bool LiveRangeCalc::findReachingDefs(LiveRange &LR, MachineBasicBlock &KillMBB,
SlotIndex Kill, unsigned PhysReg) {
unsigned KillMBBNum = KillMBB.getNumber();
// Block numbers where LR should be live-in.
SmallVector<unsigned, 16> WorkList(1, KillMBBNum);
// Remember if we have seen more than one value.
bool UniqueVNI = true;
VNInfo *TheVNI = 0;
// Using Seen as a visited set, perform a BFS for all reaching defs.
for (unsigned i = 0; i != WorkList.size(); ++i) {
MachineBasicBlock *MBB = MF->getBlockNumbered(WorkList[i]);
#ifndef NDEBUG
if (MBB->pred_empty()) {
MBB->getParent()->verify();
llvm_unreachable("Use not jointly dominated by defs.");
}
if (TargetRegisterInfo::isPhysicalRegister(PhysReg) &&
!MBB->isLiveIn(PhysReg)) {
MBB->getParent()->verify();
errs() << "The register needs to be live in to BB#" << MBB->getNumber()
<< ", but is missing from the live-in list.\n";
llvm_unreachable("Invalid global physical register");
}
#endif
for (MachineBasicBlock::pred_iterator PI = MBB->pred_begin(),
PE = MBB->pred_end(); PI != PE; ++PI) {
MachineBasicBlock *Pred = *PI;
// Is this a known live-out block?
if (Seen.test(Pred->getNumber())) {
if (VNInfo *VNI = LiveOut[Pred].first) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
}
continue;
}
SlotIndex Start, End;
std::tie(Start, End) = Indexes->getMBBRange(Pred);
// First time we see Pred. Try to determine the live-out value, but set
// it as null if Pred is live-through with an unknown value.
VNInfo *VNI = LR.extendInBlock(Start, End);
setLiveOutValue(Pred, VNI);
if (VNI) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
continue;
}
// No, we need a live-in value for Pred as well
if (Pred != &KillMBB)
WorkList.push_back(Pred->getNumber());
else
// Loopback to KillMBB, so value is really live through.
Kill = SlotIndex();
}
}
LiveIn.clear();
// Both updateSSA() and LiveRangeUpdater benefit from ordered blocks, but
// neither require it. Skip the sorting overhead for small updates.
if (WorkList.size() > 4)
array_pod_sort(WorkList.begin(), WorkList.end());
// If a unique reaching def was found, blit in the live ranges immediately.
if (UniqueVNI) {
LiveRangeUpdater Updater(&LR);
for (SmallVectorImpl<unsigned>::const_iterator I = WorkList.begin(),
E = WorkList.end(); I != E; ++I) {
SlotIndex Start, End;
std::tie(Start, End) = Indexes->getMBBRange(*I);
// Trim the live range in KillMBB.
if (*I == KillMBBNum && Kill.isValid())
End = Kill;
else
LiveOut[MF->getBlockNumbered(*I)] =
LiveOutPair(TheVNI, (MachineDomTreeNode *)0);
Updater.add(Start, End, TheVNI);
}
return true;
}
// Multiple values were found, so transfer the work list to the LiveIn array
// where UpdateSSA will use it as a work list.
LiveIn.reserve(WorkList.size());
for (SmallVectorImpl<unsigned>::const_iterator
I = WorkList.begin(), E = WorkList.end(); I != E; ++I) {
MachineBasicBlock *MBB = MF->getBlockNumbered(*I);
addLiveInBlock(LR, DomTree->getNode(MBB));
if (MBB == &KillMBB)
LiveIn.back().Kill = Kill;
}
return false;
}
MachineInstr *TargetInstrInfo::foldMemoryOperand(MachineInstr &MI,
ArrayRef<unsigned> Ops, int FI,
LiveIntervals *LIS) const {
auto Flags = MachineMemOperand::MONone;
for (unsigned OpIdx : Ops)
Flags |= MI.getOperand(OpIdx).isDef() ? MachineMemOperand::MOStore
: MachineMemOperand::MOLoad;
MachineBasicBlock *MBB = MI.getParent();
assert(MBB && "foldMemoryOperand needs an inserted instruction");
MachineFunction &MF = *MBB->getParent();
// If we're not folding a load into a subreg, the size of the load is the
// size of the spill slot. But if we are, we need to figure out what the
// actual load size is.
int64_t MemSize = 0;
const MachineFrameInfo &MFI = MF.getFrameInfo();
const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
if (Flags & MachineMemOperand::MOStore) {
MemSize = MFI.getObjectSize(FI);
} else {
for (unsigned OpIdx : Ops) {
int64_t OpSize = MFI.getObjectSize(FI);
if (auto SubReg = MI.getOperand(OpIdx).getSubReg()) {
unsigned SubRegSize = TRI->getSubRegIdxSize(SubReg);
if (SubRegSize > 0 && !(SubRegSize % 8))
OpSize = SubRegSize / 8;
}
MemSize = std::max(MemSize, OpSize);
}
}
assert(MemSize && "Did not expect a zero-sized stack slot");
MachineInstr *NewMI = nullptr;
if (MI.getOpcode() == TargetOpcode::STACKMAP ||
MI.getOpcode() == TargetOpcode::PATCHPOINT ||
MI.getOpcode() == TargetOpcode::STATEPOINT) {
// Fold stackmap/patchpoint.
NewMI = foldPatchpoint(MF, MI, Ops, FI, *this);
if (NewMI)
MBB->insert(MI, NewMI);
} else {
// Ask the target to do the actual folding.
NewMI = foldMemoryOperandImpl(MF, MI, Ops, MI, FI, LIS);
}
if (NewMI) {
NewMI->setMemRefs(MF, MI.memoperands());
// Add a memory operand, foldMemoryOperandImpl doesn't do that.
assert((!(Flags & MachineMemOperand::MOStore) ||
NewMI->mayStore()) &&
"Folded a def to a non-store!");
assert((!(Flags & MachineMemOperand::MOLoad) ||
NewMI->mayLoad()) &&
"Folded a use to a non-load!");
assert(MFI.getObjectOffset(FI) != -1);
MachineMemOperand *MMO = MF.getMachineMemOperand(
MachinePointerInfo::getFixedStack(MF, FI), Flags, MemSize,
MFI.getObjectAlignment(FI));
NewMI->addMemOperand(MF, MMO);
return NewMI;
}
// Straight COPY may fold as load/store.
if (!MI.isCopy() || Ops.size() != 1)
return nullptr;
const TargetRegisterClass *RC = canFoldCopy(MI, Ops[0]);
if (!RC)
return nullptr;
const MachineOperand &MO = MI.getOperand(1 - Ops[0]);
MachineBasicBlock::iterator Pos = MI;
if (Flags == MachineMemOperand::MOStore)
storeRegToStackSlot(*MBB, Pos, MO.getReg(), MO.isKill(), FI, RC, TRI);
else
loadRegFromStackSlot(*MBB, Pos, MO.getReg(), FI, RC, TRI);
return &*--Pos;
}
bool MipsExpandPseudo::expandAtomicCmpSwap(MachineBasicBlock &BB,
MachineBasicBlock::iterator I,
MachineBasicBlock::iterator &NMBBI) {
const unsigned Size =
I->getOpcode() == Mips::ATOMIC_CMP_SWAP_I32_POSTRA ? 4 : 8;
MachineFunction *MF = BB.getParent();
const bool ArePtrs64bit = STI->getABI().ArePtrs64bit();
DebugLoc DL = I->getDebugLoc();
unsigned LL, SC, ZERO, BNE, BEQ, MOVE;
if (Size == 4) {
if (STI->inMicroMipsMode()) {
LL = STI->hasMips32r6() ? Mips::LL_MMR6 : Mips::LL_MM;
SC = STI->hasMips32r6() ? Mips::SC_MMR6 : Mips::SC_MM;
BNE = STI->hasMips32r6() ? Mips::BNEC_MMR6 : Mips::BNE_MM;
BEQ = STI->hasMips32r6() ? Mips::BEQC_MMR6 : Mips::BEQ_MM;
} else {
LL = STI->hasMips32r6()
? (ArePtrs64bit ? Mips::LL64_R6 : Mips::LL_R6)
: (ArePtrs64bit ? Mips::LL64 : Mips::LL);
SC = STI->hasMips32r6()
? (ArePtrs64bit ? Mips::SC64_R6 : Mips::SC_R6)
: (ArePtrs64bit ? Mips::SC64 : Mips::SC);
BNE = Mips::BNE;
BEQ = Mips::BEQ;
}
ZERO = Mips::ZERO;
MOVE = Mips::OR;
} else {
LL = STI->hasMips64r6() ? Mips::LLD_R6 : Mips::LLD;
SC = STI->hasMips64r6() ? Mips::SCD_R6 : Mips::SCD;
ZERO = Mips::ZERO_64;
BNE = Mips::BNE64;
BEQ = Mips::BEQ64;
MOVE = Mips::OR64;
}
unsigned Dest = I->getOperand(0).getReg();
unsigned Ptr = I->getOperand(1).getReg();
unsigned OldVal = I->getOperand(2).getReg();
unsigned NewVal = I->getOperand(3).getReg();
unsigned Scratch = I->getOperand(4).getReg();
// insert new blocks after the current block
const BasicBlock *LLVM_BB = BB.getBasicBlock();
MachineBasicBlock *loop1MBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB.getIterator();
MF->insert(It, loop1MBB);
MF->insert(It, loop2MBB);
MF->insert(It, exitMBB);
// Transfer the remainder of BB and its successor edges to exitMBB.
exitMBB->splice(exitMBB->begin(), &BB,
std::next(MachineBasicBlock::iterator(I)), BB.end());
exitMBB->transferSuccessorsAndUpdatePHIs(&BB);
// thisMBB:
// ...
// fallthrough --> loop1MBB
BB.addSuccessor(loop1MBB, BranchProbability::getOne());
loop1MBB->addSuccessor(exitMBB);
loop1MBB->addSuccessor(loop2MBB);
loop1MBB->normalizeSuccProbs();
loop2MBB->addSuccessor(loop1MBB);
loop2MBB->addSuccessor(exitMBB);
loop2MBB->normalizeSuccProbs();
// loop1MBB:
// ll dest, 0(ptr)
// bne dest, oldval, exitMBB
BuildMI(loop1MBB, DL, TII->get(LL), Dest).addReg(Ptr).addImm(0);
BuildMI(loop1MBB, DL, TII->get(BNE))
.addReg(Dest, RegState::Kill).addReg(OldVal).addMBB(exitMBB);
// loop2MBB:
// move scratch, NewVal
// sc Scratch, Scratch, 0(ptr)
// beq Scratch, $0, loop1MBB
BuildMI(loop2MBB, DL, TII->get(MOVE), Scratch).addReg(NewVal).addReg(ZERO);
BuildMI(loop2MBB, DL, TII->get(SC), Scratch)
.addReg(Scratch).addReg(Ptr).addImm(0);
BuildMI(loop2MBB, DL, TII->get(BEQ))
.addReg(Scratch, RegState::Kill).addReg(ZERO).addMBB(loop1MBB);
LivePhysRegs LiveRegs;
computeAndAddLiveIns(LiveRegs, *loop1MBB);
computeAndAddLiveIns(LiveRegs, *loop2MBB);
computeAndAddLiveIns(LiveRegs, *exitMBB);
NMBBI = BB.end();
I->eraseFromParent();
return true;
}
bool MipsExpandPseudo::expandAtomicBinOp(MachineBasicBlock &BB,
MachineBasicBlock::iterator I,
MachineBasicBlock::iterator &NMBBI,
unsigned Size) {
MachineFunction *MF = BB.getParent();
const bool ArePtrs64bit = STI->getABI().ArePtrs64bit();
DebugLoc DL = I->getDebugLoc();
unsigned LL, SC, ZERO, BEQ;
if (Size == 4) {
if (STI->inMicroMipsMode()) {
LL = STI->hasMips32r6() ? Mips::LL_MMR6 : Mips::LL_MM;
SC = STI->hasMips32r6() ? Mips::SC_MMR6 : Mips::SC_MM;
BEQ = STI->hasMips32r6() ? Mips::BEQC_MMR6 : Mips::BEQ_MM;
} else {
LL = STI->hasMips32r6()
? (ArePtrs64bit ? Mips::LL64_R6 : Mips::LL_R6)
: (ArePtrs64bit ? Mips::LL64 : Mips::LL);
SC = STI->hasMips32r6()
? (ArePtrs64bit ? Mips::SC64_R6 : Mips::SC_R6)
: (ArePtrs64bit ? Mips::SC64 : Mips::SC);
BEQ = Mips::BEQ;
}
ZERO = Mips::ZERO;
} else {
LL = STI->hasMips64r6() ? Mips::LLD_R6 : Mips::LLD;
SC = STI->hasMips64r6() ? Mips::SCD_R6 : Mips::SCD;
ZERO = Mips::ZERO_64;
BEQ = Mips::BEQ64;
}
unsigned OldVal = I->getOperand(0).getReg();
unsigned Ptr = I->getOperand(1).getReg();
unsigned Incr = I->getOperand(2).getReg();
unsigned Scratch = I->getOperand(3).getReg();
unsigned Opcode = 0;
unsigned OR = 0;
unsigned AND = 0;
unsigned NOR = 0;
bool IsNand = false;
switch (I->getOpcode()) {
case Mips::ATOMIC_LOAD_ADD_I32_POSTRA:
Opcode = Mips::ADDu;
break;
case Mips::ATOMIC_LOAD_SUB_I32_POSTRA:
Opcode = Mips::SUBu;
break;
case Mips::ATOMIC_LOAD_AND_I32_POSTRA:
Opcode = Mips::AND;
break;
case Mips::ATOMIC_LOAD_OR_I32_POSTRA:
Opcode = Mips::OR;
break;
case Mips::ATOMIC_LOAD_XOR_I32_POSTRA:
Opcode = Mips::XOR;
break;
case Mips::ATOMIC_LOAD_NAND_I32_POSTRA:
IsNand = true;
AND = Mips::AND;
NOR = Mips::NOR;
break;
case Mips::ATOMIC_SWAP_I32_POSTRA:
OR = Mips::OR;
break;
case Mips::ATOMIC_LOAD_ADD_I64_POSTRA:
Opcode = Mips::DADDu;
break;
case Mips::ATOMIC_LOAD_SUB_I64_POSTRA:
Opcode = Mips::DSUBu;
break;
case Mips::ATOMIC_LOAD_AND_I64_POSTRA:
Opcode = Mips::AND64;
break;
case Mips::ATOMIC_LOAD_OR_I64_POSTRA:
Opcode = Mips::OR64;
break;
case Mips::ATOMIC_LOAD_XOR_I64_POSTRA:
Opcode = Mips::XOR64;
break;
case Mips::ATOMIC_LOAD_NAND_I64_POSTRA:
IsNand = true;
AND = Mips::AND64;
NOR = Mips::NOR64;
break;
case Mips::ATOMIC_SWAP_I64_POSTRA:
OR = Mips::OR64;
break;
default:
llvm_unreachable("Unknown pseudo atomic!");
}
const BasicBlock *LLVM_BB = BB.getBasicBlock();
MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB.getIterator();
MF->insert(It, loopMBB);
MF->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), &BB, std::next(I), BB.end());
exitMBB->transferSuccessorsAndUpdatePHIs(&BB);
BB.addSuccessor(loopMBB, BranchProbability::getOne());
loopMBB->addSuccessor(exitMBB);
loopMBB->addSuccessor(loopMBB);
loopMBB->normalizeSuccProbs();
BuildMI(loopMBB, DL, TII->get(LL), OldVal).addReg(Ptr).addImm(0);
assert((OldVal != Ptr) && "Clobbered the wrong ptr reg!");
assert((OldVal != Incr) && "Clobbered the wrong reg!");
if (Opcode) {
BuildMI(loopMBB, DL, TII->get(Opcode), Scratch).addReg(OldVal).addReg(Incr);
} else if (IsNand) {
assert(AND && NOR &&
"Unknown nand instruction for atomic pseudo expansion");
BuildMI(loopMBB, DL, TII->get(AND), Scratch).addReg(OldVal).addReg(Incr);
BuildMI(loopMBB, DL, TII->get(NOR), Scratch).addReg(ZERO).addReg(Scratch);
} else {
assert(OR && "Unknown instruction for atomic pseudo expansion!");
BuildMI(loopMBB, DL, TII->get(OR), Scratch).addReg(Incr).addReg(ZERO);
}
BuildMI(loopMBB, DL, TII->get(SC), Scratch).addReg(Scratch).addReg(Ptr).addImm(0);
BuildMI(loopMBB, DL, TII->get(BEQ)).addReg(Scratch).addReg(ZERO).addMBB(loopMBB);
NMBBI = BB.end();
I->eraseFromParent();
LivePhysRegs LiveRegs;
computeAndAddLiveIns(LiveRegs, *loopMBB);
computeAndAddLiveIns(LiveRegs, *exitMBB);
return true;
}
bool AArch64FrameLowering::restoreCalleeSavedRegisters(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
DebugLoc DL;
SmallVector<RegPairInfo, 8> RegPairs;
if (MI != MBB.end())
DL = MI->getDebugLoc();
computeCalleeSaveRegisterPairs(MF, CSI, TRI, RegPairs);
for (auto RPII = RegPairs.begin(), RPIE = RegPairs.end(); RPII != RPIE;
++RPII) {
RegPairInfo RPI = *RPII;
unsigned Reg1 = RPI.Reg1;
unsigned Reg2 = RPI.Reg2;
// Issue sequence of non-sp increment and sp-pi restores for cs regs. Only
// the last load is sp-pi post-increment and de-allocates the stack:
// For example:
// ldp fp, lr, [sp, #32] // addImm(+4)
// ldp x20, x19, [sp, #16] // addImm(+2)
// ldp x22, x21, [sp], #48 // addImm(+6)
// Note: see comment in spillCalleeSavedRegisters()
unsigned LdrOpc;
bool BumpSP = RPII == std::prev(RegPairs.end());
if (RPI.IsGPR) {
if (BumpSP)
LdrOpc = RPI.isPaired() ? AArch64::LDPXpost : AArch64::LDRXpost;
else
LdrOpc = RPI.isPaired() ? AArch64::LDPXi : AArch64::LDRXui;
} else {
if (BumpSP)
LdrOpc = RPI.isPaired() ? AArch64::LDPDpost : AArch64::LDRDpost;
else
LdrOpc = RPI.isPaired() ? AArch64::LDPDi : AArch64::LDRDui;
}
DEBUG(dbgs() << "CSR restore: (" << TRI->getName(Reg1);
if (RPI.isPaired())
dbgs() << ", " << TRI->getName(Reg2);
dbgs() << ") -> fi#(" << RPI.FrameIdx;
if (RPI.isPaired())
dbgs() << ", " << RPI.FrameIdx+1;
dbgs() << ")\n");
const int Offset = RPI.Offset;
MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(LdrOpc));
if (BumpSP)
MIB.addReg(AArch64::SP, RegState::Define);
if (RPI.isPaired())
MIB.addReg(Reg2, getDefRegState(true))
.addReg(Reg1, getDefRegState(true))
.addReg(AArch64::SP)
.addImm(Offset) // [sp], #offset * 8 or [sp, #offset * 8]
// where the factor * 8 is implicit
.setMIFlag(MachineInstr::FrameDestroy);
else
MIB.addReg(Reg1, getDefRegState(true))
.addReg(AArch64::SP)
.addImm(BumpSP ? Offset * 8 : Offset) // post-dec version is unscaled
.setMIFlag(MachineInstr::FrameDestroy);
}
return true;
}
/// finalizeBundle - Finalize a machine instruction bundle which includes
/// a sequence of instructions starting from FirstMI to LastMI (exclusive).
/// This routine adds a BUNDLE instruction to represent the bundle, it adds
/// IsInternalRead markers to MachineOperands which are defined inside the
/// bundle, and it copies externally visible defs and uses to the BUNDLE
/// instruction.
void llvm::finalizeBundle(MachineBasicBlock &MBB,
MachineBasicBlock::instr_iterator FirstMI,
MachineBasicBlock::instr_iterator LastMI) {
assert(FirstMI != LastMI && "Empty bundle?");
MIBundleBuilder Bundle(MBB, FirstMI, LastMI);
const TargetMachine &TM = MBB.getParent()->getTarget();
const TargetInstrInfo *TII = TM.getInstrInfo();
const TargetRegisterInfo *TRI = TM.getRegisterInfo();
MachineInstrBuilder MIB = BuildMI(*MBB.getParent(), FirstMI->getDebugLoc(),
TII->get(TargetOpcode::BUNDLE));
Bundle.prepend(MIB);
SmallVector<unsigned, 32> LocalDefs;
SmallSet<unsigned, 32> LocalDefSet;
SmallSet<unsigned, 8> DeadDefSet;
SmallSet<unsigned, 16> KilledDefSet;
SmallVector<unsigned, 8> ExternUses;
SmallSet<unsigned, 8> ExternUseSet;
SmallSet<unsigned, 8> KilledUseSet;
SmallSet<unsigned, 8> UndefUseSet;
SmallVector<MachineOperand*, 4> Defs;
for (; FirstMI != LastMI; ++FirstMI) {
for (unsigned i = 0, e = FirstMI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = FirstMI->getOperand(i);
if (!MO.isReg())
continue;
if (MO.isDef()) {
Defs.push_back(&MO);
continue;
}
unsigned Reg = MO.getReg();
if (!Reg)
continue;
assert(TargetRegisterInfo::isPhysicalRegister(Reg));
if (LocalDefSet.count(Reg)) {
MO.setIsInternalRead();
if (MO.isKill())
// Internal def is now killed.
KilledDefSet.insert(Reg);
} else {
if (ExternUseSet.insert(Reg)) {
ExternUses.push_back(Reg);
if (MO.isUndef())
UndefUseSet.insert(Reg);
}
if (MO.isKill())
// External def is now killed.
KilledUseSet.insert(Reg);
}
}
for (unsigned i = 0, e = Defs.size(); i != e; ++i) {
MachineOperand &MO = *Defs[i];
unsigned Reg = MO.getReg();
if (!Reg)
continue;
if (LocalDefSet.insert(Reg)) {
LocalDefs.push_back(Reg);
if (MO.isDead()) {
DeadDefSet.insert(Reg);
}
} else {
// Re-defined inside the bundle, it's no longer killed.
KilledDefSet.erase(Reg);
if (!MO.isDead())
// Previously defined but dead.
DeadDefSet.erase(Reg);
}
if (!MO.isDead()) {
for (MCSubRegIterator SubRegs(Reg, TRI); SubRegs.isValid(); ++SubRegs) {
unsigned SubReg = *SubRegs;
if (LocalDefSet.insert(SubReg))
LocalDefs.push_back(SubReg);
}
}
}
Defs.clear();
}
SmallSet<unsigned, 32> Added;
for (unsigned i = 0, e = LocalDefs.size(); i != e; ++i) {
unsigned Reg = LocalDefs[i];
if (Added.insert(Reg)) {
// If it's not live beyond end of the bundle, mark it dead.
bool isDead = DeadDefSet.count(Reg) || KilledDefSet.count(Reg);
MIB.addReg(Reg, getDefRegState(true) | getDeadRegState(isDead) |
getImplRegState(true));
}
}
for (unsigned i = 0, e = ExternUses.size(); i != e; ++i) {
unsigned Reg = ExternUses[i];
bool isKill = KilledUseSet.count(Reg);
bool isUndef = UndefUseSet.count(Reg);
MIB.addReg(Reg, getKillRegState(isKill) | getUndefRegState(isUndef) |
getImplRegState(true));
}
}
void ARMFrameLowering::emitPushInst(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
unsigned StmOpc, unsigned StrOpc,
bool NoGap,
bool(*Func)(unsigned, bool),
unsigned MIFlags) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
DebugLoc DL;
if (MI != MBB.end()) DL = MI->getDebugLoc();
SmallVector<std::pair<unsigned,bool>, 4> Regs;
unsigned i = CSI.size();
while (i != 0) {
unsigned LastReg = 0;
for (; i != 0; --i) {
unsigned Reg = CSI[i-1].getReg();
if (!(Func)(Reg, STI.isTargetDarwin())) continue;
// Add the callee-saved register as live-in unless it's LR and
// @llvm.returnaddress is called. If LR is returned for
// @llvm.returnaddress then it's already added to the function and
// entry block live-in sets.
bool isKill = true;
if (Reg == ARM::LR) {
if (MF.getFrameInfo()->isReturnAddressTaken() &&
MF.getRegInfo().isLiveIn(Reg))
isKill = false;
}
if (isKill)
MBB.addLiveIn(Reg);
// If NoGap is true, push consecutive registers and then leave the rest
// for other instructions. e.g.
// vpush {d8, d10, d11} -> vpush {d8}, vpush {d10, d11}
if (NoGap && LastReg && LastReg != Reg-1)
break;
LastReg = Reg;
Regs.push_back(std::make_pair(Reg, isKill));
}
if (Regs.empty())
continue;
if (Regs.size() > 1 || StrOpc== 0) {
MachineInstrBuilder MIB =
AddDefaultPred(BuildMI(MBB, MI, DL, TII.get(StmOpc), ARM::SP)
.addReg(ARM::SP).setMIFlags(MIFlags));
for (unsigned i = 0, e = Regs.size(); i < e; ++i)
MIB.addReg(Regs[i].first, getKillRegState(Regs[i].second));
} else if (Regs.size() == 1) {
MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(StrOpc),
ARM::SP)
.addReg(Regs[0].first, getKillRegState(Regs[0].second))
.addReg(ARM::SP).setMIFlags(MIFlags);
// ARM mode needs an extra reg0 here due to addrmode2. Will go away once
// that refactoring is complete (eventually).
if (StrOpc == ARM::STR_PRE) {
MIB.addReg(0);
MIB.addImm(ARM_AM::getAM2Opc(ARM_AM::sub, 4, ARM_AM::no_shift));
} else
MIB.addImm(-4);
AddDefaultPred(MIB);
}
Regs.clear();
}
}
void MipsSEInstrInfo::
loadRegFromStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
unsigned DestReg, int FI, const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI, int64_t Offset) const {
DebugLoc DL;
if (I != MBB.end()) DL = I->getDebugLoc();
MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOLoad);
unsigned Opc = 0;
const Function *Func = MBB.getParent()->getFunction();
bool ReqIndirectLoad = Func->hasFnAttribute("interrupt") &&
(DestReg == Mips::LO0 || DestReg == Mips::LO0_64 ||
DestReg == Mips::HI0 || DestReg == Mips::HI0_64);
if (Mips::GPR32RegClass.hasSubClassEq(RC))
Opc = Mips::LW;
else if (Mips::GPR64RegClass.hasSubClassEq(RC))
Opc = Mips::LD;
else if (Mips::ACC64RegClass.hasSubClassEq(RC))
Opc = Mips::LOAD_ACC64;
else if (Mips::ACC64DSPRegClass.hasSubClassEq(RC))
Opc = Mips::LOAD_ACC64DSP;
else if (Mips::ACC128RegClass.hasSubClassEq(RC))
Opc = Mips::LOAD_ACC128;
else if (Mips::DSPCCRegClass.hasSubClassEq(RC))
Opc = Mips::LOAD_CCOND_DSP;
else if (Mips::FGR32RegClass.hasSubClassEq(RC))
Opc = Mips::LWC1;
else if (Mips::AFGR64RegClass.hasSubClassEq(RC))
Opc = Mips::LDC1;
else if (Mips::FGR64RegClass.hasSubClassEq(RC))
Opc = Mips::LDC164;
else if (TRI->isTypeLegalForClass(*RC, MVT::v16i8))
Opc = Mips::LD_B;
else if (TRI->isTypeLegalForClass(*RC, MVT::v8i16) ||
TRI->isTypeLegalForClass(*RC, MVT::v8f16))
Opc = Mips::LD_H;
else if (TRI->isTypeLegalForClass(*RC, MVT::v4i32) ||
TRI->isTypeLegalForClass(*RC, MVT::v4f32))
Opc = Mips::LD_W;
else if (TRI->isTypeLegalForClass(*RC, MVT::v2i64) ||
TRI->isTypeLegalForClass(*RC, MVT::v2f64))
Opc = Mips::LD_D;
else if (Mips::HI32RegClass.hasSubClassEq(RC))
Opc = Mips::LW;
else if (Mips::HI64RegClass.hasSubClassEq(RC))
Opc = Mips::LD;
else if (Mips::LO32RegClass.hasSubClassEq(RC))
Opc = Mips::LW;
else if (Mips::LO64RegClass.hasSubClassEq(RC))
Opc = Mips::LD;
assert(Opc && "Register class not handled!");
if (!ReqIndirectLoad)
BuildMI(MBB, I, DL, get(Opc), DestReg)
.addFrameIndex(FI)
.addImm(Offset)
.addMemOperand(MMO);
else {
// Load HI/LO through K0. Notably the DestReg is encoded into the
// instruction itself.
unsigned Reg = Mips::K0;
unsigned LdOp = Mips::MTLO;
if (DestReg == Mips::HI0)
LdOp = Mips::MTHI;
if (Subtarget.getABI().ArePtrs64bit()) {
Reg = Mips::K0_64;
if (DestReg == Mips::HI0_64)
LdOp = Mips::MTHI64;
else
LdOp = Mips::MTLO64;
}
BuildMI(MBB, I, DL, get(Opc), Reg)
.addFrameIndex(FI)
.addImm(Offset)
.addMemOperand(MMO);
BuildMI(MBB, I, DL, get(LdOp)).addReg(Reg);
}
}
// Replace uses of FromReg with ToReg if they are dominated by MI.
static bool ReplaceDominatedUses(MachineBasicBlock &MBB, MachineInstr &MI,
unsigned FromReg, unsigned ToReg,
const MachineRegisterInfo &MRI,
MachineDominatorTree &MDT,
LiveIntervals &LIS) {
bool Changed = false;
LiveInterval *FromLI = &LIS.getInterval(FromReg);
LiveInterval *ToLI = &LIS.getInterval(ToReg);
SlotIndex FromIdx = LIS.getInstructionIndex(MI).getRegSlot();
VNInfo *FromVNI = FromLI->getVNInfoAt(FromIdx);
SmallVector<SlotIndex, 4> Indices;
for (auto I = MRI.use_nodbg_begin(FromReg), E = MRI.use_nodbg_end();
I != E;) {
MachineOperand &O = *I++;
MachineInstr *Where = O.getParent();
// Check that MI dominates the instruction in the normal way.
if (&MI == Where || !MDT.dominates(&MI, Where))
continue;
// If this use gets a different value, skip it.
SlotIndex WhereIdx = LIS.getInstructionIndex(*Where);
VNInfo *WhereVNI = FromLI->getVNInfoAt(WhereIdx);
if (WhereVNI && WhereVNI != FromVNI)
continue;
// Make sure ToReg isn't clobbered before it gets there.
VNInfo *ToVNI = ToLI->getVNInfoAt(WhereIdx);
if (ToVNI && ToVNI != FromVNI)
continue;
Changed = true;
LLVM_DEBUG(dbgs() << "Setting operand " << O << " in " << *Where << " from "
<< MI << "\n");
O.setReg(ToReg);
// If the store's def was previously dead, it is no longer.
if (!O.isUndef()) {
MI.getOperand(0).setIsDead(false);
Indices.push_back(WhereIdx.getRegSlot());
}
}
if (Changed) {
// Extend ToReg's liveness.
LIS.extendToIndices(*ToLI, Indices);
// Shrink FromReg's liveness.
LIS.shrinkToUses(FromLI);
// If we replaced all dominated uses, FromReg is now killed at MI.
if (!FromLI->liveAt(FromIdx.getDeadSlot()))
MI.addRegisterKilled(FromReg, MBB.getParent()
->getSubtarget<WebAssemblySubtarget>()
.getRegisterInfo());
}
return Changed;
}
void MIPrinter::print(const MachineBasicBlock &MBB) {
assert(MBB.getNumber() >= 0 && "Invalid MBB number");
OS << "bb." << MBB.getNumber();
bool HasAttributes = false;
if (const auto *BB = MBB.getBasicBlock()) {
if (BB->hasName()) {
OS << "." << BB->getName();
} else {
HasAttributes = true;
OS << " (";
int Slot = MST.getLocalSlot(BB);
if (Slot == -1)
OS << "<ir-block badref>";
else
OS << (Twine("%ir-block.") + Twine(Slot)).str();
}
}
if (MBB.hasAddressTaken()) {
OS << (HasAttributes ? ", " : " (");
OS << "address-taken";
HasAttributes = true;
}
if (MBB.isEHPad()) {
OS << (HasAttributes ? ", " : " (");
OS << "landing-pad";
HasAttributes = true;
}
if (MBB.getAlignment()) {
OS << (HasAttributes ? ", " : " (");
OS << "align " << MBB.getAlignment();
HasAttributes = true;
}
if (HasAttributes)
OS << ")";
OS << ":\n";
bool HasLineAttributes = false;
// Print the successors
if (!MBB.succ_empty()) {
OS.indent(2) << "successors: ";
for (auto I = MBB.succ_begin(), E = MBB.succ_end(); I != E; ++I) {
if (I != MBB.succ_begin())
OS << ", ";
printMBBReference(**I);
if (MBB.hasSuccessorProbabilities())
OS << '(' << MBB.getSuccProbability(I) << ')';
}
OS << "\n";
HasLineAttributes = true;
}
// Print the live in registers.
const auto *TRI = MBB.getParent()->getSubtarget().getRegisterInfo();
assert(TRI && "Expected target register info");
if (!MBB.livein_empty()) {
OS.indent(2) << "liveins: ";
bool First = true;
for (const auto &LI : MBB.liveins()) {
if (!First)
OS << ", ";
First = false;
printReg(LI.PhysReg, OS, TRI);
if (LI.LaneMask != ~0u)
OS << ':' << PrintLaneMask(LI.LaneMask);
}
OS << "\n";
HasLineAttributes = true;
}
if (HasLineAttributes)
OS << "\n";
bool IsInBundle = false;
for (auto I = MBB.instr_begin(), E = MBB.instr_end(); I != E; ++I) {
const MachineInstr &MI = *I;
if (IsInBundle && !MI.isInsideBundle()) {
OS.indent(2) << "}\n";
IsInBundle = false;
}
OS.indent(IsInBundle ? 4 : 2);
print(MI);
if (!IsInBundle && MI.getFlag(MachineInstr::BundledSucc)) {
OS << " {";
IsInBundle = true;
}
OS << "\n";
}
if (IsInBundle)
OS.indent(2) << "}\n";
}
void MIPrinter::print(const MachineBasicBlock &MBB) {
assert(MBB.getNumber() >= 0 && "Invalid MBB number");
OS << "bb." << MBB.getNumber();
bool HasAttributes = false;
if (const auto *BB = MBB.getBasicBlock()) {
if (BB->hasName()) {
OS << "." << BB->getName();
} else {
HasAttributes = true;
OS << " (";
int Slot = MST.getLocalSlot(BB);
if (Slot == -1)
OS << "<ir-block badref>";
else
OS << (Twine("%ir-block.") + Twine(Slot)).str();
}
}
if (MBB.hasAddressTaken()) {
OS << (HasAttributes ? ", " : " (");
OS << "address-taken";
HasAttributes = true;
}
if (MBB.isEHPad()) {
OS << (HasAttributes ? ", " : " (");
OS << "landing-pad";
HasAttributes = true;
}
if (MBB.getAlignment()) {
OS << (HasAttributes ? ", " : " (");
OS << "align " << MBB.getAlignment();
HasAttributes = true;
}
if (HasAttributes)
OS << ")";
OS << ":\n";
bool HasLineAttributes = false;
// Print the successors
bool canPredictProbs = canPredictBranchProbabilities(MBB);
if (!MBB.succ_empty() && (!SimplifyMIR || !canPredictProbs ||
!canPredictSuccessors(MBB))) {
OS.indent(2) << "successors: ";
for (auto I = MBB.succ_begin(), E = MBB.succ_end(); I != E; ++I) {
if (I != MBB.succ_begin())
OS << ", ";
printMBBReference(**I);
if (!SimplifyMIR || !canPredictProbs)
OS << '('
<< format("0x%08" PRIx32, MBB.getSuccProbability(I).getNumerator())
<< ')';
}
OS << "\n";
HasLineAttributes = true;
}
// Print the live in registers.
const MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
if (MRI.tracksLiveness() && !MBB.livein_empty()) {
const TargetRegisterInfo &TRI = *MRI.getTargetRegisterInfo();
OS.indent(2) << "liveins: ";
bool First = true;
for (const auto &LI : MBB.liveins()) {
if (!First)
OS << ", ";
First = false;
printReg(LI.PhysReg, OS, &TRI);
if (!LI.LaneMask.all())
OS << ":0x" << PrintLaneMask(LI.LaneMask);
}
OS << "\n";
HasLineAttributes = true;
}
if (HasLineAttributes)
OS << "\n";
bool IsInBundle = false;
for (auto I = MBB.instr_begin(), E = MBB.instr_end(); I != E; ++I) {
const MachineInstr &MI = *I;
if (IsInBundle && !MI.isInsideBundle()) {
OS.indent(2) << "}\n";
IsInBundle = false;
}
OS.indent(IsInBundle ? 4 : 2);
print(MI);
if (!IsInBundle && MI.getFlag(MachineInstr::BundledSucc)) {
OS << " {";
IsInBundle = true;
}
OS << "\n";
}
if (IsInBundle)
OS.indent(2) << "}\n";
}
/// If \p MBBI is a pseudo instruction, this method expands
/// it to the corresponding (sequence of) actual instruction(s).
/// \returns true if \p MBBI has been expanded.
bool X86ExpandPseudo::ExpandMI(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MBBI) {
MachineInstr &MI = *MBBI;
unsigned Opcode = MI.getOpcode();
DebugLoc DL = MBBI->getDebugLoc();
switch (Opcode) {
default:
return false;
case X86::TCRETURNdi:
case X86::TCRETURNdicc:
case X86::TCRETURNri:
case X86::TCRETURNmi:
case X86::TCRETURNdi64:
case X86::TCRETURNdi64cc:
case X86::TCRETURNri64:
case X86::TCRETURNmi64: {
bool isMem = Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64;
MachineOperand &JumpTarget = MBBI->getOperand(0);
MachineOperand &StackAdjust = MBBI->getOperand(isMem ? 5 : 1);
assert(StackAdjust.isImm() && "Expecting immediate value.");
// Adjust stack pointer.
int StackAdj = StackAdjust.getImm();
int MaxTCDelta = X86FI->getTCReturnAddrDelta();
int Offset = 0;
assert(MaxTCDelta <= 0 && "MaxTCDelta should never be positive");
// Incoporate the retaddr area.
Offset = StackAdj - MaxTCDelta;
assert(Offset >= 0 && "Offset should never be negative");
if (Opcode == X86::TCRETURNdicc || Opcode == X86::TCRETURNdi64cc) {
assert(Offset == 0 && "Conditional tail call cannot adjust the stack.");
}
if (Offset) {
// Check for possible merge with preceding ADD instruction.
Offset += X86FL->mergeSPUpdates(MBB, MBBI, true);
X86FL->emitSPUpdate(MBB, MBBI, Offset, /*InEpilogue=*/true);
}
// Jump to label or value in register.
bool IsWin64 = STI->isTargetWin64();
if (Opcode == X86::TCRETURNdi || Opcode == X86::TCRETURNdicc ||
Opcode == X86::TCRETURNdi64 || Opcode == X86::TCRETURNdi64cc) {
unsigned Op;
switch (Opcode) {
case X86::TCRETURNdi:
Op = X86::TAILJMPd;
break;
case X86::TCRETURNdicc:
Op = X86::TAILJMPd_CC;
break;
case X86::TCRETURNdi64cc:
assert(!IsWin64 && "Conditional tail calls confuse the Win64 unwinder.");
// TODO: We could do it for Win64 "leaf" functions though; PR30337.
Op = X86::TAILJMPd64_CC;
break;
default:
// Note: Win64 uses REX prefixes indirect jumps out of functions, but
// not direct ones.
Op = X86::TAILJMPd64;
break;
}
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
if (JumpTarget.isGlobal()) {
MIB.addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(),
JumpTarget.getTargetFlags());
} else {
assert(JumpTarget.isSymbol());
MIB.addExternalSymbol(JumpTarget.getSymbolName(),
JumpTarget.getTargetFlags());
}
if (Op == X86::TAILJMPd_CC || Op == X86::TAILJMPd64_CC) {
MIB.addImm(MBBI->getOperand(2).getImm());
}
} else if (Opcode == X86::TCRETURNmi || Opcode == X86::TCRETURNmi64) {
unsigned Op = (Opcode == X86::TCRETURNmi)
? X86::TAILJMPm
: (IsWin64 ? X86::TAILJMPm64_REX : X86::TAILJMPm64);
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, DL, TII->get(Op));
for (unsigned i = 0; i != 5; ++i)
MIB.addOperand(MBBI->getOperand(i));
} else if (Opcode == X86::TCRETURNri64) {
BuildMI(MBB, MBBI, DL,
TII->get(IsWin64 ? X86::TAILJMPr64_REX : X86::TAILJMPr64))
.addReg(JumpTarget.getReg(), RegState::Kill);
} else {
BuildMI(MBB, MBBI, DL, TII->get(X86::TAILJMPr))
.addReg(JumpTarget.getReg(), RegState::Kill);
}
MachineInstr &NewMI = *std::prev(MBBI);
NewMI.copyImplicitOps(*MBBI->getParent()->getParent(), *MBBI);
// Delete the pseudo instruction TCRETURN.
MBB.erase(MBBI);
return true;
}
case X86::EH_RETURN:
case X86::EH_RETURN64: {
MachineOperand &DestAddr = MBBI->getOperand(0);
assert(DestAddr.isReg() && "Offset should be in register!");
const bool Uses64BitFramePtr =
STI->isTarget64BitLP64() || STI->isTargetNaCl64();
unsigned StackPtr = TRI->getStackRegister();
BuildMI(MBB, MBBI, DL,
TII->get(Uses64BitFramePtr ? X86::MOV64rr : X86::MOV32rr), StackPtr)
.addReg(DestAddr.getReg());
// The EH_RETURN pseudo is really removed during the MC Lowering.
return true;
}
case X86::IRET: {
// Adjust stack to erase error code
int64_t StackAdj = MBBI->getOperand(0).getImm();
X86FL->emitSPUpdate(MBB, MBBI, StackAdj, true);
// Replace pseudo with machine iret
BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::IRET64 : X86::IRET32));
MBB.erase(MBBI);
return true;
}
case X86::RET: {
// Adjust stack to erase error code
int64_t StackAdj = MBBI->getOperand(0).getImm();
MachineInstrBuilder MIB;
if (StackAdj == 0) {
MIB = BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::RETQ : X86::RETL));
} else if (isUInt<16>(StackAdj)) {
MIB = BuildMI(MBB, MBBI, DL,
TII->get(STI->is64Bit() ? X86::RETIQ : X86::RETIL))
.addImm(StackAdj);
} else {
assert(!STI->is64Bit() &&
"shouldn't need to do this for x86_64 targets!");
// A ret can only handle immediates as big as 2**16-1. If we need to pop
// off bytes before the return address, we must do it manually.
BuildMI(MBB, MBBI, DL, TII->get(X86::POP32r)).addReg(X86::ECX, RegState::Define);
X86FL->emitSPUpdate(MBB, MBBI, StackAdj, /*InEpilogue=*/true);
BuildMI(MBB, MBBI, DL, TII->get(X86::PUSH32r)).addReg(X86::ECX);
MIB = BuildMI(MBB, MBBI, DL, TII->get(X86::RETL));
}
for (unsigned I = 1, E = MBBI->getNumOperands(); I != E; ++I)
MIB.addOperand(MBBI->getOperand(I));
MBB.erase(MBBI);
return true;
}
case X86::EH_RESTORE: {
// Restore ESP and EBP, and optionally ESI if required.
bool IsSEH = isAsynchronousEHPersonality(classifyEHPersonality(
MBB.getParent()->getFunction()->getPersonalityFn()));
X86FL->restoreWin32EHStackPointers(MBB, MBBI, DL, /*RestoreSP=*/IsSEH);
MBBI->eraseFromParent();
return true;
}
case X86::LCMPXCHG8B_SAVE_EBX:
case X86::LCMPXCHG16B_SAVE_RBX: {
// Perform the following transformation.
// SaveRbx = pseudocmpxchg Addr, <4 opds for the address>, InArg, SaveRbx
// =>
// [E|R]BX = InArg
// actualcmpxchg Addr
// [E|R]BX = SaveRbx
const MachineOperand &InArg = MBBI->getOperand(6);
unsigned SaveRbx = MBBI->getOperand(7).getReg();
unsigned ActualInArg =
Opcode == X86::LCMPXCHG8B_SAVE_EBX ? X86::EBX : X86::RBX;
// Copy the input argument of the pseudo into the argument of the
// actual instruction.
TII->copyPhysReg(MBB, MBBI, DL, ActualInArg, InArg.getReg(),
InArg.isKill());
// Create the actual instruction.
unsigned ActualOpc =
Opcode == X86::LCMPXCHG8B_SAVE_EBX ? X86::LCMPXCHG8B : X86::LCMPXCHG16B;
MachineInstr *NewInstr = BuildMI(MBB, MBBI, DL, TII->get(ActualOpc));
// Copy the operands related to the address.
for (unsigned Idx = 1; Idx < 6; ++Idx)
NewInstr->addOperand(MBBI->getOperand(Idx));
// Finally, restore the value of RBX.
TII->copyPhysReg(MBB, MBBI, DL, ActualInArg, SaveRbx,
/*SrcIsKill*/ true);
// Delete the pseudo.
MBBI->eraseFromParent();
return true;
}
}
llvm_unreachable("Previous switch has a fallthrough?");
}
/// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
/// We assume there is only one delay slot per delayed instruction.
bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
bool Changed = false;
const MipsSubtarget &STI = MBB.getParent()->getSubtarget<MipsSubtarget>();
bool InMicroMipsMode = STI.inMicroMipsMode();
const MipsInstrInfo *TII = STI.getInstrInfo();
for (Iter I = MBB.begin(); I != MBB.end(); ++I) {
if (!hasUnoccupiedSlot(&*I))
continue;
++FilledSlots;
Changed = true;
// Delay slot filling is disabled at -O0.
if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None)) {
bool Filled = false;
if (searchBackward(MBB, I)) {
Filled = true;
} else if (I->isTerminator()) {
if (searchSuccBBs(MBB, I)) {
Filled = true;
}
} else if (searchForward(MBB, I)) {
Filled = true;
}
if (Filled) {
// Get instruction with delay slot.
MachineBasicBlock::instr_iterator DSI(I);
if (InMicroMipsMode && TII->GetInstSizeInBytes(std::next(DSI)) == 2 &&
DSI->isCall()) {
// If instruction in delay slot is 16b change opcode to
// corresponding instruction with short delay slot.
DSI->setDesc(TII->get(getEquivalentCallShort(DSI->getOpcode())));
}
continue;
}
}
// If instruction is BEQ or BNE with one ZERO register, then instead of
// adding NOP replace this instruction with the corresponding compact
// branch instruction, i.e. BEQZC or BNEZC.
unsigned Opcode = I->getOpcode();
if (InMicroMipsMode &&
(Opcode == Mips::BEQ || Opcode == Mips::BNE) &&
((unsigned) I->getOperand(1).getReg()) == Mips::ZERO) {
I = replaceWithCompactBranch(MBB, I, I->getDebugLoc());
} else {
// Bundle the NOP to the instruction with the delay slot.
BuildMI(MBB, std::next(I), I->getDebugLoc(), TII->get(Mips::NOP));
MIBundleBuilder(MBB, I, std::next(I, 2));
}
}
return Changed;
}
/// converToHardwareLoop - check if the loop is a candidate for
/// converting to a hardware loop. If so, then perform the
/// transformation.
///
/// This function works on innermost loops first. A loop can
/// be converted if it is a counting loop; either a register
/// value or an immediate.
///
/// The code makes several assumptions about the representation
/// of the loop in llvm.
bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L) {
bool Changed = false;
// Process nested loops first.
for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
Changed |= convertToHardwareLoop(*I);
}
// If a nested loop has been converted, then we can't convert this loop.
if (Changed) {
return Changed;
}
// Are we able to determine the trip count for the loop?
CountValue *TripCount = getTripCount(L);
if (TripCount == 0) {
return false;
}
// Does the loop contain any invalid instructions?
if (containsInvalidInstruction(L)) {
return false;
}
MachineBasicBlock *Preheader = L->getLoopPreheader();
// No preheader means there's not place for the loop instr.
if (Preheader == 0) {
return false;
}
MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
MachineBasicBlock *LastMBB = L->getExitingBlock();
// Don't generate hw loop if the loop has more than one exit.
if (LastMBB == 0) {
return false;
}
MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
// Determine the loop start.
MachineBasicBlock *LoopStart = L->getTopBlock();
if (L->getLoopLatch() != LastMBB) {
// When the exit and latch are not the same, use the latch block as the
// start.
// The loop start address is used only after the 1st iteration, and the loop
// latch may contains instrs. that need to be executed after the 1st iter.
LoopStart = L->getLoopLatch();
// Make sure the latch is a successor of the exit, otherwise it won't work.
if (!LastMBB->isSuccessor(LoopStart)) {
return false;
}
}
// Convert the loop to a hardware loop
DEBUG(dbgs() << "Change to hardware loop at "; L->dump());
if (TripCount->isReg()) {
// Create a copy of the loop count register.
MachineFunction *MF = LastMBB->getParent();
const TargetRegisterClass *RC =
MF->getRegInfo().getRegClass(TripCount->getReg());
unsigned CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(TargetOpcode::COPY), CountReg).addReg(TripCount->getReg());
if (TripCount->isNeg()) {
unsigned CountReg1 = CountReg;
CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::NEG), CountReg).addReg(CountReg1);
}
// Add the Loop instruction to the begining of the loop.
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::LOOP0_r)).addMBB(LoopStart).addReg(CountReg);
} else {
assert(TripCount->isImm() && "Expecting immedate vaule for trip count");
// Add the Loop immediate instruction to the beginning of the loop.
int64_t CountImm = TripCount->getImm();
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::LOOP0_i)).addMBB(LoopStart).addImm(CountImm);
}
// Make sure the loop start always has a reference in the CFG. We need to
// create a BlockAddress operand to get this mechanism to work both the
// MachineBasicBlock and BasicBlock objects need the flag set.
LoopStart->setHasAddressTaken();
// This line is needed to set the hasAddressTaken flag on the BasicBlock
// object
BlockAddress::get(const_cast<BasicBlock *>(LoopStart->getBasicBlock()));
// Replace the loop branch with an endloop instruction.
DebugLoc dl = LastI->getDebugLoc();
BuildMI(*LastMBB, LastI, dl, TII->get(Hexagon::ENDLOOP0)).addMBB(LoopStart);
// The loop ends with either:
// - a conditional branch followed by an unconditional branch, or
// - a conditional branch to the loop start.
if (LastI->getOpcode() == Hexagon::JMP_c ||
LastI->getOpcode() == Hexagon::JMP_cNot) {
// delete one and change/add an uncond. branch to out of the loop
MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB();
LastI = LastMBB->erase(LastI);
if (!L->contains(BranchTarget)) {
if (LastI != LastMBB->end()) {
TII->RemoveBranch(*LastMBB);
}
SmallVector<MachineOperand, 0> Cond;
TII->InsertBranch(*LastMBB, BranchTarget, 0, Cond, dl);
}
} else {
// Conditional branch to loop start; just delete it.
LastMBB->erase(LastI);
}
delete TripCount;
++NumHWLoops;
return true;
}
void ARMFrameLowering::emitPopInst(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
unsigned LdmOpc, unsigned LdrOpc,
bool isVarArg, bool NoGap,
bool(*Func)(unsigned, bool)) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
DebugLoc DL = MI->getDebugLoc();
unsigned RetOpcode = MI->getOpcode();
bool isTailCall = (RetOpcode == ARM::TCRETURNdi ||
RetOpcode == ARM::TCRETURNdiND ||
RetOpcode == ARM::TCRETURNri ||
RetOpcode == ARM::TCRETURNriND);
SmallVector<unsigned, 4> Regs;
unsigned i = CSI.size();
while (i != 0) {
unsigned LastReg = 0;
bool DeleteRet = false;
for (; i != 0; --i) {
unsigned Reg = CSI[i-1].getReg();
if (!(Func)(Reg, STI.isTargetDarwin())) continue;
if (Reg == ARM::LR && !isTailCall && !isVarArg && STI.hasV5TOps()) {
Reg = ARM::PC;
LdmOpc = AFI->isThumbFunction() ? ARM::t2LDMIA_RET : ARM::LDMIA_RET;
// Fold the return instruction into the LDM.
DeleteRet = true;
}
// If NoGap is true, pop consecutive registers and then leave the rest
// for other instructions. e.g.
// vpop {d8, d10, d11} -> vpop {d8}, vpop {d10, d11}
if (NoGap && LastReg && LastReg != Reg-1)
break;
LastReg = Reg;
Regs.push_back(Reg);
}
if (Regs.empty())
continue;
if (Regs.size() > 1 || LdrOpc == 0) {
MachineInstrBuilder MIB =
AddDefaultPred(BuildMI(MBB, MI, DL, TII.get(LdmOpc), ARM::SP)
.addReg(ARM::SP));
for (unsigned i = 0, e = Regs.size(); i < e; ++i)
MIB.addReg(Regs[i], getDefRegState(true));
if (DeleteRet)
MI->eraseFromParent();
MI = MIB;
} else if (Regs.size() == 1) {
// If we adjusted the reg to PC from LR above, switch it back here. We
// only do that for LDM.
if (Regs[0] == ARM::PC)
Regs[0] = ARM::LR;
MachineInstrBuilder MIB =
BuildMI(MBB, MI, DL, TII.get(LdrOpc), Regs[0])
.addReg(ARM::SP, RegState::Define)
.addReg(ARM::SP);
// ARM mode needs an extra reg0 here due to addrmode2. Will go away once
// that refactoring is complete (eventually).
if (LdrOpc == ARM::LDR_POST) {
MIB.addReg(0);
MIB.addImm(ARM_AM::getAM2Opc(ARM_AM::add, 4, ARM_AM::no_shift));
} else
MIB.addImm(4);
AddDefaultPred(MIB);
}
Regs.clear();
}
}
void MipsSEInstrInfo::
storeRegToStack(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
unsigned SrcReg, bool isKill, int FI,
const TargetRegisterClass *RC, const TargetRegisterInfo *TRI,
int64_t Offset) const {
DebugLoc DL;
MachineMemOperand *MMO = GetMemOperand(MBB, FI, MachineMemOperand::MOStore);
unsigned Opc = 0;
if (Mips::GPR32RegClass.hasSubClassEq(RC))
Opc = Mips::SW;
else if (Mips::GPR64RegClass.hasSubClassEq(RC))
Opc = Mips::SD;
else if (Mips::ACC64RegClass.hasSubClassEq(RC))
Opc = Mips::STORE_ACC64;
else if (Mips::ACC64DSPRegClass.hasSubClassEq(RC))
Opc = Mips::STORE_ACC64DSP;
else if (Mips::ACC128RegClass.hasSubClassEq(RC))
Opc = Mips::STORE_ACC128;
else if (Mips::DSPCCRegClass.hasSubClassEq(RC))
Opc = Mips::STORE_CCOND_DSP;
else if (Mips::FGR32RegClass.hasSubClassEq(RC))
Opc = Mips::SWC1;
else if (Mips::AFGR64RegClass.hasSubClassEq(RC))
Opc = Mips::SDC1;
else if (Mips::FGR64RegClass.hasSubClassEq(RC))
Opc = Mips::SDC164;
else if (TRI->isTypeLegalForClass(*RC, MVT::v16i8))
Opc = Mips::ST_B;
else if (TRI->isTypeLegalForClass(*RC, MVT::v8i16) ||
TRI->isTypeLegalForClass(*RC, MVT::v8f16))
Opc = Mips::ST_H;
else if (TRI->isTypeLegalForClass(*RC, MVT::v4i32) ||
TRI->isTypeLegalForClass(*RC, MVT::v4f32))
Opc = Mips::ST_W;
else if (TRI->isTypeLegalForClass(*RC, MVT::v2i64) ||
TRI->isTypeLegalForClass(*RC, MVT::v2f64))
Opc = Mips::ST_D;
else if (Mips::LO32RegClass.hasSubClassEq(RC))
Opc = Mips::SW;
else if (Mips::LO64RegClass.hasSubClassEq(RC))
Opc = Mips::SD;
else if (Mips::HI32RegClass.hasSubClassEq(RC))
Opc = Mips::SW;
else if (Mips::HI64RegClass.hasSubClassEq(RC))
Opc = Mips::SD;
// Hi, Lo are normally caller save but they are callee save
// for interrupt handling.
const Function *Func = MBB.getParent()->getFunction();
if (Func->hasFnAttribute("interrupt")) {
if (Mips::HI32RegClass.hasSubClassEq(RC)) {
BuildMI(MBB, I, DL, get(Mips::MFHI), Mips::K0);
SrcReg = Mips::K0;
} else if (Mips::HI64RegClass.hasSubClassEq(RC)) {
BuildMI(MBB, I, DL, get(Mips::MFHI64), Mips::K0_64);
SrcReg = Mips::K0_64;
} else if (Mips::LO32RegClass.hasSubClassEq(RC)) {
BuildMI(MBB, I, DL, get(Mips::MFLO), Mips::K0);
SrcReg = Mips::K0;
} else if (Mips::LO64RegClass.hasSubClassEq(RC)) {
BuildMI(MBB, I, DL, get(Mips::MFLO64), Mips::K0_64);
SrcReg = Mips::K0_64;
}
}
assert(Opc && "Register class not handled!");
BuildMI(MBB, I, DL, get(Opc)).addReg(SrcReg, getKillRegState(isKill))
.addFrameIndex(FI).addImm(Offset).addMemOperand(MMO);
}
/// LowerAtomicPHINode - Lower the PHI node at the top of the specified block,
/// under the assuption that it needs to be lowered in a way that supports
/// atomic execution of PHIs. This lowering method is always correct all of the
/// time.
///
void PHIElimination::LowerAtomicPHINode(
MachineBasicBlock &MBB,
MachineBasicBlock::iterator AfterPHIsIt) {
++NumAtomic;
// Unlink the PHI node from the basic block, but don't delete the PHI yet.
MachineInstr *MPhi = MBB.remove(MBB.begin());
unsigned NumSrcs = (MPhi->getNumOperands() - 1) / 2;
unsigned DestReg = MPhi->getOperand(0).getReg();
assert(MPhi->getOperand(0).getSubReg() == 0 && "Can't handle sub-reg PHIs");
bool isDead = MPhi->getOperand(0).isDead();
// Create a new register for the incoming PHI arguments.
MachineFunction &MF = *MBB.getParent();
unsigned IncomingReg = 0;
bool reusedIncoming = false; // Is IncomingReg reused from an earlier PHI?
// Insert a register to register copy at the top of the current block (but
// after any remaining phi nodes) which copies the new incoming register
// into the phi node destination.
const TargetInstrInfo *TII = MF.getTarget().getInstrInfo();
if (isSourceDefinedByImplicitDef(MPhi, MRI))
// If all sources of a PHI node are implicit_def, just emit an
// implicit_def instead of a copy.
BuildMI(MBB, AfterPHIsIt, MPhi->getDebugLoc(),
TII->get(TargetOpcode::IMPLICIT_DEF), DestReg);
else {
// Can we reuse an earlier PHI node? This only happens for critical edges,
// typically those created by tail duplication.
unsigned &entry = LoweredPHIs[MPhi];
if (entry) {
// An identical PHI node was already lowered. Reuse the incoming register.
IncomingReg = entry;
reusedIncoming = true;
++NumReused;
DEBUG(dbgs() << "Reusing %reg" << IncomingReg << " for " << *MPhi);
} else {
const TargetRegisterClass *RC = MF.getRegInfo().getRegClass(DestReg);
entry = IncomingReg = MF.getRegInfo().createVirtualRegister(RC);
}
BuildMI(MBB, AfterPHIsIt, MPhi->getDebugLoc(),
TII->get(TargetOpcode::COPY), DestReg)
.addReg(IncomingReg);
}
// Update live variable information if there is any.
LiveVariables *LV = getAnalysisIfAvailable<LiveVariables>();
if (LV) {
MachineInstr *PHICopy = prior(AfterPHIsIt);
if (IncomingReg) {
LiveVariables::VarInfo &VI = LV->getVarInfo(IncomingReg);
// Increment use count of the newly created virtual register.
VI.NumUses++;
LV->setPHIJoin(IncomingReg);
// When we are reusing the incoming register, it may already have been
// killed in this block. The old kill will also have been inserted at
// AfterPHIsIt, so it appears before the current PHICopy.
if (reusedIncoming)
if (MachineInstr *OldKill = VI.findKill(&MBB)) {
DEBUG(dbgs() << "Remove old kill from " << *OldKill);
LV->removeVirtualRegisterKilled(IncomingReg, OldKill);
DEBUG(MBB.dump());
}
// Add information to LiveVariables to know that the incoming value is
// killed. Note that because the value is defined in several places (once
// each for each incoming block), the "def" block and instruction fields
// for the VarInfo is not filled in.
LV->addVirtualRegisterKilled(IncomingReg, PHICopy);
}
// Since we are going to be deleting the PHI node, if it is the last use of
// any registers, or if the value itself is dead, we need to move this
// information over to the new copy we just inserted.
LV->removeVirtualRegistersKilled(MPhi);
// If the result is dead, update LV.
if (isDead) {
LV->addVirtualRegisterDead(DestReg, PHICopy);
LV->removeVirtualRegisterDead(DestReg, MPhi);
}
}
// Adjust the VRegPHIUseCount map to account for the removal of this PHI node.
for (unsigned i = 1; i != MPhi->getNumOperands(); i += 2)
--VRegPHIUseCount[BBVRegPair(MPhi->getOperand(i+1).getMBB()->getNumber(),
MPhi->getOperand(i).getReg())];
// Now loop over all of the incoming arguments, changing them to copy into the
// IncomingReg register in the corresponding predecessor basic block.
SmallPtrSet<MachineBasicBlock*, 8> MBBsInsertedInto;
for (int i = NumSrcs - 1; i >= 0; --i) {
unsigned SrcReg = MPhi->getOperand(i*2+1).getReg();
unsigned SrcSubReg = MPhi->getOperand(i*2+1).getSubReg();
assert(TargetRegisterInfo::isVirtualRegister(SrcReg) &&
"Machine PHI Operands must all be virtual registers!");
// Get the MachineBasicBlock equivalent of the BasicBlock that is the source
// path the PHI.
MachineBasicBlock &opBlock = *MPhi->getOperand(i*2+2).getMBB();
// If source is defined by an implicit def, there is no need to insert a
// copy.
MachineInstr *DefMI = MRI->getVRegDef(SrcReg);
if (DefMI->isImplicitDef()) {
ImpDefs.insert(DefMI);
continue;
}
// Check to make sure we haven't already emitted the copy for this block.
// This can happen because PHI nodes may have multiple entries for the same
// basic block.
if (!MBBsInsertedInto.insert(&opBlock))
continue; // If the copy has already been emitted, we're done.
// Find a safe location to insert the copy, this may be the first terminator
// in the block (or end()).
MachineBasicBlock::iterator InsertPos =
findPHICopyInsertPoint(&opBlock, &MBB, SrcReg);
// Insert the copy.
if (!reusedIncoming && IncomingReg)
BuildMI(opBlock, InsertPos, MPhi->getDebugLoc(),
TII->get(TargetOpcode::COPY), IncomingReg).addReg(SrcReg, 0, SrcSubReg);
// Now update live variable information if we have it. Otherwise we're done
if (!LV) continue;
// We want to be able to insert a kill of the register if this PHI (aka, the
// copy we just inserted) is the last use of the source value. Live
// variable analysis conservatively handles this by saying that the value is
// live until the end of the block the PHI entry lives in. If the value
// really is dead at the PHI copy, there will be no successor blocks which
// have the value live-in.
// Also check to see if this register is in use by another PHI node which
// has not yet been eliminated. If so, it will be killed at an appropriate
// point later.
// Is it used by any PHI instructions in this block?
bool ValueIsUsed = VRegPHIUseCount[BBVRegPair(opBlock.getNumber(), SrcReg)];
// Okay, if we now know that the value is not live out of the block, we can
// add a kill marker in this block saying that it kills the incoming value!
if (!ValueIsUsed && !LV->isLiveOut(SrcReg, opBlock)) {
// In our final twist, we have to decide which instruction kills the
// register. In most cases this is the copy, however, the first
// terminator instruction at the end of the block may also use the value.
// In this case, we should mark *it* as being the killing block, not the
// copy.
MachineBasicBlock::iterator KillInst;
MachineBasicBlock::iterator Term = opBlock.getFirstTerminator();
if (Term != opBlock.end() && Term->readsRegister(SrcReg)) {
KillInst = Term;
// Check that no other terminators use values.
#ifndef NDEBUG
for (MachineBasicBlock::iterator TI = llvm::next(Term);
TI != opBlock.end(); ++TI) {
assert(!TI->readsRegister(SrcReg) &&
"Terminator instructions cannot use virtual registers unless"
"they are the first terminator in a block!");
}
#endif
} else if (reusedIncoming || !IncomingReg) {
// We may have to rewind a bit if we didn't insert a copy this time.
KillInst = Term;
while (KillInst != opBlock.begin())
if ((--KillInst)->readsRegister(SrcReg))
break;
} else {
// We just inserted this copy.
KillInst = prior(InsertPos);
}
assert(KillInst->readsRegister(SrcReg) && "Cannot find kill instruction");
// Finally, mark it killed.
LV->addVirtualRegisterKilled(SrcReg, KillInst);
// This vreg no longer lives all of the way through opBlock.
unsigned opBlockNum = opBlock.getNumber();
LV->getVarInfo(SrcReg).AliveBlocks.reset(opBlockNum);
}
}
// Really delete the PHI instruction now, if it is not in the LoweredPHIs map.
if (reusedIncoming || !IncomingReg)
MF.DeleteMachineInstr(MPhi);
}
bool AArch64FrameLowering::spillCalleeSavedRegisters(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
DebugLoc DL;
SmallVector<RegPairInfo, 8> RegPairs;
computeCalleeSaveRegisterPairs(MF, CSI, TRI, RegPairs);
for (auto RPII = RegPairs.rbegin(), RPIE = RegPairs.rend(); RPII != RPIE;
++RPII) {
RegPairInfo RPI = *RPII;
unsigned Reg1 = RPI.Reg1;
unsigned Reg2 = RPI.Reg2;
unsigned StrOpc;
// Issue sequence of non-sp increment and pi sp spills for cs regs. The
// first spill is a pre-increment that allocates the stack.
// For example:
// stp x22, x21, [sp, #-48]! // addImm(-6)
// stp x20, x19, [sp, #16] // addImm(+2)
// stp fp, lr, [sp, #32] // addImm(+4)
// Rationale: This sequence saves uop updates compared to a sequence of
// pre-increment spills like stp xi,xj,[sp,#-16]!
// Note: Similar rationale and sequence for restores in epilog.
bool BumpSP = RPII == RegPairs.rbegin();
if (RPI.IsGPR) {
// For first spill use pre-increment store.
if (BumpSP)
StrOpc = RPI.isPaired() ? AArch64::STPXpre : AArch64::STRXpre;
else
StrOpc = RPI.isPaired() ? AArch64::STPXi : AArch64::STRXui;
} else {
// For first spill use pre-increment store.
if (BumpSP)
StrOpc = RPI.isPaired() ? AArch64::STPDpre : AArch64::STRDpre;
else
StrOpc = RPI.isPaired() ? AArch64::STPDi : AArch64::STRDui;
}
DEBUG(dbgs() << "CSR spill: (" << TRI->getName(Reg1);
if (RPI.isPaired())
dbgs() << ", " << TRI->getName(Reg2);
dbgs() << ") -> fi#(" << RPI.FrameIdx;
if (RPI.isPaired())
dbgs() << ", " << RPI.FrameIdx+1;
dbgs() << ")\n");
const int Offset = BumpSP ? -RPI.Offset : RPI.Offset;
MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(StrOpc));
if (BumpSP)
MIB.addReg(AArch64::SP, RegState::Define);
if (RPI.isPaired()) {
MBB.addLiveIn(Reg1);
MBB.addLiveIn(Reg2);
MIB.addReg(Reg2, getPrologueDeath(MF, Reg2))
.addReg(Reg1, getPrologueDeath(MF, Reg1))
.addReg(AArch64::SP)
.addImm(Offset) // [sp, #offset * 8], where factor * 8 is implicit
.setMIFlag(MachineInstr::FrameSetup);
} else {
MBB.addLiveIn(Reg1);
MIB.addReg(Reg1, getPrologueDeath(MF, Reg1))
.addReg(AArch64::SP)
.addImm(BumpSP ? Offset * 8 : Offset) // pre-inc version is unscaled
.setMIFlag(MachineInstr::FrameSetup);
}
}
return true;
}
/// Splits a MachineBasicBlock to branch before \p SplitBefore. The original
/// branch is \p OrigBranch. The target of the new branch can either be the same
/// as the target of the original branch or the fallthrough successor of the
/// original block as determined by \p BranchToFallThrough. The branch
/// conditions will be inverted according to \p InvertNewBranch and
/// \p InvertOrigBranch. If an instruction that previously fed the branch is to
/// be deleted, it is provided in \p MIToDelete and \p NewCond will be used as
/// the branch condition. The branch probabilities will be set if the
/// MachineBranchProbabilityInfo isn't null.
static bool splitMBB(BlockSplitInfo &BSI) {
assert(BSI.allInstrsInSameMBB() &&
"All instructions must be in the same block.");
MachineBasicBlock *ThisMBB = BSI.OrigBranch->getParent();
MachineFunction *MF = ThisMBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
assert(MRI->isSSA() && "Can only do this while the function is in SSA form.");
if (ThisMBB->succ_size() != 2) {
LLVM_DEBUG(
dbgs() << "Don't know how to handle blocks that don't have exactly"
<< " two succesors.\n");
return false;
}
const PPCInstrInfo *TII = MF->getSubtarget<PPCSubtarget>().getInstrInfo();
unsigned OrigBROpcode = BSI.OrigBranch->getOpcode();
unsigned InvertedOpcode =
OrigBROpcode == PPC::BC
? PPC::BCn
: OrigBROpcode == PPC::BCn
? PPC::BC
: OrigBROpcode == PPC::BCLR ? PPC::BCLRn : PPC::BCLR;
unsigned NewBROpcode = BSI.InvertNewBranch ? InvertedOpcode : OrigBROpcode;
MachineBasicBlock *OrigTarget = BSI.OrigBranch->getOperand(1).getMBB();
MachineBasicBlock *OrigFallThrough = OrigTarget == *ThisMBB->succ_begin()
? *ThisMBB->succ_rbegin()
: *ThisMBB->succ_begin();
MachineBasicBlock *NewBRTarget =
BSI.BranchToFallThrough ? OrigFallThrough : OrigTarget;
BranchProbability ProbToNewTarget =
!BSI.MBPI ? BranchProbability::getUnknown()
: BSI.MBPI->getEdgeProbability(ThisMBB, NewBRTarget);
// Create a new basic block.
MachineBasicBlock::iterator InsertPoint = BSI.SplitBefore;
const BasicBlock *LLVM_BB = ThisMBB->getBasicBlock();
MachineFunction::iterator It = ThisMBB->getIterator();
MachineBasicBlock *NewMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MF->insert(++It, NewMBB);
// Move everything after SplitBefore into the new block.
NewMBB->splice(NewMBB->end(), ThisMBB, InsertPoint, ThisMBB->end());
NewMBB->transferSuccessors(ThisMBB);
// Add the two successors to ThisMBB. The probabilities come from the
// existing blocks if available.
ThisMBB->addSuccessor(NewBRTarget, ProbToNewTarget);
ThisMBB->addSuccessor(NewMBB, ProbToNewTarget.getCompl());
// Add the branches to ThisMBB.
BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
TII->get(NewBROpcode))
.addReg(BSI.SplitCond->getOperand(0).getReg())
.addMBB(NewBRTarget);
BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
TII->get(PPC::B))
.addMBB(NewMBB);
if (BSI.MIToDelete)
BSI.MIToDelete->eraseFromParent();
// Change the condition on the original branch and invert it if requested.
auto FirstTerminator = NewMBB->getFirstTerminator();
if (BSI.NewCond) {
assert(FirstTerminator->getOperand(0).isReg() &&
"Can't update condition of unconditional branch.");
FirstTerminator->getOperand(0).setReg(BSI.NewCond->getOperand(0).getReg());
}
if (BSI.InvertOrigBranch)
FirstTerminator->setDesc(TII->get(InvertedOpcode));
// If any of the PHIs in the successors of NewMBB reference values that
// now come from NewMBB, they need to be updated.
for (auto *Succ : NewMBB->successors()) {
updatePHIs(Succ, ThisMBB, NewMBB, MRI);
}
addIncomingValuesToPHIs(NewBRTarget, ThisMBB, NewMBB, MRI);
LLVM_DEBUG(dbgs() << "After splitting, ThisMBB:\n"; ThisMBB->dump());
LLVM_DEBUG(dbgs() << "NewMBB:\n"; NewMBB->dump());
LLVM_DEBUG(dbgs() << "New branch-to block:\n"; NewBRTarget->dump());
return true;
}
/// foldMemoryOperand - Attempt to fold a load or store of the specified stack
/// slot into the specified machine instruction for the specified operand(s).
/// If this is possible, a new instruction is returned with the specified
/// operand folded, otherwise NULL is returned. The client is responsible for
/// removing the old instruction and adding the new one in the instruction
/// stream.
MachineInstr*
TargetInstrInfo::foldMemoryOperand(MachineBasicBlock::iterator MI,
const SmallVectorImpl<unsigned> &Ops,
int FI) const {
unsigned Flags = 0;
for (unsigned i = 0, e = Ops.size(); i != e; ++i)
if (MI->getOperand(Ops[i]).isDef())
Flags |= MachineMemOperand::MOStore;
else
Flags |= MachineMemOperand::MOLoad;
MachineBasicBlock *MBB = MI->getParent();
assert(MBB && "foldMemoryOperand needs an inserted instruction");
MachineFunction &MF = *MBB->getParent();
MachineInstr *NewMI = nullptr;
if (MI->getOpcode() == TargetOpcode::STACKMAP ||
MI->getOpcode() == TargetOpcode::PATCHPOINT) {
// Fold stackmap/patchpoint.
NewMI = foldPatchpoint(MF, MI, Ops, FI, *this);
} else {
// Ask the target to do the actual folding.
NewMI =foldMemoryOperandImpl(MF, MI, Ops, FI);
}
if (NewMI) {
NewMI->setMemRefs(MI->memoperands_begin(), MI->memoperands_end());
// Add a memory operand, foldMemoryOperandImpl doesn't do that.
assert((!(Flags & MachineMemOperand::MOStore) ||
NewMI->mayStore()) &&
"Folded a def to a non-store!");
assert((!(Flags & MachineMemOperand::MOLoad) ||
NewMI->mayLoad()) &&
"Folded a use to a non-load!");
const MachineFrameInfo &MFI = *MF.getFrameInfo();
assert(MFI.getObjectOffset(FI) != -1);
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FI),
Flags, MFI.getObjectSize(FI),
MFI.getObjectAlignment(FI));
NewMI->addMemOperand(MF, MMO);
// FIXME: change foldMemoryOperandImpl semantics to also insert NewMI.
return MBB->insert(MI, NewMI);
}
// Straight COPY may fold as load/store.
if (!MI->isCopy() || Ops.size() != 1)
return nullptr;
const TargetRegisterClass *RC = canFoldCopy(MI, Ops[0]);
if (!RC)
return nullptr;
const MachineOperand &MO = MI->getOperand(1-Ops[0]);
MachineBasicBlock::iterator Pos = MI;
const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
if (Flags == MachineMemOperand::MOStore)
storeRegToStackSlot(*MBB, Pos, MO.getReg(), MO.isKill(), FI, RC, TRI);
else
loadRegFromStackSlot(*MBB, Pos, MO.getReg(), FI, RC, TRI);
return --Pos;
}
/// LowerPHINode - Lower the PHI node at the top of the specified block,
///
void PHIElimination::LowerPHINode(MachineBasicBlock &MBB,
MachineBasicBlock::iterator LastPHIIt) {
++NumLowered;
MachineBasicBlock::iterator AfterPHIsIt = llvm::next(LastPHIIt);
// Unlink the PHI node from the basic block, but don't delete the PHI yet.
MachineInstr *MPhi = MBB.remove(MBB.begin());
unsigned NumSrcs = (MPhi->getNumOperands() - 1) / 2;
unsigned DestReg = MPhi->getOperand(0).getReg();
assert(MPhi->getOperand(0).getSubReg() == 0 && "Can't handle sub-reg PHIs");
bool isDead = MPhi->getOperand(0).isDead();
// Create a new register for the incoming PHI arguments.
MachineFunction &MF = *MBB.getParent();
unsigned IncomingReg = 0;
bool reusedIncoming = false; // Is IncomingReg reused from an earlier PHI?
// Insert a register to register copy at the top of the current block (but
// after any remaining phi nodes) which copies the new incoming register
// into the phi node destination.
const TargetInstrInfo *TII = MF.getTarget().getInstrInfo();
if (isSourceDefinedByImplicitDef(MPhi, MRI))
// If all sources of a PHI node are implicit_def, just emit an
// implicit_def instead of a copy.
BuildMI(MBB, AfterPHIsIt, MPhi->getDebugLoc(),
TII->get(TargetOpcode::IMPLICIT_DEF), DestReg);
else {
// Can we reuse an earlier PHI node? This only happens for critical edges,
// typically those created by tail duplication.
unsigned &entry = LoweredPHIs[MPhi];
if (entry) {
// An identical PHI node was already lowered. Reuse the incoming register.
IncomingReg = entry;
reusedIncoming = true;
++NumReused;
DEBUG(dbgs() << "Reusing " << PrintReg(IncomingReg) << " for " << *MPhi);
} else {
const TargetRegisterClass *RC = MF.getRegInfo().getRegClass(DestReg);
entry = IncomingReg = MF.getRegInfo().createVirtualRegister(RC);
}
BuildMI(MBB, AfterPHIsIt, MPhi->getDebugLoc(),
TII->get(TargetOpcode::COPY), DestReg)
.addReg(IncomingReg);
}
// Update live variable information if there is any.
if (LV) {
MachineInstr *PHICopy = prior(AfterPHIsIt);
if (IncomingReg) {
LiveVariables::VarInfo &VI = LV->getVarInfo(IncomingReg);
// Increment use count of the newly created virtual register.
LV->setPHIJoin(IncomingReg);
// When we are reusing the incoming register, it may already have been
// killed in this block. The old kill will also have been inserted at
// AfterPHIsIt, so it appears before the current PHICopy.
if (reusedIncoming)
if (MachineInstr *OldKill = VI.findKill(&MBB)) {
DEBUG(dbgs() << "Remove old kill from " << *OldKill);
LV->removeVirtualRegisterKilled(IncomingReg, OldKill);
DEBUG(MBB.dump());
}
// Add information to LiveVariables to know that the incoming value is
// killed. Note that because the value is defined in several places (once
// each for each incoming block), the "def" block and instruction fields
// for the VarInfo is not filled in.
LV->addVirtualRegisterKilled(IncomingReg, PHICopy);
}
// Since we are going to be deleting the PHI node, if it is the last use of
// any registers, or if the value itself is dead, we need to move this
// information over to the new copy we just inserted.
LV->removeVirtualRegistersKilled(MPhi);
// If the result is dead, update LV.
if (isDead) {
LV->addVirtualRegisterDead(DestReg, PHICopy);
LV->removeVirtualRegisterDead(DestReg, MPhi);
}
}
// Update LiveIntervals for the new copy or implicit def.
if (LIS) {
MachineInstr *NewInstr = prior(AfterPHIsIt);
SlotIndex DestCopyIndex = LIS->InsertMachineInstrInMaps(NewInstr);
SlotIndex MBBStartIndex = LIS->getMBBStartIdx(&MBB);
if (IncomingReg) {
// Add the region from the beginning of MBB to the copy instruction to
// IncomingReg's live interval.
LiveInterval &IncomingLI = LIS->createEmptyInterval(IncomingReg);
VNInfo *IncomingVNI = IncomingLI.getVNInfoAt(MBBStartIndex);
if (!IncomingVNI)
IncomingVNI = IncomingLI.getNextValue(MBBStartIndex,
LIS->getVNInfoAllocator());
IncomingLI.addSegment(LiveInterval::Segment(MBBStartIndex,
DestCopyIndex.getRegSlot(),
IncomingVNI));
}
LiveInterval &DestLI = LIS->getInterval(DestReg);
assert(DestLI.begin() != DestLI.end() &&
"PHIs should have nonempty LiveIntervals.");
if (DestLI.endIndex().isDead()) {
// A dead PHI's live range begins and ends at the start of the MBB, but
// the lowered copy, which will still be dead, needs to begin and end at
// the copy instruction.
VNInfo *OrigDestVNI = DestLI.getVNInfoAt(MBBStartIndex);
assert(OrigDestVNI && "PHI destination should be live at block entry.");
DestLI.removeSegment(MBBStartIndex, MBBStartIndex.getDeadSlot());
DestLI.createDeadDef(DestCopyIndex.getRegSlot(),
LIS->getVNInfoAllocator());
DestLI.removeValNo(OrigDestVNI);
} else {
// Otherwise, remove the region from the beginning of MBB to the copy
// instruction from DestReg's live interval.
DestLI.removeSegment(MBBStartIndex, DestCopyIndex.getRegSlot());
VNInfo *DestVNI = DestLI.getVNInfoAt(DestCopyIndex.getRegSlot());
assert(DestVNI && "PHI destination should be live at its definition.");
DestVNI->def = DestCopyIndex.getRegSlot();
}
}
// Adjust the VRegPHIUseCount map to account for the removal of this PHI node.
for (unsigned i = 1; i != MPhi->getNumOperands(); i += 2)
--VRegPHIUseCount[BBVRegPair(MPhi->getOperand(i+1).getMBB()->getNumber(),
MPhi->getOperand(i).getReg())];
// Now loop over all of the incoming arguments, changing them to copy into the
// IncomingReg register in the corresponding predecessor basic block.
SmallPtrSet<MachineBasicBlock*, 8> MBBsInsertedInto;
for (int i = NumSrcs - 1; i >= 0; --i) {
unsigned SrcReg = MPhi->getOperand(i*2+1).getReg();
unsigned SrcSubReg = MPhi->getOperand(i*2+1).getSubReg();
bool SrcUndef = MPhi->getOperand(i*2+1).isUndef() ||
isImplicitlyDefined(SrcReg, MRI);
assert(TargetRegisterInfo::isVirtualRegister(SrcReg) &&
"Machine PHI Operands must all be virtual registers!");
// Get the MachineBasicBlock equivalent of the BasicBlock that is the source
// path the PHI.
MachineBasicBlock &opBlock = *MPhi->getOperand(i*2+2).getMBB();
// Check to make sure we haven't already emitted the copy for this block.
// This can happen because PHI nodes may have multiple entries for the same
// basic block.
if (!MBBsInsertedInto.insert(&opBlock))
continue; // If the copy has already been emitted, we're done.
// Find a safe location to insert the copy, this may be the first terminator
// in the block (or end()).
MachineBasicBlock::iterator InsertPos =
findPHICopyInsertPoint(&opBlock, &MBB, SrcReg);
// Insert the copy.
MachineInstr *NewSrcInstr = 0;
if (!reusedIncoming && IncomingReg) {
if (SrcUndef) {
// The source register is undefined, so there is no need for a real
// COPY, but we still need to ensure joint dominance by defs.
// Insert an IMPLICIT_DEF instruction.
NewSrcInstr = BuildMI(opBlock, InsertPos, MPhi->getDebugLoc(),
TII->get(TargetOpcode::IMPLICIT_DEF),
IncomingReg);
// Clean up the old implicit-def, if there even was one.
if (MachineInstr *DefMI = MRI->getVRegDef(SrcReg))
if (DefMI->isImplicitDef())
ImpDefs.insert(DefMI);
} else {
NewSrcInstr = BuildMI(opBlock, InsertPos, MPhi->getDebugLoc(),
TII->get(TargetOpcode::COPY), IncomingReg)
.addReg(SrcReg, 0, SrcSubReg);
}
}
// We only need to update the LiveVariables kill of SrcReg if this was the
// last PHI use of SrcReg to be lowered on this CFG edge and it is not live
// out of the predecessor. We can also ignore undef sources.
if (LV && !SrcUndef &&
!VRegPHIUseCount[BBVRegPair(opBlock.getNumber(), SrcReg)] &&
!LV->isLiveOut(SrcReg, opBlock)) {
// We want to be able to insert a kill of the register if this PHI (aka,
// the copy we just inserted) is the last use of the source value. Live
// variable analysis conservatively handles this by saying that the value
// is live until the end of the block the PHI entry lives in. If the value
// really is dead at the PHI copy, there will be no successor blocks which
// have the value live-in.
// Okay, if we now know that the value is not live out of the block, we
// can add a kill marker in this block saying that it kills the incoming
// value!
// In our final twist, we have to decide which instruction kills the
// register. In most cases this is the copy, however, terminator
// instructions at the end of the block may also use the value. In this
// case, we should mark the last such terminator as being the killing
// block, not the copy.
MachineBasicBlock::iterator KillInst = opBlock.end();
MachineBasicBlock::iterator FirstTerm = opBlock.getFirstTerminator();
for (MachineBasicBlock::iterator Term = FirstTerm;
Term != opBlock.end(); ++Term) {
if (Term->readsRegister(SrcReg))
KillInst = Term;
}
if (KillInst == opBlock.end()) {
// No terminator uses the register.
if (reusedIncoming || !IncomingReg) {
// We may have to rewind a bit if we didn't insert a copy this time.
KillInst = FirstTerm;
while (KillInst != opBlock.begin()) {
--KillInst;
if (KillInst->isDebugValue())
continue;
if (KillInst->readsRegister(SrcReg))
break;
}
} else {
// We just inserted this copy.
KillInst = prior(InsertPos);
}
}
assert(KillInst->readsRegister(SrcReg) && "Cannot find kill instruction");
// Finally, mark it killed.
LV->addVirtualRegisterKilled(SrcReg, KillInst);
// This vreg no longer lives all of the way through opBlock.
unsigned opBlockNum = opBlock.getNumber();
LV->getVarInfo(SrcReg).AliveBlocks.reset(opBlockNum);
}
if (LIS) {
if (NewSrcInstr) {
LIS->InsertMachineInstrInMaps(NewSrcInstr);
LIS->addSegmentToEndOfBlock(IncomingReg, NewSrcInstr);
}
if (!SrcUndef &&
!VRegPHIUseCount[BBVRegPair(opBlock.getNumber(), SrcReg)]) {
LiveInterval &SrcLI = LIS->getInterval(SrcReg);
bool isLiveOut = false;
for (MachineBasicBlock::succ_iterator SI = opBlock.succ_begin(),
SE = opBlock.succ_end(); SI != SE; ++SI) {
SlotIndex startIdx = LIS->getMBBStartIdx(*SI);
VNInfo *VNI = SrcLI.getVNInfoAt(startIdx);
// Definitions by other PHIs are not truly live-in for our purposes.
if (VNI && VNI->def != startIdx) {
isLiveOut = true;
break;
}
}
if (!isLiveOut) {
MachineBasicBlock::iterator KillInst = opBlock.end();
MachineBasicBlock::iterator FirstTerm = opBlock.getFirstTerminator();
for (MachineBasicBlock::iterator Term = FirstTerm;
Term != opBlock.end(); ++Term) {
if (Term->readsRegister(SrcReg))
KillInst = Term;
}
if (KillInst == opBlock.end()) {
// No terminator uses the register.
if (reusedIncoming || !IncomingReg) {
// We may have to rewind a bit if we didn't just insert a copy.
KillInst = FirstTerm;
while (KillInst != opBlock.begin()) {
--KillInst;
if (KillInst->isDebugValue())
continue;
if (KillInst->readsRegister(SrcReg))
break;
}
} else {
// We just inserted this copy.
KillInst = prior(InsertPos);
}
}
assert(KillInst->readsRegister(SrcReg) &&
"Cannot find kill instruction");
SlotIndex LastUseIndex = LIS->getInstructionIndex(KillInst);
SrcLI.removeSegment(LastUseIndex.getRegSlot(),
LIS->getMBBEndIdx(&opBlock));
}
}
}
}
// Really delete the PHI instruction now, if it is not in the LoweredPHIs map.
if (reusedIncoming || !IncomingReg) {
if (LIS)
LIS->RemoveMachineInstrFromMaps(MPhi);
MF.DeleteMachineInstr(MPhi);
}
}
void
AArch64FrameLowering::emitFrameMemOps(bool isPrologue, MachineBasicBlock &MBB,
MachineBasicBlock::iterator MBBI,
const std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI,
LoadStoreMethod PossClasses[],
unsigned NumClasses) const {
DebugLoc DL = MBB.findDebugLoc(MBBI);
MachineFunction &MF = *MBB.getParent();
MachineFrameInfo &MFI = *MF.getFrameInfo();
const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
// A certain amount of implicit contract is present here. The actual stack
// offsets haven't been allocated officially yet, so for strictly correct code
// we rely on the fact that the elements of CSI are allocated in order
// starting at SP, purely as dictated by size and alignment. In practice since
// this function handles the only accesses to those slots it's not quite so
// important.
//
// We have also ordered the Callee-saved register list in AArch64CallingConv
// so that the above scheme puts registers in order: in particular we want
// &X30 to be &X29+8 for an ABI-correct frame record (PCS 5.2.2)
for (unsigned i = 0, e = CSI.size(); i < e; ++i) {
unsigned Reg = CSI[i].getReg();
// First we need to find out which register class the register belongs to so
// that we can use the correct load/store instrucitons.
unsigned ClassIdx;
for (ClassIdx = 0; ClassIdx < NumClasses; ++ClassIdx) {
if (PossClasses[ClassIdx].RegClass->contains(Reg))
break;
}
assert(ClassIdx != NumClasses
&& "Asked to store register in unexpected class");
const TargetRegisterClass &TheClass = *PossClasses[ClassIdx].RegClass;
// Now we need to decide whether it's possible to emit a paired instruction:
// for this we want the next register to be in the same class.
MachineInstrBuilder NewMI;
bool Pair = false;
if (i + 1 < CSI.size() && TheClass.contains(CSI[i+1].getReg())) {
Pair = true;
unsigned StLow = 0, StHigh = 0;
if (isPrologue) {
// Most of these registers will be live-in to the MBB and killed by our
// store, though there are exceptions (see determinePrologueDeath).
StLow = getKillRegState(determinePrologueDeath(MBB, CSI[i+1].getReg()));
StHigh = getKillRegState(determinePrologueDeath(MBB, CSI[i].getReg()));
} else {
StLow = RegState::Define;
StHigh = RegState::Define;
}
NewMI = BuildMI(MBB, MBBI, DL, TII.get(PossClasses[ClassIdx].PairOpcode))
.addReg(CSI[i+1].getReg(), StLow)
.addReg(CSI[i].getReg(), StHigh);
// If it's a paired op, we've consumed two registers
++i;
} else {
unsigned State;
if (isPrologue) {
State = getKillRegState(determinePrologueDeath(MBB, CSI[i].getReg()));
} else {
State = RegState::Define;
}
NewMI = BuildMI(MBB, MBBI, DL,
TII.get(PossClasses[ClassIdx].SingleOpcode))
.addReg(CSI[i].getReg(), State);
}
// Note that the FrameIdx refers to the second register in a pair: it will
// be allocated the smaller numeric address and so is the one an LDP/STP
// address must use.
int FrameIdx = CSI[i].getFrameIdx();
MachineMemOperand::MemOperandFlags Flags;
Flags = isPrologue ? MachineMemOperand::MOStore : MachineMemOperand::MOLoad;
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(FrameIdx),
Flags,
Pair ? TheClass.getSize() * 2 : TheClass.getSize(),
MFI.getObjectAlignment(FrameIdx));
NewMI.addFrameIndex(FrameIdx)
.addImm(0) // address-register offset
.addMemOperand(MMO);
if (isPrologue)
NewMI.setMIFlags(MachineInstr::FrameSetup);
// For aesthetic reasons, during an epilogue we want to emit complementary
// operations to the prologue, but in the opposite order. So we still
// iterate through the CalleeSavedInfo list in order, but we put the
// instructions successively earlier in the MBB.
if (!isPrologue)
--MBBI;
}
}
bool Thumb1FrameLowering::emitPopSpecialFixUp(MachineBasicBlock &MBB,
bool DoIt) const {
MachineFunction &MF = *MBB.getParent();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ArgRegsSaveSize = AFI->getArgRegsSaveSize();
const TargetInstrInfo &TII = *STI.getInstrInfo();
const ThumbRegisterInfo *RegInfo =
static_cast<const ThumbRegisterInfo *>(STI.getRegisterInfo());
// If MBBI is a return instruction, or is a tPOP followed by a return
// instruction in the successor BB, we may be able to directly restore
// LR in the PC.
// This is only possible with v5T ops (v4T can't change the Thumb bit via
// a POP PC instruction), and only if we do not need to emit any SP update.
// Otherwise, we need a temporary register to pop the value
// and copy that value into LR.
auto MBBI = MBB.getFirstTerminator();
bool CanRestoreDirectly = STI.hasV5TOps() && !ArgRegsSaveSize;
if (CanRestoreDirectly) {
if (MBBI != MBB.end())
CanRestoreDirectly = (MBBI->getOpcode() == ARM::tBX_RET ||
MBBI->getOpcode() == ARM::tPOP_RET);
else {
assert(MBB.back().getOpcode() == ARM::tPOP);
assert(MBB.succ_size() == 1);
if ((*MBB.succ_begin())->begin()->getOpcode() == ARM::tBX_RET)
MBBI--; // Replace the final tPOP with a tPOP_RET.
else
CanRestoreDirectly = false;
}
}
if (CanRestoreDirectly) {
if (!DoIt || MBBI->getOpcode() == ARM::tPOP_RET)
return true;
MachineInstrBuilder MIB =
AddDefaultPred(
BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII.get(ARM::tPOP_RET)));
// Copy implicit ops and popped registers, if any.
for (auto MO: MBBI->operands())
if (MO.isReg() && (MO.isImplicit() || MO.isDef()) &&
MO.getReg() != ARM::LR)
MIB.addOperand(MO);
MIB.addReg(ARM::PC, RegState::Define);
// Erase the old instruction (tBX_RET or tPOP).
MBB.erase(MBBI);
return true;
}
// Look for a temporary register to use.
// First, compute the liveness information.
LivePhysRegs UsedRegs(STI.getRegisterInfo());
UsedRegs.addLiveOuts(&MBB, /*AddPristines*/ true);
// The semantic of pristines changed recently and now,
// the callee-saved registers that are touched in the function
// are not part of the pristines set anymore.
// Add those callee-saved now.
const TargetRegisterInfo *TRI = STI.getRegisterInfo();
const MCPhysReg *CSRegs = TRI->getCalleeSavedRegs(&MF);
for (unsigned i = 0; CSRegs[i]; ++i)
UsedRegs.addReg(CSRegs[i]);
DebugLoc dl = DebugLoc();
if (MBBI != MBB.end()) {
dl = MBBI->getDebugLoc();
auto InstUpToMBBI = MBB.end();
while (InstUpToMBBI != MBBI)
// The pre-decrement is on purpose here.
// We want to have the liveness right before MBBI.
UsedRegs.stepBackward(*--InstUpToMBBI);
}
// Look for a register that can be directly use in the POP.
unsigned PopReg = 0;
// And some temporary register, just in case.
unsigned TemporaryReg = 0;
BitVector PopFriendly =
TRI->getAllocatableSet(MF, TRI->getRegClass(ARM::tGPRRegClassID));
assert(PopFriendly.any() && "No allocatable pop-friendly register?!");
// Rebuild the GPRs from the high registers because they are removed
// form the GPR reg class for thumb1.
BitVector GPRsNoLRSP =
TRI->getAllocatableSet(MF, TRI->getRegClass(ARM::hGPRRegClassID));
GPRsNoLRSP |= PopFriendly;
GPRsNoLRSP.reset(ARM::LR);
GPRsNoLRSP.reset(ARM::SP);
GPRsNoLRSP.reset(ARM::PC);
for (int Register = GPRsNoLRSP.find_first(); Register != -1;
Register = GPRsNoLRSP.find_next(Register)) {
if (!UsedRegs.contains(Register)) {
// Remember the first pop-friendly register and exit.
if (PopFriendly.test(Register)) {
PopReg = Register;
TemporaryReg = 0;
break;
}
// Otherwise, remember that the register will be available to
// save a pop-friendly register.
TemporaryReg = Register;
}
}
if (!DoIt && !PopReg && !TemporaryReg)
return false;
assert((PopReg || TemporaryReg) && "Cannot get LR");
if (TemporaryReg) {
assert(!PopReg && "Unnecessary MOV is about to be inserted");
PopReg = PopFriendly.find_first();
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(TemporaryReg, RegState::Define)
.addReg(PopReg, RegState::Kill));
}
if (MBBI == MBB.end()) {
MachineInstr& Pop = MBB.back();
assert(Pop.getOpcode() == ARM::tPOP);
Pop.RemoveOperand(Pop.findRegisterDefOperandIdx(ARM::LR));
}
assert(PopReg && "Do not know how to get LR");
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tPOP)))
.addReg(PopReg, RegState::Define);
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, ArgRegsSaveSize);
if (!TemporaryReg && MBBI != MBB.end() && MBBI->getOpcode() == ARM::tBX_RET) {
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(ARM::tBX))
.addReg(PopReg, RegState::Kill);
AddDefaultPred(MIB);
MIB.copyImplicitOps(&*MBBI);
// erase the old tBX_RET instruction
MBB.erase(MBBI);
return true;
}
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(ARM::LR, RegState::Define)
.addReg(PopReg, RegState::Kill));
if (TemporaryReg) {
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(PopReg, RegState::Define)
.addReg(TemporaryReg, RegState::Kill));
}
return true;
}
bool MipsExpandPseudo::expandAtomicBinOpSubword(
MachineBasicBlock &BB, MachineBasicBlock::iterator I,
MachineBasicBlock::iterator &NMBBI) {
MachineFunction *MF = BB.getParent();
const bool ArePtrs64bit = STI->getABI().ArePtrs64bit();
DebugLoc DL = I->getDebugLoc();
unsigned LL, SC;
unsigned BEQ = Mips::BEQ;
unsigned SEOp = Mips::SEH;
if (STI->inMicroMipsMode()) {
LL = STI->hasMips32r6() ? Mips::LL_MMR6 : Mips::LL_MM;
SC = STI->hasMips32r6() ? Mips::SC_MMR6 : Mips::SC_MM;
BEQ = STI->hasMips32r6() ? Mips::BEQC_MMR6 : Mips::BEQ_MM;
} else {
LL = STI->hasMips32r6() ? (ArePtrs64bit ? Mips::LL64_R6 : Mips::LL_R6)
: (ArePtrs64bit ? Mips::LL64 : Mips::LL);
SC = STI->hasMips32r6() ? (ArePtrs64bit ? Mips::SC64_R6 : Mips::SC_R6)
: (ArePtrs64bit ? Mips::SC64 : Mips::SC);
}
bool IsSwap = false;
bool IsNand = false;
unsigned Opcode = 0;
switch (I->getOpcode()) {
case Mips::ATOMIC_LOAD_NAND_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_NAND_I16_POSTRA:
IsNand = true;
break;
case Mips::ATOMIC_SWAP_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_SWAP_I16_POSTRA:
IsSwap = true;
break;
case Mips::ATOMIC_LOAD_ADD_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_ADD_I16_POSTRA:
Opcode = Mips::ADDu;
break;
case Mips::ATOMIC_LOAD_SUB_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_SUB_I16_POSTRA:
Opcode = Mips::SUBu;
break;
case Mips::ATOMIC_LOAD_AND_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_AND_I16_POSTRA:
Opcode = Mips::AND;
break;
case Mips::ATOMIC_LOAD_OR_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_OR_I16_POSTRA:
Opcode = Mips::OR;
break;
case Mips::ATOMIC_LOAD_XOR_I8_POSTRA:
SEOp = Mips::SEB;
LLVM_FALLTHROUGH;
case Mips::ATOMIC_LOAD_XOR_I16_POSTRA:
Opcode = Mips::XOR;
break;
default:
llvm_unreachable("Unknown subword atomic pseudo for expansion!");
}
unsigned Dest = I->getOperand(0).getReg();
unsigned Ptr = I->getOperand(1).getReg();
unsigned Incr = I->getOperand(2).getReg();
unsigned Mask = I->getOperand(3).getReg();
unsigned Mask2 = I->getOperand(4).getReg();
unsigned ShiftAmnt = I->getOperand(5).getReg();
unsigned OldVal = I->getOperand(6).getReg();
unsigned BinOpRes = I->getOperand(7).getReg();
unsigned StoreVal = I->getOperand(8).getReg();
const BasicBlock *LLVM_BB = BB.getBasicBlock();
MachineBasicBlock *loopMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB.getIterator();
MF->insert(It, loopMBB);
MF->insert(It, sinkMBB);
MF->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), &BB, std::next(I), BB.end());
exitMBB->transferSuccessorsAndUpdatePHIs(&BB);
BB.addSuccessor(loopMBB, BranchProbability::getOne());
loopMBB->addSuccessor(sinkMBB);
loopMBB->addSuccessor(loopMBB);
loopMBB->normalizeSuccProbs();
BuildMI(loopMBB, DL, TII->get(LL), OldVal).addReg(Ptr).addImm(0);
if (IsNand) {
// and andres, oldval, incr2
// nor binopres, $0, andres
// and newval, binopres, mask
BuildMI(loopMBB, DL, TII->get(Mips::AND), BinOpRes)
.addReg(OldVal)
.addReg(Incr);
BuildMI(loopMBB, DL, TII->get(Mips::NOR), BinOpRes)
.addReg(Mips::ZERO)
.addReg(BinOpRes);
BuildMI(loopMBB, DL, TII->get(Mips::AND), BinOpRes)
.addReg(BinOpRes)
.addReg(Mask);
} else if (!IsSwap) {
// <binop> binopres, oldval, incr2
// and newval, binopres, mask
BuildMI(loopMBB, DL, TII->get(Opcode), BinOpRes)
.addReg(OldVal)
.addReg(Incr);
BuildMI(loopMBB, DL, TII->get(Mips::AND), BinOpRes)
.addReg(BinOpRes)
.addReg(Mask);
} else { // atomic.swap
// and newval, incr2, mask
BuildMI(loopMBB, DL, TII->get(Mips::AND), BinOpRes)
.addReg(Incr)
.addReg(Mask);
}
// and StoreVal, OlddVal, Mask2
// or StoreVal, StoreVal, BinOpRes
// StoreVal<tied1> = sc StoreVal, 0(Ptr)
// beq StoreVal, zero, loopMBB
BuildMI(loopMBB, DL, TII->get(Mips::AND), StoreVal)
.addReg(OldVal).addReg(Mask2);
BuildMI(loopMBB, DL, TII->get(Mips::OR), StoreVal)
.addReg(StoreVal).addReg(BinOpRes);
BuildMI(loopMBB, DL, TII->get(SC), StoreVal)
.addReg(StoreVal).addReg(Ptr).addImm(0);
BuildMI(loopMBB, DL, TII->get(BEQ))
.addReg(StoreVal).addReg(Mips::ZERO).addMBB(loopMBB);
// sinkMBB:
// and maskedoldval1,oldval,mask
// srl srlres,maskedoldval1,shiftamt
// sign_extend dest,srlres
sinkMBB->addSuccessor(exitMBB, BranchProbability::getOne());
BuildMI(sinkMBB, DL, TII->get(Mips::AND), Dest)
.addReg(OldVal).addReg(Mask);
BuildMI(sinkMBB, DL, TII->get(Mips::SRLV), Dest)
.addReg(Dest).addReg(ShiftAmnt);
if (STI->hasMips32r2()) {
BuildMI(sinkMBB, DL, TII->get(SEOp), Dest).addReg(Dest);
} else {
const unsigned ShiftImm = SEOp == Mips::SEH ? 16 : 24;
BuildMI(sinkMBB, DL, TII->get(Mips::SLL), Dest)
.addReg(Dest, RegState::Kill)
.addImm(ShiftImm);
BuildMI(sinkMBB, DL, TII->get(Mips::SRA), Dest)
.addReg(Dest, RegState::Kill)
.addImm(ShiftImm);
}
LivePhysRegs LiveRegs;
computeAndAddLiveIns(LiveRegs, *loopMBB);
computeAndAddLiveIns(LiveRegs, *sinkMBB);
computeAndAddLiveIns(LiveRegs, *exitMBB);
NMBBI = BB.end();
I->eraseFromParent();
return true;
}
bool AArch64FrameLowering::spillCalleeSavedRegisters(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
unsigned Count = CSI.size();
DebugLoc DL;
assert((Count & 1) == 0 && "Odd number of callee-saved regs to spill!");
if (MI != MBB.end())
DL = MI->getDebugLoc();
for (unsigned i = 0; i < Count; i += 2) {
unsigned idx = Count - i - 2;
unsigned Reg1 = CSI[idx].getReg();
unsigned Reg2 = CSI[idx + 1].getReg();
// GPRs and FPRs are saved in pairs of 64-bit regs. We expect the CSI
// list to come in sorted by frame index so that we can issue the store
// pair instructions directly. Assert if we see anything otherwise.
//
// The order of the registers in the list is controlled by
// getCalleeSavedRegs(), so they will always be in-order, as well.
assert(CSI[idx].getFrameIdx() + 1 == CSI[idx + 1].getFrameIdx() &&
"Out of order callee saved regs!");
unsigned StrOpc;
assert((Count & 1) == 0 && "Odd number of callee-saved regs to spill!");
assert((i & 1) == 0 && "Odd index for callee-saved reg spill!");
// Issue sequence of non-sp increment and pi sp spills for cs regs. The
// first spill is a pre-increment that allocates the stack.
// For example:
// stp x22, x21, [sp, #-48]! // addImm(-6)
// stp x20, x19, [sp, #16] // addImm(+2)
// stp fp, lr, [sp, #32] // addImm(+4)
// Rationale: This sequence saves uop updates compared to a sequence of
// pre-increment spills like stp xi,xj,[sp,#-16]!
// Note: Similar rational and sequence for restores in epilog.
if (AArch64::GPR64RegClass.contains(Reg1)) {
assert(AArch64::GPR64RegClass.contains(Reg2) &&
"Expected GPR64 callee-saved register pair!");
// For first spill use pre-increment store.
if (i == 0)
StrOpc = AArch64::STPXpre;
else
StrOpc = AArch64::STPXi;
} else if (AArch64::FPR64RegClass.contains(Reg1)) {
assert(AArch64::FPR64RegClass.contains(Reg2) &&
"Expected FPR64 callee-saved register pair!");
// For first spill use pre-increment store.
if (i == 0)
StrOpc = AArch64::STPDpre;
else
StrOpc = AArch64::STPDi;
} else
llvm_unreachable("Unexpected callee saved register!");
DEBUG(dbgs() << "CSR spill: (" << TRI->getName(Reg1) << ", "
<< TRI->getName(Reg2) << ") -> fi#(" << CSI[idx].getFrameIdx()
<< ", " << CSI[idx + 1].getFrameIdx() << ")\n");
// Compute offset: i = 0 => offset = -Count;
// i = 2 => offset = -(Count - 2) + Count = 2 = i; etc.
const int Offset = (i == 0) ? -Count : i;
assert((Offset >= -64 && Offset <= 63) &&
"Offset out of bounds for STP immediate");
MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(StrOpc));
if (StrOpc == AArch64::STPDpre || StrOpc == AArch64::STPXpre)
MIB.addReg(AArch64::SP, RegState::Define);
MIB.addReg(Reg2, getPrologueDeath(MF, Reg2))
.addReg(Reg1, getPrologueDeath(MF, Reg1))
.addReg(AArch64::SP)
.addImm(Offset) // [sp, #offset * 8], where factor * 8 is implicit
.setMIFlag(MachineInstr::FrameSetup);
}
return true;
}
bool MipsExpandPseudo::expandAtomicCmpSwapSubword(
MachineBasicBlock &BB, MachineBasicBlock::iterator I,
MachineBasicBlock::iterator &NMBBI) {
MachineFunction *MF = BB.getParent();
const bool ArePtrs64bit = STI->getABI().ArePtrs64bit();
DebugLoc DL = I->getDebugLoc();
unsigned LL, SC;
unsigned ZERO = Mips::ZERO;
unsigned BNE = Mips::BNE;
unsigned BEQ = Mips::BEQ;
unsigned SEOp =
I->getOpcode() == Mips::ATOMIC_CMP_SWAP_I8_POSTRA ? Mips::SEB : Mips::SEH;
if (STI->inMicroMipsMode()) {
LL = STI->hasMips32r6() ? Mips::LL_MMR6 : Mips::LL_MM;
SC = STI->hasMips32r6() ? Mips::SC_MMR6 : Mips::SC_MM;
BNE = STI->hasMips32r6() ? Mips::BNEC_MMR6 : Mips::BNE_MM;
BEQ = STI->hasMips32r6() ? Mips::BEQC_MMR6 : Mips::BEQ_MM;
} else {
LL = STI->hasMips32r6() ? (ArePtrs64bit ? Mips::LL64_R6 : Mips::LL_R6)
: (ArePtrs64bit ? Mips::LL64 : Mips::LL);
SC = STI->hasMips32r6() ? (ArePtrs64bit ? Mips::SC64_R6 : Mips::SC_R6)
: (ArePtrs64bit ? Mips::SC64 : Mips::SC);
}
unsigned Dest = I->getOperand(0).getReg();
unsigned Ptr = I->getOperand(1).getReg();
unsigned Mask = I->getOperand(2).getReg();
unsigned ShiftCmpVal = I->getOperand(3).getReg();
unsigned Mask2 = I->getOperand(4).getReg();
unsigned ShiftNewVal = I->getOperand(5).getReg();
unsigned ShiftAmnt = I->getOperand(6).getReg();
unsigned Scratch = I->getOperand(7).getReg();
unsigned Scratch2 = I->getOperand(8).getReg();
// insert new blocks after the current block
const BasicBlock *LLVM_BB = BB.getBasicBlock();
MachineBasicBlock *loop1MBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++BB.getIterator();
MF->insert(It, loop1MBB);
MF->insert(It, loop2MBB);
MF->insert(It, sinkMBB);
MF->insert(It, exitMBB);
// Transfer the remainder of BB and its successor edges to exitMBB.
exitMBB->splice(exitMBB->begin(), &BB,
std::next(MachineBasicBlock::iterator(I)), BB.end());
exitMBB->transferSuccessorsAndUpdatePHIs(&BB);
// thisMBB:
// ...
// fallthrough --> loop1MBB
BB.addSuccessor(loop1MBB, BranchProbability::getOne());
loop1MBB->addSuccessor(sinkMBB);
loop1MBB->addSuccessor(loop2MBB);
loop1MBB->normalizeSuccProbs();
loop2MBB->addSuccessor(loop1MBB);
loop2MBB->addSuccessor(sinkMBB);
loop2MBB->normalizeSuccProbs();
sinkMBB->addSuccessor(exitMBB, BranchProbability::getOne());
// loop1MBB:
// ll dest, 0(ptr)
// and Mask', dest, Mask
// bne Mask', ShiftCmpVal, exitMBB
BuildMI(loop1MBB, DL, TII->get(LL), Scratch).addReg(Ptr).addImm(0);
BuildMI(loop1MBB, DL, TII->get(Mips::AND), Scratch2)
.addReg(Scratch)
.addReg(Mask);
BuildMI(loop1MBB, DL, TII->get(BNE))
.addReg(Scratch2).addReg(ShiftCmpVal).addMBB(sinkMBB);
// loop2MBB:
// and dest, dest, mask2
// or dest, dest, ShiftNewVal
// sc dest, dest, 0(ptr)
// beq dest, $0, loop1MBB
BuildMI(loop2MBB, DL, TII->get(Mips::AND), Scratch)
.addReg(Scratch, RegState::Kill)
.addReg(Mask2);
BuildMI(loop2MBB, DL, TII->get(Mips::OR), Scratch)
.addReg(Scratch, RegState::Kill)
.addReg(ShiftNewVal);
BuildMI(loop2MBB, DL, TII->get(SC), Scratch)
.addReg(Scratch, RegState::Kill)
.addReg(Ptr)
.addImm(0);
BuildMI(loop2MBB, DL, TII->get(BEQ))
.addReg(Scratch, RegState::Kill)
.addReg(ZERO)
.addMBB(loop1MBB);
// sinkMBB:
// srl srlres, Mask', shiftamt
// sign_extend dest,srlres
BuildMI(sinkMBB, DL, TII->get(Mips::SRLV), Dest)
.addReg(Scratch2)
.addReg(ShiftAmnt);
if (STI->hasMips32r2()) {
BuildMI(sinkMBB, DL, TII->get(SEOp), Dest).addReg(Dest);
} else {
const unsigned ShiftImm =
I->getOpcode() == Mips::ATOMIC_CMP_SWAP_I16_POSTRA ? 16 : 24;
BuildMI(sinkMBB, DL, TII->get(Mips::SLL), Dest)
.addReg(Dest, RegState::Kill)
.addImm(ShiftImm);
BuildMI(sinkMBB, DL, TII->get(Mips::SRA), Dest)
.addReg(Dest, RegState::Kill)
.addImm(ShiftImm);
}
LivePhysRegs LiveRegs;
computeAndAddLiveIns(LiveRegs, *loop1MBB);
computeAndAddLiveIns(LiveRegs, *loop2MBB);
computeAndAddLiveIns(LiveRegs, *sinkMBB);
computeAndAddLiveIns(LiveRegs, *exitMBB);
NMBBI = BB.end();
I->eraseFromParent();
return true;
}
bool AArch64FrameLowering::restoreCalleeSavedRegisters(
MachineBasicBlock &MBB, MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI) const {
MachineFunction &MF = *MBB.getParent();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
unsigned Count = CSI.size();
DebugLoc DL;
assert((Count & 1) == 0 && "Odd number of callee-saved regs to spill!");
if (MI != MBB.end())
DL = MI->getDebugLoc();
for (unsigned i = 0; i < Count; i += 2) {
unsigned Reg1 = CSI[i].getReg();
unsigned Reg2 = CSI[i + 1].getReg();
// GPRs and FPRs are saved in pairs of 64-bit regs. We expect the CSI
// list to come in sorted by frame index so that we can issue the store
// pair instructions directly. Assert if we see anything otherwise.
assert(CSI[i].getFrameIdx() + 1 == CSI[i + 1].getFrameIdx() &&
"Out of order callee saved regs!");
// Issue sequence of non-sp increment and sp-pi restores for cs regs. Only
// the last load is sp-pi post-increment and de-allocates the stack:
// For example:
// ldp fp, lr, [sp, #32] // addImm(+4)
// ldp x20, x19, [sp, #16] // addImm(+2)
// ldp x22, x21, [sp], #48 // addImm(+6)
// Note: see comment in spillCalleeSavedRegisters()
unsigned LdrOpc;
assert((Count & 1) == 0 && "Odd number of callee-saved regs to spill!");
assert((i & 1) == 0 && "Odd index for callee-saved reg spill!");
if (AArch64::GPR64RegClass.contains(Reg1)) {
assert(AArch64::GPR64RegClass.contains(Reg2) &&
"Expected GPR64 callee-saved register pair!");
if (i == Count - 2)
LdrOpc = AArch64::LDPXpost;
else
LdrOpc = AArch64::LDPXi;
} else if (AArch64::FPR64RegClass.contains(Reg1)) {
assert(AArch64::FPR64RegClass.contains(Reg2) &&
"Expected FPR64 callee-saved register pair!");
if (i == Count - 2)
LdrOpc = AArch64::LDPDpost;
else
LdrOpc = AArch64::LDPDi;
} else
llvm_unreachable("Unexpected callee saved register!");
DEBUG(dbgs() << "CSR restore: (" << TRI->getName(Reg1) << ", "
<< TRI->getName(Reg2) << ") -> fi#(" << CSI[i].getFrameIdx()
<< ", " << CSI[i + 1].getFrameIdx() << ")\n");
// Compute offset: i = 0 => offset = Count - 2; i = 2 => offset = Count - 4;
// etc.
const int Offset = (i == Count - 2) ? Count : Count - i - 2;
assert((Offset >= -64 && Offset <= 63) &&
"Offset out of bounds for LDP immediate");
MachineInstrBuilder MIB = BuildMI(MBB, MI, DL, TII.get(LdrOpc));
if (LdrOpc == AArch64::LDPXpost || LdrOpc == AArch64::LDPDpost)
MIB.addReg(AArch64::SP, RegState::Define);
MIB.addReg(Reg2, getDefRegState(true))
.addReg(Reg1, getDefRegState(true))
.addReg(AArch64::SP)
.addImm(Offset); // [sp], #offset * 8 or [sp, #offset * 8]
// where the factor * 8 is implicit
}
return true;
}
VNInfo *LiveRangeCalc::findReachingDefs(LiveInterval *LI,
MachineBasicBlock *KillMBB,
SlotIndex Kill,
unsigned PhysReg) {
// Blocks where LI should be live-in.
SmallVector<MachineBasicBlock*, 16> WorkList(1, KillMBB);
// Remember if we have seen more than one value.
bool UniqueVNI = true;
VNInfo *TheVNI = 0;
// Using Seen as a visited set, perform a BFS for all reaching defs.
for (unsigned i = 0; i != WorkList.size(); ++i) {
MachineBasicBlock *MBB = WorkList[i];
#ifndef NDEBUG
if (MBB->pred_empty()) {
MBB->getParent()->verify();
llvm_unreachable("Use not jointly dominated by defs.");
}
if (TargetRegisterInfo::isPhysicalRegister(PhysReg) &&
!MBB->isLiveIn(PhysReg)) {
MBB->getParent()->verify();
errs() << "The register needs to be live in to BB#" << MBB->getNumber()
<< ", but is missing from the live-in list.\n";
llvm_unreachable("Invalid global physical register");
}
#endif
for (MachineBasicBlock::pred_iterator PI = MBB->pred_begin(),
PE = MBB->pred_end(); PI != PE; ++PI) {
MachineBasicBlock *Pred = *PI;
// Is this a known live-out block?
if (Seen.test(Pred->getNumber())) {
if (VNInfo *VNI = LiveOut[Pred].first) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
}
continue;
}
SlotIndex Start, End;
tie(Start, End) = Indexes->getMBBRange(Pred);
// First time we see Pred. Try to determine the live-out value, but set
// it as null if Pred is live-through with an unknown value.
VNInfo *VNI = LI->extendInBlock(Start, End);
setLiveOutValue(Pred, VNI);
if (VNI) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
continue;
}
// No, we need a live-in value for Pred as well
if (Pred != KillMBB)
WorkList.push_back(Pred);
else
// Loopback to KillMBB, so value is really live through.
Kill = SlotIndex();
}
}
// Transfer WorkList to LiveInBlocks in reverse order.
// This ordering works best with updateSSA().
LiveIn.clear();
LiveIn.reserve(WorkList.size());
while(!WorkList.empty())
addLiveInBlock(LI, DomTree->getNode(WorkList.pop_back_val()));
// The kill block may not be live-through.
assert(LiveIn.back().DomNode->getBlock() == KillMBB);
LiveIn.back().Kill = Kill;
return UniqueVNI ? TheVNI : 0;
}
bool LiveRangeCalc::findReachingDefs(LiveRange &LR, MachineBasicBlock &UseMBB,
SlotIndex Use, unsigned PhysReg,
ArrayRef<SlotIndex> Undefs) {
unsigned UseMBBNum = UseMBB.getNumber();
// Block numbers where LR should be live-in.
SmallVector<unsigned, 16> WorkList(1, UseMBBNum);
// Remember if we have seen more than one value.
bool UniqueVNI = true;
VNInfo *TheVNI = nullptr;
bool FoundUndef = false;
// Using Seen as a visited set, perform a BFS for all reaching defs.
for (unsigned i = 0; i != WorkList.size(); ++i) {
MachineBasicBlock *MBB = MF->getBlockNumbered(WorkList[i]);
#ifndef NDEBUG
if (Undefs.size() > 0 && MBB->pred_empty()) {
MBB->getParent()->verify();
errs() << "Use of " << PrintReg(PhysReg)
<< " does not have a corresponding definition on every path:\n";
const MachineInstr *MI = Indexes->getInstructionFromIndex(Use);
if (MI != nullptr)
errs() << Use << " " << *MI;
llvm_unreachable("Use not jointly dominated by defs.");
}
if (TargetRegisterInfo::isPhysicalRegister(PhysReg) &&
!MBB->isLiveIn(PhysReg)) {
MBB->getParent()->verify();
const TargetRegisterInfo *TRI = MRI->getTargetRegisterInfo();
errs() << "The register " << PrintReg(PhysReg, TRI)
<< " needs to be live in to BB#" << MBB->getNumber()
<< ", but is missing from the live-in list.\n";
llvm_unreachable("Invalid global physical register");
}
#endif
FoundUndef |= MBB->pred_empty();
for (MachineBasicBlock::pred_iterator PI = MBB->pred_begin(),
PE = MBB->pred_end(); PI != PE; ++PI) {
MachineBasicBlock *Pred = *PI;
// Is this a known live-out block?
if (Seen.test(Pred->getNumber())) {
if (VNInfo *VNI = Map[Pred].first) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
}
continue;
}
SlotIndex Start, End;
std::tie(Start, End) = Indexes->getMBBRange(Pred);
// First time we see Pred. Try to determine the live-out value, but set
// it as null if Pred is live-through with an unknown value.
auto EP = LR.extendInBlock(Undefs, Start, End);
VNInfo *VNI = EP.first;
FoundUndef |= EP.second;
setLiveOutValue(Pred, VNI);
if (VNI) {
if (TheVNI && TheVNI != VNI)
UniqueVNI = false;
TheVNI = VNI;
}
if (VNI || EP.second)
continue;
// No, we need a live-in value for Pred as well
if (Pred != &UseMBB)
WorkList.push_back(Pred->getNumber());
else
// Loopback to UseMBB, so value is really live through.
Use = SlotIndex();
}
}
LiveIn.clear();
FoundUndef |= (TheVNI == nullptr);
if (Undefs.size() > 0 && FoundUndef)
UniqueVNI = false;
// Both updateSSA() and LiveRangeUpdater benefit from ordered blocks, but
// neither require it. Skip the sorting overhead for small updates.
if (WorkList.size() > 4)
array_pod_sort(WorkList.begin(), WorkList.end());
// If a unique reaching def was found, blit in the live ranges immediately.
if (UniqueVNI) {
assert(TheVNI != nullptr);
LiveRangeUpdater Updater(&LR);
for (unsigned BN : WorkList) {
SlotIndex Start, End;
std::tie(Start, End) = Indexes->getMBBRange(BN);
// Trim the live range in UseMBB.
if (BN == UseMBBNum && Use.isValid())
End = Use;
else
Map[MF->getBlockNumbered(BN)] = LiveOutPair(TheVNI, nullptr);
Updater.add(Start, End, TheVNI);
}
return true;
}
// Prepare the defined/undefined bit vectors.
auto EF = EntryInfoMap.find(&LR);
if (EF == EntryInfoMap.end()) {
unsigned N = MF->getNumBlockIDs();
EF = EntryInfoMap.insert({&LR, {BitVector(), BitVector()}}).first;
EF->second.first.resize(N);
EF->second.second.resize(N);
}
BitVector &DefOnEntry = EF->second.first;
BitVector &UndefOnEntry = EF->second.second;
// Multiple values were found, so transfer the work list to the LiveIn array
// where UpdateSSA will use it as a work list.
LiveIn.reserve(WorkList.size());
for (unsigned BN : WorkList) {
MachineBasicBlock *MBB = MF->getBlockNumbered(BN);
if (Undefs.size() > 0 && !isDefOnEntry(LR, Undefs, *MBB, DefOnEntry, UndefOnEntry))
continue;
addLiveInBlock(LR, DomTree->getNode(MBB));
if (MBB == &UseMBB)
LiveIn.back().Kill = Use;
}
return false;
}
