void WebAssemblyFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
uint64_t StackSize = MF.getFrameInfo()->getStackSize();
if (!StackSize)
return;
const auto *TII = MF.getSubtarget().getInstrInfo();
auto &MRI = MF.getRegInfo();
unsigned OffsetReg = MRI.createVirtualRegister(&WebAssembly::I32RegClass);
auto InsertPt = MBB.getFirstTerminator();
DebugLoc DL;
if (InsertPt != MBB.end()) {
DL = InsertPt->getDebugLoc();
}
// Restore the stack pointer. Without FP its value is just SP32 - stacksize
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::CONST_I32), OffsetReg)
.addImm(StackSize);
auto *SPSymbol = MF.createExternalSymbolName("__stack_pointer");
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::ADD_I32), WebAssembly::SP32)
.addReg(WebAssembly::SP32)
.addReg(OffsetReg);
// Re-use OffsetReg to hold the address of the stacktop
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::CONST_I32), OffsetReg)
.addExternalSymbol(SPSymbol);
auto *MMO = new MachineMemOperand(MachinePointerInfo(),
MachineMemOperand::MOStore, 4, 4);
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::STORE_I32), WebAssembly::SP32)
.addImm(0)
.addReg(OffsetReg)
.addReg(WebAssembly::SP32)
.addMemOperand(MMO);
}
void HexagonFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = std::prev(MBB.end());
DebugLoc dl = MBBI->getDebugLoc();
//
// Only insert deallocframe if we need to. Also at -O0. See comment
// in emitPrologue above.
//
if (hasFP(MF) || MF.getTarget().getOptLevel() == CodeGenOpt::None) {
MachineBasicBlock::iterator MBBI = std::prev(MBB.end());
MachineBasicBlock::iterator MBBI_end = MBB.end();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
// Handle EH_RETURN.
if (MBBI->getOpcode() == Hexagon::EH_RETURN_JMPR) {
assert(MBBI->getOperand(0).isReg() && "Offset should be in register!");
BuildMI(MBB, MBBI, dl, TII.get(Hexagon::L2_deallocframe));
BuildMI(MBB, MBBI, dl, TII.get(Hexagon::A2_add),
Hexagon::R29).addReg(Hexagon::R29).addReg(Hexagon::R28);
return;
}
// Replace 'jumpr r31' instruction with dealloc_return for V4 and higher
// versions.
if (MF.getSubtarget<HexagonSubtarget>().hasV4TOps() &&
MBBI->getOpcode() == Hexagon::JMPret && !DisableDeallocRet) {
// Check for RESTORE_DEALLOC_RET_JMP_V4 call. Don't emit an extra DEALLOC
// instruction if we encounter it.
MachineBasicBlock::iterator BeforeJMPR =
MBB.begin() == MBBI ? MBBI : std::prev(MBBI);
if (BeforeJMPR != MBBI &&
BeforeJMPR->getOpcode() == Hexagon::RESTORE_DEALLOC_RET_JMP_V4) {
// Remove the JMPR node.
MBB.erase(MBBI);
return;
}
// Add dealloc_return.
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI_end, dl, TII.get(Hexagon::L4_return));
// Transfer the function live-out registers.
MIB->copyImplicitOps(*MBB.getParent(), &*MBBI);
// Remove the JUMPR node.
MBB.erase(MBBI);
} else { // Add deallocframe for V2 and V3, and V4 tail calls.
// Check for RESTORE_DEALLOC_BEFORE_TAILCALL_V4. We don't need an extra
// DEALLOCFRAME instruction after it.
MachineBasicBlock::iterator Term = MBB.getFirstTerminator();
MachineBasicBlock::iterator I =
Term == MBB.begin() ? MBB.end() : std::prev(Term);
if (I != MBB.end() &&
I->getOpcode() == Hexagon::RESTORE_DEALLOC_BEFORE_TAILCALL_V4)
return;
BuildMI(MBB, MBBI, dl, TII.get(Hexagon::L2_deallocframe));
}
}
}
static bool hasTerminatorThatModifiesExec(const MachineBasicBlock &MBB,
const TargetRegisterInfo &TRI) {
for (MachineBasicBlock::const_iterator I = MBB.getFirstTerminator(),
E = MBB.end(); I != E; ++I) {
if (I->modifiesRegister(AMDGPU::EXEC, &TRI))
return true;
}
return false;
}
/// enterIntvAtEnd - Enter openli at the end of MBB.
/// PhiMBB is a successor inside openli where a PHI value is created.
/// Currently, all entries must share the same PhiMBB.
void SplitEditor::enterIntvAtEnd(MachineBasicBlock &A, MachineBasicBlock &B) {
assert(openli_ && "openIntv not called before enterIntvAtEnd");
SlotIndex EndA = lis_.getMBBEndIdx(&A);
VNInfo *CurVNIA = curli_->getVNInfoAt(EndA.getPrevIndex());
if (!CurVNIA) {
DEBUG(dbgs() << " enterIntvAtEnd, curli not live out of BB#"
<< A.getNumber() << ".\n");
return;
}
// Add a phi kill value and live range out of A.
VNInfo *VNIA = insertCopy(*openli_, A, A.getFirstTerminator());
openli_->addRange(LiveRange(VNIA->def, EndA, VNIA));
// FIXME: If this is the only entry edge, we don't need the extra PHI value.
// FIXME: If there are multiple entry blocks (so not a loop), we need proper
// SSA update.
// Now look at the start of B.
SlotIndex StartB = lis_.getMBBStartIdx(&B);
SlotIndex EndB = lis_.getMBBEndIdx(&B);
const LiveRange *CurB = curli_->getLiveRangeContaining(StartB);
if (!CurB) {
DEBUG(dbgs() << " enterIntvAtEnd: curli not live in to BB#"
<< B.getNumber() << ".\n");
return;
}
VNInfo *VNIB = openli_->getVNInfoAt(StartB);
if (!VNIB) {
// Create a phi value.
VNIB = openli_->getNextValue(SlotIndex(StartB, true), 0, false,
lis_.getVNInfoAllocator());
VNIB->setIsPHIDef(true);
VNInfo *&mapVNI = valueMap_[CurB->valno];
if (mapVNI) {
// Multiple copies - must create PHI value.
abort();
} else {
// This is the first copy of dupLR. Mark the mapping.
mapVNI = VNIB;
}
}
DEBUG(dbgs() << " enterIntvAtEnd: " << *openli_ << '\n');
}
void RegAllocFast::allocateBasicBlock(MachineBasicBlock &MBB) {
this->MBB = &MBB;
LLVM_DEBUG(dbgs() << "\nAllocating " << MBB);
PhysRegState.assign(TRI->getNumRegs(), regDisabled);
assert(LiveVirtRegs.empty() && "Mapping not cleared from last block?");
MachineBasicBlock::iterator MII = MBB.begin();
// Add live-in registers as live.
for (const MachineBasicBlock::RegisterMaskPair LI : MBB.liveins())
if (MRI->isAllocatable(LI.PhysReg))
definePhysReg(MII, LI.PhysReg, regReserved);
VirtDead.clear();
Coalesced.clear();
// Otherwise, sequentially allocate each instruction in the MBB.
for (MachineInstr &MI : MBB) {
LLVM_DEBUG(
dbgs() << "\n>> " << MI << "Regs:";
dumpState()
);
// Special handling for debug values. Note that they are not allowed to
// affect codegen of the other instructions in any way.
if (MI.isDebugValue()) {
handleDebugValue(MI);
continue;
}
allocateInstruction(MI);
}
// Spill all physical registers holding virtual registers now.
LLVM_DEBUG(dbgs() << "Spilling live registers at end of block.\n");
spillAll(MBB.getFirstTerminator());
// Erase all the coalesced copies. We are delaying it until now because
// LiveVirtRegs might refer to the instrs.
for (MachineInstr *MI : Coalesced)
MBB.erase(MI);
NumCoalesced += Coalesced.size();
LLVM_DEBUG(MBB.dump());
}
/// When an instruction is found to only use loop invariant operands that is
/// safe to hoist, this instruction is called to do the dirty work.
void MachineLICM::HoistPostRA(MachineInstr *MI, unsigned Def) {
MachineBasicBlock *Preheader = getCurPreheader();
// Now move the instructions to the predecessor, inserting it before any
// terminator instructions.
DEBUG(dbgs() << "Hoisting to BB#" << Preheader->getNumber() << " from BB#"
<< MI->getParent()->getNumber() << ": " << *MI);
// Splice the instruction to the preheader.
MachineBasicBlock *MBB = MI->getParent();
Preheader->splice(Preheader->getFirstTerminator(), MBB, MI);
// Add register to livein list to all the BBs in the current loop since a
// loop invariant must be kept live throughout the whole loop. This is
// important to ensure later passes do not scavenge the def register.
AddToLiveIns(Def);
++NumPostRAHoisted;
Changed = true;
}
// FindCopyInsertPoint - Find a safe place in MBB to insert a copy from SrcReg
// when following the CFG edge to SuccMBB. This needs to be after any def of
// SrcReg, but before any subsequent point where control flow might jump out of
// the basic block.
MachineBasicBlock::iterator
llvm::PHIElimination::FindCopyInsertPoint(MachineBasicBlock &MBB,
MachineBasicBlock &SuccMBB,
unsigned SrcReg) {
// Handle the trivial case trivially.
if (MBB.empty())
return MBB.begin();
// Usually, we just want to insert the copy before the first terminator
// instruction. However, for the edge going to a landing pad, we must insert
// the copy before the call/invoke instruction.
if (!SuccMBB.isLandingPad())
return MBB.getFirstTerminator();
// Discover any defs/uses in this basic block.
SmallPtrSet<MachineInstr*, 8> DefUsesInMBB;
for (MachineRegisterInfo::reg_iterator RI = MRI->reg_begin(SrcReg),
RE = MRI->reg_end(); RI != RE; ++RI) {
MachineInstr *DefUseMI = &*RI;
if (DefUseMI->getParent() == &MBB)
DefUsesInMBB.insert(DefUseMI);
}
MachineBasicBlock::iterator InsertPoint;
if (DefUsesInMBB.empty()) {
// No defs. Insert the copy at the start of the basic block.
InsertPoint = MBB.begin();
} else if (DefUsesInMBB.size() == 1) {
// Insert the copy immediately after the def/use.
InsertPoint = *DefUsesInMBB.begin();
++InsertPoint;
} else {
// Insert the copy immediately after the last def/use.
InsertPoint = MBB.end();
while (!DefUsesInMBB.count(&*--InsertPoint)) {}
++InsertPoint;
}
// Make sure the copy goes after any phi nodes however.
return SkipPHIsAndLabels(MBB, InsertPoint);
}
// FindCopyInsertPoint - Find a safe place in MBB to insert a copy from SrcReg.
// This needs to be after any def or uses of SrcReg, but before any subsequent
// point where control flow might jump out of the basic block.
MachineBasicBlock::iterator
llvm::PHIElimination::FindCopyInsertPoint(MachineBasicBlock &MBB,
unsigned SrcReg) {
// Handle the trivial case trivially.
if (MBB.empty())
return MBB.begin();
// If this basic block does not contain an invoke, then control flow always
// reaches the end of it, so place the copy there. The logic below works in
// this case too, but is more expensive.
if (!isa<InvokeInst>(MBB.getBasicBlock()->getTerminator()))
return MBB.getFirstTerminator();
// Discover any definition/uses in this basic block.
SmallPtrSet<MachineInstr*, 8> DefUsesInMBB;
for (MachineRegisterInfo::reg_iterator RI = MRI->reg_begin(SrcReg),
RE = MRI->reg_end(); RI != RE; ++RI) {
MachineInstr *DefUseMI = &*RI;
if (DefUseMI->getParent() == &MBB)
DefUsesInMBB.insert(DefUseMI);
}
MachineBasicBlock::iterator InsertPoint;
if (DefUsesInMBB.empty()) {
// No def/uses. Insert the copy at the start of the basic block.
InsertPoint = MBB.begin();
} else if (DefUsesInMBB.size() == 1) {
// Insert the copy immediately after the definition/use.
InsertPoint = *DefUsesInMBB.begin();
++InsertPoint;
} else {
// Insert the copy immediately after the last definition/use.
InsertPoint = MBB.end();
while (!DefUsesInMBB.count(&*--InsertPoint)) {}
++InsertPoint;
}
// Make sure the copy goes after any phi nodes however.
return SkipPHIsAndLabels(MBB, InsertPoint);
}
void WebAssemblyFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
auto &MFI = MF.getFrameInfo();
uint64_t StackSize = MFI.getStackSize();
if (!needsSP(MF, MFI) || !needsSPWriteback(MF, MFI)) return;
const auto *TII = MF.getSubtarget<WebAssemblySubtarget>().getInstrInfo();
auto &MRI = MF.getRegInfo();
auto InsertPt = MBB.getFirstTerminator();
DebugLoc DL;
if (InsertPt != MBB.end())
DL = InsertPt->getDebugLoc();
// Restore the stack pointer. If we had fixed-size locals, add the offset
// subtracted in the prolog.
unsigned SPReg = 0;
MachineBasicBlock::iterator InsertAddr = InsertPt;
if (hasBP(MF)) {
auto FI = MF.getInfo<WebAssemblyFunctionInfo>();
SPReg = FI->getBasePointerVreg();
} else if (StackSize) {
const TargetRegisterClass *PtrRC =
MRI.getTargetRegisterInfo()->getPointerRegClass(MF);
unsigned OffsetReg = MRI.createVirtualRegister(PtrRC);
InsertAddr =
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::CONST_I32), OffsetReg)
.addImm(StackSize);
// In the epilog we don't need to write the result back to the SP32 physreg
// because it won't be used again. We can use a stackified register instead.
SPReg = MRI.createVirtualRegister(PtrRC);
BuildMI(MBB, InsertPt, DL, TII->get(WebAssembly::ADD_I32), SPReg)
.addReg(hasFP(MF) ? WebAssembly::FP32 : WebAssembly::SP32)
.addReg(OffsetReg);
} else {
SPReg = hasFP(MF) ? WebAssembly::FP32 : WebAssembly::SP32;
}
writeSPToMemory(SPReg, MF, MBB, InsertAddr, InsertPt, DL);
}
/// Insert restore code for the callee-saved registers used in the function.
static void insertCSRRestores(MachineBasicBlock &RestoreBlock,
std::vector<CalleeSavedInfo> &CSI) {
MachineFunction &MF = *RestoreBlock.getParent();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering();
const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
// Restore all registers immediately before the return and any
// terminators that precede it.
MachineBasicBlock::iterator I = RestoreBlock.getFirstTerminator();
if (!TFI->restoreCalleeSavedRegisters(RestoreBlock, I, CSI, TRI)) {
for (const CalleeSavedInfo &CI : reverse(CSI)) {
unsigned Reg = CI.getReg();
const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
TII.loadRegFromStackSlot(RestoreBlock, I, Reg, CI.getFrameIdx(), RC, TRI);
assert(I != RestoreBlock.begin() &&
"loadRegFromStackSlot didn't insert any code!");
// Insert in reverse order. loadRegFromStackSlot can insert
// multiple instructions.
}
}
}
bool PPCCTRLoopsVerify::runOnMachineFunction(MachineFunction &MF) {
MDT = &getAnalysis<MachineDominatorTree>();
// Verify that all bdnz/bdz instructions are dominated by a loop mtctr before
// any other instructions that might clobber the ctr register.
for (MachineFunction::iterator I = MF.begin(), IE = MF.end();
I != IE; ++I) {
MachineBasicBlock *MBB = I;
if (!MDT->isReachableFromEntry(MBB))
continue;
for (MachineBasicBlock::iterator MII = MBB->getFirstTerminator(),
MIIE = MBB->end(); MII != MIIE; ++MII) {
unsigned Opc = MII->getOpcode();
if (Opc == PPC::BDNZ8 || Opc == PPC::BDNZ ||
Opc == PPC::BDZ8 || Opc == PPC::BDZ)
if (!verifyCTRBranch(MBB, MII))
llvm_unreachable("Invalid PPC CTR loop!");
}
}
return false;
}
/// Splits a MachineBasicBlock to branch before \p SplitBefore. The original
/// branch is \p OrigBranch. The target of the new branch can either be the same
/// as the target of the original branch or the fallthrough successor of the
/// original block as determined by \p BranchToFallThrough. The branch
/// conditions will be inverted according to \p InvertNewBranch and
/// \p InvertOrigBranch. If an instruction that previously fed the branch is to
/// be deleted, it is provided in \p MIToDelete and \p NewCond will be used as
/// the branch condition. The branch probabilities will be set if the
/// MachineBranchProbabilityInfo isn't null.
static bool splitMBB(BlockSplitInfo &BSI) {
assert(BSI.allInstrsInSameMBB() &&
"All instructions must be in the same block.");
MachineBasicBlock *ThisMBB = BSI.OrigBranch->getParent();
MachineFunction *MF = ThisMBB->getParent();
MachineRegisterInfo *MRI = &MF->getRegInfo();
assert(MRI->isSSA() && "Can only do this while the function is in SSA form.");
if (ThisMBB->succ_size() != 2) {
LLVM_DEBUG(
dbgs() << "Don't know how to handle blocks that don't have exactly"
<< " two successors.\n");
return false;
}
const PPCInstrInfo *TII = MF->getSubtarget<PPCSubtarget>().getInstrInfo();
unsigned OrigBROpcode = BSI.OrigBranch->getOpcode();
unsigned InvertedOpcode =
OrigBROpcode == PPC::BC
? PPC::BCn
: OrigBROpcode == PPC::BCn
? PPC::BC
: OrigBROpcode == PPC::BCLR ? PPC::BCLRn : PPC::BCLR;
unsigned NewBROpcode = BSI.InvertNewBranch ? InvertedOpcode : OrigBROpcode;
MachineBasicBlock *OrigTarget = BSI.OrigBranch->getOperand(1).getMBB();
MachineBasicBlock *OrigFallThrough = OrigTarget == *ThisMBB->succ_begin()
? *ThisMBB->succ_rbegin()
: *ThisMBB->succ_begin();
MachineBasicBlock *NewBRTarget =
BSI.BranchToFallThrough ? OrigFallThrough : OrigTarget;
BranchProbability ProbToNewTarget =
!BSI.MBPI ? BranchProbability::getUnknown()
: BSI.MBPI->getEdgeProbability(ThisMBB, NewBRTarget);
// Create a new basic block.
MachineBasicBlock::iterator InsertPoint = BSI.SplitBefore;
const BasicBlock *LLVM_BB = ThisMBB->getBasicBlock();
MachineFunction::iterator It = ThisMBB->getIterator();
MachineBasicBlock *NewMBB = MF->CreateMachineBasicBlock(LLVM_BB);
MF->insert(++It, NewMBB);
// Move everything after SplitBefore into the new block.
NewMBB->splice(NewMBB->end(), ThisMBB, InsertPoint, ThisMBB->end());
NewMBB->transferSuccessors(ThisMBB);
// Add the two successors to ThisMBB. The probabilities come from the
// existing blocks if available.
ThisMBB->addSuccessor(NewBRTarget, ProbToNewTarget);
ThisMBB->addSuccessor(NewMBB, ProbToNewTarget.getCompl());
// Add the branches to ThisMBB.
BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
TII->get(NewBROpcode))
.addReg(BSI.SplitCond->getOperand(0).getReg())
.addMBB(NewBRTarget);
BuildMI(*ThisMBB, ThisMBB->end(), BSI.SplitBefore->getDebugLoc(),
TII->get(PPC::B))
.addMBB(NewMBB);
if (BSI.MIToDelete)
BSI.MIToDelete->eraseFromParent();
// Change the condition on the original branch and invert it if requested.
auto FirstTerminator = NewMBB->getFirstTerminator();
if (BSI.NewCond) {
assert(FirstTerminator->getOperand(0).isReg() &&
"Can't update condition of unconditional branch.");
FirstTerminator->getOperand(0).setReg(BSI.NewCond->getOperand(0).getReg());
}
if (BSI.InvertOrigBranch)
FirstTerminator->setDesc(TII->get(InvertedOpcode));
// If any of the PHIs in the successors of NewMBB reference values that
// now come from NewMBB, they need to be updated.
for (auto *Succ : NewMBB->successors()) {
updatePHIs(Succ, ThisMBB, NewMBB, MRI);
}
addIncomingValuesToPHIs(NewBRTarget, ThisMBB, NewMBB, MRI);
LLVM_DEBUG(dbgs() << "After splitting, ThisMBB:\n"; ThisMBB->dump());
LLVM_DEBUG(dbgs() << "NewMBB:\n"; NewMBB->dump());
LLVM_DEBUG(dbgs() << "New branch-to block:\n"; NewBRTarget->dump());
return true;
}
/// Insert epilog code into the function.
/// For ARC, this inserts a call to a function that restores callee saved
/// registers onto the stack, when enough callee saved registers are required.
void ARCFrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
LLVM_DEBUG(dbgs() << "Emit Epilogue: " << MF.getName() << "\n");
auto *AFI = MF.getInfo<ARCFunctionInfo>();
const ARCInstrInfo *TII = MF.getSubtarget<ARCSubtarget>().getInstrInfo();
MachineBasicBlock::iterator MBBI = MBB.getFirstTerminator();
MachineFrameInfo &MFI = MF.getFrameInfo();
uint64_t StackSize = MF.getFrameInfo().getStackSize();
bool SavedBlink = false;
unsigned AmountAboveFunclet = 0;
// If we have variable sized frame objects, then we have to move
// the stack pointer to a known spot (fp - StackSize).
// Then, replace the frame pointer by (new) [sp,StackSize-4].
// Then, move the stack pointer the rest of the way (sp = sp + StackSize).
if (hasFP(MF)) {
BuildMI(MBB, MBBI, DebugLoc(), TII->get(ARC::SUB_rru6), ARC::SP)
.addReg(ARC::FP)
.addImm(StackSize);
AmountAboveFunclet += 4;
}
// Now, move the stack pointer to the bottom of the save area for the funclet.
const std::vector<CalleeSavedInfo> &CSI = MFI.getCalleeSavedInfo();
unsigned Last = determineLastCalleeSave(CSI);
unsigned StackSlotsUsedByFunclet = 0;
// Now, restore the callee save registers.
if (UseSaveRestoreFunclet && Last > ARC::R14) {
// BL to __ld_r13_to_<TRI->getRegAsmName()>
StackSlotsUsedByFunclet = Last - ARC::R12;
AmountAboveFunclet += 4 * (StackSlotsUsedByFunclet + 1);
SavedBlink = true;
}
if (MFI.hasCalls() && !SavedBlink) {
AmountAboveFunclet += 4;
SavedBlink = true;
}
// Move the stack pointer up to the point of the funclet.
if (StackSize - AmountAboveFunclet) {
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::ADD_rru6))
.addReg(ARC::SP)
.addReg(ARC::SP)
.addImm(StackSize - AmountAboveFunclet);
}
if (StackSlotsUsedByFunclet) {
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::BL))
.addExternalSymbol(load_funclet_name[Last - ARC::R15])
.addReg(ARC::BLINK, RegState::Implicit | RegState::Kill);
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::ADD_rru6))
.addReg(ARC::SP)
.addReg(ARC::SP)
.addImm(4 * (StackSlotsUsedByFunclet));
}
// Now, pop blink if necessary.
if (SavedBlink) {
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::POP_S_BLINK));
}
// Now, pop fp if necessary.
if (hasFP(MF)) {
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::LD_AB_rs9))
.addReg(ARC::SP, RegState::Define)
.addReg(ARC::FP, RegState::Define)
.addReg(ARC::SP)
.addImm(4);
}
// Relieve the varargs area if necessary.
if (MF.getFunction().isVarArg()) {
// Add in the varargs area here first.
LLVM_DEBUG(dbgs() << "Varargs\n");
unsigned VarArgsBytes = MFI.getObjectSize(AFI->getVarArgsFrameIndex());
BuildMI(MBB, MBBI, MBB.findDebugLoc(MBBI), TII->get(ARC::ADD_rru6))
.addReg(ARC::SP)
.addReg(ARC::SP)
.addImm(VarArgsBytes);
}
}
/// Insert a BLOCK marker for branches to MBB (if needed).
static void PlaceBlockMarker(MachineBasicBlock &MBB, MachineFunction &MF,
SmallVectorImpl<MachineBasicBlock *> &ScopeTops,
const WebAssemblyInstrInfo &TII,
const MachineLoopInfo &MLI,
MachineDominatorTree &MDT,
WebAssemblyFunctionInfo &MFI) {
// First compute the nearest common dominator of all forward non-fallthrough
// predecessors so that we minimize the time that the BLOCK is on the stack,
// which reduces overall stack height.
MachineBasicBlock *Header = nullptr;
bool IsBranchedTo = false;
int MBBNumber = MBB.getNumber();
for (MachineBasicBlock *Pred : MBB.predecessors())
if (Pred->getNumber() < MBBNumber) {
Header = Header ? MDT.findNearestCommonDominator(Header, Pred) : Pred;
if (ExplicitlyBranchesTo(Pred, &MBB))
IsBranchedTo = true;
}
if (!Header)
return;
if (!IsBranchedTo)
return;
assert(&MBB != &MF.front() && "Header blocks shouldn't have predecessors");
MachineBasicBlock *LayoutPred = &*prev(MachineFunction::iterator(&MBB));
// If the nearest common dominator is inside a more deeply nested context,
// walk out to the nearest scope which isn't more deeply nested.
for (MachineFunction::iterator I(LayoutPred), E(Header); I != E; --I) {
if (MachineBasicBlock *ScopeTop = ScopeTops[I->getNumber()]) {
if (ScopeTop->getNumber() > Header->getNumber()) {
// Skip over an intervening scope.
I = next(MachineFunction::iterator(ScopeTop));
} else {
// We found a scope level at an appropriate depth.
Header = ScopeTop;
break;
}
}
}
// If there's a loop which ends just before MBB which contains Header, we can
// reuse its label instead of inserting a new BLOCK.
for (MachineLoop *Loop = MLI.getLoopFor(LayoutPred);
Loop && Loop->contains(LayoutPred); Loop = Loop->getParentLoop())
if (Loop && LoopBottom(Loop) == LayoutPred && Loop->contains(Header))
return;
// Decide where in Header to put the BLOCK.
MachineBasicBlock::iterator InsertPos;
MachineLoop *HeaderLoop = MLI.getLoopFor(Header);
if (HeaderLoop && MBB.getNumber() > LoopBottom(HeaderLoop)->getNumber()) {
// Header is the header of a loop that does not lexically contain MBB, so
// the BLOCK needs to be above the LOOP, after any END constructs.
InsertPos = Header->begin();
while (InsertPos->getOpcode() != WebAssembly::LOOP)
++InsertPos;
} else {
// Otherwise, insert the BLOCK as late in Header as we can, but before the
// beginning of the local expression tree and any nested BLOCKs.
InsertPos = Header->getFirstTerminator();
while (InsertPos != Header->begin() && IsChild(prev(InsertPos), MFI) &&
prev(InsertPos)->getOpcode() != WebAssembly::LOOP &&
prev(InsertPos)->getOpcode() != WebAssembly::END_BLOCK &&
prev(InsertPos)->getOpcode() != WebAssembly::END_LOOP)
--InsertPos;
}
// Add the BLOCK.
BuildMI(*Header, InsertPos, DebugLoc(), TII.get(WebAssembly::BLOCK));
// Mark the end of the block.
InsertPos = MBB.begin();
while (InsertPos != MBB.end() &&
InsertPos->getOpcode() == WebAssembly::END_LOOP)
++InsertPos;
BuildMI(MBB, InsertPos, DebugLoc(), TII.get(WebAssembly::END_BLOCK));
// Track the farthest-spanning scope that ends at this point.
int Number = MBB.getNumber();
if (!ScopeTops[Number] ||
ScopeTops[Number]->getNumber() > Header->getNumber())
ScopeTops[Number] = Header;
}
void AArch64FrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = MBB.getLastNonDebugInstr();
MachineFrameInfo *MFI = MF.getFrameInfo();
const AArch64Subtarget &Subtarget = MF.getSubtarget<AArch64Subtarget>();
const AArch64RegisterInfo *RegInfo = Subtarget.getRegisterInfo();
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL;
bool IsTailCallReturn = false;
if (MBB.end() != MBBI) {
DL = MBBI->getDebugLoc();
unsigned RetOpcode = MBBI->getOpcode();
IsTailCallReturn = RetOpcode == AArch64::TCRETURNdi ||
RetOpcode == AArch64::TCRETURNri;
}
int NumBytes = MFI->getStackSize();
const AArch64FunctionInfo *AFI = MF.getInfo<AArch64FunctionInfo>();
// All calls are tail calls in GHC calling conv, and functions have no
// prologue/epilogue.
if (MF.getFunction()->getCallingConv() == CallingConv::GHC)
return;
// Initial and residual are named for consistency with the prologue. Note that
// in the epilogue, the residual adjustment is executed first.
uint64_t ArgumentPopSize = 0;
if (IsTailCallReturn) {
MachineOperand &StackAdjust = MBBI->getOperand(1);
// For a tail-call in a callee-pops-arguments environment, some or all of
// the stack may actually be in use for the call's arguments, this is
// calculated during LowerCall and consumed here...
ArgumentPopSize = StackAdjust.getImm();
} else {
// ... otherwise the amount to pop is *all* of the argument space,
// conveniently stored in the MachineFunctionInfo by
// LowerFormalArguments. This will, of course, be zero for the C calling
// convention.
ArgumentPopSize = AFI->getArgumentStackToRestore();
}
// The stack frame should be like below,
//
// ---------------------- ---
// | | |
// | BytesInStackArgArea| CalleeArgStackSize
// | (NumReusableBytes) | (of tail call)
// | | ---
// | | |
// ---------------------| --- |
// | | | |
// | CalleeSavedReg | | |
// | (NumRestores * 8) | | |
// | | | |
// ---------------------| | NumBytes
// | | StackSize (StackAdjustUp)
// | LocalStackSize | | |
// | (covering callee | | |
// | args) | | |
// | | | |
// ---------------------- --- ---
//
// So NumBytes = StackSize + BytesInStackArgArea - CalleeArgStackSize
// = StackSize + ArgumentPopSize
//
// AArch64TargetLowering::LowerCall figures out ArgumentPopSize and keeps
// it as the 2nd argument of AArch64ISD::TC_RETURN.
NumBytes += ArgumentPopSize;
unsigned NumRestores = 0;
// Move past the restores of the callee-saved registers.
MachineBasicBlock::iterator LastPopI = MBB.getFirstTerminator();
const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(&MF);
MachineBasicBlock::iterator Begin = MBB.begin();
while (LastPopI != Begin) {
--LastPopI;
unsigned Restores = getNumCSRestores(*LastPopI, CSRegs);
NumRestores += Restores;
if (Restores == 0) {
++LastPopI;
break;
}
}
NumBytes -= NumRestores * 8;
assert(NumBytes >= 0 && "Negative stack allocation size!?");
if (!hasFP(MF)) {
// If this was a redzone leaf function, we don't need to restore the
// stack pointer.
if (!canUseRedZone(MF))
emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::SP, NumBytes,
TII);
return;
}
// Restore the original stack pointer.
// FIXME: Rather than doing the math here, we should instead just use
// non-post-indexed loads for the restores if we aren't actually going to
// be able to save any instructions.
if (NumBytes || MFI->hasVarSizedObjects())
emitFrameOffset(MBB, LastPopI, DL, AArch64::SP, AArch64::FP,
-(NumRestores - 2) * 8, TII, MachineInstr::NoFlags);
}
/// EmitSchedule - Emit the machine code in scheduled order. Return the new
/// InsertPos and MachineBasicBlock that contains this insertion
/// point. ScheduleDAGSDNodes holds a BB pointer for convenience, but this does
/// not necessarily refer to returned BB. The emitter may split blocks.
MachineBasicBlock *ScheduleDAGSDNodes::
EmitSchedule(MachineBasicBlock::iterator &InsertPos) {
InstrEmitter Emitter(BB, InsertPos);
DenseMap<SDValue, unsigned> VRBaseMap;
DenseMap<SUnit*, unsigned> CopyVRBaseMap;
SmallVector<std::pair<unsigned, MachineInstr*>, 32> Orders;
SmallSet<unsigned, 8> Seen;
bool HasDbg = DAG->hasDebugValues();
// Emit a node, and determine where its first instruction is for debuginfo.
// Zero, one, or multiple instructions can be created when emitting a node.
auto EmitNode =
[&](SDNode *Node, bool IsClone, bool IsCloned,
DenseMap<SDValue, unsigned> &VRBaseMap) -> MachineInstr * {
// Fetch instruction prior to this, or end() if nonexistant.
auto GetPrevInsn = [&](MachineBasicBlock::iterator I) {
if (I == BB->begin())
return BB->end();
else
return std::prev(Emitter.getInsertPos());
};
MachineBasicBlock::iterator Before = GetPrevInsn(Emitter.getInsertPos());
Emitter.EmitNode(Node, IsClone, IsCloned, VRBaseMap);
MachineBasicBlock::iterator After = GetPrevInsn(Emitter.getInsertPos());
// If the iterator did not change, no instructions were inserted.
if (Before == After)
return nullptr;
if (Before == BB->end()) {
// There were no prior instructions; the new ones must start at the
// beginning of the block.
return &Emitter.getBlock()->instr_front();
} else {
// Return first instruction after the pre-existing instructions.
return &*std::next(Before);
}
};
// If this is the first BB, emit byval parameter dbg_value's.
if (HasDbg && BB->getParent()->begin() == MachineFunction::iterator(BB)) {
SDDbgInfo::DbgIterator PDI = DAG->ByvalParmDbgBegin();
SDDbgInfo::DbgIterator PDE = DAG->ByvalParmDbgEnd();
for (; PDI != PDE; ++PDI) {
MachineInstr *DbgMI= Emitter.EmitDbgValue(*PDI, VRBaseMap);
if (DbgMI) {
BB->insert(InsertPos, DbgMI);
// We re-emit the dbg_value closer to its use, too, after instructions
// are emitted to the BB.
(*PDI)->clearIsEmitted();
}
}
}
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
SUnit *SU = Sequence[i];
if (!SU) {
// Null SUnit* is a noop.
TII->insertNoop(*Emitter.getBlock(), InsertPos);
continue;
}
// For pre-regalloc scheduling, create instructions corresponding to the
// SDNode and any glued SDNodes and append them to the block.
if (!SU->getNode()) {
// Emit a copy.
EmitPhysRegCopy(SU, CopyVRBaseMap, InsertPos);
continue;
}
SmallVector<SDNode *, 4> GluedNodes;
for (SDNode *N = SU->getNode()->getGluedNode(); N; N = N->getGluedNode())
GluedNodes.push_back(N);
while (!GluedNodes.empty()) {
SDNode *N = GluedNodes.back();
auto NewInsn = EmitNode(N, SU->OrigNode != SU, SU->isCloned, VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(N, DAG, Emitter, VRBaseMap, Orders, Seen, NewInsn);
GluedNodes.pop_back();
}
auto NewInsn =
EmitNode(SU->getNode(), SU->OrigNode != SU, SU->isCloned, VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(SU->getNode(), DAG, Emitter, VRBaseMap, Orders, Seen,
NewInsn);
}
// Insert all the dbg_values which have not already been inserted in source
// order sequence.
if (HasDbg) {
MachineBasicBlock::iterator BBBegin = BB->getFirstNonPHI();
// Sort the source order instructions and use the order to insert debug
// values. Use stable_sort so that DBG_VALUEs are inserted in the same order
// regardless of the host's implementation fo std::sort.
std::stable_sort(Orders.begin(), Orders.end(), less_first());
std::stable_sort(DAG->DbgBegin(), DAG->DbgEnd(),
[](const SDDbgValue *LHS, const SDDbgValue *RHS) {
return LHS->getOrder() < RHS->getOrder();
});
SDDbgInfo::DbgIterator DI = DAG->DbgBegin();
SDDbgInfo::DbgIterator DE = DAG->DbgEnd();
// Now emit the rest according to source order.
unsigned LastOrder = 0;
for (unsigned i = 0, e = Orders.size(); i != e && DI != DE; ++i) {
unsigned Order = Orders[i].first;
MachineInstr *MI = Orders[i].second;
// Insert all SDDbgValue's whose order(s) are before "Order".
assert(MI);
for (; DI != DE; ++DI) {
if ((*DI)->getOrder() < LastOrder || (*DI)->getOrder() >= Order)
break;
if ((*DI)->isEmitted())
continue;
MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap);
if (DbgMI) {
if (!LastOrder)
// Insert to start of the BB (after PHIs).
BB->insert(BBBegin, DbgMI);
else {
// Insert at the instruction, which may be in a different
// block, if the block was split by a custom inserter.
MachineBasicBlock::iterator Pos = MI;
MI->getParent()->insert(Pos, DbgMI);
}
}
}
LastOrder = Order;
}
// Add trailing DbgValue's before the terminator. FIXME: May want to add
// some of them before one or more conditional branches?
SmallVector<MachineInstr*, 8> DbgMIs;
for (; DI != DE; ++DI) {
if ((*DI)->isEmitted())
continue;
assert((*DI)->getOrder() >= LastOrder &&
"emitting DBG_VALUE out of order");
if (MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap))
DbgMIs.push_back(DbgMI);
}
MachineBasicBlock *InsertBB = Emitter.getBlock();
MachineBasicBlock::iterator Pos = InsertBB->getFirstTerminator();
InsertBB->insert(Pos, DbgMIs.begin(), DbgMIs.end());
SDDbgInfo::DbgLabelIterator DLI = DAG->DbgLabelBegin();
SDDbgInfo::DbgLabelIterator DLE = DAG->DbgLabelEnd();
// Now emit the rest according to source order.
LastOrder = 0;
for (const auto &InstrOrder : Orders) {
unsigned Order = InstrOrder.first;
MachineInstr *MI = InstrOrder.second;
if (!MI)
continue;
// Insert all SDDbgLabel's whose order(s) are before "Order".
for (; DLI != DLE &&
(*DLI)->getOrder() >= LastOrder && (*DLI)->getOrder() < Order;
++DLI) {
MachineInstr *DbgMI = Emitter.EmitDbgLabel(*DLI);
if (DbgMI) {
if (!LastOrder)
// Insert to start of the BB (after PHIs).
BB->insert(BBBegin, DbgMI);
else {
// Insert at the instruction, which may be in a different
// block, if the block was split by a custom inserter.
MachineBasicBlock::iterator Pos = MI;
MI->getParent()->insert(Pos, DbgMI);
}
}
}
if (DLI == DLE)
break;
LastOrder = Order;
}
}
InsertPos = Emitter.getInsertPos();
return Emitter.getBlock();
}
/// TailDuplicate - If it is profitable, duplicate TailBB's contents in each
/// of its predecessors.
bool
TailDuplicatePass::TailDuplicate(MachineBasicBlock *TailBB, MachineFunction &MF,
SmallVector<MachineBasicBlock*, 8> &TDBBs,
SmallVector<MachineInstr*, 16> &Copies) {
if (!shouldTailDuplicate(MF, *TailBB))
return false;
DEBUG(dbgs() << "\n*** Tail-duplicating BB#" << TailBB->getNumber() << '\n');
// Iterate through all the unique predecessors and tail-duplicate this
// block into them, if possible. Copying the list ahead of time also
// avoids trouble with the predecessor list reallocating.
bool Changed = false;
SmallSetVector<MachineBasicBlock*, 8> Preds(TailBB->pred_begin(),
TailBB->pred_end());
DenseSet<unsigned> UsedByPhi;
getRegsUsedByPHIs(*TailBB, &UsedByPhi);
for (SmallSetVector<MachineBasicBlock *, 8>::iterator PI = Preds.begin(),
PE = Preds.end(); PI != PE; ++PI) {
MachineBasicBlock *PredBB = *PI;
assert(TailBB != PredBB &&
"Single-block loop should have been rejected earlier!");
// EH edges are ignored by AnalyzeBranch.
if (PredBB->succ_size() > 1)
continue;
MachineBasicBlock *PredTBB, *PredFBB;
SmallVector<MachineOperand, 4> PredCond;
if (TII->AnalyzeBranch(*PredBB, PredTBB, PredFBB, PredCond, true))
continue;
if (!PredCond.empty())
continue;
// Don't duplicate into a fall-through predecessor (at least for now).
if (PredBB->isLayoutSuccessor(TailBB) && PredBB->canFallThrough())
continue;
DEBUG(dbgs() << "\nTail-duplicating into PredBB: " << *PredBB
<< "From Succ: " << *TailBB);
TDBBs.push_back(PredBB);
// Remove PredBB's unconditional branch.
TII->RemoveBranch(*PredBB);
// Clone the contents of TailBB into PredBB.
DenseMap<unsigned, unsigned> LocalVRMap;
SmallVector<std::pair<unsigned,unsigned>, 4> CopyInfos;
MachineBasicBlock::iterator I = TailBB->begin();
while (I != TailBB->end()) {
MachineInstr *MI = &*I;
++I;
if (MI->isPHI()) {
// Replace the uses of the def of the PHI with the register coming
// from PredBB.
ProcessPHI(MI, TailBB, PredBB, LocalVRMap, CopyInfos, UsedByPhi, true);
} else {
// Replace def of virtual registers with new registers, and update
// uses with PHI source register or the new registers.
DuplicateInstruction(MI, TailBB, PredBB, MF, LocalVRMap, UsedByPhi);
}
}
MachineBasicBlock::iterator Loc = PredBB->getFirstTerminator();
for (unsigned i = 0, e = CopyInfos.size(); i != e; ++i) {
Copies.push_back(BuildMI(*PredBB, Loc, DebugLoc(),
TII->get(TargetOpcode::COPY),
CopyInfos[i].first).addReg(CopyInfos[i].second));
}
// Simplify
TII->AnalyzeBranch(*PredBB, PredTBB, PredFBB, PredCond, true);
NumInstrDups += TailBB->size() - 1; // subtract one for removed branch
// Update the CFG.
PredBB->removeSuccessor(PredBB->succ_begin());
assert(PredBB->succ_empty() &&
"TailDuplicate called on block with multiple successors!");
for (MachineBasicBlock::succ_iterator I = TailBB->succ_begin(),
E = TailBB->succ_end(); I != E; ++I)
PredBB->addSuccessor(*I);
Changed = true;
++NumTailDups;
}
// If TailBB was duplicated into all its predecessors except for the prior
// block, which falls through unconditionally, move the contents of this
// block into the prior block.
MachineBasicBlock *PrevBB = prior(MachineFunction::iterator(TailBB));
MachineBasicBlock *PriorTBB = 0, *PriorFBB = 0;
SmallVector<MachineOperand, 4> PriorCond;
// This has to check PrevBB->succ_size() because EH edges are ignored by
// AnalyzeBranch.
if (PrevBB->succ_size() == 1 &&
!TII->AnalyzeBranch(*PrevBB, PriorTBB, PriorFBB, PriorCond, true) &&
PriorCond.empty() && !PriorTBB && TailBB->pred_size() == 1 &&
!TailBB->hasAddressTaken()) {
DEBUG(dbgs() << "\nMerging into block: " << *PrevBB
<< "From MBB: " << *TailBB);
if (PreRegAlloc) {
DenseMap<unsigned, unsigned> LocalVRMap;
SmallVector<std::pair<unsigned,unsigned>, 4> CopyInfos;
MachineBasicBlock::iterator I = TailBB->begin();
// Process PHI instructions first.
while (I != TailBB->end() && I->isPHI()) {
// Replace the uses of the def of the PHI with the register coming
// from PredBB.
MachineInstr *MI = &*I++;
ProcessPHI(MI, TailBB, PrevBB, LocalVRMap, CopyInfos, UsedByPhi, true);
if (MI->getParent())
MI->eraseFromParent();
}
// Now copy the non-PHI instructions.
while (I != TailBB->end()) {
// Replace def of virtual registers with new registers, and update
// uses with PHI source register or the new registers.
MachineInstr *MI = &*I++;
DuplicateInstruction(MI, TailBB, PrevBB, MF, LocalVRMap, UsedByPhi);
MI->eraseFromParent();
}
MachineBasicBlock::iterator Loc = PrevBB->getFirstTerminator();
for (unsigned i = 0, e = CopyInfos.size(); i != e; ++i) {
Copies.push_back(BuildMI(*PrevBB, Loc, DebugLoc(),
TII->get(TargetOpcode::COPY),
CopyInfos[i].first)
.addReg(CopyInfos[i].second));
}
} else {
// No PHIs to worry about, just splice the instructions over.
PrevBB->splice(PrevBB->end(), TailBB, TailBB->begin(), TailBB->end());
}
PrevBB->removeSuccessor(PrevBB->succ_begin());
assert(PrevBB->succ_empty());
PrevBB->transferSuccessors(TailBB);
TDBBs.push_back(PrevBB);
Changed = true;
}
// If this is after register allocation, there are no phis to fix.
if (!PreRegAlloc)
return Changed;
// If we made no changes so far, we are safe.
if (!Changed)
return Changed;
// Handle the nasty case in that we duplicated a block that is part of a loop
// into some but not all of its predecessors. For example:
// 1 -> 2 <-> 3 |
// \ |
// \---> rest |
// if we duplicate 2 into 1 but not into 3, we end up with
// 12 -> 3 <-> 2 -> rest |
// \ / |
// \----->-----/ |
// If there was a "var = phi(1, 3)" in 2, it has to be ultimately replaced
// with a phi in 3 (which now dominates 2).
// What we do here is introduce a copy in 3 of the register defined by the
// phi, just like when we are duplicating 2 into 3, but we don't copy any
// real instructions or remove the 3 -> 2 edge from the phi in 2.
for (SmallSetVector<MachineBasicBlock *, 8>::iterator PI = Preds.begin(),
PE = Preds.end(); PI != PE; ++PI) {
MachineBasicBlock *PredBB = *PI;
if (std::find(TDBBs.begin(), TDBBs.end(), PredBB) != TDBBs.end())
continue;
// EH edges
if (PredBB->succ_size() != 1)
continue;
DenseMap<unsigned, unsigned> LocalVRMap;
SmallVector<std::pair<unsigned,unsigned>, 4> CopyInfos;
MachineBasicBlock::iterator I = TailBB->begin();
// Process PHI instructions first.
while (I != TailBB->end() && I->isPHI()) {
// Replace the uses of the def of the PHI with the register coming
// from PredBB.
MachineInstr *MI = &*I++;
ProcessPHI(MI, TailBB, PredBB, LocalVRMap, CopyInfos, UsedByPhi, false);
}
MachineBasicBlock::iterator Loc = PredBB->getFirstTerminator();
for (unsigned i = 0, e = CopyInfos.size(); i != e; ++i) {
Copies.push_back(BuildMI(*PredBB, Loc, DebugLoc(),
TII->get(TargetOpcode::COPY),
CopyInfos[i].first).addReg(CopyInfos[i].second));
}
}
return Changed;
}
bool Thumb1FrameLowering::emitPopSpecialFixUp(MachineBasicBlock &MBB,
bool DoIt) const {
MachineFunction &MF = *MBB.getParent();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ArgRegsSaveSize = AFI->getArgRegsSaveSize();
const TargetInstrInfo &TII = *STI.getInstrInfo();
const ThumbRegisterInfo *RegInfo =
static_cast<const ThumbRegisterInfo *>(STI.getRegisterInfo());
// If MBBI is a return instruction, we may be able to directly restore
// LR in the PC.
// This is possible if we do not need to emit any SP update.
// Otherwise, we need a temporary register to pop the value
// and copy that value into LR.
auto MBBI = MBB.getFirstTerminator();
if (!ArgRegsSaveSize && MBBI != MBB.end() &&
MBBI->getOpcode() == ARM::tBX_RET) {
if (!DoIt)
return true;
MachineInstrBuilder MIB =
AddDefaultPred(
BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII.get(ARM::tPOP_RET)))
.addReg(ARM::PC, RegState::Define);
MIB.copyImplicitOps(&*MBBI);
// erase the old tBX_RET instruction
MBB.erase(MBBI);
return true;
}
// Look for a temporary register to use.
// First, compute the liveness information.
LivePhysRegs UsedRegs(STI.getRegisterInfo());
UsedRegs.addLiveOuts(&MBB, /*AddPristines*/ true);
// The semantic of pristines changed recently and now,
// the callee-saved registers that are touched in the function
// are not part of the pristines set anymore.
// Add those callee-saved now.
const TargetRegisterInfo *TRI = STI.getRegisterInfo();
const MCPhysReg *CSRegs = TRI->getCalleeSavedRegs(&MF);
for (unsigned i = 0; CSRegs[i]; ++i)
UsedRegs.addReg(CSRegs[i]);
DebugLoc dl = DebugLoc();
if (MBBI != MBB.end()) {
dl = MBBI->getDebugLoc();
auto InstUpToMBBI = MBB.end();
// The post-decrement is on purpose here.
// We want to have the liveness right before MBBI.
while (InstUpToMBBI-- != MBBI)
UsedRegs.stepBackward(*InstUpToMBBI);
}
// Look for a register that can be directly use in the POP.
unsigned PopReg = 0;
// And some temporary register, just in case.
unsigned TemporaryReg = 0;
BitVector PopFriendly =
TRI->getAllocatableSet(MF, TRI->getRegClass(ARM::tGPRRegClassID));
assert(PopFriendly.any() && "No allocatable pop-friendly register?!");
// Rebuild the GPRs from the high registers because they are removed
// form the GPR reg class for thumb1.
BitVector GPRsNoLRSP =
TRI->getAllocatableSet(MF, TRI->getRegClass(ARM::hGPRRegClassID));
GPRsNoLRSP |= PopFriendly;
GPRsNoLRSP.reset(ARM::LR);
GPRsNoLRSP.reset(ARM::SP);
GPRsNoLRSP.reset(ARM::PC);
for (int Register = GPRsNoLRSP.find_first(); Register != -1;
Register = GPRsNoLRSP.find_next(Register)) {
if (!UsedRegs.contains(Register)) {
// Remember the first pop-friendly register and exit.
if (PopFriendly.test(Register)) {
PopReg = Register;
TemporaryReg = 0;
break;
}
// Otherwise, remember that the register will be available to
// save a pop-friendly register.
TemporaryReg = Register;
}
}
if (!DoIt && !PopReg && !TemporaryReg)
return false;
assert((PopReg || TemporaryReg) && "Cannot get LR");
if (TemporaryReg) {
assert(!PopReg && "Unnecessary MOV is about to be inserted");
PopReg = PopFriendly.find_first();
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(TemporaryReg, RegState::Define)
.addReg(PopReg, RegState::Kill));
}
assert(PopReg && "Do not know how to get LR");
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tPOP)))
.addReg(PopReg, RegState::Define);
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, ArgRegsSaveSize);
if (!TemporaryReg && MBBI != MBB.end() && MBBI->getOpcode() == ARM::tBX_RET) {
MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(ARM::tBX))
.addReg(PopReg, RegState::Kill);
AddDefaultPred(MIB);
MIB.copyImplicitOps(&*MBBI);
// erase the old tBX_RET instruction
MBB.erase(MBBI);
return true;
}
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(ARM::LR, RegState::Define)
.addReg(PopReg, RegState::Kill));
if (TemporaryReg) {
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(PopReg, RegState::Define)
.addReg(TemporaryReg, RegState::Kill));
}
return true;
}
/// 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, Int64Cmp;
MachineInstr *MI = IV_Opnd->getParent();
if (L->contains(MI) && isCompareEqualsImm(MI, SignedCmp, Int64Cmp) &&
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();
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 = DefInstr->getOperand(2).getImm();
MachineInstr *DefInstr2 =
MRI->getVRegDef(DefInstr->getOperand(1).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;
}
OldInsts.push_back(DefInstr);
OldInsts.push_back(DefInstr2);
// count/iv_value, the trip count, should be positive here. If it
// is negative, that indicates that the counter will wrap.
if (Int64Cmp)
return new CountValue(count/iv_value);
else
return new CountValue(uint32_t(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;
}
OldInsts.push_back(DefInstr);
if (Int64Cmp)
return new CountValue(count/iv_value);
else
return new CountValue(uint32_t(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;
}
/// converToCTRLoop - check if the loop is a candidate for
/// converting to a CTR loop. If so, then perform the
/// transformation.
///
/// This function works on innermost loops first. A loop can
/// be converted if it is a counting loop; either a register
/// value or an immediate.
///
/// The code makes several assumptions about the representation
/// of the loop in llvm.
bool PPCCTRLoops::convertToCTRLoop(MachineLoop *L) {
bool Changed = false;
// Process nested loops first.
for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
Changed |= convertToCTRLoop(*I);
}
// If a nested loop has been converted, then we can't convert this loop.
if (Changed) {
return Changed;
}
SmallVector<MachineInstr *, 2> OldInsts;
// Are we able to determine the trip count for the loop?
CountValue *TripCount = getTripCount(L, OldInsts);
if (TripCount == 0) {
DEBUG(dbgs() << "failed to get trip count!\n");
return false;
}
if (TripCount->isImm()) {
DEBUG(dbgs() << "constant trip count: " << TripCount->getImm() << "\n");
// FIXME: We currently can't form 64-bit constants
// (including 32-bit unsigned constants)
if (!isInt<32>(TripCount->getImm()))
return false;
}
// Does the loop contain any invalid instructions?
if (containsInvalidInstruction(L)) {
return false;
}
MachineBasicBlock *Preheader = L->getLoopPreheader();
// No preheader means there's not place for the loop instr.
if (Preheader == 0) {
return false;
}
MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
DebugLoc dl;
if (InsertPos != Preheader->end())
dl = InsertPos->getDebugLoc();
MachineBasicBlock *LastMBB = L->getExitingBlock();
// Don't generate CTR loop if the loop has more than one exit.
if (LastMBB == 0) {
return false;
}
MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
// Determine the loop start.
MachineBasicBlock *LoopStart = L->getTopBlock();
if (L->getLoopLatch() != LastMBB) {
// When the exit and latch are not the same, use the latch block as the
// start.
// The loop start address is used only after the 1st iteration, and the loop
// latch may contains instrs. that need to be executed after the 1st iter.
LoopStart = L->getLoopLatch();
// Make sure the latch is a successor of the exit, otherwise it won't work.
if (!LastMBB->isSuccessor(LoopStart)) {
return false;
}
}
// Convert the loop to a CTR loop
DEBUG(dbgs() << "Change to CTR loop at "; L->dump());
MachineFunction *MF = LastMBB->getParent();
const PPCSubtarget &Subtarget = MF->getTarget().getSubtarget<PPCSubtarget>();
bool isPPC64 = Subtarget.isPPC64();
const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
const TargetRegisterClass *RC = isPPC64 ? G8RC : GPRC;
unsigned CountReg;
if (TripCount->isReg()) {
// Create a copy of the loop count register.
const TargetRegisterClass *SrcRC =
MF->getRegInfo().getRegClass(TripCount->getReg());
CountReg = MF->getRegInfo().createVirtualRegister(RC);
unsigned CopyOp = (isPPC64 && GPRC->hasSubClassEq(SrcRC)) ?
(unsigned) PPC::EXTSW_32_64 :
(unsigned) TargetOpcode::COPY;
BuildMI(*Preheader, InsertPos, dl,
TII->get(CopyOp), CountReg).addReg(TripCount->getReg());
if (TripCount->isNeg()) {
unsigned CountReg1 = CountReg;
CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, dl,
TII->get(isPPC64 ? PPC::NEG8 : PPC::NEG),
CountReg).addReg(CountReg1);
}
} else {
assert(TripCount->isImm() && "Expecting immedate vaule for trip count");
// Put the trip count in a register for transfer into the count register.
int64_t CountImm = TripCount->getImm();
if (TripCount->isNeg())
CountImm = -CountImm;
CountReg = MF->getRegInfo().createVirtualRegister(RC);
if (abs64(CountImm) > 0x7FFF) {
BuildMI(*Preheader, InsertPos, dl,
TII->get(isPPC64 ? PPC::LIS8 : PPC::LIS),
CountReg).addImm((CountImm >> 16) & 0xFFFF);
unsigned CountReg1 = CountReg;
CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, dl,
TII->get(isPPC64 ? PPC::ORI8 : PPC::ORI),
CountReg).addReg(CountReg1).addImm(CountImm & 0xFFFF);
} else {
void Thumb1FrameLowering::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = MBB.getFirstTerminator();
DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
const ThumbRegisterInfo *RegInfo =
static_cast<const ThumbRegisterInfo *>(STI.getRegisterInfo());
const Thumb1InstrInfo &TII =
*static_cast<const Thumb1InstrInfo *>(STI.getInstrInfo());
unsigned ArgRegsSaveSize = AFI->getArgRegsSaveSize();
int NumBytes = (int)MFI->getStackSize();
assert((unsigned)NumBytes >= ArgRegsSaveSize &&
"ArgRegsSaveSize is included in NumBytes");
const MCPhysReg *CSRegs = RegInfo->getCalleeSavedRegs(&MF);
unsigned FramePtr = RegInfo->getFrameRegister(MF);
if (!AFI->hasStackFrame()) {
if (NumBytes - ArgRegsSaveSize != 0)
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, NumBytes - ArgRegsSaveSize);
} else {
// Unwind MBBI to point to first LDR / VLDRD.
if (MBBI != MBB.begin()) {
do
--MBBI;
while (MBBI != MBB.begin() && isCSRestore(MBBI, CSRegs));
if (!isCSRestore(MBBI, CSRegs))
++MBBI;
}
// Move SP to start of FP callee save spill area.
NumBytes -= (AFI->getGPRCalleeSavedArea1Size() +
AFI->getGPRCalleeSavedArea2Size() +
AFI->getDPRCalleeSavedAreaSize() +
ArgRegsSaveSize);
if (AFI->shouldRestoreSPFromFP()) {
NumBytes = AFI->getFramePtrSpillOffset() - NumBytes;
// Reset SP based on frame pointer only if the stack frame extends beyond
// frame pointer stack slot, the target is ELF and the function has FP, or
// the target uses var sized objects.
if (NumBytes) {
assert(!MFI->getPristineRegs(MF).test(ARM::R4) &&
"No scratch register to restore SP from FP!");
emitThumbRegPlusImmediate(MBB, MBBI, dl, ARM::R4, FramePtr, -NumBytes,
TII, *RegInfo);
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr),
ARM::SP)
.addReg(ARM::R4));
} else
AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr),
ARM::SP)
.addReg(FramePtr));
} else {
if (MBBI != MBB.end() && MBBI->getOpcode() == ARM::tBX_RET &&
&MBB.front() != MBBI && std::prev(MBBI)->getOpcode() == ARM::tPOP) {
MachineBasicBlock::iterator PMBBI = std::prev(MBBI);
if (!tryFoldSPUpdateIntoPushPop(STI, MF, PMBBI, NumBytes))
emitSPUpdate(MBB, PMBBI, TII, dl, *RegInfo, NumBytes);
} else if (!tryFoldSPUpdateIntoPushPop(STI, MF, MBBI, NumBytes))
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, NumBytes);
}
}
if (needPopSpecialFixUp(MF)) {
bool Done = emitPopSpecialFixUp(MBB, /* DoIt */ true);
(void)Done;
assert(Done && "Emission of the special fixup failed!?");
}
}
/// ComputeLocalLiveness - Computes liveness of registers within a basic
/// block, setting the killed/dead flags as appropriate.
void RALocal::ComputeLocalLiveness(MachineBasicBlock& MBB) {
MachineRegisterInfo &MRI = MBB.getParent()->getRegInfo();
// Keep track of the most recently seen previous use or def of each reg,
// so that we can update them with dead/kill markers.
DenseMap<unsigned, std::pair<MachineInstr*, unsigned> > LastUseDef;
for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
I != E; ++I) {
if (I->isDebugValue())
continue;
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
MachineOperand &MO = I->getOperand(i);
// Uses don't trigger any flags, but we need to save
// them for later. Also, we have to process these
// _before_ processing the defs, since an instr
// uses regs before it defs them.
if (!MO.isReg() || !MO.getReg() || !MO.isUse())
continue;
LastUseDef[MO.getReg()] = std::make_pair(I, i);
if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) continue;
const unsigned *Aliases = TRI->getAliasSet(MO.getReg());
if (Aliases == 0)
continue;
while (*Aliases) {
DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
alias = LastUseDef.find(*Aliases);
if (alias != LastUseDef.end() && alias->second.first != I)
LastUseDef[*Aliases] = std::make_pair(I, i);
++Aliases;
}
}
for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
MachineOperand &MO = I->getOperand(i);
// Defs others than 2-addr redefs _do_ trigger flag changes:
// - A def followed by a def is dead
// - A use followed by a def is a kill
if (!MO.isReg() || !MO.getReg() || !MO.isDef()) continue;
DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
last = LastUseDef.find(MO.getReg());
if (last != LastUseDef.end()) {
// Check if this is a two address instruction. If so, then
// the def does not kill the use.
if (last->second.first == I &&
I->isRegTiedToUseOperand(i))
continue;
MachineOperand &lastUD =
last->second.first->getOperand(last->second.second);
if (lastUD.isDef())
lastUD.setIsDead(true);
else
lastUD.setIsKill(true);
}
LastUseDef[MO.getReg()] = std::make_pair(I, i);
}
}
// Live-out (of the function) registers contain return values of the function,
// so we need to make sure they are alive at return time.
MachineBasicBlock::iterator Ret = MBB.getFirstTerminator();
bool BBEndsInReturn = (Ret != MBB.end() && Ret->getDesc().isReturn());
if (BBEndsInReturn)
for (MachineRegisterInfo::liveout_iterator
I = MF->getRegInfo().liveout_begin(),
E = MF->getRegInfo().liveout_end(); I != E; ++I)
if (!Ret->readsRegister(*I)) {
Ret->addOperand(MachineOperand::CreateReg(*I, false, true));
LastUseDef[*I] = std::make_pair(Ret, Ret->getNumOperands()-1);
}
// Finally, loop over the final use/def of each reg
// in the block and determine if it is dead.
for (DenseMap<unsigned, std::pair<MachineInstr*, unsigned> >::iterator
I = LastUseDef.begin(), E = LastUseDef.end(); I != E; ++I) {
MachineInstr *MI = I->second.first;
unsigned idx = I->second.second;
MachineOperand &MO = MI->getOperand(idx);
bool isPhysReg = TargetRegisterInfo::isPhysicalRegister(MO.getReg());
// A crude approximation of "live-out" calculation
bool usedOutsideBlock = isPhysReg ? false :
UsedInMultipleBlocks.test(MO.getReg() -
TargetRegisterInfo::FirstVirtualRegister);
// If the machine BB ends in a return instruction, then the value isn't used
// outside of the BB.
if (!isPhysReg && (!usedOutsideBlock || BBEndsInReturn)) {
// DBG_VALUE complicates this: if the only refs of a register outside
// this block are DBG_VALUE, we can't keep the reg live just for that,
// as it will cause the reg to be spilled at the end of this block when
// it wouldn't have been otherwise. Nullify the DBG_VALUEs when that
// happens.
bool UsedByDebugValueOnly = false;
for (MachineRegisterInfo::reg_iterator UI = MRI.reg_begin(MO.getReg()),
UE = MRI.reg_end(); UI != UE; ++UI) {
// Two cases:
// - used in another block
// - used in the same block before it is defined (loop)
if (UI->getParent() == &MBB &&
!(MO.isDef() && UI.getOperand().isUse() && precedes(&*UI, MI)))
continue;
if (UI->isDebugValue()) {
UsedByDebugValueOnly = true;
continue;
}
// A non-DBG_VALUE use means we can leave DBG_VALUE uses alone.
UsedInMultipleBlocks.set(MO.getReg() -
TargetRegisterInfo::FirstVirtualRegister);
usedOutsideBlock = true;
UsedByDebugValueOnly = false;
break;
}
if (UsedByDebugValueOnly)
for (MachineRegisterInfo::reg_iterator UI = MRI.reg_begin(MO.getReg()),
UE = MRI.reg_end(); UI != UE; ++UI)
if (UI->isDebugValue() &&
(UI->getParent() != &MBB ||
(MO.isDef() && precedes(&*UI, MI))))
UI.getOperand().setReg(0U);
}
// Physical registers and those that are not live-out of the block are
// killed/dead at their last use/def within this block.
if (isPhysReg || !usedOutsideBlock || BBEndsInReturn) {
if (MO.isUse()) {
// Don't mark uses that are tied to defs as kills.
if (!MI->isRegTiedToDefOperand(idx))
MO.setIsKill(true);
} else {
MO.setIsDead(true);
}
}
}
}
bool Thumb1FrameLowering::
restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
std::vector<CalleeSavedInfo> &CSI,
const TargetRegisterInfo *TRI) const {
if (CSI.empty())
return false;
MachineFunction &MF = *MBB.getParent();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
const TargetInstrInfo &TII = *STI.getInstrInfo();
const ARMBaseRegisterInfo *RegInfo = static_cast<const ARMBaseRegisterInfo *>(
MF.getSubtarget().getRegisterInfo());
bool isVarArg = AFI->getArgRegsSaveSize() > 0;
DebugLoc DL = MI != MBB.end() ? MI->getDebugLoc() : DebugLoc();
ARMRegSet LoRegsToRestore;
ARMRegSet HiRegsToRestore;
// Low registers (r0-r7) which can be used to restore the high registers.
ARMRegSet CopyRegs;
for (CalleeSavedInfo I : CSI) {
unsigned Reg = I.getReg();
if (ARM::tGPRRegClass.contains(Reg) || Reg == ARM::LR) {
LoRegsToRestore[Reg] = true;
} else if (ARM::hGPRRegClass.contains(Reg) && Reg != ARM::LR) {
HiRegsToRestore[Reg] = true;
} else {
llvm_unreachable("callee-saved register of unexpected class");
}
// If this is a low register not used as the frame pointer, we may want to
// use it for restoring the high registers.
if ((ARM::tGPRRegClass.contains(Reg)) &&
!(hasFP(MF) && Reg == RegInfo->getFrameRegister(MF)))
CopyRegs[Reg] = true;
}
// If this is a return block, we may be able to use some unused return value
// registers for restoring the high regs.
auto Terminator = MBB.getFirstTerminator();
if (Terminator != MBB.end() && Terminator->getOpcode() == ARM::tBX_RET) {
CopyRegs[ARM::R0] = true;
CopyRegs[ARM::R1] = true;
CopyRegs[ARM::R2] = true;
CopyRegs[ARM::R3] = true;
for (auto Op : Terminator->implicit_operands()) {
if (Op.isReg())
CopyRegs[Op.getReg()] = false;
}
}
static const unsigned AllCopyRegs[] = {ARM::R0, ARM::R1, ARM::R2, ARM::R3,
ARM::R4, ARM::R5, ARM::R6, ARM::R7};
static const unsigned AllHighRegs[] = {ARM::R8, ARM::R9, ARM::R10, ARM::R11};
const unsigned *AllCopyRegsEnd = std::end(AllCopyRegs);
const unsigned *AllHighRegsEnd = std::end(AllHighRegs);
// Find the first register to restore.
auto HiRegToRestore = findNextOrderedReg(std::begin(AllHighRegs),
HiRegsToRestore, AllHighRegsEnd);
while (HiRegToRestore != AllHighRegsEnd) {
assert(!CopyRegs.none());
// Find the first low register to use.
auto CopyReg =
findNextOrderedReg(std::begin(AllCopyRegs), CopyRegs, AllCopyRegsEnd);
// Create the POP instruction.
MachineInstrBuilder PopMIB =
BuildMI(MBB, MI, DL, TII.get(ARM::tPOP)).add(predOps(ARMCC::AL));
while (HiRegToRestore != AllHighRegsEnd && CopyReg != AllCopyRegsEnd) {
// Add the low register to the POP.
PopMIB.addReg(*CopyReg, RegState::Define);
// Create the MOV from low to high register.
BuildMI(MBB, MI, DL, TII.get(ARM::tMOVr))
.addReg(*HiRegToRestore, RegState::Define)
.addReg(*CopyReg, RegState::Kill)
.add(predOps(ARMCC::AL));
CopyReg = findNextOrderedReg(++CopyReg, CopyRegs, AllCopyRegsEnd);
HiRegToRestore =
findNextOrderedReg(++HiRegToRestore, HiRegsToRestore, AllHighRegsEnd);
}
}
MachineInstrBuilder MIB =
BuildMI(MF, DL, TII.get(ARM::tPOP)).add(predOps(ARMCC::AL));
bool NeedsPop = false;
for (unsigned i = CSI.size(); i != 0; --i) {
CalleeSavedInfo &Info = CSI[i-1];
unsigned Reg = Info.getReg();
// High registers (excluding lr) have already been dealt with
if (!(ARM::tGPRRegClass.contains(Reg) || Reg == ARM::LR))
continue;
if (Reg == ARM::LR) {
Info.setRestored(false);
if (!MBB.succ_empty() ||
MI->getOpcode() == ARM::TCRETURNdi ||
MI->getOpcode() == ARM::TCRETURNri)
// LR may only be popped into PC, as part of return sequence.
// If this isn't the return sequence, we'll need emitPopSpecialFixUp
// to restore LR the hard way.
// FIXME: if we don't pass any stack arguments it would be actually
// advantageous *and* correct to do the conversion to an ordinary call
// instruction here.
continue;
// Special epilogue for vararg functions. See emitEpilogue
if (isVarArg)
continue;
// ARMv4T requires BX, see emitEpilogue
if (!STI.hasV5TOps())
continue;
// Pop LR into PC.
Reg = ARM::PC;
(*MIB).setDesc(TII.get(ARM::tPOP_RET));
if (MI != MBB.end())
MIB.copyImplicitOps(*MI);
MI = MBB.erase(MI);
}
MIB.addReg(Reg, getDefRegState(true));
NeedsPop = true;
}
// It's illegal to emit pop instruction without operands.
if (NeedsPop)
MBB.insert(MI, &*MIB);
else
MF.DeleteMachineInstr(MIB);
return true;
}
bool Thumb1FrameLowering::emitPopSpecialFixUp(MachineBasicBlock &MBB,
bool DoIt) const {
MachineFunction &MF = *MBB.getParent();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
unsigned ArgRegsSaveSize = AFI->getArgRegsSaveSize();
const TargetInstrInfo &TII = *STI.getInstrInfo();
const ThumbRegisterInfo *RegInfo =
static_cast<const ThumbRegisterInfo *>(STI.getRegisterInfo());
// If MBBI is a return instruction, or is a tPOP followed by a return
// instruction in the successor BB, we may be able to directly restore
// LR in the PC.
// This is only possible with v5T ops (v4T can't change the Thumb bit via
// a POP PC instruction), and only if we do not need to emit any SP update.
// Otherwise, we need a temporary register to pop the value
// and copy that value into LR.
auto MBBI = MBB.getFirstTerminator();
bool CanRestoreDirectly = STI.hasV5TOps() && !ArgRegsSaveSize;
if (CanRestoreDirectly) {
if (MBBI != MBB.end() && MBBI->getOpcode() != ARM::tB)
CanRestoreDirectly = (MBBI->getOpcode() == ARM::tBX_RET ||
MBBI->getOpcode() == ARM::tPOP_RET);
else {
auto MBBI_prev = MBBI;
MBBI_prev--;
assert(MBBI_prev->getOpcode() == ARM::tPOP);
assert(MBB.succ_size() == 1);
if ((*MBB.succ_begin())->begin()->getOpcode() == ARM::tBX_RET)
MBBI = MBBI_prev; // Replace the final tPOP with a tPOP_RET.
else
CanRestoreDirectly = false;
}
}
if (CanRestoreDirectly) {
if (!DoIt || MBBI->getOpcode() == ARM::tPOP_RET)
return true;
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII.get(ARM::tPOP_RET))
.add(predOps(ARMCC::AL));
// Copy implicit ops and popped registers, if any.
for (auto MO: MBBI->operands())
if (MO.isReg() && (MO.isImplicit() || MO.isDef()))
MIB.add(MO);
MIB.addReg(ARM::PC, RegState::Define);
// Erase the old instruction (tBX_RET or tPOP).
MBB.erase(MBBI);
return true;
}
// Look for a temporary register to use.
// First, compute the liveness information.
const TargetRegisterInfo &TRI = *STI.getRegisterInfo();
LivePhysRegs UsedRegs(TRI);
UsedRegs.addLiveOuts(MBB);
// The semantic of pristines changed recently and now,
// the callee-saved registers that are touched in the function
// are not part of the pristines set anymore.
// Add those callee-saved now.
const MCPhysReg *CSRegs = TRI.getCalleeSavedRegs(&MF);
for (unsigned i = 0; CSRegs[i]; ++i)
UsedRegs.addReg(CSRegs[i]);
DebugLoc dl = DebugLoc();
if (MBBI != MBB.end()) {
dl = MBBI->getDebugLoc();
auto InstUpToMBBI = MBB.end();
while (InstUpToMBBI != MBBI)
// The pre-decrement is on purpose here.
// We want to have the liveness right before MBBI.
UsedRegs.stepBackward(*--InstUpToMBBI);
}
// Look for a register that can be directly use in the POP.
unsigned PopReg = 0;
// And some temporary register, just in case.
unsigned TemporaryReg = 0;
BitVector PopFriendly =
TRI.getAllocatableSet(MF, TRI.getRegClass(ARM::tGPRRegClassID));
assert(PopFriendly.any() && "No allocatable pop-friendly register?!");
// Rebuild the GPRs from the high registers because they are removed
// form the GPR reg class for thumb1.
BitVector GPRsNoLRSP =
TRI.getAllocatableSet(MF, TRI.getRegClass(ARM::hGPRRegClassID));
GPRsNoLRSP |= PopFriendly;
GPRsNoLRSP.reset(ARM::LR);
GPRsNoLRSP.reset(ARM::SP);
GPRsNoLRSP.reset(ARM::PC);
findTemporariesForLR(GPRsNoLRSP, PopFriendly, UsedRegs, PopReg, TemporaryReg);
// If we couldn't find a pop-friendly register, restore LR before popping the
// other callee-saved registers, so we can use one of them as a temporary.
bool UseLDRSP = false;
if (!PopReg && MBBI != MBB.begin()) {
auto PrevMBBI = MBBI;
PrevMBBI--;
if (PrevMBBI->getOpcode() == ARM::tPOP) {
MBBI = PrevMBBI;
UsedRegs.stepBackward(*MBBI);
findTemporariesForLR(GPRsNoLRSP, PopFriendly, UsedRegs, PopReg, TemporaryReg);
UseLDRSP = true;
}
}
if (!DoIt && !PopReg && !TemporaryReg)
return false;
assert((PopReg || TemporaryReg) && "Cannot get LR");
if (UseLDRSP) {
assert(PopReg && "Do not know how to get LR");
// Load the LR via LDR tmp, [SP, #off]
BuildMI(MBB, MBBI, dl, TII.get(ARM::tLDRspi))
.addReg(PopReg, RegState::Define)
.addReg(ARM::SP)
.addImm(MBBI->getNumExplicitOperands() - 2)
.add(predOps(ARMCC::AL));
// Move from the temporary register to the LR.
BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(ARM::LR, RegState::Define)
.addReg(PopReg, RegState::Kill)
.add(predOps(ARMCC::AL));
// Advance past the pop instruction.
MBBI++;
// Increment the SP.
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, ArgRegsSaveSize + 4);
return true;
}
if (TemporaryReg) {
assert(!PopReg && "Unnecessary MOV is about to be inserted");
PopReg = PopFriendly.find_first();
BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(TemporaryReg, RegState::Define)
.addReg(PopReg, RegState::Kill)
.add(predOps(ARMCC::AL));
}
if (MBBI != MBB.end() && MBBI->getOpcode() == ARM::tPOP_RET) {
// We couldn't use the direct restoration above, so
// perform the opposite conversion: tPOP_RET to tPOP.
MachineInstrBuilder MIB =
BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII.get(ARM::tPOP))
.add(predOps(ARMCC::AL));
bool Popped = false;
for (auto MO: MBBI->operands())
if (MO.isReg() && (MO.isImplicit() || MO.isDef()) &&
MO.getReg() != ARM::PC) {
MIB.add(MO);
if (!MO.isImplicit())
Popped = true;
}
// Is there anything left to pop?
if (!Popped)
MBB.erase(MIB.getInstr());
// Erase the old instruction.
MBB.erase(MBBI);
MBBI = BuildMI(MBB, MBB.end(), dl, TII.get(ARM::tBX_RET))
.add(predOps(ARMCC::AL));
}
assert(PopReg && "Do not know how to get LR");
BuildMI(MBB, MBBI, dl, TII.get(ARM::tPOP))
.add(predOps(ARMCC::AL))
.addReg(PopReg, RegState::Define);
emitSPUpdate(MBB, MBBI, TII, dl, *RegInfo, ArgRegsSaveSize);
BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(ARM::LR, RegState::Define)
.addReg(PopReg, RegState::Kill)
.add(predOps(ARMCC::AL));
if (TemporaryReg)
BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr))
.addReg(PopReg, RegState::Define)
.addReg(TemporaryReg, RegState::Kill)
.add(predOps(ARMCC::AL));
return true;
}
/// EmitSchedule - Emit the machine code in scheduled order. Return the new
/// InsertPos and MachineBasicBlock that contains this insertion
/// point. ScheduleDAGSDNodes holds a BB pointer for convenience, but this does
/// not necessarily refer to returned BB. The emitter may split blocks.
MachineBasicBlock *ScheduleDAGSDNodes::
EmitSchedule(MachineBasicBlock::iterator &InsertPos) {
InstrEmitter Emitter(BB, InsertPos);
DenseMap<SDValue, unsigned> VRBaseMap;
DenseMap<SUnit*, unsigned> CopyVRBaseMap;
SmallVector<std::pair<unsigned, MachineInstr*>, 32> Orders;
SmallSet<unsigned, 8> Seen;
bool HasDbg = DAG->hasDebugValues();
// If this is the first BB, emit byval parameter dbg_value's.
if (HasDbg && BB->getParent()->begin() == MachineFunction::iterator(BB)) {
SDDbgInfo::DbgIterator PDI = DAG->ByvalParmDbgBegin();
SDDbgInfo::DbgIterator PDE = DAG->ByvalParmDbgEnd();
for (; PDI != PDE; ++PDI) {
MachineInstr *DbgMI= Emitter.EmitDbgValue(*PDI, VRBaseMap);
if (DbgMI)
BB->insert(InsertPos, DbgMI);
}
}
for (unsigned i = 0, e = Sequence.size(); i != e; i++) {
SUnit *SU = Sequence[i];
if (!SU) {
// Null SUnit* is a noop.
TII->insertNoop(*Emitter.getBlock(), InsertPos);
continue;
}
// For pre-regalloc scheduling, create instructions corresponding to the
// SDNode and any glued SDNodes and append them to the block.
if (!SU->getNode()) {
// Emit a copy.
EmitPhysRegCopy(SU, CopyVRBaseMap, InsertPos);
continue;
}
SmallVector<SDNode *, 4> GluedNodes;
for (SDNode *N = SU->getNode()->getGluedNode(); N;
N = N->getGluedNode())
GluedNodes.push_back(N);
while (!GluedNodes.empty()) {
SDNode *N = GluedNodes.back();
Emitter.EmitNode(GluedNodes.back(), SU->OrigNode != SU, SU->isCloned,
VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(N, DAG, Emitter, VRBaseMap, Orders, Seen);
GluedNodes.pop_back();
}
Emitter.EmitNode(SU->getNode(), SU->OrigNode != SU, SU->isCloned,
VRBaseMap);
// Remember the source order of the inserted instruction.
if (HasDbg)
ProcessSourceNode(SU->getNode(), DAG, Emitter, VRBaseMap, Orders,
Seen);
}
// Insert all the dbg_values which have not already been inserted in source
// order sequence.
if (HasDbg) {
MachineBasicBlock::iterator BBBegin = BB->getFirstNonPHI();
// Sort the source order instructions and use the order to insert debug
// values.
std::sort(Orders.begin(), Orders.end(), OrderSorter());
SDDbgInfo::DbgIterator DI = DAG->DbgBegin();
SDDbgInfo::DbgIterator DE = DAG->DbgEnd();
// Now emit the rest according to source order.
unsigned LastOrder = 0;
for (unsigned i = 0, e = Orders.size(); i != e && DI != DE; ++i) {
unsigned Order = Orders[i].first;
MachineInstr *MI = Orders[i].second;
// Insert all SDDbgValue's whose order(s) are before "Order".
if (!MI)
continue;
for (; DI != DE &&
(*DI)->getOrder() >= LastOrder && (*DI)->getOrder() < Order; ++DI) {
if ((*DI)->isInvalidated())
continue;
MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap);
if (DbgMI) {
if (!LastOrder)
// Insert to start of the BB (after PHIs).
BB->insert(BBBegin, DbgMI);
else {
// Insert at the instruction, which may be in a different
// block, if the block was split by a custom inserter.
MachineBasicBlock::iterator Pos = MI;
MI->getParent()->insert(llvm::next(Pos), DbgMI);
}
}
}
LastOrder = Order;
}
// Add trailing DbgValue's before the terminator. FIXME: May want to add
// some of them before one or more conditional branches?
SmallVector<MachineInstr*, 8> DbgMIs;
while (DI != DE) {
if (!(*DI)->isInvalidated())
if (MachineInstr *DbgMI = Emitter.EmitDbgValue(*DI, VRBaseMap))
DbgMIs.push_back(DbgMI);
++DI;
}
MachineBasicBlock *InsertBB = Emitter.getBlock();
MachineBasicBlock::iterator Pos = InsertBB->getFirstTerminator();
InsertBB->insert(Pos, DbgMIs.begin(), DbgMIs.end());
}
InsertPos = Emitter.getInsertPos();
return Emitter.getBlock();
}
/// TailDuplicate - If it is profitable, duplicate TailBB's contents in each
/// of its predecessors.
bool
TailDuplicatePass::TailDuplicate(MachineBasicBlock *TailBB, MachineFunction &MF,
SmallVector<MachineBasicBlock*, 8> &TDBBs,
SmallVector<MachineInstr*, 16> &Copies) {
// Set the limit on the number of instructions to duplicate, with a default
// of one less than the tail-merge threshold. When optimizing for size,
// duplicate only one, because one branch instruction can be eliminated to
// compensate for the duplication.
unsigned MaxDuplicateCount;
if (TailDuplicateSize.getNumOccurrences() == 0 &&
MF.getFunction()->hasFnAttr(Attribute::OptimizeForSize))
MaxDuplicateCount = 1;
else
MaxDuplicateCount = TailDuplicateSize;
if (PreRegAlloc) {
if (TailBB->empty())
return false;
const TargetInstrDesc &TID = TailBB->back().getDesc();
// Pre-regalloc tail duplication hurts compile time and doesn't help
// much except for indirect branches and returns.
if (!TID.isIndirectBranch() && !TID.isReturn())
return false;
// If the target has hardware branch prediction that can handle indirect
// branches, duplicating them can often make them predictable when there
// are common paths through the code. The limit needs to be high enough
// to allow undoing the effects of tail merging and other optimizations
// that rearrange the predecessors of the indirect branch.
MaxDuplicateCount = 20;
}
// Don't try to tail-duplicate single-block loops.
if (TailBB->isSuccessor(TailBB))
return false;
// Check the instructions in the block to determine whether tail-duplication
// is invalid or unlikely to be profitable.
unsigned InstrCount = 0;
bool HasCall = false;
for (MachineBasicBlock::iterator I = TailBB->begin();
I != TailBB->end(); ++I) {
// Non-duplicable things shouldn't be tail-duplicated.
if (I->getDesc().isNotDuplicable()) return false;
// Do not duplicate 'return' instructions if this is a pre-regalloc run.
// A return may expand into a lot more instructions (e.g. reload of callee
// saved registers) after PEI.
if (PreRegAlloc && I->getDesc().isReturn()) return false;
// Don't duplicate more than the threshold.
if (InstrCount == MaxDuplicateCount) return false;
// Remember if we saw a call.
if (I->getDesc().isCall()) HasCall = true;
if (!I->isPHI() && !I->isDebugValue())
InstrCount += 1;
}
// Don't tail-duplicate calls before register allocation. Calls presents a
// barrier to register allocation so duplicating them may end up increasing
// spills.
if (InstrCount > 1 && (PreRegAlloc && HasCall))
return false;
DEBUG(dbgs() << "\n*** Tail-duplicating BB#" << TailBB->getNumber() << '\n');
// Iterate through all the unique predecessors and tail-duplicate this
// block into them, if possible. Copying the list ahead of time also
// avoids trouble with the predecessor list reallocating.
bool Changed = false;
SmallSetVector<MachineBasicBlock*, 8> Preds(TailBB->pred_begin(),
TailBB->pred_end());
for (SmallSetVector<MachineBasicBlock *, 8>::iterator PI = Preds.begin(),
PE = Preds.end(); PI != PE; ++PI) {
MachineBasicBlock *PredBB = *PI;
assert(TailBB != PredBB &&
"Single-block loop should have been rejected earlier!");
if (PredBB->succ_size() > 1) continue;
MachineBasicBlock *PredTBB, *PredFBB;
SmallVector<MachineOperand, 4> PredCond;
if (TII->AnalyzeBranch(*PredBB, PredTBB, PredFBB, PredCond, true))
continue;
if (!PredCond.empty())
continue;
// EH edges are ignored by AnalyzeBranch.
if (PredBB->succ_size() != 1)
continue;
// Don't duplicate into a fall-through predecessor (at least for now).
if (PredBB->isLayoutSuccessor(TailBB) && PredBB->canFallThrough())
continue;
DEBUG(dbgs() << "\nTail-duplicating into PredBB: " << *PredBB
<< "From Succ: " << *TailBB);
TDBBs.push_back(PredBB);
// Remove PredBB's unconditional branch.
TII->RemoveBranch(*PredBB);
// Clone the contents of TailBB into PredBB.
DenseMap<unsigned, unsigned> LocalVRMap;
SmallVector<std::pair<unsigned,unsigned>, 4> CopyInfos;
MachineBasicBlock::iterator I = TailBB->begin();
while (I != TailBB->end()) {
MachineInstr *MI = &*I;
++I;
if (MI->isPHI()) {
// Replace the uses of the def of the PHI with the register coming
// from PredBB.
ProcessPHI(MI, TailBB, PredBB, LocalVRMap, CopyInfos);
} else {
// Replace def of virtual registers with new registers, and update
// uses with PHI source register or the new registers.
DuplicateInstruction(MI, TailBB, PredBB, MF, LocalVRMap);
}
}
MachineBasicBlock::iterator Loc = PredBB->getFirstTerminator();
for (unsigned i = 0, e = CopyInfos.size(); i != e; ++i) {
Copies.push_back(BuildMI(*PredBB, Loc, DebugLoc(),
TII->get(TargetOpcode::COPY),
CopyInfos[i].first).addReg(CopyInfos[i].second));
}
NumInstrDups += TailBB->size() - 1; // subtract one for removed branch
// Update the CFG.
PredBB->removeSuccessor(PredBB->succ_begin());
assert(PredBB->succ_empty() &&
"TailDuplicate called on block with multiple successors!");
for (MachineBasicBlock::succ_iterator I = TailBB->succ_begin(),
E = TailBB->succ_end(); I != E; ++I)
PredBB->addSuccessor(*I);
Changed = true;
++NumTailDups;
}
// If TailBB was duplicated into all its predecessors except for the prior
// block, which falls through unconditionally, move the contents of this
// block into the prior block.
MachineBasicBlock *PrevBB = prior(MachineFunction::iterator(TailBB));
MachineBasicBlock *PriorTBB = 0, *PriorFBB = 0;
SmallVector<MachineOperand, 4> PriorCond;
bool PriorUnAnalyzable =
TII->AnalyzeBranch(*PrevBB, PriorTBB, PriorFBB, PriorCond, true);
// This has to check PrevBB->succ_size() because EH edges are ignored by
// AnalyzeBranch.
if (!PriorUnAnalyzable && PriorCond.empty() && !PriorTBB &&
TailBB->pred_size() == 1 && PrevBB->succ_size() == 1 &&
!TailBB->hasAddressTaken()) {
DEBUG(dbgs() << "\nMerging into block: " << *PrevBB
<< "From MBB: " << *TailBB);
if (PreRegAlloc) {
DenseMap<unsigned, unsigned> LocalVRMap;
SmallVector<std::pair<unsigned,unsigned>, 4> CopyInfos;
MachineBasicBlock::iterator I = TailBB->begin();
// Process PHI instructions first.
while (I != TailBB->end() && I->isPHI()) {
// Replace the uses of the def of the PHI with the register coming
// from PredBB.
MachineInstr *MI = &*I++;
ProcessPHI(MI, TailBB, PrevBB, LocalVRMap, CopyInfos);
if (MI->getParent())
MI->eraseFromParent();
}
// Now copy the non-PHI instructions.
while (I != TailBB->end()) {
// Replace def of virtual registers with new registers, and update
// uses with PHI source register or the new registers.
MachineInstr *MI = &*I++;
DuplicateInstruction(MI, TailBB, PrevBB, MF, LocalVRMap);
MI->eraseFromParent();
}
MachineBasicBlock::iterator Loc = PrevBB->getFirstTerminator();
for (unsigned i = 0, e = CopyInfos.size(); i != e; ++i) {
Copies.push_back(BuildMI(*PrevBB, Loc, DebugLoc(),
TII->get(TargetOpcode::COPY),
CopyInfos[i].first)
.addReg(CopyInfos[i].second));
}
} else {
// No PHIs to worry about, just splice the instructions over.
PrevBB->splice(PrevBB->end(), TailBB, TailBB->begin(), TailBB->end());
}
PrevBB->removeSuccessor(PrevBB->succ_begin());
assert(PrevBB->succ_empty());
PrevBB->transferSuccessors(TailBB);
TDBBs.push_back(PrevBB);
Changed = true;
}
return Changed;
}
bool SIFixSGPRLiveRanges::runOnMachineFunction(MachineFunction &MF) {
MachineRegisterInfo &MRI = MF.getRegInfo();
const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
const SIRegisterInfo *TRI = static_cast<const SIRegisterInfo *>(
MF.getSubtarget().getRegisterInfo());
LiveIntervals *LIS = &getAnalysis<LiveIntervals>();
MachinePostDominatorTree *PDT = &getAnalysis<MachinePostDominatorTree>();
std::vector<std::pair<unsigned, LiveRange *>> SGPRLiveRanges;
// First pass, collect all live intervals for SGPRs
for (const MachineBasicBlock &MBB : MF) {
for (const MachineInstr &MI : MBB) {
for (const MachineOperand &MO : MI.defs()) {
if (MO.isImplicit())
continue;
unsigned Def = MO.getReg();
if (TargetRegisterInfo::isVirtualRegister(Def)) {
if (TRI->isSGPRClass(MRI.getRegClass(Def)))
SGPRLiveRanges.push_back(
std::make_pair(Def, &LIS->getInterval(Def)));
} else if (TRI->isSGPRClass(TRI->getPhysRegClass(Def))) {
SGPRLiveRanges.push_back(
std::make_pair(Def, &LIS->getRegUnit(Def)));
}
}
}
}
// Second pass fix the intervals
for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
BI != BE; ++BI) {
MachineBasicBlock &MBB = *BI;
if (MBB.succ_size() < 2)
continue;
// We have structured control flow, so number of succesors should be two.
assert(MBB.succ_size() == 2);
MachineBasicBlock *SuccA = *MBB.succ_begin();
MachineBasicBlock *SuccB = *(++MBB.succ_begin());
MachineBasicBlock *NCD = PDT->findNearestCommonDominator(SuccA, SuccB);
if (!NCD)
continue;
MachineBasicBlock::iterator NCDTerm = NCD->getFirstTerminator();
if (NCDTerm != NCD->end() && NCDTerm->getOpcode() == AMDGPU::SI_ELSE) {
assert(NCD->succ_size() == 2);
// We want to make sure we insert the Use after the ENDIF, not after
// the ELSE.
NCD = PDT->findNearestCommonDominator(*NCD->succ_begin(),
*(++NCD->succ_begin()));
}
assert(SuccA && SuccB);
for (std::pair<unsigned, LiveRange*> RegLR : SGPRLiveRanges) {
unsigned Reg = RegLR.first;
LiveRange *LR = RegLR.second;
// FIXME: We could be smarter here. If the register is Live-In to
// one block, but the other doesn't have any SGPR defs, then there
// won't be a conflict. Also, if the branch decision is based on
// a value in an SGPR, then there will be no conflict.
bool LiveInToA = LIS->isLiveInToMBB(*LR, SuccA);
bool LiveInToB = LIS->isLiveInToMBB(*LR, SuccB);
if ((!LiveInToA && !LiveInToB) ||
(LiveInToA && LiveInToB))
continue;
// This interval is live in to one successor, but not the other, so
// we need to update its range so it is live in to both.
DEBUG(dbgs() << "Possible SGPR conflict detected " << " in " << *LR <<
" BB#" << SuccA->getNumber() << ", BB#" <<
SuccB->getNumber() <<
" with NCD = " << NCD->getNumber() << '\n');
// FIXME: Need to figure out how to update LiveRange here so this pass
// will be able to preserve LiveInterval analysis.
BuildMI(*NCD, NCD->getFirstNonPHI(), DebugLoc(),
TII->get(AMDGPU::SGPR_USE))
.addReg(Reg, RegState::Implicit);
DEBUG(NCD->getFirstNonPHI()->dump());
}
}
return false;
}
bool SIFixSGPRLiveRanges::runOnMachineFunction(MachineFunction &MF) {
MachineRegisterInfo &MRI = MF.getRegInfo();
const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
const SIRegisterInfo *TRI = static_cast<const SIRegisterInfo *>(
MF.getSubtarget().getRegisterInfo());
bool MadeChange = false;
MachinePostDominatorTree *PDT = &getAnalysis<MachinePostDominatorTree>();
std::vector<std::pair<unsigned, LiveRange *>> SGPRLiveRanges;
LiveIntervals *LIS = &getAnalysis<LiveIntervals>();
LiveVariables *LV = getAnalysisIfAvailable<LiveVariables>();
MachineBasicBlock *Entry = MF.begin();
// Use a depth first order so that in SSA, we encounter all defs before
// uses. Once the defs of the block have been found, attempt to insert
// SGPR_USE instructions in successor blocks if required.
for (MachineBasicBlock *MBB : depth_first(Entry)) {
for (const MachineInstr &MI : *MBB) {
for (const MachineOperand &MO : MI.defs()) {
if (MO.isImplicit())
continue;
unsigned Def = MO.getReg();
if (TargetRegisterInfo::isVirtualRegister(Def)) {
if (TRI->isSGPRClass(MRI.getRegClass(Def))) {
// Only consider defs that are live outs. We don't care about def /
// use within the same block.
LiveRange &LR = LIS->getInterval(Def);
if (LIS->isLiveOutOfMBB(LR, MBB))
SGPRLiveRanges.push_back(std::make_pair(Def, &LR));
}
} else if (TRI->isSGPRClass(TRI->getPhysRegClass(Def))) {
SGPRLiveRanges.push_back(std::make_pair(Def, &LIS->getRegUnit(Def)));
}
}
}
if (MBB->succ_size() < 2)
continue;
// We have structured control flow, so the number of successors should be
// two.
assert(MBB->succ_size() == 2);
MachineBasicBlock *SuccA = *MBB->succ_begin();
MachineBasicBlock *SuccB = *(++MBB->succ_begin());
MachineBasicBlock *NCD = PDT->findNearestCommonDominator(SuccA, SuccB);
if (!NCD)
continue;
MachineBasicBlock::iterator NCDTerm = NCD->getFirstTerminator();
if (NCDTerm != NCD->end() && NCDTerm->getOpcode() == AMDGPU::SI_ELSE) {
assert(NCD->succ_size() == 2);
// We want to make sure we insert the Use after the ENDIF, not after
// the ELSE.
NCD = PDT->findNearestCommonDominator(*NCD->succ_begin(),
*(++NCD->succ_begin()));
}
for (std::pair<unsigned, LiveRange*> RegLR : SGPRLiveRanges) {
unsigned Reg = RegLR.first;
LiveRange *LR = RegLR.second;
// FIXME: We could be smarter here. If the register is Live-In to one
// block, but the other doesn't have any SGPR defs, then there won't be a
// conflict. Also, if the branch condition is uniform then there will be
// no conflict.
bool LiveInToA = LIS->isLiveInToMBB(*LR, SuccA);
bool LiveInToB = LIS->isLiveInToMBB(*LR, SuccB);
if (!LiveInToA && !LiveInToB) {
DEBUG(dbgs() << PrintReg(Reg, TRI, 0)
<< " is live into neither successor\n");
continue;
}
if (LiveInToA && LiveInToB) {
DEBUG(dbgs() << PrintReg(Reg, TRI, 0)
<< " is live into both successors\n");
continue;
}
// This interval is live in to one successor, but not the other, so
// we need to update its range so it is live in to both.
DEBUG(dbgs() << "Possible SGPR conflict detected for "
<< PrintReg(Reg, TRI, 0) << " in " << *LR
<< " BB#" << SuccA->getNumber() << ", BB#"
<< SuccB->getNumber()
<< " with NCD = BB#" << NCD->getNumber() << '\n');
assert(TargetRegisterInfo::isVirtualRegister(Reg) &&
"Not expecting to extend live range of physreg");
// FIXME: Need to figure out how to update LiveRange here so this pass
// will be able to preserve LiveInterval analysis.
MachineInstr *NCDSGPRUse =
BuildMI(*NCD, NCD->getFirstNonPHI(), DebugLoc(),
TII->get(AMDGPU::SGPR_USE))
.addReg(Reg, RegState::Implicit);
MadeChange = true;
SlotIndex SI = LIS->InsertMachineInstrInMaps(NCDSGPRUse);
LIS->extendToIndices(*LR, SI.getRegSlot());
if (LV) {
// TODO: This won't work post-SSA
LV->HandleVirtRegUse(Reg, NCD, NCDSGPRUse);
}
DEBUG(NCDSGPRUse->dump());
}
}
return MadeChange;
}
/// converToHardwareLoop - check if the loop is a candidate for
/// converting to a hardware loop. If so, then perform the
/// transformation.
///
/// This function works on innermost loops first. A loop can
/// be converted if it is a counting loop; either a register
/// value or an immediate.
///
/// The code makes several assumptions about the representation
/// of the loop in llvm.
bool HexagonHardwareLoops::convertToHardwareLoop(MachineLoop *L) {
bool Changed = false;
// Process nested loops first.
for (MachineLoop::iterator I = L->begin(), E = L->end(); I != E; ++I) {
Changed |= convertToHardwareLoop(*I);
}
// If a nested loop has been converted, then we can't convert this loop.
if (Changed) {
return Changed;
}
// Are we able to determine the trip count for the loop?
CountValue *TripCount = getTripCount(L);
if (TripCount == 0) {
return false;
}
// Does the loop contain any invalid instructions?
if (containsInvalidInstruction(L)) {
return false;
}
MachineBasicBlock *Preheader = L->getLoopPreheader();
// No preheader means there's not place for the loop instr.
if (Preheader == 0) {
return false;
}
MachineBasicBlock::iterator InsertPos = Preheader->getFirstTerminator();
MachineBasicBlock *LastMBB = L->getExitingBlock();
// Don't generate hw loop if the loop has more than one exit.
if (LastMBB == 0) {
return false;
}
MachineBasicBlock::iterator LastI = LastMBB->getFirstTerminator();
// Determine the loop start.
MachineBasicBlock *LoopStart = L->getTopBlock();
if (L->getLoopLatch() != LastMBB) {
// When the exit and latch are not the same, use the latch block as the
// start.
// The loop start address is used only after the 1st iteration, and the loop
// latch may contains instrs. that need to be executed after the 1st iter.
LoopStart = L->getLoopLatch();
// Make sure the latch is a successor of the exit, otherwise it won't work.
if (!LastMBB->isSuccessor(LoopStart)) {
return false;
}
}
// Convert the loop to a hardware loop
DEBUG(dbgs() << "Change to hardware loop at "; L->dump());
if (TripCount->isReg()) {
// Create a copy of the loop count register.
MachineFunction *MF = LastMBB->getParent();
const TargetRegisterClass *RC =
MF->getRegInfo().getRegClass(TripCount->getReg());
unsigned CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(TargetOpcode::COPY), CountReg).addReg(TripCount->getReg());
if (TripCount->isNeg()) {
unsigned CountReg1 = CountReg;
CountReg = MF->getRegInfo().createVirtualRegister(RC);
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::NEG), CountReg).addReg(CountReg1);
}
// Add the Loop instruction to the begining of the loop.
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::LOOP0_r)).addMBB(LoopStart).addReg(CountReg);
} else {
assert(TripCount->isImm() && "Expecting immedate vaule for trip count");
// Add the Loop immediate instruction to the beginning of the loop.
int64_t CountImm = TripCount->getImm();
BuildMI(*Preheader, InsertPos, InsertPos->getDebugLoc(),
TII->get(Hexagon::LOOP0_i)).addMBB(LoopStart).addImm(CountImm);
}
// Make sure the loop start always has a reference in the CFG. We need to
// create a BlockAddress operand to get this mechanism to work both the
// MachineBasicBlock and BasicBlock objects need the flag set.
LoopStart->setHasAddressTaken();
// This line is needed to set the hasAddressTaken flag on the BasicBlock
// object
BlockAddress::get(const_cast<BasicBlock *>(LoopStart->getBasicBlock()));
// Replace the loop branch with an endloop instruction.
DebugLoc dl = LastI->getDebugLoc();
BuildMI(*LastMBB, LastI, dl, TII->get(Hexagon::ENDLOOP0)).addMBB(LoopStart);
// The loop ends with either:
// - a conditional branch followed by an unconditional branch, or
// - a conditional branch to the loop start.
if (LastI->getOpcode() == Hexagon::JMP_c ||
LastI->getOpcode() == Hexagon::JMP_cNot) {
// delete one and change/add an uncond. branch to out of the loop
MachineBasicBlock *BranchTarget = LastI->getOperand(1).getMBB();
LastI = LastMBB->erase(LastI);
if (!L->contains(BranchTarget)) {
if (LastI != LastMBB->end()) {
TII->RemoveBranch(*LastMBB);
}
SmallVector<MachineOperand, 0> Cond;
TII->InsertBranch(*LastMBB, BranchTarget, 0, Cond, dl);
}
} else {
// Conditional branch to loop start; just delete it.
LastMBB->erase(LastI);
}
delete TripCount;
++NumHWLoops;
return true;
}
/// Walk the specified region of the CFG and hoist loop invariants out to the
/// preheader.
void MachineLICM::HoistRegionPostRA() {
MachineBasicBlock *Preheader = getCurPreheader();
if (!Preheader)
return;
unsigned NumRegs = TRI->getNumRegs();
BitVector PhysRegDefs(NumRegs); // Regs defined once in the loop.
BitVector PhysRegClobbers(NumRegs); // Regs defined more than once.
SmallVector<CandidateInfo, 32> Candidates;
SmallSet<int, 32> StoredFIs;
// Walk the entire region, count number of defs for each register, and
// collect potential LICM candidates.
const std::vector<MachineBasicBlock *> &Blocks = CurLoop->getBlocks();
for (MachineBasicBlock *BB : Blocks) {
// If the header of the loop containing this basic block is a landing pad,
// then don't try to hoist instructions out of this loop.
const MachineLoop *ML = MLI->getLoopFor(BB);
if (ML && ML->getHeader()->isEHPad()) continue;
// Conservatively treat live-in's as an external def.
// FIXME: That means a reload that're reused in successor block(s) will not
// be LICM'ed.
for (const auto &LI : BB->liveins()) {
for (MCRegAliasIterator AI(LI.PhysReg, TRI, true); AI.isValid(); ++AI)
PhysRegDefs.set(*AI);
}
SpeculationState = SpeculateUnknown;
for (MachineInstr &MI : *BB)
ProcessMI(&MI, PhysRegDefs, PhysRegClobbers, StoredFIs, Candidates);
}
// Gather the registers read / clobbered by the terminator.
BitVector TermRegs(NumRegs);
MachineBasicBlock::iterator TI = Preheader->getFirstTerminator();
if (TI != Preheader->end()) {
for (const MachineOperand &MO : TI->operands()) {
if (!MO.isReg())
continue;
unsigned Reg = MO.getReg();
if (!Reg)
continue;
for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
TermRegs.set(*AI);
}
}
// Now evaluate whether the potential candidates qualify.
// 1. Check if the candidate defined register is defined by another
// instruction in the loop.
// 2. If the candidate is a load from stack slot (always true for now),
// check if the slot is stored anywhere in the loop.
// 3. Make sure candidate def should not clobber
// registers read by the terminator. Similarly its def should not be
// clobbered by the terminator.
for (CandidateInfo &Candidate : Candidates) {
if (Candidate.FI != INT_MIN &&
StoredFIs.count(Candidate.FI))
continue;
unsigned Def = Candidate.Def;
if (!PhysRegClobbers.test(Def) && !TermRegs.test(Def)) {
bool Safe = true;
MachineInstr *MI = Candidate.MI;
for (const MachineOperand &MO : MI->operands()) {
if (!MO.isReg() || MO.isDef() || !MO.getReg())
continue;
unsigned Reg = MO.getReg();
if (PhysRegDefs.test(Reg) ||
PhysRegClobbers.test(Reg)) {
// If it's using a non-loop-invariant register, then it's obviously
// not safe to hoist.
Safe = false;
break;
}
}
if (Safe)
HoistPostRA(MI, Candidate.Def);
}
}
}
