* bi)
{
if (taken == bi->tagePred) {
// correct prediction
switch (bi->provider) {
case BIMODAL_ONLY: tageBimodalProviderCorrect++; break;
case TAGE_LONGEST_MATCH: tageLongestMatchProviderCorrect++; break;
case BIMODAL_ALT_MATCH: bimodalAltMatchProviderCorrect++; break;
case TAGE_ALT_MATCH: tageAltMatchProviderCorrect++; break;
}
} else {
// wrong prediction
switch (bi->provider) {
case BIMODAL_ONLY: tageBimodalProviderWrong++; break;
case TAGE_LONGEST_MATCH:
tageLongestMatchProviderWrong++;
if (bi->altTaken == taken) {
tageAltMatchProviderWouldHaveHit++;
}
break;
case BIMODAL_ALT_MATCH:
bimodalAltMatchProviderWrong++;
break;
case TAGE_ALT_MATCH:
tageAltMatchProviderWrong++;
break;
}
switch (bi->provider) {
case BIMODAL_ALT_MATCH:
case TAGE_ALT_MATCH:
if (bi->longestMatchPred == taken) {
tageLongestMatchProviderWouldHaveHit++;
}
}
}
switch (bi->provider) {
case TAGE_LONGEST_MATCH:
case TAGE_ALT_MATCH:
tageLongestMatchProvider[bi->hitBank]++;
tageAltMatchProvider[bi->altBank]++;
break;
}
}
void
TAGE::regStats()
{
BPredUnit::regStats();
tageLongestMatchProviderCorrect
.name(name() + ".tageLongestMatchProviderCorrect")
.desc("Number of times TAGE Longest Match is the provider and "
"the prediction is correct");
tageAltMatchProviderCorrect
.name(name() + ".tageAltMatchProviderCorrect")
.desc("Number of times TAGE Alt Match is the provider and "
"the prediction is correct");
bimodalAltMatchProviderCorrect
.name(name() + ".bimodalAltMatchProviderCorrect")
.desc("Number of times TAGE Alt Match is the bimodal and it is the "
"provider and the prediction is correct");
tageBimodalProviderCorrect
.name(name() + ".tageBimodalProviderCorrect")
.desc("Number of times there are no hits on the TAGE tables "
"and the bimodal prediction is correct");
tageLongestMatchProviderWrong
.name(name() + ".tageLongestMatchProviderWrong")
.desc("Number of times TAGE Longest Match is the provider and "
"the prediction is wrong");
tageAltMatchProviderWrong
.name(name() + ".tageAltMatchProviderWrong")
.desc("Number of times TAGE Alt Match is the provider and "
"the prediction is wrong");
bimodalAltMatchProviderWrong
.name(name() + ".bimodalAltMatchProviderWrong")
.desc("Number of times TAGE Alt Match is the bimodal and it is the "
"provider and the prediction is wrong");
tageBimodalProviderWrong
.name(name() + ".tageBimodalProviderWrong")
.desc("Number of times there are no hits on the TAGE tables "
"and the bimodal prediction is wrong");
tageAltMatchProviderWouldHaveHit
.name(name() + ".tageAltMatchProviderWouldHaveHit")
.desc("Number of times TAGE Longest Match is the provider, "
"the prediction is wrong and Alt Match prediction was correct");
tageLongestMatchProviderWouldHaveHit
.name(name() + ".tageLongestMatchProviderWouldHaveHit")
.desc("Number of times TAGE Alt Match is the provider, the "
"prediction is wrong and Longest Match prediction was correct");
tageLongestMatchProvider
.init(nHistoryTables + 1)
.name(name() + ".tageLongestMatchProvider")
.desc("TAGE provider for longest match");
tageAltMatchProvider
.init(nHistoryTables + 1)
.name(name() + ".tageAltMatchProvider")
.desc("TAGE provider for alt match");
TageBranchInfo *bi = static_cast<TageBranchInfo*>(bp_history);
tage->updateHistories(tid, br_pc, true, bi->tageBranchInfo, true);
}
TAGE*

359
src/cpu/pred/tage.hh
View File

@@ -55,11 +55,30 @@
#include "base/types.hh"
#include "cpu/pred/bpred_unit.hh"
#include "cpu/pred/tage_base.hh"
#include "params/TAGE.hh"
class TAGE: public BPredUnit
{
protected:
TAGEBase *tage;
struct TageBranchInfo {
TAGEBase::BranchInfo *tageBranchInfo;
TageBranchInfo(TAGEBase &tage) : tageBranchInfo(tage.makeBranchInfo())
{}
virtual ~TageBranchInfo()
{
delete tageBranchInfo;
}
};
virtual bool predict(ThreadID tid, Addr branch_pc, bool cond_branch,
void* &b);
public:
TAGE(const TAGEParams *params);
// Base class methods.
@@ -67,346 +86,10 @@ class TAGE: public BPredUnit
bool lookup(ThreadID tid, Addr branch_addr, void* &bp_history) override;
void btbUpdate(ThreadID tid, Addr branch_addr, void* &bp_history) override;
void update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed) override;
bool squashed, const StaticInstPtr & inst,
Addr corrTarget) override;
virtual void squash(ThreadID tid, void *bp_history) override;
unsigned getGHR(ThreadID tid, void *bp_history) const override;
virtual void regStats() override;
protected:
// Prediction Structures
// Tage Entry
struct TageEntry
{
int8_t ctr;
uint16_t tag;
uint8_t u;
TageEntry() : ctr(0), tag(0), u(0) { }
};
// Folded History Table - compressed history
// to mix with instruction PC to index partially
// tagged tables.
struct FoldedHistory
{
unsigned comp;
int compLength;
int origLength;
int outpoint;
void init(int original_length, int compressed_length)
{
comp = 0;
origLength = original_length;
compLength = compressed_length;
outpoint = original_length % compressed_length;
}
void update(uint8_t * h)
{
comp = (comp << 1) | h[0];
comp ^= h[origLength] << outpoint;
comp ^= (comp >> compLength);
comp &= (ULL(1) << compLength) - 1;
}
};
// provider type
enum {
BIMODAL_ONLY = 0,
TAGE_LONGEST_MATCH,
BIMODAL_ALT_MATCH,
TAGE_ALT_MATCH,
LAST_TAGE_PROVIDER_TYPE = TAGE_ALT_MATCH
};
// Primary branch history entry
struct TageBranchInfo
{
int pathHist;
int ptGhist;
int hitBank;
int hitBankIndex;
int altBank;
int altBankIndex;
int bimodalIndex;
bool tagePred;
bool altTaken;
bool condBranch;
bool longestMatchPred;
bool pseudoNewAlloc;
Addr branchPC;
// Pointer to dynamically allocated storage
// to save table indices and folded histories.
// To do one call to new instead of five.
int *storage;
// Pointers to actual saved array within the dynamically
// allocated storage.
int *tableIndices;
int *tableTags;
int *ci;
int *ct0;
int *ct1;
// for stats purposes
unsigned provider;
TageBranchInfo(int sz)
: pathHist(0), ptGhist(0),
hitBank(0), hitBankIndex(0),
altBank(0), altBankIndex(0),
bimodalIndex(0),
tagePred(false), altTaken(false),
condBranch(false), longestMatchPred(false),
pseudoNewAlloc(false), branchPC(0),
provider(-1)
{
storage = new int [sz * 5];
tableIndices = storage;
tableTags = storage + sz;
ci = tableTags + sz;
ct0 = ci + sz;
ct1 = ct0 + sz;
}
virtual ~TageBranchInfo()
{
delete[] storage;
}
};
/**
* Computes the index used to access the
* bimodal table.
* @param pc_in The unshifted branch PC.
*/
int bindex(Addr pc_in) const;
/**
* Computes the index used to access a
* partially tagged table.
* @param tid The thread ID used to select the
* global histories to use.
* @param pc The unshifted branch PC.
* @param bank The partially tagged table to access.
*/
inline int gindex(ThreadID tid, Addr pc, int bank) const;
/**
* Utility function to shuffle the path history
* depending on which tagged table we are accessing.
* @param phist The path history.
* @param size Number of path history bits to use.
* @param bank The partially tagged table to access.
*/
int F(int phist, int size, int bank) const;
/**
* Computes the partial tag of a tagged table.
* @param tid the thread ID used to select the
* global histories to use.
* @param pc The unshifted branch PC.
* @param bank The partially tagged table to access.
*/
inline uint16_t gtag(ThreadID tid, Addr pc, int bank) const;
/**
* Updates a direction counter based on the actual
* branch outcome.
* @param ctr Reference to counter to update.
* @param taken Actual branch outcome.
* @param nbits Counter width.
*/
void ctrUpdate(int8_t & ctr, bool taken, int nbits);
/**
* Updates an unsigned counter based on up/down parameter
* @param ctr Reference to counter to update.
* @param up Boolean indicating if the counter is incremented/decremented
* If true it is incremented, if false it is decremented
* @param nbits Counter width.
*/
void unsignedCtrUpdate(uint8_t & ctr, bool up, unsigned nbits);
/**
* Get a branch prediction from the bimodal
* predictor.
* @param pc The unshifted branch PC.
* @param bi Pointer to information on the
* prediction.
*/
bool getBimodePred(Addr pc, TageBranchInfo* bi) const;
/**
* Updates the bimodal predictor.
* @param pc The unshifted branch PC.
* @param taken The actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
void baseUpdate(Addr pc, bool taken, TageBranchInfo* bi);
/**
* (Speculatively) updates the global branch history.
* @param h Reference to pointer to global branch history.
* @param dir (Predicted) outcome to update the histories
* with.
* @param tab
* @param PT Reference to path history.
*/
void updateGHist(uint8_t * &h, bool dir, uint8_t * tab, int &PT);
/**
* Get a branch prediction from TAGE. *NOT* an override of
* BpredUnit::predict().
* @param tid The thread ID to select the global
* histories to use.
* @param branch_pc The unshifted branch PC.
* @param cond_branch True if the branch is conditional.
* @param b Reference to wrapping pointer to allow storing
* derived class prediction information in the base class.
*/
virtual bool predict(
ThreadID tid, Addr branch_pc, bool cond_branch, void* &b);
/**
* Update TAGE. Called at execute to repair histories on a misprediction
* and at commit to update the tables.
* @param tid The thread ID to select the global
* histories to use.
* @param branch_pc The unshifted branch PC.
* @param taken Actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
void update(ThreadID tid, Addr branch_pc, bool taken, TageBranchInfo* bi);
/**
* (Speculatively) updates global histories (path and direction).
* Also recomputes compressed (folded) histories based on the
* branch direction.
* @param tid The thread ID to select the histories
* to update.
* @param branch_pc The unshifted branch PC.
* @param taken (Predicted) branch direction.
* @param b Wrapping pointer to TageBranchInfo (to allow
* storing derived class prediction information in the
* base class).
*/
void updateHistories(ThreadID tid, Addr branch_pc, bool taken, void* b);
/**
* Restores speculatively updated path and direction histories.
* Also recomputes compressed (folded) histories based on the
* correct branch outcome.
* This version of squash() is called once on a branch misprediction.
* @param tid The Thread ID to select the histories to rollback.
* @param taken The correct branch outcome.
* @param bp_history Wrapping pointer to TageBranchInfo (to allow
* storing derived class prediction information in the
* base class).
* @post bp_history points to valid memory.
*/
virtual void squash(ThreadID tid, bool taken, void *bp_history);
/**
* Update TAGE for conditional branches.
* @param branch_pc The unshifted branch PC.
* @param taken Actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
* @nrand Random int number from 0 to 3
*/
virtual void condBranchUpdate(
Addr branch_pc, bool taken, TageBranchInfo* bi, int nrand);
/**
* TAGE prediction called from TAGE::predict
* @param tid The thread ID to select the global
* histories to use.
* @param branch_pc The unshifted branch PC.
* @param cond_branch True if the branch is conditional.
* @param bi Pointer to the TageBranchInfo
*/
bool tagePredict(
ThreadID tid, Addr branch_pc, bool cond_branch, TageBranchInfo* bi);
/**
* Update the stats
* @param taken Actual branch outcome
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
virtual void updateStats(bool taken, TageBranchInfo* bi);
const unsigned logRatioBiModalHystEntries;
const unsigned nHistoryTables;
const unsigned tagTableCounterBits;
const unsigned tagTableUBits;
const unsigned histBufferSize;
const unsigned minHist;
const unsigned maxHist;
const unsigned pathHistBits;
const std::vector<unsigned> tagTableTagWidths;
const std::vector<int> logTagTableSizes;
std::vector<bool> btablePrediction;
std::vector<bool> btableHysteresis;
TageEntry **gtable;
// Keep per-thread histories to
// support SMT.
struct ThreadHistory {
// Speculative path history
// (LSB of branch address)
int pathHist;
// Speculative branch direction
// history (circular buffer)
// @TODO Convert to std::vector<bool>
uint8_t *globalHistory;
// Pointer to most recent branch outcome
uint8_t* gHist;
// Index to most recent branch outcome
int ptGhist;
// Speculative folded histories.
FoldedHistory *computeIndices;
FoldedHistory *computeTags[2];
};
std::vector<ThreadHistory> threadHistory;
int *histLengths;
int *tableIndices;
int *tableTags;
int8_t useAltPredForNewlyAllocated;
uint64_t tCounter;
uint64_t logUResetPeriod;
unsigned useAltOnNaBits;
// stats
Stats::Scalar tageLongestMatchProviderCorrect;
Stats::Scalar tageAltMatchProviderCorrect;
Stats::Scalar bimodalAltMatchProviderCorrect;
Stats::Scalar tageBimodalProviderCorrect;
Stats::Scalar tageLongestMatchProviderWrong;
Stats::Scalar tageAltMatchProviderWrong;
Stats::Scalar bimodalAltMatchProviderWrong;
Stats::Scalar tageBimodalProviderWrong;
Stats::Scalar tageAltMatchProviderWouldHaveHit;
Stats::Scalar tageLongestMatchProviderWouldHaveHit;
Stats::Vector tageLongestMatchProvider;
Stats::Vector tageAltMatchProvider;
};
#endif // __CPU_PRED_TAGE

803
src/cpu/pred/tage_base.cc Normal file
View File

@@ -0,0 +1,803 @@
/*
* Copyright (c) 2014 The University of Wisconsin
*
* Copyright (c) 2006 INRIA (Institut National de Recherche en
* Informatique et en Automatique / French National Research Institute
* for Computer Science and Applied Mathematics)
*
* 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: Vignyan Reddy, Dibakar Gope and Arthur Perais,
* from André Seznec's code.
*/
/* @file
* Implementation of a TAGE branch predictor
*/
#include "cpu/pred/tage_base.hh"
#include "base/intmath.hh"
#include "base/logging.hh"
#include "base/random.hh"
#include "debug/Fetch.hh"
#include "debug/Tage.hh"
TAGEBase::TAGEBase(const TAGEBaseParams *p)
: SimObject(p),
logRatioBiModalHystEntries(p->logRatioBiModalHystEntries),
nHistoryTables(p->nHistoryTables),
tagTableCounterBits(p->tagTableCounterBits),
tagTableUBits(p->tagTableUBits),
histBufferSize(p->histBufferSize),
minHist(p->minHist),
maxHist(p->maxHist),
pathHistBits(p->pathHistBits),
tagTableTagWidths(p->tagTableTagWidths),
logTagTableSizes(p->logTagTableSizes),
threadHistory(p->numThreads),
logUResetPeriod(p->logUResetPeriod),
numUseAltOnNa(p->numUseAltOnNa),
useAltOnNaBits(p->useAltOnNaBits),
maxNumAlloc(p->maxNumAlloc),
noSkip(p->noSkip),
speculativeHistUpdate(p->speculativeHistUpdate),
instShiftAmt(p->instShiftAmt)
{
if (noSkip.empty()) {
// Set all the table to enabled by default
noSkip.resize(nHistoryTables + 1, true);
}
}
TAGEBase::BranchInfo*
TAGEBase::makeBranchInfo() {
return new BranchInfo(*this);
}
void
TAGEBase::init()
{
// Current method for periodically resetting the u counter bits only
// works for 1 or 2 bits
// Also make sure that it is not 0
assert(tagTableUBits <= 2 && (tagTableUBits > 0));
// we use int type for the path history, so it cannot be more than
// its size
assert(pathHistBits <= (sizeof(int)*8));
// initialize the counter to half of the period
assert(logUResetPeriod != 0);
tCounter = ULL(1) << (logUResetPeriod - 1);
assert(histBufferSize > maxHist * 2);
useAltPredForNewlyAllocated.resize(numUseAltOnNa, 0);
for (auto& history : threadHistory) {
history.pathHist = 0;
history.globalHistory = new uint8_t[histBufferSize];
history.gHist = history.globalHistory;
memset(history.gHist, 0, histBufferSize);
history.ptGhist = 0;
}
histLengths = new int [nHistoryTables+1];
calculateParameters();
assert(tagTableTagWidths.size() == (nHistoryTables+1));
assert(logTagTableSizes.size() == (nHistoryTables+1));
// First entry is for the Bimodal table and it is untagged in this
// implementation
assert(tagTableTagWidths[0] == 0);
for (auto& history : threadHistory) {
history.computeIndices = new FoldedHistory[nHistoryTables+1];
history.computeTags[0] = new FoldedHistory[nHistoryTables+1];
history.computeTags[1] = new FoldedHistory[nHistoryTables+1];
initFoldedHistories(history);
}
const uint64_t bimodalTableSize = ULL(1) << logTagTableSizes[0];
btablePrediction.resize(bimodalTableSize, false);
btableHysteresis.resize(bimodalTableSize >> logRatioBiModalHystEntries,
true);
gtable = new TageEntry*[nHistoryTables + 1];
buildTageTables();
tableIndices = new int [nHistoryTables+1];
tableTags = new int [nHistoryTables+1];
}
void
TAGEBase::initFoldedHistories(ThreadHistory & history)
{
for (int i = 1; i <= nHistoryTables; i++) {
history.computeIndices[i].init(
histLengths[i], (logTagTableSizes[i]));
history.computeTags[0][i].init(
history.computeIndices[i].origLength, tagTableTagWidths[i]);
history.computeTags[1][i].init(
history.computeIndices[i].origLength, tagTableTagWidths[i]-1);
DPRINTF(Tage, "HistLength:%d, TTSize:%d, TTTWidth:%d\n",
histLengths[i], logTagTableSizes[i], tagTableTagWidths[i]);
}
}
void
TAGEBase::buildTageTables()
{
for (int i = 1; i <= nHistoryTables; i++) {
gtable[i] = new TageEntry[1<<(logTagTableSizes[i])];
}
}
void
TAGEBase::calculateParameters()
{
histLengths[1] = minHist;
histLengths[nHistoryTables] = maxHist;
for (int i = 2; i <= nHistoryTables; i++) {
histLengths[i] = (int) (((double) minHist *
pow ((double) (maxHist) / (double) minHist,
(double) (i - 1) / (double) ((nHistoryTables- 1))))
+ 0.5);
}
}
void
TAGEBase::btbUpdate(ThreadID tid, Addr branch_pc, BranchInfo* &bi)
{
if (speculativeHistUpdate) {
ThreadHistory& tHist = threadHistory[tid];
DPRINTF(Tage, "BTB miss resets prediction: %lx\n", branch_pc);
assert(tHist.gHist == &tHist.globalHistory[tHist.ptGhist]);
tHist.gHist[0] = 0;
for (int i = 1; i <= nHistoryTables; i++) {
tHist.computeIndices[i].comp = bi->ci[i];
tHist.computeTags[0][i].comp = bi->ct0[i];
tHist.computeTags[1][i].comp = bi->ct1[i];
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
}
}
int
TAGEBase::bindex(Addr pc_in) const
{
return ((pc_in >> instShiftAmt) & ((ULL(1) << (logTagTableSizes[0])) - 1));
}
int
TAGEBase::F(int A, int size, int bank) const
{
int A1, A2;
A = A & ((ULL(1) << size) - 1);
A1 = (A & ((ULL(1) << logTagTableSizes[bank]) - 1));
A2 = (A >> logTagTableSizes[bank]);
A2 = ((A2 << bank) & ((ULL(1) << logTagTableSizes[bank]) - 1))
+ (A2 >> (logTagTableSizes[bank] - bank));
A = A1 ^ A2;
A = ((A << bank) & ((ULL(1) << logTagTableSizes[bank]) - 1))
+ (A >> (logTagTableSizes[bank] - bank));
return (A);
}
// gindex computes a full hash of pc, ghist and pathHist
int
TAGEBase::gindex(ThreadID tid, Addr pc, int bank) const
{
int index;
int hlen = (histLengths[bank] > pathHistBits) ? pathHistBits :
histLengths[bank];
const unsigned int shiftedPc = pc >> instShiftAmt;
index =
shiftedPc ^
(shiftedPc >> ((int) abs(logTagTableSizes[bank] - bank) + 1)) ^
threadHistory[tid].computeIndices[bank].comp ^
F(threadHistory[tid].pathHist, hlen, bank);
return (index & ((ULL(1) << (logTagTableSizes[bank])) - 1));
}
// Tag computation
uint16_t
TAGEBase::gtag(ThreadID tid, Addr pc, int bank) const
{
int tag = (pc >> instShiftAmt) ^
threadHistory[tid].computeTags[0][bank].comp ^
(threadHistory[tid].computeTags[1][bank].comp << 1);
return (tag & ((ULL(1) << tagTableTagWidths[bank]) - 1));
}
// Up-down saturating counter
template<typename T>
void
TAGEBase::ctrUpdate(T & ctr, bool taken, int nbits)
{
assert(nbits <= sizeof(T) << 3);
if (taken) {
if (ctr < ((1 << (nbits - 1)) - 1))
ctr++;
} else {
if (ctr > -(1 << (nbits - 1)))
ctr--;
}
}
// int8_t and int versions of this function may be needed
template void TAGEBase::ctrUpdate(int8_t & ctr, bool taken, int nbits);
template void TAGEBase::ctrUpdate(int & ctr, bool taken, int nbits);
// Up-down unsigned saturating counter
void
TAGEBase::unsignedCtrUpdate(uint8_t & ctr, bool up, unsigned nbits)
{
assert(nbits <= sizeof(uint8_t) << 3);
if (up) {
if (ctr < ((1 << nbits) - 1))
ctr++;
} else {
if (ctr)
ctr--;
}
}
// Bimodal prediction
bool
TAGEBase::getBimodePred(Addr pc, BranchInfo* bi) const
{
return btablePrediction[bi->bimodalIndex];
}
// Update the bimodal predictor: a hysteresis bit is shared among N prediction
// bits (N = 2 ^ logRatioBiModalHystEntries)
void
TAGEBase::baseUpdate(Addr pc, bool taken, BranchInfo* bi)
{
int inter = (btablePrediction[bi->bimodalIndex] << 1)
+ btableHysteresis[bi->bimodalIndex >> logRatioBiModalHystEntries];
if (taken) {
if (inter < 3)
inter++;
} else if (inter > 0) {
inter--;
}
const bool pred = inter >> 1;
const bool hyst = inter & 1;
btablePrediction[bi->bimodalIndex] = pred;
btableHysteresis[bi->bimodalIndex >> logRatioBiModalHystEntries] = hyst;
DPRINTF(Tage, "Updating branch %lx, pred:%d, hyst:%d\n", pc, pred, hyst);
}
// shifting the global history: we manage the history in a big table in order
// to reduce simulation time
void
TAGEBase::updateGHist(uint8_t * &h, bool dir, uint8_t * tab, int &pt)
{
if (pt == 0) {
DPRINTF(Tage, "Rolling over the histories\n");
// Copy beginning of globalHistoryBuffer to end, such that
// the last maxHist outcomes are still reachable
// through pt[0 .. maxHist - 1].
for (int i = 0; i < maxHist; i++)
tab[histBufferSize - maxHist + i] = tab[i];
pt = histBufferSize - maxHist;
h = &tab[pt];
}
pt--;
h--;
h[0] = (dir) ? 1 : 0;
}
void
TAGEBase::calculateIndicesAndTags(ThreadID tid, Addr branch_pc,
BranchInfo* bi)
{
// computes the table addresses and the partial tags
for (int i = 1; i <= nHistoryTables; i++) {
tableIndices[i] = gindex(tid, branch_pc, i);
bi->tableIndices[i] = tableIndices[i];
tableTags[i] = gtag(tid, branch_pc, i);
bi->tableTags[i] = tableTags[i];
}
}
unsigned
TAGEBase::getUseAltIdx(BranchInfo* bi)
{
// There is only 1 counter on the base TAGE implementation
return 0;
}
bool
TAGEBase::tagePredict(ThreadID tid, Addr branch_pc,
bool cond_branch, BranchInfo* bi)
{
Addr pc = branch_pc;
bool pred_taken = true;
if (cond_branch) {
// TAGE prediction
calculateIndicesAndTags(tid, pc, bi);
bi->bimodalIndex = bindex(pc);
bi->hitBank = 0;
bi->altBank = 0;
//Look for the bank with longest matching history
for (int i = nHistoryTables; i > 0; i--) {
if (noSkip[i] &&
gtable[i][tableIndices[i]].tag == tableTags[i]) {
bi->hitBank = i;
bi->hitBankIndex = tableIndices[bi->hitBank];
break;
}
}
//Look for the alternate bank
for (int i = bi->hitBank - 1; i > 0; i--) {
if (noSkip[i] &&
gtable[i][tableIndices[i]].tag == tableTags[i]) {
bi->altBank = i;
bi->altBankIndex = tableIndices[bi->altBank];
break;
}
}
//computes the prediction and the alternate prediction
if (bi->hitBank > 0) {
if (bi->altBank > 0) {
bi->altTaken =
gtable[bi->altBank][tableIndices[bi->altBank]].ctr >= 0;
extraAltCalc(bi);
}else {
bi->altTaken = getBimodePred(pc, bi);
}
bi->longestMatchPred =
gtable[bi->hitBank][tableIndices[bi->hitBank]].ctr >= 0;
bi->pseudoNewAlloc =
abs(2 * gtable[bi->hitBank][bi->hitBankIndex].ctr + 1) <= 1;
//if the entry is recognized as a newly allocated entry and
//useAltPredForNewlyAllocated is positive use the alternate
//prediction
if ((useAltPredForNewlyAllocated[getUseAltIdx(bi)] < 0) ||
! bi->pseudoNewAlloc) {
bi->tagePred = bi->longestMatchPred;
bi->provider = TAGE_LONGEST_MATCH;
} else {
bi->tagePred = bi->altTaken;
bi->provider = bi->altBank ? TAGE_ALT_MATCH
: BIMODAL_ALT_MATCH;
}
} else {
bi->altTaken = getBimodePred(pc, bi);
bi->tagePred = bi->altTaken;
bi->longestMatchPred = bi->altTaken;
bi->provider = BIMODAL_ONLY;
}
//end TAGE prediction
pred_taken = (bi->tagePred);
DPRINTF(Tage, "Predict for %lx: taken?:%d, tagePred:%d, altPred:%d\n",
branch_pc, pred_taken, bi->tagePred, bi->altTaken);
}
bi->branchPC = branch_pc;
bi->condBranch = cond_branch;
return pred_taken;
}
void
TAGEBase::adjustAlloc(bool & alloc, bool taken)
{
// Nothing for this base class implementation
}
void
TAGEBase::handleAllocAndUReset(bool alloc, bool taken, BranchInfo* bi,
int nrand)
{
if (alloc) {
// is there some "unuseful" entry to allocate
uint8_t min = 1;
for (int i = nHistoryTables; i > bi->hitBank; i--) {
if (gtable[i][bi->tableIndices[i]].u < min) {
min = gtable[i][bi->tableIndices[i]].u;
}
}
// we allocate an entry with a longer history
// to avoid ping-pong, we do not choose systematically the next
// entry, but among the 3 next entries
int Y = nrand &
((ULL(1) << (nHistoryTables - bi->hitBank - 1)) - 1);
int X = bi->hitBank + 1;
if (Y & 1) {
X++;
if (Y & 2)
X++;
}
// No entry available, forces one to be available
if (min > 0) {
gtable[X][bi->tableIndices[X]].u = 0;
}
//Allocate entries
unsigned numAllocated = 0;
for (int i = X; i <= nHistoryTables; i++) {
if ((gtable[i][bi->tableIndices[i]].u == 0)) {
gtable[i][bi->tableIndices[i]].tag = bi->tableTags[i];
gtable[i][bi->tableIndices[i]].ctr = (taken) ? 0 : -1;
++numAllocated;
if (numAllocated == maxNumAlloc) {
break;
}
}
}
}
tCounter++;
handleUReset();
}
void
TAGEBase::handleUReset()
{
//periodic reset of u: reset is not complete but bit by bit
if ((tCounter & ((ULL(1) << logUResetPeriod) - 1)) == 0) {
// reset least significant bit
// most significant bit becomes least significant bit
for (int i = 1; i <= nHistoryTables; i++) {
for (int j = 0; j < (ULL(1) << logTagTableSizes[i]); j++) {
resetUctr(gtable[i][j].u);
}
}
}
}
void
TAGEBase::resetUctr(uint8_t & u)
{
u >>= 1;
}
void
TAGEBase::condBranchUpdate(ThreadID tid, Addr branch_pc, bool taken,
BranchInfo* bi, int nrand, Addr corrTarget)
{
// TAGE UPDATE
// try to allocate a new entries only if prediction was wrong
bool alloc = (bi->tagePred != taken) && (bi->hitBank < nHistoryTables);
if (bi->hitBank > 0) {
// Manage the selection between longest matching and alternate
// matching for "pseudo"-newly allocated longest matching entry
bool PseudoNewAlloc = bi->pseudoNewAlloc;
// an entry is considered as newly allocated if its prediction
// counter is weak
if (PseudoNewAlloc) {
if (bi->longestMatchPred == taken) {
alloc = false;
}
// if it was delivering the correct prediction, no need to
// allocate new entry even if the overall prediction was false
if (bi->longestMatchPred != bi->altTaken) {
ctrUpdate(useAltPredForNewlyAllocated[getUseAltIdx(bi)],
bi->altTaken == taken, useAltOnNaBits);
}
}
}
adjustAlloc(alloc, taken);
handleAllocAndUReset(alloc, taken, bi, nrand);
handleTAGEUpdate(branch_pc, taken, bi);
}
void
TAGEBase::handleTAGEUpdate(Addr branch_pc, bool taken, BranchInfo* bi)
{
if (bi->hitBank > 0) {
DPRINTF(Tage, "Updating tag table entry (%d,%d) for branch %lx\n",
bi->hitBank, bi->hitBankIndex, branch_pc);
ctrUpdate(gtable[bi->hitBank][bi->hitBankIndex].ctr, taken,
tagTableCounterBits);
// if the provider entry is not certified to be useful also update
// the alternate prediction
if (gtable[bi->hitBank][bi->hitBankIndex].u == 0) {
if (bi->altBank > 0) {
ctrUpdate(gtable[bi->altBank][bi->altBankIndex].ctr, taken,
tagTableCounterBits);
DPRINTF(Tage, "Updating tag table entry (%d,%d) for"
" branch %lx\n", bi->hitBank, bi->hitBankIndex,
branch_pc);
}
if (bi->altBank == 0) {
baseUpdate(branch_pc, taken, bi);
}
}
// update the u counter
if (bi->tagePred != bi->altTaken) {
unsignedCtrUpdate(gtable[bi->hitBank][bi->hitBankIndex].u,
bi->tagePred == taken, tagTableUBits);
}
} else {
baseUpdate(branch_pc, taken, bi);
}
}
void
TAGEBase::updateHistories(ThreadID tid, Addr branch_pc, bool taken,
BranchInfo* bi, bool speculative,
const StaticInstPtr &inst, Addr target)
{
if (speculative != speculativeHistUpdate) {
return;
}
ThreadHistory& tHist = threadHistory[tid];
// UPDATE HISTORIES
bool pathbit = ((branch_pc >> instShiftAmt) & 1);
//on a squash, return pointers to this and recompute indices.
//update user history
updateGHist(tHist.gHist, taken, tHist.globalHistory, tHist.ptGhist);
tHist.pathHist = (tHist.pathHist << 1) + pathbit;
tHist.pathHist = (tHist.pathHist & ((ULL(1) << pathHistBits) - 1));
if (speculative) {
bi->ptGhist = tHist.ptGhist;
bi->pathHist = tHist.pathHist;
}
//prepare next index and tag computations for user branchs
for (int i = 1; i <= nHistoryTables; i++)
{
if (speculative) {
bi->ci[i] = tHist.computeIndices[i].comp;
bi->ct0[i] = tHist.computeTags[0][i].comp;
bi->ct1[i] = tHist.computeTags[1][i].comp;
}
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
DPRINTF(Tage, "Updating global histories with branch:%lx; taken?:%d, "
"path Hist: %x; pointer:%d\n", branch_pc, taken, tHist.pathHist,
tHist.ptGhist);
assert(threadHistory[tid].gHist ==
&threadHistory[tid].globalHistory[threadHistory[tid].ptGhist]);
}
void
TAGEBase::squash(ThreadID tid, bool taken, TAGEBase::BranchInfo *bi,
Addr target)
{
if (!speculativeHistUpdate) {
/* If there are no speculative updates, no actions are needed */
return;
}
ThreadHistory& tHist = threadHistory[tid];
DPRINTF(Tage, "Restoring branch info: %lx; taken? %d; PathHistory:%x, "
"pointer:%d\n", bi->branchPC,taken, bi->pathHist, bi->ptGhist);
tHist.pathHist = bi->pathHist;
tHist.ptGhist = bi->ptGhist;
tHist.gHist = &(tHist.globalHistory[tHist.ptGhist]);
tHist.gHist[0] = (taken ? 1 : 0);
for (int i = 1; i <= nHistoryTables; i++) {
tHist.computeIndices[i].comp = bi->ci[i];
tHist.computeTags[0][i].comp = bi->ct0[i];
tHist.computeTags[1][i].comp = bi->ct1[i];
tHist.computeIndices[i].update(tHist.gHist);
tHist.computeTags[0][i].update(tHist.gHist);
tHist.computeTags[1][i].update(tHist.gHist);
}
}
void
TAGEBase::extraAltCalc(BranchInfo* bi)
{
// do nothing. This is only used in some derived classes
return;
}
int
TAGEBase::getRandom()
{
return random_mt.random<int>();
}
void
TAGEBase::updateStats(bool taken, BranchInfo* bi)
{
if (taken == bi->tagePred) {
// correct prediction
switch (bi->provider) {
case BIMODAL_ONLY: tageBimodalProviderCorrect++; break;
case TAGE_LONGEST_MATCH: tageLongestMatchProviderCorrect++; break;
case BIMODAL_ALT_MATCH: bimodalAltMatchProviderCorrect++; break;
case TAGE_ALT_MATCH: tageAltMatchProviderCorrect++; break;
}
} else {
// wrong prediction
switch (bi->provider) {
case BIMODAL_ONLY: tageBimodalProviderWrong++; break;
case TAGE_LONGEST_MATCH:
tageLongestMatchProviderWrong++;
if (bi->altTaken == taken) {
tageAltMatchProviderWouldHaveHit++;
}
break;
case BIMODAL_ALT_MATCH:
bimodalAltMatchProviderWrong++;
break;
case TAGE_ALT_MATCH:
tageAltMatchProviderWrong++;
break;
}
switch (bi->provider) {
case BIMODAL_ALT_MATCH:
case TAGE_ALT_MATCH:
if (bi->longestMatchPred == taken) {
tageLongestMatchProviderWouldHaveHit++;
}
}
}
switch (bi->provider) {
case TAGE_LONGEST_MATCH:
case TAGE_ALT_MATCH:
tageLongestMatchProvider[bi->hitBank]++;
tageAltMatchProvider[bi->altBank]++;
break;
}
}
unsigned
TAGEBase::getGHR(ThreadID tid, BranchInfo *bi) const
{
unsigned val = 0;
for (unsigned i = 0; i < 32; i++) {
// Make sure we don't go out of bounds
int gh_offset = bi->ptGhist + i;
assert(&(threadHistory[tid].globalHistory[gh_offset]) <
threadHistory[tid].globalHistory + histBufferSize);
val |= ((threadHistory[tid].globalHistory[gh_offset] & 0x1) << i);
}
return val;
}
void
TAGEBase::regStats()
{
tageLongestMatchProviderCorrect
.name(name() + ".tageLongestMatchProviderCorrect")
.desc("Number of times TAGE Longest Match is the provider and "
"the prediction is correct");
tageAltMatchProviderCorrect
.name(name() + ".tageAltMatchProviderCorrect")
.desc("Number of times TAGE Alt Match is the provider and "
"the prediction is correct");
bimodalAltMatchProviderCorrect
.name(name() + ".bimodalAltMatchProviderCorrect")
.desc("Number of times TAGE Alt Match is the bimodal and it is the "
"provider and the prediction is correct");
tageBimodalProviderCorrect
.name(name() + ".tageBimodalProviderCorrect")
.desc("Number of times there are no hits on the TAGE tables "
"and the bimodal prediction is correct");
tageLongestMatchProviderWrong
.name(name() + ".tageLongestMatchProviderWrong")
.desc("Number of times TAGE Longest Match is the provider and "
"the prediction is wrong");
tageAltMatchProviderWrong
.name(name() + ".tageAltMatchProviderWrong")
.desc("Number of times TAGE Alt Match is the provider and "
"the prediction is wrong");
bimodalAltMatchProviderWrong
.name(name() + ".bimodalAltMatchProviderWrong")
.desc("Number of times TAGE Alt Match is the bimodal and it is the "
"provider and the prediction is wrong");
tageBimodalProviderWrong
.name(name() + ".tageBimodalProviderWrong")
.desc("Number of times there are no hits on the TAGE tables "
"and the bimodal prediction is wrong");
tageAltMatchProviderWouldHaveHit
.name(name() + ".tageAltMatchProviderWouldHaveHit")
.desc("Number of times TAGE Longest Match is the provider, "
"the prediction is wrong and Alt Match prediction was correct");
tageLongestMatchProviderWouldHaveHit
.name(name() + ".tageLongestMatchProviderWouldHaveHit")
.desc("Number of times TAGE Alt Match is the provider, the "
"prediction is wrong and Longest Match prediction was correct");
tageLongestMatchProvider
.init(nHistoryTables + 1)
.name(name() + ".tageLongestMatchProvider")
.desc("TAGE provider for longest match");
tageAltMatchProvider
.init(nHistoryTables + 1)
.name(name() + ".tageAltMatchProvider")
.desc("TAGE provider for alt match");
}
int8_t
TAGEBase::getCtr(int hitBank, int hitBankIndex) const
{
return gtable[hitBank][hitBankIndex].ctr;
}
unsigned
TAGEBase::getTageCtrBits() const
{
return tagTableCounterBits;
}
int
TAGEBase::getPathHist(ThreadID tid) const
{
return threadHistory[tid].pathHist;
}
bool
TAGEBase::isSpeculativeUpdateEnabled() const
{
return speculativeHistUpdate;
}
TAGEBase*
TAGEBaseParams::create()
{
return new TAGEBase(this);
}

500
src/cpu/pred/tage_base.hh Normal file
View File

@@ -0,0 +1,500 @@
/*
* Copyright (c) 2014 The University of Wisconsin
*
* Copyright (c) 2006 INRIA (Institut National de Recherche en
* Informatique et en Automatique / French National Research Institute
* for Computer Science and Applied Mathematics)
*
* 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: Vignyan Reddy, Dibakar Gope and Arthur Perais,
* from André Seznec's code.
*/
/* @file
* Implementation of a TAGE branch predictor. TAGE is a global-history based
* branch predictor. It features a PC-indexed bimodal predictor and N
* partially tagged tables, indexed with a hash of the PC and the global
* branch history. The different lengths of global branch history used to
* index the partially tagged tables grow geometrically. A small path history
* is also used in the hash.
*
* All TAGE tables are accessed in parallel, and the one using the longest
* history that matches provides the prediction (some exceptions apply).
* Entries are allocated in components using a longer history than the
* one that predicted when the prediction is incorrect.
*/
#ifndef __CPU_PRED_TAGE_BASE
#define __CPU_PRED_TAGE_BASE
#include <vector>
#include "base/statistics.hh"
#include "cpu/static_inst.hh"
#include "params/TAGEBase.hh"
#include "sim/sim_object.hh"
class TAGEBase : public SimObject
{
public:
TAGEBase(const TAGEBaseParams *p);
void regStats() override;
void init() override;
protected:
// Prediction Structures
// Tage Entry
struct TageEntry
{
int8_t ctr;
uint16_t tag;
uint8_t u;
TageEntry() : ctr(0), tag(0), u(0) { }
};
// Folded History Table - compressed history
// to mix with instruction PC to index partially
// tagged tables.
struct FoldedHistory
{
unsigned comp;
int compLength;
int origLength;
int outpoint;
int bufferSize;
FoldedHistory()
{
comp = 0;
}
void init(int original_length, int compressed_length)
{
origLength = original_length;
compLength = compressed_length;
outpoint = original_length % compressed_length;
}
void update(uint8_t * h)
{
comp = (comp << 1) | h[0];
comp ^= h[origLength] << outpoint;
comp ^= (comp >> compLength);
comp &= (ULL(1) << compLength) - 1;
}
};
public:
// provider type
enum {
BIMODAL_ONLY = 0,
TAGE_LONGEST_MATCH,
BIMODAL_ALT_MATCH,
TAGE_ALT_MATCH,
LAST_TAGE_PROVIDER_TYPE = TAGE_ALT_MATCH
};
// Primary branch history entry
struct BranchInfo
{
int pathHist;
int ptGhist;
int hitBank;
int hitBankIndex;
int altBank;
int altBankIndex;
int bimodalIndex;
bool tagePred;
bool altTaken;
bool condBranch;
bool longestMatchPred;
bool pseudoNewAlloc;
Addr branchPC;
// Pointer to dynamically allocated storage
// to save table indices and folded histories.
// To do one call to new instead of five.
int *storage;
// Pointers to actual saved array within the dynamically
// allocated storage.
int *tableIndices;
int *tableTags;
int *ci;
int *ct0;
int *ct1;
// for stats purposes
unsigned provider;
BranchInfo(const TAGEBase &tage)
: pathHist(0), ptGhist(0),
hitBank(0), hitBankIndex(0),
altBank(0), altBankIndex(0),
bimodalIndex(0),
tagePred(false), altTaken(false),
condBranch(false), longestMatchPred(false),
pseudoNewAlloc(false), branchPC(0),
provider(-1)
{
int sz = tage.nHistoryTables + 1;
storage = new int [sz * 5];
tableIndices = storage;
tableTags = storage + sz;
ci = tableTags + sz;
ct0 = ci + sz;
ct1 = ct0 + sz;
}
virtual ~BranchInfo()
{
delete[] storage;
}
};
virtual BranchInfo *makeBranchInfo();
/**
* Computes the index used to access the
* bimodal table.
* @param pc_in The unshifted branch PC.
*/
virtual int bindex(Addr pc_in) const;
/**
* Computes the index used to access a
* partially tagged table.
* @param tid The thread ID used to select the
* global histories to use.
* @param pc The unshifted branch PC.
* @param bank The partially tagged table to access.
*/
virtual int gindex(ThreadID tid, Addr pc, int bank) const;
/**
* Utility function to shuffle the path history
* depending on which tagged table we are accessing.
* @param phist The path history.
* @param size Number of path history bits to use.
* @param bank The partially tagged table to access.
*/
virtual int F(int phist, int size, int bank) const;
/**
* Computes the partial tag of a tagged table.
* @param tid the thread ID used to select the
* global histories to use.
* @param pc The unshifted branch PC.
* @param bank The partially tagged table to access.
*/
virtual uint16_t gtag(ThreadID tid, Addr pc, int bank) const;
/**
* Updates a direction counter based on the actual
* branch outcome.
* @param ctr Reference to counter to update.
* @param taken Actual branch outcome.
* @param nbits Counter width.
*/
template<typename T>
static void ctrUpdate(T & ctr, bool taken, int nbits);
/**
* Updates an unsigned counter based on up/down parameter
* @param ctr Reference to counter to update.
* @param up Boolean indicating if the counter is incremented/decremented
* If true it is incremented, if false it is decremented
* @param nbits Counter width.
*/
static void unsignedCtrUpdate(uint8_t & ctr, bool up, unsigned nbits);
/**
* Get a branch prediction from the bimodal
* predictor.
* @param pc The unshifted branch PC.
* @param bi Pointer to information on the
* prediction.
*/
virtual bool getBimodePred(Addr pc, BranchInfo* bi) const;
/**
* Updates the bimodal predictor.
* @param pc The unshifted branch PC.
* @param taken The actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
void baseUpdate(Addr pc, bool taken, BranchInfo* bi);
/**
* (Speculatively) updates the global branch history.
* @param h Reference to pointer to global branch history.
* @param dir (Predicted) outcome to update the histories
* with.
* @param tab
* @param PT Reference to path history.
*/
void updateGHist(uint8_t * &h, bool dir, uint8_t * tab, int &PT);
/**
* Update TAGE. Called at execute to repair histories on a misprediction
* and at commit to update the tables.
* @param tid The thread ID to select the global
* histories to use.
* @param branch_pc The unshifted branch PC.
* @param taken Actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
void update(ThreadID tid, Addr branch_pc, bool taken, BranchInfo* bi);
/**
* (Speculatively) updates global histories (path and direction).
* Also recomputes compressed (folded) histories based on the
* branch direction.
* @param tid The thread ID to select the histories
* to update.
* @param branch_pc The unshifted branch PC.
* @param taken (Predicted) branch direction.
* @param b Wrapping pointer to BranchInfo (to allow
* storing derived class prediction information in the
* base class).
*/
virtual void updateHistories(
ThreadID tid, Addr branch_pc, bool taken, BranchInfo* b,
bool speculative,
const StaticInstPtr & inst = StaticInst::nullStaticInstPtr,
Addr target = MaxAddr);
/**
* Restores speculatively updated path and direction histories.
* Also recomputes compressed (folded) histories based on the
* correct branch outcome.
* This version of squash() is called once on a branch misprediction.
* @param tid The Thread ID to select the histories to rollback.
* @param taken The correct branch outcome.
* @param bp_history Wrapping pointer to BranchInfo (to allow
* storing derived class prediction information in the
* base class).
* @param target The correct branch target
* @post bp_history points to valid memory.
*/
virtual void squash(
ThreadID tid, bool taken, BranchInfo *bi, Addr target);
/**
* Update TAGE for conditional branches.
* @param branch_pc The unshifted branch PC.
* @param taken Actual branch outcome.
* @param bi Pointer to information on the prediction
* recorded at prediction time.
* @nrand Random int number from 0 to 3
* @param corrTarget The correct branch target
*/
virtual void condBranchUpdate(
ThreadID tid, Addr branch_pc, bool taken, BranchInfo* bi,
int nrand, Addr corrTarget);
/**
* TAGE prediction called from TAGE::predict
* @param tid The thread ID to select the global
* histories to use.
* @param branch_pc The unshifted branch PC.
* @param cond_branch True if the branch is conditional.
* @param bi Pointer to the BranchInfo
*/
bool tagePredict(
ThreadID tid, Addr branch_pc, bool cond_branch, BranchInfo* bi);
/**
* Update the stats
* @param taken Actual branch outcome
* @param bi Pointer to information on the prediction
* recorded at prediction time.
*/
virtual void updateStats(bool taken, BranchInfo* bi);
/**
* Instantiates the TAGE table entries
*/
virtual void buildTageTables();
/**
* Calculates the history lengths
* and some other paramters in derived classes
*/
virtual void calculateParameters();
/**
* On a prediction, calculates the TAGE indices and tags for
* all the different history lengths
*/
virtual void calculateIndicesAndTags(
ThreadID tid, Addr branch_pc, BranchInfo* bi);
/**
* Calculation of the index for useAltPredForNewlyAllocated
* On this base TAGE implementation it is always 0
*/
virtual unsigned getUseAltIdx(BranchInfo* bi);
/**
* Extra calculation to tell whether TAGE allocaitons may happen or not
* on an update
* For this base TAGE implementation it does nothing
*/
virtual void adjustAlloc(bool & alloc, bool taken);
/**
* Handles Allocation and U bits reset on an update
*/
virtual void handleAllocAndUReset(
bool alloc, bool taken, BranchInfo* bi, int nrand);
/**
* Handles the U bits reset
*/
virtual void handleUReset();
/**
* Handles the update of the TAGE entries
*/
virtual void handleTAGEUpdate(
Addr branch_pc, bool taken, BranchInfo* bi);
/**
* Algorithm for resetting a single U counter
*/
virtual void resetUctr(uint8_t & u);
/**
* Extra steps for calculating altTaken
* For this base TAGE class it does nothing
*/
virtual void extraAltCalc(BranchInfo* bi);
/**
* Algorithm for returning a random number
* This base TAGE class just uses random_mt, but some derived classes
* may want to use a more realistic implementation or force some values
*/
static int getRandom();
void btbUpdate(ThreadID tid, Addr branch_addr, BranchInfo* &bi);
unsigned getGHR(ThreadID tid, BranchInfo *bi) const;
int8_t getCtr(int hitBank, int hitBankIndex) const;
unsigned getTageCtrBits() const;
int getPathHist(ThreadID tid) const;
bool isSpeculativeUpdateEnabled() const;
protected:
const unsigned logRatioBiModalHystEntries;
const unsigned nHistoryTables;
const unsigned tagTableCounterBits;
const unsigned tagTableUBits;
const unsigned histBufferSize;
const unsigned minHist;
const unsigned maxHist;
const unsigned pathHistBits;
std::vector<unsigned> tagTableTagWidths;
std::vector<int> logTagTableSizes;
std::vector<bool> btablePrediction;
std::vector<bool> btableHysteresis;
TageEntry **gtable;
// Keep per-thread histories to
// support SMT.
struct ThreadHistory {
// Speculative path history
// (LSB of branch address)
int pathHist;
// Speculative branch direction
// history (circular buffer)
// @TODO Convert to std::vector<bool>
uint8_t *globalHistory;
// Pointer to most recent branch outcome
uint8_t* gHist;
// Index to most recent branch outcome
int ptGhist;
// Speculative folded histories.
FoldedHistory *computeIndices;
FoldedHistory *computeTags[2];
};
std::vector<ThreadHistory> threadHistory;
/**
* Initialization of the folded histories
*/
virtual void initFoldedHistories(ThreadHistory & history);
int *histLengths;
int *tableIndices;
int *tableTags;
std::vector<int8_t> useAltPredForNewlyAllocated;
int64_t tCounter;
uint64_t logUResetPeriod;
unsigned numUseAltOnNa;
unsigned useAltOnNaBits;
unsigned maxNumAlloc;
// Tells which tables are active
// (for the base TAGE implementation all are active)
// Some other classes use this for handling associativity
std::vector<bool> noSkip;
const bool speculativeHistUpdate;
const unsigned instShiftAmt;
// stats
Stats::Scalar tageLongestMatchProviderCorrect;
Stats::Scalar tageAltMatchProviderCorrect;
Stats::Scalar bimodalAltMatchProviderCorrect;
Stats::Scalar tageBimodalProviderCorrect;
Stats::Scalar tageLongestMatchProviderWrong;
Stats::Scalar tageAltMatchProviderWrong;
Stats::Scalar bimodalAltMatchProviderWrong;
Stats::Scalar tageBimodalProviderWrong;
Stats::Scalar tageAltMatchProviderWouldHaveHit;
Stats::Scalar tageLongestMatchProviderWouldHaveHit;
Stats::Vector tageLongestMatchProvider;
Stats::Vector tageAltMatchProvider;
};
#endif // __CPU_PRED_TAGE_BASE

3
src/cpu/pred/tournament.cc
View File

@@ -266,7 +266,8 @@ TournamentBP::uncondBranch(ThreadID tid, Addr pc, void * &bp_history)
void
TournamentBP::update(ThreadID tid, Addr branch_addr, bool taken,
void *bp_history, bool squashed)
void *bp_history, bool squashed,
const StaticInstPtr & inst, Addr corrTarget)
{
assert(bp_history);

5
src/cpu/pred/tournament.hh
View File

@@ -101,9 +101,12 @@ class TournamentBP : public BPredUnit
* when the branch was predicted.
* @param squashed is set when this function is called during a squash
* operation.
* @param inst Static instruction information
* @param corrTarget Resolved target of the branch (only needed if
* squashed)
*/
void update(ThreadID tid, Addr branch_addr, bool taken, void *bp_history,
bool squashed);
bool squashed, const StaticInstPtr & inst, Addr corrTarget);
/**
* Restores the global branch history on a squash.