//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "function-lowering-info"
-#include "FunctionLoweringInfo.h"
-#include "llvm/CallingConv.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Module.h"
-#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/FunctionLoweringInfo.h"
+#include "llvm/ADT/PostOrderIterator.h"
+#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/Analysis/DebugInfo.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetFrameInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetIntrinsicInfo.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetOptions.h"
-#include "llvm/Support/Compiler.h"
+#include "llvm/CodeGen/WinEHFuncInfo.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DebugInfo.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetFrameLowering.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/TargetOptions.h"
+#include "llvm/Target/TargetRegisterInfo.h"
+#include "llvm/Target/TargetSubtargetInfo.h"
#include <algorithm>
using namespace llvm;
-/// ComputeLinearIndex - Given an LLVM IR aggregate type and a sequence
-/// of insertvalue or extractvalue indices that identify a member, return
-/// the linearized index of the start of the member.
-///
-unsigned llvm::ComputeLinearIndex(const TargetLowering &TLI, const Type *Ty,
- const unsigned *Indices,
- const unsigned *IndicesEnd,
- unsigned CurIndex) {
- // Base case: We're done.
- if (Indices && Indices == IndicesEnd)
- return CurIndex;
-
- // Given a struct type, recursively traverse the elements.
- if (const StructType *STy = dyn_cast<StructType>(Ty)) {
- for (StructType::element_iterator EB = STy->element_begin(),
- EI = EB,
- EE = STy->element_end();
- EI != EE; ++EI) {
- if (Indices && *Indices == unsigned(EI - EB))
- return ComputeLinearIndex(TLI, *EI, Indices+1, IndicesEnd, CurIndex);
- CurIndex = ComputeLinearIndex(TLI, *EI, 0, 0, CurIndex);
- }
- return CurIndex;
- }
- // Given an array type, recursively traverse the elements.
- else if (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
- const Type *EltTy = ATy->getElementType();
- for (unsigned i = 0, e = ATy->getNumElements(); i != e; ++i) {
- if (Indices && *Indices == i)
- return ComputeLinearIndex(TLI, EltTy, Indices+1, IndicesEnd, CurIndex);
- CurIndex = ComputeLinearIndex(TLI, EltTy, 0, 0, CurIndex);
- }
- return CurIndex;
- }
- // We haven't found the type we're looking for, so keep searching.
- return CurIndex + 1;
-}
-
-/// 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, const Type *Ty,
- SmallVectorImpl<EVT> &ValueVTs,
- SmallVectorImpl<uint64_t> *Offsets,
- uint64_t StartingOffset) {
- // Given a struct type, recursively traverse the elements.
- if (const StructType *STy = dyn_cast<StructType>(Ty)) {
- const StructLayout *SL = TLI.getTargetData()->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 (const ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
- const Type *EltTy = ATy->getElementType();
- uint64_t EltSize = TLI.getTargetData()->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);
-}
+#define DEBUG_TYPE "function-lowering-info"
/// isUsedOutsideOfDefiningBlock - Return true if this instruction is used by
/// PHI nodes or outside of the basic block that defines it, or used by a
/// switch or atomic instruction, which may expand to multiple basic blocks.
static bool isUsedOutsideOfDefiningBlock(const Instruction *I) {
+ if (I->use_empty()) return false;
if (isa<PHINode>(I)) return true;
const BasicBlock *BB = I->getParent();
- for (Value::const_use_iterator UI = I->use_begin(), E = I->use_end();
- UI != E; ++UI)
- if (cast<Instruction>(*UI)->getParent() != BB || isa<PHINode>(*UI))
+ for (const User *U : I->users())
+ if (cast<Instruction>(U)->getParent() != BB || isa<PHINode>(U))
return true;
+
return false;
}
-/// isOnlyUsedInEntryBlock - If the specified argument is only used in the
-/// entry block, return true. This includes arguments used by switches, since
-/// the switch may expand into multiple basic blocks.
-static bool isOnlyUsedInEntryBlock(const Argument *A, bool EnableFastISel) {
- // With FastISel active, we may be splitting blocks, so force creation
- // of virtual registers for all non-dead arguments.
- // Don't force virtual registers for byval arguments though, because
- // fast-isel can't handle those in all cases.
- if (EnableFastISel && !A->hasByValAttr())
- return A->use_empty();
-
- const BasicBlock *Entry = A->getParent()->begin();
- for (Value::const_use_iterator UI = A->use_begin(), E = A->use_end();
- UI != E; ++UI)
- if (cast<Instruction>(*UI)->getParent() != Entry || isa<SwitchInst>(*UI))
- return false; // Use not in entry block.
- return true;
+static ISD::NodeType getPreferredExtendForValue(const Value *V) {
+ // For the users of the source value being used for compare instruction, if
+ // the number of signed predicate is greater than unsigned predicate, we
+ // prefer to use SIGN_EXTEND.
+ //
+ // With this optimization, we would be able to reduce some redundant sign or
+ // zero extension instruction, and eventually more machine CSE opportunities
+ // can be exposed.
+ ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
+ unsigned NumOfSigned = 0, NumOfUnsigned = 0;
+ for (const User *U : V->users()) {
+ if (const auto *CI = dyn_cast<CmpInst>(U)) {
+ NumOfSigned += CI->isSigned();
+ NumOfUnsigned += CI->isUnsigned();
+ }
+ }
+ if (NumOfSigned > NumOfUnsigned)
+ ExtendKind = ISD::SIGN_EXTEND;
+
+ return ExtendKind;
}
-FunctionLoweringInfo::FunctionLoweringInfo(const TargetLowering &tli)
- : TLI(tli) {
+namespace {
+struct WinEHNumbering {
+ WinEHNumbering(WinEHFuncInfo &FuncInfo) : FuncInfo(FuncInfo), NextState(0) {}
+
+ WinEHFuncInfo &FuncInfo;
+ int NextState;
+
+ SmallVector<ActionHandler *, 4> HandlerStack;
+ SmallPtrSet<const Function *, 4> VisitedHandlers;
+
+ int currentEHNumber() const {
+ return HandlerStack.empty() ? -1 : HandlerStack.back()->getEHState();
+ }
+
+ void createUnwindMapEntry(int ToState, ActionHandler *AH);
+ void createTryBlockMapEntry(int TryLow, int TryHigh,
+ ArrayRef<CatchHandler *> Handlers);
+ void processCallSite(ArrayRef<ActionHandler *> Actions, ImmutableCallSite CS);
+ void calculateStateNumbers(const Function &F);
+};
}
void FunctionLoweringInfo::set(const Function &fn, MachineFunction &mf,
- bool EnableFastISel) {
+ SelectionDAG *DAG) {
Fn = &fn;
MF = &mf;
+ TLI = MF->getSubtarget().getTargetLowering();
RegInfo = &MF->getRegInfo();
+ MachineModuleInfo &MMI = MF->getMMI();
- // Create a vreg for each argument register that is not dead and is used
- // outside of the entry block for the function.
- for (Function::const_arg_iterator AI = Fn->arg_begin(), E = Fn->arg_end();
- AI != E; ++AI)
- if (!isOnlyUsedInEntryBlock(AI, EnableFastISel))
- InitializeRegForValue(AI);
+ // Check whether the function can return without sret-demotion.
+ SmallVector<ISD::OutputArg, 4> Outs;
+ GetReturnInfo(Fn->getReturnType(), Fn->getAttributes(), Outs, *TLI);
+ CanLowerReturn = TLI->CanLowerReturn(Fn->getCallingConv(), *MF,
+ Fn->isVarArg(), Outs, Fn->getContext());
// Initialize the mapping of values to registers. This is only set up for
// instruction values that are used outside of the block that defines
// them.
Function::const_iterator BB = Fn->begin(), EB = Fn->end();
- for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I)
- if (const AllocaInst *AI = dyn_cast<AllocaInst>(I))
- if (const ConstantInt *CUI = dyn_cast<ConstantInt>(AI->getArraySize())) {
- const Type *Ty = AI->getAllocatedType();
- uint64_t TySize = TLI.getTargetData()->getTypeAllocSize(Ty);
- unsigned Align =
- std::max((unsigned)TLI.getTargetData()->getPrefTypeAlignment(Ty),
- AI->getAlignment());
-
- TySize *= CUI->getZExtValue(); // Get total allocated size.
- if (TySize == 0) TySize = 1; // Don't create zero-sized stack objects.
- StaticAllocaMap[AI] =
- MF->getFrameInfo()->CreateStackObject(TySize, Align, false);
+ for (; BB != EB; ++BB)
+ for (BasicBlock::const_iterator I = BB->begin(), E = BB->end();
+ I != E; ++I) {
+ if (const AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
+ // Static allocas can be folded into the initial stack frame adjustment.
+ if (AI->isStaticAlloca()) {
+ const ConstantInt *CUI = cast<ConstantInt>(AI->getArraySize());
+ Type *Ty = AI->getAllocatedType();
+ uint64_t TySize = TLI->getDataLayout()->getTypeAllocSize(Ty);
+ unsigned Align =
+ std::max((unsigned)TLI->getDataLayout()->getPrefTypeAlignment(Ty),
+ AI->getAlignment());
+
+ TySize *= CUI->getZExtValue(); // Get total allocated size.
+ if (TySize == 0) TySize = 1; // Don't create zero-sized stack objects.
+
+ StaticAllocaMap[AI] =
+ MF->getFrameInfo()->CreateStackObject(TySize, Align, false, AI);
+
+ } else {
+ unsigned Align = std::max(
+ (unsigned)TLI->getDataLayout()->getPrefTypeAlignment(
+ AI->getAllocatedType()),
+ AI->getAlignment());
+ unsigned StackAlign =
+ MF->getSubtarget().getFrameLowering()->getStackAlignment();
+ if (Align <= StackAlign)
+ Align = 0;
+ // Inform the Frame Information that we have variable-sized objects.
+ MF->getFrameInfo()->CreateVariableSizedObject(Align ? Align : 1, AI);
+ }
}
- for (; BB != EB; ++BB)
- for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I != E; ++I)
- if (!I->use_empty() && isUsedOutsideOfDefiningBlock(I))
+ // Look for inline asm that clobbers the SP register.
+ if (isa<CallInst>(I) || isa<InvokeInst>(I)) {
+ ImmutableCallSite CS(I);
+ if (isa<InlineAsm>(CS.getCalledValue())) {
+ unsigned SP = TLI->getStackPointerRegisterToSaveRestore();
+ const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
+ std::vector<TargetLowering::AsmOperandInfo> Ops =
+ TLI->ParseConstraints(TRI, CS);
+ for (size_t I = 0, E = Ops.size(); I != E; ++I) {
+ TargetLowering::AsmOperandInfo &Op = Ops[I];
+ if (Op.Type == InlineAsm::isClobber) {
+ // Clobbers don't have SDValue operands, hence SDValue().
+ TLI->ComputeConstraintToUse(Op, SDValue(), DAG);
+ std::pair<unsigned, const TargetRegisterClass *> PhysReg =
+ TLI->getRegForInlineAsmConstraint(TRI, Op.ConstraintCode,
+ Op.ConstraintVT);
+ if (PhysReg.first == SP)
+ MF->getFrameInfo()->setHasInlineAsmWithSPAdjust(true);
+ }
+ }
+ }
+ }
+
+ // Look for calls to the @llvm.va_start intrinsic. We can omit some
+ // prologue boilerplate for variadic functions that don't examine their
+ // arguments.
+ if (const auto *II = dyn_cast<IntrinsicInst>(I)) {
+ if (II->getIntrinsicID() == Intrinsic::vastart)
+ MF->getFrameInfo()->setHasVAStart(true);
+ }
+
+ // If we have a musttail call in a variadic funciton, we need to ensure we
+ // forward implicit register parameters.
+ if (const auto *CI = dyn_cast<CallInst>(I)) {
+ if (CI->isMustTailCall() && Fn->isVarArg())
+ MF->getFrameInfo()->setHasMustTailInVarArgFunc(true);
+ }
+
+ // Mark values used outside their block as exported, by allocating
+ // a virtual register for them.
+ if (isUsedOutsideOfDefiningBlock(I))
if (!isa<AllocaInst>(I) ||
!StaticAllocaMap.count(cast<AllocaInst>(I)))
InitializeRegForValue(I);
+ // Collect llvm.dbg.declare information. This is done now instead of
+ // during the initial isel pass through the IR so that it is done
+ // in a predictable order.
+ if (const DbgDeclareInst *DI = dyn_cast<DbgDeclareInst>(I)) {
+ DIVariable DIVar = DI->getVariable();
+ if (MMI.hasDebugInfo() && DIVar && DI->getDebugLoc()) {
+ // Don't handle byval struct arguments or VLAs, for example.
+ // Non-byval arguments are handled here (they refer to the stack
+ // temporary alloca at this point).
+ const Value *Address = DI->getAddress();
+ if (Address) {
+ if (const BitCastInst *BCI = dyn_cast<BitCastInst>(Address))
+ Address = BCI->getOperand(0);
+ if (const AllocaInst *AI = dyn_cast<AllocaInst>(Address)) {
+ DenseMap<const AllocaInst *, int>::iterator SI =
+ StaticAllocaMap.find(AI);
+ if (SI != StaticAllocaMap.end()) { // Check for VLAs.
+ int FI = SI->second;
+ MMI.setVariableDbgInfo(DI->getVariable(), DI->getExpression(),
+ FI, DI->getDebugLoc());
+ }
+ }
+ }
+ }
+ }
+
+ // Decide the preferred extend type for a value.
+ PreferredExtendType[I] = getPreferredExtendForValue(I);
+ }
+
// Create an initial MachineBasicBlock for each LLVM BasicBlock in F. This
// also creates the initial PHI MachineInstrs, though none of the input
// operands are populated.
const PHINode *PN = dyn_cast<PHINode>(I); ++I) {
if (PN->use_empty()) continue;
+ // Skip empty types
+ if (PN->getType()->isEmptyTy())
+ continue;
+
DebugLoc DL = PN->getDebugLoc();
unsigned PHIReg = ValueMap[PN];
assert(PHIReg && "PHI node does not have an assigned virtual register!");
SmallVector<EVT, 4> ValueVTs;
- ComputeValueVTs(TLI, PN->getType(), ValueVTs);
+ ComputeValueVTs(*TLI, PN->getType(), ValueVTs);
for (unsigned vti = 0, vte = ValueVTs.size(); vti != vte; ++vti) {
EVT VT = ValueVTs[vti];
- unsigned NumRegisters = TLI.getNumRegisters(Fn->getContext(), VT);
- const TargetInstrInfo *TII = MF->getTarget().getInstrInfo();
+ unsigned NumRegisters = TLI->getNumRegisters(Fn->getContext(), VT);
+ const TargetInstrInfo *TII = MF->getSubtarget().getInstrInfo();
for (unsigned i = 0; i != NumRegisters; ++i)
BuildMI(MBB, DL, TII->get(TargetOpcode::PHI), PHIReg + i);
PHIReg += NumRegisters;
}
// Mark landing pad blocks.
- for (BB = Fn->begin(); BB != EB; ++BB)
- if (const InvokeInst *Invoke = dyn_cast<InvokeInst>(BB->getTerminator()))
+ const LandingPadInst *LP = nullptr;
+ for (BB = Fn->begin(); BB != EB; ++BB) {
+ if (const auto *Invoke = dyn_cast<InvokeInst>(BB->getTerminator()))
MBBMap[Invoke->getSuccessor(1)]->setIsLandingPad();
+ if (BB->isLandingPad())
+ LP = BB->getLandingPadInst();
+ }
+
+ // Calculate EH numbers for MSVC C++ EH and save SEH handlers if necessary.
+ EHPersonality Personality = EHPersonality::Unknown;
+ if (LP)
+ Personality = classifyEHPersonality(LP->getPersonalityFn());
+ if (Personality == EHPersonality::MSVC_Win64SEH) {
+ addSEHHandlersForLPads();
+ } else if (Personality == EHPersonality::MSVC_CXX) {
+ const Function *WinEHParentFn = MMI.getWinEHParent(&fn);
+ WinEHFuncInfo &FI = MMI.getWinEHFuncInfo(WinEHParentFn);
+ if (FI.LandingPadStateMap.empty()) {
+ WinEHNumbering Num(FI);
+ Num.calculateStateNumbers(*WinEHParentFn);
+ // Pop everything on the handler stack.
+ Num.processCallSite(None, ImmutableCallSite());
+ }
+ }
+}
+
+void FunctionLoweringInfo::addSEHHandlersForLPads() {
+ MachineModuleInfo &MMI = MF->getMMI();
+
+ // Iterate over all landing pads with llvm.eh.actions calls.
+ for (const BasicBlock &BB : *Fn) {
+ const LandingPadInst *LP = BB.getLandingPadInst();
+ if (!LP)
+ continue;
+ const IntrinsicInst *ActionsCall =
+ dyn_cast<IntrinsicInst>(LP->getNextNode());
+ if (!ActionsCall ||
+ ActionsCall->getIntrinsicID() != Intrinsic::eh_actions)
+ continue;
+
+ // Parse the llvm.eh.actions call we found.
+ MachineBasicBlock *LPadMBB = MBBMap[LP->getParent()];
+ SmallVector<ActionHandler *, 4> Actions;
+ parseEHActions(ActionsCall, Actions);
+
+ // Iterate EH actions from most to least precedence, which means
+ // iterating in reverse.
+ for (auto I = Actions.rbegin(), E = Actions.rend(); I != E; ++I) {
+ ActionHandler *Action = *I;
+ if (auto *CH = dyn_cast<CatchHandler>(Action)) {
+ const auto *Filter =
+ dyn_cast<Function>(CH->getSelector()->stripPointerCasts());
+ assert((Filter || CH->getSelector()->isNullValue()) &&
+ "expected function or catch-all");
+ const auto *RecoverBA =
+ cast<BlockAddress>(CH->getHandlerBlockOrFunc());
+ MMI.addSEHCatchHandler(LPadMBB, Filter, RecoverBA);
+ } else {
+ assert(isa<CleanupHandler>(Action));
+ const auto *Fini = cast<Function>(Action->getHandlerBlockOrFunc());
+ MMI.addSEHCleanupHandler(LPadMBB, Fini);
+ }
+ }
+ DeleteContainerPointers(Actions);
+ }
+}
+
+void WinEHNumbering::createUnwindMapEntry(int ToState, ActionHandler *AH) {
+ WinEHUnwindMapEntry UME;
+ UME.ToState = ToState;
+ if (auto *CH = dyn_cast_or_null<CleanupHandler>(AH))
+ UME.Cleanup = cast<Function>(CH->getHandlerBlockOrFunc());
+ else
+ UME.Cleanup = nullptr;
+ FuncInfo.UnwindMap.push_back(UME);
+}
+
+void WinEHNumbering::createTryBlockMapEntry(int TryLow, int TryHigh,
+ ArrayRef<CatchHandler *> Handlers) {
+ WinEHTryBlockMapEntry TBME;
+ TBME.TryLow = TryLow;
+ TBME.TryHigh = TryHigh;
+ assert(TBME.TryLow <= TBME.TryHigh);
+ for (CatchHandler *CH : Handlers) {
+ WinEHHandlerType HT;
+ if (CH->getSelector()->isNullValue()) {
+ HT.Adjectives = 0x40;
+ HT.TypeDescriptor = nullptr;
+ } else {
+ auto *GV = cast<GlobalVariable>(CH->getSelector()->stripPointerCasts());
+ // Selectors are always pointers to GlobalVariables with 'struct' type.
+ // The struct has two fields, adjectives and a type descriptor.
+ auto *CS = cast<ConstantStruct>(GV->getInitializer());
+ HT.Adjectives =
+ cast<ConstantInt>(CS->getAggregateElement(0U))->getZExtValue();
+ HT.TypeDescriptor =
+ cast<GlobalVariable>(CS->getAggregateElement(1)->stripPointerCasts());
+ }
+ HT.Handler = cast<Function>(CH->getHandlerBlockOrFunc());
+ HT.CatchObjRecoverIdx = CH->getExceptionVarIndex();
+ TBME.HandlerArray.push_back(HT);
+ }
+ FuncInfo.TryBlockMap.push_back(TBME);
+}
+
+static void print_name(const Value *V) {
+#ifndef NDEBUG
+ if (!V) {
+ DEBUG(dbgs() << "null");
+ return;
+ }
+
+ if (const auto *F = dyn_cast<Function>(V))
+ DEBUG(dbgs() << F->getName());
+ else
+ DEBUG(V->dump());
+#endif
+}
+
+void WinEHNumbering::processCallSite(ArrayRef<ActionHandler *> Actions,
+ ImmutableCallSite CS) {
+ int FirstMismatch = 0;
+ for (int E = std::min(HandlerStack.size(), Actions.size()); FirstMismatch < E;
+ ++FirstMismatch) {
+ if (HandlerStack[FirstMismatch]->getHandlerBlockOrFunc() !=
+ Actions[FirstMismatch]->getHandlerBlockOrFunc())
+ break;
+ delete Actions[FirstMismatch];
+ }
+
+ bool EnteringScope = (int)Actions.size() > FirstMismatch;
+
+ // Don't recurse while we are looping over the handler stack. Instead, defer
+ // the numbering of the catch handlers until we are done popping.
+ SmallVector<CatchHandler *, 4> PoppedCatches;
+ for (int I = HandlerStack.size() - 1; I >= FirstMismatch; --I) {
+ if (auto *CH = dyn_cast<CatchHandler>(HandlerStack.back())) {
+ PoppedCatches.push_back(CH);
+ } else {
+ // Delete cleanup handlers
+ delete HandlerStack.back();
+ }
+ HandlerStack.pop_back();
+ }
+
+ // We need to create a new state number if we are exiting a try scope and we
+ // will not push any more actions.
+ int TryHigh = NextState - 1;
+ if (!EnteringScope && !PoppedCatches.empty()) {
+ createUnwindMapEntry(currentEHNumber(), nullptr);
+ ++NextState;
+ }
+
+ int LastTryLowIdx = 0;
+ for (int I = 0, E = PoppedCatches.size(); I != E; ++I) {
+ CatchHandler *CH = PoppedCatches[I];
+ if (I + 1 == E || CH->getEHState() != PoppedCatches[I + 1]->getEHState()) {
+ int TryLow = CH->getEHState();
+ auto Handlers =
+ makeArrayRef(&PoppedCatches[LastTryLowIdx], I - LastTryLowIdx + 1);
+ createTryBlockMapEntry(TryLow, TryHigh, Handlers);
+ LastTryLowIdx = I + 1;
+ }
+ }
+
+ for (CatchHandler *CH : PoppedCatches) {
+ if (auto *F = dyn_cast<Function>(CH->getHandlerBlockOrFunc()))
+ calculateStateNumbers(*F);
+ delete CH;
+ }
+
+ bool LastActionWasCatch = false;
+ for (size_t I = FirstMismatch; I != Actions.size(); ++I) {
+ // We can reuse eh states when pushing two catches for the same invoke.
+ bool CurrActionIsCatch = isa<CatchHandler>(Actions[I]);
+ // FIXME: Reenable this optimization!
+ if (CurrActionIsCatch && LastActionWasCatch && false) {
+ Actions[I]->setEHState(currentEHNumber());
+ } else {
+ createUnwindMapEntry(currentEHNumber(), Actions[I]);
+ Actions[I]->setEHState(NextState);
+ NextState++;
+ DEBUG(dbgs() << "Creating unwind map entry for: (");
+ print_name(Actions[I]->getHandlerBlockOrFunc());
+ DEBUG(dbgs() << ", " << currentEHNumber() << ")\n");
+ }
+ HandlerStack.push_back(Actions[I]);
+ LastActionWasCatch = CurrActionIsCatch;
+ }
+
+ DEBUG(dbgs() << "In EHState " << currentEHNumber() << " for CallSite: ");
+ print_name(CS ? CS.getCalledValue() : nullptr);
+ DEBUG(dbgs() << '\n');
+}
+
+void WinEHNumbering::calculateStateNumbers(const Function &F) {
+ auto I = VisitedHandlers.insert(&F);
+ if (!I.second)
+ return; // We've already visited this handler, don't renumber it.
+
+ DEBUG(dbgs() << "Calculating state numbers for: " << F.getName() << '\n');
+ SmallVector<ActionHandler *, 4> ActionList;
+ for (const BasicBlock &BB : F) {
+ for (const Instruction &I : BB) {
+ const auto *CI = dyn_cast<CallInst>(&I);
+ if (!CI || CI->doesNotThrow())
+ continue;
+ processCallSite(None, CI);
+ }
+ const auto *II = dyn_cast<InvokeInst>(BB.getTerminator());
+ if (!II)
+ continue;
+ const LandingPadInst *LPI = II->getLandingPadInst();
+ auto *ActionsCall = dyn_cast<IntrinsicInst>(LPI->getNextNode());
+ if (!ActionsCall)
+ continue;
+ assert(ActionsCall->getIntrinsicID() == Intrinsic::eh_actions);
+ parseEHActions(ActionsCall, ActionList);
+ processCallSite(ActionList, II);
+ ActionList.clear();
+ FuncInfo.LandingPadStateMap[LPI] = currentEHNumber();
+ }
+
+ FuncInfo.CatchHandlerMaxState[&F] = NextState - 1;
}
/// clear - Clear out all the function-specific state. This returns this
CatchInfoFound.clear();
#endif
LiveOutRegInfo.clear();
+ VisitedBBs.clear();
+ ArgDbgValues.clear();
+ ByValArgFrameIndexMap.clear();
+ RegFixups.clear();
+ StatepointStackSlots.clear();
+ PreferredExtendType.clear();
}
-unsigned FunctionLoweringInfo::MakeReg(EVT VT) {
- return RegInfo->createVirtualRegister(TLI.getRegClassFor(VT));
+/// CreateReg - Allocate a single virtual register for the given type.
+unsigned FunctionLoweringInfo::CreateReg(MVT VT) {
+ return RegInfo->createVirtualRegister(
+ MF->getSubtarget().getTargetLowering()->getRegClassFor(VT));
}
-/// CreateRegForValue - Allocate the appropriate number of virtual registers of
+/// CreateRegs - Allocate the appropriate number of virtual registers of
/// the correctly promoted or expanded types. Assign these registers
/// consecutive vreg numbers and return the first assigned number.
///
/// In the case that the given value has struct or array type, this function
/// will assign registers for each member or element.
///
-unsigned FunctionLoweringInfo::CreateRegForValue(const Value *V) {
+unsigned FunctionLoweringInfo::CreateRegs(Type *Ty) {
+ const TargetLowering *TLI = MF->getSubtarget().getTargetLowering();
+
SmallVector<EVT, 4> ValueVTs;
- ComputeValueVTs(TLI, V->getType(), ValueVTs);
+ ComputeValueVTs(*TLI, Ty, ValueVTs);
unsigned FirstReg = 0;
for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
EVT ValueVT = ValueVTs[Value];
- EVT RegisterVT = TLI.getRegisterType(V->getContext(), ValueVT);
+ MVT RegisterVT = TLI->getRegisterType(Ty->getContext(), ValueVT);
- unsigned NumRegs = TLI.getNumRegisters(V->getContext(), ValueVT);
+ unsigned NumRegs = TLI->getNumRegisters(Ty->getContext(), ValueVT);
for (unsigned i = 0; i != NumRegs; ++i) {
- unsigned R = MakeReg(RegisterVT);
+ unsigned R = CreateReg(RegisterVT);
if (!FirstReg) FirstReg = R;
}
}
return FirstReg;
}
-/// ExtractTypeInfo - Returns the type info, possibly bitcast, encoded in V.
-GlobalVariable *llvm::ExtractTypeInfo(Value *V) {
- V = V->stripPointerCasts();
- GlobalVariable *GV = dyn_cast<GlobalVariable>(V);
-
- if (GV && GV->getName() == ".llvm.eh.catch.all.value") {
- assert(GV->hasInitializer() &&
- "The EH catch-all value must have an initializer");
- Value *Init = GV->getInitializer();
- GV = dyn_cast<GlobalVariable>(Init);
- if (!GV) V = cast<ConstantPointerNull>(Init);
+/// GetLiveOutRegInfo - Gets LiveOutInfo for a register, returning NULL if the
+/// register is a PHI destination and the PHI's LiveOutInfo is not valid. If
+/// the register's LiveOutInfo is for a smaller bit width, it is extended to
+/// the larger bit width by zero extension. The bit width must be no smaller
+/// than the LiveOutInfo's existing bit width.
+const FunctionLoweringInfo::LiveOutInfo *
+FunctionLoweringInfo::GetLiveOutRegInfo(unsigned Reg, unsigned BitWidth) {
+ if (!LiveOutRegInfo.inBounds(Reg))
+ return nullptr;
+
+ LiveOutInfo *LOI = &LiveOutRegInfo[Reg];
+ if (!LOI->IsValid)
+ return nullptr;
+
+ if (BitWidth > LOI->KnownZero.getBitWidth()) {
+ LOI->NumSignBits = 1;
+ LOI->KnownZero = LOI->KnownZero.zextOrTrunc(BitWidth);
+ LOI->KnownOne = LOI->KnownOne.zextOrTrunc(BitWidth);
}
- assert((GV || isa<ConstantPointerNull>(V)) &&
- "TypeInfo must be a global variable or NULL");
- return GV;
+ return LOI;
}
-/// AddCatchInfo - Extract the personality and type infos from an eh.selector
-/// call, and add them to the specified machine basic block.
-void llvm::AddCatchInfo(const CallInst &I, MachineModuleInfo *MMI,
- MachineBasicBlock *MBB) {
- // Inform the MachineModuleInfo of the personality for this landing pad.
- const ConstantExpr *CE = cast<ConstantExpr>(I.getOperand(2));
- assert(CE->getOpcode() == Instruction::BitCast &&
- isa<Function>(CE->getOperand(0)) &&
- "Personality should be a function");
- MMI->addPersonality(MBB, cast<Function>(CE->getOperand(0)));
-
- // Gather all the type infos for this landing pad and pass them along to
- // MachineModuleInfo.
- std::vector<const GlobalVariable *> TyInfo;
- unsigned N = I.getNumOperands();
-
- for (unsigned i = N - 1; i > 2; --i) {
- if (const ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(i))) {
- unsigned FilterLength = CI->getZExtValue();
- unsigned FirstCatch = i + FilterLength + !FilterLength;
- assert (FirstCatch <= N && "Invalid filter length");
-
- if (FirstCatch < N) {
- TyInfo.reserve(N - FirstCatch);
- for (unsigned j = FirstCatch; j < N; ++j)
- TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
- MMI->addCatchTypeInfo(MBB, TyInfo);
- TyInfo.clear();
- }
+/// ComputePHILiveOutRegInfo - Compute LiveOutInfo for a PHI's destination
+/// register based on the LiveOutInfo of its operands.
+void FunctionLoweringInfo::ComputePHILiveOutRegInfo(const PHINode *PN) {
+ Type *Ty = PN->getType();
+ if (!Ty->isIntegerTy() || Ty->isVectorTy())
+ return;
- if (!FilterLength) {
- // Cleanup.
- MMI->addCleanup(MBB);
- } else {
- // Filter.
- TyInfo.reserve(FilterLength - 1);
- for (unsigned j = i + 1; j < FirstCatch; ++j)
- TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
- MMI->addFilterTypeInfo(MBB, TyInfo);
- TyInfo.clear();
- }
+ SmallVector<EVT, 1> ValueVTs;
+ ComputeValueVTs(*TLI, Ty, ValueVTs);
+ assert(ValueVTs.size() == 1 &&
+ "PHIs with non-vector integer types should have a single VT.");
+ EVT IntVT = ValueVTs[0];
- N = i;
- }
+ if (TLI->getNumRegisters(PN->getContext(), IntVT) != 1)
+ return;
+ IntVT = TLI->getTypeToTransformTo(PN->getContext(), IntVT);
+ unsigned BitWidth = IntVT.getSizeInBits();
+
+ unsigned DestReg = ValueMap[PN];
+ if (!TargetRegisterInfo::isVirtualRegister(DestReg))
+ return;
+ LiveOutRegInfo.grow(DestReg);
+ LiveOutInfo &DestLOI = LiveOutRegInfo[DestReg];
+
+ Value *V = PN->getIncomingValue(0);
+ if (isa<UndefValue>(V) || isa<ConstantExpr>(V)) {
+ DestLOI.NumSignBits = 1;
+ APInt Zero(BitWidth, 0);
+ DestLOI.KnownZero = Zero;
+ DestLOI.KnownOne = Zero;
+ return;
}
- if (N > 3) {
- TyInfo.reserve(N - 3);
- for (unsigned j = 3; j < N; ++j)
- TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
- MMI->addCatchTypeInfo(MBB, TyInfo);
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+ APInt Val = CI->getValue().zextOrTrunc(BitWidth);
+ DestLOI.NumSignBits = Val.getNumSignBits();
+ DestLOI.KnownZero = ~Val;
+ DestLOI.KnownOne = Val;
+ } else {
+ assert(ValueMap.count(V) && "V should have been placed in ValueMap when its"
+ "CopyToReg node was created.");
+ unsigned SrcReg = ValueMap[V];
+ if (!TargetRegisterInfo::isVirtualRegister(SrcReg)) {
+ DestLOI.IsValid = false;
+ return;
+ }
+ const LiveOutInfo *SrcLOI = GetLiveOutRegInfo(SrcReg, BitWidth);
+ if (!SrcLOI) {
+ DestLOI.IsValid = false;
+ return;
+ }
+ DestLOI = *SrcLOI;
}
-}
-void llvm::CopyCatchInfo(const BasicBlock *SrcBB, const BasicBlock *DestBB,
- MachineModuleInfo *MMI, FunctionLoweringInfo &FLI) {
- for (BasicBlock::const_iterator I = SrcBB->begin(), E = --SrcBB->end();
- I != E; ++I)
- if (const EHSelectorInst *EHSel = dyn_cast<EHSelectorInst>(I)) {
- // Apply the catch info to DestBB.
- AddCatchInfo(*EHSel, MMI, FLI.MBBMap[DestBB]);
-#ifndef NDEBUG
- if (!FLI.MBBMap[SrcBB]->isLandingPad())
- FLI.CatchInfoFound.insert(EHSel);
-#endif
+ assert(DestLOI.KnownZero.getBitWidth() == BitWidth &&
+ DestLOI.KnownOne.getBitWidth() == BitWidth &&
+ "Masks should have the same bit width as the type.");
+
+ for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
+ Value *V = PN->getIncomingValue(i);
+ if (isa<UndefValue>(V) || isa<ConstantExpr>(V)) {
+ DestLOI.NumSignBits = 1;
+ APInt Zero(BitWidth, 0);
+ DestLOI.KnownZero = Zero;
+ DestLOI.KnownOne = Zero;
+ return;
}
-}
-/// hasInlineAsmMemConstraint - Return true if the inline asm instruction being
-/// processed uses a memory 'm' constraint.
-bool
-llvm::hasInlineAsmMemConstraint(std::vector<InlineAsm::ConstraintInfo> &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;
+ if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
+ APInt Val = CI->getValue().zextOrTrunc(BitWidth);
+ DestLOI.NumSignBits = std::min(DestLOI.NumSignBits, Val.getNumSignBits());
+ DestLOI.KnownZero &= ~Val;
+ DestLOI.KnownOne &= Val;
+ continue;
}
- // Indirect operand accesses access memory.
- if (CI.isIndirect)
- return true;
+ assert(ValueMap.count(V) && "V should have been placed in ValueMap when "
+ "its CopyToReg node was created.");
+ unsigned SrcReg = ValueMap[V];
+ if (!TargetRegisterInfo::isVirtualRegister(SrcReg)) {
+ DestLOI.IsValid = false;
+ return;
+ }
+ const LiveOutInfo *SrcLOI = GetLiveOutRegInfo(SrcReg, BitWidth);
+ if (!SrcLOI) {
+ DestLOI.IsValid = false;
+ return;
+ }
+ DestLOI.NumSignBits = std::min(DestLOI.NumSignBits, SrcLOI->NumSignBits);
+ DestLOI.KnownZero &= SrcLOI->KnownZero;
+ DestLOI.KnownOne &= SrcLOI->KnownOne;
}
+}
- return false;
+/// setArgumentFrameIndex - Record frame index for the byval
+/// argument. This overrides previous frame index entry for this argument,
+/// if any.
+void FunctionLoweringInfo::setArgumentFrameIndex(const Argument *A,
+ int FI) {
+ ByValArgFrameIndexMap[A] = FI;
}
-/// getFCmpCondCode - Return the ISD condition code corresponding to
-/// the given LLVM IR floating-point condition code. This includes
-/// consideration of global floating-point math flags.
-///
-ISD::CondCode llvm::getFCmpCondCode(FCmpInst::Predicate Pred) {
- ISD::CondCode FPC, FOC;
- switch (Pred) {
- case FCmpInst::FCMP_FALSE: FOC = FPC = ISD::SETFALSE; break;
- case FCmpInst::FCMP_OEQ: FOC = ISD::SETEQ; FPC = ISD::SETOEQ; break;
- case FCmpInst::FCMP_OGT: FOC = ISD::SETGT; FPC = ISD::SETOGT; break;
- case FCmpInst::FCMP_OGE: FOC = ISD::SETGE; FPC = ISD::SETOGE; break;
- case FCmpInst::FCMP_OLT: FOC = ISD::SETLT; FPC = ISD::SETOLT; break;
- case FCmpInst::FCMP_OLE: FOC = ISD::SETLE; FPC = ISD::SETOLE; break;
- case FCmpInst::FCMP_ONE: FOC = ISD::SETNE; FPC = ISD::SETONE; break;
- case FCmpInst::FCMP_ORD: FOC = FPC = ISD::SETO; break;
- case FCmpInst::FCMP_UNO: FOC = FPC = ISD::SETUO; break;
- case FCmpInst::FCMP_UEQ: FOC = ISD::SETEQ; FPC = ISD::SETUEQ; break;
- case FCmpInst::FCMP_UGT: FOC = ISD::SETGT; FPC = ISD::SETUGT; break;
- case FCmpInst::FCMP_UGE: FOC = ISD::SETGE; FPC = ISD::SETUGE; break;
- case FCmpInst::FCMP_ULT: FOC = ISD::SETLT; FPC = ISD::SETULT; break;
- case FCmpInst::FCMP_ULE: FOC = ISD::SETLE; FPC = ISD::SETULE; break;
- case FCmpInst::FCMP_UNE: FOC = ISD::SETNE; FPC = ISD::SETUNE; break;
- case FCmpInst::FCMP_TRUE: FOC = FPC = ISD::SETTRUE; break;
- default:
- llvm_unreachable("Invalid FCmp predicate opcode!");
- FOC = FPC = ISD::SETFALSE;
- break;
- }
- if (FiniteOnlyFPMath())
- return FOC;
- else
- return FPC;
+/// getArgumentFrameIndex - Get frame index for the byval argument.
+/// If the argument does not have any assigned frame index then 0 is
+/// returned.
+int FunctionLoweringInfo::getArgumentFrameIndex(const Argument *A) {
+ DenseMap<const Argument *, int>::iterator I =
+ ByValArgFrameIndexMap.find(A);
+ if (I != ByValArgFrameIndexMap.end())
+ return I->second;
+ DEBUG(dbgs() << "Argument does not have assigned frame index!\n");
+ return 0;
}
-/// getICmpCondCode - Return the ISD condition code corresponding to
-/// the given LLVM IR integer condition code.
-///
-ISD::CondCode llvm::getICmpCondCode(ICmpInst::Predicate Pred) {
- switch (Pred) {
- case ICmpInst::ICMP_EQ: return ISD::SETEQ;
- case ICmpInst::ICMP_NE: return ISD::SETNE;
- case ICmpInst::ICMP_SLE: return ISD::SETLE;
- case ICmpInst::ICMP_ULE: return ISD::SETULE;
- case ICmpInst::ICMP_SGE: return ISD::SETGE;
- case ICmpInst::ICMP_UGE: return ISD::SETUGE;
- case ICmpInst::ICMP_SLT: return ISD::SETLT;
- case ICmpInst::ICMP_ULT: return ISD::SETULT;
- case ICmpInst::ICMP_SGT: return ISD::SETGT;
- case ICmpInst::ICMP_UGT: return ISD::SETUGT;
- default:
- llvm_unreachable("Invalid ICmp predicate opcode!");
- return ISD::SETNE;
+/// ComputeUsesVAFloatArgument - Determine if any floating-point values are
+/// being passed to this variadic function, and set the MachineModuleInfo's
+/// usesVAFloatArgument flag if so. This flag is used to emit an undefined
+/// reference to _fltused on Windows, which will link in MSVCRT's
+/// floating-point support.
+void llvm::ComputeUsesVAFloatArgument(const CallInst &I,
+ MachineModuleInfo *MMI)
+{
+ FunctionType *FT = cast<FunctionType>(
+ I.getCalledValue()->getType()->getContainedType(0));
+ if (FT->isVarArg() && !MMI->usesVAFloatArgument()) {
+ for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
+ Type* T = I.getArgOperand(i)->getType();
+ for (auto i : post_order(T)) {
+ if (i->isFloatingPointTy()) {
+ MMI->setUsesVAFloatArgument(true);
+ return;
+ }
+ }
+ }
}
}
-/// 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);
- const Function *F = ExitBB->getParent();
-
- // 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;
+/// AddLandingPadInfo - Extract the exception handling information from the
+/// landingpad instruction and add them to the specified machine module info.
+void llvm::AddLandingPadInfo(const LandingPadInst &I, MachineModuleInfo &MMI,
+ MachineBasicBlock *MBB) {
+ MMI.addPersonality(MBB,
+ cast<Function>(I.getPersonalityFn()->stripPointerCasts()));
+
+ if (I.isCleanup())
+ MMI.addCleanup(MBB);
+
+ // FIXME: New EH - Add the clauses in reverse order. This isn't 100% correct,
+ // but we need to do it this way because of how the DWARF EH emitter
+ // processes the clauses.
+ for (unsigned i = I.getNumClauses(); i != 0; --i) {
+ Value *Val = I.getClause(i - 1);
+ if (I.isCatch(i - 1)) {
+ MMI.addCatchTypeInfo(MBB,
+ dyn_cast<GlobalValue>(Val->stripPointerCasts()));
+ } else {
+ // Add filters in a list.
+ Constant *CVal = cast<Constant>(Val);
+ SmallVector<const GlobalValue*, 4> FilterList;
+ for (User::op_iterator
+ II = CVal->op_begin(), IE = CVal->op_end(); II != IE; ++II)
+ FilterList.push_back(cast<GlobalValue>((*II)->stripPointerCasts()));
+
+ MMI.addFilterTypeInfo(MBB, FilterList);
}
-
- // 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.
- 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;
}