unsigned AlphaCodeEmitter::getMachineOpValue(const MachineInstr &MI,
const MachineOperand &MO) {
unsigned rv = 0; // Return value; defaults to 0 for unhandled cases
// or things that get fixed up later by the JIT.
if (MO.isReg()) {
rv = getAlphaRegNumber(MO.getReg());
} else if (MO.isImm()) {
rv = MO.getImm();
} else if (MO.isGlobal() || MO.isSymbol() || MO.isCPI()) {
DEBUG(errs() << MO << " is a relocated op for " << MI << "\n");
unsigned Reloc = 0;
int Offset = 0;
bool useGOT = false;
switch (MI.getOpcode()) {
case Alpha::BSR:
Reloc = Alpha::reloc_bsr;
break;
case Alpha::LDLr:
case Alpha::LDQr:
case Alpha::LDBUr:
case Alpha::LDWUr:
case Alpha::LDSr:
case Alpha::LDTr:
case Alpha::LDAr:
case Alpha::STQr:
case Alpha::STLr:
case Alpha::STWr:
case Alpha::STBr:
case Alpha::STSr:
case Alpha::STTr:
Reloc = Alpha::reloc_gprellow;
break;
case Alpha::LDAHr:
Reloc = Alpha::reloc_gprelhigh;
break;
case Alpha::LDQl:
Reloc = Alpha::reloc_literal;
useGOT = true;
break;
case Alpha::LDAg:
case Alpha::LDAHg:
Reloc = Alpha::reloc_gpdist;
Offset = MI.getOperand(3).getImm();
break;
default:
llvm_unreachable("unknown relocatable instruction");
}
if (MO.isGlobal())
MCE.addRelocation(MachineRelocation::getGV(MCE.getCurrentPCOffset(),
Reloc, MO.getGlobal(), Offset,
isa<Function>(MO.getGlobal()),
useGOT));
else if (MO.isSymbol())
MCE.addRelocation(MachineRelocation::getExtSym(MCE.getCurrentPCOffset(),
Reloc, MO.getSymbolName(),
Offset, true));
else
MCE.addRelocation(MachineRelocation::getConstPool(MCE.getCurrentPCOffset(),
Reloc, MO.getIndex(), Offset));
} else if (MO.isMBB()) {
MCE.addRelocation(MachineRelocation::getBB(MCE.getCurrentPCOffset(),
Alpha::reloc_bsr, MO.getMBB()));
} else {
#ifndef NDEBUG
errs() << "ERROR: Unknown type of MachineOperand: " << MO << "\n";
#endif
llvm_unreachable(0);
}
return rv;
}
bool SIFoldOperands::runOnMachineFunction(MachineFunction &MF) {
MachineRegisterInfo &MRI = MF.getRegInfo();
const SIInstrInfo *TII =
static_cast<const SIInstrInfo *>(MF.getSubtarget().getInstrInfo());
const SIRegisterInfo &TRI = TII->getRegisterInfo();
for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
BI != BE; ++BI) {
MachineBasicBlock &MBB = *BI;
MachineBasicBlock::iterator I, Next;
for (I = MBB.begin(); I != MBB.end(); I = Next) {
Next = std::next(I);
MachineInstr &MI = *I;
if (!isSafeToFold(MI.getOpcode()))
continue;
unsigned OpSize = TII->getOpSize(MI, 1);
MachineOperand &OpToFold = MI.getOperand(1);
bool FoldingImm = OpToFold.isImm();
// FIXME: We could also be folding things like FrameIndexes and
// TargetIndexes.
if (!FoldingImm && !OpToFold.isReg())
continue;
// Folding immediates with more than one use will increase program size.
// FIXME: This will also reduce register usage, which may be better
// in some cases. A better heuristic is needed.
if (FoldingImm && !TII->isInlineConstant(OpToFold, OpSize) &&
!MRI.hasOneUse(MI.getOperand(0).getReg()))
continue;
// FIXME: Fold operands with subregs.
if (OpToFold.isReg() &&
(!TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()) ||
OpToFold.getSubReg()))
continue;
std::vector<FoldCandidate> FoldList;
for (MachineRegisterInfo::use_iterator
Use = MRI.use_begin(MI.getOperand(0).getReg()), E = MRI.use_end();
Use != E; ++Use) {
MachineInstr *UseMI = Use->getParent();
const MachineOperand &UseOp = UseMI->getOperand(Use.getOperandNo());
// FIXME: Fold operands with subregs.
if (UseOp.isReg() && ((UseOp.getSubReg() && OpToFold.isReg()) ||
UseOp.isImplicit())) {
continue;
}
APInt Imm;
if (FoldingImm) {
unsigned UseReg = UseOp.getReg();
const TargetRegisterClass *UseRC
= TargetRegisterInfo::isVirtualRegister(UseReg) ?
MRI.getRegClass(UseReg) :
TRI.getRegClass(UseReg);
Imm = APInt(64, OpToFold.getImm());
// Split 64-bit constants into 32-bits for folding.
if (UseOp.getSubReg()) {
if (UseRC->getSize() != 8)
continue;
if (UseOp.getSubReg() == AMDGPU::sub0) {
Imm = Imm.getLoBits(32);
} else {
assert(UseOp.getSubReg() == AMDGPU::sub1);
Imm = Imm.getHiBits(32);
}
}
// In order to fold immediates into copies, we need to change the
// copy to a MOV.
if (UseMI->getOpcode() == AMDGPU::COPY) {
unsigned DestReg = UseMI->getOperand(0).getReg();
const TargetRegisterClass *DestRC
= TargetRegisterInfo::isVirtualRegister(DestReg) ?
MRI.getRegClass(DestReg) :
TRI.getRegClass(DestReg);
unsigned MovOp = TII->getMovOpcode(DestRC);
if (MovOp == AMDGPU::COPY)
continue;
UseMI->setDesc(TII->get(MovOp));
}
}
const MCInstrDesc &UseDesc = UseMI->getDesc();
// Don't fold into target independent nodes. Target independent opcodes
// don't have defined register classes.
if (UseDesc.isVariadic() ||
UseDesc.OpInfo[Use.getOperandNo()].RegClass == -1)
continue;
if (FoldingImm) {
MachineOperand ImmOp = MachineOperand::CreateImm(Imm.getSExtValue());
tryAddToFoldList(FoldList, UseMI, Use.getOperandNo(), &ImmOp, TII);
continue;
}
tryAddToFoldList(FoldList, UseMI, Use.getOperandNo(), &OpToFold, TII);
// FIXME: We could try to change the instruction from 64-bit to 32-bit
// to enable more folding opportunites. The shrink operands pass
// already does this.
}
for (FoldCandidate &Fold : FoldList) {
if (updateOperand(Fold, TRI)) {
// Clear kill flags.
if (!Fold.isImm()) {
assert(Fold.OpToFold && Fold.OpToFold->isReg());
Fold.OpToFold->setIsKill(false);
}
DEBUG(dbgs() << "Folded source from " << MI << " into OpNo " <<
Fold.UseOpNo << " of " << *Fold.UseMI << '\n');
}
}
}
}
return false;
}
void PEI::replaceFrameIndices(MachineBasicBlock *BB, MachineFunction &Fn,
int &SPAdj) {
assert(Fn.getSubtarget().getRegisterInfo() &&
"getRegisterInfo() must be implemented!");
const TargetInstrInfo &TII = *Fn.getSubtarget().getInstrInfo();
const TargetRegisterInfo &TRI = *Fn.getSubtarget().getRegisterInfo();
const TargetFrameLowering *TFI = Fn.getSubtarget().getFrameLowering();
unsigned FrameSetupOpcode = TII.getCallFrameSetupOpcode();
unsigned FrameDestroyOpcode = TII.getCallFrameDestroyOpcode();
if (RS && !FrameIndexVirtualScavenging) RS->enterBasicBlock(BB);
bool InsideCallSequence = false;
for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
if (I->getOpcode() == FrameSetupOpcode ||
I->getOpcode() == FrameDestroyOpcode) {
InsideCallSequence = (I->getOpcode() == FrameSetupOpcode);
SPAdj += TII.getSPAdjust(I);
MachineBasicBlock::iterator PrevI = BB->end();
if (I != BB->begin()) PrevI = std::prev(I);
TFI->eliminateCallFramePseudoInstr(Fn, *BB, I);
// Visit the instructions created by eliminateCallFramePseudoInstr().
if (PrevI == BB->end())
I = BB->begin(); // The replaced instr was the first in the block.
else
I = std::next(PrevI);
continue;
}
MachineInstr *MI = I;
bool DoIncr = true;
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
if (!MI->getOperand(i).isFI())
continue;
// Frame indicies in debug values are encoded in a target independent
// way with simply the frame index and offset rather than any
// target-specific addressing mode.
if (MI->isDebugValue()) {
assert(i == 0 && "Frame indicies can only appear as the first "
"operand of a DBG_VALUE machine instruction");
unsigned Reg;
MachineOperand &Offset = MI->getOperand(1);
Offset.setImm(Offset.getImm() +
TFI->getFrameIndexReference(
Fn, MI->getOperand(0).getIndex(), Reg));
MI->getOperand(0).ChangeToRegister(Reg, false /*isDef*/);
continue;
}
// TODO: This code should be commoned with the code for
// PATCHPOINT. There's no good reason for the difference in
// implementation other than historical accident. The only
// remaining difference is the unconditional use of the stack
// pointer as the base register.
if (MI->getOpcode() == TargetOpcode::STATEPOINT) {
assert((!MI->isDebugValue() || i == 0) &&
"Frame indicies can only appear as the first operand of a "
"DBG_VALUE machine instruction");
unsigned Reg;
MachineOperand &Offset = MI->getOperand(i + 1);
const unsigned refOffset =
TFI->getFrameIndexReferenceFromSP(Fn, MI->getOperand(i).getIndex(),
Reg);
Offset.setImm(Offset.getImm() + refOffset);
MI->getOperand(i).ChangeToRegister(Reg, false /*isDef*/);
continue;
}
// Some instructions (e.g. inline asm instructions) can have
// multiple frame indices and/or cause eliminateFrameIndex
// to insert more than one instruction. We need the register
// scavenger to go through all of these instructions so that
// it can update its register information. We keep the
// iterator at the point before insertion so that we can
// revisit them in full.
bool AtBeginning = (I == BB->begin());
if (!AtBeginning) --I;
// If this instruction has a FrameIndex operand, we need to
// use that target machine register info object to eliminate
// it.
TRI.eliminateFrameIndex(MI, SPAdj, i,
FrameIndexVirtualScavenging ? nullptr : RS);
// Reset the iterator if we were at the beginning of the BB.
if (AtBeginning) {
I = BB->begin();
DoIncr = false;
}
MI = nullptr;
break;
}
// If we are looking at a call sequence, we need to keep track of
// the SP adjustment made by each instruction in the sequence.
// This includes both the frame setup/destroy pseudos (handled above),
// as well as other instructions that have side effects w.r.t the SP.
// Note that this must come after eliminateFrameIndex, because
// if I itself referred to a frame index, we shouldn't count its own
// adjustment.
if (MI && InsideCallSequence)
SPAdj += TII.getSPAdjust(MI);
if (DoIncr && I != BB->end()) ++I;
// Update register states.
if (RS && !FrameIndexVirtualScavenging && MI) RS->forward(MI);
}
}
/// Attempt the reassociation transformation to reduce critical path length.
/// See the above comments before getMachineCombinerPatterns().
void TargetInstrInfo::reassociateOps(
MachineInstr &Root, MachineInstr &Prev,
MachineCombinerPattern::MC_PATTERN Pattern,
SmallVectorImpl<MachineInstr *> &InsInstrs,
SmallVectorImpl<MachineInstr *> &DelInstrs,
DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) const {
MachineFunction *MF = Root.getParent()->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = Root.getRegClassConstraint(0, TII, TRI);
// This array encodes the operand index for each parameter because the
// operands may be commuted. Each row corresponds to a pattern value,
// and each column specifies the index of A, B, X, Y.
unsigned OpIdx[4][4] = {
{ 1, 1, 2, 2 },
{ 1, 2, 2, 1 },
{ 2, 1, 1, 2 },
{ 2, 2, 1, 1 }
};
MachineOperand &OpA = Prev.getOperand(OpIdx[Pattern][0]);
MachineOperand &OpB = Root.getOperand(OpIdx[Pattern][1]);
MachineOperand &OpX = Prev.getOperand(OpIdx[Pattern][2]);
MachineOperand &OpY = Root.getOperand(OpIdx[Pattern][3]);
MachineOperand &OpC = Root.getOperand(0);
unsigned RegA = OpA.getReg();
unsigned RegB = OpB.getReg();
unsigned RegX = OpX.getReg();
unsigned RegY = OpY.getReg();
unsigned RegC = OpC.getReg();
if (TargetRegisterInfo::isVirtualRegister(RegA))
MRI.constrainRegClass(RegA, RC);
if (TargetRegisterInfo::isVirtualRegister(RegB))
MRI.constrainRegClass(RegB, RC);
if (TargetRegisterInfo::isVirtualRegister(RegX))
MRI.constrainRegClass(RegX, RC);
if (TargetRegisterInfo::isVirtualRegister(RegY))
MRI.constrainRegClass(RegY, RC);
if (TargetRegisterInfo::isVirtualRegister(RegC))
MRI.constrainRegClass(RegC, RC);
// Create a new virtual register for the result of (X op Y) instead of
// recycling RegB because the MachineCombiner's computation of the critical
// path requires a new register definition rather than an existing one.
unsigned NewVR = MRI.createVirtualRegister(RC);
InstrIdxForVirtReg.insert(std::make_pair(NewVR, 0));
unsigned Opcode = Root.getOpcode();
bool KillA = OpA.isKill();
bool KillX = OpX.isKill();
bool KillY = OpY.isKill();
// Create new instructions for insertion.
MachineInstrBuilder MIB1 =
BuildMI(*MF, Prev.getDebugLoc(), TII->get(Opcode), NewVR)
.addReg(RegX, getKillRegState(KillX))
.addReg(RegY, getKillRegState(KillY));
MachineInstrBuilder MIB2 =
BuildMI(*MF, Root.getDebugLoc(), TII->get(Opcode), RegC)
.addReg(RegA, getKillRegState(KillA))
.addReg(NewVR, getKillRegState(true));
setSpecialOperandAttr(Root, Prev, *MIB1, *MIB2);
// Record new instructions for insertion and old instructions for deletion.
InsInstrs.push_back(MIB1);
InsInstrs.push_back(MIB2);
DelInstrs.push_back(&Prev);
DelInstrs.push_back(&Root);
}
void
UserValue::addDefsFromCopies(LiveInterval *LI, unsigned LocNo,
const SmallVectorImpl<SlotIndex> &Kills,
SmallVectorImpl<std::pair<SlotIndex, unsigned> > &NewDefs,
MachineRegisterInfo &MRI, LiveIntervals &LIS) {
if (Kills.empty())
return;
// Don't track copies from physregs, there are too many uses.
if (!TargetRegisterInfo::isVirtualRegister(LI->reg))
return;
// Collect all the (vreg, valno) pairs that are copies of LI.
SmallVector<std::pair<LiveInterval*, const VNInfo*>, 8> CopyValues;
for (MachineOperand &MO : MRI.use_nodbg_operands(LI->reg)) {
MachineInstr *MI = MO.getParent();
// Copies of the full value.
if (MO.getSubReg() || !MI->isCopy())
continue;
unsigned DstReg = MI->getOperand(0).getReg();
// Don't follow copies to physregs. These are usually setting up call
// arguments, and the argument registers are always call clobbered. We are
// better off in the source register which could be a callee-saved register,
// or it could be spilled.
if (!TargetRegisterInfo::isVirtualRegister(DstReg))
continue;
// Is LocNo extended to reach this copy? If not, another def may be blocking
// it, or we are looking at a wrong value of LI.
SlotIndex Idx = LIS.getInstructionIndex(MI);
LocMap::iterator I = locInts.find(Idx.getRegSlot(true));
if (!I.valid() || I.value() != LocNo)
continue;
if (!LIS.hasInterval(DstReg))
continue;
LiveInterval *DstLI = &LIS.getInterval(DstReg);
const VNInfo *DstVNI = DstLI->getVNInfoAt(Idx.getRegSlot());
assert(DstVNI && DstVNI->def == Idx.getRegSlot() && "Bad copy value");
CopyValues.push_back(std::make_pair(DstLI, DstVNI));
}
if (CopyValues.empty())
return;
DEBUG(dbgs() << "Got " << CopyValues.size() << " copies of " << *LI << '\n');
// Try to add defs of the copied values for each kill point.
for (unsigned i = 0, e = Kills.size(); i != e; ++i) {
SlotIndex Idx = Kills[i];
for (unsigned j = 0, e = CopyValues.size(); j != e; ++j) {
LiveInterval *DstLI = CopyValues[j].first;
const VNInfo *DstVNI = CopyValues[j].second;
if (DstLI->getVNInfoAt(Idx) != DstVNI)
continue;
// Check that there isn't already a def at Idx
LocMap::iterator I = locInts.find(Idx);
if (I.valid() && I.start() <= Idx)
continue;
DEBUG(dbgs() << "Kill at " << Idx << " covered by valno #"
<< DstVNI->id << " in " << *DstLI << '\n');
MachineInstr *CopyMI = LIS.getInstructionFromIndex(DstVNI->def);
assert(CopyMI && CopyMI->isCopy() && "Bad copy value");
unsigned LocNo = getLocationNo(CopyMI->getOperand(0));
I.insert(Idx, Idx.getNextSlot(), LocNo);
NewDefs.push_back(std::make_pair(Idx, LocNo));
break;
}
}
}
bool LiveVariables::runOnMachineFunction(MachineFunction &mf) {
MF = &mf;
MRI = &mf.getRegInfo();
TRI = MF->getTarget().getRegisterInfo();
unsigned NumRegs = TRI->getNumRegs();
PhysRegDef = new MachineInstr*[NumRegs];
PhysRegUse = new MachineInstr*[NumRegs];
PHIVarInfo = new SmallVector<unsigned, 4>[MF->getNumBlockIDs()];
std::fill(PhysRegDef, PhysRegDef + NumRegs, nullptr);
std::fill(PhysRegUse, PhysRegUse + NumRegs, nullptr);
PHIJoins.clear();
// FIXME: LiveIntervals will be updated to remove its dependence on
// LiveVariables to improve compilation time and eliminate bizarre pass
// dependencies. Until then, we can't change much in -O0.
if (!MRI->isSSA())
report_fatal_error("regalloc=... not currently supported with -O0");
analyzePHINodes(mf);
// Calculate live variable information in depth first order on the CFG of the
// function. This guarantees that we will see the definition of a virtual
// register before its uses due to dominance properties of SSA (except for PHI
// nodes, which are treated as a special case).
MachineBasicBlock *Entry = MF->begin();
SmallPtrSet<MachineBasicBlock*,16> Visited;
for (df_ext_iterator<MachineBasicBlock*, SmallPtrSet<MachineBasicBlock*,16> >
DFI = df_ext_begin(Entry, Visited), E = df_ext_end(Entry, Visited);
DFI != E; ++DFI) {
MachineBasicBlock *MBB = *DFI;
// Mark live-in registers as live-in.
SmallVector<unsigned, 4> Defs;
for (MachineBasicBlock::livein_iterator II = MBB->livein_begin(),
EE = MBB->livein_end(); II != EE; ++II) {
assert(TargetRegisterInfo::isPhysicalRegister(*II) &&
"Cannot have a live-in virtual register!");
HandlePhysRegDef(*II, nullptr, Defs);
}
// Loop over all of the instructions, processing them.
DistanceMap.clear();
unsigned Dist = 0;
for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end();
I != E; ++I) {
MachineInstr *MI = I;
if (MI->isDebugValue())
continue;
DistanceMap.insert(std::make_pair(MI, Dist++));
// Process all of the operands of the instruction...
unsigned NumOperandsToProcess = MI->getNumOperands();
// Unless it is a PHI node. In this case, ONLY process the DEF, not any
// of the uses. They will be handled in other basic blocks.
if (MI->isPHI())
NumOperandsToProcess = 1;
// Clear kill and dead markers. LV will recompute them.
SmallVector<unsigned, 4> UseRegs;
SmallVector<unsigned, 4> DefRegs;
SmallVector<unsigned, 1> RegMasks;
for (unsigned i = 0; i != NumOperandsToProcess; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (MO.isRegMask()) {
RegMasks.push_back(i);
continue;
}
if (!MO.isReg() || MO.getReg() == 0)
continue;
unsigned MOReg = MO.getReg();
if (MO.isUse()) {
MO.setIsKill(false);
if (MO.readsReg())
UseRegs.push_back(MOReg);
} else /*MO.isDef()*/ {
MO.setIsDead(false);
DefRegs.push_back(MOReg);
}
}
// Process all uses.
for (unsigned i = 0, e = UseRegs.size(); i != e; ++i) {
unsigned MOReg = UseRegs[i];
if (TargetRegisterInfo::isVirtualRegister(MOReg))
HandleVirtRegUse(MOReg, MBB, MI);
else if (!MRI->isReserved(MOReg))
HandlePhysRegUse(MOReg, MI);
}
// Process all masked registers. (Call clobbers).
for (unsigned i = 0, e = RegMasks.size(); i != e; ++i)
HandleRegMask(MI->getOperand(RegMasks[i]));
// Process all defs.
for (unsigned i = 0, e = DefRegs.size(); i != e; ++i) {
unsigned MOReg = DefRegs[i];
if (TargetRegisterInfo::isVirtualRegister(MOReg))
HandleVirtRegDef(MOReg, MI);
else if (!MRI->isReserved(MOReg))
HandlePhysRegDef(MOReg, MI, Defs);
}
UpdatePhysRegDefs(MI, Defs);
}
// Handle any virtual assignments from PHI nodes which might be at the
// bottom of this basic block. We check all of our successor blocks to see
// if they have PHI nodes, and if so, we simulate an assignment at the end
// of the current block.
if (!PHIVarInfo[MBB->getNumber()].empty()) {
SmallVectorImpl<unsigned> &VarInfoVec = PHIVarInfo[MBB->getNumber()];
for (SmallVectorImpl<unsigned>::iterator I = VarInfoVec.begin(),
E = VarInfoVec.end(); I != E; ++I)
// Mark it alive only in the block we are representing.
MarkVirtRegAliveInBlock(getVarInfo(*I),MRI->getVRegDef(*I)->getParent(),
MBB);
}
// MachineCSE may CSE instructions which write to non-allocatable physical
// registers across MBBs. Remember if any reserved register is liveout.
SmallSet<unsigned, 4> LiveOuts;
for (MachineBasicBlock::const_succ_iterator SI = MBB->succ_begin(),
SE = MBB->succ_end(); SI != SE; ++SI) {
MachineBasicBlock *SuccMBB = *SI;
if (SuccMBB->isLandingPad())
continue;
for (MachineBasicBlock::livein_iterator LI = SuccMBB->livein_begin(),
LE = SuccMBB->livein_end(); LI != LE; ++LI) {
unsigned LReg = *LI;
if (!TRI->isInAllocatableClass(LReg))
// Ignore other live-ins, e.g. those that are live into landing pads.
LiveOuts.insert(LReg);
}
}
// Loop over PhysRegDef / PhysRegUse, killing any registers that are
// available at the end of the basic block.
for (unsigned i = 0; i != NumRegs; ++i)
if ((PhysRegDef[i] || PhysRegUse[i]) && !LiveOuts.count(i))
HandlePhysRegDef(i, nullptr, Defs);
std::fill(PhysRegDef, PhysRegDef + NumRegs, nullptr);
std::fill(PhysRegUse, PhysRegUse + NumRegs, nullptr);
}
// Convert and transfer the dead / killed information we have gathered into
// VirtRegInfo onto MI's.
for (unsigned i = 0, e1 = VirtRegInfo.size(); i != e1; ++i) {
const unsigned Reg = TargetRegisterInfo::index2VirtReg(i);
for (unsigned j = 0, e2 = VirtRegInfo[Reg].Kills.size(); j != e2; ++j)
if (VirtRegInfo[Reg].Kills[j] == MRI->getVRegDef(Reg))
VirtRegInfo[Reg].Kills[j]->addRegisterDead(Reg, TRI);
else
VirtRegInfo[Reg].Kills[j]->addRegisterKilled(Reg, TRI);
}
// Check to make sure there are no unreachable blocks in the MC CFG for the
// function. If so, it is due to a bug in the instruction selector or some
// other part of the code generator if this happens.
#ifndef NDEBUG
for(MachineFunction::iterator i = MF->begin(), e = MF->end(); i != e; ++i)
assert(Visited.count(&*i) != 0 && "unreachable basic block found");
#endif
delete[] PhysRegDef;
delete[] PhysRegUse;
delete[] PHIVarInfo;
return false;
}
bool
AArch64InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
MachineBasicBlock *&FBB,
SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify) const {
// If the block has no terminators, it just falls into the block after it.
MachineBasicBlock::iterator I = MBB.end();
if (I == MBB.begin())
return false;
--I;
while (I->isDebugValue()) {
if (I == MBB.begin())
return false;
--I;
}
if (!isUnpredicatedTerminator(I))
return false;
// Get the last instruction in the block.
MachineInstr *LastInst = I;
// If there is only one terminator instruction, process it.
unsigned LastOpc = LastInst->getOpcode();
if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
if (LastOpc == AArch64::Bimm) {
TBB = LastInst->getOperand(0).getMBB();
return false;
}
if (isCondBranch(LastOpc)) {
classifyCondBranch(LastInst, TBB, Cond);
return false;
}
return true; // Can't handle indirect branch.
}
// Get the instruction before it if it is a terminator.
MachineInstr *SecondLastInst = I;
unsigned SecondLastOpc = SecondLastInst->getOpcode();
// If AllowModify is true and the block ends with two or more unconditional
// branches, delete all but the first unconditional branch.
if (AllowModify && LastOpc == AArch64::Bimm) {
while (SecondLastOpc == AArch64::Bimm) {
LastInst->eraseFromParent();
LastInst = SecondLastInst;
LastOpc = LastInst->getOpcode();
if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
// Return now the only terminator is an unconditional branch.
TBB = LastInst->getOperand(0).getMBB();
return false;
} else {
SecondLastInst = I;
SecondLastOpc = SecondLastInst->getOpcode();
}
}
}
// If there are three terminators, we don't know what sort of block this is.
if (SecondLastInst && I != MBB.begin() && isUnpredicatedTerminator(--I))
return true;
// If the block ends with a B and a Bcc, handle it.
if (LastOpc == AArch64::Bimm) {
if (SecondLastOpc == AArch64::Bcc) {
TBB = SecondLastInst->getOperand(1).getMBB();
Cond.push_back(MachineOperand::CreateImm(AArch64::Bcc));
Cond.push_back(SecondLastInst->getOperand(0));
FBB = LastInst->getOperand(0).getMBB();
return false;
} else if (isCondBranch(SecondLastOpc)) {
classifyCondBranch(SecondLastInst, TBB, Cond);
FBB = LastInst->getOperand(0).getMBB();
return false;
}
}
// If the block ends with two unconditional branches, handle it. The second
// one is not executed, so remove it.
if (SecondLastOpc == AArch64::Bimm && LastOpc == AArch64::Bimm) {
TBB = SecondLastInst->getOperand(0).getMBB();
I = LastInst;
if (AllowModify)
I->eraseFromParent();
return false;
}
// Otherwise, can't handle this.
return true;
}
MachineInstr *
ARMBSISimplifyIndexMemOpsPass::convertToSimpleInstrs(MachineFunction::iterator &MFI,
MachineBasicBlock::iterator &MBBI,
LiveVariables *LV) const {
// FIXME: Thumb2 support.
llvm::errs() << "about to convert to three address\n";
llvm::errs() << "arg passed to convert to three address\n";
MachineInstr *MI = MBBI;
MachineFunction &MF = *MI->getParent()->getParent();
uint64_t TSFlags = MI->getDesc().TSFlags;
bool isPre = false;
switch ((TSFlags & ARMII::IndexModeMask) >> ARMII::IndexModeShift) {
default: return NULL;
case ARMII::IndexModePre:
isPre = true;
break;
case ARMII::IndexModePost:
break;
}
// Try splitting an indexed load/store to an un-indexed one plus an add/sub
// operation.
unsigned MemOpc = TII->getUnindexedOpcode(MI->getOpcode());
if (MemOpc == 0)
return NULL;
MachineInstr *UpdateMI = NULL;
MachineInstr *MemMI = NULL;
MachineInstrBuilder MemMIB;
unsigned AddrMode = (TSFlags & ARMII::AddrModeMask);
const MCInstrDesc &MCID = MI->getDesc();
unsigned NumOps = MCID.getNumOperands();
bool isLoad = !MI->mayStore();
const MachineOperand &WB = isLoad ? MI->getOperand(1) : MI->getOperand(0);
const MachineOperand &Base = MI->getOperand(2);
const MachineOperand &Offset = MI->getOperand(NumOps-3);
unsigned WBReg = WB.getReg();
unsigned BaseReg = Base.getReg();
unsigned OffReg = -1;
unsigned OffImm = -1;
if (Offset.isReg())
OffReg = Offset.getReg();
else if (MI->getOperand(NumOps-3).isImm()) {
OffImm = MI->getOperand(NumOps-3).getImm();
OffReg = 0;
}
else {
llvm::errs() << "FAIL\n";
return NULL;//FAIL
}
ARMCC::CondCodes Pred = (ARMCC::CondCodes)MI->getOperand(NumOps-2).getImm();
switch (AddrMode) {
default: llvm_unreachable("Unknown indexed op!");
case ARMII::AddrMode2: {
bool isSub = ARM_AM::getAM2Op(OffImm) == ARM_AM::sub;
uint64_t Amt = ARM_AM::getAM2Offset(OffImm);
if (isSub) {
Amt = ~Amt;
Amt++;
Amt &= 0x0fff;
return NULL;
}
if (OffReg == 0) {
if (ARM_AM::getSOImmVal(Amt) == -1) {
// Can't encode it in a so_imm operand. This transformation will
// add more than 1 instruction. Abandon!
llvm::errs() << "here amt = " << Amt << " is sub: " << isSub << "\n";
return NULL;
}
llvm::errs() << "NONFAIL offImm amt " << OffImm << " pred = " << Pred << "here amt = " << Amt << " is sub: " << isSub << "\n";
UpdateMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(isSub ? ARM::SUBrsi : ARM::ADDrsi), WBReg)
.addReg(BaseReg).addImm(Amt)
.addImm(Pred).addReg(0).addReg(0);
} else if (Amt != 0) {
ARM_AM::ShiftOpc ShOpc = ARM_AM::getAM2ShiftOpc(OffImm);
unsigned SOOpc = ARM_AM::getSORegOpc(ShOpc, Amt);
UpdateMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(isSub ? ARM::SUBri : ARM::ADDri), WBReg)
.addReg(BaseReg).addReg(OffReg).addReg(0).addImm(SOOpc)
.addImm(Pred).addReg(0).addReg(0);
} else
UpdateMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(isSub ? ARM::SUBrr : ARM::ADDrr), WBReg)
.addReg(BaseReg).addReg(OffReg)
.addImm(Pred).addReg(0).addReg(0);
break;
}
case ARMII::AddrMode3 : {
bool isSub = ARM_AM::getAM3Op(OffImm) == ARM_AM::sub;
unsigned Amt = ARM_AM::getAM3Offset(OffImm);
llvm::errs() << " in addressing mode 3\n";
if (OffReg == 0)
// Immediate is 8-bits. It's guaranteed to fit in a so_imm operand.
UpdateMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(isSub ? ARM::SUBri : ARM::ADDri), WBReg)
.addReg(BaseReg).addImm(Amt)
.addImm(Pred).addReg(0).addReg(0);
else
UpdateMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(isSub ? ARM::SUBrr : ARM::ADDrr), WBReg)
.addReg(BaseReg).addReg(OffReg)
.addImm(Pred).addReg(0).addReg(0);
break;
}
}
std::vector<MachineInstr*> NewMIs;
if (isPre) {
if (isLoad)
MemMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(MemOpc), MI->getOperand(0).getReg())
.addReg(WBReg).addImm(0).addImm(Pred).addReg(0);
else {
MemMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(MemOpc)).addReg(MI->getOperand(1).getReg())
.addReg(WBReg).addImm(0).addImm(Pred).addReg(0);
//AddDefaultPred(MemMIB);
}
NewMIs.push_back(MemMI);
} else {
if (isLoad)
MemMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(MemOpc), MI->getOperand(0).getReg())
.addReg(BaseReg).addImm(0).addImm(Pred).addReg(0);
else {
MemMI = BuildMI(MF, MI->getDebugLoc(),
TII->get(MemOpc)).addReg(MI->getOperand(1).getReg())
.addReg(BaseReg).addImm(0).addImm(Pred).addReg(0);
//AddDefaultPred(MemMIB);
}
if (WB.isDead())
UpdateMI->getOperand(0).setIsDead();
NewMIs.push_back(UpdateMI);
NewMIs.push_back(MemMI);
}
// Transfer LiveVariables states, kill / dead info.
if (LV) {
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
unsigned Reg = MO.getReg();
LiveVariables::VarInfo &VI = LV->getVarInfo(Reg);
if (MO.isDef()) {
MachineInstr *NewMI = (Reg == WBReg) ? UpdateMI : MemMI;
if (MO.isDead())
LV->addVirtualRegisterDead(Reg, NewMI);
}
if (MO.isUse() && MO.isKill()) {
for (unsigned j = 0; j < 2; ++j) {
// Look at the two new MI's in reverse order.
MachineInstr *NewMI = NewMIs[j];
if (!NewMI->readsRegister(Reg))
continue;
LV->addVirtualRegisterKilled(Reg, NewMI);
if (VI.removeKill(MI))
VI.Kills.push_back(NewMI);
break;
}
}
}
}
}
MFI->insert(MBBI, NewMIs[1]);
MFI->insert(MBBI, NewMIs[0]);
return NewMIs[0];
}
bool WebAssemblyLowerBrUnless::runOnMachineFunction(MachineFunction &MF) {
DEBUG(dbgs() << "********** Lowering br_unless **********\n"
"********** Function: "
<< MF.getName() << '\n');
auto &MFI = *MF.getInfo<WebAssemblyFunctionInfo>();
const auto &TII = *MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo();
auto &MRI = MF.getRegInfo();
for (auto &MBB : MF) {
for (auto MII = MBB.begin(); MII != MBB.end();) {
MachineInstr *MI = &*MII++;
if (MI->getOpcode() != WebAssembly::BR_UNLESS)
continue;
unsigned Cond = MI->getOperand(1).getReg();
bool Inverted = false;
// Attempt to invert the condition in place.
if (MFI.isVRegStackified(Cond)) {
assert(MRI.hasOneDef(Cond));
MachineInstr *Def = MRI.getVRegDef(Cond);
switch (Def->getOpcode()) {
using namespace WebAssembly;
case EQ_I32: Def->setDesc(TII.get(NE_I32)); Inverted = true; break;
case NE_I32: Def->setDesc(TII.get(EQ_I32)); Inverted = true; break;
case GT_S_I32: Def->setDesc(TII.get(LE_S_I32)); Inverted = true; break;
case GE_S_I32: Def->setDesc(TII.get(LT_S_I32)); Inverted = true; break;
case LT_S_I32: Def->setDesc(TII.get(GE_S_I32)); Inverted = true; break;
case LE_S_I32: Def->setDesc(TII.get(GT_S_I32)); Inverted = true; break;
case GT_U_I32: Def->setDesc(TII.get(LE_U_I32)); Inverted = true; break;
case GE_U_I32: Def->setDesc(TII.get(LT_U_I32)); Inverted = true; break;
case LT_U_I32: Def->setDesc(TII.get(GE_U_I32)); Inverted = true; break;
case LE_U_I32: Def->setDesc(TII.get(GT_U_I32)); Inverted = true; break;
case EQ_I64: Def->setDesc(TII.get(NE_I64)); Inverted = true; break;
case NE_I64: Def->setDesc(TII.get(EQ_I64)); Inverted = true; break;
case GT_S_I64: Def->setDesc(TII.get(LE_S_I64)); Inverted = true; break;
case GE_S_I64: Def->setDesc(TII.get(LT_S_I64)); Inverted = true; break;
case LT_S_I64: Def->setDesc(TII.get(GE_S_I64)); Inverted = true; break;
case LE_S_I64: Def->setDesc(TII.get(GT_S_I64)); Inverted = true; break;
case GT_U_I64: Def->setDesc(TII.get(LE_U_I64)); Inverted = true; break;
case GE_U_I64: Def->setDesc(TII.get(LT_U_I64)); Inverted = true; break;
case LT_U_I64: Def->setDesc(TII.get(GE_U_I64)); Inverted = true; break;
case LE_U_I64: Def->setDesc(TII.get(GT_U_I64)); Inverted = true; break;
case EQ_F32: Def->setDesc(TII.get(NE_F32)); Inverted = true; break;
case NE_F32: Def->setDesc(TII.get(EQ_F32)); Inverted = true; break;
case EQ_F64: Def->setDesc(TII.get(NE_F64)); Inverted = true; break;
case NE_F64: Def->setDesc(TII.get(EQ_F64)); Inverted = true; break;
default: break;
}
}
// If we weren't able to invert the condition in place. Insert an
// instruction to invert it.
if (!Inverted) {
unsigned Tmp = MRI.createVirtualRegister(&WebAssembly::I32RegClass);
BuildMI(MBB, MI, MI->getDebugLoc(), TII.get(WebAssembly::EQZ_I32), Tmp)
.addReg(Cond);
MFI.stackifyVReg(Tmp);
Cond = Tmp;
Inverted = true;
}
// The br_unless condition has now been inverted. Insert a br_if and
// delete the br_unless.
assert(Inverted);
BuildMI(MBB, MI, MI->getDebugLoc(), TII.get(WebAssembly::BR_IF))
.addOperand(MI->getOperand(0))
.addReg(Cond);
MBB.erase(MI);
}
}
return true;
}
// Creates the more optimized patterns and generally does all the code
// transformations in this pass.
unsigned A15SDOptimizer::optimizeSDPattern(MachineInstr *MI) {
if (MI->isCopy()) {
return optimizeAllLanesPattern(MI, MI->getOperand(1).getReg());
}
if (MI->isInsertSubreg()) {
unsigned DPRReg = MI->getOperand(1).getReg();
unsigned SPRReg = MI->getOperand(2).getReg();
if (TRI->isVirtualRegister(DPRReg) && TRI->isVirtualRegister(SPRReg)) {
MachineInstr *DPRMI = MRI->getVRegDef(MI->getOperand(1).getReg());
MachineInstr *SPRMI = MRI->getVRegDef(MI->getOperand(2).getReg());
if (DPRMI && SPRMI) {
// See if the first operand of this insert_subreg is IMPLICIT_DEF
MachineInstr *ECDef = elideCopies(DPRMI);
if (ECDef && ECDef->isImplicitDef()) {
// Another corner case - if we're inserting something that is purely
// a subreg copy of a DPR, just use that DPR.
MachineInstr *EC = elideCopies(SPRMI);
// Is it a subreg copy of ssub_0?
if (EC && EC->isCopy() &&
EC->getOperand(1).getSubReg() == ARM::ssub_0) {
LLVM_DEBUG(dbgs() << "Found a subreg copy: " << *SPRMI);
// Find the thing we're subreg copying out of - is it of the same
// regclass as DPRMI? (i.e. a DPR or QPR).
unsigned FullReg = SPRMI->getOperand(1).getReg();
const TargetRegisterClass *TRC =
MRI->getRegClass(MI->getOperand(1).getReg());
if (TRC->hasSuperClassEq(MRI->getRegClass(FullReg))) {
LLVM_DEBUG(dbgs() << "Subreg copy is compatible - returning ");
LLVM_DEBUG(dbgs() << printReg(FullReg) << "\n");
eraseInstrWithNoUses(MI);
return FullReg;
}
}
return optimizeAllLanesPattern(MI, MI->getOperand(2).getReg());
}
}
}
return optimizeAllLanesPattern(MI, MI->getOperand(0).getReg());
}
if (MI->isRegSequence() && usesRegClass(MI->getOperand(1),
&ARM::SPRRegClass)) {
// See if all bar one of the operands are IMPLICIT_DEF and insert the
// optimizer pattern accordingly.
unsigned NumImplicit = 0, NumTotal = 0;
unsigned NonImplicitReg = ~0U;
for (unsigned I = 1; I < MI->getNumExplicitOperands(); ++I) {
if (!MI->getOperand(I).isReg())
continue;
++NumTotal;
unsigned OpReg = MI->getOperand(I).getReg();
if (!TRI->isVirtualRegister(OpReg))
break;
MachineInstr *Def = MRI->getVRegDef(OpReg);
if (!Def)
break;
if (Def->isImplicitDef())
++NumImplicit;
else
NonImplicitReg = MI->getOperand(I).getReg();
}
if (NumImplicit == NumTotal - 1)
return optimizeAllLanesPattern(MI, NonImplicitReg);
else
return optimizeAllLanesPattern(MI, MI->getOperand(0).getReg());
}
llvm_unreachable("Unhandled update pattern!");
}
unsigned CriticalAntiDepBreaker::
BreakAntiDependencies(const std::vector<SUnit>& SUnits,
MachineBasicBlock::iterator Begin,
MachineBasicBlock::iterator End,
unsigned InsertPosIndex) {
// The code below assumes that there is at least one instruction,
// so just duck out immediately if the block is empty.
if (SUnits.empty()) return 0;
// Keep a map of the MachineInstr*'s back to the SUnit representing them.
// This is used for updating debug information.
DenseMap<MachineInstr*,const SUnit*> MISUnitMap;
// Find the node at the bottom of the critical path.
const SUnit *Max = 0;
for (unsigned i = 0, e = SUnits.size(); i != e; ++i) {
const SUnit *SU = &SUnits[i];
MISUnitMap[SU->getInstr()] = SU;
if (!Max || SU->getDepth() + SU->Latency > Max->getDepth() + Max->Latency)
Max = SU;
}
#ifndef NDEBUG
{
DEBUG(dbgs() << "Critical path has total latency "
<< (Max->getDepth() + Max->Latency) << "\n");
DEBUG(dbgs() << "Available regs:");
for (unsigned Reg = 0; Reg < TRI->getNumRegs(); ++Reg) {
if (KillIndices[Reg] == ~0u)
DEBUG(dbgs() << " " << TRI->getName(Reg));
}
DEBUG(dbgs() << '\n');
}
#endif
// Track progress along the critical path through the SUnit graph as we walk
// the instructions.
const SUnit *CriticalPathSU = Max;
MachineInstr *CriticalPathMI = CriticalPathSU->getInstr();
// Consider this pattern:
// A = ...
// ... = A
// A = ...
// ... = A
// A = ...
// ... = A
// A = ...
// ... = A
// There are three anti-dependencies here, and without special care,
// we'd break all of them using the same register:
// A = ...
// ... = A
// B = ...
// ... = B
// B = ...
// ... = B
// B = ...
// ... = B
// because at each anti-dependence, B is the first register that
// isn't A which is free. This re-introduces anti-dependencies
// at all but one of the original anti-dependencies that we were
// trying to break. To avoid this, keep track of the most recent
// register that each register was replaced with, avoid
// using it to repair an anti-dependence on the same register.
// This lets us produce this:
// A = ...
// ... = A
// B = ...
// ... = B
// C = ...
// ... = C
// B = ...
// ... = B
// This still has an anti-dependence on B, but at least it isn't on the
// original critical path.
//
// TODO: If we tracked more than one register here, we could potentially
// fix that remaining critical edge too. This is a little more involved,
// because unlike the most recent register, less recent registers should
// still be considered, though only if no other registers are available.
std::vector<unsigned> LastNewReg(TRI->getNumRegs(), 0);
// Attempt to break anti-dependence edges on the critical path. Walk the
// instructions from the bottom up, tracking information about liveness
// as we go to help determine which registers are available.
unsigned Broken = 0;
unsigned Count = InsertPosIndex - 1;
for (MachineBasicBlock::iterator I = End, E = Begin;
I != E; --Count) {
MachineInstr *MI = --I;
if (MI->isDebugValue())
continue;
// Check if this instruction has a dependence on the critical path that
// is an anti-dependence that we may be able to break. If it is, set
// AntiDepReg to the non-zero register associated with the anti-dependence.
//
// We limit our attention to the critical path as a heuristic to avoid
// breaking anti-dependence edges that aren't going to significantly
// impact the overall schedule. There are a limited number of registers
// and we want to save them for the important edges.
//
// TODO: Instructions with multiple defs could have multiple
// anti-dependencies. The current code here only knows how to break one
// edge per instruction. Note that we'd have to be able to break all of
// the anti-dependencies in an instruction in order to be effective.
unsigned AntiDepReg = 0;
if (MI == CriticalPathMI) {
if (const SDep *Edge = CriticalPathStep(CriticalPathSU)) {
const SUnit *NextSU = Edge->getSUnit();
// Only consider anti-dependence edges.
if (Edge->getKind() == SDep::Anti) {
AntiDepReg = Edge->getReg();
assert(AntiDepReg != 0 && "Anti-dependence on reg0?");
if (!AllocatableSet.test(AntiDepReg))
// Don't break anti-dependencies on non-allocatable registers.
AntiDepReg = 0;
else if (KeepRegs.count(AntiDepReg))
// Don't break anti-dependencies if an use down below requires
// this exact register.
AntiDepReg = 0;
else {
// If the SUnit has other dependencies on the SUnit that it
// anti-depends on, don't bother breaking the anti-dependency
// since those edges would prevent such units from being
// scheduled past each other regardless.
//
// Also, if there are dependencies on other SUnits with the
// same register as the anti-dependency, don't attempt to
// break it.
for (SUnit::const_pred_iterator P = CriticalPathSU->Preds.begin(),
PE = CriticalPathSU->Preds.end(); P != PE; ++P)
if (P->getSUnit() == NextSU ?
(P->getKind() != SDep::Anti || P->getReg() != AntiDepReg) :
(P->getKind() == SDep::Data && P->getReg() == AntiDepReg)) {
AntiDepReg = 0;
break;
}
}
}
CriticalPathSU = NextSU;
CriticalPathMI = CriticalPathSU->getInstr();
} else {
// We've reached the end of the critical path.
CriticalPathSU = 0;
CriticalPathMI = 0;
}
}
PrescanInstruction(MI);
// If MI's defs have a special allocation requirement, don't allow
// any def registers to be changed. Also assume all registers
// defined in a call must not be changed (ABI).
if (MI->getDesc().isCall() || MI->getDesc().hasExtraDefRegAllocReq() ||
TII->isPredicated(MI))
// If this instruction's defs have special allocation requirement, don't
// break this anti-dependency.
AntiDepReg = 0;
else if (AntiDepReg) {
// If this instruction has a use of AntiDepReg, breaking it
// is invalid.
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg()) continue;
unsigned Reg = MO.getReg();
if (Reg == 0) continue;
if (MO.isUse() && TRI->regsOverlap(AntiDepReg, Reg)) {
AntiDepReg = 0;
break;
}
}
}
// Determine AntiDepReg's register class, if it is live and is
// consistently used within a single class.
const TargetRegisterClass *RC = AntiDepReg != 0 ? Classes[AntiDepReg] : 0;
assert((AntiDepReg == 0 || RC != NULL) &&
"Register should be live if it's causing an anti-dependence!");
if (RC == reinterpret_cast<TargetRegisterClass *>(-1))
AntiDepReg = 0;
// Look for a suitable register to use to break the anti-depenence.
//
// TODO: Instead of picking the first free register, consider which might
// be the best.
if (AntiDepReg != 0) {
std::pair<std::multimap<unsigned, MachineOperand *>::iterator,
std::multimap<unsigned, MachineOperand *>::iterator>
Range = RegRefs.equal_range(AntiDepReg);
if (unsigned NewReg = findSuitableFreeRegister(Range.first, Range.second,
AntiDepReg,
LastNewReg[AntiDepReg],
RC)) {
DEBUG(dbgs() << "Breaking anti-dependence edge on "
<< TRI->getName(AntiDepReg)
<< " with " << RegRefs.count(AntiDepReg) << " references"
<< " using " << TRI->getName(NewReg) << "!\n");
// Update the references to the old register to refer to the new
// register.
for (std::multimap<unsigned, MachineOperand *>::iterator
Q = Range.first, QE = Range.second; Q != QE; ++Q) {
Q->second->setReg(NewReg);
// If the SU for the instruction being updated has debug information
// related to the anti-dependency register, make sure to update that
// as well.
const SUnit *SU = MISUnitMap[Q->second->getParent()];
if (!SU) continue;
for (unsigned i = 0, e = SU->DbgInstrList.size() ; i < e ; ++i) {
MachineInstr *DI = SU->DbgInstrList[i];
assert (DI->getNumOperands()==3 && DI->getOperand(0).isReg() &&
DI->getOperand(0).getReg()
&& "Non register dbg_value attached to SUnit!");
if (DI->getOperand(0).getReg() == AntiDepReg)
DI->getOperand(0).setReg(NewReg);
}
}
// We just went back in time and modified history; the
// liveness information for the anti-dependence reg is now
// inconsistent. Set the state as if it were dead.
Classes[NewReg] = Classes[AntiDepReg];
DefIndices[NewReg] = DefIndices[AntiDepReg];
KillIndices[NewReg] = KillIndices[AntiDepReg];
assert(((KillIndices[NewReg] == ~0u) !=
(DefIndices[NewReg] == ~0u)) &&
"Kill and Def maps aren't consistent for NewReg!");
Classes[AntiDepReg] = 0;
DefIndices[AntiDepReg] = KillIndices[AntiDepReg];
KillIndices[AntiDepReg] = ~0u;
assert(((KillIndices[AntiDepReg] == ~0u) !=
(DefIndices[AntiDepReg] == ~0u)) &&
"Kill and Def maps aren't consistent for AntiDepReg!");
RegRefs.erase(AntiDepReg);
LastNewReg[AntiDepReg] = NewReg;
++Broken;
}
}
ScanInstruction(MI, Count);
}
return Broken;
}
void AVRRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
int SPAdj, unsigned FIOperandNum,
RegScavenger *RS) const {
assert(SPAdj == 0 && "Unexpected SPAdj value");
MachineInstr &MI = *II;
DebugLoc dl = MI.getDebugLoc();
MachineBasicBlock &MBB = *MI.getParent();
const MachineFunction &MF = *MBB.getParent();
const AVRTargetMachine &TM = (const AVRTargetMachine &)MF.getTarget();
const TargetInstrInfo &TII = *TM.getSubtargetImpl()->getInstrInfo();
const MachineFrameInfo &MFI = MF.getFrameInfo();
const TargetFrameLowering *TFI = TM.getSubtargetImpl()->getFrameLowering();
int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
int Offset = MFI.getObjectOffset(FrameIndex);
// Add one to the offset because SP points to an empty slot.
Offset += MFI.getStackSize() - TFI->getOffsetOfLocalArea() + 1;
// Fold incoming offset.
Offset += MI.getOperand(FIOperandNum + 1).getImm();
// This is actually "load effective address" of the stack slot
// instruction. We have only two-address instructions, thus we need to
// expand it into move + add.
if (MI.getOpcode() == AVR::FRMIDX) {
MI.setDesc(TII.get(AVR::MOVWRdRr));
MI.getOperand(FIOperandNum).ChangeToRegister(AVR::R29R28, false);
assert(Offset > 0 && "Invalid offset");
// We need to materialize the offset via an add instruction.
unsigned Opcode;
unsigned DstReg = MI.getOperand(0).getReg();
assert(DstReg != AVR::R29R28 && "Dest reg cannot be the frame pointer");
// Generally, to load a frame address two add instructions are emitted that
// could get folded into a single one:
// movw r31:r30, r29:r28
// adiw r31:r30, 29
// adiw r31:r30, 16
// to:
// movw r31:r30, r29:r28
// adiw r31:r30, 45
foldFrameOffset(*std::next(II), Offset, DstReg);
// Select the best opcode based on DstReg and the offset size.
switch (DstReg) {
case AVR::R25R24:
case AVR::R27R26:
case AVR::R31R30: {
if (isUInt<6>(Offset)) {
Opcode = AVR::ADIWRdK;
break;
}
LLVM_FALLTHROUGH;
}
default: {
// This opcode will get expanded into a pair of subi/sbci.
Opcode = AVR::SUBIWRdK;
Offset = -Offset;
break;
}
}
MachineInstr *New = BuildMI(MBB, std::next(II), dl, TII.get(Opcode), DstReg)
.addReg(DstReg, RegState::Kill)
.addImm(Offset);
New->getOperand(3).setIsDead();
return;
}
// If the offset is too big we have to adjust and restore the frame pointer
// to materialize a valid load/store with displacement.
//:TODO: consider using only one adiw/sbiw chain for more than one frame index
if (Offset > 63) {
unsigned AddOpc = AVR::ADIWRdK, SubOpc = AVR::SBIWRdK;
int AddOffset = Offset - 63 + 1;
// For huge offsets where adiw/sbiw cannot be used use a pair of subi/sbci.
if ((Offset - 63 + 1) > 63) {
AddOpc = AVR::SUBIWRdK;
SubOpc = AVR::SUBIWRdK;
AddOffset = -AddOffset;
}
// It is possible that the spiller places this frame instruction in between
// a compare and branch, invalidating the contents of SREG set by the
// compare instruction because of the add/sub pairs. Conservatively save and
// restore SREG before and after each add/sub pair.
BuildMI(MBB, II, dl, TII.get(AVR::INRdA), AVR::R0).addImm(0x3f);
MachineInstr *New = BuildMI(MBB, II, dl, TII.get(AddOpc), AVR::R29R28)
.addReg(AVR::R29R28, RegState::Kill)
.addImm(AddOffset);
New->getOperand(3).setIsDead();
// Restore SREG.
BuildMI(MBB, std::next(II), dl, TII.get(AVR::OUTARr))
.addImm(0x3f)
.addReg(AVR::R0, RegState::Kill);
// No need to set SREG as dead here otherwise if the next instruction is a
// cond branch it will be using a dead register.
New = BuildMI(MBB, std::next(II), dl, TII.get(SubOpc), AVR::R29R28)
.addReg(AVR::R29R28, RegState::Kill)
.addImm(Offset - 63 + 1);
Offset = 62;
}
MI.getOperand(FIOperandNum).ChangeToRegister(AVR::R29R28, false);
assert(isUInt<6>(Offset) && "Offset is out of range");
MI.getOperand(FIOperandNum + 1).ChangeToImmediate(Offset);
}
void X86RegisterInfo::
eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const {
const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
bool reseveCallFrame = TFI->hasReservedCallFrame(MF);
int Opcode = I->getOpcode();
bool isDestroy = Opcode == TII.getCallFrameDestroyOpcode();
DebugLoc DL = I->getDebugLoc();
uint64_t Amount = !reseveCallFrame ? I->getOperand(0).getImm() : 0;
uint64_t CalleeAmt = isDestroy ? I->getOperand(1).getImm() : 0;
I = MBB.erase(I);
if (!reseveCallFrame) {
// If the stack pointer can be changed after prologue, turn the
// adjcallstackup instruction into a 'sub ESP, <amt>' and the
// adjcallstackdown instruction into 'add ESP, <amt>'
// TODO: consider using push / pop instead of sub + store / add
if (Amount == 0)
return;
// We need to keep the stack aligned properly. To do this, we round the
// amount of space needed for the outgoing arguments up to the next
// alignment boundary.
unsigned StackAlign = TM.getFrameLowering()->getStackAlignment();
Amount = (Amount + StackAlign - 1) / StackAlign * StackAlign;
MachineInstr *New = 0;
if (Opcode == TII.getCallFrameSetupOpcode()) {
New = BuildMI(MF, DL, TII.get(getSUBriOpcode(Is64Bit, Amount)),
StackPtr)
.addReg(StackPtr)
.addImm(Amount);
} else {
assert(Opcode == TII.getCallFrameDestroyOpcode());
// Factor out the amount the callee already popped.
Amount -= CalleeAmt;
if (Amount) {
unsigned Opc = getADDriOpcode(Is64Bit, Amount);
New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
.addReg(StackPtr).addImm(Amount);
}
}
if (New) {
// The EFLAGS implicit def is dead.
New->getOperand(3).setIsDead();
// Replace the pseudo instruction with a new instruction.
MBB.insert(I, New);
}
return;
}
if (Opcode == TII.getCallFrameDestroyOpcode() && CalleeAmt) {
// If we are performing frame pointer elimination and if the callee pops
// something off the stack pointer, add it back. We do this until we have
// more advanced stack pointer tracking ability.
unsigned Opc = getSUBriOpcode(Is64Bit, CalleeAmt);
MachineInstr *New = BuildMI(MF, DL, TII.get(Opc), StackPtr)
.addReg(StackPtr).addImm(CalleeAmt);
// The EFLAGS implicit def is dead.
New->getOperand(3).setIsDead();
// We are not tracking the stack pointer adjustment by the callee, so make
// sure we restore the stack pointer immediately after the call, there may
// be spill code inserted between the CALL and ADJCALLSTACKUP instructions.
MachineBasicBlock::iterator B = MBB.begin();
while (I != B && !llvm::prior(I)->isCall())
--I;
MBB.insert(I, New);
}
}
static void LowerTlsAddr(MCStreamer &OutStreamer,
X86MCInstLower &MCInstLowering,
const MachineInstr &MI) {
bool is64Bits = MI.getOpcode() == X86::TLS_addr64 ||
MI.getOpcode() == X86::TLS_base_addr64;
bool needsPadding = MI.getOpcode() == X86::TLS_addr64;
MCContext &context = OutStreamer.getContext();
if (needsPadding)
OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
MCSymbolRefExpr::VariantKind SRVK;
switch (MI.getOpcode()) {
case X86::TLS_addr32:
case X86::TLS_addr64:
SRVK = MCSymbolRefExpr::VK_TLSGD;
break;
case X86::TLS_base_addr32:
SRVK = MCSymbolRefExpr::VK_TLSLDM;
break;
case X86::TLS_base_addr64:
SRVK = MCSymbolRefExpr::VK_TLSLD;
break;
default:
llvm_unreachable("unexpected opcode");
}
MCSymbol *sym = MCInstLowering.GetSymbolFromOperand(MI.getOperand(3));
const MCSymbolRefExpr *symRef = MCSymbolRefExpr::Create(sym, SRVK, context);
MCInst LEA;
if (is64Bits) {
LEA.setOpcode(X86::LEA64r);
LEA.addOperand(MCOperand::CreateReg(X86::RDI)); // dest
LEA.addOperand(MCOperand::CreateReg(X86::RIP)); // base
LEA.addOperand(MCOperand::CreateImm(1)); // scale
LEA.addOperand(MCOperand::CreateReg(0)); // index
LEA.addOperand(MCOperand::CreateExpr(symRef)); // disp
LEA.addOperand(MCOperand::CreateReg(0)); // seg
} else if (SRVK == MCSymbolRefExpr::VK_TLSLDM) {
LEA.setOpcode(X86::LEA32r);
LEA.addOperand(MCOperand::CreateReg(X86::EAX)); // dest
LEA.addOperand(MCOperand::CreateReg(X86::EBX)); // base
LEA.addOperand(MCOperand::CreateImm(1)); // scale
LEA.addOperand(MCOperand::CreateReg(0)); // index
LEA.addOperand(MCOperand::CreateExpr(symRef)); // disp
LEA.addOperand(MCOperand::CreateReg(0)); // seg
} else {
LEA.setOpcode(X86::LEA32r);
LEA.addOperand(MCOperand::CreateReg(X86::EAX)); // dest
LEA.addOperand(MCOperand::CreateReg(0)); // base
LEA.addOperand(MCOperand::CreateImm(1)); // scale
LEA.addOperand(MCOperand::CreateReg(X86::EBX)); // index
LEA.addOperand(MCOperand::CreateExpr(symRef)); // disp
LEA.addOperand(MCOperand::CreateReg(0)); // seg
}
OutStreamer.EmitInstruction(LEA);
if (needsPadding) {
OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX));
OutStreamer.EmitInstruction(MCInstBuilder(X86::REX64_PREFIX));
}
StringRef name = is64Bits ? "__tls_get_addr" : "___tls_get_addr";
MCSymbol *tlsGetAddr = context.GetOrCreateSymbol(name);
const MCSymbolRefExpr *tlsRef =
MCSymbolRefExpr::Create(tlsGetAddr,
MCSymbolRefExpr::VK_PLT,
context);
OutStreamer.EmitInstruction(MCInstBuilder(is64Bits ? X86::CALL64pcrel32
: X86::CALLpcrel32)
.addExpr(tlsRef));
}
void MSP430FrameLowering::emitPrologue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
assert(&MF.front() == &MBB && "Shrink-wrapping not yet supported");
MachineFrameInfo *MFI = MF.getFrameInfo();
MSP430MachineFunctionInfo *MSP430FI = MF.getInfo<MSP430MachineFunctionInfo>();
const MSP430InstrInfo &TII =
*static_cast<const MSP430InstrInfo *>(MF.getSubtarget().getInstrInfo());
MachineBasicBlock::iterator MBBI = MBB.begin();
DebugLoc DL = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
// Get the number of bytes to allocate from the FrameInfo.
uint64_t StackSize = MFI->getStackSize();
uint64_t NumBytes = 0;
if (hasFP(MF)) {
// Calculate required stack adjustment
uint64_t FrameSize = StackSize - 2;
NumBytes = FrameSize - MSP430FI->getCalleeSavedFrameSize();
// Get the offset of the stack slot for the EBP register... which is
// guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
// Update the frame offset adjustment.
MFI->setOffsetAdjustment(-NumBytes);
// Save FP into the appropriate stack slot...
BuildMI(MBB, MBBI, DL, TII.get(MSP430::PUSH16r))
.addReg(MSP430::FP, RegState::Kill);
// Update FP with the new base value...
BuildMI(MBB, MBBI, DL, TII.get(MSP430::MOV16rr), MSP430::FP)
.addReg(MSP430::SP);
// Mark the FramePtr as live-in in every block except the entry.
for (MachineFunction::iterator I = std::next(MF.begin()), E = MF.end();
I != E; ++I)
I->addLiveIn(MSP430::FP);
} else
NumBytes = StackSize - MSP430FI->getCalleeSavedFrameSize();
// Skip the callee-saved push instructions.
while (MBBI != MBB.end() && (MBBI->getOpcode() == MSP430::PUSH16r))
++MBBI;
if (MBBI != MBB.end())
DL = MBBI->getDebugLoc();
if (NumBytes) { // adjust stack pointer: SP -= numbytes
// If there is an SUB16ri of SP immediately before this instruction, merge
// the two.
//NumBytes -= mergeSPUpdates(MBB, MBBI, true);
// If there is an ADD16ri or SUB16ri of SP immediately after this
// instruction, merge the two instructions.
// mergeSPUpdatesDown(MBB, MBBI, &NumBytes);
if (NumBytes) {
MachineInstr *MI =
BuildMI(MBB, MBBI, DL, TII.get(MSP430::SUB16ri), MSP430::SP)
.addReg(MSP430::SP).addImm(NumBytes);
// The SRW implicit def is dead.
MI->getOperand(3).setIsDead();
}
}
}
void RegScavenger::forward() {
// Move ptr forward.
if (!Tracking) {
MBBI = MBB->begin();
Tracking = true;
} else {
assert(MBBI != MBB->end() && "Already at the end of the basic block!");
MBBI = next(MBBI);
}
MachineInstr *MI = MBBI;
if (MI == ScavengeRestore) {
ScavengedReg = 0;
ScavengedRC = NULL;
ScavengeRestore = NULL;
}
// Find out which registers are early clobbered, killed, defined, and marked
// def-dead in this instruction.
BitVector EarlyClobberRegs(NumPhysRegs);
BitVector KillRegs(NumPhysRegs);
BitVector DefRegs(NumPhysRegs);
BitVector DeadRegs(NumPhysRegs);
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg() || MO.isUndef())
continue;
unsigned Reg = MO.getReg();
if (!Reg || isReserved(Reg))
continue;
if (MO.isUse()) {
// Two-address operands implicitly kill.
if (MO.isKill() || MI->isRegTiedToDefOperand(i))
addRegWithSubRegs(KillRegs, Reg);
} else {
assert(MO.isDef());
if (MO.isDead())
addRegWithSubRegs(DeadRegs, Reg);
else
addRegWithSubRegs(DefRegs, Reg);
if (MO.isEarlyClobber())
addRegWithAliases(EarlyClobberRegs, Reg);
}
}
// Verify uses and defs.
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg() || MO.isUndef())
continue;
unsigned Reg = MO.getReg();
if (!Reg || isReserved(Reg))
continue;
if (MO.isUse()) {
assert(isUsed(Reg) && "Using an undefined register!");
assert((!EarlyClobberRegs.test(Reg) || MI->isRegTiedToDefOperand(i)) &&
"Using an early clobbered register!");
} else {
assert(MO.isDef());
assert((KillRegs.test(Reg) || isUnused(Reg) ||
isLiveInButUnusedBefore(Reg, MI, MBB, TRI, MRI)) &&
"Re-defining a live register!");
}
}
// Commit the changes.
setUnused(KillRegs);
setUnused(DeadRegs);
setUsed(DefRegs);
}
/// replaceFrameIndices - Replace all MO_FrameIndex operands with physical
/// register references and actual offsets.
///
void PEI::replaceFrameIndices(MachineFunction &Fn) {
if (!Fn.getFrameInfo()->hasStackObjects()) return; // Nothing to do?
const TargetMachine &TM = Fn.getTarget();
assert(TM.getRegisterInfo() && "TM::getRegisterInfo() must be implemented!");
const TargetInstrInfo &TII = *Fn.getTarget().getInstrInfo();
const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
const TargetFrameLowering *TFI = TM.getFrameLowering();
bool StackGrowsDown =
TFI->getStackGrowthDirection() == TargetFrameLowering::StackGrowsDown;
int FrameSetupOpcode = TII.getCallFrameSetupOpcode();
int FrameDestroyOpcode = TII.getCallFrameDestroyOpcode();
for (MachineFunction::iterator BB = Fn.begin(),
E = Fn.end(); BB != E; ++BB) {
#ifndef NDEBUG
int SPAdjCount = 0; // frame setup / destroy count.
#endif
int SPAdj = 0; // SP offset due to call frame setup / destroy.
if (RS && !FrameIndexVirtualScavenging) RS->enterBasicBlock(BB);
for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
if (I->getOpcode() == FrameSetupOpcode ||
I->getOpcode() == FrameDestroyOpcode) {
#ifndef NDEBUG
// Track whether we see even pairs of them
SPAdjCount += I->getOpcode() == FrameSetupOpcode ? 1 : -1;
#endif
// Remember how much SP has been adjusted to create the call
// frame.
int Size = I->getOperand(0).getImm();
if ((!StackGrowsDown && I->getOpcode() == FrameSetupOpcode) ||
(StackGrowsDown && I->getOpcode() == FrameDestroyOpcode))
Size = -Size;
SPAdj += Size;
MachineBasicBlock::iterator PrevI = BB->end();
if (I != BB->begin()) PrevI = prior(I);
TRI.eliminateCallFramePseudoInstr(Fn, *BB, I);
// Visit the instructions created by eliminateCallFramePseudoInstr().
if (PrevI == BB->end())
I = BB->begin(); // The replaced instr was the first in the block.
else
I = llvm::next(PrevI);
continue;
}
MachineInstr *MI = I;
bool DoIncr = true;
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
if (!MI->getOperand(i).isFI())
continue;
// Some instructions (e.g. inline asm instructions) can have
// multiple frame indices and/or cause eliminateFrameIndex
// to insert more than one instruction. We need the register
// scavenger to go through all of these instructions so that
// it can update its register information. We keep the
// iterator at the point before insertion so that we can
// revisit them in full.
bool AtBeginning = (I == BB->begin());
if (!AtBeginning) --I;
// If this instruction has a FrameIndex operand, we need to
// use that target machine register info object to eliminate
// it.
TRI.eliminateFrameIndex(MI, SPAdj, i,
FrameIndexVirtualScavenging ? NULL : RS);
// Reset the iterator if we were at the beginning of the BB.
if (AtBeginning) {
I = BB->begin();
DoIncr = false;
}
MI = 0;
break;
}
if (DoIncr && I != BB->end()) ++I;
// Update register states.
if (RS && !FrameIndexVirtualScavenging && MI) RS->forward(MI);
}
// If we have evenly matched pairs of frame setup / destroy instructions,
// make sure the adjustments come out to zero. If we don't have matched
// pairs, we can't be sure the missing bit isn't in another basic block
// due to a custom inserter playing tricks, so just asserting SPAdj==0
// isn't sufficient. See tMOVCC on Thumb1, for example.
assert((SPAdjCount || SPAdj == 0) &&
"Unbalanced call frame setup / destroy pairs?");
}
}
bool ARMInstructionSelector::select(MachineInstr &I) const {
assert(I.getParent() && "Instruction should be in a basic block!");
assert(I.getParent()->getParent() && "Instruction should be in a function!");
auto &MBB = *I.getParent();
auto &MF = *MBB.getParent();
auto &MRI = MF.getRegInfo();
if (!isPreISelGenericOpcode(I.getOpcode())) {
if (I.isCopy())
return selectCopy(I, TII, MRI, TRI, RBI);
return true;
}
if (selectImpl(I))
return true;
MachineInstrBuilder MIB{MF, I};
bool isSExt = false;
using namespace TargetOpcode;
switch (I.getOpcode()) {
case G_SEXT:
isSExt = true;
LLVM_FALLTHROUGH;
case G_ZEXT: {
LLT DstTy = MRI.getType(I.getOperand(0).getReg());
// FIXME: Smaller destination sizes coming soon!
if (DstTy.getSizeInBits() != 32) {
DEBUG(dbgs() << "Unsupported destination size for extension");
return false;
}
LLT SrcTy = MRI.getType(I.getOperand(1).getReg());
unsigned SrcSize = SrcTy.getSizeInBits();
switch (SrcSize) {
case 1: {
// ZExt boils down to & 0x1; for SExt we also subtract that from 0
I.setDesc(TII.get(ARM::ANDri));
MIB.addImm(1).add(predOps(ARMCC::AL)).add(condCodeOp());
if (isSExt) {
unsigned SExtResult = I.getOperand(0).getReg();
// Use a new virtual register for the result of the AND
unsigned AndResult = MRI.createVirtualRegister(&ARM::GPRRegClass);
I.getOperand(0).setReg(AndResult);
auto InsertBefore = std::next(I.getIterator());
auto SubI =
BuildMI(MBB, InsertBefore, I.getDebugLoc(), TII.get(ARM::RSBri))
.addDef(SExtResult)
.addUse(AndResult)
.addImm(0)
.add(predOps(ARMCC::AL))
.add(condCodeOp());
if (!constrainSelectedInstRegOperands(*SubI, TII, TRI, RBI))
return false;
}
break;
}
case 8:
case 16: {
unsigned NewOpc = selectSimpleExtOpc(I.getOpcode(), SrcSize);
if (NewOpc == I.getOpcode())
return false;
I.setDesc(TII.get(NewOpc));
MIB.addImm(0).add(predOps(ARMCC::AL));
break;
}
default:
DEBUG(dbgs() << "Unsupported source size for extension");
return false;
}
break;
}
case G_ANYEXT:
case G_TRUNC: {
// The high bits are undefined, so there's nothing special to do, just
// treat it as a copy.
auto SrcReg = I.getOperand(1).getReg();
auto DstReg = I.getOperand(0).getReg();
const auto &SrcRegBank = *RBI.getRegBank(SrcReg, MRI, TRI);
const auto &DstRegBank = *RBI.getRegBank(DstReg, MRI, TRI);
if (SrcRegBank.getID() != DstRegBank.getID()) {
DEBUG(dbgs() << "G_TRUNC/G_ANYEXT operands on different register banks\n");
return false;
}
if (SrcRegBank.getID() != ARM::GPRRegBankID) {
DEBUG(dbgs() << "G_TRUNC/G_ANYEXT on non-GPR not supported yet\n");
return false;
}
I.setDesc(TII.get(COPY));
return selectCopy(I, TII, MRI, TRI, RBI);
}
case G_SELECT:
return selectSelect(MIB, MRI);
case G_ICMP: {
CmpConstants Helper(ARM::CMPrr, ARM::INSTRUCTION_LIST_END,
ARM::GPRRegBankID, 32);
return selectCmp(Helper, MIB, MRI);
}
case G_FCMP: {
assert(TII.getSubtarget().hasVFP2() && "Can't select fcmp without VFP");
unsigned OpReg = I.getOperand(2).getReg();
unsigned Size = MRI.getType(OpReg).getSizeInBits();
if (Size == 64 && TII.getSubtarget().isFPOnlySP()) {
DEBUG(dbgs() << "Subtarget only supports single precision");
return false;
}
if (Size != 32 && Size != 64) {
DEBUG(dbgs() << "Unsupported size for G_FCMP operand");
return false;
}
CmpConstants Helper(Size == 32 ? ARM::VCMPS : ARM::VCMPD, ARM::FMSTAT,
ARM::FPRRegBankID, Size);
return selectCmp(Helper, MIB, MRI);
}
case G_GEP:
I.setDesc(TII.get(ARM::ADDrr));
MIB.add(predOps(ARMCC::AL)).add(condCodeOp());
break;
case G_FRAME_INDEX:
// Add 0 to the given frame index and hope it will eventually be folded into
// the user(s).
I.setDesc(TII.get(ARM::ADDri));
MIB.addImm(0).add(predOps(ARMCC::AL)).add(condCodeOp());
break;
case G_CONSTANT: {
unsigned Reg = I.getOperand(0).getReg();
if (!validReg(MRI, Reg, 32, ARM::GPRRegBankID))
return false;
I.setDesc(TII.get(ARM::MOVi));
MIB.add(predOps(ARMCC::AL)).add(condCodeOp());
auto &Val = I.getOperand(1);
if (Val.isCImm()) {
if (Val.getCImm()->getBitWidth() > 32)
return false;
Val.ChangeToImmediate(Val.getCImm()->getZExtValue());
}
if (!Val.isImm()) {
return false;
}
break;
}
case G_GLOBAL_VALUE:
return selectGlobal(MIB, MRI);
case G_STORE:
case G_LOAD: {
const auto &MemOp = **I.memoperands_begin();
if (MemOp.getOrdering() != AtomicOrdering::NotAtomic) {
DEBUG(dbgs() << "Atomic load/store not supported yet\n");
return false;
}
unsigned Reg = I.getOperand(0).getReg();
unsigned RegBank = RBI.getRegBank(Reg, MRI, TRI)->getID();
LLT ValTy = MRI.getType(Reg);
const auto ValSize = ValTy.getSizeInBits();
assert((ValSize != 64 || TII.getSubtarget().hasVFP2()) &&
"Don't know how to load/store 64-bit value without VFP");
const auto NewOpc = selectLoadStoreOpCode(I.getOpcode(), RegBank, ValSize);
if (NewOpc == G_LOAD || NewOpc == G_STORE)
return false;
I.setDesc(TII.get(NewOpc));
if (NewOpc == ARM::LDRH || NewOpc == ARM::STRH)
// LDRH has a funny addressing mode (there's already a FIXME for it).
MIB.addReg(0);
MIB.addImm(0).add(predOps(ARMCC::AL));
break;
}
case G_MERGE_VALUES: {
if (!selectMergeValues(MIB, TII, MRI, TRI, RBI))
return false;
break;
}
case G_UNMERGE_VALUES: {
if (!selectUnmergeValues(MIB, TII, MRI, TRI, RBI))
return false;
break;
}
case G_BRCOND: {
if (!validReg(MRI, I.getOperand(0).getReg(), 1, ARM::GPRRegBankID)) {
DEBUG(dbgs() << "Unsupported condition register for G_BRCOND");
return false;
}
// Set the flags.
auto Test = BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(ARM::TSTri))
.addReg(I.getOperand(0).getReg())
.addImm(1)
.add(predOps(ARMCC::AL));
if (!constrainSelectedInstRegOperands(*Test, TII, TRI, RBI))
return false;
// Branch conditionally.
auto Branch = BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(ARM::Bcc))
.add(I.getOperand(1))
.add(predOps(ARMCC::EQ, ARM::CPSR));
if (!constrainSelectedInstRegOperands(*Branch, TII, TRI, RBI))
return false;
I.eraseFromParent();
return true;
}
default:
return false;
}
return constrainSelectedInstRegOperands(I, TII, TRI, RBI);
}
bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
DEBUG(dbgs() << "********** Register Stackifying **********\n"
"********** Function: "
<< MF.getName() << '\n');
bool Changed = false;
MachineRegisterInfo &MRI = MF.getRegInfo();
WebAssemblyFunctionInfo &MFI = *MF.getInfo<WebAssemblyFunctionInfo>();
// Walk the instructions from the bottom up. Currently we don't look past
// block boundaries, and the blocks aren't ordered so the block visitation
// order isn't significant, but we may want to change this in the future.
for (MachineBasicBlock &MBB : MF) {
for (MachineInstr &MI : reverse(MBB)) {
MachineInstr *Insert = &MI;
// Don't nest anything inside a phi.
if (Insert->getOpcode() == TargetOpcode::PHI)
break;
// Iterate through the inputs in reverse order, since we'll be pulling
// operands off the stack in FIFO order.
for (MachineOperand &Op : reverse(Insert->uses())) {
// We're only interested in explicit virtual register operands.
if (!Op.isReg() || Op.isImplicit())
continue;
unsigned Reg = Op.getReg();
if (!TargetRegisterInfo::isVirtualRegister(Reg))
continue;
// Only consider registers with a single definition.
// TODO: Eventually we may relax this, to stackify phi transfers.
MachineInstr *Def = MRI.getVRegDef(Reg);
if (!Def)
continue;
// There's no use in nesting implicit defs inside anything.
if (Def->getOpcode() == TargetOpcode::IMPLICIT_DEF)
continue;
// Argument instructions represent live-in registers and not real
// instructions.
if (Def->getOpcode() == WebAssembly::ARGUMENT_I32 ||
Def->getOpcode() == WebAssembly::ARGUMENT_I64 ||
Def->getOpcode() == WebAssembly::ARGUMENT_F32 ||
Def->getOpcode() == WebAssembly::ARGUMENT_F64)
continue;
// Single-use expression trees require defs that have one use, or that
// they be trivially clonable.
// TODO: Eventually we'll relax this, to take advantage of set_local
// returning its result.
bool OneUse = MRI.hasOneUse(Reg);
if (!OneUse && !Def->isMoveImmediate())
continue;
// For now, be conservative and don't look across block boundaries,
// unless we have something trivially clonable.
// TODO: Be more aggressive.
if (Def->getParent() != &MBB && !Def->isMoveImmediate())
continue;
// For now, be simple and don't reorder loads, stores, or side effects.
// TODO: Be more aggressive.
if ((Def->mayLoad() || Def->mayStore() ||
Def->hasUnmodeledSideEffects()))
continue;
Changed = true;
if (OneUse) {
// Move the def down and nest it in the current instruction.
MBB.insert(MachineBasicBlock::instr_iterator(Insert),
Def->removeFromParent());
MFI.stackifyVReg(Reg);
Insert = Def;
} else {
// Clone the def down and nest it in the current instruction.
MachineInstr *Clone = MF.CloneMachineInstr(Def);
unsigned OldReg = Def->getOperand(0).getReg();
unsigned NewReg = MRI.createVirtualRegister(MRI.getRegClass(OldReg));
assert(Op.getReg() == OldReg);
assert(Clone->getOperand(0).getReg() == OldReg);
Op.setReg(NewReg);
Clone->getOperand(0).setReg(NewReg);
MBB.insert(MachineBasicBlock::instr_iterator(Insert), Clone);
MFI.stackifyVReg(Reg);
Insert = Clone;
}
}
}
}
return Changed;
}
// The CC users in CCUsers are testing the result of a comparison of some
// value X against zero and we know that any CC value produced by MI
// would also reflect the value of X. Try to adjust CCUsers so that
// they test the result of MI directly, returning true on success.
// Leave everything unchanged on failure.
bool SystemZElimCompare::
adjustCCMasksForInstr(MachineInstr *MI, MachineInstr *Compare,
SmallVectorImpl<MachineInstr *> &CCUsers) {
int Opcode = MI->getOpcode();
const MCInstrDesc &Desc = TII->get(Opcode);
unsigned MIFlags = Desc.TSFlags;
// See which compare-style condition codes are available.
unsigned ReusableCCMask = SystemZII::getCompareZeroCCMask(MIFlags);
// For unsigned comparisons with zero, only equality makes sense.
unsigned CompareFlags = Compare->getDesc().TSFlags;
if (CompareFlags & SystemZII::IsLogical)
ReusableCCMask &= SystemZ::CCMASK_CMP_EQ;
if (ReusableCCMask == 0)
return false;
unsigned CCValues = SystemZII::getCCValues(MIFlags);
assert((ReusableCCMask & ~CCValues) == 0 && "Invalid CCValues");
// Now check whether these flags are enough for all users.
SmallVector<MachineOperand *, 4> AlterMasks;
for (unsigned int I = 0, E = CCUsers.size(); I != E; ++I) {
MachineInstr *MI = CCUsers[I];
// Fail if this isn't a use of CC that we understand.
unsigned Flags = MI->getDesc().TSFlags;
unsigned FirstOpNum;
if (Flags & SystemZII::CCMaskFirst)
FirstOpNum = 0;
else if (Flags & SystemZII::CCMaskLast)
FirstOpNum = MI->getNumExplicitOperands() - 2;
else
return false;
// Check whether the instruction predicate treats all CC values
// outside of ReusableCCMask in the same way. In that case it
// doesn't matter what those CC values mean.
unsigned CCValid = MI->getOperand(FirstOpNum).getImm();
unsigned CCMask = MI->getOperand(FirstOpNum + 1).getImm();
unsigned OutValid = ~ReusableCCMask & CCValid;
unsigned OutMask = ~ReusableCCMask & CCMask;
if (OutMask != 0 && OutMask != OutValid)
return false;
AlterMasks.push_back(&MI->getOperand(FirstOpNum));
AlterMasks.push_back(&MI->getOperand(FirstOpNum + 1));
}
// All users are OK. Adjust the masks for MI.
for (unsigned I = 0, E = AlterMasks.size(); I != E; I += 2) {
AlterMasks[I]->setImm(CCValues);
unsigned CCMask = AlterMasks[I + 1]->getImm();
if (CCMask & ~ReusableCCMask)
AlterMasks[I + 1]->setImm((CCMask & ReusableCCMask) |
(CCValues & ~ReusableCCMask));
}
// CC is now live after MI.
int CCDef = MI->findRegisterDefOperandIdx(SystemZ::CC, false, true, TRI);
assert(CCDef >= 0 && "Couldn't find CC set");
MI->getOperand(CCDef).setIsDead(false);
// Clear any intervening kills of CC.
MachineBasicBlock::iterator MBBI = MI, MBBE = Compare;
for (++MBBI; MBBI != MBBE; ++MBBI)
MBBI->clearRegisterKills(SystemZ::CC, TRI);
return true;
}
/// getTripCount - Return a loop-invariant LLVM value indicating the
/// number of times the loop will be executed. The trip count can
/// be either a register or a constant value. If the trip-count
/// cannot be determined, this returns null.
///
/// We find the trip count from the phi instruction that defines the
/// induction variable. We follow the links to the CMP instruction
/// to get the trip count.
///
/// Based upon getTripCount in LoopInfo.
///
CountValue *PPCCTRLoops::getTripCount(MachineLoop *L,
SmallVector<MachineInstr *, 2> &OldInsts) const {
MachineBasicBlock *LastMBB = L->getExitingBlock();
// Don't generate a CTR loop if the loop has more than one exit.
if (LastMBB == 0)
return 0;
MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
if (LastI->getOpcode() != PPC::BCC)
return 0;
// We need to make sure that this compare is defining the condition
// register actually used by the terminating branch.
unsigned PredReg = LastI->getOperand(1).getReg();
DEBUG(dbgs() << "Examining loop with first terminator: " << *LastI);
unsigned PredCond = LastI->getOperand(0).getImm();
if (PredCond != PPC::PRED_EQ && PredCond != PPC::PRED_NE)
return 0;
// Check that the loop has a induction variable.
SmallVector<MachineInstr *, 4> IVars, IOps;
getCanonicalInductionVariable(L, IVars, IOps);
for (unsigned i = 0; i < IVars.size(); ++i) {
MachineInstr *IOp = IOps[i];
MachineInstr *IV_Inst = IVars[i];
// Canonical loops will end with a 'cmpwi/cmpdi cr, IV, Imm',
// if Imm is 0, get the count from the PHI opnd
// if Imm is -M, than M is the count
// Otherwise, Imm is the count
MachineOperand *IV_Opnd;
const MachineOperand *InitialValue;
if (!L->contains(IV_Inst->getOperand(2).getMBB())) {
InitialValue = &IV_Inst->getOperand(1);
IV_Opnd = &IV_Inst->getOperand(3);
} else {
InitialValue = &IV_Inst->getOperand(3);
IV_Opnd = &IV_Inst->getOperand(1);
}
DEBUG(dbgs() << "Considering:\n");
DEBUG(dbgs() << " induction operation: " << *IOp);
DEBUG(dbgs() << " induction variable: " << *IV_Inst);
DEBUG(dbgs() << " initial value: " << *InitialValue << "\n");
// Look for the cmp instruction to determine if we
// can get a useful trip count. The trip count can
// be either a register or an immediate. The location
// of the value depends upon the type (reg or imm).
for (MachineRegisterInfo::reg_iterator
RI = MRI->reg_begin(IV_Opnd->getReg()), RE = MRI->reg_end();
RI != RE; ++RI) {
IV_Opnd = &RI.getOperand();
bool SignedCmp;
MachineInstr *MI = IV_Opnd->getParent();
if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp) &&
MI->getOperand(0).getReg() == PredReg) {
OldInsts.push_back(MI);
OldInsts.push_back(IOp);
DEBUG(dbgs() << " compare: " << *MI);
const MachineOperand &MO = MI->getOperand(2);
assert(MO.isImm() && "IV Cmp Operand should be an immediate");
int64_t ImmVal;
if (SignedCmp)
ImmVal = (short) MO.getImm();
else
ImmVal = MO.getImm();
const MachineInstr *IV_DefInstr = MRI->getVRegDef(IV_Opnd->getReg());
assert(L->contains(IV_DefInstr->getParent()) &&
"IV definition should occurs in loop");
int64_t iv_value = (short) IV_DefInstr->getOperand(2).getImm();
assert(InitialValue->isReg() && "Expecting register for init value");
unsigned InitialValueReg = InitialValue->getReg();
const MachineInstr *DefInstr = MRI->getVRegDef(InitialValueReg);
// Here we need to look for an immediate load (an li or lis/ori pair).
if (DefInstr && (DefInstr->getOpcode() == PPC::ORI8 ||
DefInstr->getOpcode() == PPC::ORI)) {
int64_t start = (short) DefInstr->getOperand(2).getImm();
const MachineInstr *DefInstr2 =
MRI->getVRegDef(DefInstr->getOperand(0).getReg());
if (DefInstr2 && (DefInstr2->getOpcode() == PPC::LIS8 ||
DefInstr2->getOpcode() == PPC::LIS)) {
DEBUG(dbgs() << " initial constant: " << *DefInstr);
DEBUG(dbgs() << " initial constant: " << *DefInstr2);
start |= int64_t(short(DefInstr2->getOperand(1).getImm())) << 16;
int64_t count = ImmVal - start;
if ((count % iv_value) != 0) {
return 0;
}
return new CountValue(count/iv_value);
}
} else if (DefInstr && (DefInstr->getOpcode() == PPC::LI8 ||
DefInstr->getOpcode() == PPC::LI)) {
DEBUG(dbgs() << " initial constant: " << *DefInstr);
int64_t count = ImmVal - int64_t(short(DefInstr->getOperand(1).getImm()));
if ((count % iv_value) != 0) {
return 0;
}
return new CountValue(count/iv_value);
} else if (iv_value == 1 || iv_value == -1) {
// We can't determine a constant starting value.
if (ImmVal == 0) {
return new CountValue(InitialValueReg, iv_value > 0);
}
// FIXME: handle non-zero end value.
}
// FIXME: handle non-unit increments (we might not want to introduce division
// but we can handle some 2^n cases with shifts).
}
}
}
return 0;
}
MipsInstrInfo::BranchType MipsInstrInfo::
AnalyzeBranch(MachineBasicBlock &MBB, MachineBasicBlock *&TBB,
MachineBasicBlock *&FBB, SmallVectorImpl<MachineOperand> &Cond,
bool AllowModify,
SmallVectorImpl<MachineInstr*> &BranchInstrs) const {
MachineBasicBlock::reverse_iterator I = MBB.rbegin(), REnd = MBB.rend();
// Skip all the debug instructions.
while (I != REnd && I->isDebugValue())
++I;
if (I == REnd || !isUnpredicatedTerminator(&*I)) {
// This block ends with no branches (it just falls through to its succ).
// Leave TBB/FBB null.
TBB = FBB = NULL;
return BT_NoBranch;
}
MachineInstr *LastInst = &*I;
unsigned LastOpc = LastInst->getOpcode();
BranchInstrs.push_back(LastInst);
// Not an analyzable branch (e.g., indirect jump).
if (!GetAnalyzableBrOpc(LastOpc))
return LastInst->isIndirectBranch() ? BT_Indirect : BT_None;
// Get the second to last instruction in the block.
unsigned SecondLastOpc = 0;
MachineInstr *SecondLastInst = NULL;
if (++I != REnd) {
SecondLastInst = &*I;
SecondLastOpc = GetAnalyzableBrOpc(SecondLastInst->getOpcode());
// Not an analyzable branch (must be an indirect jump).
if (isUnpredicatedTerminator(SecondLastInst) && !SecondLastOpc)
return BT_None;
}
// If there is only one terminator instruction, process it.
if (!SecondLastOpc) {
// Unconditional branch
if (LastOpc == UncondBrOpc) {
TBB = LastInst->getOperand(0).getMBB();
return BT_Uncond;
}
// Conditional branch
AnalyzeCondBr(LastInst, LastOpc, TBB, Cond);
return BT_Cond;
}
// If we reached here, there are two branches.
// If there are three terminators, we don't know what sort of block this is.
if (++I != REnd && isUnpredicatedTerminator(&*I))
return BT_None;
BranchInstrs.insert(BranchInstrs.begin(), SecondLastInst);
// If second to last instruction is an unconditional branch,
// analyze it and remove the last instruction.
if (SecondLastOpc == UncondBrOpc) {
// Return if the last instruction cannot be removed.
if (!AllowModify)
return BT_None;
TBB = SecondLastInst->getOperand(0).getMBB();
LastInst->eraseFromParent();
BranchInstrs.pop_back();
return BT_Uncond;
}
// Conditional branch followed by an unconditional branch.
// The last one must be unconditional.
if (LastOpc != UncondBrOpc)
return BT_None;
AnalyzeCondBr(SecondLastInst, SecondLastOpc, TBB, Cond);
FBB = LastInst->getOperand(0).getMBB();
return BT_CondUncond;
}
bool X86CallFrameOptimization::adjustCallSequence(MachineFunction &MF,
MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) {
// Check that this particular call sequence is amenable to the
// transformation.
const X86RegisterInfo &RegInfo = *static_cast<const X86RegisterInfo *>(
MF.getSubtarget().getRegisterInfo());
unsigned StackPtr = RegInfo.getStackRegister();
int FrameDestroyOpcode = TII->getCallFrameDestroyOpcode();
// We expect to enter this at the beginning of a call sequence
assert(I->getOpcode() == TII->getCallFrameSetupOpcode());
MachineBasicBlock::iterator FrameSetup = I++;
// For globals in PIC mode, we can have some LEAs here.
// Ignore them, they don't bother us.
// TODO: Extend this to something that covers more cases.
while (I->getOpcode() == X86::LEA32r)
++I;
// We expect a copy instruction here.
// TODO: The copy instruction is a lowering artifact.
// We should also support a copy-less version, where the stack
// pointer is used directly.
if (!I->isCopy() || !I->getOperand(0).isReg())
return false;
MachineBasicBlock::iterator SPCopy = I++;
StackPtr = SPCopy->getOperand(0).getReg();
// Scan the call setup sequence for the pattern we're looking for.
// We only handle a simple case - a sequence of MOV32mi or MOV32mr
// instructions, that push a sequence of 32-bit values onto the stack, with
// no gaps between them.
SmallVector<MachineInstr*, 4> MovVector(4, nullptr);
unsigned int MaxAdjust = FrameSetup->getOperand(0).getImm() / 4;
if (MaxAdjust > 4)
MovVector.resize(MaxAdjust, nullptr);
do {
int Opcode = I->getOpcode();
if (Opcode != X86::MOV32mi && Opcode != X86::MOV32mr)
break;
// We only want movs of the form:
// movl imm/r32, k(%esp)
// If we run into something else, bail.
// Note that AddrBaseReg may, counter to its name, not be a register,
// but rather a frame index.
// TODO: Support the fi case. This should probably work now that we
// have the infrastructure to track the stack pointer within a call
// sequence.
if (!I->getOperand(X86::AddrBaseReg).isReg() ||
(I->getOperand(X86::AddrBaseReg).getReg() != StackPtr) ||
!I->getOperand(X86::AddrScaleAmt).isImm() ||
(I->getOperand(X86::AddrScaleAmt).getImm() != 1) ||
(I->getOperand(X86::AddrIndexReg).getReg() != X86::NoRegister) ||
(I->getOperand(X86::AddrSegmentReg).getReg() != X86::NoRegister) ||
!I->getOperand(X86::AddrDisp).isImm())
return false;
int64_t StackDisp = I->getOperand(X86::AddrDisp).getImm();
assert(StackDisp >= 0 && "Negative stack displacement when passing parameters");
// We really don't want to consider the unaligned case.
if (StackDisp % 4)
return false;
StackDisp /= 4;
assert((size_t)StackDisp < MovVector.size() &&
"Function call has more parameters than the stack is adjusted for.");
// If the same stack slot is being filled twice, something's fishy.
if (MovVector[StackDisp] != nullptr)
return false;
MovVector[StackDisp] = I;
++I;
} while (I != MBB.end());
// We now expect the end of the sequence - a call and a stack adjust.
if (I == MBB.end())
return false;
// For PCrel calls, we expect an additional COPY of the basereg.
// If we find one, skip it.
if (I->isCopy()) {
if (I->getOperand(1).getReg() ==
MF.getInfo<X86MachineFunctionInfo>()->getGlobalBaseReg())
++I;
else
return false;
}
if (!I->isCall())
return false;
MachineBasicBlock::iterator Call = I;
if ((++I)->getOpcode() != FrameDestroyOpcode)
return false;
// Now, go through the vector, and see that we don't have any gaps,
// but only a series of 32-bit MOVs.
int64_t ExpectedDist = 0;
auto MMI = MovVector.begin(), MME = MovVector.end();
for (; MMI != MME; ++MMI, ExpectedDist += 4)
if (*MMI == nullptr)
break;
// If the call had no parameters, do nothing
if (!ExpectedDist)
return false;
// We are either at the last parameter, or a gap.
// Make sure it's not a gap
for (; MMI != MME; ++MMI)
if (*MMI != nullptr)
return false;
// Ok, we can in fact do the transformation for this call.
// Do not remove the FrameSetup instruction, but adjust the parameters.
// PEI will end up finalizing the handling of this.
FrameSetup->getOperand(1).setImm(ExpectedDist);
DebugLoc DL = I->getDebugLoc();
// Now, iterate through the vector in reverse order, and replace the movs
// with pushes. MOVmi/MOVmr doesn't have any defs, so no need to
// replace uses.
for (int Idx = (ExpectedDist / 4) - 1; Idx >= 0; --Idx) {
MachineBasicBlock::iterator MOV = *MovVector[Idx];
MachineOperand PushOp = MOV->getOperand(X86::AddrNumOperands);
if (MOV->getOpcode() == X86::MOV32mi) {
unsigned PushOpcode = X86::PUSHi32;
// If the operand is a small (8-bit) immediate, we can use a
// PUSH instruction with a shorter encoding.
// Note that isImm() may fail even though this is a MOVmi, because
// the operand can also be a symbol.
if (PushOp.isImm()) {
int64_t Val = PushOp.getImm();
if (isInt<8>(Val))
PushOpcode = X86::PUSH32i8;
}
BuildMI(MBB, Call, DL, TII->get(PushOpcode)).addOperand(PushOp);
} else {
unsigned int Reg = PushOp.getReg();
// If PUSHrmm is not slow on this target, try to fold the source of the
// push into the instruction.
const X86Subtarget &ST = MF.getSubtarget<X86Subtarget>();
bool SlowPUSHrmm = ST.isAtom() || ST.isSLM();
// Check that this is legal to fold. Right now, we're extremely
// conservative about that.
MachineInstr *DefMov = nullptr;
if (!SlowPUSHrmm && (DefMov = canFoldIntoRegPush(FrameSetup, Reg))) {
MachineInstr *Push = BuildMI(MBB, Call, DL, TII->get(X86::PUSH32rmm));
unsigned NumOps = DefMov->getDesc().getNumOperands();
for (unsigned i = NumOps - X86::AddrNumOperands; i != NumOps; ++i)
Push->addOperand(DefMov->getOperand(i));
DefMov->eraseFromParent();
} else {
BuildMI(MBB, Call, DL, TII->get(X86::PUSH32r)).addReg(Reg).getInstr();
}
}
MBB.erase(MOV);
}
// The stack-pointer copy is no longer used in the call sequences.
// There should not be any other users, but we can't commit to that, so:
if (MRI->use_empty(SPCopy->getOperand(0).getReg()))
SPCopy->eraseFromParent();
// Once we've done this, we need to make sure PEI doesn't assume a reserved
// frame.
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
FuncInfo->setHasPushSequences(true);
return true;
}
bool DeadMachineInstructionElim::runOnMachineFunction(MachineFunction &MF) {
bool AnyChanges = false;
MRI = &MF.getRegInfo();
TRI = MF.getTarget().getRegisterInfo();
TII = MF.getTarget().getInstrInfo();
// Treat reserved registers as always live.
BitVector ReservedRegs = TRI->getReservedRegs(MF);
// Loop over all instructions in all blocks, from bottom to top, so that it's
// more likely that chains of dependent but ultimately dead instructions will
// be cleaned up.
for (MachineFunction::reverse_iterator I = MF.rbegin(), E = MF.rend();
I != E; ++I) {
MachineBasicBlock *MBB = &*I;
// Start out assuming that reserved registers are live out of this block.
LivePhysRegs = ReservedRegs;
// Also add any explicit live-out physregs for this block.
if (!MBB->empty() && MBB->back().getDesc().isReturn())
for (MachineRegisterInfo::liveout_iterator LOI = MRI->liveout_begin(),
LOE = MRI->liveout_end(); LOI != LOE; ++LOI) {
unsigned Reg = *LOI;
if (TargetRegisterInfo::isPhysicalRegister(Reg))
LivePhysRegs.set(Reg);
}
// FIXME: Add live-ins from sucessors to LivePhysRegs. Normally, physregs
// are not live across blocks, but some targets (x86) can have flags live
// out of a block.
// Now scan the instructions and delete dead ones, tracking physreg
// liveness as we go.
for (MachineBasicBlock::reverse_iterator MII = MBB->rbegin(),
MIE = MBB->rend(); MII != MIE; ) {
MachineInstr *MI = &*MII;
// If the instruction is dead, delete it!
if (isDead(MI)) {
DEBUG(dbgs() << "DeadMachineInstructionElim: DELETING: " << *MI);
// It is possible that some DBG_VALUE instructions refer to this
// instruction. Examine each def operand for such references;
// if found, mark the DBG_VALUE as undef (but don't delete it).
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (!MO.isReg() || !MO.isDef())
continue;
unsigned Reg = MO.getReg();
if (!TargetRegisterInfo::isVirtualRegister(Reg))
continue;
MachineRegisterInfo::use_iterator nextI;
for (MachineRegisterInfo::use_iterator I = MRI->use_begin(Reg),
E = MRI->use_end(); I!=E; I=nextI) {
nextI = llvm::next(I); // I is invalidated by the setReg
MachineOperand& Use = I.getOperand();
MachineInstr *UseMI = Use.getParent();
if (UseMI==MI)
continue;
assert(Use.isDebug());
UseMI->getOperand(0).setReg(0U);
}
}
AnyChanges = true;
MI->eraseFromParent();
++NumDeletes;
MIE = MBB->rend();
// MII is now pointing to the next instruction to process,
// so don't increment it.
continue;
}
// Record the physreg defs.
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isDef()) {
unsigned Reg = MO.getReg();
if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
LivePhysRegs.reset(Reg);
// Check the subreg set, not the alias set, because a def
// of a super-register may still be partially live after
// this def.
for (const unsigned *SubRegs = TRI->getSubRegisters(Reg);
*SubRegs; ++SubRegs)
LivePhysRegs.reset(*SubRegs);
}
}
}
// Record the physreg uses, after the defs, in case a physreg is
// both defined and used in the same instruction.
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
const MachineOperand &MO = MI->getOperand(i);
if (MO.isReg() && MO.isUse()) {
unsigned Reg = MO.getReg();
if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
LivePhysRegs.set(Reg);
for (const unsigned *AliasSet = TRI->getAliasSet(Reg);
*AliasSet; ++AliasSet)
LivePhysRegs.set(*AliasSet);
}
}
}
// We didn't delete the current instruction, so increment MII to
// the next one.
++MII;
}
}
LivePhysRegs.clear();
return AnyChanges;
}
RegisterBankInfo::InstructionMapping
RegisterBankInfo::getInstrMappingImpl(const MachineInstr &MI) const {
// For copies we want to walk over the operands and try to find one
// that has a register bank since the instruction itself will not get
// us any constraint.
bool IsCopyLike = isCopyLike(MI);
// For copy like instruction, only the mapping of the definition
// is important. The rest is not constrained.
unsigned NumOperandsForMapping = IsCopyLike ? 1 : MI.getNumOperands();
RegisterBankInfo::InstructionMapping Mapping(DefaultMappingID, /*Cost*/ 1,
/*OperandsMapping*/ nullptr,
NumOperandsForMapping);
const MachineFunction &MF = *MI.getParent()->getParent();
const TargetSubtargetInfo &STI = MF.getSubtarget();
const TargetRegisterInfo &TRI = *STI.getRegisterInfo();
const MachineRegisterInfo &MRI = MF.getRegInfo();
// We may need to query the instruction encoding to guess the mapping.
const TargetInstrInfo &TII = *STI.getInstrInfo();
// Before doing anything complicated check if the mapping is not
// directly available.
bool CompleteMapping = true;
SmallVector<const ValueMapping *, 8> OperandsMapping(NumOperandsForMapping);
for (unsigned OpIdx = 0, EndIdx = MI.getNumOperands(); OpIdx != EndIdx;
++OpIdx) {
const MachineOperand &MO = MI.getOperand(OpIdx);
if (!MO.isReg())
continue;
unsigned Reg = MO.getReg();
if (!Reg)
continue;
// The register bank of Reg is just a side effect of the current
// excution and in particular, there is no reason to believe this
// is the best default mapping for the current instruction. Keep
// it as an alternative register bank if we cannot figure out
// something.
const RegisterBank *AltRegBank = getRegBank(Reg, MRI, TRI);
// For copy-like instruction, we want to reuse the register bank
// that is already set on Reg, if any, since those instructions do
// not have any constraints.
const RegisterBank *CurRegBank = IsCopyLike ? AltRegBank : nullptr;
if (!CurRegBank) {
// If this is a target specific instruction, we can deduce
// the register bank from the encoding constraints.
CurRegBank = getRegBankFromConstraints(MI, OpIdx, TII, TRI);
if (!CurRegBank) {
// All our attempts failed, give up.
CompleteMapping = false;
if (!IsCopyLike)
// MI does not carry enough information to guess the mapping.
return InstructionMapping();
continue;
}
}
const ValueMapping *ValMapping =
&getValueMapping(0, getSizeInBits(Reg, MRI, TRI), *CurRegBank);
if (IsCopyLike) {
OperandsMapping[0] = ValMapping;
CompleteMapping = true;
break;
}
OperandsMapping[OpIdx] = ValMapping;
}
if (IsCopyLike && !CompleteMapping)
// No way to deduce the type from what we have.
return InstructionMapping();
assert(CompleteMapping && "Setting an uncomplete mapping");
Mapping.setOperandsMapping(getOperandsMapping(OperandsMapping));
return Mapping;
}
void MSP430FrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
MSP430MachineFunctionInfo *MSP430FI = MF.getInfo<MSP430MachineFunctionInfo>();
const MSP430InstrInfo &TII =
*static_cast<const MSP430InstrInfo *>(MF.getSubtarget().getInstrInfo());
MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
unsigned RetOpcode = MBBI->getOpcode();
DebugLoc DL = MBBI->getDebugLoc();
switch (RetOpcode) {
case MSP430::RET:
case MSP430::RETI: break; // These are ok
default:
llvm_unreachable("Can only insert epilog into returning blocks");
}
// Get the number of bytes to allocate from the FrameInfo
uint64_t StackSize = MFI->getStackSize();
unsigned CSSize = MSP430FI->getCalleeSavedFrameSize();
uint64_t NumBytes = 0;
if (hasFP(MF)) {
// Calculate required stack adjustment
uint64_t FrameSize = StackSize - 2;
NumBytes = FrameSize - CSSize;
// pop FP.
BuildMI(MBB, MBBI, DL, TII.get(MSP430::POP16r), MSP430::FP);
} else
NumBytes = StackSize - CSSize;
// Skip the callee-saved pop instructions.
while (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PI = std::prev(MBBI);
unsigned Opc = PI->getOpcode();
if (Opc != MSP430::POP16r && !PI->isTerminator())
break;
--MBBI;
}
DL = MBBI->getDebugLoc();
// If there is an ADD16ri or SUB16ri of SP immediately before this
// instruction, merge the two instructions.
//if (NumBytes || MFI->hasVarSizedObjects())
// mergeSPUpdatesUp(MBB, MBBI, StackPtr, &NumBytes);
if (MFI->hasVarSizedObjects()) {
BuildMI(MBB, MBBI, DL,
TII.get(MSP430::MOV16rr), MSP430::SP).addReg(MSP430::FP);
if (CSSize) {
MachineInstr *MI =
BuildMI(MBB, MBBI, DL,
TII.get(MSP430::SUB16ri), MSP430::SP)
.addReg(MSP430::SP).addImm(CSSize);
// The SRW implicit def is dead.
MI->getOperand(3).setIsDead();
}
} else {
// adjust stack pointer back: SP += numbytes
if (NumBytes) {
MachineInstr *MI =
BuildMI(MBB, MBBI, DL, TII.get(MSP430::ADD16ri), MSP430::SP)
.addReg(MSP430::SP).addImm(NumBytes);
// The SRW implicit def is dead.
MI->getOperand(3).setIsDead();
}
}
}
void X86ExpandPseudo::ExpandICallBranchFunnel(
MachineBasicBlock *MBB, MachineBasicBlock::iterator MBBI) {
MachineBasicBlock *JTMBB = MBB;
MachineInstr *JTInst = &*MBBI;
MachineFunction *MF = MBB->getParent();
const BasicBlock *BB = MBB->getBasicBlock();
auto InsPt = MachineFunction::iterator(MBB);
++InsPt;
std::vector<std::pair<MachineBasicBlock *, unsigned>> TargetMBBs;
DebugLoc DL = JTInst->getDebugLoc();
MachineOperand Selector = JTInst->getOperand(0);
const GlobalValue *CombinedGlobal = JTInst->getOperand(1).getGlobal();
auto CmpTarget = [&](unsigned Target) {
BuildMI(*MBB, MBBI, DL, TII->get(X86::LEA64r), X86::R11)
.addReg(X86::RIP)
.addImm(1)
.addReg(0)
.addGlobalAddress(CombinedGlobal,
JTInst->getOperand(2 + 2 * Target).getImm())
.addReg(0);
BuildMI(*MBB, MBBI, DL, TII->get(X86::CMP64rr))
.add(Selector)
.addReg(X86::R11);
};
auto CreateMBB = [&]() {
auto *NewMBB = MF->CreateMachineBasicBlock(BB);
MBB->addSuccessor(NewMBB);
return NewMBB;
};
auto EmitCondJump = [&](unsigned Opcode, MachineBasicBlock *ThenMBB) {
BuildMI(*MBB, MBBI, DL, TII->get(Opcode)).addMBB(ThenMBB);
auto *ElseMBB = CreateMBB();
MF->insert(InsPt, ElseMBB);
MBB = ElseMBB;
MBBI = MBB->end();
};
auto EmitCondJumpTarget = [&](unsigned Opcode, unsigned Target) {
auto *ThenMBB = CreateMBB();
TargetMBBs.push_back({ThenMBB, Target});
EmitCondJump(Opcode, ThenMBB);
};
auto EmitTailCall = [&](unsigned Target) {
BuildMI(*MBB, MBBI, DL, TII->get(X86::TAILJMPd64))
.add(JTInst->getOperand(3 + 2 * Target));
};
std::function<void(unsigned, unsigned)> EmitBranchFunnel =
[&](unsigned FirstTarget, unsigned NumTargets) {
if (NumTargets == 1) {
EmitTailCall(FirstTarget);
return;
}
if (NumTargets == 2) {
CmpTarget(FirstTarget + 1);
EmitCondJumpTarget(X86::JB_1, FirstTarget);
EmitTailCall(FirstTarget + 1);
return;
}
if (NumTargets < 6) {
CmpTarget(FirstTarget + 1);
EmitCondJumpTarget(X86::JB_1, FirstTarget);
EmitCondJumpTarget(X86::JE_1, FirstTarget + 1);
EmitBranchFunnel(FirstTarget + 2, NumTargets - 2);
return;
}
auto *ThenMBB = CreateMBB();
CmpTarget(FirstTarget + (NumTargets / 2));
EmitCondJump(X86::JB_1, ThenMBB);
EmitCondJumpTarget(X86::JE_1, FirstTarget + (NumTargets / 2));
EmitBranchFunnel(FirstTarget + (NumTargets / 2) + 1,
NumTargets - (NumTargets / 2) - 1);
MF->insert(InsPt, ThenMBB);
MBB = ThenMBB;
MBBI = MBB->end();
EmitBranchFunnel(FirstTarget, NumTargets / 2);
};
EmitBranchFunnel(0, (JTInst->getNumOperands() - 2) / 2);
for (auto P : TargetMBBs) {
MF->insert(InsPt, P.first);
BuildMI(P.first, DL, TII->get(X86::TAILJMPd64))
.add(JTInst->getOperand(3 + 2 * P.second));
}
JTMBB->erase(JTInst);
}
MachineBasicBlock::iterator MSP430FrameLowering::eliminateCallFramePseudoInstr(
MachineFunction &MF, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const {
const MSP430InstrInfo &TII =
*static_cast<const MSP430InstrInfo *>(MF.getSubtarget().getInstrInfo());
unsigned StackAlign = getStackAlignment();
if (!hasReservedCallFrame(MF)) {
// If the stack pointer can be changed after prologue, turn the
// adjcallstackup instruction into a 'sub SP, <amt>' and the
// adjcallstackdown instruction into 'add SP, <amt>'
// TODO: consider using push / pop instead of sub + store / add
MachineInstr *Old = I;
uint64_t Amount = Old->getOperand(0).getImm();
if (Amount != 0) {
// We need to keep the stack aligned properly. To do this, we round the
// amount of space needed for the outgoing arguments up to the next
// alignment boundary.
Amount = (Amount+StackAlign-1)/StackAlign*StackAlign;
MachineInstr *New = nullptr;
if (Old->getOpcode() == TII.getCallFrameSetupOpcode()) {
New = BuildMI(MF, Old->getDebugLoc(),
TII.get(MSP430::SUB16ri), MSP430::SP)
.addReg(MSP430::SP).addImm(Amount);
} else {
assert(Old->getOpcode() == TII.getCallFrameDestroyOpcode());
// factor out the amount the callee already popped.
uint64_t CalleeAmt = Old->getOperand(1).getImm();
Amount -= CalleeAmt;
if (Amount)
New = BuildMI(MF, Old->getDebugLoc(),
TII.get(MSP430::ADD16ri), MSP430::SP)
.addReg(MSP430::SP).addImm(Amount);
}
if (New) {
// The SRW implicit def is dead.
New->getOperand(3).setIsDead();
// Replace the pseudo instruction with a new instruction...
MBB.insert(I, New);
}
}
} else if (I->getOpcode() == TII.getCallFrameDestroyOpcode()) {
// If we are performing frame pointer elimination and if the callee pops
// something off the stack pointer, add it back.
if (uint64_t CalleeAmt = I->getOperand(1).getImm()) {
MachineInstr *Old = I;
MachineInstr *New =
BuildMI(MF, Old->getDebugLoc(), TII.get(MSP430::SUB16ri),
MSP430::SP).addReg(MSP430::SP).addImm(CalleeAmt);
// The SRW implicit def is dead.
New->getOperand(3).setIsDead();
MBB.insert(I, New);
}
}
return MBB.erase(I);
}
/// scavengeFrameVirtualRegs - Replace all frame index virtual registers
/// with physical registers. Use the register scavenger to find an
/// appropriate register to use.
///
/// FIXME: Iterating over the instruction stream is unnecessary. We can simply
/// iterate over the vreg use list, which at this point only contains machine
/// operands for which eliminateFrameIndex need a new scratch reg.
void
PEI::scavengeFrameVirtualRegs(MachineFunction &Fn) {
// Run through the instructions and find any virtual registers.
for (MachineFunction::iterator BB = Fn.begin(),
E = Fn.end(); BB != E; ++BB) {
RS->enterBasicBlock(BB);
int SPAdj = 0;
// The instruction stream may change in the loop, so check BB->end()
// directly.
for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); ) {
// We might end up here again with a NULL iterator if we scavenged a
// register for which we inserted spill code for definition by what was
// originally the first instruction in BB.
if (I == MachineBasicBlock::iterator(nullptr))
I = BB->begin();
MachineInstr *MI = I;
MachineBasicBlock::iterator J = std::next(I);
MachineBasicBlock::iterator P =
I == BB->begin() ? MachineBasicBlock::iterator(nullptr)
: std::prev(I);
// RS should process this instruction before we might scavenge at this
// location. This is because we might be replacing a virtual register
// defined by this instruction, and if so, registers killed by this
// instruction are available, and defined registers are not.
RS->forward(I);
for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
if (MI->getOperand(i).isReg()) {
MachineOperand &MO = MI->getOperand(i);
unsigned Reg = MO.getReg();
if (Reg == 0)
continue;
if (!TargetRegisterInfo::isVirtualRegister(Reg))
continue;
// When we first encounter a new virtual register, it
// must be a definition.
assert(MI->getOperand(i).isDef() &&
"frame index virtual missing def!");
// Scavenge a new scratch register
const TargetRegisterClass *RC = Fn.getRegInfo().getRegClass(Reg);
unsigned ScratchReg = RS->scavengeRegister(RC, J, SPAdj);
++NumScavengedRegs;
// Replace this reference to the virtual register with the
// scratch register.
assert (ScratchReg && "Missing scratch register!");
MachineRegisterInfo &MRI = Fn.getRegInfo();
Fn.getRegInfo().replaceRegWith(Reg, ScratchReg);
// Make sure MRI now accounts this register as used.
MRI.setPhysRegUsed(ScratchReg);
// Because this instruction was processed by the RS before this
// register was allocated, make sure that the RS now records the
// register as being used.
RS->setRegUsed(ScratchReg);
}
}
// If the scavenger needed to use one of its spill slots, the
// spill code will have been inserted in between I and J. This is a
// problem because we need the spill code before I: Move I to just
// prior to J.
if (I != std::prev(J)) {
BB->splice(J, BB, I);
// Before we move I, we need to prepare the RS to visit I again.
// Specifically, RS will assert if it sees uses of registers that
// it believes are undefined. Because we have already processed
// register kills in I, when it visits I again, it will believe that
// those registers are undefined. To avoid this situation, unprocess
// the instruction I.
assert(RS->getCurrentPosition() == I &&
"The register scavenger has an unexpected position");
I = P;
RS->unprocess(P);
} else
++I;
}
}
}
/// TailDuplicateAndUpdate - Tail duplicate the block and cleanup.
bool
TailDuplicatePass::TailDuplicateAndUpdate(MachineBasicBlock *MBB,
bool IsSimple,
MachineFunction &MF) {
// Save the successors list.
SmallSetVector<MachineBasicBlock*, 8> Succs(MBB->succ_begin(),
MBB->succ_end());
SmallVector<MachineBasicBlock*, 8> TDBBs;
SmallVector<MachineInstr*, 16> Copies;
if (!TailDuplicate(MBB, IsSimple, MF, TDBBs, Copies))
return false;
++NumTails;
SmallVector<MachineInstr*, 8> NewPHIs;
MachineSSAUpdater SSAUpdate(MF, &NewPHIs);
// TailBB's immediate successors are now successors of those predecessors
// which duplicated TailBB. Add the predecessors as sources to the PHI
// instructions.
bool isDead = MBB->pred_empty() && !MBB->hasAddressTaken();
if (PreRegAlloc)
UpdateSuccessorsPHIs(MBB, isDead, TDBBs, Succs);
// If it is dead, remove it.
if (isDead) {
NumInstrDups -= MBB->size();
RemoveDeadBlock(MBB);
++NumDeadBlocks;
}
// Update SSA form.
if (!SSAUpdateVRs.empty()) {
for (unsigned i = 0, e = SSAUpdateVRs.size(); i != e; ++i) {
unsigned VReg = SSAUpdateVRs[i];
SSAUpdate.Initialize(VReg);
// If the original definition is still around, add it as an available
// value.
MachineInstr *DefMI = MRI->getVRegDef(VReg);
MachineBasicBlock *DefBB = 0;
if (DefMI) {
DefBB = DefMI->getParent();
SSAUpdate.AddAvailableValue(DefBB, VReg);
}
// Add the new vregs as available values.
DenseMap<unsigned, AvailableValsTy>::iterator LI =
SSAUpdateVals.find(VReg);
for (unsigned j = 0, ee = LI->second.size(); j != ee; ++j) {
MachineBasicBlock *SrcBB = LI->second[j].first;
unsigned SrcReg = LI->second[j].second;
SSAUpdate.AddAvailableValue(SrcBB, SrcReg);
}
// Rewrite uses that are outside of the original def's block.
MachineRegisterInfo::use_iterator UI = MRI->use_begin(VReg);
while (UI != MRI->use_end()) {
MachineOperand &UseMO = UI.getOperand();
MachineInstr *UseMI = &*UI;
++UI;
if (UseMI->isDebugValue()) {
// SSAUpdate can replace the use with an undef. That creates
// a debug instruction that is a kill.
// FIXME: Should it SSAUpdate job to delete debug instructions
// instead of replacing the use with undef?
UseMI->eraseFromParent();
continue;
}
if (UseMI->getParent() == DefBB && !UseMI->isPHI())
continue;
SSAUpdate.RewriteUse(UseMO);
}
}
SSAUpdateVRs.clear();
SSAUpdateVals.clear();
}
// Eliminate some of the copies inserted by tail duplication to maintain
// SSA form.
for (unsigned i = 0, e = Copies.size(); i != e; ++i) {
MachineInstr *Copy = Copies[i];
if (!Copy->isCopy())
continue;
unsigned Dst = Copy->getOperand(0).getReg();
unsigned Src = Copy->getOperand(1).getReg();
if (MRI->hasOneNonDBGUse(Src) &&
MRI->constrainRegClass(Src, MRI->getRegClass(Dst))) {
// Copy is the only use. Do trivial copy propagation here.
MRI->replaceRegWith(Dst, Src);
Copy->eraseFromParent();
}
}
if (NewPHIs.size())
NumAddedPHIs += NewPHIs.size();
return true;
}
