ScheduleDAGInstrs.cpp [plain text]
#define DEBUG_TYPE "misched"
#include "llvm/CodeGen/ScheduleDAGInstrs.h"
#include "llvm/ADT/MapVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/LiveIntervalAnalysis.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/RegisterPressure.h"
#include "llvm/CodeGen/ScheduleDFS.h"
#include "llvm/IR/Operator.h"
#include "llvm/MC/MCInstrItineraries.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetRegisterInfo.h"
#include "llvm/Target/TargetSubtargetInfo.h"
using namespace llvm;
static cl::opt<bool> EnableAASchedMI("enable-aa-sched-mi", cl::Hidden,
cl::ZeroOrMore, cl::init(false),
cl::desc("Enable use of AA during MI GAD construction"));
ScheduleDAGInstrs::ScheduleDAGInstrs(MachineFunction &mf,
const MachineLoopInfo &mli,
const MachineDominatorTree &mdt,
bool IsPostRAFlag,
LiveIntervals *lis)
: ScheduleDAG(mf), MLI(mli), MDT(mdt), MFI(mf.getFrameInfo()), LIS(lis),
IsPostRA(IsPostRAFlag), CanHandleTerminators(false), FirstDbgValue(0) {
assert((IsPostRA || LIS) && "PreRA scheduling requires LiveIntervals");
DbgValues.clear();
assert(!(IsPostRA && MRI.getNumVirtRegs()) &&
"Virtual registers must be removed prior to PostRA scheduling");
const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
SchedModel.init(*ST.getSchedModel(), &ST, TII);
}
static const Value *getUnderlyingObjectFromInt(const Value *V) {
do {
if (const Operator *U = dyn_cast<Operator>(V)) {
if (U->getOpcode() == Instruction::PtrToInt)
return U->getOperand(0);
if (U->getOpcode() != Instruction::Add ||
(!isa<ConstantInt>(U->getOperand(1)) &&
Operator::getOpcode(U->getOperand(1)) != Instruction::Mul &&
!isa<PHINode>(U->getOperand(1))))
return V;
V = U->getOperand(0);
} else {
return V;
}
assert(V->getType()->isIntegerTy() && "Unexpected operand type!");
} while (1);
}
static void getUnderlyingObjects(const Value *V,
SmallVectorImpl<Value *> &Objects) {
SmallPtrSet<const Value*, 16> Visited;
SmallVector<const Value *, 4> Working(1, V);
do {
V = Working.pop_back_val();
SmallVector<Value *, 4> Objs;
GetUnderlyingObjects(const_cast<Value *>(V), Objs);
for (SmallVector<Value *, 4>::iterator I = Objs.begin(), IE = Objs.end();
I != IE; ++I) {
V = *I;
if (!Visited.insert(V))
continue;
if (Operator::getOpcode(V) == Instruction::IntToPtr) {
const Value *O =
getUnderlyingObjectFromInt(cast<User>(V)->getOperand(0));
if (O->getType()->isPointerTy()) {
Working.push_back(O);
continue;
}
}
Objects.push_back(const_cast<Value *>(V));
}
} while (!Working.empty());
}
static void getUnderlyingObjectsForInstr(const MachineInstr *MI,
const MachineFrameInfo *MFI,
SmallVectorImpl<std::pair<const Value *, bool> > &Objects) {
if (!MI->hasOneMemOperand() ||
!(*MI->memoperands_begin())->getValue() ||
(*MI->memoperands_begin())->isVolatile())
return;
const Value *V = (*MI->memoperands_begin())->getValue();
if (!V)
return;
SmallVector<Value *, 4> Objs;
getUnderlyingObjects(V, Objs);
for (SmallVector<Value *, 4>::iterator I = Objs.begin(), IE = Objs.end();
I != IE; ++I) {
bool MayAlias = true;
V = *I;
if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V)) {
if (PSV->isAliased(MFI)) {
Objects.clear();
return;
}
MayAlias = PSV->mayAlias(MFI);
} else if (!isIdentifiedObject(V)) {
Objects.clear();
return;
}
Objects.push_back(std::make_pair(V, MayAlias));
}
}
void ScheduleDAGInstrs::startBlock(MachineBasicBlock *bb) {
BB = bb;
}
void ScheduleDAGInstrs::finishBlock() {
BB = 0;
}
void ScheduleDAGInstrs::enterRegion(MachineBasicBlock *bb,
MachineBasicBlock::iterator begin,
MachineBasicBlock::iterator end,
unsigned endcount) {
assert(bb == BB && "startBlock should set BB");
RegionBegin = begin;
RegionEnd = end;
EndIndex = endcount;
MISUnitMap.clear();
ScheduleDAG::clearDAG();
}
void ScheduleDAGInstrs::exitRegion() {
}
void ScheduleDAGInstrs::addSchedBarrierDeps() {
MachineInstr *ExitMI = RegionEnd != BB->end() ? &*RegionEnd : 0;
ExitSU.setInstr(ExitMI);
bool AllDepKnown = ExitMI &&
(ExitMI->isCall() || ExitMI->isBarrier());
if (ExitMI && AllDepKnown) {
for (unsigned i = 0, e = ExitMI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = ExitMI->getOperand(i);
if (!MO.isReg() || MO.isDef()) continue;
unsigned Reg = MO.getReg();
if (Reg == 0) continue;
if (TRI->isPhysicalRegister(Reg))
Uses.insert(PhysRegSUOper(&ExitSU, -1, Reg));
else {
assert(!IsPostRA && "Virtual register encountered after regalloc.");
if (MO.readsReg()) addVRegUseDeps(&ExitSU, i);
}
}
} else {
assert(Uses.empty() && "Uses in set before adding deps?");
for (MachineBasicBlock::succ_iterator SI = BB->succ_begin(),
SE = BB->succ_end(); SI != SE; ++SI)
for (MachineBasicBlock::livein_iterator I = (*SI)->livein_begin(),
E = (*SI)->livein_end(); I != E; ++I) {
unsigned Reg = *I;
if (!Uses.contains(Reg))
Uses.insert(PhysRegSUOper(&ExitSU, -1, Reg));
}
}
}
void ScheduleDAGInstrs::addPhysRegDataDeps(SUnit *SU, unsigned OperIdx) {
const MachineOperand &MO = SU->getInstr()->getOperand(OperIdx);
assert(MO.isDef() && "expect physreg def");
const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
for (MCRegAliasIterator Alias(MO.getReg(), TRI, true);
Alias.isValid(); ++Alias) {
if (!Uses.contains(*Alias))
continue;
for (Reg2SUnitsMap::iterator I = Uses.find(*Alias); I != Uses.end(); ++I) {
SUnit *UseSU = I->SU;
if (UseSU == SU)
continue;
int UseOp = I->OpIdx;
MachineInstr *RegUse = 0;
SDep Dep;
if (UseOp < 0)
Dep = SDep(SU, SDep::Artificial);
else {
SU->hasPhysRegDefs = true;
Dep = SDep(SU, SDep::Data, *Alias);
RegUse = UseSU->getInstr();
Dep.setMinLatency(
SchedModel.computeOperandLatency(SU->getInstr(), OperIdx,
RegUse, UseOp, true));
}
Dep.setLatency(
SchedModel.computeOperandLatency(SU->getInstr(), OperIdx,
RegUse, UseOp, false));
ST.adjustSchedDependency(SU, UseSU, Dep);
UseSU->addPred(Dep);
}
}
}
void ScheduleDAGInstrs::addPhysRegDeps(SUnit *SU, unsigned OperIdx) {
const MachineInstr *MI = SU->getInstr();
const MachineOperand &MO = MI->getOperand(OperIdx);
SDep::Kind Kind = MO.isUse() ? SDep::Anti : SDep::Output;
for (MCRegAliasIterator Alias(MO.getReg(), TRI, true);
Alias.isValid(); ++Alias) {
if (!Defs.contains(*Alias))
continue;
for (Reg2SUnitsMap::iterator I = Defs.find(*Alias); I != Defs.end(); ++I) {
SUnit *DefSU = I->SU;
if (DefSU == &ExitSU)
continue;
if (DefSU != SU &&
(Kind != SDep::Output || !MO.isDead() ||
!DefSU->getInstr()->registerDefIsDead(*Alias))) {
if (Kind == SDep::Anti)
DefSU->addPred(SDep(SU, Kind, *Alias));
else {
SDep Dep(SU, Kind, *Alias);
unsigned OutLatency =
SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr());
Dep.setMinLatency(OutLatency);
Dep.setLatency(OutLatency);
DefSU->addPred(Dep);
}
}
}
}
if (!MO.isDef()) {
SU->hasPhysRegUses = true;
Uses.insert(PhysRegSUOper(SU, OperIdx, MO.getReg()));
}
else {
addPhysRegDataDeps(SU, OperIdx);
unsigned Reg = MO.getReg();
if (Uses.contains(Reg))
Uses.eraseAll(Reg);
if (!MO.isDead()) {
Defs.eraseAll(Reg);
} else if (SU->isCall) {
Reg2SUnitsMap::RangePair P = Defs.equal_range(Reg);
Reg2SUnitsMap::iterator B = P.first;
Reg2SUnitsMap::iterator I = P.second;
for (bool isBegin = I == B; !isBegin; ) {
isBegin = (--I) == B;
if (!I->SU->isCall)
break;
I = Defs.erase(I);
}
}
Defs.insert(PhysRegSUOper(SU, OperIdx, Reg));
}
}
void ScheduleDAGInstrs::addVRegDefDeps(SUnit *SU, unsigned OperIdx) {
const MachineInstr *MI = SU->getInstr();
unsigned Reg = MI->getOperand(OperIdx).getReg();
if (MRI.hasOneDef(Reg))
return;
VReg2SUnitMap::iterator DefI = VRegDefs.find(Reg);
if (DefI == VRegDefs.end())
VRegDefs.insert(VReg2SUnit(Reg, SU));
else {
SUnit *DefSU = DefI->SU;
if (DefSU != SU && DefSU != &ExitSU) {
SDep Dep(SU, SDep::Output, Reg);
unsigned OutLatency =
SchedModel.computeOutputLatency(MI, OperIdx, DefSU->getInstr());
Dep.setMinLatency(OutLatency);
Dep.setLatency(OutLatency);
DefSU->addPred(Dep);
}
DefI->SU = SU;
}
}
void ScheduleDAGInstrs::addVRegUseDeps(SUnit *SU, unsigned OperIdx) {
MachineInstr *MI = SU->getInstr();
unsigned Reg = MI->getOperand(OperIdx).getReg();
assert(LIS && "vreg dependencies requires LiveIntervals");
LiveRangeQuery LRQ(LIS->getInterval(Reg), LIS->getInstructionIndex(MI));
VNInfo *VNI = LRQ.valueIn();
assert(VNI && "No value to read by operand");
MachineInstr *Def = LIS->getInstructionFromIndex(VNI->def);
if (Def) {
SUnit *DefSU = getSUnit(Def);
if (DefSU) {
SDep dep(DefSU, SDep::Data, Reg);
int DefOp = Def->findRegisterDefOperandIdx(Reg);
dep.setLatency(
SchedModel.computeOperandLatency(Def, DefOp, MI, OperIdx, false));
dep.setMinLatency(
SchedModel.computeOperandLatency(Def, DefOp, MI, OperIdx, true));
const TargetSubtargetInfo &ST = TM.getSubtarget<TargetSubtargetInfo>();
ST.adjustSchedDependency(DefSU, SU, const_cast<SDep &>(dep));
SU->addPred(dep);
}
}
VReg2SUnitMap::iterator DefI = VRegDefs.find(Reg);
if (DefI != VRegDefs.end() && DefI->SU != SU)
DefI->SU->addPred(SDep(SU, SDep::Anti, Reg));
}
static inline bool isGlobalMemoryObject(AliasAnalysis *AA, MachineInstr *MI) {
if (MI->isCall() || MI->hasUnmodeledSideEffects() ||
(MI->hasOrderedMemoryRef() &&
(!MI->mayLoad() || !MI->isInvariantLoad(AA))))
return true;
return false;
}
static inline bool isUnsafeMemoryObject(MachineInstr *MI,
const MachineFrameInfo *MFI) {
if (!MI || MI->memoperands_empty())
return true;
if ((*MI->memoperands_begin())->isVolatile() ||
MI->hasUnmodeledSideEffects())
return true;
const Value *V = (*MI->memoperands_begin())->getValue();
if (!V)
return true;
SmallVector<Value *, 4> Objs;
getUnderlyingObjects(V, Objs);
for (SmallVector<Value *, 4>::iterator I = Objs.begin(),
IE = Objs.end(); I != IE; ++I) {
V = *I;
if (const PseudoSourceValue *PSV = dyn_cast<PseudoSourceValue>(V)) {
if (PSV->isAliased(MFI))
return true;
}
if (!isIdentifiedObject(V))
return true;
}
return false;
}
static bool MIsNeedChainEdge(AliasAnalysis *AA, const MachineFrameInfo *MFI,
MachineInstr *MIa,
MachineInstr *MIb) {
if (MIa == MIb)
return false;
if (isUnsafeMemoryObject(MIa, MFI) || isUnsafeMemoryObject(MIb, MFI))
return true;
if (!MIa->mayStore() && !MIb->mayStore())
return false;
if (!AA)
return true;
MachineMemOperand *MMOa = *MIa->memoperands_begin();
MachineMemOperand *MMOb = *MIb->memoperands_begin();
if (!MIa->hasOneMemOperand() || !MIb->hasOneMemOperand())
llvm_unreachable("Multiple memory operands.");
assert ((MMOa->getOffset() >= 0) && "Negative MachineMemOperand offset");
assert ((MMOb->getOffset() >= 0) && "Negative MachineMemOperand offset");
int64_t MinOffset = std::min(MMOa->getOffset(), MMOb->getOffset());
int64_t Overlapa = MMOa->getSize() + MMOa->getOffset() - MinOffset;
int64_t Overlapb = MMOb->getSize() + MMOb->getOffset() - MinOffset;
AliasAnalysis::AliasResult AAResult = AA->alias(
AliasAnalysis::Location(MMOa->getValue(), Overlapa,
MMOa->getTBAAInfo()),
AliasAnalysis::Location(MMOb->getValue(), Overlapb,
MMOb->getTBAAInfo()));
return (AAResult != AliasAnalysis::NoAlias);
}
static unsigned
iterateChainSucc(AliasAnalysis *AA, const MachineFrameInfo *MFI,
SUnit *SUa, SUnit *SUb, SUnit *ExitSU, unsigned *Depth,
SmallPtrSet<const SUnit*, 16> &Visited) {
if (!SUa || !SUb || SUb == ExitSU)
return *Depth;
if (!Visited.insert(SUb))
return *Depth;
if (SUa->isSucc(SUb) ||
isGlobalMemoryObject(AA, SUb->getInstr()))
return *Depth;
if (*Depth > 200 ||
MIsNeedChainEdge(AA, MFI, SUa->getInstr(), SUb->getInstr())) {
SUb->addPred(SDep(SUa, SDep::MayAliasMem));
return *Depth;
}
(*Depth)++;
for (SUnit::const_succ_iterator I = SUb->Succs.begin(), E = SUb->Succs.end();
I != E; ++I)
if (I->isCtrl())
iterateChainSucc (AA, MFI, SUa, I->getSUnit(), ExitSU, Depth, Visited);
return *Depth;
}
static void adjustChainDeps(AliasAnalysis *AA, const MachineFrameInfo *MFI,
SUnit *SU, SUnit *ExitSU, std::set<SUnit *> &CheckList,
unsigned LatencyToLoad) {
if (!SU)
return;
SmallPtrSet<const SUnit*, 16> Visited;
unsigned Depth = 0;
for (std::set<SUnit *>::iterator I = CheckList.begin(), IE = CheckList.end();
I != IE; ++I) {
if (SU == *I)
continue;
if (MIsNeedChainEdge(AA, MFI, SU->getInstr(), (*I)->getInstr())) {
SDep Dep(SU, SDep::MayAliasMem);
Dep.setLatency(((*I)->getInstr()->mayLoad()) ? LatencyToLoad : 0);
(*I)->addPred(Dep);
}
for (SUnit::const_succ_iterator J = (*I)->Succs.begin(),
JE = (*I)->Succs.end(); J != JE; ++J)
if (J->isCtrl())
iterateChainSucc (AA, MFI, SU, J->getSUnit(),
ExitSU, &Depth, Visited);
}
}
static inline
void addChainDependency (AliasAnalysis *AA, const MachineFrameInfo *MFI,
SUnit *SUa, SUnit *SUb,
std::set<SUnit *> &RejectList,
unsigned TrueMemOrderLatency = 0,
bool isNormalMemory = false) {
if (!EnableAASchedMI ||
MIsNeedChainEdge(AA, MFI, SUa->getInstr(), SUb->getInstr())) {
SDep Dep(SUa, isNormalMemory ? SDep::MayAliasMem : SDep::Barrier);
Dep.setLatency(TrueMemOrderLatency);
SUb->addPred(Dep);
}
else {
RejectList.insert(SUb);
DEBUG(dbgs() << "\tReject chain dep between SU("
<< SUa->NodeNum << ") and SU("
<< SUb->NodeNum << ")\n");
}
}
void ScheduleDAGInstrs::initSUnits() {
SUnits.reserve(BB->size());
for (MachineBasicBlock::iterator I = RegionBegin; I != RegionEnd; ++I) {
MachineInstr *MI = I;
if (MI->isDebugValue())
continue;
SUnit *SU = newSUnit(MI);
MISUnitMap[MI] = SU;
SU->isCall = MI->isCall();
SU->isCommutable = MI->isCommutable();
SU->Latency = SchedModel.computeInstrLatency(SU->getInstr());
}
}
void ScheduleDAGInstrs::buildSchedGraph(AliasAnalysis *AA,
RegPressureTracker *RPTracker) {
initSUnits();
SUnit *BarrierChain = 0, *AliasChain = 0;
MapVector<const Value *, SUnit *> AliasMemDefs, NonAliasMemDefs;
MapVector<const Value *, std::vector<SUnit *> > AliasMemUses, NonAliasMemUses;
std::set<SUnit*> RejectMemNodes;
DbgValues.clear();
FirstDbgValue = NULL;
assert(Defs.empty() && Uses.empty() &&
"Only BuildGraph should update Defs/Uses");
Defs.setUniverse(TRI->getNumRegs());
Uses.setUniverse(TRI->getNumRegs());
assert(VRegDefs.empty() && "Only BuildSchedGraph may access VRegDefs");
VRegDefs.setUniverse(MRI.getNumVirtRegs());
addSchedBarrierDeps();
MachineInstr *DbgMI = NULL;
for (MachineBasicBlock::iterator MII = RegionEnd, MIE = RegionBegin;
MII != MIE; --MII) {
MachineInstr *MI = prior(MII);
if (MI && DbgMI) {
DbgValues.push_back(std::make_pair(DbgMI, MI));
DbgMI = NULL;
}
if (MI->isDebugValue()) {
DbgMI = MI;
continue;
}
if (RPTracker) {
RPTracker->recede();
assert(RPTracker->getPos() == prior(MII) && "RPTracker can't find MI");
}
assert((!MI->isTerminator() || CanHandleTerminators) && !MI->isLabel() &&
"Cannot schedule terminators or labels!");
SUnit *SU = MISUnitMap[MI];
assert(SU && "No SUnit mapped to this MI");
bool HasVRegDef = false;
for (unsigned j = 0, n = MI->getNumOperands(); j != n; ++j) {
const MachineOperand &MO = MI->getOperand(j);
if (!MO.isReg()) continue;
unsigned Reg = MO.getReg();
if (Reg == 0) continue;
if (TRI->isPhysicalRegister(Reg))
addPhysRegDeps(SU, j);
else {
assert(!IsPostRA && "Virtual register encountered!");
if (MO.isDef()) {
HasVRegDef = true;
addVRegDefDeps(SU, j);
}
else if (MO.readsReg()) addVRegUseDeps(SU, j);
}
}
if (SU->NumSuccs == 0 && SU->Latency > 1
&& (HasVRegDef || MI->mayLoad())) {
SDep Dep(SU, SDep::Artificial);
Dep.setLatency(SU->Latency - 1);
ExitSU.addPred(Dep);
}
unsigned TrueMemOrderLatency = MI->mayStore() ? 1 : 0;
if (isGlobalMemoryObject(AA, MI)) {
for (MapVector<const Value *, SUnit *>::iterator I =
NonAliasMemDefs.begin(), E = NonAliasMemDefs.end(); I != E; ++I) {
I->second->addPred(SDep(SU, SDep::Barrier));
}
for (MapVector<const Value *, std::vector<SUnit *> >::iterator I =
NonAliasMemUses.begin(), E = NonAliasMemUses.end(); I != E; ++I) {
for (unsigned i = 0, e = I->second.size(); i != e; ++i) {
SDep Dep(SU, SDep::Barrier);
Dep.setLatency(TrueMemOrderLatency);
I->second[i]->addPred(Dep);
}
}
if (BarrierChain)
BarrierChain->addPred(SDep(SU, SDep::Barrier));
BarrierChain = SU;
adjustChainDeps(AA, MFI, SU, &ExitSU, RejectMemNodes,
TrueMemOrderLatency);
RejectMemNodes.clear();
NonAliasMemDefs.clear();
NonAliasMemUses.clear();
new_alias_chain:
if (AliasChain) {
unsigned ChainLatency = 0;
if (AliasChain->getInstr()->mayLoad())
ChainLatency = TrueMemOrderLatency;
addChainDependency(AA, MFI, SU, AliasChain, RejectMemNodes,
ChainLatency);
}
AliasChain = SU;
for (unsigned k = 0, m = PendingLoads.size(); k != m; ++k)
addChainDependency(AA, MFI, SU, PendingLoads[k], RejectMemNodes,
TrueMemOrderLatency);
for (MapVector<const Value *, SUnit *>::iterator I = AliasMemDefs.begin(),
E = AliasMemDefs.end(); I != E; ++I)
addChainDependency(AA, MFI, SU, I->second, RejectMemNodes);
for (MapVector<const Value *, std::vector<SUnit *> >::iterator I =
AliasMemUses.begin(), E = AliasMemUses.end(); I != E; ++I) {
for (unsigned i = 0, e = I->second.size(); i != e; ++i)
addChainDependency(AA, MFI, SU, I->second[i], RejectMemNodes,
TrueMemOrderLatency);
}
adjustChainDeps(AA, MFI, SU, &ExitSU, RejectMemNodes,
TrueMemOrderLatency);
PendingLoads.clear();
AliasMemDefs.clear();
AliasMemUses.clear();
} else if (MI->mayStore()) {
SmallVector<std::pair<const Value *, bool>, 4> Objs;
getUnderlyingObjectsForInstr(MI, MFI, Objs);
if (Objs.empty()) {
goto new_alias_chain;
}
bool MayAlias = false;
for (SmallVector<std::pair<const Value *, bool>, 4>::iterator
K = Objs.begin(), KE = Objs.end(); K != KE; ++K) {
const Value *V = K->first;
bool ThisMayAlias = K->second;
if (ThisMayAlias)
MayAlias = true;
MapVector<const Value *, SUnit *>::iterator I =
((ThisMayAlias) ? AliasMemDefs.find(V) : NonAliasMemDefs.find(V));
MapVector<const Value *, SUnit *>::iterator IE =
((ThisMayAlias) ? AliasMemDefs.end() : NonAliasMemDefs.end());
if (I != IE) {
addChainDependency(AA, MFI, SU, I->second, RejectMemNodes, 0, true);
I->second = SU;
} else {
if (ThisMayAlias)
AliasMemDefs[V] = SU;
else
NonAliasMemDefs[V] = SU;
}
MapVector<const Value *, std::vector<SUnit *> >::iterator J =
((ThisMayAlias) ? AliasMemUses.find(V) : NonAliasMemUses.find(V));
MapVector<const Value *, std::vector<SUnit *> >::iterator JE =
((ThisMayAlias) ? AliasMemUses.end() : NonAliasMemUses.end());
if (J != JE) {
for (unsigned i = 0, e = J->second.size(); i != e; ++i)
addChainDependency(AA, MFI, SU, J->second[i], RejectMemNodes,
TrueMemOrderLatency, true);
J->second.clear();
}
}
if (MayAlias) {
for (unsigned k = 0, m = PendingLoads.size(); k != m; ++k)
addChainDependency(AA, MFI, SU, PendingLoads[k], RejectMemNodes,
TrueMemOrderLatency);
if (AliasChain)
addChainDependency(AA, MFI, SU, AliasChain, RejectMemNodes);
adjustChainDeps(AA, MFI, SU, &ExitSU, RejectMemNodes,
TrueMemOrderLatency);
}
if (BarrierChain)
BarrierChain->addPred(SDep(SU, SDep::Barrier));
if (!ExitSU.isPred(SU))
ExitSU.addPred(SDep(SU, SDep::Artificial));
} else if (MI->mayLoad()) {
bool MayAlias = true;
if (MI->isInvariantLoad(AA)) {
} else {
SmallVector<std::pair<const Value *, bool>, 4> Objs;
getUnderlyingObjectsForInstr(MI, MFI, Objs);
if (Objs.empty()) {
for (MapVector<const Value *, SUnit *>::iterator I =
AliasMemDefs.begin(), E = AliasMemDefs.end(); I != E; ++I)
addChainDependency(AA, MFI, SU, I->second, RejectMemNodes);
PendingLoads.push_back(SU);
MayAlias = true;
} else {
MayAlias = false;
}
for (SmallVector<std::pair<const Value *, bool>, 4>::iterator
J = Objs.begin(), JE = Objs.end(); J != JE; ++J) {
const Value *V = J->first;
bool ThisMayAlias = J->second;
if (ThisMayAlias)
MayAlias = true;
MapVector<const Value *, SUnit *>::iterator I =
((ThisMayAlias) ? AliasMemDefs.find(V) : NonAliasMemDefs.find(V));
MapVector<const Value *, SUnit *>::iterator IE =
((ThisMayAlias) ? AliasMemDefs.end() : NonAliasMemDefs.end());
if (I != IE)
addChainDependency(AA, MFI, SU, I->second, RejectMemNodes, 0, true);
if (ThisMayAlias)
AliasMemUses[V].push_back(SU);
else
NonAliasMemUses[V].push_back(SU);
}
if (MayAlias)
adjustChainDeps(AA, MFI, SU, &ExitSU, RejectMemNodes, 0);
if (MayAlias && AliasChain)
addChainDependency(AA, MFI, SU, AliasChain, RejectMemNodes);
if (BarrierChain)
BarrierChain->addPred(SDep(SU, SDep::Barrier));
}
}
}
if (DbgMI)
FirstDbgValue = DbgMI;
Defs.clear();
Uses.clear();
VRegDefs.clear();
PendingLoads.clear();
}
void ScheduleDAGInstrs::dumpNode(const SUnit *SU) const {
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
SU->getInstr()->dump();
#endif
}
std::string ScheduleDAGInstrs::getGraphNodeLabel(const SUnit *SU) const {
std::string s;
raw_string_ostream oss(s);
if (SU == &EntrySU)
oss << "<entry>";
else if (SU == &ExitSU)
oss << "<exit>";
else
SU->getInstr()->print(oss, &TM, true);
return oss.str();
}
std::string ScheduleDAGInstrs::getDAGName() const {
return "dag." + BB->getFullName();
}
namespace llvm {
class SchedDFSImpl {
SchedDFSResult &R;
IntEqClasses SubtreeClasses;
std::vector<std::pair<const SUnit*, const SUnit*> > ConnectionPairs;
struct RootData {
unsigned NodeID;
unsigned ParentNodeID; unsigned SubInstrCount;
RootData(unsigned id): NodeID(id),
ParentNodeID(SchedDFSResult::InvalidSubtreeID),
SubInstrCount(0) {}
unsigned getSparseSetIndex() const { return NodeID; }
};
SparseSet<RootData> RootSet;
public:
SchedDFSImpl(SchedDFSResult &r): R(r), SubtreeClasses(R.DFSNodeData.size()) {
RootSet.setUniverse(R.DFSNodeData.size());
}
bool isVisited(const SUnit *SU) const {
return R.DFSNodeData[SU->NodeNum].SubtreeID
!= SchedDFSResult::InvalidSubtreeID;
}
void visitPreorder(const SUnit *SU) {
R.DFSNodeData[SU->NodeNum].InstrCount =
SU->getInstr()->isTransient() ? 0 : 1;
}
void visitPostorderNode(const SUnit *SU) {
R.DFSNodeData[SU->NodeNum].SubtreeID = SU->NodeNum;
RootData RData(SU->NodeNum);
RData.SubInstrCount = SU->getInstr()->isTransient() ? 0 : 1;
unsigned InstrCount = R.DFSNodeData[SU->NodeNum].InstrCount;
for (SUnit::const_pred_iterator
PI = SU->Preds.begin(), PE = SU->Preds.end(); PI != PE; ++PI) {
if (PI->getKind() != SDep::Data)
continue;
unsigned PredNum = PI->getSUnit()->NodeNum;
if ((InstrCount - R.DFSNodeData[PredNum].InstrCount) < R.SubtreeLimit)
joinPredSubtree(*PI, SU, false);
if (R.DFSNodeData[PredNum].SubtreeID == PredNum) {
if (RootSet[PredNum].ParentNodeID == SchedDFSResult::InvalidSubtreeID)
RootSet[PredNum].ParentNodeID = SU->NodeNum;
}
else if (RootSet.count(PredNum)) {
RData.SubInstrCount += RootSet[PredNum].SubInstrCount;
RootSet.erase(PredNum);
}
}
RootSet[SU->NodeNum] = RData;
}
void visitPostorderEdge(const SDep &PredDep, const SUnit *Succ) {
R.DFSNodeData[Succ->NodeNum].InstrCount
+= R.DFSNodeData[PredDep.getSUnit()->NodeNum].InstrCount;
joinPredSubtree(PredDep, Succ);
}
void visitCrossEdge(const SDep &PredDep, const SUnit *Succ) {
ConnectionPairs.push_back(std::make_pair(PredDep.getSUnit(), Succ));
}
void finalize() {
SubtreeClasses.compress();
R.DFSTreeData.resize(SubtreeClasses.getNumClasses());
assert(SubtreeClasses.getNumClasses() == RootSet.size()
&& "number of roots should match trees");
for (SparseSet<RootData>::const_iterator
RI = RootSet.begin(), RE = RootSet.end(); RI != RE; ++RI) {
unsigned TreeID = SubtreeClasses[RI->NodeID];
if (RI->ParentNodeID != SchedDFSResult::InvalidSubtreeID)
R.DFSTreeData[TreeID].ParentTreeID = SubtreeClasses[RI->ParentNodeID];
R.DFSTreeData[TreeID].SubInstrCount = RI->SubInstrCount;
}
R.SubtreeConnections.resize(SubtreeClasses.getNumClasses());
R.SubtreeConnectLevels.resize(SubtreeClasses.getNumClasses());
DEBUG(dbgs() << R.getNumSubtrees() << " subtrees:\n");
for (unsigned Idx = 0, End = R.DFSNodeData.size(); Idx != End; ++Idx) {
R.DFSNodeData[Idx].SubtreeID = SubtreeClasses[Idx];
DEBUG(dbgs() << " SU(" << Idx << ") in tree "
<< R.DFSNodeData[Idx].SubtreeID << '\n');
}
for (std::vector<std::pair<const SUnit*, const SUnit*> >::const_iterator
I = ConnectionPairs.begin(), E = ConnectionPairs.end();
I != E; ++I) {
unsigned PredTree = SubtreeClasses[I->first->NodeNum];
unsigned SuccTree = SubtreeClasses[I->second->NodeNum];
if (PredTree == SuccTree)
continue;
unsigned Depth = I->first->getDepth();
addConnection(PredTree, SuccTree, Depth);
addConnection(SuccTree, PredTree, Depth);
}
}
protected:
bool joinPredSubtree(const SDep &PredDep, const SUnit *Succ,
bool CheckLimit = true) {
assert(PredDep.getKind() == SDep::Data && "Subtrees are for data edges");
const SUnit *PredSU = PredDep.getSUnit();
unsigned PredNum = PredSU->NodeNum;
if (R.DFSNodeData[PredNum].SubtreeID != PredNum)
return false;
unsigned NumDataSucs = 0;
for (SUnit::const_succ_iterator SI = PredSU->Succs.begin(),
SE = PredSU->Succs.end(); SI != SE; ++SI) {
if (SI->getKind() == SDep::Data) {
if (++NumDataSucs >= 4)
return false;
}
}
if (CheckLimit && R.DFSNodeData[PredNum].InstrCount > R.SubtreeLimit)
return false;
R.DFSNodeData[PredNum].SubtreeID = Succ->NodeNum;
SubtreeClasses.join(Succ->NodeNum, PredNum);
return true;
}
void addConnection(unsigned FromTree, unsigned ToTree, unsigned Depth) {
if (!Depth)
return;
do {
SmallVectorImpl<SchedDFSResult::Connection> &Connections =
R.SubtreeConnections[FromTree];
for (SmallVectorImpl<SchedDFSResult::Connection>::iterator
I = Connections.begin(), E = Connections.end(); I != E; ++I) {
if (I->TreeID == ToTree) {
I->Level = std::max(I->Level, Depth);
return;
}
}
Connections.push_back(SchedDFSResult::Connection(ToTree, Depth));
FromTree = R.DFSTreeData[FromTree].ParentTreeID;
} while (FromTree != SchedDFSResult::InvalidSubtreeID);
}
};
}
namespace {
class SchedDAGReverseDFS {
std::vector<std::pair<const SUnit*, SUnit::const_pred_iterator> > DFSStack;
public:
bool isComplete() const { return DFSStack.empty(); }
void follow(const SUnit *SU) {
DFSStack.push_back(std::make_pair(SU, SU->Preds.begin()));
}
void advance() { ++DFSStack.back().second; }
const SDep *backtrack() {
DFSStack.pop_back();
return DFSStack.empty() ? 0 : llvm::prior(DFSStack.back().second);
}
const SUnit *getCurr() const { return DFSStack.back().first; }
SUnit::const_pred_iterator getPred() const { return DFSStack.back().second; }
SUnit::const_pred_iterator getPredEnd() const {
return getCurr()->Preds.end();
}
};
}
static bool hasDataSucc(const SUnit *SU) {
for (SUnit::const_succ_iterator
SI = SU->Succs.begin(), SE = SU->Succs.end(); SI != SE; ++SI) {
if (SI->getKind() == SDep::Data && !SI->getSUnit()->isBoundaryNode())
return true;
}
return false;
}
void SchedDFSResult::compute(ArrayRef<SUnit> SUnits) {
if (!IsBottomUp)
llvm_unreachable("Top-down ILP metric is unimplemnted");
SchedDFSImpl Impl(*this);
for (ArrayRef<SUnit>::const_iterator
SI = SUnits.begin(), SE = SUnits.end(); SI != SE; ++SI) {
const SUnit *SU = &*SI;
if (Impl.isVisited(SU) || hasDataSucc(SU))
continue;
SchedDAGReverseDFS DFS;
Impl.visitPreorder(SU);
DFS.follow(SU);
for (;;) {
while (DFS.getPred() != DFS.getPredEnd()) {
const SDep &PredDep = *DFS.getPred();
DFS.advance();
if (PredDep.getKind() != SDep::Data
|| PredDep.getSUnit()->isBoundaryNode()) {
continue;
}
if (Impl.isVisited(PredDep.getSUnit())) {
Impl.visitCrossEdge(PredDep, DFS.getCurr());
continue;
}
Impl.visitPreorder(PredDep.getSUnit());
DFS.follow(PredDep.getSUnit());
}
const SUnit *Child = DFS.getCurr();
const SDep *PredDep = DFS.backtrack();
Impl.visitPostorderNode(Child);
if (PredDep)
Impl.visitPostorderEdge(*PredDep, DFS.getCurr());
if (DFS.isComplete())
break;
}
}
Impl.finalize();
}
void SchedDFSResult::scheduleTree(unsigned SubtreeID) {
for (SmallVectorImpl<Connection>::const_iterator
I = SubtreeConnections[SubtreeID].begin(),
E = SubtreeConnections[SubtreeID].end(); I != E; ++I) {
SubtreeConnectLevels[I->TreeID] =
std::max(SubtreeConnectLevels[I->TreeID], I->Level);
DEBUG(dbgs() << " Tree: " << I->TreeID
<< " @" << SubtreeConnectLevels[I->TreeID] << '\n');
}
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
void ILPValue::print(raw_ostream &OS) const {
OS << InstrCount << " / " << Length << " = ";
if (!Length)
OS << "BADILP";
else
OS << format("%g", ((double)InstrCount / Length));
}
void ILPValue::dump() const {
dbgs() << *this << '\n';
}
namespace llvm {
raw_ostream &operator<<(raw_ostream &OS, const ILPValue &Val) {
Val.print(OS);
return OS;
}
} #endif // !NDEBUG || LLVM_ENABLE_DUMP