/// ComputeValueVTs - Given an LLVM IR type, compute a sequence of
/// EVTs that represent all the individual underlying
/// non-aggregate types that comprise it.
///
/// If Offsets is non-null, it points to a vector to be filled in
/// with the in-memory offsets of each of the individual values.
///
void llvm::ComputeValueVTs(const TargetLowering &TLI, Type *Ty,
SmallVectorImpl<EVT> &ValueVTs,
SmallVectorImpl<uint64_t> *Offsets,
uint64_t StartingOffset) {
// Given a struct type, recursively traverse the elements.
if (StructType *STy = dyn_cast<StructType>(Ty)) {
const StructLayout *SL = TLI.getDataLayout()->getStructLayout(STy);
for (StructType::element_iterator EB = STy->element_begin(),
EI = EB,
EE = STy->element_end();
EI != EE; ++EI)
ComputeValueVTs(TLI, *EI, ValueVTs, Offsets,
StartingOffset + SL->getElementOffset(EI - EB));
return;
}
// Given an array type, recursively traverse the elements.
if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
Type *EltTy = ATy->getElementType();
uint64_t EltSize = TLI.getDataLayout()->getTypeAllocSize(EltTy);
for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i)
ComputeValueVTs(TLI, EltTy, ValueVTs, Offsets,
StartingOffset + i * EltSize);
return;
}
// Interpret void as zero return values.
if (Ty->isVoidTy())
return;
// Base case: we can get an EVT for this LLVM IR type.
ValueVTs.push_back(TLI.getValueType(Ty));
if (Offsets)
Offsets->push_back(StartingOffset);
}
/// Get the EVTs and ArgFlags collections that represent the legalized return
/// type of the given function. This does not require a DAG or a return value,
/// and is suitable for use before any DAGs for the function are constructed.
/// TODO: Move this out of TargetLowering.cpp.
void llvm::GetReturnInfo(Type* ReturnType, AttributeSet attr,
SmallVectorImpl<ISD::OutputArg> &Outs,
const TargetLowering &TLI) {
SmallVector<EVT, 4> ValueVTs;
ComputeValueVTs(TLI, ReturnType, ValueVTs);
unsigned NumValues = ValueVTs.size();
if (NumValues == 0) return;
for (unsigned j = 0, f = NumValues; j != f; ++j) {
EVT VT = ValueVTs[j];
ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::SExt))
ExtendKind = ISD::SIGN_EXTEND;
else if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::ZExt))
ExtendKind = ISD::ZERO_EXTEND;
// FIXME: C calling convention requires the return type to be promoted to
// at least 32-bit. But this is not necessary for non-C calling
// conventions. The frontend should mark functions whose return values
// require promoting with signext or zeroext attributes.
if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger()) {
MVT MinVT = TLI.getRegisterType(ReturnType->getContext(), MVT::i32);
if (VT.bitsLT(MinVT))
VT = MinVT;
}
unsigned NumParts = TLI.getNumRegisters(ReturnType->getContext(), VT);
MVT PartVT = TLI.getRegisterType(ReturnType->getContext(), VT);
// 'inreg' on function refers to return value
ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::InReg))
Flags.setInReg();
// Propagate extension type if any
if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::SExt))
Flags.setSExt();
else if (attr.hasAttribute(AttributeSet::ReturnIndex, Attribute::ZExt))
Flags.setZExt();
for (unsigned i = 0; i < NumParts; ++i)
Outs.push_back(ISD::OutputArg(Flags, PartVT, /*isFixed=*/true, 0, 0));
}
}
/// IsOperandAMemoryOperand - Check to see if all uses of OpVal by the specified
/// inline asm call are due to memory operands. If so, return true, otherwise
/// return false.
static bool IsOperandAMemoryOperand(CallInst *CI, InlineAsm *IA, Value *OpVal,
const TargetLowering &TLI) {
std::vector<TargetLowering::AsmOperandInfo> TargetConstraints = TLI.ParseConstraints(ImmutableCallSite(CI));
for (unsigned i = 0, e = TargetConstraints.size(); i != e; ++i) {
TargetLowering::AsmOperandInfo &OpInfo = TargetConstraints[i];
// Compute the constraint code and ConstraintType to use.
TLI.ComputeConstraintToUse(OpInfo, SDValue());
// If this asm operand is our Value*, and if it isn't an indirect memory
// operand, we can't fold it!
if (OpInfo.CallOperandVal == OpVal &&
(OpInfo.ConstraintType != TargetLowering::C_Memory ||
!OpInfo.isIndirect))
return false;
}
return true;
}
bool llvm::isInTailCallPosition(SelectionDAG &DAG, SDNode *Node,
const TargetLowering &TLI) {
const Function *F = DAG.getMachineFunction().getFunction();
// Conservatively require the attributes of the call to match those of
// the return. Ignore noalias because it doesn't affect the call sequence.
unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
if (CallerRetAttr & ~Attribute::NoAlias)
return false;
// It's not safe to eliminate the sign / zero extension of the return value.
if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
return false;
// Check if the only use is a function return node.
return TLI.isUsedByReturnOnly(Node);
}
/// hasInlineAsmMemConstraint - Return true if the inline asm instruction being
/// processed uses a memory 'm' constraint.
bool
llvm::hasInlineAsmMemConstraint(InlineAsm::ConstraintInfoVector &CInfos,
const TargetLowering &TLI) {
for (unsigned i = 0, e = CInfos.size(); i != e; ++i) {
InlineAsm::ConstraintInfo &CI = CInfos[i];
for (unsigned j = 0, ee = CI.Codes.size(); j != ee; ++j) {
TargetLowering::ConstraintType CType = TLI.getConstraintType(CI.Codes[j]);
if (CType == TargetLowering::C_Memory)
return true;
}
// Indirect operand accesses access memory.
if (CI.isIndirect)
return true;
}
return false;
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
!isa<UnreachableInst>(Term)))
return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(BBI))
return false;
}
return returnTypeIsEligibleForTailCall(ExitBB->getParent(), I, Ret, TLI);
}
/// IsOperandAMemoryOperand - Check to see if all uses of OpVal by the specified
/// inline asm call are due to memory operands. If so, return true, otherwise
/// return false.
static bool IsOperandAMemoryOperand(CallInst *CI, InlineAsm *IA, Value *OpVal,
const TargetLowering &TLI) {
std::vector<InlineAsm::ConstraintInfo>
Constraints = IA->ParseConstraints();
unsigned ArgNo = 1; // ArgNo - The operand of the CallInst.
for (unsigned i = 0, e = Constraints.size(); i != e; ++i) {
TargetLowering::AsmOperandInfo OpInfo(Constraints[i]);
// Compute the value type for each operand.
switch (OpInfo.Type) {
case InlineAsm::isOutput:
if (OpInfo.isIndirect)
OpInfo.CallOperandVal = CI->getOperand(ArgNo++);
break;
case InlineAsm::isInput:
OpInfo.CallOperandVal = CI->getOperand(ArgNo++);
break;
case InlineAsm::isClobber:
// Nothing to do.
break;
}
// Compute the constraint code and ConstraintType to use.
TLI.ComputeConstraintToUse(OpInfo, SDValue(),
OpInfo.ConstraintType == TargetLowering::C_Memory);
// If this asm operand is our Value*, and if it isn't an indirect memory
// operand, we can't fold it!
if (OpInfo.CallOperandVal == OpVal &&
(OpInfo.ConstraintType != TargetLowering::C_Memory ||
!OpInfo.isIndirect))
return false;
}
return true;
}
/// OptimizeNoopCopyExpression - If the specified cast instruction is a noop
/// copy (e.g. it's casting from one pointer type to another, i32->i8 on PPC),
/// sink it into user blocks to reduce the number of virtual
/// registers that must be created and coalesced.
///
/// Return true if any changes are made.
///
static bool OptimizeNoopCopyExpression(CastInst *CI, const TargetLowering &TLI) {
// If this is a noop copy,
EVT SrcVT = TLI.getValueType(CI->getOperand(0)->getType());
EVT DstVT = TLI.getValueType(CI->getType());
// This is an fp<->int conversion?
if (SrcVT.isInteger() != DstVT.isInteger())
return false;
// If this is an extension, it will be a zero or sign extension, which
// isn't a noop.
if (SrcVT.bitsLT(DstVT)) return false;
// If these values will be promoted, find out what they will be promoted
// to. This helps us consider truncates on PPC as noop copies when they
// are.
if (TLI.getTypeAction(CI->getContext(), SrcVT) ==
TargetLowering::TypePromoteInteger)
SrcVT = TLI.getTypeToTransformTo(CI->getContext(), SrcVT);
if (TLI.getTypeAction(CI->getContext(), DstVT) ==
TargetLowering::TypePromoteInteger)
DstVT = TLI.getTypeToTransformTo(CI->getContext(), DstVT);
// If, after promotion, these are the same types, this is a noop copy.
if (SrcVT != DstVT)
return false;
BasicBlock *DefBB = CI->getParent();
/// InsertedCasts - Only insert a cast in each block once.
DenseMap<BasicBlock*, CastInst*> InsertedCasts;
bool MadeChange = false;
for (Value::use_iterator UI = CI->use_begin(), E = CI->use_end();
UI != E; ) {
Use &TheUse = UI.getUse();
Instruction *User = cast<Instruction>(*UI);
// Figure out which BB this cast is used in. For PHI's this is the
// appropriate predecessor block.
BasicBlock *UserBB = User->getParent();
if (PHINode *PN = dyn_cast<PHINode>(User)) {
UserBB = PN->getIncomingBlock(UI);
}
// Preincrement use iterator so we don't invalidate it.
++UI;
// If this user is in the same block as the cast, don't change the cast.
if (UserBB == DefBB) continue;
// If we have already inserted a cast into this block, use it.
CastInst *&InsertedCast = InsertedCasts[UserBB];
if (!InsertedCast) {
BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedCast =
CastInst::Create(CI->getOpcode(), CI->getOperand(0), CI->getType(), "",
InsertPt);
MadeChange = true;
}
// Replace a use of the cast with a use of the new cast.
TheUse = InsertedCast;
++NumCastUses;
}
// If we removed all uses, nuke the cast.
if (CI->use_empty()) {
CI->eraseFromParent();
MadeChange = true;
}
return MadeChange;
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attribute CalleeRetAttr,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
!isa<UnreachableInst>(Term)))
return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(BBI))
return false;
}
// If the block ends with a void return or unreachable, it doesn't matter
// what the call's return type is.
if (!Ret || Ret->getNumOperands() == 0) return true;
// If the return value is undef, it doesn't matter what the call's
// return type is.
if (isa<UndefValue>(Ret->getOperand(0))) return true;
// Conservatively require the attributes of the call to match those of
// the return. Ignore noalias because it doesn't affect the call sequence.
const Function *F = ExitBB->getParent();
Attribute CallerRetAttr = F->getAttributes().getRetAttributes();
if (AttrBuilder(CalleeRetAttr).removeAttribute(Attribute::NoAlias) !=
AttrBuilder(CallerRetAttr).removeAttribute(Attribute::NoAlias))
return false;
// It's not safe to eliminate the sign / zero extension of the return value.
if (CallerRetAttr.hasAttribute(Attribute::ZExt) ||
CallerRetAttr.hasAttribute(Attribute::SExt))
return false;
// Otherwise, make sure the unmodified return value of I is the return value.
// We handle two cases: multiple return values + scalars.
Value *RetVal = Ret->getOperand(0);
if (!isa<InsertValueInst>(RetVal) || !isa<StructType>(RetVal->getType()))
// Handle scalars first.
return getNoopInput(Ret->getOperand(0), TLI) == I;
// If this is an aggregate return, look through the insert/extract values and
// see if each is transparent.
for (unsigned i = 0, e =cast<StructType>(RetVal->getType())->getNumElements();
i != e; ++i) {
const Value *InScalar = FindInsertedValue(RetVal, i);
if (InScalar == 0) return false;
InScalar = getNoopInput(InScalar, TLI);
// If the scalar value being inserted is an extractvalue of the right index
// from the call, then everything is good.
const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(InScalar);
if (EVI == 0 || EVI->getOperand(0) != I || EVI->getNumIndices() != 1 ||
EVI->getIndices()[0] != i)
return false;
}
return true;
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS, Attributes CalleeRetAttr,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!GuaranteedTailCallOpt || !isa<UnreachableInst>(Term))) return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!I->isSafeToSpeculativelyExecute())
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!BBI->isSafeToSpeculativelyExecute())
return false;
}
// If the block ends with a void return or unreachable, it doesn't matter
// what the call's return type is.
if (!Ret || Ret->getNumOperands() == 0) return true;
// If the return value is undef, it doesn't matter what the call's
// return type is.
if (isa<UndefValue>(Ret->getOperand(0))) return true;
// Conservatively require the attributes of the call to match those of
// the return. Ignore noalias because it doesn't affect the call sequence.
const Function *F = ExitBB->getParent();
unsigned CallerRetAttr = F->getAttributes().getRetAttributes();
if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
return false;
// It's not safe to eliminate the sign / zero extension of the return value.
if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
return false;
// Otherwise, make sure the unmodified return value of I is the return value.
for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
U = dyn_cast<Instruction>(U->getOperand(0))) {
if (!U)
return false;
if (!U->hasOneUse())
return false;
if (U == I)
break;
// Check for a truly no-op truncate.
if (isa<TruncInst>(U) &&
TLI.isTruncateFree(U->getOperand(0)->getType(), U->getType()))
continue;
// Check for a truly no-op bitcast.
if (isa<BitCastInst>(U) &&
(U->getOperand(0)->getType() == U->getType() ||
(U->getOperand(0)->getType()->isPointerTy() &&
U->getType()->isPointerTy())))
continue;
// Otherwise it's not a true no-op.
return false;
}
return true;
}
/// EmitJumpTableInfo - Print assembly representations of the jump tables used
/// by the current function to the current output stream.
///
void AsmPrinter::EmitJumpTableInfo(MachineJumpTableInfo *MJTI,
MachineFunction &MF) {
const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables();
if (JT.empty()) return;
bool IsPic = TM.getRelocationModel() == Reloc::PIC_;
// Use JumpTableDirective otherwise honor the entry size from the jump table
// info.
const char *JTEntryDirective = TAI->getJumpTableDirective();
bool HadJTEntryDirective = JTEntryDirective != NULL;
if (!HadJTEntryDirective) {
JTEntryDirective = MJTI->getEntrySize() == 4 ?
TAI->getData32bitsDirective() : TAI->getData64bitsDirective();
}
// Pick the directive to use to print the jump table entries, and switch to
// the appropriate section.
TargetLowering *LoweringInfo = TM.getTargetLowering();
const char* JumpTableDataSection = TAI->getJumpTableDataSection();
if ((IsPic && !(LoweringInfo && LoweringInfo->usesGlobalOffsetTable())) ||
!JumpTableDataSection) {
// In PIC mode, we need to emit the jump table to the same section as the
// function body itself, otherwise the label differences won't make sense.
// We should also do if the section name is NULL.
const Function *F = MF.getFunction();
SwitchToTextSection(getSectionForFunction(*F).c_str(), F);
} else {
SwitchToDataSection(JumpTableDataSection);
}
EmitAlignment(Log2_32(MJTI->getAlignment()));
for (unsigned i = 0, e = JT.size(); i != e; ++i) {
const std::vector<MachineBasicBlock*> &JTBBs = JT[i].MBBs;
// If this jump table was deleted, ignore it.
if (JTBBs.empty()) continue;
// For PIC codegen, if possible we want to use the SetDirective to reduce
// the number of relocations the assembler will generate for the jump table.
// Set directives are all printed before the jump table itself.
std::set<MachineBasicBlock*> EmittedSets;
if (TAI->getSetDirective() && IsPic)
for (unsigned ii = 0, ee = JTBBs.size(); ii != ee; ++ii)
if (EmittedSets.insert(JTBBs[ii]).second)
printSetLabel(i, JTBBs[ii]);
// On some targets (e.g. darwin) we want to emit two consequtive labels
// before each jump table. The first label is never referenced, but tells
// the assembler and linker the extents of the jump table object. The
// second label is actually referenced by the code.
if (const char *JTLabelPrefix = TAI->getJumpTableSpecialLabelPrefix())
O << JTLabelPrefix << "JTI" << getFunctionNumber() << '_' << i << ":\n";
O << TAI->getPrivateGlobalPrefix() << "JTI" << getFunctionNumber()
<< '_' << i << ":\n";
for (unsigned ii = 0, ee = JTBBs.size(); ii != ee; ++ii) {
O << JTEntryDirective << ' ';
// If we have emitted set directives for the jump table entries, print
// them rather than the entries themselves. If we're emitting PIC, then
// emit the table entries as differences between two text section labels.
// If we're emitting non-PIC code, then emit the entries as direct
// references to the target basic blocks.
if (!EmittedSets.empty()) {
O << TAI->getPrivateGlobalPrefix() << getFunctionNumber()
<< '_' << i << "_set_" << JTBBs[ii]->getNumber();
} else if (IsPic) {
printBasicBlockLabel(JTBBs[ii], false, false);
// If the arch uses custom Jump Table directives, don't calc relative to
// JT
if (!HadJTEntryDirective)
O << '-' << TAI->getPrivateGlobalPrefix() << "JTI"
<< getFunctionNumber() << '_' << i;
} else {
printBasicBlockLabel(JTBBs[ii], false, false);
}
O << '\n';
}
}
}
/// Test if the given instruction is in a position to be optimized
/// with a tail-call. This roughly means that it's in a block with
/// a return and there's nothing that needs to be scheduled
/// between it and the return.
///
/// This function only tests target-independent requirements.
bool llvm::isInTailCallPosition(ImmutableCallSite CS,
const TargetLowering &TLI) {
const Instruction *I = CS.getInstruction();
const BasicBlock *ExitBB = I->getParent();
const TerminatorInst *Term = ExitBB->getTerminator();
const ReturnInst *Ret = dyn_cast<ReturnInst>(Term);
// The block must end in a return statement or unreachable.
//
// FIXME: Decline tailcall if it's not guaranteed and if the block ends in
// an unreachable, for now. The way tailcall optimization is currently
// implemented means it will add an epilogue followed by a jump. That is
// not profitable. Also, if the callee is a special function (e.g.
// longjmp on x86), it can end up causing miscompilation that has not
// been fully understood.
if (!Ret &&
(!TLI.getTargetMachine().Options.GuaranteedTailCallOpt ||
!isa<UnreachableInst>(Term)))
return false;
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(I))
for (BasicBlock::const_iterator BBI = prior(prior(ExitBB->end())); ;
--BBI) {
if (&*BBI == I)
break;
// Debug info intrinsics do not get in the way of tail call optimization.
if (isa<DbgInfoIntrinsic>(BBI))
continue;
if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
!isSafeToSpeculativelyExecute(BBI))
return false;
}
// If the block ends with a void return or unreachable, it doesn't matter
// what the call's return type is.
if (!Ret || Ret->getNumOperands() == 0) return true;
// If the return value is undef, it doesn't matter what the call's
// return type is.
if (isa<UndefValue>(Ret->getOperand(0))) return true;
// Make sure the attributes attached to each return are compatible.
AttrBuilder CallerAttrs(ExitBB->getParent()->getAttributes(),
AttributeSet::ReturnIndex);
AttrBuilder CalleeAttrs(cast<CallInst>(I)->getAttributes(),
AttributeSet::ReturnIndex);
// Noalias is completely benign as far as calling convention goes, it
// shouldn't affect whether the call is a tail call.
CallerAttrs = CallerAttrs.removeAttribute(Attribute::NoAlias);
CalleeAttrs = CalleeAttrs.removeAttribute(Attribute::NoAlias);
bool AllowDifferingSizes = true;
if (CallerAttrs.contains(Attribute::ZExt)) {
if (!CalleeAttrs.contains(Attribute::ZExt))
return false;
AllowDifferingSizes = false;
CallerAttrs.removeAttribute(Attribute::ZExt);
CalleeAttrs.removeAttribute(Attribute::ZExt);
} else if (CallerAttrs.contains(Attribute::SExt)) {
if (!CalleeAttrs.contains(Attribute::SExt))
return false;
AllowDifferingSizes = false;
CallerAttrs.removeAttribute(Attribute::SExt);
CalleeAttrs.removeAttribute(Attribute::SExt);
}
// If they're still different, there's some facet we don't understand
// (currently only "inreg", but in future who knows). It may be OK but the
// only safe option is to reject the tail call.
if (CallerAttrs != CalleeAttrs)
return false;
const Value *RetVal = Ret->getOperand(0), *CallVal = I;
SmallVector<unsigned, 4> RetPath, CallPath;
SmallVector<CompositeType *, 4> RetSubTypes, CallSubTypes;
bool RetEmpty = !firstRealType(RetVal->getType(), RetSubTypes, RetPath);
bool CallEmpty = !firstRealType(CallVal->getType(), CallSubTypes, CallPath);
// Nothing's actually returned, it doesn't matter what the callee put there
// it's a valid tail call.
if (RetEmpty)
return true;
// Iterate pairwise through each of the value types making up the tail call
// and the corresponding return. For each one we want to know whether it's
// essentially going directly from the tail call to the ret, via operations
// that end up not generating any code.
//
// We allow a certain amount of covariance here. For example it's permitted
// for the tail call to define more bits than the ret actually cares about
// (e.g. via a truncate).
do {
if (CallEmpty) {
// We've exhausted the values produced by the tail call instruction, the
// rest are essentially undef. The type doesn't really matter, but we need
// *something*.
Type *SlotType = RetSubTypes.back()->getTypeAtIndex(RetPath.back());
CallVal = UndefValue::get(SlotType);
}
// The manipulations performed when we're looking through an insertvalue or
// an extractvalue would happen at the front of the RetPath list, so since
// we have to copy it anyway it's more efficient to create a reversed copy.
using std::copy;
SmallVector<unsigned, 4> TmpRetPath, TmpCallPath;
copy(RetPath.rbegin(), RetPath.rend(), std::back_inserter(TmpRetPath));
copy(CallPath.rbegin(), CallPath.rend(), std::back_inserter(TmpCallPath));
// Finally, we can check whether the value produced by the tail call at this
// index is compatible with the value we return.
if (!slotOnlyDiscardsData(RetVal, CallVal, TmpRetPath, TmpCallPath,
AllowDifferingSizes, TLI))
return false;
CallEmpty = !nextRealType(CallSubTypes, CallPath);
} while(nextRealType(RetSubTypes, RetPath));
return true;
}
